Gretl Guide

Gretl Users Guide
Gnu Regression, Econometrics and Time-series Library

Allin Cottrell
Department of Economics
Wake Forest University
Riccardo Jack Lucchetti
Dipartimento di Economia
Universit Politecnica delle Marche
October, 2013
Permission is granted to copy, distribute and/or modify this document under the terms of the
GNU Free Documentation License, Version 1.1 or any later version published by the Free Software
Foundation (see http://www.gnu.org/licenses/fdl.html).
Contents
1 Introduction 1
1.1 Features at a glance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.3 Installing the programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
I Running the program 3
2 Getting started 4
2.1 Lets run a regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Estimation output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 The main window menus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 Keyboard shortcuts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5 The gretl toolbar . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3 Modes of working 12
3.1 Command scripts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Saving script objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 The gretl console . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.4 The Session concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4 Data les 18
4.1 Data le formats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.2 Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.3 Creating a dataset from scratch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.4 Structuring a dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.5 Panel data specics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
4.6 Missing data values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.7 Maximum size of data sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.8 Data le collections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.9 Assembling data from multiple sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5 Sub-sampling a dataset 31
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.2 Setting the sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.3 Restricting the sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
i
Contents ii
5.4 Panel data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.5 Resampling and bootstrapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6 Graphs and plots 36
6.1 Gnuplot graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.2 Boxplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
7 Joining data sources 41
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2 Basic syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.3 Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
7.4 Matching with keys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7.5 Aggregation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7.6 String-valued key variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7.7 A real-world case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
7.8 The representation of dates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
7.9 Time-series data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
7.10 Special handling of time columns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
7.11 Panel data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
8 Realtime data 58
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
8.2 Atomic format for realtime data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
8.3 More on time-related options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
8.4 Getting a certain data vintage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
8.5 Getting the n-th release for each observation period . . . . . . . . . . . . . . . . . . . . . 61
8.6 Getting the values at a xed lag after the observation period . . . . . . . . . . . . . . . . 62
8.7 Getting the revision history for an observation . . . . . . . . . . . . . . . . . . . . . . . . 63
9 Special functions in genr 66
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
9.2 Long-run variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
9.3 Cumulative densities and p-values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
9.4 Retrieving internal variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
9.5 The discrete Fourier transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
10 Gretl data types 71
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
10.2 Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
10.3 Scalars . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
10.4 Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
Contents iii
10.5 Lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
10.6 Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
10.7 Bundles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
10.8 The life cycle of gretl objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
11 Discrete variables 78
11.1 Declaring variables as discrete . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
11.2 Commands for discrete variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
12 Loop constructs 83
12.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
12.2 Loop control variants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
12.3 Progressive mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
12.4 Loop examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
13 User-dened functions 91
13.1 Dening a function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
13.2 Calling a function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
13.3 Deleting a function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
13.4 Function programming details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
13.5 Function packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
13.6 Memo: updating old-style functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
14 Named lists and strings 109
14.1 Named lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
14.2 Named strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
15 Matrix manipulation 118
15.1 Creating matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
15.2 Empty matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
15.3 Selecting sub-matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
15.4 Matrix operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
15.5 Matrixscalar operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
15.6 Matrix functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
15.7 Matrix accessors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
15.8 Namespace issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
15.9 Creating a data series from a matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
15.10Matrices and lists . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
15.11Deleting a matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
15.12Printing a matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
15.13Example: OLS using matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
Contents iv
16 Cheat sheet 135
16.1 Dataset handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
16.2 Creating/modifying variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
16.3 Neat tricks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
II Econometric methods 141
17 Robust covariance matrix estimation 142
17.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
17.2 Cross-sectional data and the HCCME . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
17.3 Time series data and HAC covariance matrices . . . . . . . . . . . . . . . . . . . . . . . . 144
17.4 Special issues with panel data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
17.5 The cluster-robust estimator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
18 Panel data 151
18.1 Estimation of panel models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
18.2 Autoregressive panel models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
19 Dynamic panel models 158
19.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
19.2 Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
19.3 Replication of DPD results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
19.4 Cross-country growth example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
19.5 Auxiliary test statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
19.6 Memo: dpanel options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
20 Nonlinear least squares 170
20.1 Introduction and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
20.2 Initializing the parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
20.3 NLS dialog window . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
20.4 Analytical and numerical derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
20.5 Controlling termination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
20.6 Details on the code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
20.7 Numerical accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
21 Maximum likelihood estimation 175
21.1 Generic ML estimation with gretl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
21.2 Gamma estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
21.3 Stochastic frontier cost function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
21.4 GARCH models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
21.5 Analytical derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
Contents v
21.6 Debugging ML scripts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
21.7 Using functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
21.8 Advanced use of mle: functions, analytical derivatives, algorithm choice . . . . . . . . 187
22 GMM estimation 192
22.1 Introduction and terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
22.2 OLS as GMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
22.3 TSLS as GMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
22.4 Covariance matrix options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
22.5 A real example: the Consumption Based Asset Pricing Model . . . . . . . . . . . . . . . . 197
22.6 Caveats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
23 Model selection criteria 202
23.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
23.2 Information criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
24 Time series lters 204
24.1 Fractional dierencing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
24.2 The HodrickPrescott lter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
24.3 The Baxter and King lter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
24.4 The Butterworth lter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
25 Univariate time series models 208
25.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
25.2 ARIMA models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
25.3 Unit root tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
25.4 Cointegration tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
25.5 ARCH and GARCH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
26 Vector Autoregressions 222
26.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
26.2 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
26.3 Structural VARs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
27 Cointegration and Vector Error Correction Models 229
27.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
27.2 Vector Error Correction Models as representation of a cointegrated system . . . . . . . 230
27.3 Interpretation of the deterministic components . . . . . . . . . . . . . . . . . . . . . . . . 231
27.4 The Johansen cointegration tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
27.5 Identication of the cointegration vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
27.6 Over-identifying restrictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Contents vi
27.7 Numerical solution methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
28 Multivariate models 246
28.1 The system command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
28.2 Restriction and estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
28.3 System accessors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
29 Forecasting 251
29.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
29.2 Saving and inspecting tted values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
29.3 The fcast command . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
29.4 Univariate forecast evaluation statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
29.5 Forecasts based on VAR models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
29.6 Forecasting from simultaneous systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254
30 The Kalman Filter 255
30.1 Preamble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
30.2 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
30.3 Intended usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
30.4 Overview of syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
30.5 Dening the lter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
30.6 The kfilter function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
30.7 The ksmooth function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
30.8 The ksimul function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
30.9 Example 1: ARMA estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
30.10Example 2: local level model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
31 Numerical methods 267
31.1 BFGS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
31.2 NewtonRaphson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
31.3 Simulated Annealing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
31.4 Computing a Jacobian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
32 Discrete and censored dependent variables 276
32.1 Logit and probit models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
32.2 Ordered response models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
32.3 Multinomial logit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
32.4 Bivariate probit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
32.5 The Tobit model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
32.6 Interval regression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
32.7 Sample selection model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Contents vii
32.8 Count data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
32.9 Duration models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
33 Quantile regression 296
33.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
33.2 Basic syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
33.3 Condence intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
33.4 Multiple quantiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
33.5 Large datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
34 Nonparametric methods 300
34.1 Locally weighted regression (loess) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
34.2 The NadarayaWatson estimator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
III Technical details 305
35 Gretl and T
E
X 306
35.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
35.2 T
E
X-related menu items . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
35.3 Fine-tuning typeset output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
35.4 Installing and learning T
E
X . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
36 Gretl and R 312
36.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
36.2 Starting an interactive R session . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
36.3 Running an R script . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
36.4 Taking stu back and forth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
36.5 Interacting with R from the command line . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
36.6 Performance issues with R . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
36.7 Further use of the R library . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
37 Gretl and Ox 323
37.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
37.2 Ox support in gretl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
37.3 Illustration: replication of DPD model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325
38 Gretl and Octave 327
38.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
38.2 Octave support in gretl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
38.3 Illustration: spectral methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
39 Gretl and Stata 331
Contents viii
40 Gretl and Python 332
40.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
40.2 Python support in gretl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
40.3 Illustration: linear regression with multicollinearity . . . . . . . . . . . . . . . . . . . . . 332
41 Troubleshooting gretl 335
41.1 Bug reports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
41.2 Auxiliary programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
42 The command line interface 337
IV Appendices 338
A Data le details 339
A.1 Basic native format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
A.2 Traditional ESL format . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
A.3 Binary database details . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
B Data import via ODBC 342
B.1 ODBC support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
B.2 ODBC base concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342
B.3 Syntax . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
B.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
C Building gretl 348
C.1 Installing the prerequisites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
C.2 Getting the source: release or CVS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
C.3 Congure the source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
C.4 Build and install . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
D Numerical accuracy 353
E Related free software 354
F Listing of URLs 355
Bibliography 356
Chapter 1
Introduction
1.1 Features at a glance
Gretl is an econometrics package, including a shared library, a command-line client program and a
graphical user interface.
User-friendly Gretl oers an intuitive user interface; it is very easy to get up and running with
econometric analysis. Thanks to its association with the econometrics textbooks by Ramu
Ramanathan, Jerey Wooldridge, and James Stock and Mark Watson, the package oers many
practice data les and command scripts. These are well annotated and accessible. Two other
useful resources for gretl users are the available documentation and the gretl-users mailing
list.
Flexible You can choose your preferred point on the spectrum from interactive point-and-click to
complex scripting, and can easily combine these approaches.
Cross-platform Gretls home platform is Linux but it is also available for MS Windows and Mac
OS X, and should work on any unix-like system that has the appropriate basic libraries (see
Appendix C).
Open source The full source code for gretl is available to anyone who wants to critique it, patch it,
or extend it. See Appendix C.
Sophisticated Gretl oers a full range of least-squares based estimators, either for single equations
and for systems, including vector autoregressions and vector error correction models. Sev-
eral specic maximum likelihood estimators (e.g. probit, ARIMA, GARCH) are also provided
natively; more advanced estimation methods can be implemented by the user via generic
maximum likelihood or nonlinear GMM.
Extensible Users can enhance gretl by writing their own functions and procedures in gretls script-
ing language, which includes a wide range of matrix functions.
Accurate Gretl has been thoroughly tested on several benchmarks, among which the NIST refer-
ence datasets. See Appendix D.
Internet ready Gretl can fetch materials such databases, collections of textbook datales and add-
on packages over the internet.
International Gretl will produce its output in English, French, Italian, Spanish, Polish, Portuguese,
German, Basque, Turkish, Russian, Albanian or Greek depending on your computers native
language setting.
1.2 Acknowledgements
The gretl code base originally derived from the program ESL (Econometrics Software Library),
written by Professor Ramu Ramanathan of the University of California, San Diego. We are much in
debt to Professor Ramanathan for making this code available under the GNU General Public Licence
and for helping to steer gretls early development.
1
Chapter 1. Introduction 2
We are also grateful to the authors of several econometrics textbooks for permission to package for
gretl various datasets associated with their texts. This list currently includes William Greene, au-
thor of Econometric Analysis; Jerey Wooldridge (Introductory Econometrics: A Modern Approach);
James Stock and Mark Watson (Introduction to Econometrics); Damodar Gujarati (Basic Economet-
rics); Russell Davidson and James MacKinnon (Econometric Theory and Methods); and Marno Ver-
beek (A Guide to Modern Econometrics).
GARCH estimation in gretl is based on code deposited in the archive of the Journal of Applied
Econometrics by Professors Fiorentini, Calzolari and Panattoni, and the code to generate p-values
for DickeyFuller tests is due to James MacKinnon. In each case we are grateful to the authors for
permission to use their work.
With regard to the internationalization of gretl, thanks go to Ignacio Daz-Emparanza (Spanish),
Michel Robitaille and Florent Bresson (French), Cristian Rigamonti (Italian), Tadeusz Kufel and Pawel
Kufel (Polish), Markus Hahn and Sven Schreiber (German), Hlio Guilherme and Henrique Andrade
(Portuguese), Susan Orbe (Basque), Talha Yalta (Turkish) and Alexander Gedranovich (Russian).
Gretl has benetted greatly from the work of numerous developers of free, open-source software:
for specics please see Appendix C. Our thanks are due to Richard Stallman of the Free Software
Foundation, for his support of free software in general and for agreeing to adopt gretl as a GNU
program in particular.
Many users of gretl have submitted useful suggestions and bug reports. In this connection par-
ticular thanks are due to Ignacio Daz-Emparanza, Tadeusz Kufel, Pawel Kufel, Alan Isaac, Cri
Rigamonti, Sven Schreiber, Talha Yalta, Andreas Rosenblad, and Dirk Eddelbuettel, who maintains
the gretl package for Debian GNU/Linux.
1.3 Installing the programs
Linux
On the Linux
1
platform you have the choice of compiling the gretl code yourself or making use of a
pre-built package. Building gretl from the source is necessary if you want to access the development
version or customize gretl to your needs, but this takes quite a few skills; most users will want to
go for a pre-built package.
Some Linux distributions feature gretl as part of their standard oering: Debian, Ubuntu and Fe-
dora, for example. If this is the case, all you need to do is install gretl through your package
manager of choice. In addition the gretl webpage at http://gretl.sourceforge.net oers a
generic package in rpm format for modern Linux systems.
If you prefer to compile your own (or are using a unix system for which pre-built packages are not
available), instructions on building gretl can be found in Appendix C.
MS Windows
The MS Windows version comes as a self-extracting executable. Installation is just a matter of
downloading gretl_install.exe and running this program. You will be prompted for a location
to install the package.
Mac OS X
The Mac version comes as a gzipped disk image. Installation is a matter of downloading the image
le, opening it in the Finder, and dragging Gretl.app to the Applications folder. However, when
installing for the rst time two prerequisite packages must be put in place rst; details are given
on the gretl website.
1
In this manual we use Linux as shorthand to refer to the GNU/Linux operating system. What is said herein about
Linux mostly applies to other unix-type systems too, though some local modications may be needed.
Part I
Running the program
3
Chapter 2
Getting started
2.1 Lets run a regression
This introduction is mostly angled towards the graphical client program; please see Chapter 42
below and the Gretl Command Reference for details on the command-line program, gretlcli.
You can supply the name of a data le to open as an argument to gretl, but for the moment lets
not do that: just re up the program.
1
You should see a main window (which will hold information
on the data set but which is at rst blank) and various menus, some of them disabled at rst.
What can you do at this point? You can browse the supplied data les (or databases), open a data
le, create a new data le, read the help items, or open a command script. For now lets browse the
supplied data les. Under the File menu choose Open data, Sample le. A second notebook-type
window will open, presenting the sets of data les supplied with the package (see Figure 2.1). Select
the rst tab, Ramanathan. The numbering of the les in this section corresponds to the chapter
organization of Ramanathan (2002), which contains discussion of the analysis of these data. The
data will be useful for practice purposes even without the text.
Figure 2.1: Practice data les window
If you select a row in this window and click on Info this opens a window showing information on
the data set in question (for example, on the sources and denitions of the variables). If you nd
a le that is of interest, you may open it by clicking on Open, or just double-clicking on the le
name. For the moment lets open data3-6.
In gretl windows containing lists, double-clicking on a line launches a default action for the associated list
entry: e.g. displaying the values of a data series, opening a le.
1
For convenience we refer to the graphical client program simply as gretl in this manual. Note, however, that the
specic name of the program diers according to the computer platform. On Linux it is called gretl_x11 while on
MS Windows it is gretl.exe. On Linux systems a wrapper script named gretl is also installed see also the Gretl
Command Reference.
4
Chapter 2. Getting started 5
This le contains data pertaining to a classic econometric chestnut, the consumption function.
The data window should now display the name of the current data le, the overall data range and
sample range, and the names of the variables along with brief descriptive tagssee Figure 2.2.
Figure 2.2: Main window, with a practice data le open
OK, what can we do now? Hopefully the various menu options should be fairly self explanatory. For
now well dip into the Model menu; a brief tour of all the main window menus is given in Section 2.3
below.
Gretls Model menu oers numerous various econometric estimation routines. The simplest and
most standard is Ordinary Least Squares (OLS). Selecting OLS pops up a dialog box calling for a
model specicationsee Figure 2.3.
Figure 2.3: Model specication dialog
To select the dependent variable, highlight the variable you want in the list on the left and click
the arrow that points to the Dependent variable slot. If you check the Set as default box this
variable will be pre-selected as dependent when you next open the model dialog box. Shortcut:
double-clicking on a variable on the left selects it as dependent and also sets it as the default. To
select independent variables, highlight them on the left and click the green arrow (or right-click the
highlighted variable); to remove variables from the selected list, use the rad arrow. To select several
variable in the list box, drag the mouse over them; to select several non-contiguous variables, hold
down the Ctrl key and click on the variables you want. To run a regression with consumption as
the dependent variable and income as independent, click Ct into the Dependent slot and add Yt to
the Independent variables list.
2.2 Estimation output
Once youve specied a model, a window displaying the regression output will appear. The output
is reasonably comprehensive and in a standard format (Figure 2.4).
Figure 2.4: Model output window
The output window contains menus that allow you to inspect or graph the residuals and tted
values, and to run various diagnostic tests on the model.
For most models there is also an option to print the regression output in L
A
T
E
X format. See Chap-
ter 35 for details.
To import gretl output into a word processor, you may copy and paste from an output window,
using its Edit menu (or Copy button, in some contexts) to the target program. Many (not all) gretl
windows oer the option of copying in RTF (Microsofts Rich Text Format) or as L
A
T
E
X. If you are
pasting into a word processor, RTF may be a good option because the tabular formatting of the
output is preserved.
2
Alternatively, you can save the output to a (plain text) le then import the
le into the target program. When you nish a gretl session you are given the option of saving all
the output from the session to a single le.
Note that on the gnome desktop and under MS Windows, the File menu includes a command to
send the output directly to a printer.
When pasting or importing plain text gretl output into a word processor, select a monospaced or typewriter-
style font (e.g. Courier) to preserve the outputs tabular formatting. Select a small font (10-point Courier
should do) to prevent the output lines from being broken in the wrong place.
2
Note that when you copy as RTF under MS Windows, Windows will only allow you to paste the material into ap-
plications that understand RTF. Thus you will be able to paste into MS Word, but not into notepad. Note also that
there appears to be a bug in some versions of Windows, whereby the paste will not work properly unless the target
application (e.g. MS Word) is already running prior to copying the material in question.
2.3 The main window menus
Reading left to right along the main windows menu bar, we nd the File, Tools, Data, View, Add,
Sample, Variable, Model and Help menus.
File menu
Open data: Open a native gretl data le or import from other formats. See Chapter 4.
Append data: Add data to the current working data set, from a gretl data le, a comma-
separated values le or a spreadsheet le.
Save data: Save the currently open native gretl data le.
Save data as: Write out the current data set in native format, with the option of using
gzip data compression. See Chapter 4.
Export data: Write out the current data set in Comma Separated Values (CSV) format, or
the formats of GNU R or GNU Octave. See Chapter 4 and also Appendix E.
Send to: Send the current data set as an e-mail attachment.
New data set: Allows you to create a blank data set, ready for typing in values or for
importing series from a database. See below for more on databases.
Clear data set: Clear the current data set out of memory. Generally you dont have to do
this (since opening a new data le automatically clears the old one) but sometimes its
useful.
Script les: A script is a le containing a sequence of gretl commands. This item
contains entries that let you open a script you have created previously (User le), open
a sample script, or open an editor window in which you can create a new script.
Session les: A session le contains a snapshot of a previous gretl session, including
the data set used and any models or graphs that you saved. Under this item you can
open a saved session or save the current session.
Databases: Allows you to browse various large databases, either on your own computer
or, if you are connected to the internet, on the gretl database server. See Section 4.2 for
details.
Function les: Handles function packages (see Section 13.5), which allow you to access
functions written by other users and share the ones written by you.
Exit: Quit the program. Youll be prompted to save any unsaved work.
Tools menu
Statistical tables: Look up critical values for commonly used distributions (normal or
Gaussian, t, chi-square, F and DurbinWatson).
P-value nder: Look up p-values from the Gaussian, t, chi-square, F, gamma, binomial or
Poisson distributions. See also the pvalue command in the Gretl Command Reference.
Distribution graphs: Produce graphs of various probability distributions. In the resulting
graph window, the pop-up menu includes an item Add another curve, which enables
you to superimpose a further plot (for example, you can draw the t distribution with
various dierent degrees of freedom).
Test statistic calculator: Calculate test statistics and p-values for a range of common hy-
pothesis tests (population mean, variance and proportion; dierence of means, variances
and proportions).
Nonparametric tests: Calculate test statistics for various nonparametric tests (Sign test,
Wilcoxon rank sum test, Wilcoxon signed rank test, Runs test).
Seed for random numbers: Set the seed for the random number generator (by default
this is set based on the system time when the program is started).
Command log: Open a window containing a record of the commands executed so far.
Gretl console: Open a console window into which you can type commands as you would
using the command-line program, gretlcli (as opposed to using point-and-click).
Start Gnu R: Start R (if it is installed on your system), and load a copy of the data set
currently open in gretl. See Appendix E.
Sort variables: Rearrange the listing of variables in the main window, either by ID number
or alphabetically by name.
NIST test suite: Check the numerical accuracy of gretl against the reference results for
linear regression made available by the (US) National Institute of Standards and Technol-
ogy.
Preferences: Set the paths to various les gretl needs to access. Choose the font in which
gretl displays text output. Activate or suppress gretls messaging about the availability
of program updates, and so on. See the Gretl Command Reference for further details.
Data menu
Select all: Several menu items act upon those variables that are currently selected in the
main window. This item lets you select all the variables.
Display values: Pops up a window with a simple (not editable) printout of the values of
the selected variable or variables.
Edit values: Opens a spreadsheet window where you can edit the values of the selected
variables.
Add observations: Gives a dialog box in which you can choose a number of observations
to add at the end of the current dataset; for use with forecasting.
Remove extra observations: Active only if extra observations have been added automati-
cally in the process of forecasting; deletes these extra observations.
Read info, Edit info: Read info just displays the summary information for the current
data le; Edit info allows you to make changes to it (if you have permission to do so).
Print description: Opens a window containing a full account of the current dataset, in-
cluding the summary information and any specic information on each of the variables.
Add case markers: Prompts for the name of a text le containing case markers (short
strings identifying the individual observations) and adds this information to the data set.
See Chapter 4.
Remove case markers: Active only if the dataset has case markers identifying the obser-
vations; removes these case markers.
Dataset structure: invokes a series of dialog boxes which allow you to change the struc-
tural interpretation of the current dataset. For example, if data were read in as a cross
section you can get the program to interpret them as time series or as a panel. See also
section 4.4.
Compact data: For time-series data of higher than annual frequency, gives you the option
of compacting the data to a lower frequency, using one of four compaction methods
(average, sum, start of period or end of period).
Expand data: For time-series data, gives you the option of expanding the data to a higher
frequency.
Transpose data: Turn each observation into a variable and vice versa (or in other words,
each row of the data matrix becomes a column in the modied data matrix); can be useful
with imported data that have been read in sideways.
View menu
Icon view: Opens a window showing the content of the current session as a set of icons;
see section 3.4.
Graph specied vars: Gives a choice between a time series plot, a regular XY scatter
plot, an XY plot using impulses (vertical bars), an XY plot with factor separation (i.e.
with the points colored dierently depending to the value of a given dummy variable),
boxplots, and a 3-D graph. Serves up a dialog box where you specify the variables to
graph. See Chapter 6 for details.
Multiple graphs: Allows you to compose a set of up to six small graphs, either pairwise
scatter-plots or time-series graphs. These are displayed together in a single window.
Summary statistics: Shows a full set of descriptive statistics for the variables selected in
the main window.
Correlation matrix: Shows the pairwise correlation coecients for the selected variables.
Cross Tabulation: Shows a cross-tabulation of the selected variables. This works only if
at least two variables in the data set have been marked as discrete (see Chapter 11).
Principal components: Produces a Principal Components Analysis for the selected vari-
ables.
Mahalanobis distances: Computes the Mahalanobis distance of each observation from
the centroid of the selected set of variables.
Cross-correlogram: Computes and graphs the cross-correlogram for two selected vari-
ables.
Add menu Oers various standard transformations of variables (logs, lags, squares, etc.) that
you may wish to add to the data set. Also gives the option of adding random variables, and
(for time-series data) adding seasonal dummy variables (e.g. quarterly dummy variables for
quarterly data).
Sample menu
Set range: Select a dierent starting and/or ending point for the current sample, within
the range of data available.
Restore full range: self-explanatory.
Dene, based on dummy: Given a dummy (indicator) variable with values 0 or 1, this
drops from the current sample all observations for which the dummy variable has value
0.
Restrict, based on criterion: Similar to the item above, except that you dont need a pre-
dened variable: you supply a Boolean expression (e.g. sqft > 1400) and the sample is
restricted to observations satisfying that condition. See the entry for genr in the Gretl
Command Reference for details on the Boolean operators that can be used.
Random sub-sample: Draw a random sample from the full dataset.
Drop all obs with missing values: Drop from the current sample all observations for
which at least one variable has a missing value (see Section 4.6).
Count missing values: Give a report on observations where data values are missing. May
be useful in examining a panel data set, where its quite common to encounter missing
values.
Set missing value code: Set a numerical value that will be interpreted as missing or not
available. This is intended for use with imported data, when gretl has not recognized
the missing-value code used.
Variable menu Most items under here operate on a single variable at a time. The active
variable is set by highlighting it (clicking on its row) in the main data window. Most options
will be self-explanatory. Note that you can rename a variable and can edit its descriptive label
under Edit attributes. You can also Dene a new variable via a formula (e.g. involving
some function of one or more existing variables). For the syntax of such formulae, look at the
online help for Generate variable syntax or see the genr command in the Gretl Command
Reference. One simple example:
foo = x1 * x2
will create a new variable foo as the product of the existing variables x1 and x2. In these
formulae, variables must be referenced by name, not number.
Model menu For details on the various estimators oered under this menu please consult the
Gretl Command Reference. Also see Chapter 20 regarding the estimation of nonlinear models.
Help menu Please use this as needed! It gives details on the syntax required in various dialog
entries.
2.4 Keyboard shortcuts
When working in the main gretl window, some common operations may be performed using the
keyboard, as shown in the table below.
Return Opens a window displaying the values of the currently selected variables: it is
the same as selecting Data, Display Values.
Delete Pressing this key has the eect of deleting the selected variables. A conrma-
tion is required, to prevent accidental deletions.
e Has the same eect as selecting Edit attributes from the Variable menu.
F2 Same as e. Included for compatibility with other programs.
g Has the same eect as selecting Dene new variable from the Variable
menu (which maps onto the genr command).
h Opens a help window for gretl commands.
F1 Same as h. Included for compatibility with other programs.
r Refreshes the variable list in the main window.
t Graphs the selected variable; a line graph is used for time-series datasets,
whereas a distribution plot is used for cross-sectional data.
2.5 The gretl toolbar
At the bottom left of the main window sits the toolbar.
The icons have the following functions, reading from left to right:
1. Launch a calculator program. A convenience function in case you want quick access to a
calculator when youre working in gretl. The default program is calc.exe under MS Win-
dows, or xcalc under the X window system. You can change the program under the Tools,
Preferences, General menu, Programs tab.
2. Start a new script. Opens an editor window in which you can type a series of commands to be
sent to the program as a batch.
3. Open the gretl console. A shortcut to the Gretl console menu item (Section 2.3 above).
4. Open the session icon window.
5. Open a window displaying available gretl function packages.
6. Open this manual in PDF format.
7. Open the help item for script commands syntax (i.e. a listing with details of all available
commands).
8. Open the dialog box for dening a graph.
9. Open the dialog box for estimating a model using ordinary least squares.
10. Open a window listing the sample datasets supplied with gretl, and any other data le collec-
tions that have been installed.
Chapter 3
Modes of working
3.1 Command scripts
As you execute commands in gretl, using the GUI and lling in dialog entries, those commands are
recorded in the form of a script or batch le. Such scripts can be edited and re-run, using either
gretl or the command-line client, gretlcli.
To view the current state of the script at any point in a gretl session, choose Command log under
the Tools menu. This log le is called session.inp and it is overwritten whenever you start a new
session. To preserve it, save the script under a dierent name. Script les will be found most easily,
using the GUI le selector, if you name them with the extension .inp.
To open a script you have written independently, use the File, Script les menu item; to create a
script from scratch use the File, Script les, New script item or the new script toolbar button.
In either case a script window will open (see Figure 3.1).
Figure 3.1: Script window, editing a command le
The toolbar at the top of the script window oers the following functions (left to right): (1) Save
the le; (2) Save the le under a specied name; (3) Print the le (this option is not available on all
platforms); (4) Execute the commands in the le; (5) Copy selected text; (6) Paste the selected text;
(7) Find and replace text; (8) Undo the last Paste or Replace action; (9) Help (if you place the cursor
in a command word and press the question mark you will get help on that command); (10) Close
the window.
When you execute the script, by clicking on the Execute icon or by pressing Ctrl-r, all output is
directed to a single window, where it can be edited, saved or copied to the clipboard. To learn
more about the possibilities of scripting, take a look at the gretl Help item Command reference,
12
Chapter 3. Modes of working 13
or start up the command-line program gretlcli and consult its help, or consult the Gretl Command
Reference.
If you run the script when part of it is highlighted, gretl will only run that portion. Moreover, if you
want to run just the current line, you can do so by pressing Ctrl-Enter.
1
Clicking the right mouse button in the script editor window produces a pop-up menu. This gives
you the option of executing either the line on which the cursor is located, or the selected region of
the script if theres a selection in place. If the script is editable, this menu also gives the option of
adding or removing comment markers from the start of the line or lines.
The gretl package includes over 70 practice scripts. Most of these relate to Ramanathan (2002),
but they may also be used as a free-standing introduction to scripting in gretl and to various points
of econometric theory. You can explore the practice les under File, Script les, Practice le There
you will nd a listing of the les along with a brief description of the points they illustrate and the
data they employ. Open any le and run it to see the output. Note that long commands in a script
can be broken over two or more lines, using backslash as a continuation character.
You can, if you wish, use the GUI controls and the scripting approach in tandem, exploiting each
method where it oers greater convenience. Here are two suggestions.
Open a data le in the GUI. Explore the datagenerate graphs, run regressions, perform tests.
Then open the Command log, edit out any redundant commands, and save it under a specic
name. Run the script to generate a single le containing a concise record of your work.
Start by establishing a new script le. Type in any commands that may be required to set
up transformations of the data (see the genr command in the Gretl Command Reference).
Typically this sort of thing can be accomplished more eciently via commands assembled
with forethought rather than point-and-click. Then save and run the script: the GUI data
window will be updated accordingly. Now you can carry out further exploration of the data
via the GUI. To revisit the data at a later point, open and rerun the preparatory script rst.
Scripts and data les
One common way of doing econometric research with gretl is as follows: compose a script; execute
the script; inspect the output; modify the script; run it againwith the last three steps repeated as
many times as necessary. In this context, note that when you open a data le this clears out most
of gretls internal state. Its therefore probably a good idea to have your script start with an open
command: the data le will be re-opened each time, and you can be condent youre getting fresh
results.
One further point should be noted. When you go to open a new data le via the graphical interface,
you are always prompted: opening a new data le will lose any unsaved work, do you really want
to do this? When you execute a script that opens a data le, however, you are not prompted. The
assumption is that in this case youre not going to lose any work, because the work is embodied
in the script itself (and it would be annoying to be prompted at each iteration of the work cycle
described above).
This means you should be careful if youve done work using the graphical interface and then decide
to run a script: the current data le will be replaced without any questions asked, and its your
responsibility to save any changes to your data rst.
1
This feature is not unique to gretl; other econometric packages oer the same facility. However, experience shows
that while this can be remarkably useful, it can also lead to writing dinosaur scripts that are never meant to be executed
all at once, but rather used as a chaotic repository to cherry-pick snippets from. Since gretl allows you to have several
script windows open at the same time, you may want to keep your scripts tidy and reasonably small.
3.2 Saving script objects
When you estimate a model using point-and-click, the model results are displayed in a separate
window, oering menus which let you perform tests, draw graphs, save data from the model, and
so on. Ordinarily, when you estimate a model using a script you just get a non-interactive printout
of the results. You can, however, arrange for models estimated in a script to be captured, so that
you can examine them interactively when the script is nished. Here is an example of the syntax
for achieving this eect:
Model1 <- ols Ct 0 Yt
That is, you type a name for the model to be saved under, then a back-pointing assignment arrow,
then the model command. The assignment arrow is composed of the less-than sign followed by a
dash; it must be separated by spaces from both the preceding name and the following command.
The name for a saved object may include spaces, but in that case it must be wrapped in double
quotes:
"Model 1" <- ols Ct 0 Yt
Models saved in this way will appear as icons in the gretl icon view window (see Section 3.4) after
the script is executed. In addition, you can arrange to have a named model displayed (in its own
window) automatically as follows:
Model1.show
Again, if the name contains spaces it must be quoted:
"Model 1".show
The same facility can be used for graphs. For example the following will create a plot of Ct against
Yt, save it under the name CrossPlot (it will appear under this name in the icon view window),
and have it displayed:
CrossPlot <- gnuplot Ct Yt
CrossPlot.show
You can also save the output from selected commands as named pieces of text (again, these will
appear in the session icon window, from where you can open them later). For example this com-
mand sends the output from an augmented DickeyFuller test to a text object named ADF1 and
displays it in a window:
ADF1 <- adf 2 x1
ADF1.show
Objects saved in this way (whether models, graphs or pieces of text output) can be destroyed using
the command .free appended to the name of the object, as in ADF1.free.
3.3 The gretl console
A further option is available for your computing convenience. Under gretls Tools menu you will
nd the item Gretl console (there is also an open gretl console button on the toolbar in the
main window). This opens up a window in which you can type commands and execute them one
by one (by pressing the Enter key) interactively. This is essentially the same as gretlclis mode of
operation, except that the GUI is updated based on commands executed from the console, enabling
you to work back and forth as you wish.
In the console, you have command history; that is, you can use the up and down arrow keys to
navigate the list of command you have entered to date. You can retrieve, edit and then re-enter a
previous command.
In console mode, you can create, display and free objects (models, graphs or text) aa described
above for script mode.
3.4 The Session concept
Gretl oers the idea of a session as a way of keeping track of your work and revisiting it later.
The basic idea is to provide an iconic space containing various objects pertaining to your current
working session (see Figure 3.2). You can add objects (represented by icons) to this space as you
go along. If you save the session, these added objects should be available again if you re-open the
session later.
Figure 3.2: Icon view: one model and one graph have been added to the default icons
If you start gretl and open a data set, then select Icon view from the View menu, you should see
the basic default set of icons: these give you quick access to information on the data set (if any),
correlation matrix (Correlations) and descriptive summary statistics (Summary). All of these
are activated by double-clicking the relevant icon. The Data set icon is a little more complex:
double-clicking opens up the data in the built-in spreadsheet, but you can also right-click on the
icon for a menu of other actions.
To add a model to the Icon view, rst estimate it using the Model menu. Then pull down the File
menu in the model window and select Save to session as icon. . . or Save as icon and close.
Simply hitting the S key over the model window is a shortcut to the latter action.
To add a graph, rst create it (under the View menu, Graph specied vars, or via one of gretls
other graph-generating commands). Click on the graph window to bring up the graph menu, and
select Save to session as icon.
Once a model or graph is added its icon will appear in the Icon view window. Double-clicking on the
icon redisplays the object, while right-clicking brings up a menu which lets you display or delete
the object. This popup menu also gives you the option of editing graphs.
The model table
In econometric research it is common to estimate several models with a common dependent
variablethe models diering in respect of which independent variables are included, or per-
haps in respect of the estimator used. In this situation it is convenient to present the regression
results in the form of a table, where each column contains the results (coecient estimates and
standard errors) for a given model, and each row contains the estimates for a given variable across
the models.
In the Icon view window gretl provides a means of constructing such a table (and copying it in plain
text, L
A
T
E
X or Rich Text Format). The procedure is outlined below. (The model table can also be built
non-interactively, in script mode. For details, see the entry for modeltab in the Gretl Command
Reference.)
1. Estimate a model which you wish to include in the table, and in the model display window,
under the File menu, select Save to session as icon or Save as icon and close.
2. Repeat step 1 for the other models to be included in the table (up to a total of six models).
3. When you are done estimating the models, open the icon view of your gretl session, by se-
lecting Icon view under the View menu in the main gretl window, or by clicking the session
icon view icon on the gretl toolbar.
4. In the Icon view, there is an icon labeled Model table. Decide which model you wish to
appear in the left-most column of the model table and add it to the table, either by dragging
its icon onto the Model table icon, or by right-clicking on the model icon and selecting Add
to model table from the pop-up menu.
5. Repeat step 4 for the other models you wish to include in the table. The second model selected
will appear in the second column from the left, and so on.
6. When you are nished composing the model table, display it by double-clicking on its icon.
Under the Edit menu in the window which appears, you have the option of copying the table
to the clipboard in various formats.
7. If the ordering of the models in the table is not what you wanted, right-click on the model
table icon and select Clear table. Then go back to step 4 above and try again.
A simple instance of gretls model table is shown in Figure 3.3.
Figure 3.3: Example of model table
The graph page
The graph page icon in the session window oers a means of putting together several graphs
for printing on a single page. This facility will work only if you have the L
A
T
E
X typesetting system
installed, and are able to generate and view either PDF or PostScript output. The output format
is controlled by your choice of program for compiling T
E
X les, which can be found under the
Programs tab in the Preferences dialog box (under the Tools menu in the main window). Usually
this should be pdatex for PDF output or latex for PostScript. In the latter case you must have a
working set-up for handling PostScript, which will usually include dvips, ghostscript and a viewer
such as gv, ggv or kghostview.
In the Icon view window, you can drag up to eight graphs onto the graph page icon. When you
double-click on the icon (or right-click and select Display), a page containing the selected graphs
(in PDF or EPS format) will be composed and opened in your viewer. From there you should be able
to print the page.
To clear the graph page, right-click on its icon and select Clear.
As with the model table, it is also possible to manipulate the graph page via commands in script or
console modesee the entry for the graphpg command in the Gretl Command Reference.
Saving and re-opening sessions
If you create models or graphs that you think you may wish to re-examine later, then before quitting
gretl select Session les, Save session from the File menu and give a name under which to save
the session. To re-open the session later, either
Start gretl then re-open the session le by going to the File, Session les, Open session, or
From the command line, type gretl -r sessionle, where sessionle is the name under which
the session was saved, or
Drag the icon representing a session le onto gretl.
Chapter 4
Data les
4.1 Data le formats
Gretl has its own native format for data les. Most users will probably not want to read or write
such les outside of gretl itself, but occasionally this may be useful and details on the le formats
are given in Appendix A. The program can also import data from a variety of other formats. In the
GUI program this can be done via the File, Open Data, Import menu; in script mode, simply use
the open command. The supported import formats are as follows.
Plain text les (comma-separated or CSV being the most common type). For details on what
gretl expects of such les, see Section 4.3.
Spreadsheets: MS Excel, Gnumeric and Open Document (ODS). The requirements for such les
are given in Section 4.3.
Stata data les (.dta).
SPSS data les (.sav).
SAS xport les (.xpt).
Eviews workles (.wf1).
1
JMulTi data les.
When you import data from a plain text format, gretl opens a diagnostic window, reporting on its
progress in reading the data. If you encounter a problem with ill-formatted data, the messages in
this window should give you a handle on xing the problem.
Note that gretl has a facility for writing out data in the native formats of GNU R, Octave, JMulTi and
PcGive (see Appendix E). In the GUI client this option is found under the File, Export data menu;
in the command-line client use the store command with the appropriate option ag.
4.2 Databases
For working with large amounts of data gretl is supplied with a database-handling routine. A
database, as opposed to a data le, is not read directly into the programs workspace. A database
can contain series of mixed frequencies and sample ranges. You open the database and select
series to import into the working dataset. You can then save those series in a native format data
le if you wish. Databases can be accessed via the menu item File, Databases.
For details on the format of gretl databases, see Appendix A.
Online access to databases
Several gretl databases are available from Wake Forest University. Your computer must be con-
nected to the internet for this option to work. Please see the description of the data command
under the Help menu.
1
See http://www.ecn.wfu.edu/eviews_format/.
18
Chapter 4. Data les 19
Visit the gretl data page for details and updates on available data.
Foreign database formats
Thanks to Thomas Doan of Estima, who made available the specication of the database format
used by RATS 4 (Regression Analysis of Time Series), gretl can handle such databasesor at least,
a subset of same, namely time-series databases containing monthly and quarterly series.
Gretl can also import data from PcGive databases. These take the form of a pair of les, one
containing the actual data (with sux .bn7) and one containing supplementary information (.in7).
In addition, gretl oers ODBC connectivity. Be warned: this feature is meant for somewhat ad-
vanced users; there is currently no graphical interface. Interested readers will nd more info in
appendix B.
4.3 Creating a dataset from scratch
There are several ways of doing this:
1. Find, or create using a text editor, a plain text data le and open it via Import.
2. Use your favorite spreadsheet to establish the data le, save it in comma-separated format if
necessary (this may not be necessary if the spreadsheet format is MS Excel, Gnumeric or Open
Document), then use one of the Import options.
3. Use gretls built-in spreadsheet.
4. Select data series from a suitable database.
5. Use your favorite text editor or other software tools to a create data le in gretl format inde-
pendently.
Here are a few comments and details on these methods.
Common points on imported data
Options (1) and (2) involve using gretls import mechanism. For the program to read such data
successfully, certain general conditions must be satised:
The rst row must contain valid variable names. A valid variable name is of 31 characters
maximum; starts with a letter; and contains nothing but letters, numbers and the underscore
character, _. (Longer variable names will be truncated to 31 characters.) Qualications to
the above: First, in the case of an plain text import, if the le contains no row with variable
names the program will automatically add names, v1, v2 and so on. Second, by the rst
row is meant the rst relevant row. In the case of plain text imports, blank rows and rows
beginning with a hash mark, #, are ignored. In the case of Excel and Gnumeric imports, you
are presented with a dialog box where you can select an oset into the spreadsheet, so that
gretl will ignore a specied number of rows and/or columns.
Data values: these should constitute a rectangular block, with one variable per column (and
one observation per row). The number of variables (data columns) must match the number
of variable names given. See also section 4.6. Numeric data are expected, but in the case of
importing from plain text, the program oers limited handling of character (string) data: if
a given column contains character data only, consecutive numeric codes are substituted for
the strings, and once the import is complete a table is printed showing the correspondence
between the strings and the codes.
Dates (or observation labels): Optionally, the rst column may contain strings such as dates,
or labels for cross-sectional observations. Such strings have a maximum of 8 characters (as
with variable names, longer strings will be truncated). A column of this sort should be headed
with the string obs or date, or the rst row entry may be left blank.
For dates to be recognized as such, the date strings must adhere to one or other of a set of
specic formats, as follows. For annual data: 4-digit years. For quarterly data: a 4-digit year,
followed by a separator (either a period, a colon, or the letter Q), followed by a 1-digit quarter.
Examples: 1997.1, 2002:3, 1947Q1. For monthly data: a 4-digit year, followed by a period or
a colon, followed by a two-digit month. Examples: 1997.01, 2002:10.
Plain text (CSV) les can use comma, space, tab or semicolon as the column separator. When you
use the Import menu item you given the option of specifying the separator, though in most cases
it should be detected automatically.
If you use a spreadsheet to prepare your data you are able to carry out various transformations of
the raw data with ease (adding things up, taking percentages or whatever): note, however, that
you can also do this sort of thing easilyperhaps more easilywithin gretl, by using the tools
under the Add menu.
Appending imported data
You may wish to establish a dataset piece by piece, by incremental importation of data from other
sources. This is supported via the File, Append data menu items: gretl will check the new data for
conformability with the existing dataset and, if everything seems OK, will merge the data. You can
add new variables in this way, provided the data frequency matches that of the existing dataset. Or
you can append new observations for data series that are already present; in this case the variable
names must match up correctly. Note that by default (that is, if you choose Open data rather
than Append data), opening a new data le closes the current one.
Using the built-in spreadsheet
Under the File, New data set menu you can choose the sort of dataset you want to establish (e.g.
quarterly time series, cross-sectional). You will then be prompted for starting and ending dates (or
observation numbers) and the name of the rst variable to add to the dataset. After supplying this
information you will be faced with a simple spreadsheet into which you can type data values. In
the spreadsheet window, clicking the right mouse button will invoke a popup menu which enables
you to add a new variable (column), to add an observation (append a row at the foot of the sheet),
or to insert an observation at the selected point (move the data down and insert a blank row.)
Once you have entered data into the spreadsheet you import these into gretls workspace using the
spreadsheets Apply changes button.
Please note that gretls spreadsheet is quite basic and has no support for functions or formulas.
Data transformations are done via the Add or Variable menus in the main window.
Selecting from a database
Another alternative is to establish your dataset by selecting variables from a database.
Begin with the File, Databases menu item. This has four forks: Gretl native, RATS 4, PcGive
and On database server. You should be able to nd the le fedstl.bin in the le selector that
opens if you choose the Gretl native option since this le, which contains a large collection of US
macroeconomic time series, is supplied with the distribution.
You wont nd anything under RATS 4 unless you have purchased RATS data.
2
If you do possess
RATS data you should go into the Tools, Preferences, General dialog, select the Databases tab,
2
See www.estima.com
and ll in the correct path to your RATS les.
If your computer is connected to the internet you should nd several databases (at Wake Forest
University) under On database server. You can browse these remotely; you also have the option
of installing them onto your own computer. The initial remote databases window has an item
showing, for each le, whether it is already installed locally (and if so, if the local version is up to
date with the version at Wake Forest).
Assuming you have managed to open a database you can import selected series into gretls workspace
by using the Series, Import menu item in the database window, or via the popup menu that ap-
pears if you click the right mouse button, or by dragging the series into the programs main window.
Creating a gretl data le independently
It is possible to create a data le in one or other of gretls own formats using a text editor or
software tools such as awk, sed or perl. This may be a good choice if you have large amounts
of data already in machine readable form. You will, of course, need to study these data formats
(XML-based or traditional) as described in Appendix A.
4.4 Structuring a dataset
Once your data are read by gretl, it may be necessary to supply some information on the nature of
the data. We distinguish between three kinds of datasets:
1. Cross section
2. Time series
3. Panel data
The primary tool for doing this is the Data, Dataset structure menu entry in the graphical inter-
face, or the setobs command for scripts and the command-line interface.
Cross sectional data
By a cross section we mean observations on a set of units (which may be rms, countries, indi-
viduals, or whatever) at a common point in time. This is the default interpretation for a data le:
if there is insucient information to interpret data as time-series or panel data, they are automat-
ically interpreted as a cross section. In the unlikely event that cross-sectional data are wrongly
interpreted as time series, you can correct this by selecting the Data, Dataset structure menu
item. Click the cross-sectional radio button in the dialog box that appears, then click Forward.
Click OK to conrm your selection.
Time series data
When you import data from a spreadsheet or plain text le, gretl will make fairly strenuous eorts
to glean time-series information from the rst column of the data, if it looks at all plausible that
such information may be present. If time-series structure is present but not recognized, again you
can use the Data, Dataset structure menu item. Select Time series and click Forward; select the
appropriate data frequency and click Forward again; then select or enter the starting observation
and click Forward once more. Finally, click OK to conrm the time-series interpretation if it is
correct (or click Back to make adjustments if need be).
Besides the basic business of getting a data set interpreted as time series, further issues may arise
relating to the frequency of time-series data. In a gretl time-series data set, all the series must
have the same frequency. Suppose you wish to make a combined dataset using series that, in their
original state, are not all of the same frequency. For example, some series are monthly and some
are quarterly.
Your rst step is to formulate a strategy: Do you want to end up with a quarterly or a monthly data
set? A basic point to note here is that compacting data from a higher frequency (e.g. monthly) to
a lower frequency (e.g. quarterly) is usually unproblematic. You lose information in doing so, but
in general it is perfectly legitimate to take (say) the average of three monthly observations to create
a quarterly observation. On the other hand, expanding data from a lower to a higher frequency is
not, in general, a valid operation.
In most cases, then, the best strategy is to start by creating a data set of the lower frequency, and
then to compact the higher frequency data to match. When you import higher-frequency data from
a database into the current data set, you are given a choice of compaction method (average, sum,
start of period, or end of period). In most instances average is likely to be appropriate.
You can also import lower-frequency data into a high-frequency data set, but this is generally not
recommended. What gretl does in this case is simply replicate the values of the lower-frequency
series as many times as required. For example, suppose we have a quarterly series with the value
35.5 in 1990:1, the rst quarter of 1990. On expansion to monthly, the value 35.5 will be assigned
to the observations for January, February and March of 1990. The expanded variable is therefore
useless for ne-grained time-series analysis, outside of the special case where you know that the
variable in question does in fact remain constant over the sub-periods.
When the current data frequency is appropriate, gretl oers both Compact data and Expand
data options under the Data menu. These options operate on the whole data set, compacting or
exanding all series. They should be considered expert options and should be used with caution.
Panel data
Panel data are inherently three dimensional the dimensions being variable, cross-sectional unit,
and time-period. For example, a particular number in a panel data set might be identied as the
observation on capital stock for General Motors in 1980. (A note on terminology: we use the
terms cross-sectional unit, unit and group interchangeably below to refer to the entities that
compose the cross-sectional dimension of the panel. These might, for instance, be rms, countries
or persons.)
For representation in a textual computer le (and also for gretls internal calculations) the three
dimensions must somehow be attened into two. This attening involves taking layers of the
data that would naturally stack in a third dimension, and stacking them in the vertical dimension.
gretl always expects data to be arranged by observation, that is, such that each row represents an
observation (and each variable occupies one and only one column). In this context the attening of
a panel data set can be done in either of two ways:
Stacked time series: the successive vertical blocks each comprise a time series for a given
unit.
Stacked cross sections: the successive vertical blocks each comprise a cross-section for a
given period.
You may input data in whichever arrangement is more convenient. Internally, however, gretl always
stores panel data in the form of stacked time series.
4.5 Panel data specics
When you import panel data into gretl from a spreadsheet or comma separated format, the panel
nature of the data will not be recognized automatically (most likely the data will be treated as
undated). A panel interpretation can be imposed on the data using the graphical interface or via
the setobs command.
In the graphical interface, use the menu item Data, Dataset structure. In the rst dialog box
that appears, select Panel. In the next dialog you have a three-way choice. The rst two options,
Stacked time series and Stacked cross sections are applicable if the data set is already organized
in one of these two ways. If you select either of these options, the next step is to specify the number
of cross-sectional units in the data set. The third option, Use index variables, is applicable if the
data set contains two variables that index the units and the time periods respectively; the next step
is then to select those variables. For example, a data le might contain a country code variable and
a variable representing the year of the observation. In that case gretl can reconstruct the panel
structure of the data regardless of how the observation rows are organized.
The setobs command has options that parallel those in the graphical interface. If suitable index
variables are available you can do, for example
setobs unitvar timevar --panel-vars
where unitvar is a variable that indexes the units and timevar is a variable indexing the periods.
Alternatively you can use the form setobs freq 1:1 structure, where freq is replaced by the block
size of the data (that is, the number of periods in the case of stacked time series, or the number
of units in the case of stacked cross-sections) and structure is either --stacked-time-series or
--stacked-cross-section. Two examples are given below: the rst is suitable for a panel in
the form of stacked time series with observations from 20 periods; the second for stacked cross
sections with 5 units.
setobs 20 1:1 --stacked-time-series
setobs 5 1:1 --stacked-cross-section
Panel data arranged by variable
Publicly available panel data sometimes come arranged by variable. Suppose we have data on two
variables, x1 and x2, for each of 50 states in each of 5 years (giving a total of 250 observations
per variable). One textual representation of such a data set would start with a block for x1, with
50 rows corresponding to the states and 5 columns corresponding to the years. This would be
followed, vertically, by a block with the same structure for variable x2. A fragment of such a data
le is shown below, with quinquennial observations 19651985. Imagine the table continued for
48 more states, followed by another 50 rows for variable x2.
x1
1965 1970 1975 1980 1985
AR 100.0 110.5 118.7 131.2 160.4
AZ 100.0 104.3 113.8 120.9 140.6
If a datale with this sort of structure is read into gretl,
3
the program will interpret the columns as
distinct variables, so the data will not be usable as is. But there is a mechanism for correcting the
situation, namely the stack function within the genr command.
Consider the rst data column in the fragment above: the rst 50 rows of this column constitute a
cross-section for the variable x1 in the year 1965. If we could create a new variable by stacking the
rst 50 entries in the second column underneath the rst 50 entries in the rst, we would be on the
way to making a data set by observation (in the rst of the two forms mentioned above, stacked
3
Note that you will have to modify such a datale slightly before it can be read at all. The line containing the variable
name (in this example x1) will have to be removed, and so will the initial row containing the years, otherwise they will be
taken as numerical data.
cross-sections). That is, wed have a column comprising a cross-section for x1 in 1965, followed by
a cross-section for the same variable in 1970.
The following gretl script illustrates how we can accomplish the stacking, for both x1 and x2. We
assume that the original data le is called panel.txt, and that in this le the columns are headed
with variable names p1, p2, . . . , p5. (The columns are not really variables, but in the rst instance
we pretend that they are.)
open panel.txt
genr x1 = stack(p1..p5) --length=50
genr x2 = stack(p1..p5) --offset=50 --length=50
setobs 50 1:1 --stacked-cross-section
store panel.gdt x1 x2
The second line illustrates the syntax of the stack function. The double dots within the parenthe-
ses indicate a range of variables to be stacked: here we want to stack all 5 columns (for all 5 years).
4
The full data set contains 100 rows; in the stacking of variable x1 we wish to read only the rst 50
rows from each column: we achieve this by adding --length=50. Note that if you want to stack a
non-contiguous set of columns you can give a comma-separated list of variable names, as in
genr x = stack(p1,p3,p5)
or you can provide within the parentheses the name of a previously created list (see chapter 14).
On line 3 we do the stacking for variable x2. Again we want a length of 50 for the components of
the stacked series, but this time we want gretl to start reading from the 50th row of the original
data, and we specify --offset=50. Line 4 imposes a panel interpretation on the data; nally, we
save the data in gretl format, with the panel interpretation, discarding the original variables p1
through p5.
The illustrative script above is appropriate when the number of variable to be processed is small.
When then are many variables in the data set it will be more ecient to use a command loop to
accomplish the stacking, as shown in the following script. The setup is presumed to be the same
as in the previous section (50 units, 5 periods), but with 20 variables rather than 2.
open panel.txt
loop i=1..20
genr k = ($i - 1) * 50
genr x$i = stack(p1..p5) --offset=k --length=50
endloop
setobs 50 1.01 --stacked-cross-section
store panel.gdt x1 x2 x3 x4 x5 x6 x7 x8 x9 x10 \
x11 x12 x13 x14 x15 x16 x17 x18 x19 x20
Panel data marker strings
It can be helpful with panel data to have the observations identied by mnemonic markers. A
special function in the genr command is available for this purpose.
In the example above, suppose all the states are identied by two-letter codes in the left-most
column of the original datale. When the stacking operation is performed, these codes will be
stacked along with the data values. If the rst row is marked AR for Arkansas, then the marker AR
will end up being shown on each row containing an observation for Arkansas. Thats all very well,
but these markers dont tell us anything about the date of the observation. To rectify this we could
do:
4
You can also specify a list of series using the wildcard *; for example stack(p*) would stack all series whose
names begin with p.
genr time
genr year = 1960 + (5 * time)
genr markers = "%s:%d", marker, year
The rst line generates a 1-based index representing the period of each observation, and the second
line uses the time variable to generate a variable representing the year of the observation. The
third line contains this special feature: if (and only if) the name of the new variable to generate is
markers, the portion of the command following the equals sign is taken as a C-style format string
(which must be wrapped in double quotes), followed by a comma-separated list of arguments.
The arguments will be printed according to the given format to create a new set of observation
markers. Valid arguments are either the names of variables in the dataset, or the string marker
which denotes the pre-existing observation marker. The format speciers which are likely to be
useful in this context are %s for a string and %d for an integer. Strings can be truncated: for
example %.3s will use just the rst three characters of the string. To chop initial characters o
an existing observation marker when constructing a new one, you can use the syntax marker + n,
where n is a positive integer: in the case the rst n characters will be skipped.
After the commands above are processed, then, the observation markers will look like, for example,
AR:1965, where the two-letter state code and the year of the observation are spliced together with
a colon.
Panel dummy variables
In a panel study you may wish to construct dummy variables of one or both of the following sorts:
(a) dummies as unique identiers for the units or groups, and (b) dummies as unique identiers for
the time periods. The former may be used to allow the intercept of the regression to dier across
the units, the latter to allow the intercept to dier across periods.
Two special functions are available to create such dummies. These are found under the Add
menu in the GUI, or under the genr command in script mode or gretlcli.
1. unit dummies (script command genr unitdum). This command creates a set of dummy
variables identifying the cross-sectional units. The variable du_1 will have value 1 in each
row corresponding to a unit 1 observation, 0 otherwise; du_2 will have value 1 in each row
corresponding to a unit 2 observation, 0 otherwise; and so on.
2. time dummies (script command genr timedum). This command creates a set of dummy
variables identifying the periods. The variable dt_1 will have value 1 in each row correspond-
ing to a period 1 observation, 0 otherwise; dt_2 will have value 1 in each row corresponding
to a period 2 observation, 0 otherwise; and so on.
If a panel data set has the YEAR of the observation entered as one of the variables you can create a
periodic dummy to pick out a particular year, e.g. genr dum = (YEAR=1960). You can also create
periodic dummy variables using the modulus operator, %. For instance, to create a dummy with
value 1 for the rst observation and every thirtieth observation thereafter, 0 otherwise, do
genr index
genr dum = ((index-1) % 30) = 0
Lags, dierences, trends
If the time periods are evenly spaced you may want to use lagged values of variables in a panel
regression (but see also chapter 19); you may also wish to construct rst dierences of variables of
interest.
Once a dataset is identied as a panel, gretl will handle the generation of such variables correctly.
For example the command genr x1_1 = x1(-1) will create a variable that contains the rst lag
of x1 where available, and the missing value code where the lag is not available (e.g. at the start of
the time series for each group). When you run a regression using such variables, the program will
automatically skip the missing observations.
When a panel data set has a fairly substantial time dimension, you may wish to include a trend in
the analysis. The command genr time creates a variable named time which runs from 1 to T for
each unit, where T is the length of the time-series dimension of the panel. If you want to create an
index that runs consecutively from 1 to m T, where m is the number of units in the panel, use
genr index.
Basic statistics by unit
gretl contains functions which can be used to generate basic descriptive statistics for a given vari-
able, on a per-unit basis; these are pnobs() (number of valid cases), pmin() and pmax() (minimum
and maximum) and pmean() and psd() (mean and standard deviation).
As a brief illustration, suppose we have a panel data set comprising 8 time-series observations on
each of N units or groups. Then the command
genr pmx = pmean(x)
creates a series of this form: the rst 8 values (corresponding to unit 1) contain the mean of x for
unit 1, the next 8 values contain the mean for unit 2, and so on. The psd() function works in a
similar manner. The sample standard deviation for group i is computed as
s
i
=
(x x
i
)
2
T
i
1
where T
i
denotes the number of valid observations on x for the given unit, x
i
denotes the group
mean, and the summation is across valid observations for the group. If T
i
< 2, however, the
standard deviation is recorded as 0.
One particular use of psd() may be worth noting. If you want to form a sub-sample of a panel that
contains only those units for which the variable x is time-varying, you can either use
smpl (pmin(x) < pmax(x)) --restrict
or
smpl (psd(x) > 0) --restrict
4.6 Missing data values
Representation and handling
Missing values are represented internally as DBL_MAX, the largest oating-point number that can be
represented on the system (which is likely to be at least 10 to the power 300, and so should not
be confused with legitimate data values). In a native-format data le they should be represented
as NA. When importing CSV data gretl accepts several common representations of missing values
including 999, the string NA (in upper or lower case), a single dot, or simply a blank cell. Blank cells
should, of course, be properly delimited, e.g. 120.6,,5.38, in which the middle value is presumed
missing.
As for handling of missing values in the course of statistical analysis, gretl does the following:
In calculating descriptive statistics (mean, standard deviation, etc.) under the summary com-
mand, missing values are simply skipped and the sample size adjusted appropriately.
In running regressions gretl rst adjusts the beginning and end of the sample range, trun-
cating the sample if need be. Missing values at the beginning of the sample are common in
time series work due to the inclusion of lags, rst dierences and so on; missing values at the
end of the range are not uncommon due to dierential updating of series and possibly the
inclusion of leads.
If gretl detects any missing values inside the (possibly truncated) sample range for a regression,
the result depends on the character of the dataset and the estimator chosen. In many cases, the
program will automatically skip the missing observations when calculating the regression results.
In this situation a message is printed stating how many observations were dropped. On the other
hand, the skipping of missing observations is not supported for all procedures: exceptions include
all autoregressive estimators, system estimators such as SUR, and nonlinear least squares. In the
case of panel data, the skipping of missing observations is supported only if their omission leaves
a balanced panel. If missing observations are found in cases where they are not supported, gretl
gives an error message and refuses to produce estimates.
Manipulating missing values
Some special functions are available for the handling of missing values. The boolean function
missing() takes the name of a variable as its single argument; it returns a series with value 1 for
each observation at which the given variable has a missing value, and value 0 otherwise (that is, if
the given variable has a valid value at that observation). The function ok() is complementary to
missing; it is just a shorthand for !missing (where ! is the boolean NOT operator). For example,
one can count the missing values for variable x using
scalar nmiss_x = sum(missing(x))
The function zeromiss(), which again takes a single series as its argument, returns a series where
all zero values are set to the missing code. This should be used with cautionone does not want
to confuse missing values and zerosbut it can be useful in some contexts. For example, one can
determine the rst valid observation for a variable x using
genr time
scalar x0 = min(zeromiss(time * ok(x)))
The function misszero() does the opposite of zeromiss, that is, it converts all missing values to
zero.
It may be worth commenting on the propagation of missing values within genr formulae. The
general rule is that in arithmetical operations involving two variables, if either of the variables has
a missing value at observation t then the resulting series will also have a missing value at t. The
one exception to this rule is multiplication by zero: zero times a missing value produces zero (since
this is mathematically valid regardless of the unknown value).
4.7 Maximum size of data sets
Basically, the size of data sets (both the number of variables and the number of observations per
variable) is limited only by the characteristics of your computer. gretl allocates memory dynami-
cally, and will ask the operating system for as much memory as your data require. Obviously, then,
you are ultimately limited by the size of RAM.
Aside from the multiple-precision OLS option, gretl uses double-precision oating-point numbers
throughout. The size of such numbers in bytes depends on the computer platform, but is typically
eight. To give a rough notion of magnitudes, suppose we have a data set with 10,000 observations
on 500 variables. Thats 5 million oating-point numbers or 40 million bytes. If we dene the
megabyte (MB) as 1024 1024 bytes, as is standard in talking about RAM, its slightly over 38 MB.
The program needs additional memory for workspace, but even so, handling a data set of this size
should be quite feasible on a current PC, which at the time of writing is likely to have at least 256
MB of RAM.
If RAM is not an issue, there is one further limitation on data size (though its very unlikely to
be a binding constraint). That is, variables and observations are indexed by signed integers, and
on a typical PC these will be 32-bit values, capable of representing a maximum positive value of
2
31
1 = 2, 147, 483, 647.
The limits mentioned above apply to gretls native functionality. There are tighter limits with
regard to two third-party programs that are available as add-ons to gretl for certain sorts of time-
series analysis including seasonal adjustment, namely TRAMO/SEATS and X-12-ARIMA. These pro-
grams employ a xed-size memory allocation, and cant handle series of more than 600 observa-
tions.
4.8 Data le collections
If youre using gretl in a teaching context you may be interested in adding a collection of data les
and/or scripts that relate specically to your course, in such a way that students can browse and
access them easily.
There are three ways to access such collections of les:
For data les: select the menu item File, Open data, Sample le, or click on the folder icon
on the gretl toolbar.
For script les: select the menu item File, Script les, Practice le.
When a user selects one of the items:
The data or script les included in the gretl distribution are automatically shown (this includes
les relating to Ramanathans Introductory Econometrics and Greenes Econometric Analysis).
The program looks for certain known collections of data les available as optional extras,
for instance the datales from various econometrics textbooks (Davidson and MacKinnon,
Gujarati, Stock and Watson, Verbeek, Wooldridge) and the Penn World Table (PWT 5.6). (See
the data page at the gretl website for information on these collections.) If the additional les
are found, they are added to the selection windows.
The program then searches for valid le collections (not necessarily known in advance) in
these places: the system data directory, the system script directory, the user directory,
and all rst-level subdirectories of these. For reference, typical values for these directories
are shown in Table 4.1. (Note that PERSONAL is a placeholder that is expanded by Windows,
corresponding to My Documents on English-language systems.)
Linux MS Windows
system data dir /usr/share/gretl/data c:\Program Files\gretl\data
system script dir /usr/share/gretl/scripts c:\Program Files\gretl\scripts
user dir $HOME/gretl PERSONAL\gretl
Table 4.1: Typical locations for le collections
Any valid collections will be added to the selection windows. So what constitutes a valid le collec-
tion? This comprises either a set of data les in gretl XML format (with the .gdt sux) or a set of
script les containing gretl commands (with .inp sux), in each case accompanied by a master
le or catalog. The gretl distribution contains several example catalog les, for instance the le
descriptions in the misc sub-directory of the gretl data directory and ps_descriptions in the
misc sub-directory of the scripts directory.
If you are adding your own collection, data catalogs should be named descriptions and script
catalogs should be be named ps_descriptions. In each case the catalog should be placed (along
with the associated data or script les) in its own specic sub-directory (e.g. /usr/share/gretl/
data/mydata or c:\userdata\gretl\data\mydata).
The catalog les are plain text; if they contain non-ASCII characters they must be encoded as UTF-
8. The syntax of such les is straightforward. Here, for example, are the rst few lines of gretls
misc data catalog:
# Gretl: various illustrative datafiles
"arma","artificial data for ARMA script example"
"ects_nls","Nonlinear least squares example"
"hamilton","Prices and exchange rate, U.S. and Italy"
The rst line, which must start with a hash mark, contains a short name, here Gretl, which
will appear as the label for this collections tab in the data browser window, followed by a colon,
followed by an optional short description of the collection.
Subsequent lines contain two elements, separated by a comma and wrapped in double quotation
marks. The rst is a datale name (leave o the .gdt sux here) and the second is a short de-
scription of the content of that datale. There should be one such line for each datale in the
collection.
A script catalog le looks very similar, except that there are three elds in the le lines: a lename
(without its .inp sux), a brief description of the econometric point illustrated in the script, and
a brief indication of the nature of the data used. Again, here are the rst few lines of the supplied
misc script catalog:
# Gretl: various sample scripts
"arma","ARMA modeling","artificial data"
"ects_nls","Nonlinear least squares (Davidson)","artificial data"
"leverage","Influential observations","artificial data"
"longley","Multicollinearity","US employment"
If you want to make your own data collection available to users, these are the steps:
1. Assemble the data, in whatever format is convenient.
2. Convert the data to gretl format and save as gdt les. It is probably easiest to convert the data
by importing them into the program from plain text, CSV, or a spreadsheet format (MS Excel
or Gnumeric) then saving them. You may wish to add descriptions of the individual variables
(the Variable, Edit attributes menu item), and add information on the source of the data (the
Data, Edit info menu item).
3. Write a descriptions le for the collection using a text editor.
4. Put the datales plus the descriptions le in a subdirectory of the gretl data directory (or user
directory).
5. If the collection is to be distributed to other people, package the data les and catalog in some
suitable manner, e.g. as a ziple.
If you assemble such a collection, and the data are not proprietary, we would encourage you to
submit the collection for packaging as a gretl optional extra.
4.9 Assembling data from multiple sources
In many contexts researchers need to bring together data from multiple source les, and in some
cases these sources are not organized such that the data can simply be stuck together by append-
ing rows or columns to a base dataset. In gretl, the join command can be used for this purpose;
this command is discussed in detail in chapter 7.
Chapter 5
Sub-sampling a dataset
5.1 Introduction
Some subtle issues can arise here; this chapter attempts to explain the issues.
A sub-sample may be dened in relation to a full dataset in two dierent ways: we will refer to these
as setting the sample and restricting the sample; these methods are discussed in sections 5.2
and 5.3 respectively. In addition section 5.4 discusses some special issues relating to panel data,
and section 5.5 covers resampling with replacement, which is useful in the context of bootstrapping
test statistics.
The following discussion focuses on the command-line approach. But you can also invoke the
methods outlined here via the items under the Sample menu in the GUI program.
5.2 Setting the sample
By setting the sample we mean dening a sub-sample simply by means of adjusting the starting
and/or ending point of the current sample range. This is likely to be most relevant for time-series
data. For example, one has quarterly data from 1960:1 to 2003:4, and one wants to run a regression
using only data from the 1970s. A suitable command is then
smpl 1970:1 1979:4
Or one wishes to set aside a block of observations at the end of the data period for out-of-sample
forecasting. In that case one might do
smpl ; 2000:4
where the semicolon is shorthand for leave the starting observation unchanged. (The semicolon
may also be used in place of the second parameter, to mean that the ending observation should be
unchanged.) By unchanged here, we mean unchanged relative to the last smpl setting, or relative
to the full dataset if no sub-sample has been dened up to this point. For example, after
smpl 1970:1 2003:4
smpl ; 2000:4
the sample range will be 1970:1 to 2000:4.
An incremental or relative form of setting the sample range is also supported. In this case a relative
oset should be given, in the form of a signed integer (or a semicolon to indicate no change), for
both the starting and ending point. For example
smpl +1 ;
will advance the starting observation by one while preserving the ending observation, and
smpl +2 -1
31
Chapter 5. Sub-sampling a dataset 32
will both advance the starting observation by two and retard the ending observation by one.
An important feature of setting the sample as described above is that it necessarily results in
the selection of a subset of observations that are contiguous in the full dataset. The structure of
the dataset is therefore unaected (for example, if it is a quarterly time series before setting the
sample, it remains a quarterly time series afterwards).
5.3 Restricting the sample
By restricting the sample we mean selecting observations on the basis of some Boolean (logical)
criterion, or by means of a random number generator. This is likely to be most relevant for cross-
sectional or panel data.
Suppose we have data on a cross-section of individuals, recording their gender, income and other
characteristics. We wish to select for analysis only the women. If we have a gender dummy variable
with value 1 for men and 0 for women we could do
smpl gender==0 --restrict
to this eect. Or suppose we want to restrict the sample to respondents with incomes over $50,000.
Then we could use
smpl income>50000 --restrict
A question arises: if we issue the two commands above in sequence, what do we end up with in
our sub-sample: all cases with income over 50000, or just women with income over 50000? By
default, the answer is the latter: women with income over 50000. The second restriction augments
the rst, or in other words the nal restriction is the logical product of the new restriction and any
restriction that is already in place. If you want a new restriction to replace any existing restrictions
you can rst recreate the full dataset using
smpl --full
Alternatively, you can add the replace option to the smpl command:
smpl income>50000 --restrict --replace
This option has the eect of automatically re-establishing the full dataset before applying the new
restriction.
Unlike a simple setting of the sample, restricting the sample may result in selection of non-
contiguous observations from the full data set. It may therefore change the structure of the data
set.
This can be seen in the case of panel data. Say we have a panel of ve rms (indexed by the variable
firm) observed in each of several years (identied by the variable year). Then the restriction
smpl year==1995 --restrict
produces a dataset that is not a panel, but a cross-section for the year 1995. Similarly
smpl firm==3 --restrict
produces a time-series dataset for rm number 3.
For these reasons (possible non-contiguity in the observations, possible change in the structure of
the data), gretl acts dierently when you restrict the sample as opposed to simply setting it. In
the case of setting, the program merely records the starting and ending observations and uses these
as parameters to the various commands calling for the estimation of models, the computation of
statistics, and so on. In the case of restriction, the program makes a reduced copy of the dataset
and by default treats this reduced copy as a simple, undated cross-sectionbut see the further
discussion of panel data in section 5.4.
If you wish to re-impose a time-series interpretation of the reduced dataset you can do so using the
setobs command, or the GUI menu item Data, Dataset structure.
The fact that restricting the sample results in the creation of a reduced copy of the original
dataset may raise an issue when the dataset is very large. With such a dataset in memory, the
creation of a copy may lead to a situation where the computer runs low on memory for calculating
regression results. You can work around this as follows:
1. Open the full data set, and impose the sample restriction.
2. Save a copy of the reduced data set to disk.
3. Close the full dataset and open the reduced one.
4. Proceed with your analysis.
Random sub-sampling
Besides restricting the sample on some deterministic criterion, it may sometimes be useful (when
working with very large datasets, or perhaps to study the properties of an estimator) to draw a
random sub-sample from the full dataset. This can be done using, for example,
smpl 100 --random
to select 100 cases. If you want the sample to be reproducible, you should set the seed for the
random number generator rst, using the set command. This sort of sampling falls under the
restriction category: a reduced copy of the dataset is made.
5.4 Panel data
Consider for concreteness the ArellanoBond dataset supplied with gretl (abdata.gdt). This com-
prises data on 140 rms (n = 140) observed over the years 19761984 (T = 9). The dataset is
nominally balanced in the sense that that the time-series length is the same for all countries (this
being a requirement for a dataset to count as a panel in gretl), but in fact there are many missing
values (NAs).
You may want to sub-sample such a dataset in either the cross-sectional dimension (limit the sam-
ple to a subset of rms) or the time dimension (e.g. use data from the 1980s only). The simplest
(but limited) way to sub-sample on rms keys o the notation used by gretl for panel observations.
The full data range is printed as 1:1 (rm 1, period 1) to 140:9 (rm 140, period 9). The eect of
smpl 1:1 80:9
is to limit the sample to the rst 80 rms. Note that if you instead tried smpl 1:1 80:4, gretl
would insist on preserving the balance of the panel and would truncate the range to complete
rms, as if you had typed smpl 1:1 79:9.
The rms in the ArellanoBond dataset are anonymous, but suppose you had a panel with ve
named countries. With such a panel you can inform gretl of the names of the groups using the
setobs command. For example, given
string cstr = "Portugal Italy Ireland Greece Spain"
setobs country cstr --panel-groups
gretl creates a string-valued series named country with group names taken from the variable cstr.
Then, to include only Italy and Spain you could do
smpl country=="Italy" || country=="Spain" --restrict
or to exclude one country,
smpl country!="Ireland" --restrict
To sub-sample in the time dimension, use of --restrict is required. For example, the Arellano
Bond dataset contains a variable named YEAR that records the year of the observations and if one
wanted to omit the rst two years of data one could do
smpl YEAR >= 1978 --restrict
If a dataset does not already incude a suitable variable for this purpose one can use the command
genr time to create a simple 1-based time index.
Note that if you apply a sample restriction that just selects certain units (rms, countries or what-
ever), or selects certain contiguous time-periodssuch that n > 1, T > 1 and the time-series length
is still the same across all included unitsyour sub-sample will still be interpreted by gretl as a
panel.
Unbalancing restrictions
In some cases one wants to sub-sample according to a criterion that cuts across the grain of
a panel dataset. For instance, suppose you have a micro dataset with thousands of individuals
observed over several years and you want to restrict the sample to observations on employed
women.
If we simply extracted from the total nT rows of the dataset those that pertain to women who were
employed at time t (t = 1, . . . , T) we would likely end up with a dataset that doesnt count as a
panel in gretl (because the specic time-series length, T
i
, would dier across individuals). In some
contexts it might be OK that gretl doesnt take your sub-sample to be a panel, but if you want to
apply panel-specic methods this is a problem. You can solve it by giving the --balanced option
with smpl. For example, supposing your dataset contained dummy variables gender (with the value
1 coding for women) and employed, you could do
smpl gender==1 && employed==1 --restrict --balanced
What exactly does this do? Well, lets say the years of your data are 2000, 2005 and 2010, and that
some women were employed in all of those years, giving a maximum T
i
value of 3. But individual
526 is a women who was employed only in the year 2000 (T
i
= 1). The eect of the --balanced
option is then to insert padding rows of NAs for the years 2005 and 2010 for individual 526, and
similarly for all individuals with 0 < T
i
< 3. Your sub-sample then qualies as a panel.
5.5 Resampling and bootstrapping
Given an original data series x, the command
series xr = resample(x)
creates a new series each of whose elements is drawn at random from the elements of x. If the
original series has 100 observations, each element of x is selected with probability 1/100 at each
drawing. Thus the eect is to shue the elements of x, with the twist that each element of x may
appear more than once, or not at all, in xr.
The primary use of this function is in the construction of bootstrap condence intervals or p-values.
Here is a simple example. Suppose we estimate a simple regression of y on x via OLS and nd that
the slope coecient has a reported t-ratio of 2.5 with 40 degrees of freedom. The two-tailed p-
value for the null hypothesis that the slope parameter equals zero is then 0.0166, using the t(40)
distribution. Depending on the context, however, we may doubt whether the ratio of coecient to
standard error truly follows the t(40) distribution. In that case we could derive a bootstrap p-value
as shown in Example 5.1.
Under the null hypothesis that the slope with respect to x is zero, y is simply equal to its mean plus
an error term. We simulate y by resampling the residuals from the initial OLS and re-estimate the
model. We repeat this procedure a large number of times, and record the number of cases where
the absolute value of the t-ratio is greater than 2.5: the proportion of such cases is our bootstrap
p-value. For a good discussion of simulation-based tests and bootstrapping, see Davidson and
MacKinnon (2004, chapter 4).
Example 5.1: Calculation of bootstrap p-value
ols y 0 x
# save the residuals
genr ui = $uhat
scalar ybar = mean(y)
# number of replications for bootstrap
scalar replics = 10000
scalar tcount = 0
series ysim
loop replics --quiet
# generate simulated y by resampling
ysim = ybar + resample(ui)
ols ysim 0 x
scalar tsim = abs($coeff(x) / $stderr(x))
tcount += (tsim > 2.5)
endloop
printf "proportion of cases with |t| > 2.5 = %g\n", tcount / replics
Chapter 6
Graphs and plots
6.1 Gnuplot graphs
A separate program, gnuplot, is called to generate graphs. Gnuplot is a very full-featured graphing
program with myriad options. It is available from www.gnuplot.info (but note that a suitable copy
of gnuplot is bundled with the packaged versions of gretl for MS Windows and Mac OS X). Gretl
gives you direct access, via a graphical interface, to a subset of gnuplots options and it tries to
choose sensible values for you; it also allows you to take complete control over graph details if you
wish.
With a graph displayed, you can click on the graph window for a pop-up menu with the following
options.
Save as PNG: Save the graph in Portable Network Graphics format (the same format that you
see on screen).
Save as postscript: Save in encapsulated postscript (EPS) format.
Save as Windows metale: Save in Enhanced Metale (EMF) format.
Save to session as icon: The graph will appear in iconic form when you select Icon view from
the View menu.
Zoom: Lets you select an area within the graph for closer inspection (not available for all
graphs).
Print: (Current GTK or MS Windows only) lets you print the graph directly.
Copy to clipboard: MS Windows only, lets you paste the graph into Windows applications such
as MS Word.
Edit: Opens a controller for the plot which lets you adjust many aspects of its appearance.
Close: Closes the graph window.
Displaying data labels
For simple X-Y scatter plots, some further options are available if the dataset includes case mark-
ers (that is, labels identifying each observation).
1
With a scatter plot displayed, when you move
the mouse pointer over a data point its label is shown on the graph. By default these labels are
transient: they do not appear in the printed or copied version of the graph. They can be removed by
selecting Clear data labels from the graph pop-up menu. If you want the labels to be axed per-
manently (so they will show up when the graph is printed or copied), select the option Freeze data
labels from the pop-up menu; Clear data labels cancels this operation. The other label-related
option, All data labels, requests that case markers be shown for all observations. At present the
display of case markers is disabled for graphs containing more than 250 data points.
1
For an example of such a dataset, see the Ramanathan le data4-10: this contains data on private school enrollment
for the 50 states of the USA plus Washington, DC; the case markers are the two-letter codes for the states.
36
Chapter 6. Graphs and plots 37
GUI plot editor
Selecting the Edit option in the graph popup menu opens an editing dialog box, shown in Figure 6.1.
Notice that there are several tabs, allowing you to adjust many aspects of a graphs appearance:
font, title, axis scaling, line colors and types, and so on. You can also add lines or descriptive labels
to a graph (under the Lines and Labels tabs). The Apply button applies your changes without
closing the editor; OK applies the changes and closes the dialog.
Figure 6.1: gretls gnuplot controller
Publication-quality graphics: advanced options
The GUI plot editor has two limitations. First, it cannot represent all the myriad options that
gnuplot oers. Users who are suciently familiar with gnuplot to know what theyre missing in
the plot editor presumably dont need much help from gretl, so long as they can get hold of the
gnuplot command le that gretl has put together. Second, even if the plot editor meets your needs,
in terms of ne-tuning the graph you see on screen, a few details may need further work in order
to get optimal results for publication.
Either way, the rst step in advanced tweaking of a graph is to get access to the graph command
le.
In the graph display window, right-click and choose Save to session as icon.
If its not already open, open the icon view windoweither via the menu item View/Icon view,
or by clicking the session icon view button on the main-window toolbar.
Right-click on the icon representing the newly added graph and select Edit plot commands
from the pop-up menu.
You get a window displaying the plot le (Figure 6.2).
Here are the basic things you can do in this window. Obviously, you can edit the le you just
opened. You can also send it for processing by gnuplot, by clicking the Execute (cogwheel) icon
in the toolbar. Or you can use the Save as button to save a copy for editing and processing as you
wish.
Figure 6.2: Plot commands editor
Unless youre a gnuplot expert, most likely youll only need to edit a couple of lines at the top of
the le, specifying a driver (plus options) and an output le. We oer here a brief summary of some
points that may be useful.
First, gnuplots output mode is set via the command set term followed by the name of a supported
driver (terminal in gnuplot parlance) plus various possible options. (The top line in the plot
commands window shows the set term line that gretl used to make a PNG le, commented out.)
The graphic formats that are most suitable for publication are PDF and EPS. These are supported
by the gnuplot term types pdf, pdfcairo and postscript (with the eps option). The pdfcairo
driver has the virtue that is behaves in a very similar manner to the PNG one, the output of which
you see on screen. This is provided by the version of gnuplot that is included in the gretl packages
for MS Windows and Mac OS X; if youre on Linux it may or may be supported. If pdfcairo is not
available, the pdf terminal may be available; the postscript terminal is almost certainly available.
Besides selecting a term type, if you want to get gnuplot to write the actual output le you need
to append a set output line giving a lename. Here are a few examples of the rst two lines you
might type in the window editing your plot commands. Well make these more realistic shortly.
set term pdfcairo
set output mygraph.pdf
set term pdf
set output mygraph.pdf
set term postscript eps
set output mygraph.eps
There are a couple of things worth remarking here. First, you may want to adjust the size of the
graph, and second you may want to change the font. The default sizes produced by the above
drivers are 5 inches by 3 inches for pdfcairo and pdf, and 5 inches by 3.5 inches for postscript
eps. In each case you can change this by giving a size specication, which takes the form XX,YY
(examples below).
You may ask, why bother changing the size in the gnuplot command le? After all, PDF and EPS are
both vector formats, so the graphs can be scaled at will. True, but a uniform scaling will also aect
the font size, which may end looking wrong. You can get optimal results by experimenting with
the font and size options to gnuplots set term command. Here are some examples (comments
follow below).
# pdfcairo, regular size, slightly amended
set term pdfcairo font "Sans,6" size 5in,3.5in
# or small size
set term pdfcairo font "Sans,5" size 3in,2in
# pdf, regular size, slightly amended
set term pdf font "Helvetica,8" size 5in,3.5in
# or small
set term pdf font "Helvetica,6" size 3in,2in
# postscript, regular
set term post eps solid font "Helvetica,16"
# or small
set term post eps solid font "Helvetica,12" size 3in,2in
On the rst line we set a sans serif font for pdfcairo at a suitable size for a 5 3.5 inch plot
(which you may nd looks better than the rather letterboxy default of 5 3). And on the second
we illustrate what you might do to get a smaller 3 2 inch plot. You can specify the plot size in
centimeters if you prefer, as in
set term pdfcairo font "Sans,6" size 6cm,4cm
We then repeat the exercise for the pdf terminal. Notice that here were specifying one of the 35
standard PostScript fonts, namely Helvetica. Unlike pdfcairo, the plain pdf driver is unlikely to
be able to nd fonts other than these.
In the third pair of lines we illustrate options for the postscript driver (which, as you see, can
be abbreviated as post). Note that here we have added the option solid. Unlike most other
drivers, this one uses dashed lines unless you specify the solid option. Also note that weve
(apparently) specied a much larger font in this case. Thats because the eps option in eect tells
the postscript driver to work at half-size (among other things), so we need to double the font
size.
Table 6.1 summarizes the basics for the three drivers we have mentioned.
Terminal default size (inches) suggested font
pdfcairo 5 3 Sans,6
pdf 5 3 Helvetica,8
post eps 5 3.5 Helvetica,16
Table 6.1: Drivers for publication-quality graphics
To nd out more about gnuplot visit www.gnuplot.info. This site has documentation for the current
version of the program in various formats.
Additional tips
To be written. Line widths, enhanced text. Show a before and after example.
6.2 Boxplots
These plots (after Tukey and Chambers) display the distribution of a variable. The central box
encloses the middle 50 percent of the data, i.e. it is bounded by the rst and third quartiles. The
whiskers extend to the minimum and maximum values. A line is drawn across the box at the
median and a + sign identies the meansee Figure 6.3.
0.05
0.1
0.15
0.2
0.25
ENROLL
median
Q1
Q3
mean
Figure 6.3: Sample boxplot
In the case of boxplots with condence intervals, dotted lines show the limits of an approximate 90
percent condence interval for the median. This is obtained by the bootstrap method, which can
take a while if the data series is very long.
After each variable specied in the boxplot command, a parenthesized boolean expression may
be added, to limit the sample for the variable in question. A space must be inserted between the
variable name or number and the expression. Suppose you have salary gures for men and women,
and you have a dummy variable GENDER with value 1 for men and 0 for women. In that case you
could draw comparative boxplots with the following line in the boxplots dialog:
salary (GENDER=1) salary (GENDER=0)
Chapter 7
Joining data sources
7.1 Introduction
Gretl provides two commands for adding data from le to an existing dataset in the programs
workspace, namely append and join. The append command, which has been available for a long
time, is relatively simple and is described in the Gretl Command Reference. Here we focus on the
join command, which is much more exible and sophisticated. This chapter gives an overview
of the functionality of join along with a detailed account of its syntax and options. We provide
several toy examples and discuss one real-world case at length.
First, a note on terminology: in the following we use the terms left-hand and inner to refer
to the dataset that is already in memory, and the terms right-hand and outer to refer to the
dataset in the le from which additional data are to be drawn.
Two main features of join are worth emphasizing at the outset:
Key variables can be used to match specic observations (rows) in the inner and outer
datasets, and this match need not be 1 to 1.
A row lter may be applied to screen out unwanted observations in the outer dataset.
As will be explained below, these features support rather complex concatenation and manipulation
of data from dierent sources. The potential complexity of a join operation, when several options
are activated, means that it is not possible to support batch importation of several series with
a single command. In contrast with the append command, which by default imports the entire
content of a data le, join imports one series at a time. A command loop can be used to apply
join to multiple series.
A related aspect of join should be notedone that makes this command particularly useful when
dealing with very large data les. That is, when gretl executes a join operation it does not, in gen-
eral, read into memory the entire content of the right-hand side dataset. Only those columns that
are actually needed for the operation are read in full. This makes join faster and less demanding
of RAM than the methods available in most other software. On the other hand, gretls asymmetrical
treatment of the inner and outer datasets in join may require some getting used to, for users
of other packages.
7.2 Basic syntax
The minimal invocation of join is
join lename varname
where lename is the name of a delimited text data le and varname is the name of the series to be
imported. Only column-delimited text les are supported at present; the delimiter may be comma,
space, tab or semicolon. A series named varname may already be present in the left-hand dataset,
but that is not required. The series to be imported may be numerical or string-valued.
The eect of the minimal version of join is this: gretl looks for a data column labeled varname in
the specied le; if such a column is found and the number of observations on the right matches
the number of observations in the current sample range on the left, then the values from the right
41
Chapter 7. Joining data sources 42
are copied into the relevant range of observations on the left. If varname does not already exist
on the left, any observations outside of the current sample are set to NA; if it exists already then
observations outside of the current sample are left unchanged.
The case where you want to rename a series on import is handled by the --data option. This option
has one required argument, the name by which the series is known on the right. At this point we
need to explain something about right-hand variable names (column headings).
Right-hand names
We accept on input arbitrary column heading strings, but if these strings do not qualify as valid
gretl identiers they are automatically converted, and in the context of join you must use the
converted names. A gretl identier must start with a letter, contain nothing but (ASCII) letters,
digits and the underscore character, and must not exceed 31 characters. The rules used in name
conversion are:
1. Skip any leading non-letters.
2. Until the 31-character is reached or the input is exhausted: transcribe legal characters; skip
illegal characters apart from spaces; and replace one or more consecutive spaces with an
underscore, unless the last character transcribed is an underscore in which case space is
skipped.
In the unlikely event that this policy yields an empty string, we replace the original with coln,
where n is replaced by the 1-based index of the column in question among those used in the
join operation. If you are in doubt regarding the converted name of a given column, the function
fixname() can be used as a check: it takes the original string as an argument and returns the
converted name. Examples:
? eval fixname("valid_identifier")
valid_identifier
? eval fixname("12. Some name")
Some_name
Returning to the use of the --data option, suppose we have a column headed "12. Some name"
on the right and wish to import it as x. After guring how the right-hand name converts, we can do
join foo.csv x --data="Some_name"
7.3 Filtering
Rows from the outer dataset can be ltered using the --filter option. The required parameter
for this option is a Boolean condition, that is, an expression which evaluates to non-zero (true,
include the row) or zero (false, skip the row) for each of the outer rows. The lter expression may
include any of the following terms: up to three right-hand series (under their converted names as
explained above); scalar or string variables dened on the left; any of the operators and functions
available in gretl (including user-dened functions); and numeric or string constants.
Here are a few simple examples of potentially valid lter options (assuming that the specied right-
hand side columns are found):
# 1. relationship between two right-hand variables
--filter="x15<=x17"
# 2. comparison of right-hand variable with constant
--filter="nkids>2"
# 3. comparison of string-valued right-hand variable with string constant
--filter="SEX==\"F\""
# 4. filter on valid values of a right-hand variable
--filter=!missing(income)
# 5. compound condition
--filter="x < 100 && (x > 0 || y > 0)"
Note that if you are comparing against a string constant (as in example 3 above) it is necessary
to put the string in escaped double-quotes (each double-quote preceded by a backslash) so the
interpreter knows that F is not supposed to be the name of a variable.
It is safest to enclose the whole lter expression in double quotes, however this is not strictly
required unless the expression contains spaces or the equals sign.
In general, an error is agged if a missing value is encountered in a series referenced in a lter
expression. This is because the condition then becomes indeterminate; taking example 2 above, if
the nkids value is NA on any given row we are not in a position to evaluate the condition nkids>2.
However, you can use the missing() functionor ok(), which is a shorthand for !missing()if
you need a lter that keys o the missing or non-missing status of a variable.
7.4 Matching with keys
Things get interesting when we come to key-matching. The purpose of this facility is perhaps best
introduced by example. Suppose that (as with many survey and census-based datasets) we have a
dataset that is composed of two or more related les, each having a dierent unit of observation;
for example we have a persons data le and a households data le. Table 7.1 shows a simple,
articial case. The le people.csv contains a unique identier for the individuals, pid. The
households le, hholds.csv, contains the unique household identier hid, which is also present
in the persons le.
As a rst example of join with keys, lets add the household-level variable xh to the persons
dataset:
open people.csv --quiet
join hholds.csv xh --ikey=hid
print --byobs
The basic key option is named ikey; this indicates inner key, that is, the key variable found in the
left-hand or inner dataset. By default it is assumed that the right-hand dataset contains a column of
the same name, though as well see below that assumption can be overridden. The join command
above says, nd a series named xh in the right-hand dataset and add it to the left-hand one, using
the values of hid to match rows. Looking at the data in Table 7.1 we can see how this should
work. Persons 1 and 2 are both members of household 1, so they should both get values of 1 for
xh; persons 3 and 4 are members of household 2, so that xh = 4; and so on. Note that the order
in which the key values occur on the right-hand side does not matter. The gretl output from the
print command is shown in the lower panel of Table 7.1.
Note that key variables are treated conceptually as integers. If a specied key contains fractional
values these are truncated.
Three extensions of the basic key mechanism are available.
If the outer dataset contains a relevant key variable but it goes under a dierent name from
the inner key, you can use the --okey option to specify the outer key. (As with other right-
hand names, this does not have to be a valid gretl identier.) So, for example, if hholds.csv
people.csv hholds.csv
pid,hid,gender,age,xp hid,country,xh
1,1,M,50,1 1,US,1
2,1,F,40,2 6,IT,12
3,2,M,30,3 3,UK,6
4,2,F,25,2 4,IT,8
5,3,M,40,3 2,US,4
6,4,F,35,4 5,IT,10
7,4,M,70,3
8,4,F,60,3
9,5,F,20,4
10,6,M,40,4
pid hid xh
1 1 1
2 1 1
3 2 4
4 2 4
5 3 6
6 4 8
7 4 8
8 4 8
9 5 10
10 6 12
Table 7.1: Two linked CSV data les, and the eect of a join
contained the hid information, but under the name HHOLD, the join command above could
be modied as
join hholds.csv xh --ikey=hid --okey=HHOLD
If a single key is not sucient to generate the matches you want, you can specify a double key
in the form of two series names separated by a comma; in this case the importation of data is
restricted to those rows on which both keys match. The syntax here is, for example
join foo.csv x --ikey=key1,key2
Again, the --okey option may be used if the corresponding right-hand columns are named
dierently. The same number of keys must be given on the left and the right, but when a
double key is used and only one of the key names diers on the right, the name that is in
common may be omitted (although the comma separator must be retained). For example, the
second of the following lines is acceptable shorthand for the rst:
join foo.csv x --ikey=key1,Lkey2 --okey=key1,Rkey2
join foo.csv x --ikey=key1,Lkey2 --okey=,Rkey2
The example shown in Table 7.1 is an instance of a 1 to 1 match, in the sense that applying
the matching criterion produces exactly one value of the variable xh corresponding to each
row of the inner dataset. Two other possibilities arise.
First, it may be that some rows in the inner dataset have no match on the right: by default
such observations get NA for the imported data. In some cases, the appropriate value to
record in the absence of a match may be zero rather than NA. For example, consider a query
on number of hours worked when the inner dataset contains individuals and the outer le
contains data on jobs: if an individual does not appear in the jobs le, the number of hours
worked is implicitly zero. In such cases gretls misszero() function can be used to adjust
the imported data.
Second, some left-hand rows may have more than one match on the right; we refer to this as
1 to n matching. In that case it is necessary to specify an aggregation method. This is the
job of the --aggr option, discussed below.
7.5 Aggregation
The --aggr option for 1 to n matching of rows takes a single argument, which must be one of the
following:
Code Value returned
count count of matches
avg mean of matching values
sum sum of matching values
min minimum of matching values
max maximum of matching values
seq:i the i
th
matching value (e.g. seq:2)
min(aux) minimum of matching values of auxiliary variable
max(aux) maximum of matching values of auxiliary variable
Note that the count aggregation method is special, in that there is no need for a data series on
the right; the imported series is simply a function of the specied key(s). All the other methods
require that actual data are found on the right. Also note that when count is used, the value
returned when no match is found is (as one might expect) zero rather than NA.
The basic use of the seq method is shown above: following the colon you give a positive integer rep-
resenting the (1-based) position of the observation in the sequence of matched rows. Alternatively,
a negative integer can be used to count down from the last match (seq:-1 selects the last match,
seq:-2 the second-last match, and so on). If the specied sequence number is out of bounds for a
given observation this method returns NA.
Referring again to the data in Table 7.1, suppose we want to import data from the persons le into
a dataset established at household level. Heres an example where we use the individual age data
from people.csv to add the average and minimum age of household members.
open hholds.csv --quiet
join people.csv avgage --ikey=hid --data=age --aggr=avg
join people.csv minage --ikey=hid --data=age --aggr=min
Heres a further example where we add to the household data the sum of the personal data xp, with
the twist that we apply lters to get the sum specically for household members under the age of
40, and for women.
join people.csv young_xp --ikey=hid --filter="age<40" --data=xp --aggr=sum
join people.csv female_xp --ikey=hid --filter="gender==\"F\"" --data=xp --aggr=sum
The possibility of using an auxiliary variable with the min and max modes of aggregation gives extra
exibility. For example, suppose we want for each household the income of its oldest member:
join people.csv oldest_xp --ikey=hid --data=xp --aggr=max(age)
7.6 String-valued key variables
The examples above use numerical variables (household and individual ID numbers) in the match-
ing process. It is also possible to use string-valued variables, in which case a match means that the
string values of the key variables compare equal (with case sensitivity). When using double keys,
you can mix numerical and string keys, but naturally you cannot mix a string variable on the left
(via ikey) with a numerical one on the right (via okey), or vice versa.
Heres a simple example. Suppose that alongside hholds.csv we have a le countries.csv with
the following content:
country,GDP
UK,100
US,500
IT,150
FR,180
The variable country, which is also found in hholds.csv, is string-valued. We can pull the GDP of
the country in which the household resides into our households dataset with
open hholds.csv -q
join countries.csv GDP --ikey=country
which gives
hid country GDP
1 1 1 500
2 6 2 150
3 3 3 100
4 4 2 150
5 2 1 500
6 5 2 150
7.7 A real-world case
For a real use-case for join with cross-sectional data, we turn to the Bank of Italys Survey on House-
hold Income and Wealth (SHIW).
1
In ASCII form the 2010 survey results comprise 47MB of data in
29 les. In this exercise we will draw on ve of the SHIW les to construct a replica of the dataset
used in Thomas Mrozs famous paper (Mroz, 1987) on womens labor force participation, which
contains data on married women between the age of 30 and 60 along with certain characteristics
of their households and husbands.
Our general strategy is as follows: we create a core dataset by opening the le carcom10.csv,
which contains basic data on the individuals. After dropping unwanted individuals (all but married
women), we use the resulting dataset as a base for pulling in further data via the join command.
The complete script to do the job is given in the Appendix to this chapter; here we walk through
the script with comments interspersed. We assume that all the relevant les from the Bank of Italy
survey are contained in a subdirectory called SHIW.
Starting with carcom10.csv, we use the --cols option to the open command to import specic
series, namely NQUEST (household ID number), NORD (sequence number for individuals within each
household), SEX (male = 1, female = 2), PARENT (status in household: 1 = head of household, 2 =
spouse of head, etc.), STACIV (marital status: married = 1), STUDIO (educational level, coded from
1 to 8), ETA (age in years) and ACOM4C (size of town).
open SHIW/carcom10.csv --cols=1,2,3,4,9,10,29,41
We then restrict the sample to married women from 30 to 60 years of age, and additionally restrict
the sample of women to those who are either heads of households or spouses of the head.
smpl SEX==2 && ETA>=30 && ETA<=60 && STACIV==1 --restrict
smpl PARENT<3 --restrict
For compatibility with the Mroz dataset as presented in the gretl data le mroz87.gdt, we rename
the age and education variables as WA and WE respectively, we compute the CIT dummy and nally
we store the reduced base dataset in gretl format.
rename ETA WA
rename STUDIO WE
series CIT = (ACOM4C > 2)
store mroz_rep.gdt
The next step will be to get data on working hours from the jobs le allb1.csv. Theres a com-
plication here. We need the total hours worked over the course of the year (for both the women
and their husbands). This is not available as such, but the variables ORETOT and MESILAV give,
respectively, average hours worked per week and the number of months worked in 2010, each on a
per-job basis. If each person held at most one job over the year we could compute his or her annual
hours as
HRS = ORETOT * 52 * MESILAV/12
1
Details of the survey can be found at http://www.bancaditalia.it/statistiche/indcamp/bilfait/dismicro.
The ASCII (CSV) data les for the 2010 survey are available at http://www.bancaditalia.it/statistiche/indcamp/
bilfait/dismicro/annuale/ascii/ind10_ascii.zip.
However, some people had more than one job, and in this case what we want is the sum of annual
hours across their jobs. We could use join with the seq aggregation method to construct this sum,
but it is probably more straightforward to read the allb1 data, compute the HRS values per job as
shown above, and save the results to a temporary CSV le.
open SHIW/allb1.csv --cols=1,2,8,11 --quiet
series HRS = misszero(ORETOT) * 52 * misszero(MESILAV)/12
store HRS.csv NQUEST NORD HRS
Now we can reopen the base dataset and join the hours variable from HRS.csv. Note that we need
a double key here: the women are uniquely identied by the combination of NQUEST and NORD. We
dont need an okey specication since these keys go under the same names in the right-hand le.
We dene labor force participation, LFP, based on hours.
open mroz_rep.gdt
join HRS.csv WHRS --ikey=NQUEST,NORD --data=HRS --aggr=sum
WHRS = misszero(WHRS)
LFP = WHRS > 0
For reference, heres how we could have used seq to avoid writing a temporary le:
join SHIW/allb1.csv njobs --ikey=NQUEST,NORD --data=ORETOT --aggr=count
series WHRS = 0
loop i=1..max(njobs) -q
join SHIW/allb1.csv htmp --ikey=NQUEST,NORD --data=ORETOT --aggr="seq:$i"
join SHIW/allb1.csv mtmp --ikey=NQUEST,NORD --data=MESILAV --aggr="seq:$i"
WHRS += misszero(htmp) * 52 * misszero(mtmp)/12
endloop
To generate the work experience variable, AX, we use the le lavoro.csv: this contains a variable
named ETALAV which records the age at which the person rst started work.
join SHIW/lavoro.csv ETALAV --ikey=NQUEST,NORD
series AX = misszero(WA - ETALAV)
We compute the womans hourly wage, WW, as the ratio of total employment income to annual
working hours. This requires drawing the series YL (payroll income) and YM (net self-employment
income) from the persons le rper10.csv.
join SHIW/rper10.csv YL --ikey=NQUEST,NORD --aggr=sum
join SHIW/rper10.csv YM --ikey=NQUEST,NORD --aggr=sum
series WW = LFP ? (YL + YM)/WHRS : 0
The familys net disposable income is available as Y in the le rfam10.csv; we import this as
FAMINC.
join SHIW/rfam10.csv FAMINC --ikey=NQUEST --data=Y
Data on number of children are now obtained by applying the count method. For the Mroz repli-
cation we want the number of children under the age of 6, and also the number aged 6 to 18.
join SHIW/carcom10.csv KIDS --ikey=NQUEST --aggr=count --filter="ETA<=18"
join SHIW/carcom10.csv KL6 --ikey=NQUEST --aggr=count --filter=ETA<6
series K618 = KIDS - KL6
We want to add data on the womens husbands, but how do we nd them? To do this we create an
additional inner key which well call H_ID (husband ID), by sub-sampling in turn on the observations
falling into each of two classes: (a) those where the woman is recorded as head of household and
(b) those where the husband has that status. In each case we want the individual ID (NORD) of the
household member whose status is complementary to that of the woman in question. So for case
(a) we subsample using PARENT==1 (head of household) and lter the join using PARENT==2 (spouse
of head); in case (b) we do the converse. We thus construct H_ID piece-wise.
# for women who are household heads
smpl PARENT==1 --restrict --replace
join SHIW/carcom10.csv H_ID --ikey=NQUEST --data=NORD --filter="PARENT==2"
# for women who are not household heads
smpl full
Now we can use our new inner key to retrieve the husbands data, matching H_ID on the left with
NORD on the right within each household.
join SHIW/carcom10.csv HA --ikey=NQUEST,H_ID --okey=NQUEST,NORD --data=ETA
join SHIW/carcom10.csv HE --ikey=NQUEST,H_ID --okey=NQUEST,NORD --data=STUDIO
join HRS.csv HHRS --ikey=NQUEST,H_ID --okey=NQUEST,NORD --data=HRS --aggr=sum
HHRS = misszero(HHRS)
The remainder of the script is straightforward and does not require discussion here: we recode
the education variables for compatibility; delete some intermediate series that are not needed any
more; add informative labels; and save the nal product. See the Appendix for details.
To compare the results from this dataset with those from the earlier US data used by Mroz, one can
copy the input le heckit.inp (supplied with the gretl package) and substitute mroz_rep.gdt for
mroz87.gdt. It turns out that the results are qualitatively very similar.
7.8 The representation of dates
Up to this point all the data we have considered have been cross-sectional. In the following sections
we discuss data that have a time dimension, and before proceeding it may be useful to say some-
thing about the representation of dates. Gretl takes the ISO 8601 standard as its reference point
but provides mean of converting dates provided in other formats; it also oers a set of calendrical
functions for manipulating dates (isodate, isoconv, epochday and others).
ISO 8601 recognizes two formats for daily dates, extended and basic. In both formats dates
are given as 4-digit year, 2-digit month and 2-digit day, in that order. In extended format a dash
is inserted between the eldsas in 2013-10-21 or more generally YYYY-MM-DDwhile in basic
format the elds are run together (YYYYMMDD). Extended format is more easily parsed by human
readers while basic format is more suitable for computer processing, since one can apply ordinary
arithmetic to compare dates as equal, earlier or later. The standard also recognizes YYYY-MM as
representing year and month, e.g. 2010-11 for November 2010,
2
as well as a plain four-digit number
for year alone.
One problem for economists is that the quarter is not a period covered by ISO 8601. This could
be presented by YYYY-Q (with only one digit following the dash) but in gretl output we in fact use a
colon, as in 2013:2 for the second quarter of 2013. (For printed output of months gretl also uses
a colon, as in 2013:06. A diculty with following ISO here is that in a statistical context a string
such as 1980-10 may look more like a subtraction than a date.) Anyway, at present we are more
interested in the parsing of dates on input rather than in what gretl prints. And in that context note
that excess precision is acceptable: a month may be represented by its rst day (e.g. 2005-05-01
for May, 2005), and a quarter may be represented by its rst month and day (2005-07-01 for the
third quarter of 2005).
2
The form YYYYMM is not recognized for year and month.
Some additional points regarding dates will be taken up as they become relevant in practical cases
of joining data.
7.9 Time-series data
Suppose our left-hand dataset is recognized by gretl as time series with a supported frequency
(annual, quarterly, monthly, weekly, daily or hourly). This will be the case if the original data were
read from a le that contained suitable time or date information, or if a time-series interpretation
has been imposed using either the setobs command or its GUI equivalent. Thenapart, perhaps,
from some very special casesjoining additional data is bound to involve matching observations
by time-period.
In this case, contrary to the cross-sectional case, the inner dataset has a natural ordering which
gretl is aware of; hence, no inner key is required. All we need is a means of identifying the period
on the right and this is why, in a time-series context, well refer to the outer key as the time key.
Most likely, this information will appear in a single column in the outer data le, often but not
always the rst column.
The join command provides a simple (but limited) default for extracting period information from
the outer data le, plus an option that can be used if the default is not applicable, as follows.
The default assumptions are: (1) the time key appears in the rst column; (2) the heading
of this column is either left blank or is one of obs, date, year, period, observation, or
observation_date (on a case-insensitive comparison); and (3) the time format conforms
to ISO 8601 where applicable (extended daily date format YYYY-MM-DD, monthly format
YYYY-MM, or annual format YYYY).
If dates do not appear in the rst column of the outer le, or if the column heading or format
is not as just described, the --tkey option can be used to indicate which column should be
used and/or what format should be assumed.
Setting the time-key column and/or format
The --tkey option requires a parameter holding the name of the column in which the time key
is located and/or a string specifying the format in which dates/times are written in the time-key
column. This parameter should be enclosed in double-quotes. If both elements are present they
should be separated by a comma; if only a format is given it should be preceded by a comma. Some
examples:
--tkey="Period,%m/%d/%Y"
--tkey="Period"
--tkey="obsperiod"
--tkey=",%Ym%m"
The rst of these applies if Period is not the rst column on the right, and dates are given in the
US format of month, day, year, separated by slashes. The second implies that although Period is
not the rst column, the date format is ISO 8601. The third again implies that the date format is
OK; here the name is required even if obsperiod is the rst column since this heading is not one
recognized by gretls heuristic. The last example implies that dates are in the rst column (with
one of the recognized headings), but are given in the non-standard format year, m, month.
The date format string should be composed using the codes employed by the POSIX function
strptime; Table 7.2 contains a list of the most relevant codes.
3
3
The %q code for quarter is not present in strptime; it is added for use with join since quarterly data are common
in macroeconomics.
Code Meaning
%% The % character.
%b The month name according to the current locale, either abbreviated
or in full.
%C The century number (099).
%d The day of month (131).
%D Equivalent to %m/%d/%y. (This is the American style date, very con-
fusing to non-Americans, especially since %d/%m/%y is widely used in
Europe. The ISO 8601 standard format is %Y-%m-%d.)
%H The hour (023).
%j The day number in the year (1366).
%m The month number (112).
%n Arbitrary whitespace.
%q The quarter (14).
%w The weekday number (06) with Sunday = 0.
%y The year within century (099). When a century is not otherwise spec-
ied, values in the range 6999 refer to years in the twentieth century
(19691999); values in the range 0068 refer to years in the twenty-
rst century (20002068).
%Y The year, including century (for example, 1991).
Table 7.2: Date format codes
Example: daily stock prices
We show below the rst few lines of a le named IBM.csv containing stock-price data for IBM
corporation.
Date,Open,High,Low,Close,Volume,Adj Close
2013-08-02,195.50,195.50,193.22,195.16,3861000,195.16
2013-08-01,196.65,197.17,195.41,195.81,2856900,195.81
2013-07-31,194.49,196.91,194.49,195.04,3810000,195.04
Note that the data are in reverse time-series orderthat wont matter to join, the data can appear
in any order. Also note that the rst column is headed Date and holds daily dates as ISO 8601
extended. That means we can pull the data into gretl very easily. In the following fragment we
create a suitably dimensioned empty daily dataset then rely on the default behavior of join with
time-series data to import the closing stock price.
nulldata 500
setobs 5 2012-01-01
join IBM.csv Close
To make explicit what were doing, we could accomplish exactly the same using the --tkey option:
join IBM.csv Close --tkey="Date,%Y-%m-%d"
Example: OECD quarterly data
Table 7.3 shows an excerpt from a CSV le provided by the OECD statistical site (stat.oecd.org)
in response to a request for GDP at constant prices for several countries.
4
4
Retrieved 2013-08-05. The OECD les in fact contain two leading columns with very long labels; these are irrelevant
to the present example and can be omitted without altering the sample script.
Frequency,Period,Country,Value,Flags
"Quarterly","Q1-1960","France",463876.148126845,E
"Quarterly","Q1-1960","Germany",768802.119278467,E
"Quarterly","Q1-1960","Italy",414629.791450547,E
"Quarterly","Q1-1960","United Kingdom",578437.090291889,E
...
Table 7.3: Example of CSV le as provided by the OECD statistical website
This is an instance of data in what we call atomic format, that is, a format in which each line of the
outer le contains a single data-point and extracting data mainly requires ltering the appropriate
lines. The outer time key is under the Period heading, and has the format Q<quarter>-<year>.
Assuming that the le in Table 7.3 has the name oecd.csv, the following script reconstructs the
time series of Gross Domestic Product for several countries:
nulldata 220
setobs 4 1960:1
join oecd.csv FRA --tkey="Period,Q%q-%Y" --data=Value --filter="Country==\"France\""
join oecd.csv GER --tkey="Period,Q%q-%Y" --data=Value --filter="Country==\"Germany\""
join oecd.csv ITA --tkey="Period,Q%q-%Y" --data=Value --filter="Country==\"Italy\""
join oecd.csv UK --tkey="Period,Q%q-%Y" --data=Value --filter="Country==\"United Kingdom\""
Note the use of the format codes %q for the quarter and %Y for the 4-digit year. A touch of elegance
could have been added by storing the invariant part of the join command into a string and then
using string substitution, as in
sprintf optstr "--tkey=\"Period,Q%%q-%%Y\" --data=Value"
join oecd.csv FRA @optstr --filter="Country==\"France\""
join oecd.csv GER @optstr --filter="Country==\"Germany\""
join oecd.csv ITA @optstr --filter="Country==\"Italy\""
join oecd.csv UK @optstr --filter="Country==\"United Kingdom\""
If one were importing a large number of such series it might be worth rewriting the sequence of
joins as a loop, as in
sprintf optstr "--tkey=\"Period,Q%%q-%%Y\" --data=Value"
sprintf countries "France Germany Italy \"United Kingdom\""
sprintf vnames "FRA GER ITA UK"
loop foreach i @countries
string vname = strsplit(vnames, i)
join oecd.csv @vname @optstr --filter="Country==\"$i\""
endloop
7.10 Special handling of time columns
When dealing with straight time series data the tkey mechanism described above should suce
in almost all cases. In some contexts, however, time enters the picture in a more complex way;
examples include panel data (see section 7.11) and so-called realtime data (see chapter 8). To handle
such cases join provides the --tconvert option. This can be used to select certain columns in
the right-hand data le for special treatment: strings representing dates in these columns will be
converted to numerical values: 8-digit numbers on the pattern YYYYMMDD (ISO basic daily format).
Once dates are in this form it is easy to use them in key-matching or ltering.
By default it is assumed that the strings in the selected columns are in ISO extended format,
YYYY-MM-DD. If that is not the case you can supply a time-format string using the --tconv-fmt
option. The format string should be written using the codes shown in Table 7.2.
Here are some examples:
# select one column for treatment
--tconvert=start_date
# select two columns for treatment
--tconvert="start_date,end_date"
# specify US-style daily date format
--tconv-fmt="%m/%d/%Y"
# specify quarterly date-strings (as in 2004q1)
--tconv-fmt="%Yq%q"
Some points to note:
If a specied column is not selected for a substantive role in the join operation (as data to be
imported, as a key, or as an auxiliary variable for use in aggregation) the column in question
is not read and so no conversion is carried out.
If a specied column contains numerical rather than string values, no conversion is carried
out.
If a string value in a selected column fails parsing using the relevant time format (user-
specied or default), the converted value is NA.
On successful conversion, the output is always in daily-date form as stated above. If you
specify a monthly or quarterly time format, the converted date is the rst day of the month
or quarter.
7.11 Panel data
In section 7.9 we gave an example of reading quarterly GDP data for several countries from an
OECD le. In that context we imported each countrys data as a distinct time-series variable. Now
suppose we want the GDP data in panel format instead (stacked time series). How can we do this
with join?
As a reminder, heres what the OECD data look like:
Frequency,Period,Country,Value,Flags
and so on. If we have four countries and quarterly observations running from 1960:1 to 2013:2 (T
= 214 quarters) we might set up our panel workspace like this:
scalar N = 4
scalar T = 214
scalar NT = N*T
nulldata NT --preserve
setobs T 1.1 --stacked-time-series
The relevant outer keys are obvious: Country for the country and Period for the time period. Our
task is now to construct matching keys in the inner dataset. This can be done via two panel-specic
options to the setobs command. Lets work on the time dimension rst:
setobs 4 1960:1 --panel-time
series quarter = $obsdate
This variant of setobs allows us to tell gretl that time in our panel is quarterly, starting in the
rst quarter of 1960. Having set that, the accessor $obsdate will give us a series of 8-digit dates
representing the rst day of each quarter19600101, 19600401, 19600701, and so on, repeating
for each country. As we explained in section 7.10, we can use the --tconvert option on the outer
series Period to get exactly matching values (in this case using a format of Q%q-%Y for parsing the
Period values).
Now for the country names:
string cstrs
sprintf cstrs "France Germany Italy \"United Kingdom\""
setobs country cstrs --panel-groups
Here we write into the string cstrs the names of the countries, using escaped double-quotes to
handle the space in United Kingdom, then pass this string to setobs with the --panel-groups
option, preceded by the identier country. This asks gretl to construct a string-valued series
named country, in which each name will repeat T times.
Were now ready to join. Assuming the OECD le is named oecd.csv we do
join oecd.csv GDP --data=Value \
--ikey=country,quarter --okey=Country,Period \
--tconvert=Period --tconv-fmt="Q%q-%Y"
Other input formats
The OECD le discussed above is in the most convenient format for join, with one data-point per
line. But sometimes we may want to make a panel from a data le structured like this:
# Real GDP
Period,France,Germany,Italy,"United Kingdom"
"Q1-1960",463863,768757,414630,578437
"Q2-1960",465605,782438,420715,572853
"Q3-1960",469091,809484,426894,581252
"Q4-1960",474651,817758,427221,584779
"Q1-1961",482285,826031,442528,594684
...
Call this le side_by_side.csv. Assuming the same initial set-up as above, we can panelize the
data by setting the sample to each countrys time series in turn and importing the relevant column.
The only point to watch here is that the string United Kingdom, being a column heading, will
become United_Kingdom on importing (see section 7.2) so well need a slightly dierent set of
country strings.
sprintf cstrs "France Germany Italy United_Kingdom"
setobs country cstrs --panel-groups
loop foreach i @cstrs --quiet
smpl country=="$i" --restrict --replace
join side_by_side.csv GDP --data=$i \
--ikey=quarter --okey=Period \
--tconvert=Period --tconv-fmt="Q%q-%Y"
endloop
smpl full
If our working dataset and the outer data le are dimensioned such that there are just as many
time-series observations on the right as there are time slots on the leftand the observations
on the right are contiguous, in chronological order, and start on the same date as the working
datasetwe could dispense with the key apparatus and just use the rst line of the join command
shown above. However, in general it is safer to use keys to ensure that the data end up in correct
registration.
Appendix: the full Mroz data script
# start with everybody; get gender, age and a few other variables
# directly while were at it
open SHIW/carcom10.csv --cols=1,2,3,4,9,10,29,41
# subsample on married women between the ages of 30 and 60
smpl SEX==2 && ETA>=30 && ETA<=60 && STACIV==1 --restrict
# for simplicity, restrict to heads of households and their spouses
smpl PARENT<3 --restrict
# rename the age and education variables for compatibility; compute
# the "city" dummy and finally save the reduced base dataset
rename ETA WA
rename STUDIO WE
series CIT = (ACOM4C>2)
store mroz_rep.gdt
# make a temp file holding annual hours worked per job
open SHIW/allb1.csv --cols=1,2,8,11 --quiet
series HRS = misszero(ORETOT) * 52 * misszero(MESILAV)/12
store HRS.csv NQUEST NORD HRS
# reopen the base dataset and begin drawing assorted data in
open mroz_rep.gdt
# womens annual hours (summed across jobs)
join HRS.csv WHRS --ikey=NQUEST,NORD --data=HRS --aggr=sum
WHRS = misszero(WHRS)
# labor force participation
LFP = WHRS > 0
# work experience: ETALAV = age when started first job
join SHIW/lavoro.csv ETALAV --ikey=NQUEST,NORD
series AX = misszero(WA - ETALAV)
# womens hourly wages
join SHIW/rper10.csv YL --ikey=NQUEST,NORD --aggr=sum
join SHIW/rper10.csv YM --ikey=NQUEST,NORD --aggr=sum
series WW = LFP ? (YL + YM)/WHRS : 0
# family income (Y = net disposable income)
join SHIW/rfam10.csv FAMINC --ikey=NQUEST --data=Y
# get data on children using the "count" method
join SHIW/carcom10.csv KIDS --ikey=NQUEST --aggr=count --filter="ETA<=18"
join SHIW/carcom10.csv KL6 --ikey=NQUEST --aggr=count --filter=ETA<6
series K618 = KIDS - KL6
# data on husbands: we first construct an auxiliary inner key for
# husbands, using the little trick of subsampling the inner dataset
#
# for women who are household heads
# for women who are not household heads
smpl full
# add husbands data via the newly-added secondary inner key
join SHIW/carcom10.csv HA --ikey=NQUEST,H_ID --okey=NQUEST,NORD --data=ETA
join SHIW/carcom10.csv HE --ikey=NQUEST,H_ID --okey=NQUEST,NORD --data=STUDIO
join HRS.csv HHRS --ikey=NQUEST,H_ID --okey=NQUEST,NORD --data=HRS --aggr=sum
HHRS = misszero(HHRS)
# final cleanup begins
# recode educational attainment as years of education
matrix eduyrs = {0, 5, 8, 11, 13, 16, 18, 21}
series WE = replace(WE, seq(1,8), eduyrs)
series HE = replace(HE, seq(1,8), eduyrs)
# cut some cruft
delete SEX STACIV KIDS YL YM PARENT H_ID ETALAV
# add some labels for the series
setinfo LFP -d "1 if woman worked in 2010"
setinfo WHRS -d "Wifes hours of work in 2010
setinfo KL6 -d "Number of children less than 6 years old in household"
setinfo K618 -d "Number of children between ages 6 and 18 in household"
setinfo WA -d "Wifes age"
setinfo WE -d "Wifes educational attainment, in years"
setinfo WW -d "Wifes average hourly earnings, in 2010 euros"
setinfo HHRS -d "Husbands hours worked in 2010"
setinfo HA -d "Husbands age"
setinfo HE -d "Husbands educational attainment, in years"
setinfo FAMINC -d "Family income, in 2010 euros"
setinfo AX -d "Actual years of wifes previous labor market experience"
setinfo CIT -d "1 if live in large city"
# save the final product
store mroz_rep.gdt
Chapter 8
Realtime data
8.1 Introduction
As of gretl version 1.9.13 the join command (see chapter 7) has been enhanced to deal with so-
called realtime datasets in a straightforward manner. Such datasets contain information on when
the observations in a time series were actually published by the relevant statistical agency and how
they have been revised over time. Probably the most popular sources of such data are the Alfred
online database at the St. Louis Fed (http://alfred.stlouisfed.org/) and the OECDs StatEx-
tracts site, http://stats.oecd.org/. The examples in this chapter deal with les downloaded
from these sources, but should be easy to adapt to les with a slightly dierent format.
As already stated, join requires a column-oriented plain text le, where the columns may be sepa-
rated by commas, tabs, spaces or semicolons. Alfred and the OECD provide the option to download
realtime data in this format (tab-delimited les from Alfred, comma-delimited from the OECD). If
you have a realtime dataset in a spreadsheet le you must export it to a delimited text le before
using it with join.
Representing revision histories is more complex than just storing a standard time series, because
for each observation period you have in general more than one published value over time, along
with the information on when each of these values were valid or current. Sometimes this is repre-
sented in spreadsheets with two time axes, one for the observation period and another one for the
publication date or vintage. The lled cells then form an upper triangle (or a guillotine blade
shape, if the publication dates do not reach back far enough to complete the triangle). This format
can be useful for giving a human reader an overview of realtime data, but it is not optimal for
automatic processing; for that purpose atomic format is best.
8.2 Atomic format for realtime data
What we are calling atomic format is exactly the format used by Alfred if you choose the option
Observations by Real-Time Period, and by the OECD if you select all editions of a series for
download as plain text (CSV).
1
A le in this format contains one actual data-point per line, together
with associated metadata. This is illustrated in Table 8.1, where we show the rst three lines from
an Alfred le and an OECD le (slightly modied).
2
Consider the rst data line in the Alfred le: in the observation_date column we nd 1960-01-01,
indicating that the data-point on this line, namely 112.0, is an observation or measurement (in this
case, of the US index of industrial production) that refers to the period starting on January 1st
1960. The realtime_start_date value of 1960-02-16 tells us that this value was published on
February 16th 1960, and the realtime_end_date value says that this vintage remained current
through March 15th 1960. On the next day (as we can see from the following line) this data-point
was revised slightly downward to 111.0.
Daily dates in Alfred les are given in ISO extended format, YYYY-MM-DD, but below we describe
how to deal with dierently formatted dates. Note that daily dates are appropriate for the last
1
If you choose to download in Excel format from OECD you get a le in the triangular or guillotine format mentioned
above.
2
In the Alfred le we have used commas rather than tabs as the column delimiter; in the OECD example we have
shortened the name in the Variable column.
58
Chapter 8. Realtime data 59
Alfred: monthly US industrial production
observation_date,INDPRO,realtime_start_date,realtime_end_date
1960-01-01,112.0000,1960-02-16,1960-03-15
1960-01-01,111.0000,1960-03-16,1961-10-15
OECD: monthly UK industrial production
Country,Variable,Frequency,Time,Edition,Value,Flags
"United Kingdom","INDPRO","Monthly","Jan-1990","February 1999",100,
"United Kingdom","INDPRO","Monthly","Feb-1990","February 1999",99.3,
Table 8.1: Variant atomic formats for realtime data
two columns, which jointly record the interval over which a given data vintage was current. Daily
dates might, however, be considered overly precise for the rst column, since the data period may
well be the year, quarter or month (as it is in fact here). However, following Alfreds practice it is
acceptable to specify a daily date, indicating the rst day of the period, even for non-daily data.
3
Compare the rst data line of the OECD example. Theres a greater amount of leading metadata,
which is left implicit in the Alfred le. Here Time is the equivalent of Alfreds observation_date,
and Edition the equivalent of Alfreds realtime_start_date. So we read that in February 1999
a value of 100 was current for the UK index of industrial production for January 1990, and from
the next line we see that in the same vintage month a value of 99.3 was current for industrial
production in February 1990.
Besides the dierent names and ordering of the columns, there are a few more substantive dier-
ences between Alfred and OECD les, most of which are irrelevant for join but some of which are
(possibly) relevant.
The rst (irrelevant) dierence is the ordering of the lines. It appears (though were not sure how
consistent this is) that in Alfred les the lines are sorted by observation date rst and then by
publication dateso that all revisions of a given observation are grouped togetherwhile OECD
les are sorted rst by revision date (Edition) and then by observation date (Time). If we want the
next revision of UK industrial production for January 1990 in the OECD le we have to scan down
several lines until we nd
"United Kingdom","INDPRO","Monthly","Jan-1990","March 1999",100,
This dierence is basically irrelevant because join can handle the case where the lines appear in
random order, although some operations can be coded more conveniently if were able to assume
chronological ordering (either on the Alfred or the OECD pattern, it doesnt matter).
The second (also irrelevant) dierence is that the OECD seems to include periodic Edition lines
even when there is no change from the previous value (as illustrated above, where the UK industrial
production index for January 1990 is reported as 100 as of March 1999, the same value that we saw
to be current in February 1999), while Alfred reports a new value only when it diers from what
was previously current.
A third dierence lies in the dating of the revisions or editions. As we have seen, Alfred gives
a specic daily date while (in the UK industrial production le at any rate), the OECD just dates
each edition to a month. This is not necessarily relevant for join, but it does raise the question of
whether the OECD might date revisions to a ner granularity in some of their les, in which case
one would have to be on the lookout for a dierent date format.
The nal dierence is that Alfred supplies an end date for each data vintage while the OECD
3
Notice that this implies that in the Alfred example it is not clear without further information whether the observation
period is the rst quarter of 1960, the month January 1960, or the day January 1st 1960. However, we assume that this
information is always available in context.
supplies only a starting date. But there is less to this dierence than meets the eye: according to
the Alfred webmaster, by design, a new vintage must start immediately following (the day after)
the lapse of the old vintageso the end date conveys no independent information.
4
8.3 More on time-related options
Before we get properly started it is worth saying a little more about the --tkey and --tconvert
options to join (rst introduced in section 7.10), as they apply in the case of realtime data.
When youre working with regular time series data tkey is likely to be useful while tconvert is
unlikely to be applicable (see section 7.9). On the other hand, when youre working with panel data
tkey is denitely not applicable but tconvert may well be helpful (section 7.11). When working
with realtime data, however, depending on the task in hand both options may be useful. You will
likely need tkey; you may well wish to select at least one column for tconvert treatment; and in
fact you may want to name a given column in both contextsthat is, include the tkey variable
among the tconvert columns.
Why might this make sense? Well, think of the --tconvert option as a preprocessing directive:
it asks gretl to convert date strings to numerical values (8-digit ISO basic dates) at source, as they
are read from the outer datale. The --tkey option, on the other hand, singles out a column as
the one to use for matching rows with the inner dataset. So you would want to name a column in
both roles if (a) it should be used for matching periods and also (b) it is desirable to have the values
from this column in numerical form, most likely for use in ltering.
As we have seen, you can supply specic formats in connection with both tkey and tconvert (in
the latter case via the companion option --tconv-fmt) to handle the case where the date strings on
the right are not ISO-friendly at source. This raises the question of how the format specications
work if a given column is named under both options. Here are the rules that gretl applies:
1. If a format is given with the --tkey option it always applies to the tkey column alone; and
for that column it overrides any format given via the --tconv-fmt option.
2. If a format is given via tconv-fmt it is assumed to apply to all the tconvert columns, unless
this assumption is overriden by rule 1.
8.4 Getting a certain data vintage
The most common application of realtime data is to travel back in time and retrieve the data that
were current as of a certain date in the past. This would enable you to replicate a forecast or other
statistical result that could have been produced at that date.
For example, suppose we are interested in a variable of monthly frequency named INDPRO, realtime
data on which is stored in an Alfred le named INDPRO.txt, and we want to check the status quo
as of June 15th 2011.
If we dont already have a suitable dataset into which to import the INDPRO data, our rst steps will
be to create an appropriately dimensioned empty dataset using the nulldata command and then
specify its time-series character via setobs, as in
nulldata 132
setobs 12 2004:01
For convenience we can put the name of our realtime le into a string variable. On Windows this
might look like
4
Email received from Travis May of the Federal Reserve Bank of St. Louis, 2013-10-17. This closes o the possibility
that a given vintage could lapse or expire some time before the next vintage becomes available, hence giving rise to a
hole in an Alfred realtime le.
string fname = "C:/Users/yourname/Downloads/INDPRO.txt"
We can then import the data vintage 2011-06-15 using join, arbitrarily choosing the self-explanatory
identier ip_asof_20110615.
join @fname ip_asof_20110615 --tkey=observation_date --data=INDPRO \
--tconvert="realtime_start_date" \
--filter="realtime_start_date<=20110615" --aggr=max(realtime_start_date)
Here some detailed explanations of the various options are warranted:
The --tkey option species the column which should be treated as holding the observation
period identiers to be matched against the periods in the current gretl dataset.
5
The more
general form of this option is --tkey="colname,format" (note the double quotes here), so
if the dates do not come in standard format, we can tell gretl how to parse them by using
the appropriate conversion speciers as shown in Table 7.2. For example, here we could have
written --tkey="observation_date,%Y-%m-%d".
Next, --data=INDPRO tells gretl that we want to retrieve the entries stored in the column
named INDPRO.
As explained in section 7.10 the --tconvert option selects certain columns in the right-hand
data le for conversion from date strings to 8-digit numbers on the pattern YYYYMMDD. Well
need this for the next step, ltering, since the transformation to numerical values makes
it possible to perform basic arithmetic on dates. Note that since date strings in Alfred les
conform to gretls default assumption it is not necessary to use the --tconv-fmt option here.
The --filter option specication in combination with the subsequent --aggr aggregation
treatment is the central piece of our data retrieval; notice how we use the date constant
20110615 in ISO basic form to do numerical comparisons, and how we perform the numerical
max operation on the converted column realtime_start_date. It would also have been
possible to predene a scalar variable, as in
vintage = 20110615
and then use vintage in the join command instead. Here we tell join that we only want to
extract those publications that (1) already appeared before (and including) June 15th 2011,
and (2) were not yet obsoleted by a newer release.
6
As a result, your dataset will now contain a time series named ip_asof_20110615 with the values
that a researcher would have had available on June 15th 2011. Of course, all values for the observa-
tions after June 2011 will be missing (and probably a few before that, too), because they only have
become available later on.
8.5 Getting the n-th release for each observation period
For some purposes it may be useful to retrieve the n-th published value of each observation, where
n is a xed positive integer, irrespective of when each of these n-th releases was published. Sup-
pose we are interested in the third release, then the relevant join command becomes:
join @fname ip_3rdpub --tkey=observation_date --data=INDPRO --aggr="seq:3"
5
Strictly speaking, using --tkey is unnecessary in this example because we could just have relied on the default,
which is to use the rst column in the source le for the periods. However, being explicit is often a good idea.
6
By implementing the second condition through the max aggregation on the realtime_start_date column alone,
without using the realtime_end_date column, we make use of the fact that Alfred les cannot have holes as explained
before.
Since we do not need the realtime_start_date information for this retrieval, we have dropped
the --tconvert option here. Note that this formulation assumes that the source le is ordered
chronologically, otherwise using the option --aggr="seq:3", which retrieves the third value from
each sequence of matches, could have yielded a result dierent from the one intended. However,
this assumption holds for Alfred les and is probably rather safe in general.
The values of the variable imported as ip_3rdpub in this way were published at dierent dates,
so the variable is eectively a mix of dierent vintages. Depending on the type of variable, this
may also imply drastic jumps in the values; for example, index numbers are regularly re-based
to dierent base periods. This problem also carries over to ination-adjusted economic variables,
where the base period of the price index changes over time. Mixing vintages in general also means
mixing dierent scales in the output, with which you would have to deal appropriately.
7
8.6 Getting the values at a xed lag after the observation period
New data releases may take place on any day of the month, and as we have seen the specic day
of each release is recorded in realtime les from Alfred. However, if you are working with, say,
monthly or quarterly data you may sometimes want to adjust the granularity of your realtime axis
to a monthly or quarterly frequency. For example, in order to analyse the data revision process for
monthly industrial production you might be interested in the extent of revisions between the data
available two and three months after each observation period.
This is a relatively complicated task and there is more than one way of accomplishing it. Either you
have to make several passes through the outer dataset or you need a sophisticated lter, written
as a hansl function. Either way you will want to make use of some of gretls built-in calendrical
functions.
Well assume that a suitably dimensioned workspace has been set up as described above. Given
that, the key ingredients of the join are a ltering function which well call rel_ok (for release is
OK) and the join command which calls it. Heres the function:
function series rel_ok (series obsdate, series reldate, int p)
series y_obs, m_obs, y_rel, m_rel
# get year and month from observation date
isoconv(obsdate, &y_obs, &m_obs)
# get year and month from release date
isoconv(reldate, &y_rel, &m_rel)
# find the delta in months
series dm = (12*y_rel + m_rel) - (12*y_obs + m_obs)
# and implement the filter
return dm <= p
end function
And heres the command:
scalar lag = 3 # choose your fixed lag here
join @fname ip_plus3 --data=INDPRO --tkey=observation_date \
--tconvert="observation_date,realtime_start_date" \
--filter="rel_ok(observation_date, realtime_start_date, lag)" \
--aggr=max(realtime_start_date)
Note that we use --tconvert to convert both the observation date and the realtime start date (or
release date) to 8-digit numerical values. Both of these series are passed to the lter, which uses the
7
Some user-contributed functions may be available that address this issue, but it is beyond our scope here. Another
even more complicated issue in the realtime context is that of benchmark revisions applied by statistical agencies,
where the underlying denition or composition of a variable changes on some date, which goes beyond a mere rescaling.
However, this type of structural change is not, in principle, a feature of realtime data alone, but applies to any time-series
data.
built-in function isoconv to extract year and month. We can then calculate dm, the delta months
since the observation date, for each release. The lter condition is that this delta should be no
greater than the specied lag, p.
8
This lter condition may be satised by more than one release, but only the latest of those will
actually be the vintage that was current at the end of the n-th month after the observation period,
so we add the option --aggr=max(realtime_start_date). If instead you want to target the
release at the beginning of the n-th month you would have to use a slightly more complicated lter
function.
An illustration
Figure 8.1 shows four time series for the monthly index of US industrial production from October
2005 to June 2009: the value as of rst publication plus the values current 3, 6 and 12 months out
from the observation date.
9
From visual inspection it would seem that over much of this period
the Federal reserve was fairly consistently overestimating industrial production at rst release and
shortly thereafter, relative to the gure they arrived at with a lag of a year.
The script that produced this Figure is shown in full in Example 8.1. Note that in this script we are
using a somewhat more ecient version of the rel_ok function shown above, where we pass the
required series arguments in pointer form to avoid having to copy them (see chapter 13).
94
96
98
100
102
104
106
108
110
112
114
116
2006 2007 2008 2009
First publication
Plus 3 months
Plus 6 months
Plus 12 months
Figure 8.1: Successive revisions to US industrial production
8.7 Getting the revision history for an observation
For our nal example we show how to retrieve the revision history for a given observation (again
using Alfred data on US industrial production). In this exercise we are switching the time axis: the
observation period is a xed point and time is vintage time.
A suitable script is shown in Example 8.2. We rst select an observation to track (January 1970).
We start the clock in the following month, when a data-point for this period was rst published,
8
The lter is written on the assumption that the lag is expressed in months; on that understanding it could be used
with annual or quarterly data as well as monthly. The idea could be generalized to cover weekly or daily data without
much diculty.
9
Why not a longer series? Because if we try to extend it in either direction we immediately run into the index re-basing
problem mentioned in section 8.5, with big (staggered) leaps downward in all the series.
Example 8.1: Retrieving successive realtime lags of US industrial production
function series rel_ok (series *obsdate, series *reldate, int p)
series y_obs, m_obs, y_rel, m_rel
isoconv(obsdate, &y_obs, &m_obs)
isoconv(reldate, &y_rel, &m_rel)
series dm = (12*y_rel + m_rel) - (12*y_obs + m_obs)
return dm <= p
end function
nulldata 45
setobs 12 2005:10
string fname = "INDPRO.txt"
# initial published values
join @fname firstpub --data=INDPRO --tkey=observation_date \
--tconvert=realtime_start_date --aggr=min(realtime_start_date)
# plus 3 months
join @fname plus3 --data=INDPRO --tkey=observation_date \
--filter="rel_ok(&observation_date, &realtime_start_date, 3)" \
# plus 6 months
# plus 12 months
setinfo firstpub --graph-name="First publication"
setinfo plus3 --graph-name="Plus 3 months"
gnuplot firstpub plus3 plus6 plus12 --time --with-lines \
--output=realtime.pdf { set key left bottom; }
and let it run to the end of the vintage history (in this le, March 2013). Our outer time key is
the realtime start date and we lter on observation date; we name the imported INDPRO values as
ip_jan70. Since it sometimes happens that more than one revision occurs in a given month we
need to select an aggregation method: here we choose to take the last revision in the month.
Recall from section 8.2 that Alfred records a new revision only when the data-point in question
actually changes. This means that our imported series will contain missing values for all months
when no real revision took place. However, we can apply a simple autoregressive procedure to ll
in the data: each missing value equals the prior non-missing value.
Figure 8.2 displays the revision history. Over this sample period the periodic re-basing of the index
overshadows amendments due to accrual of new information.
Example 8.2: Retrieving a revision history
# choose the observation to track here (YYYYMMDD)
scalar target = 19700101
nulldata 518 --preserve
setobs 12 1970:02
join INDPRO.txt ip_jan70 --data=INDPRO --tkey=realtime_start_date \
--tconvert=observation_date \
--filter="observation_date==target" --aggr=seq:-1
ip_jan70 = ok(ip_jan70) ? ip_jan70 : ip_jan70(-1)
20
40
60
80
100
120
140
160
180
1970 1975 1980 1985 1990 1995 2000 2005 2010
i
p
_
j
a
n
7
0
Figure 8.2: Vintages of the index of US industrial production for January 1970
Chapter 9
Special functions in genr
9.1 Introduction
The genr command provides a exible means of dening new variables. It is documented in the
Gretl Command Reference. This chapter oers a more expansive discussion of some of the special
functions available via genr and some of the ner points of the command.
9.2 Long-run variance
As is well known, the variance of the average of T random variables x
1
, x
2
, . . . , x
T
with equal vari-
ance
2
equals
2
/T if the data are uncorrelated. In this case, the sample variance of x
t
over the
sample size provides a consistent estimator.
If, however, there is serial correlation among the x
t
s, the variance of

X = T
1
T
t=1
x
t
must be
estimated dierently. One of the most widely used statistics for this purpose is a nonparametric
kernel estimator with the Bartlett kernel dened as

2
(k) = T
1
Tk
_
t=k
_
_
k
_
i=k
w
i
(x
t

X)(x
ti

X)
_
_
, (9.1)
where the integer k is known as the window size and the w
i
terms are the so-called Bartlett weights,
dened as w
i
= 1
i
k+1
. It can be shown that, for k large enough,
2
(k)/T yields a consistent
estimator of the variance of

X.
Gretl implements this estimator by means of the function lrvar(), which takes two arguments:
the series whose long-run variance must be estimated and the scalar k. If k is negative, the popular
choice T
1/3
is used.
9.3 Cumulative densities and p-values
The two functions cdf and pvalue provide complementary means of examining values fromseveral
probability distributions: the standard normal, Students t,
2
, F, gamma, and binomial. The syntax
of these functions is set out in the Gretl Command Reference; here we expand on some subtleties.
The cumulative density function or CDF for a random variable is the integral of the variables
density from its lower limit (typically either or 0) to any specied value x. The p-value (at
least the one-tailed, right-hand p-value as returned by the pvalue function) is the complementary
probability, the integral from x to the upper limit of the distribution, typically +.
In principle, therefore, there is no need for two distinct functions: given a CDF value p
0
you could
easily nd the corresponding p-value as 1 p
0
(or vice versa). In practice, with nite-precision
computer arithmetic, the two functions are not redundant. This requires a little explanation. In
gretl, as in most statistical programs, oating point numbers are represented as doubles
double-precision values that typically have a storage size of eight bytes or 64 bits. Since there are
only so many bits available, only so many oating-point numbers can be represented: doubles do
not model the real line. Typically doubles can represent numbers over the range (roughly) :1.7977
10
308
, but only to about 15 digits of precision.
66
Chapter 9. Special functions in genr 67
Suppose youre interested in the left tail of the
2
distribution with 50 degrees of freedom: youd
like to know the CDF value for x = 0.9. Take a look at the following interactive session:
? genr p1 = cdf(X, 50, 0.9)
Generated scalar p1 (ID 2) = 8.94977e-35
? genr p2 = pvalue(X, 50, 0.9)
Generated scalar p2 (ID 3) = 1
? genr test = 1 - p2
Generated scalar test (ID 4) = 0
The cdf function has produced an accurate value, but the pvalue function gives an answer of 1,
from which it is not possible to retrieve the answer to the CDF question. This may seem surprising
at rst, but consider: if the value of p1 above is correct, then the correct value for p2 is 18.94977
10
35
. But theres no way that value can be represented as a double: that would require over 30
digits of precision.
Of course this is an extreme example. If the x in question is not too far o into one or other tail
of the distribution, the cdf and pvalue functions will in fact produce complementary answers, as
shown below:
? genr p1 = cdf(X, 50, 30)
Generated scalar p1 (ID 2) = 0.0111648
? genr p2 = pvalue(X, 50, 30)
Generated scalar p2 (ID 3) = 0.988835
? genr test = 1 - p2
Generated scalar test (ID 4) = 0.0111648
But the moral is that if you want to examine extreme values you should be careful in selecting the
function you need, in the knowledge that values very close to zero can be represented as doubles
while values very close to 1 cannot.
9.4 Retrieving internal variables
The genr command provides a means of retrieving various values calculated by the program in
the course of estimating models or testing hypotheses. The variables that can be retrieved in this
way are listed in the Gretl Command Reference; here we say a bit more about the special variables
$test and $pvalue.
These variables hold, respectively, the value of the last test statistic calculated using an explicit
testing command and the p-value for that test statistic. If no such test has been performed at the
time when these variables are referenced, they will produce the missing value code. The explicit
testing commands that work in this way are as follows: add (joint test for the signicance of vari-
ables added to a model); adf (Augmented DickeyFuller test, see below); arch (test for ARCH); chow
(Chow test for a structural break); coeffsum (test for the sum of specied coecients); cusum (the
HarveyCollier t-statistic); kpss (KPSS stationarity test, no p-value available); lmtest (see below);
meantest (test for dierence of means); omit (joint test for the signicance of variables omitted
from a model); reset (Ramseys RESET); restrict (general linear restriction); runs (runs test for
randomness); testuhat (test for normality of residual); and vartest (test for dierence of vari-
ances). In most cases both a $test and a $pvalue are stored; the exception is the KPSS test, for
which a p-value is not currently available.
An important point to notice about this mechanismis that the internal variables $test and $pvalue
are over-written each time one of the tests listed above is performed. If you want to reference these
values, you must do so at the correct point in the sequence of gretl commands.
A related point is that some of the test commands generate, by default, more than one test statistic
and p-value; in these cases only the last values are stored. To get proper control over the retrieval
of values via $test and $pvalue you should formulate the test command in such a way that the
result is unambiguous. This comment applies in particular to the adf and lmtest commands.
By default, the adf command generates three variants of the DickeyFuller test: one based
on a regression including a constant, one using a constant and linear trend, and one using a
constant and a quadratic trend. When you wish to reference $test or $pvalue in connection
with this command, you can control the variant that is recorded by using one of the ags
--nc, --c, --ct or --ctt with adf.
By default, the lmtest command (which must follow an OLS regression) performs several
diagnostic tests on the regression in question. To control what is recorded in $test and
$pvalue you should limit the test using one of the ags --logs, --autocorr, --squares or
--white.
As an aid in working with values retrieved using $test and $pvalue, the nature of the test to which
these values relate is written into the descriptive label for the generated variable. You can read the
label for the variable using the label command (with just one argument, the name of the variable),
to check that you have retrieved the right value. The following interactive session illustrates this
point.
? adf 4 x1 --c
Augmented Dickey-Fuller tests, order 4, for x1
sample size 59
unit-root null hypothesis: a = 1
test with constant
model: (1 - L)y = b0 + (a-1)*y(-1) + ... + e
estimated value of (a - 1): -0.216889
test statistic: t = -1.83491
asymptotic p-value 0.3638
P-values based on MacKinnon (JAE, 1996)
? genr pv = $pvalue
Generated scalar pv (ID 13) = 0.363844
? label pv
pv=Dickey-Fuller pvalue (scalar)
9.5 The discrete Fourier transform
The discrete Fourier transform can be best thought of as a linear, invertible transform of a complex
vector. Hence, if x is an n-dimensional vector whose k-th element is x
k
= a
k
+ib
k
, then the output
of the discrete Fourier transform is a vector f = (x) whose k-th element is
f
k
=
n1
_
j=0
e
i(j,k)
x
j
where (j, k) = 2i
jk
n
. Since the transformation is invertible, the vector x can be recovered from
f via the so-called inverse transform
x
k
=
1
n
n1
_
j=0
e
i(j,k)
f
j
.
The Fourier transform is used in many diverse situations on account of this key property: the
convolution of two vectors can be performed eciently by multiplying the elements of their Fourier
transforms and inverting the result. If
z
k
=
n
_
j=1
x
j
y
kj
,
then
(z) = (x) C(y).
That is, (z)
k
= (x)
k
(y)
k
.
For computing the Fourier transform, gretl uses the external library fftw3: see Frigo and Johnson
(2005). This guarantees extreme speed and accuracy. In fact, the CPU time needed to perform
the transform is O(nlogn) for any n. This is why the array of numerical techniques employed in
fftw3 is commonly known as the Fast Fourier Transform.
Gretl provides two matrix functions
1
for performing the Fourier transform and its inverse: fft and
ffti. In fact, gretls implementation of the Fourier transform is somewhat more specialized: the
input to the fft function is understood to be real. Conversely, ffti takes a complex argument and
delivers a real result. For example:
x1 = { 1 ; 2 ; 3 }
# perform the transform
f = fft(a)
# perform the inverse transform
x2 = ffti(f)
yields
x
1
=
_
_
_
_
1
2
3
_
_
_
_
f =
_
_
_
_
6 0
1.5 0.866
1.5 0.866
_
_
_
_
x
2
=
_
_
_
_
1
2
3
_
_
_
_
where the rst column of f holds the real part and the second holds the complex part. In general,
if the input to fft has n columns, the output has 2n columns, where the real parts are stored in
the odd columns and the complex parts in the even ones. Should it be necessary to compute the
Fourier transform on several vectors with the same number of elements, it is numerically more
ecient to group them into a matrix rather than invoking fft for each vector separately.
As an example, consider the multiplication of two polynomials:
a(x) = 1 +0.5x
b(x) = 1 +0.3x 0.8x
2
c(x) = a(x) b(x) = 1 +0.8x 0.65x
2
0.4x
3
The coecients of the polynomial c(x) are the convolution of the coecients of a(x) and b(x);
the following gretl code fragment illustrates how to compute the coecients of c(x):
# define the two polynomials
a = { 1, 0.5, 0, 0 }
b = { 1, 0.3, -0.8, 0 }
# perform the transforms
fa = fft(a)
fb = fft(b)
# complex-multiply the two transforms
fc = cmult(fa, fb)
# compute the coefficients of c via the inverse transform
c = ffti(fc)
Maximum eciency would have been achieved by grouping a and b into a matrix. The computa-
tional advantage is so little in this case that the exercise is a bit silly, but the following alternative
may be preferable for a large number of rows/columns:
# define the two polynomials
a = { 1 ; 0.5; 0 ; 0 }
b = { 1 ; 0.3 ; -0.8 ; 0 }
1
See chapter 15.
# perform the transforms jointly
f = fft(a ~ b)
# complex-multiply the two transforms
fc = cmult(f[,1:2], f[,3:4])
# compute the coefficients of c via the inverse transform
c = ffti(fc)
Traditionally, the Fourier transform in econometrics has been mostly used in time-series analysis,
the periodogram being the best known example. Example script 9.1 shows how to compute the
periodogram of a time series via the fft function.
Example 9.1: Periodogram via the Fourier transform
nulldata 50
# generate an AR(1) process
series e = normal()
series x = 0
x = 0.9*x(-1) + e
# compute the periodogram
scale = 2*pi*$nobs
X = { x }
F = fft(X)
S = sumr(F.^2)
S = S[2:($nobs/2)+1]/scale
omega = seq(1,($nobs/2)) .* (2*pi/$nobs)
omega = omega ~ S
# compare the built-in command
pergm x
print omega
Chapter 10
Gretl data types
10.1 Introduction
Gretl oers the following data types:
scalar holds a single numerical value
series holds n numerical values, where n is the number of observations in the current
dataset
matrix holds a rectangular array of numerical values, of any dimensions
list holds the ID numbers of a set of series
string holds an array of characters
bundle holds a variable number of objects of various types
The numerical values mentioned above are all double-precision oating point numbers.
In this chapter we give a run-down of the basic characteristics of each of these types and also
explain their life cycle (creation, modication and destruction). The list and matrix types, whose
uses are relatively complex, are discussed at greater length in the following two chapters.
10.2 Series
We begin with the series type, which is the oldest and in a sense the most basic type in gretl. When
you open a data le in the gretl GUI, what you see in the main window are the ID numbers, names
(and descriptions, if available) of the series read from the le. All the series existing at any point in
a gretl session are of the same length, although some may have missing values. The variables that
can be added via the items under the Add menu in the main window (logs, squares and so on) are
also series.
For a gretl session to contain any series, a common series length must be established. This is
usually achieved by opening a data le, or importing a series from a database, in which case the
length is set by the rst import. But one can also use the nulldata command, which takes as it
single argument the desired length, a positive integer.
Each series has these basic attributes: an ID number, a name, and of course n numerical values. In
addition a series may have a description (which is shown in the main window and is also accessible
via the labels command), a display name for use in graphs, a record of the compaction method
used in reducing the variables frequency (for time-series data only) and a ag marking the variable
as discrete. These attributes can be edited in the GUI by choosing Edit Attributes (either under the
Variable menu or via right-click), or by means of the setinfo command.
In the context of most commands you are able to reference series by name or by ID number as you
wish. The main exception is the denition or modication of variables via a formula; here you must
use names since ID numbers would get confused with numerical constants.
Note that series ID numbers are always consecutive, and the ID number for a given series will change
if you delete a lower-numbered series. In some contexts, where gretl is liable to get confused by
such changes, deletion of low-numbered series is disallowed.
71
Chapter 10. Gretl data types 72
Discrete series
It is possible to mark variables of the series type as discrete. The meaning and uses of this facility
are explained in chapter 11.
10.3 Scalars
The scalar type is relatively simple: just a convenient named holder for a single numerical value.
Scalars have none of the additional attributes pertaining to series, do not have ID numbers, and
must be referenced by name. A common use of scalar variables is to record information made
available by gretl commands for further processing, as in scalar s2 = $sigma2 to record the
square of the standard error of the regression following an estimation command such as ols.
You can dene and work with scalars in gretl without having any dataset in place.
In the gretl GUI, scalar variables can be inspected and their values edited via the Icon view (see
the View menu in the main window).
10.4 Matrices
Matrices in gretl work much as in other mathematical software (e.g. MATLAB, Octave). Like scalars
they have no ID numbers and must be referenced by name, and they can be used without any
dataset in place. Matrix indexing is 1-based: the top-left element of matrix A is A[1,1]. Matrices
are discussed at length in chapter 15; advanced users of gretl will want to study this chapter in
detail.
Matrices have two optional attribute beyond their numerical content: they may have column and/or
row names attached; these are displayed when the matrix is printed. See the colnames and
rownames functions for details.
In the gretl GUI, matrices can be inspected, analysed and edited via the Icon view item under the
View menu in the main window: each currently dened matrix is represented by an icon.
10.5 Lists
As with matrices, lists merit an explication of their own (see chapter 14). Briey, named lists can
(and should!) be used to make command scripts less verbose and repetitious, and more easily
modiable. Since lists are in fact lists of series ID numbers they can be used only when a dataset is
in place.
In the gretl GUI, named lists can be inspected and edited under the Data menu in the main window,
via the item Dene or edit list.
10.6 Strings
String variables may be used for labeling, or for constructing commands. They are discussed in
chapter 14. They must be referenced by name; they can be dened in the absence of a dataset.
Such variables can be created and modied via the command-line in the gretl console or via script;
there is no means of editing them via the gretl GUI.
10.7 Bundles
A bundle is a container or wrapper for various sorts of objectsspecically, scalars, series, matri-
ces, strings and bundles. (Yes, a bundle can contain other bundles). A bundle takes the form of a
hash table or associative array: each item placed in the bundle is associated with a key which can
used to retrieve it subsequently. We begin by explaining the mechanics of bundles then oer some
thoughts on what they are good for.
To use a bundle you rst either declare it, as in
bundle foo
or dene an empty bundle using the null keyword:
bundle foo = null
These two formulations are basically equivalent, in that they both create an empty bundle. The
dierence is that the second variant may be reusedif a bundle named foo already exists the
eect is to empty itwhile the rst may only be used once in a given gretl session; it is an error to
declare a variable that already exists.
To add an object to a bundle you assign to a compound left-hand value: the name of the bundle
followed by the key. Two forms of syntax are acceptable in this context. The original syntax (and
the only form supported prior to version 1.9.12 of gretl) requires that the key be given as a quoted
string literal enclosed in square brackets. For example, the statement
foo["matrix1"] = m
adds an object called m (presumably a matrix) to bundle foo under the key matrix1. From gretl
1.9.12, however, a simpler syntax is available, provided the key satises the rules for a gretl variable
name (31 characters maximum, starting with a letter and composed of just letters, numbers or
underscore).
1
In that case you can simply join the key to the bundle name with a dot, as in
foo.matrix1 = m
To get an item out of a bundle, again use the name of the bundle followed by the key, as in
matrix bm = foo["matrix1"]
# or using the dot notation
matrix m = foo.matrix1
Note that the key identifying an object within a given bundle is necessarily unique. If you reuse an
existing key in a new assignment, the eect is to replace the object which was previously stored
under the given key. It is not required that the type of the replacement object is the same as that
of the original.
Also note that when you add an object to a bundle, what in fact happens is that the bundle acquires
a copy of the object. The external object retains its own identity and is unaected if the bundled
object is replaced by another. Consider the following script fragment:
bundle foo
matrix m = I(3)
foo.mykey = m
scalar x = 20
foo.mykey = x
After the above commands are completed bundle foo does not contain a matrix under mykey, but
the original matrix m is still in good health.
To delete an object from a bundle use the delete command, with the bundle/key combination, as
in
1
When using the original syntax, keys do not have to be valid as variable namesfor example, they can include
spacesbut they are limited to 31 characters.
delete foo.mykey
delete foo["quoted key"]
This destroys the object associated with the key and removes the key from the hash table.
To determine whether a bundle contains an object associated with a given key, use the inbundle()
function. This takes two arguments: the name of the bundle and the key string. If the key string
contains embedded spaces it must be double-quoted, otherwise quotation is optional. The value
returned by this function is an integer which codes for the type of the object (0 for no match, 1 for
scalar, 2 for series, 3 for matrix, 4 for string and 5 for bundle). The function typestr() may be
used to get the string corresponding to this code. For example:
scalar type = inbundle(foo, x)
if type == 0
print "x: no such object"
else
printf "x is of type %s\n", typestr(type)
endif
Besides adding, accessing, replacing and deleting individual items, the other operations that are
supported for bundles are union, printing and deletion. As regards union, if bundles b1 and b2 are
dened you can say
bundle b3 = b1 + b2
to create a new bundle that is the union of the two others. The algorithm is: create a new bundle
that is a copy of b1, then add any items from b2 whose keys are not already present in the new
bundle. (This means that bundle union is not commutative if the bundles have one or more key
strings in common.)
If b is a bundle and you say print b, you get a listing of the bundles keys along with the types of
the corresponding objects, as in
? print b
bundle b:
x (scalar)
mat (matrix)
inside (bundle)
What are bundles good for?
Bundles are unlikely to be of interest in the context of standalone gretl scripts, but they can be
very useful in the context of complex function packages where a good deal of information has to
be passed around between the component functions. Instead of using a lengthy list of individual
arguments, function A can bundle up the required data and pass it to functions B and C, where
relevant information can be extracted via a mnemonic key.
In this context bundles should be passed in pointer form (see chapter 13) as illustrated in the fol-
lowing trivial example, where a bundle is created at one level then lled out by a separate function.
# modification of bundle (pointer) by user function
function void fill_out_bundle (bundle *b)
b.mat = I(3)
b.str = "foo"
b.x = 32
end function
bundle my_bundle
fill_out_bundle(&my_bundle)
The bundle type can also be used to advantage as the return value from a packaged function, in
cases where a package writer wants to give the user the option of accessing various results. In
the gretl GUI, function packages that return a bundle are treated specially: the output window that
displays the printed results acquires a menu showing the bundled items (their names and types),
from which the user can save items of interest. For example, a function package that estimates a
model might return a bundle containing a vector of parameter estimates, a residual series and a
covariance matrix for the parameter estimates, among other possibilities.
As a renement to support the use of bundles as a function return type, the setnote function can
be used to add a brief explanatory note to a bundled itemsuch notes will then be shown in the
GUI menu. This function takes three arguments: the name of a bundle, a key string, and the note.
For example
setnote(b, "vcv", "covariance matrix")
After this, the object under the key vcv in bundle b will be shown as covariance matrix in a GUI
menu.
10.8 The life cycle of gretl objects
Creation
The most basic way to create a new variable of any type is by declaration, where one states the type
followed by the name of the variable to create, as in
scalar x
series y
matrix A
and so forth. In that case the object in question is given a default initialization, as follows: a new
scalar has value NA (missing); a new series is lled with NAs; a new matrix is null (zero rows and
columns); a new string is empty; a new list has no members, and a new bundle is empty.
Declaration can be supplemented by a denite initialization, as in
scalar x = pi
series y = log(x)
matrix A = zeros(10,4)
The traditional way of creating a new variable in gretl was via the genr command (which is still
supported), as in
genr x = y/100
Here the type of x is left implicit and will be determined automatically depending on the context: if
y is a scalar, a series or a matrix x will inherit ys type (otherwise an error will be generated, since
division is applicable to these types only). Moreover, the type of a new variable can be left implicit
without use of genr:
2
x = y/100
2
Apart from the bundle type: that must always be specied.
In modern gretl scripting we recommend that you state the type of a new variable explicitly.
This makes the intent clearer to a reader of the script and also guards against errors that might
otherwise be dicult to understand (i.e. a certain variable turns out to be of the wrong type for
some subsequent calculation, but you dont notice at rst because you didnt say what type you
needed). An exception to this rule might reasonably be granted for clear and simple cases where
theres little possibility of confusion.
Modication
Typically, the values of variables of all types are modied by assignment, using the = operator with
the name of the variable on the left and a suitable value or formula on the right:
z = normal()
x = 100 * log(y) - log(y(-1))
M = qform(a, X)
By a suitable value we mean one that is conformable for the type in question. A gretl variable
acquires its type when it is rst created and this cannot be changed via assignment; for example, if
you have a matrix A and later want a string A, you will have to delete the matrix rst.
One point to watch out for in gretl scripting is type conicts having to do with the names of series brought
in from a data le. For example, in setting up a command loop (see chapter 12) it is very common to call
the loop index i. Now a loop index is a scalar (typically incremented each time round the loop). If you open
a data le that happens to contain a series named i you will get a type error (Types not conformable for
operation) when you try to use i as a loop index.
Although the type of an existing variable cannot be changed on the y, gretl nonetheless tries to be
as understanding as possible. For example if x is a series and you say
x = 100
gretl will give the series a constant value of 100 rather than complaining that you are trying to
assign a scalar to a series. This issue is particularly relevant for the matrix typesee chapter 15
for details.
Besides using the regular assignment operator you also have the option of using an inected
equals sign, as in the C programming language. This is shorthand for the case where the new value
of the variable is a function of the old value. For example,
x += 100 # in longhand: x = x + 100
x *= 100 # in longhand: x = x * 100
For scalar variables you can use a more condensed shorthand for simple increment or decrement
by 1, namely trailing ++ or -- respectively:
x = 100
x-- # x now equals 99
x++ # x now equals 100
In the case of objects holding more than one valueseries, matrices and bundlesyou can mod-
ify particular values within the object using an expression within square brackets to identify the
elements to access. We have discussed this above for the bundle type and chapter 15 goes into
details for matrices. As for series, there are two ways to specify particular values for modication:
you can use a simple 1-based index, or if the dataset is a time series or panel (or if it has marker
strings that identify the observations) you can use an appropriate observation string. Such strings
are displayed by gretl when you print data with the --byobs ag. Examples:
x[13] = 100 # simple index: the 13th observation
x[1995:4] = 100 # date: quarterly time series
x[2003:08] = 100 # date: monthly time series
x["AZ"] = 100 # the observation with marker string "AZ"
x[3:15] = 100 # panel: the 15th observation for the 3rd unit
Note that with quarterly or monthly time series there is no ambiguity between a simple index
number and a date, since dates always contain a colon. With annual time-series data, however,
such ambiguity exists and it is resolved by the rule that a number in brackets is always read as a
simple index: x[1905] means the nineteen-hundred and fth observation, not the observation for
the year 1905. You can specify a year by quotation, as in x["1905"].
Destruction
Objects of the types discussed above, with the important exception of named lists, are all destroyed
using the delete command: delete objectname.
Lists are an exception for this reason: in the context of gretl commands, a named list expands to
the ID numbers of the member series, so if you say
delete L
for L a list, the eect is to delete all the series in L; the list itself is not destroyed, but ends up
empty. To delete the list itself (without deleting the member series) you must invert the command
and use the list keyword:
list L delete
Chapter 11
Discrete variables
When a variable can take only a nite, typically small, number of values, then it is said to be discrete.
In gretl, variables of the series type (only) can be marked as discrete. (When we speak of variables
below this should be understood as referring to series.) Some gretl commands act in a slightly
dierent way when applied to discrete variables; moreover, gretl provides a few commands that
only apply to discrete variables. Specically, the dummify and xtab commands (see below) are
available only for discrete variables, while the freq (frequency distribution) command produces
dierent output for discrete variables.
11.1 Declaring variables as discrete
Gretl uses a simple heuristic to judge whether a given variable should be treated as discrete, but
you also have the option of explicitly marking a variable as discrete, in which case the heuristic
check is bypassed.
The heuristic is as follows: First, are all the values of the variable reasonably round, where this
is taken to mean that they are all integer multiples of 0.25? If this criterion is met, we then ask
whether the variable takes on a fairly small set of distinct values, where fairly small is dened
as less than or equal to 8. If both conditions are satised, the variable is automatically considered
discrete.
To mark a variable as discrete you have two options.
1. From the graphical interface, select Variable, Edit Attributes from the menu. A dialog box
will appear and, if the variable seems suitable, you will see a tick box labeled Treat this
variable as discrete. This dialog box can also be invoked via the context menu (right-click on
a variable) or by pressing the F2 key.
2. From the command-line interface, via the discrete command. The command takes one or
more arguments, which can be either variables or list of variables. For example:
list xlist = x1 x2 x3
discrete z1 xlist z2
This syntax makes it possible to declare as discrete many variables at once, which cannot
presently be done via the graphical interface. The switch --reverse reverses the declaration
of a variable as discrete, or in other words marks it as continuous. For example:
discrete foo
# now foo is discrete
discrete foo --reverse
# now foo is continuous
The command-line variant is more powerful, in that you can mark a variable as discrete even if it
does not seem to be suitable for this treatment.
Note that marking a variable as discrete does not aect its content. It is the users responsibility
to make sure that marking a variable as discrete is a sensible thing to do. Note that if you want to
recode a continuous variable into classes, you can use gretls arithmetical functionality, as in the
following example:
78
Chapter 11. Discrete variables 79
nulldata 100
# generate a series with mean 2 and variance 1
series x = normal() + 2
# split into 4 classes
series z = (x>0) + (x>2) + (x>4)
# now declare z as discrete
discrete z
Once a variable is marked as discrete, this setting is remembered when you save the data le.
11.2 Commands for discrete variables
The dummify command
The dummify command takes as argument a series x and creates dummy variables for each distinct
value present in x, which must have already been declared as discrete. Example:
open greene22_2
discrete Z5 # mark Z5 as discrete
dummify Z5
The eect of the above command is to generate 5 new dummy variables, labeled DZ5_1 through
DZ5_5, which correspond to the dierent values in Z5. Hence, the variable DZ5_4 is 1 if Z5 equals
4 and 0 otherwise. This functionality is also available through the graphical interface by selecting
the menu item Add, Dummies for selected discrete variables.
The dummify command can also be used with the following syntax:
list dlist = dummify(x)
This not only creates the dummy variables, but also a named list (see section 14.1) that can be used
afterwards. The following example computes summary statistics for the variable Y for each value
of Z5:
open greene22_2
list foo = dummify(Z5)
loop foreach i foo
smpl $i --restrict --replace
summary Y
endloop
smpl --full
Since dummify generates a list, it can be used directly in commands that call for a list as input, such
as ols. For example:
open greene22_2
ols Y 0 dummify(Z5)
The freq command
The freq command displays absolute and relative frequencies for a given variable. The way fre-
quencies are counted depends on whether the variable is continuous or discrete. This command is
also available via the graphical interface by selecting the Variable, Frequency distribution menu
entry.
For discrete variables, frequencies are counted for each distinct value that the variable takes. For
continuous variables, values are grouped into bins and then the frequencies are counted for each
bin. The number of bins, by default, is computed as a function of the number of valid observations
in the currently selected sample via the rule shown in Table 11.1. However, when the command is
invoked through the menu item Variable, Frequency Plot, this default can be overridden by the
user.
Observations Bins
8 n < 16 5
16 n < 50 7
50 n 850 [
n|
n > 850 29
Table 11.1: Number of bins for various sample sizes
For example, the following code
open greene19_1
freq TUCE
discrete TUCE # mark TUCE as discrete
freq TUCE
yields
Read datafile /usr/local/share/gretl/data/greene/greene19_1.gdt
periodicity: 1, maxobs: 32,
observations range: 1-32
Listing 5 variables:
0) const 1) GPA 2) TUCE 3) PSI 4) GRADE
? freq TUCE
Frequency distribution for TUCE, obs 1-32
number of bins = 7, mean = 21.9375, sd = 3.90151
interval midpt frequency rel. cum.
< 13.417 12.000 1 3.12% 3.12% *
13.417 - 16.250 14.833 1 3.12% 6.25% *
16.250 - 19.083 17.667 6 18.75% 25.00% ******
19.083 - 21.917 20.500 6 18.75% 43.75% ******
21.917 - 24.750 23.333 9 28.12% 71.88% **********
24.750 - 27.583 26.167 7 21.88% 93.75% *******
>= 27.583 29.000 2 6.25% 100.00% **
Test for null hypothesis of normal distribution:
Chi-square(2) = 1.872 with p-value 0.39211
? discrete TUCE # mark TUCE as discrete
? freq TUCE
Frequency distribution for TUCE, obs 1-32
frequency rel. cum.
12 1 3.12% 3.12% *
14 1 3.12% 6.25% *
17 3 9.38% 15.62% ***
19 3 9.38% 25.00% ***
20 2 6.25% 31.25% **
21 4 12.50% 43.75% ****
22 2 6.25% 50.00% **
23 4 12.50% 62.50% ****
24 3 9.38% 71.88% ***
25 4 12.50% 84.38% ****
26 2 6.25% 90.62% **
27 1 3.12% 93.75% *
28 1 3.12% 96.88% *
29 1 3.12% 100.00% *
Test for null hypothesis of normal distribution:
Chi-square(2) = 1.872 with p-value 0.39211
As can be seen from the sample output, a DoornikHansen test for normality is computed auto-
matically. This test is suppressed for discrete variables where the number of distinct values is less
than 10.
This command accepts two options: --quiet, to avoid generation of the histogram when invoked
from the command line and --gamma, for replacing the normality test with Lockes nonparametric
test, whose null hypothesis is that the data follow a Gamma distribution.
If the distinct values of a discrete variable need to be saved, the values() matrix construct can be
used (see chapter 15).
The xtab command
The xtab command cab be invoked in either of the following ways. First,
xtab ylist ; xlist
where ylist and xlist are lists of discrete variables. This produces cross-tabulations (two-way
frequencies) of each of the variables in ylist (by row) against each of the variables in xlist (by
column). Or second,
xtab xlist
In the second case a full set of cross-tabulations is generated; that is, each variable in xlist is tabu-
lated against each other variable in the list. In the graphical interface, this command is represented
by the Cross Tabulation item under the View menu, which is active if at least two variables are
selected.
Here is an example of use:
open greene22_2
discrete Z* # mark Z1-Z8 as discrete
xtab Z1 Z4 ; Z5 Z6
which produces
Cross-tabulation of Z1 (rows) against Z5 (columns)
[ 1][ 2][ 3][ 4][ 5] TOT.
[ 0] 20 91 75 93 36 315
[ 1] 28 73 54 97 34 286
TOTAL 48 164 129 190 70 601
Pearson chi-square test = 5.48233 (4 df, p-value = 0.241287)
[ 9][ 12][ 14][ 16][ 17][ 18][ 20] TOT.
[ 0] 4 36 106 70 52 45 2 315
[ 1] 3 8 48 45 37 67 78 286
TOTAL 7 44 154 115 89 112 80 601
Pearson chi-square test = 123.177 (6 df, p-value = 3.50375e-24)
[ 1][ 2][ 3][ 4][ 5] TOT.
[ 0] 17 60 35 45 14 171
[ 1] 31 104 94 145 56 430
TOTAL 48 164 129 190 70 601
[ 9][ 12][ 14][ 16][ 17][ 18][ 20] TOT.
[ 0] 1 8 39 47 30 32 14 171
[ 1] 6 36 115 68 59 80 66 430
TOTAL 7 44 154 115 89 112 80 601
Pearsons
2
test for independence is automatically displayed, provided that all cells have expected
frequencies under independence greater than 10
7
. However, a common rule of thumb states that
this statistic is valid only if the expected frequency is 5 or greater for at least 80 percent of the
cells. If this condition is not met a warning is printed.
Additionally, the --row or --column options can be given: in this case, the output displays row or
column percentages, respectively.
If you want to cut and paste the output of xtab to some other program, e.g. a spreadsheet, you
may want to use the --zeros option; this option causes cells with zero frequency to display the
number 0 instead of being empty.
Chapter 12
Loop constructs
12.1 Introduction
The command loop opens a special mode in which gretl accepts a block of commands to be re-
peated zero or more times. This feature may be useful for, among other things, Monte Carlo
simulations, bootstrapping of test statistics and iterative estimation procedures. The general form
of a loop is:
loop control-expression [ --progressive | --verbose | --quiet ]
loop body
endloop
Five forms of control-expression are available, as explained in section 12.2.
Not all gretl commands are available within loops. The commands that are not presently accepted
in this context are shown in Table 12.1.
Table 12.1: Commands not usable in loops
corrgm cusum function hurst include leverage nulldata rmplot
run setmiss vif
By default, the genr command operates quietly in the context of a loop (without printing informa-
tion on the variable generated). To force the printing of feedback from genr you may specify the
--verbose option to loop. The --quiet option suppresses the usual printout of the number of
iterations performed, which may be desirable when loops are nested.
The --progressive option to loop modies the behavior of the commands print and store,
and certain estimation commands, in a manner that may be useful with Monte Carlo analyses (see
Section 12.3).
The following sections explain the various forms of the loop control expression and provide some
examples of use of loops.
If you are carrying out a substantial Monte Carlo analysis with many thousands of repetitions, memory
capacity and processing time may be an issue. To minimize the use of computer resources, run your script
using the command-line program, gretlcli, with output redirected to a le.
12.2 Loop control variants
Count loop
The simplest form of loop control is a direct specication of the number of times the loop should
be repeated. We refer to this as a count loop. The number of repetitions may be a numerical
constant, as in loop 1000, or may be read from a scalar variable, as in loop replics.
In the case where the loop count is given by a variable, say replics, in concept replics is an
83
Chapter 12. Loop constructs 84
integer; if the value is not integral, it is converted to an integer by truncation. Note that replics is
evaluated only once, when the loop is initially compiled.
While loop
A second sort of control expression takes the form of the keyword while followed by a boolean
expression. For example,
loop while essdiff > .00001
Execution of the commands within the loop will continue so long as (a) the specied condition
evaluates as true and (b) the number of iterations does not exceed the value of the internal variable
loop_maxiter. By default this equals 100000, but you can specify a dierent value (or remove the
limit) via the set command (see the Gretl Command Reference).
Index loop
A third form of loop control uses an index variable, for example i.
1
In this case you specify starting
and ending values for the index, which is incremented by one each time round the loop. The syntax
looks like this: loop i=1..20.
The index variable may be a pre-existing scalar; if this is not the case, the variable is created
automatically and is destroyed on exit from the loop.
The index may be used within the loop body in either of two ways: you can access the integer value
of i (see Example 12.4) or you can use its string representation, $i (see Example 12.5).
The starting and ending values for the index can be given in numerical form, by reference to pre-
dened scalar variables, or as expressions that evaluate to scalars. In the latter two cases the
variables are evaluated once, at the start of the loop. In addition, with time series data you can give
the starting and ending values in the form of dates, as in loop i=1950:1..1999:4.
This form of loop control is intended to be quick and easy, and as such it is subject to certain
limitations. In particular, the index variable is always incremented by one at each iteration. If, for
example, you have
loop i=m..n
where m and n are scalar variables with values m > n at the time of execution, the index will not be
decremented; rather, the loop will simply be bypassed.
If you need more complex loop control, see the for form below.
The index loop is particularly useful in conjunction with the values() matrix function when some
operation must be carried out for each value of some discrete variable (see chapter 11). Consider
the following example:
open greene22_2
discrete Z8
v8 = values(Z8)
loop i=1..rows(v8)
scalar xi = v8[i]
smpl (Z8=xi) --restrict --replace
printf "mean(Y | Z8 = %g) = %8.5f, sd(Y | Z8 = %g) = %g\n", \
xi, mean(Y), xi, sd(Y)
endloop
1
It is common programming practice to use simple, one-character names for such variables. However, you may use
any name that is acceptable by gretl: up to 31 characters, starting with a letter, and containing nothing but letters,
numerals and the underscore character.
In this case, we evaluate the conditional mean and standard deviation of the variable Y for each
value of Z8.
Foreach loop
The fourth form of loop control also uses an index variable, in this case to index a specied list
of strings. The loop is executed once for each string in the list. This can be useful for performing
repetitive operations on a list of variables. Here is an example of the syntax:
loop foreach i peach pear plum
print "$i"
endloop
This loop will execute three times, printing out peach, pear and plum on the respective itera-
tions. The numerical value of the index starts at 1 and is incremented by 1 at each iteration.
If you wish to loop across a list of variables that are contiguous in the dataset, you can give the
names of the rst and last variables in the list, separated by .., rather than having to type all
the names. For example, say we have 50 variables AK, AL, . . . , WY, containing income levels for the
states of the US. To run a regression of income on time for each of the states we could do:
genr time
loop foreach i AL..WY
ols $i const time
endloop
This loop variant can also be used for looping across the elements in a named list (see chapter 14).
For example:
list ylist = y1 y2 y3
loop foreach i ylist
ols $i const x1 x2
endloop
Note that if you use this idiom inside a function (see chapter 13), looping across a list that has been
supplied to the function as an argument, it is necessary to use the syntax listname.$i to reference
the list-member variables. In the context of the example above, this would mean replacing the third
line with
ols ylist.$i const x1 x2
For loop
The nal form of loop control emulates the for statement in the C programming language. The
sytax is loop for, followed by three component expressions, separated by semicolons and sur-
rounded by parentheses. The three components are as follows:
1. Initialization: This is evaluated only once, at the start of the loop. Common example: setting
a scalar control variable to some starting value.
2. Continuation condition: this is evaluated at the top of each iteration (including the rst). If
the expression evaluates as true (non-zero), iteration continues, otherwise it stops. Common
example: an inequality expressing a bound on a control variable.
3. Modier: an expression which modies the value of some variable. This is evaluated prior
to checking the continuation condition, on each iteration after the rst. Common example: a
control variable is incremented or decremented.
Heres a simple example:
loop for (r=0.01; r<.991; r+=.01)
In this example the variable r will take on the values 0.01, 0.02, . . . , 0.99 across the 99 iterations.
Note that due to the nite precision of oating point arithmetic on computers it may be necessary
to use a continuation condition such as the above, r<.991, rather than the more natural r<=.99.
(Using double-precision numbers on an x86 processor, at the point where you would expect r to
equal 0.99 it may in fact have value 0.990000000000001.)
Any or all of the three expressions governing a for loop may be omittedthe minimal form is
(;;). If the continuation test is omitted it is implicitly true, so you have an innite loop unless you
arrange for some other way out, such as a break statement.
If the initialization expression in a for loop takes the common form of setting a scalar variable to
a given value, the string representation of that scalars value is made available within the loop via
the accessor $varname.
12.3 Progressive mode
If the --progressive option is given for a command loop, special behavior is invoked for certain
commands, namely, print, store and simple estimation commands. By simple here we mean
commands which (a) estimate a single equation (as opposed to a system of equations) and (b) do
so by means of a single command statement (as opposed to a block of statements, as with nls and
mle). The paradigm is ols; other possibilities include tsls, wls, logit and so on.
The special behavior is as follows.
Estimators: The results from each individual iteration of the estimator are not printed. Instead,
after the loop is completed you get a printout of (a) the mean value of each estimated coecient
across all the repetitions, (b) the standard deviation of those coecient estimates, (c) the mean
value of the estimated standard error for each coecient, and (d) the standard deviation of the
estimated standard errors. This makes sense only if there is some random input at each step.
print: When this command is used to print the value of a variable, you do not get a print each time
round the loop. Instead, when the loop is terminated you get a printout of the mean and standard
deviation of the variable, across the repetitions of the loop. This mode is intended for use with
variables that have a scalar value at each iteration, for example the error sum of squares from a
regression. Data series cannot be printed in this way, and neither can matrices.
store: This command writes out the values of the specied scalars, from each time round the
loop, to a specied le. Thus it keeps a complete record of their values across the iterations. For
example, coecient estimates could be saved in this way so as to permit subsequent examination
of their frequency distribution. Only one such store can be used in a given loop.
12.4 Loop examples
Monte Carlo example
A simple example of a Monte Carlo loop in progressive mode is shown in Example 12.1.
This loop will print out summary statistics for the a and b estimates and R
2
across the 100 rep-
etitions. After running the loop, coeffs.gdt, which contains the individual coecient estimates
from all the runs, can be opened in gretl to examine the frequency distribution of the estimates in
detail.
The command nulldata is useful for Monte Carlo work. Instead of opening a real data set,
nulldata 50 (for instance) opens a dummy data set, containing just a constant and an index vari-
able, with a series length of 50. Constructed variables can then be added using the genr command.
See the set command for information on generating repeatable pseudo-random series.
Example 12.1: Simple Monte Carlo loop
nulldata 50
set seed 547
genr x = 100 * uniform()
# open a "progressive" loop, to be repeated 100 times
loop 100 --progressive
genr u = 10 * normal()
# construct the dependent variable
genr y = 10*x + u
# run OLS regression
ols y const x
# grab the coefficient estimates and R-squared
genr a = $coeff(const)
genr b = $coeff(x)
genr r2 = $rsq
# arrange for printing of stats on these
print a b r2
# and save the coefficients to file
store coeffs.gdt a b
endloop
Iterated least squares
Example 12.2 uses a while loop to replicate the estimation of a nonlinear consumption function
of the form
C = +Y
+
as presented in Greene (2000), Example 11.3. This script is included in the gretl distribution under
the name greene11_3.inp; you can nd it in gretl under the menu item File, Script les, Practice
le, Greene....
The option --print-final for the ols command arranges matters so that the regression results
will not be printed each time round the loop, but the results fromthe regression on the last iteration
will be printed when the loop terminates.
Example 12.3 shows how a loop can be used to estimate an ARMA model, exploiting the outer
product of the gradient (OPG) regression discussed by Davidson and MacKinnon in their Estimation
and Inference in Econometrics.
Indexed loop examples
Example 12.4 shows an indexed loop in which the smpl is keyed to the index variable i. Suppose
we have a panel dataset with observations on a number of hospitals for the years 1991 to 2000
(where the year of the observation is indicated by a variable named year). We restrict the sample
to each of these years in turn and print cross-sectional summary statistics for variables 1 through
4.
Example 12.5 illustrates string substitution in an indexed loop.
The rst time round this loop the variable V will be set to equal COMP1987 and the dependent
variable for the ols will be PBT1987. The next time round V will be redened as equal to COMP1988
and the dependent variable in the regression will be PBT1988. And so on.
Example 12.2: Nonlinear consumption function
open greene11_3.gdt
# run initial OLS
ols C 0 Y
genr essbak = $ess
genr essdiff = 1
genr beta = $coeff(Y)
genr gamma = 1
# iterate OLS till the error sum of squares converges
loop while essdiff > .00001
# form the linearized variables
genr C0 = C + gamma * beta * Y^gamma * log(Y)
genr x1 = Y^gamma
genr x2 = beta * Y^gamma * log(Y)
# run OLS
ols C0 0 x1 x2 --print-final --no-df-corr --vcv
genr beta = $coeff(x1)
genr gamma = $coeff(x2)
genr ess = $ess
genr essdiff = abs(ess - essbak)/essbak
genr essbak = ess
endloop
# print parameter estimates using their "proper names"
set echo off
printf "alpha = %g\n", $coeff(0)
printf "beta = %g\n", beta
printf "gamma = %g\n", gamma
Example 12.3: ARMA 1, 1
open armaloop.gdt
genr c = 0
genr a = 0.1
genr m = 0.1
series e = 1.0
genr de_c = e
genr de_a = e
genr de_m = e
genr crit = 1
loop while crit > 1.0e-9
# one-step forecast errors
genr e = y - c - a*y(-1) - m*e(-1)
# log-likelihood
genr loglik = -0.5 * sum(e^2)
print loglik
# partials of forecast errors wrt c, a, and m
genr de_c = -1 - m * de_c(-1)
genr de_a = -y(-1) -m * de_a(-1)
genr de_m = -e(-1) -m * de_m(-1)
# partials of l wrt c, a and m
genr sc_c = -de_c * e
genr sc_a = -de_a * e
genr sc_m = -de_m * e
# OPG regression
ols const sc_c sc_a sc_m --print-final --no-df-corr --vcv
# Update the parameters
genr dc = $coeff(sc_c)
genr c = c + dc
genr da = $coeff(sc_a)
genr a = a + da
genr dm = $coeff(sc_m)
genr m = m + dm
printf " constant = %.8g (gradient = %#.6g)\n", c, dc
printf " ar1 coefficient = %.8g (gradient = %#.6g)\n", a, da
printf " ma1 coefficient = %.8g (gradient = %#.6g)\n", m, dm
genr crit = $T - $ess
print crit
endloop
genr se_c = $stderr(sc_c)
genr se_a = $stderr(sc_a)
genr se_m = $stderr(sc_m)
set echo off
printf "\n"
printf "constant = %.8g (se = %#.6g, t = %.4f)\n", c, se_c, c/se_c
printf "ar1 term = %.8g (se = %#.6g, t = %.4f)\n", a, se_a, a/se_a
printf "ma1 term = %.8g (se = %#.6g, t = %.4f)\n", m, se_m, m/se_m
Example 12.4: Panel statistics
open hospitals.gdt
loop i=1991..2000
smpl (year=i) --restrict --replace
summary 1 2 3 4
endloop
Example 12.5: String substitution
open bea.dat
loop i=1987..2001
genr V = COMP$i
genr TC = GOC$i - PBT$i
genr C = TC - V
ols PBT$i const TC V
endloop
Chapter 13
User-dened functions
13.1 Dening a function
Gretl oers a mechanism for dening functions, which may be called via the command line, in
the context of a script, or (if packaged appropriately, see section 13.5) via the programs graphical
interface.
The syntax for dening a function looks like this:
1
function return-type function-name (parameters)
function body
end function
The opening line of a function denition contains these elements, in strict order:
1. The keyword function.
2. return-type, which states the type of value returned by the function, if any. This must be one
of void (if the function does not return anything), scalar, series, matrix, list or string.
3. function-name, the unique identier for the function. Function names have a maximum length
of 31 characters; they must start with a letter and can contain only letters, numerals and the
underscore character. You will get an error if you try to dene a function having the same
name as an existing gretl command.
4. The functionss parameters, in the form of a comma-separated list enclosed in parentheses.
This may be run into the function name, or separated by white space as shown.
Function parameters can be of any of the types shown below.
2
Type Description
bool scalar variable acting as a Boolean switch
int scalar variable acting as an integer
scalar scalar variable
series data series
list named list of series
matrix matrix or vector
string string variable or string literal
bundle all-purpose container (see section 10.7)
Each element in the listing of parameters must include two terms: a type specier, and the name
by which the parameter shall be known within the function. An example follows:
1
The syntax given here diers from the standard prior to gretl version 1.8.4. For reasons of backward compatibility
the old syntax is still supported; see section 13.6 for details.
2
An additional parameter type is available for GUI use, namely obs; this is equivalent to int except for the way it is
represented in the graphical interface for calling a function.
91
Chapter 13. User-dened functions 92
function scalar myfunc (series y, list xvars, bool verbose)
Each of the type-speciers, with the exception of list and string, may be modied by prepending
an asterisk to the associated parameter name, as in
function scalar myfunc (series *y, scalar *b)
The meaning of this modication is explained below (see section 13.4); it is related to the use of
pointer arguments in the C programming language.
Function parameters: optional renements
Besides the required elements mentioned above, the specication of a function parameter may
include some additional elds, as follows:
The const modier.
For scalar or int parameters: minimum, maximum and default values; or for bool parame-
ters, just a default value.
For optional pointer and list arguments (see section 13.4), the special default value null.
For all parameters, a descriptive string.
For int parameters with minimum and maximum values specied, a set of strings to associate
with the allowed numerical values (value labels).
The rst two of these options may be useful in many contexts; the last two may be helpful if a
function is to be packaged for use in the gretl GUI (but probably not otherwise). We now expand on
each of the options.
The const modier: must be given as a prex to the basic parameter specication, as in
const matrix M
This constitutes a promise that the corresponding argument will not be modied within the
function; gretl will ag an error if the function attempts to modify the argument.
Minimum, maximum and default values for scalar or int types: These values should di-
rectly follow the name of the parameter, enclosed in square brackets and with the individual
elements separated by colons. For example, suppose we have an integer parameter order for
which we wish to specify a minimum of 1, a maximum of 12, and a default of 4. We can write
int order[1:12:4]
If you wish to omit any of the three speciers, leave the corresponding eld empty. For
example [1::4] would specify a minimum of 1 and a default of 4 while leaving the maximum
unlimited.
For a parameter of type bool (whose values are just zero or non-zero), you can specify a
default of 1 (true) or 0 (false), as in
bool verbose[0]
Descriptive string: This will show up as an aid to the user if the function is packaged (see
section 13.5 below) and called via gretls graphical interface. The string should be enclosed
in double quotes and separated from the preceding elements of the parameter specication
with a space, as in
series y "dependent variable"
Value labels: These may be used only with int parameters for which minimum and maximum
values have been specied, so there is a xed number of admissible values, and the number
of labels must match the number of values. They will show up in the graphical interface
in the form of a drop-down list, making the function writers intent clearer when an integer
argument represents a categorical selection. A set of value labels must be enclosed in braces,
and the individual labels must be enclosed in double quotes and separated by commas or
spaces. For example:
int case[1:3:1] {"Fixed effects", "Between model", "Random effects"}
If two or more of the trailing optional elds are given in a parameter specication, they must be
given in the order shown above: minmaxdefault, description, value labels. Note that there is
no facility for escaping characters within descriptive strings or value labels; these may contain
spaces but they cannot contain the double-quote character.
Here is an example of a well-formed function specication using all the elements mentioned above:
function matrix myfunc (series y "dependent variable",
list X "regressors",
int p[0::1] "lag order",
int c[1:2:1] "criterion" {"AIC", "BIC"},
bool quiet[0])
One advantage of specifying default values for parameters, where applicable, is that in script or
command-line mode users may omit trailing arguments that have defaults. For example, myfunc
above could be invoked with just two arguments, corresponding to y and X; implicitly p = 1, c = 1
and quiet is false.
Functions taking no parameters
You may dene a function that has no parameters (these are called routines in some programming
languages). In this case, use the keyword void in place of the listing of parameters:
function matrix myfunc2 (void)
The function body
The function body is composed of gretl commands, or calls to user-dened functions (that is,
function calls may be nested). A function may call itself (that is, functions may be recursive). While
the function body may contain function calls, it may not contain function denitions. That is, you
cannot dene a function inside another function. For further details, see section 13.4.
13.2 Calling a function
A user function is called by typing its name followed by zero or more arguments enclosed in
parentheses. If there are two or more arguments these should be separated by commas.
There are automatic checks in place to ensure that the number of arguments given in a function
call matches the number of parameters, and that the types of the given arguments match the types
specied in the denition of the function. An error is agged if either of these conditions is violated.
One qualication: allowance is made for omitting arguments at the end of the list, provided that
default values are specied in the function denition. To be precise, the check is that the number
of arguments is at least equal to the number of required parameters, and is no greater than the
total number of parameters.
A scalar, series or matrix argument to a function may be given either as the name of a pre-existing
variable or as an expression which evaluates to a variable of the appropriate type. Scalar arguments
may also be given as numerical values. List arguments must be specied by name.
The following trivial example illustrates a function call that correctly matches the function deni-
tion.
# function definition
function scalar ols_ess(series y, list xvars)
ols y 0 xvars --quiet
scalar myess = $ess
printf "ESS = %g\n", myess
return myess
end function
# main script
open data4-1
list xlist = 2 3 4
# function call (the return value is ignored here)
ols_ess(price, xlist)
The function call gives two arguments: the rst is a data series specied by name and the second
is a named list of regressors. Note that while the function oers the variable myess as a return
value, it is ignored by the caller in this instance. (As a side note here, if you want a function to
calculate some value having to do with a regression, but are not interested in the full results of the
regression, you may wish to use the --quiet ag with the estimation command as shown above.)
A second example shows how to write a function call that assigns a return value to a variable in the
caller:
# function definition
function series get_uhat(series y, list xvars)
series uh = $uhat
return uh
end function
# main script
open data4-1
list xlist = 2 3 4
# function call
series resid = get_uhat(price, xlist)
13.3 Deleting a function
If you have dened a function and subsequently wish to clear it out of memory, you can do so using
the keywords delete or clear, as in
function myfunc delete
function get_uhat clear
Note, however, that if myfunc is already a dened function, providing a newdenition automatically
overwrites the previous one, so it should rarely be necessary to delete functions explicitly.
13.4 Function programming details
Variables versus pointers
Series, scalar, and matrix arguments to functions can be passed in two ways: as they are, or as
pointers. For example, consider the following:
function series triple1(series x)
return 3*x
end function
function series triple2(series *x)
return 3*x
end function
These two functions are nearly identical (and yield the same result); the only dierence is that you
need to feed a series into triple1, as in triple1(myseries), while triple2 must be supplied a
pointer to a series, as in triple2(&myseries).
Why make the distinction? There are two main reasons for doing so: modularity and performance.
By modularity we mean the insulation of a function from the rest of the script which calls it. One of
the many benets of this approach is that your functions are easily reusable in other contexts. To
achieve modularity, variables created within a function are local to that function, and are destroyed
when the function exits, unless they are made available as return values and these values are picked
up or assigned by the caller.
In addition, functions do not have access to variables in outer scope (that is, variables that exist
in the script from which the function is called) except insofar as these are explicitly passed to the
function as arguments.
By default, when a variable is passed to a function as an argument, what the function actually gets
is a copy of the outer variable, which means that the value of the outer variable is not modied by
anything that goes on inside the function. But the use of pointers allows a function and its caller
to cooperate such that an outer variable can be modied by the function. In eect, this allows a
function to return more than one value (although only one variable can be returned directlysee
below). The parameter in question is marked with a prex of * in the function denition, and the
corresponding argument is marked with the complementary prex & in the caller. For example,
function series get_uhat_and_ess(series y, list xvars, scalar *ess)
ess = $ess
series uh = $uhat
return uh
end function
# main script
open data4-1
list xlist = 2 3 4
# function call
scalar SSR
series resid = get_uhat_and_ess(price, xlist, &SSR)
In the above, we may say that the function is given the address of the scalar variable SSR, and it
assigns a value to that variable (under the local name ess). (For anyone used to programming in C:
note that it is not necessary, or even possible, to dereference the variable in question within the
function using the * operator. Unadorned use of the name of the variable is sucient to access the
variable in outer scope.)
An address parameter of this sort can be used as a means of oering optional information to the
caller. (That is, the corresponding argument is not strictly needed, but will be used if present). In
that case the parameter should be given a default value of null and the the function should test to
see if the caller supplied a corresponding argument or not, using the built-in function isnull().
For example, here is the simple function shown above, modied to make the lling out of the ess
value optional.
function series get_uhat_and_ess(series y, list xvars, scalar *ess[null])
if !isnull(ess)
ess = $ess
endif
return $uhat
end function
If the caller does not care to get the ess value, it can use null in place of a real argument:
series resid = get_uhat_and_ess(price, xlist, null)
Alternatively, trailing function arguments that have default values may be omitted, so the following
would also be a valid call:
series resid = get_uhat_and_ess(price, xlist)
Pointer arguments may also be useful for optimizing performance: even if a variable is not modied
inside the function, it may be a good idea to pass it as a pointer if it occupies a lot of memory.
Otherwise, the time gretl spends transcribing the value of the variable to the local copy may be
non-negligible, compared to the time the function spends doing the job it was written for.
Example 13.1 takes this to the extreme. We dene two functions which return the number of rows
of a matrix (a pretty fast operation). One function gets a matrix as argument, the other one a pointer
to a matrix. The two functions are evaluated on a matrix with 2000 rows and 2000 columns; on a
typical system, oating-point numbers take 8 bytes of memory, so the space occupied by the matrix
is roughly 32 megabytes.
Running the code in example 13.1 will produce output similar to the following (the actual numbers
depend on the machine youre running the example on):
Elapsed time:
without pointers (copy) = 3.66 seconds,
with pointers (no copy) = 0.01 seconds.
If a pointer argument is used for this sort of purposeand the object to which the pointer points
is not modied by the functionit is a good idea to signal this to the user by adding the const
qualier, as shown for function b in Example 13.1. When a pointer argument is qualied in this
way, any attempt to modify the object within the function will generate an error.
One limitation on the use of pointer-type arguments should be noted: you cannot supply a given
variable as a pointer argument more than once in any given function call. For example, suppose we
have a function that takes two matrix-pointer arguments,
function scalar pointfunc (matrix *a, matrix *b)
And suppose we have two matrices, x and y, at the caller level. The call
pointfunc(&x, &y)
is OK, but the call
pointfunc(&x, &x) # will not work
will generate an error.
List arguments
The use of a named list as an argument to a function gives a means of supplying a function with
a set of variables whose number is unknown when the function is writtenfor example, sets of
regressors or instruments. Within the function, the list can be passed on to commands such as
ols.
Example 13.1: Performance comparison: values versus pointer
function scalar a(matrix X)
return rows(X)
end function
function scalar b(const matrix *X)
return rows(X)
end function
nulldata 10
set echo off
set messages off
X = zeros(2000,2000)
r = 0
set stopwatch
loop 100
r = a(X)
endloop
fa = $stopwatch
set stopwatch
loop 100
r = b(&X)
endloop
fb = $stopwatch
printf "Elapsed time:\n\
\twithout pointers (copy) = %g seconds,\n\
\twith pointers (no copy) = %g seconds.\n", fa, fb
A list argument can also be unpacked using a foreach loop construct, but this requires some
care. For example, suppose you have a list X and want to calculate the standard deviation of each
variable in the list. You can do:
loop foreach i X
scalar sd_$i = sd(X.$i)
endloop
Please note: a special piece of syntax is needed in this context. If we wanted to perform the above
task on a list in a regular script (not inside a function), we could do
loop foreach i X
scalar sd_$i = sd($i)
endloop
where $i gets the name of the variable at position i in the list, and sd($i) gets its standard
deviation. But inside a function, working on a list supplied as an argument, if we want to reference
an individual variable in the list we must use the syntax listname.varname. Hence in the example
above we write sd(X.$i).
This is necessary to avoid possible collisions between the name-space of the function and the name-
space of the caller script. For example, suppose we have a function that takes a list argument, and
that denes a local variable called y. Now suppose that this function is passed a list containing
a variable named y. If the two name-spaces were not separated either wed get an error, or the
external variable y would be silently over-written by the local one. It is important, therefore, that
list-argument variables should not be visible by name within functions. To get hold of such
variables you need to use the form of identication just mentioned: the name of the list, followed
by a dot, followed by the name of the variable.
Constancy of list arguments When a named list of variables is passed to a function, the function
is actually provided with a copy of the list. The function may modify this copy (for instance, adding
or removing members), but the original list at the level of the caller is not modied.
Optional list arguments If a list argument to a function is optional, this should be indicated by
appending a default value of null, as in
function scalar myfunc (scalar y, list X[null])
In that case, if the caller gives null as the list argument (or simply omits the last argument) the
named list X inside the function will be empty. This possibility can be detected using the nelem()
function, which returns 0 for an empty list.
String arguments
String arguments can be used, for example, to provide exibility in the naming of variables created
within a function. In the following example the function mavg returns a list containing two moving
averages constructed from an input series, with the names of the newly created variables governed
by the string argument.
function list mavg (series y, string vname)
series @vname_2 = (y+y(-1)) / 2
series @vname_4 = (y+y(-1)+y(-2)+y(-3)) / 4
list retlist = @vname_2 @vname_4
return retlist
end function
open data9-9
list malist = mavg(nocars, "nocars")
print malist --byobs
The last line of the script will print two variables named nocars_2 and nocars_4. For details on
the handling of named strings, see chapter 14.
If a string argument is considered optional, it may be given a null default value, as in
function scalar foo (series y, string vname[null])
Retrieving the names of arguments
The variables given as arguments to a function are known inside the function by the names of the
corresponding parameters. For example, within the function whose signature is
function void somefun (series y)
we have the series known as y. It may be useful, however, to be able to determine the names of
the variables provided as arguments. This can be done using the function argname, which takes
the name of a function parameter as its single argument and returns a string. Here is a simple
illustration:
function void namefun (series y)
printf "the series given as y was named %s\n", argname(y)
end function
open data9-7
namefun(QNC)
This produces the output
the series given as y was named QNC
Please note that this will not always work: the arguments given to functions may be anonymous
variables, created on the y, as in somefun(log(QNC)) or somefun(CPI/100). In that case the
argname function returns an empty string. Function writers who wish to make use of this facility
should check the return from argname using the strlen() function: if this returns 0, no name was
found.
Return values
Functions can return nothing (just printing a result, perhaps), or they can return a single variable
a scalar, series, list, matrix, string, or bundle (see section 10.7). The return value, if any, is specied
via a statement within the function body beginning with the keyword return, followed by either the
name of a variable (which must be of the type announced on the rst line of the function denition)
or an expression which produces a value of the correct type.
Having a function return a list or bundle is a way of permitting the return of more than one
variable. For example, you can dene several series inside a function and package them as a list;
in this case they are not destroyed when the function exits. Here is a simple example, which also
illustrates the possibility of setting the descriptive labels for variables generated in a function.
function list make_cubes (list xlist)
list cubes = null
loop foreach i xlist --quiet
series $i3 = (xlist.$i)^3
setinfo $i3 -d "cube of $i"
list cubes += $i3
endloop
return cubes
end function
open data4-1
list xlist = price sqft
list cubelist = make_cubes(xlist)
print xlist cubelist --byobs
labels
A return statement causes the function to return (exit) at the point where it appears within the
body of the function. A function may also exit when (a) the end of the function code is reached (in
the case of a function with no return value), (b) a gretl error occurs, or (c) a funcerr statement is
reached.
The funcerr keyword, which may be followed by a string enclosed in double quotes, causes a
function to exit with an error agged. If a string is provided, this is printed on exit, otherwise a
generic error message is printed. This mechanism enables the author of a function to pre-empt an
ordinary execution error and/or oer a more specic and helpful error message. For example,
if nelem(xlist) = 0
funcerr "xlist must not be empty"
endif
A function may contain more than one return statement, as in
function scalar multi (bool s)
if s
return 1000
else
return 10
endif
end function
However, it is recommended programming practice to have a single return point from a function
unless this is very inconvenient. The simple example above would be better written as
function scalar multi (bool s)
return s ? 1000 : 10
end function
Error checking
When gretl rst reads and compiles a function denition there is minimal error-checking: the
only checks are that the function name is acceptable, and, so far as the body is concerned, that you
are not trying to dene a function inside a function (see Section 13.1). Otherwise, if the function
body contains invalid commands this will become apparent only when the function is called and
its commands are executed.
Debugging
The usual mechanism whereby gretl echoes commands and reports on the creation of new variables
is by default suppressed when a function is being executed. If you want more verbose output from
a particular function you can use either or both of the following commands within the function:
set echo on
set messages on
Alternatively, you can achieve this eect for all functions via the command set debug 1. Usually
when you set the value of a state variable using the set command, the eect applies only to the
current level of function execution. For instance, if you do set messages on within function f1,
which in turn calls function f2, then messages will be printed for f1 but not f2. The debug variable,
however, acts globally; all functions become verbose regardless of their level.
Further, you can do set debug 2: in addition to command echo and the printing of messages, this
is equivalent to setting max_verbose (which produces verbose output from the BFGS maximizer) at
all levels of function execution.
13.5 Function packages
Since gretl 1.6.0 there has been a mechanism to package functions and make them available to
other users of gretl. Here is a walk-through of the process.
Load a function in memory
There are several ways to load a function:
If you have a script le containing function denitions, open that le and run it.
Create a script le from scratch. Include at least one function denition, and run the script.
Open the GUI console and type a function denition interactively. This method is not partic-
ularly recommended; you are probably better composing a function non-interactively.
For example, suppose you decide to package a function that returns the percentage change of a
time series. Open a script le and type
function series pc(series y "Series to process")
return 100 * diff(y)/y(-1)
end function
In this case, we have appended a string to the function argument, as explained in section 13.1, so
as to make our interface more informative. This is not obligatory: if you omit the descriptive string,
gretl will supply a predened one.
Now run your function. You may want to make sure it works properly by running a few tests. For
example, you may open the console and type
genr x = uniform()
genr dpcx = pc(x)
print x dpcx --byobs
You should see something similar to gure 13.1. The function seems to work ok. Once your
function is debugged, you may proceed to the next stage.
Create a package
We rst present the mechanism for creating a function package via gretls graphical interface. This
can also be done via the command line, which oers some additional functionality for package
authors; an explanation is given later in this section.
Figure 13.1: Output of function check
Start the GUI program and take a look at the File, Function les menu. This menu contains four
items: On local machine, On server, Edit package, New package.
Select New package. (This will produce an error message unless at least one user-dened function
is currently loaded in memorysee the previous point.) In the rst dialog you get to select:
A public function to package.
Zero or more private helper functions.
Public functions are directly available to users; private functions are part of the behind the scenes
mechanism in a function package.
On clicking OK a second dialog should appear (see Figure 13.2), where you get to enter the package
information (author, version, date, and a short description). You can also enter help text for the
public interface. You have a further chance to edit the code of the function(s) to be packaged, by
clicking on Edit function code. (If the package contains more than one function, a drop-down
selector will be shown.) And you get to add a sample script that exercises your package. This
will be helpful for potential users, and also for testing. A sample script is required if you want to
upload the package to the gretl server (for which a check-box is supplied).
You wont need it right now, but the button labeled Save as script allows you to reverse engineer
a function package, writing out a script that contains all the relevant function denitions.
Clicking Save in this dialog leads you to a File Save dialog. All being well, this should be pointing
towards a directory named functions, either under the gretl system directory (if you have write
permission on that) or the gretl user directory. This is the recommended place to save function
package les, since that is where the program will look in the special routine for opening such les
(see below).
Needless to say, the menu command File, Function les, Edit package allows you to make changes
to a local function package.
A word on the le you just saved. By default, it will have a .gfn extension. This is a function
package le: unlike an ordinary gretl script le, it is an XML le containing both the function code
and the extra information entered in the packager. Hackers might wish to write such a le from
scratch rather than using the GUI packager, but most people are likely to nd it awkward. Note
Figure 13.2: The package editor window
that XML-special characters in the function code have to be escaped, e.g. & must be represented as
&. Also, some elements of the function syntax dier from the standard script representation:
the parameters and return values (if any) are represented in XML. Basically, the function is pre-
parsed, and ready for fast loading using libxml.
Load a package
Why package functions in this way? To see whats on oer so far, try the next phase of the walk-
through.
Close gretl, then re-open it. Now go to File, Function les, On local machine. If the previous stage
above has gone OK, you should see the le you packaged and saved, with its short description. If
you click on Info you get a window with all the information gretl has gleaned from the function
package. If you click on the View code icon in the toolbar of this new window, you get a script
view window showing the actual function code. Now, back to the Function packages window, if
you click on the packages name, the relevant functions are loaded into gretls workspace, ready to
be called by clicking on the Call button.
After loading the function(s) from the package, open the GUI console. Try typing help foo, replac-
ing foo with the name of the public interface from the loaded function package: if any help text
was provided for the function, it should be presented.
In a similar way, you can browse and load the function packages available on the gretl server, by
selecting File, Function les, On server.
Once your package is installed on your local machine, you can use the function it contains via
the graphical interface as described above, or by using the CLI, namely in a script or through the
console. In the latter case, you load the function via the include command, specifying the package
le as the argument, complete with the .gfn extension.
To continue with our example, load the le np.gdt (supplied with gretl among the sample datasets).
Suppose you want to compute the rate of change for the variable iprod via your new function and
Figure 13.3: Using your package
store the result in a series named foo.
Go to File, Function les, On local machine. You will be shown a list of the installed packages,
including the one you have just created. If you select it and click on Execute (or double-click on
the name of the function package), a window similar to the one shown in gure 13.3 will appear.
Notice that the description string Series to process, supplied with the function denition, appears
to the left of the top series chooser.
Click Ok and the series foo will be generated (see gure 13.4). Under some circumstances you
might have to press the r key (refresh) to have your new variable show up the main window
variable list.
Figure 13.4: Percent change in industrial production
Alternatively, the same could have been accomplished by the script
include pc.gfn
open np
foo = pc(iprod)
Creating a package via the command line
The mechanism described above, for creating function packages using the GUI, is likely to be con-
venient for small to medium-sized packages but may be too cumbersome for ambitious packages
that include a large hierarchy of private functions. To facilitate the building of such packages gretl
oers the makepkg command.
To use makepkg you create three les: a driver script that loads all the functions you want to pack-
age and invokes makepkg; a small, plain-text specication le that contains the required package
details (author, version, etc.); and (in the simplest case) a plain text help le. You run the driver
script and gretl writes the package (.gfn) le.
We rst illustrate with a simple notional package. We have a gretl script le named foo.inp that
contains a function, foo, that we want to package. Our driver script would then look like this
include foo.inp
makepkg foo.gfn
Note that the makepkg command takes one argument, the name of the package le to be created.
The package specication le should have the same basename but the extension .spec. In this case
gretl will therefore look for foo.spec. It should look something like this:
# foo.spec
author = A. U. Thor
version = 1.0
date = 2011-02-01
description = Does something with time series
public = foo
help = foohelp.txt
sample-script = example.inp
min-version = 1.9.3
data-requirement = needs-time-series-data
As you can see, the format of each line in this le is key = value, with two qualications: blank
lines are permitted (and ignored, as are comment lines that start with #).
All the elds included in the above example are required, with the exception of data-requirement,
though the order in which they appear is immaterial. Heres a run-down of the basic elds:
author: the name(s) of the author(s). Accented or other non-ASCII characters should be given
as UTF-8.
version: the version number of the package, which should be limited to two integers sepa-
rated by a period.
date: the release date of the current verson of the package, in ISO 8601 format: YYYY-MM-DD.
description: a brief description of the functionality oered by the package. This will be
displayed in the GUI function packages window so it should be just one short line.
public: the listing of public functions.
help: the name of a plain text (UTF-8) le containing help; all packages must provide help.
sample-script: the name of a sample script that illustrates use of the package; all packages
must supply a sample script.
min-version: the minimum version of gretl required for the package to work correctly. If
youre unsure about this, the conservative thing is to give the current gretl version.
The public eld indicates which function or functions are to be made directly available to users (as
opposed to private helper functions). In the example above there is just one public function. Note
that any functions in memory when makepkg is invoked, other than those designated as public, are
assumed to be private functions that should also be included in the package. That is, the list of
private functions (if any) is implicit.
The data-requirement eld should be specied if the package requires time-series or panel data,
or alternatively if no dataset is required. If the data-requirement eld is omitted, the assumption
is that the package needs a dataset in place, but it doesnt matter what kind; if the packaged
functions do not use any series or lists this requirement can be explicitly relaxed. Valid values for
this eld are:
needs-time-series-data (any time-series data OK)
needs-qm-data (must be quarterly or monthly)
needs-panel-data (must be a panel)
no-data-ok (no dataset is needed)
For a more complex example, lets look at the gig (GARCH-in-gretl) package. The driver script for
building gig looks something like this:
set echo off
set messages off
include gig_mle.inp
include gig_setup.inp
include gig_estimate.inp
include gig_printout.inp
include gig_plot.inp
makepkg gig.gfn
In this case the functions to be packaged (of which there are many) are distributed across several
script les, each of which is the target of an include command. The set commands at the top are
included to cut down on the verbosity of the output.
The content of gig.spec is as follows:
author = Riccardo "Jack" Lucchetti and Stefano Balietti
version = 2.0
date = 2010-12-21
description = An assortment of univariate GARCH models
public = GUI_gig \
gig_setup gig_set_dist gig_set_pq gig_set_vQR \
gig_print gig_estimate \
gig_plot gig_dplot \
gig_bundle_print GUI_gig_plot
gui-main = GUI_gig
bundle-print = gig_bundle_print
bundle-plot = GUI_gig_plot
help = gig.pdf
sample-script = examples/example1.inp
min-version = 1.9.3
data-requirement = needs-time-series-data
Note that backslash continuation can be used for the elements of the public function listing.
In addition to the elds shown in the simple example above, gig.spec includes three optional
elds: gui-main, bundle-print and bundle-plot. These keywords are used to designate certain
functions as playing a special role in the gretl graphical interface. A function picked out in this way
must be in the public list and must satisfy certain further requirements.
gui-main: this species a function as the one which will be presented automatically to GUI
users (instead of users being faced with a choice of interfaces). This makes sense only for
packages that have multiple public functions. In addition, the gui-main function must return
a bundle (see section 10.7).
bundle-print: this picks out a function that should be used to print the contents of a bundle
returned by the gui-main function. It must take a pointer-to-bundle as its rst argument.
The second argument, if present, should be an int switch, with two or more valid values, that
controls the printing in some way. Any further arguments must have default values specied
so that they can be omitted.
bundle-plot: selects a function for the role of producing a plot or graph based on the con-
tents of a returned bundle. The requirements on this function are as for bundle-print.
The GUI special tags support a user-friendly mode of operation. On a successful call to gui-main,
gretl opens a window displaying the contents of the returned bundle (formatted via bundle-print).
Menus in this window give the user the option of saving the entire bundle (in which case its rep-
resented as an icon in the icon view window) or of extracting specic elements from the bundle
(series or matrices, for example).
If the package has a bundle-plot function, the bundle window also has a Graph menu. In gig, for
example, the bundle-plot function has this signature:
function void GUI_gig_plot(bundle *model, int ptype[0:1:0] \
"Plot type" {"Time series", "Density"})
The ptype switch is used to choose between a time-series plot of the residual and its conditional
variance, and a kernel density plot of the innovation against the theoretical distribution it is sup-
posed to follow. The use of the value-labels Time series and Density means that the Graph menu
will display these two choices.
One other feature of the gig spec le is noteworthy: the help eld species gig.pdf, documenta-
tion in PDF format. Unlike plain-text help, this cannot be rolled into the gfn (XML) le produced
by the makepkg command; rather, both gig.gfn and gig.pdf are packaged into a zip archive for
distribution. This represents a form of package which is new in gretl 1.9.4. More details will be
made available before long.
13.6 Memo: updating old-style functions
As mentioned at the start of this chapter, dierent rules were in force for dening functions prior
to gretl 1.8.4. While the old syntax is still supported to date, this may not always be the case. But
it is straightforward to convert a function to the new style. The only thing that must be changed
for compatibility with the new syntax is the declaration of the functions return type. Previously
this was placed inline in the return statement, whereas now it is placed right after the function
keyword. For example:
# old style
function triple (series x)
y = 3*x
return series y # note the "series" here
end function
# new style
function series triple (series x)
y = 3*x
return y
end function
Note also that the role of the return statement has changed (and its use has become more exible):
The return statement now causes the function to return directly, and you can have more
than one such statement, wrapped in conditionals. Before there could only be one return
statement, and its role was just to specify the type available for assignment by the caller.
The nal element in the return statement can now be an expression that evaluates to a value
of the advertised return type; before, it had to be the name of a pre-dened variable.
Chapter 14
Named lists and strings
14.1 Named lists
Many gretl commands take one or more lists of series as arguments. To make this easier to handle
in the context of command scripts, and in particular within user-dened functions, gretl oers the
possibility of named lists.
Creating and modifying named lists
A named list is created using the keyword list, followed by the name of the list, an equals sign,
and an expression that forms a list. The most basic sort of expression that works in this context is
a space-separated list of variables, given either by name or by ID number. For example,
list xlist = 1 2 3 4
list reglist = income price
Note that the variables in question must be of the series type: you cannot include scalars in a
named list.
Two abbreviations are available in dening lists:
You can use the wildcard character, *, to create a list of variables by name. For example,
dum* can be used to indicate all variables whose names begin with dum.
You can use two dots to indicate a range of variables. For example income..price indicates
the set of variables whose ID numbers are greater than or equal to that of income and less
than or equal to that of price.
In addition there are two special forms:
If you use the keyword null on the right-hand side, you get an empty list.
If you use the keyword dataset on the right, you get a list containing all the series in the
current dataset (except the pre-dened const).
The name of the list must start with a letter, and must be composed entirely of letters, numbers
or the underscore character. The maximum length of the name is 31 characters; list names cannot
contain spaces.
Once a named list has been created, it will be remembered for the duration of the gretl session
(unless you delete it), and can be used in the context of any gretl command where a list of variables
is expected. One simple example is the specication of a list of regressors:
list xlist = x1 x2 x3 x4
ols y 0 xlist
To get rid of a list, you use the following syntax:
list xlist delete
109
Chapter 14. Named lists and strings 110
Be careful: delete xlist will delete the variables contained in the list, so it implies data loss
(which may not be what you want). On the other hand, list xlist delete will simply undene
the xlist identier and the variables themselves will not be aected.
Similarly, to print the names of the variables in a list you have to invert the usual print command,
as in
list xlist print
If you just say print xlist the list will be expanded and the values of all the member variables
will be printed.
Lists can be modied in various ways. To redene an existing list altogether, use the same syntax
as for creating a list. For example
list xlist = 1 2 3
xlist = 4 5 6
After the second assignment, xlist contains just variables 4, 5 and 6.
To append or prepend variables to an existing list, we can make use of the fact that a named list
stands in for a longhand list. For example, we can do
list xlist = xlist 5 6 7
xlist = 9 10 xlist 11 12
Another option for appending a term (or a list) to an existing list is to use +=, as in
xlist += cpi
To drop a variable from a list, use -=:
xlist -= cpi
In most contexts where lists are used in gretl, it is expected that they do not contain any duplicated
elements. If you form a new list by simple concatenation, as in list L3 = L1 L2 (where L1 and
L2 are existing lists), its possible that the result may contain duplicates. To guard against this you
can form a new list as the union of two existing ones:
list L3 = L1 || L2
The result is a list that contains all the members of L1, plus any members of L2 that are not already
in L1.
In the same vein, you can construct a new list as the intersection of two existing ones:
list L3 = L1 && L2
Here L3 contains all the elements that are present in both L1 and L2.
You can also subtract one list from another:
list L3 = L1 - L2
The result contains all the elements of L1 that are not present in L2.
Lists and matrices
Another way of forming a list is by assignment from a matrix. The matrix in question must be
interpretable as a vector containing ID numbers of (series) variables. It may be either a row or
a column vector, and each of its elements must have an integer part that is no greater than the
number of variables in the data set. For example:
matrix m = {1,2,3,4}
list L = m
The above is OK provided the data set contains at least 4 variables.
Querying a list
You can determine the number of variables or elements in a list using the function nelem().
list xlist = 1 2 3
nl = nelem(xlist)
The (scalar) variable nl will be assigned a value of 3 since xlist contains 3 members.
You can determine whether a given series is a member of a specied list using the function
inlist(), as in
scalar k = inlist(L, y)
where L is a list and y a series. The series may be specied by name or ID number. The return value
is the (1-based) position of the series in the list, or zero if the series is not present in the list.
Generating lists of transformed variables
Given a named list of variables, you are able to generate lists of transformations of these variables
using the functions log, lags, diff, ldiff, sdiff or dummify. For example
list lxlist = log(xlist)
list difflist = diff(xlist)
When generating a list of lags in this way, you specify the maximum lag order inside the parenthe-
ses, before the list name and separated by a comma. For example
list laglist = lags(2, xlist)
or
scalar order = 4
list laglist = lags(order, xlist)
These commands will populate laglist with the specied number of lags of the variables in xlist.
You can give the name of a single series in place of a list as the second argument to lags: this is
equivalent to giving a list with just one member.
The dummify function creates a set of dummy variables coding for all but one of the distinct values
taken on by the original variable, which should be discrete. (The smallest value is taken as the
omitted catgory.) Like lags, this function returns a list even if the input is a single series.
Another useful operation you can perform with lists is creating interaction variables. Suppose you
have a discrete variable x
i
, taking values from 1 to n and a variable z
i
, which could be continuous
or discrete. In many cases, you want to split z
i
into a set of n variables via the rule
z
(j)
i
=
_
z
i
when x
i
= j
0 otherwise;
in practice, you create dummies for the x
i
variable rst and then you multiply them all by z
i
; these
are commonly called the interactions between x
i
and z
i
. In gretl you can do
list H = D ^ Z
where D is a list of discrete series (or a single discrete series), Z is a list (or a single series)
1
; all the
interactions will be created and listed together under the name H.
An example is provided in script 14.1
Generating series from lists
Once a list is dened, gretl oers several functions that apply to the list and return a series. In most
cases, these functions also apply to single series and behave as natural extensions when applied to
a list, but this is not always the case.
For recognizing and handling missing values, gretl oers several functions (see the Gretl Command
Reference for details). In this context, it is worth remarking that the ok() function can be used
with a list argument. For example,
series xok = ok(xlist)
After these commands, the series xok will have value 1 for observations where none of x1, x2, or
x3 has a missing value, and value 0 for any observations where this condition is not met.
The functions max, min, mean, sd, sum and var behave horizontally rather than vertically when their
argument is a list. For instance, the following commands
list Xlist = x1 x2 x3
series m = mean(Xlist)
produce a series m whose i-th element is the average of x
1,i
, x
2,i
and x
3,i
; missing values, if any, are
implicitly discarded.
In addition, gretl provides three functions for weighted operations: wmean, wsd and wvar. Consider
as an illustration Table 14.1: the rst three columns are GDP per capita for France, Germany and
Italy; columns 4 to 6 contain the population for each country. If we want to compute an aggregate
indicator of per capita GDP, all we have to do is
list Ypc = YpcFR YpcGE YpcIT
list N = NFR NGE NIT
y = wmean(Ypc, N)
so for example
y
1996
=
114.9 59830.635 +124.6 82034.771 +119.3 56890.372
59830.635 +82034.771 +56890.372
= 120.163
See the Gretl Command Reference for more details.
1
Warning: this construct does not work if neither D nor Z are of the the list type.
Example 14.1: Usage of interaction lists
Input:
open mroz87.gdt --quiet
# the coding below makes it so that
# KIDS = 0 -> no kids
# KIDS = 1 -> young kids only
# KIDS = 2 -> young or older kids
series KIDS = (KL6 > 0) + ((KL6 > 0) || (K618 > 0))
list D = CIT KIDS # interaction discrete variables
list X = WE WA # variables to "split"
list INTER = D ^ X
smpl 1 6
print D X INTER -o
Output (selected portions):
CIT KIDS WE WA WE_CIT_0
1 0 2 12 32 12
2 1 1 12 30 0
3 0 2 12 35 12
4 0 1 12 34 12
5 1 2 14 31 0
6 1 0 12 54 0
WE_CIT_1 WA_CIT_0 WA_CIT_1 WE_KIDS_0 WE_KIDS_1
1 0 32 0 0 0
2 12 0 30 0 12
3 0 35 0 0 0
4 0 34 0 0 12
5 14 0 31 0 0
6 12 0 54 12 0
WE_KIDS_2 WA_KIDS_0 WA_KIDS_1 WA_KIDS_2
1 12 0 0 32
2 0 0 30 0
3 12 0 0 35
4 0 0 34 0
5 14 0 0 31
6 0 54 0 0
14.2 Named strings
For some purposes it may be useful to save a string (that is, a sequence of characters) as a named
variable that can be reused.
Some examples of the denition of a string variable are shown below.
string s1 = "some stuff I want to save"
string s2 = getenv("HOME")
YpcFR YpcGE YpcIT NFR NGE NIT
1997 114.9 124.6 119.3 59830.635 82034.771 56890.372
1998 115.3 122.7 120.0 60046.709 82047.195 56906.744
1999 115.0 122.4 117.8 60348.255 82100.243 56916.317
2000 115.6 118.8 117.2 60750.876 82211.508 56942.108
2001 116.0 116.9 118.1 61181.560 82349.925 56977.217
2002 116.3 115.5 112.2 61615.562 82488.495 57157.406
2003 112.1 116.9 111.0 62041.798 82534.176 57604.658
2004 110.3 116.6 106.9 62444.707 82516.260 58175.310
2005 112.4 115.1 105.1 62818.185 82469.422 58607.043
2006 111.9 114.2 103.3 63195.457 82376.451 58941.499
Table 14.1: GDP per capita and population in 3 European countries (Source: Eurostat)
string s3 = s1 + 11
The rst eld after the type-name string is the name under which the string should be saved, then
comes an equals sign, then comes a specication of the string to be saved. This can be the keyword
null, to produce an empty string, or may take any of the following forms:
a string literal (enclosed in double quotes); or
the name of an existing string variable; or
a function that returns a string (see below); or
any of the above followed by + and an integer oset.
The role of the integer oset is to use a substring of the preceding element, starting at the given
character oset. An empty string is returned if the oset is greater than the length of the string in
question.
To add to the end of an existing string you can use the operator +=, as in
string s1 = "some stuff I want to "
string s1 += "save"
or you can use the ~ operator to join two or more strings, as in
string s1 = "sweet"
string s2 = "Home, " ~ s1 ~ " home."
Note that when you dene a string variable using a string literal, no characters are treated as
special (other than the double quotes that delimit the string). Specically, the backslash is not
used as an escape character. So, for example,
string s = "\"
is a valid assignment, producing a string that contains a single backslash character. If you wish to
use backslash-escapes to denote newlines, tabs, embedded double-quotes and so on, use sprintf
instead.
The sprintf command can also be used to dene a string variable. This command works in
exactly the same way as the printf command except that the format string must be preceded
by an identier: either the name of an existing string variable or a new name to which the string
should be assigned. For example,
scalar x = 8
sprintf foo "var%d", x
String variables and string substitution
String variables can be used in two ways in scripting: the name of the variable can be typed as
is, or it may be preceded by the at sign, @. In the rst variant the named string is treated as a
variable in its own right, while the second calls for string substitution. The context determines
which of these variants is appropriate.
In the following contexts the names of string variables should be given in plain form (without the
at sign):
When such a variable appears among the arguments to the commands printf or sprintf.
When such a variable is given as the argument to a function.
On the right-hand side of a string assignment.
Here is an illustration of the use of a named string argument with printf:
? string vstr = "variance"
Generated string vstr
? printf "vstr: %12s\n", vstr
vstr: variance
String substitution can be used in contexts where a string variable is not acceptable as such. If
gretl encounters the symbol @ followed directly by the name of a string variable, this notation is
treated as a macro: the value of the variable is sustituted literally into the command line before
the regular parsing of the command is carried out.
One common use of string substitution is when you want to construct and use the name of a series
programatically. For example, suppose you want to create 10 random normal series named norm1
to norm10. This can be accomplished as follows.
string sname = null
loop i=1..10
sprintf sname "norm%d", i
series @sname = normal()
endloop
Note that plain sname could not be used in the second line within the loop: the eect would be
to attempt to overwrite the string variable named sname with a series of the same name. What
we want is for the current value of sname to be dumped directly into the command that denes a
series, and the @ notation achieves that.
Another typical use of string substitution is when you want the options used with a particular
command to vary depending on some condition. For example,
function void use_optstr (series y, list xlist, int verbose)
string optstr = verbose ? "" : "--simple-print"
ols y xlist @optstr
end function
open data4-1
list X = const sqft
use_optstr(price, X, 1)
use_optstr(price, X, 0)
When printing the value of a string variable using the print command, the plain variable name
should generally be used, as in
string s = "Just testing"
print s
The following variant is equivalent, though clumsy.
print "@s"
But note that this next variant does something quite dierent.
print @s
After string substitution, the print command reads
print Just testing
which attempts to print the values of two variables, Just and testing.
Built-in strings
Apart from any strings that the user may dene, some string variables are dened by gretl itself.
These may be useful for people writing functions that include shell commands. The built-in strings
are as shown in Table 14.2.
gretldir the gretl installation directory
workdir users current gretl working directory
dotdir the directory gretl uses for temporary les
gnuplot path to, or name of, the gnuplot executable
tramo path to, or name of, the tramo executable
x12a path to, or name of, the x-12-arima executable
tramodir tramo data directory
x12adir x-12-arima data directory
Table 14.2: Built-in string variables
Reading strings from the environment
In addition, it is possible to read into gretls named strings, values that are dened in the external
environment. To do this you use the function getenv, which takes the name of an environment
variable as its argument. For example:
? string user = getenv("USER")
Generated string user
? string home = getenv("HOME")
Generated string home
? print "@users home directory is @home"
cottrells home directory is /home/cottrell
To check whether you got a non-empty value from a given call to getenv, you can use the function
strlen, which retrieves the length of the string, as in
? string temp = getenv("TEMP")
Generated string temp
? scalar x = strlen(temp)
Generated scalar x = 0
Capturing strings via the shell
If shell commands are enabled in gretl, you can capture the output from such commands using the
syntax
string stringname = $(shellcommand)
That is, you enclose a shell command in parentheses, preceded by a dollar sign.
Reading from a le into a string
You can read the content of a le into a string variable using the syntax
string stringname = readfile(lename)
The lename eld may be given as a string variable. For example
? sprintf fname "%s/QNC.rts", x12adir
Generated string fname
? string foo = readfile(fname)
Generated string foo
The above could also be accomplished using the macro variant of a string variable, provided it is
placed in quotation marks:
string foo = readfile("@x12adir/QNC.rts")
The strstr function
Invocation of this function takes the form
string stringname = strstr(s1, s2)
The eect is to search s1 for the rst occurrence of s2. If no such occurrence is found, an empty
string is returned; otherwise the portion of s1 starting with s2 is returned. For example:
? string hw = "hello world"
Generated string hw
? string w = strstr(hw, "o")
Generated string w
? print w
o world
Chapter 15
Matrix manipulation
Together with the other two basic types of data (series and scalars), gretl oers a quite compre-
hensive array of matrix methods. This chapter illustrates the peculiarities of matrix syntax and
discusses briey some of the more complex matrix functions. For a full listing of matrix functions
and a comprehensive account of their syntax, please refer to the Gretl Command Reference.
15.1 Creating matrices
Matrices can be created using any of these methods:
1. By direct specication of the scalar values that compose the matrix in numerical form, by
reference to pre-existing scalar variables, or using computed values.
2. By providing a list of data series.
3. By providing a named list of series.
4. Using a formula of the same general type that is used with the genr command, whereby a new
matrix is dened in terms of existing matrices and/or scalars, or via some special functions.
To specify a matrix directly in terms of scalars, the syntax is, for example:
matrix A = { 1, 2, 3 ; 4, 5, 6 }
The matrix is dened by rows; the elements on each row are separated by commas and the rows
are separated by semi-colons. The whole expression must be wrapped in braces. Spaces within the
braces are not signicant. The above expression denes a 2 3 matrix. Each element should be a
numerical value, the name of a scalar variable, or an expression that evaluates to a scalar. Directly
after the closing brace you can append a single quote () to obtain the transpose.
To specify a matrix in terms of data series the syntax is, for example,
matrix A = { x1, x2, x3 }
where the names of the variables are separated by commas. Besides names of existing variables,
you can use expressions that evaluate to a series. For example, given a series x you could do
matrix A = { x, x^2 }
Each variable occupies a column (and there can only be one variable per column). You cannot use
the semicolon as a row separator in this case: if you want the series arranged in rows, append the
transpose symbol. The range of data values included in the matrix depends on the current setting
of the sample range.
Instead of giving an explicit list of variables, you may instead provide the name of a saved list (see
Chapter 14), as in
118
Chapter 15. Matrix manipulation 119
matrix A = { xlist }
When you provide a named list, the data series are by default placed in columns, as is natural in an
econometric context: if you want them in rows, append the transpose symbol.
As a special case of constructing a matrix from a list of variables, you can say
matrix A = { dataset }
This builds a matrix using all the series in the current dataset, apart from the constant (variable 0).
When this dummy list is used, it must be the sole element in the matrix denition {...}. You can,
however, create a matrix that includes the constant along with all other variables using horizontal
concatenation (see below), as in
matrix A = {const}~{dataset}
By default, when you build a matrix from series that include missing values the data rows that
contain NAs are skipped. But you can modify this behavior via the command set skip_missing
off. In that case NAs are converted to NaN (Not a Number). In the IEEE oating-point stan-
dard, arithmetic operations involving NaN always produce NaN. Alternatively, you can take greater
control over the observations (data rows) that are included in the matrix using the set variable
matrix_mask, as in
set matrix_mask msk
where msk is the name of a series. Subsequent commands that formmatrices fromseries or lists will
include only observations for which msk has non-zero (and non-missing) values. You can remove
this mask via the command set matrix_mask null.
Names of matrices must satisfy the same requirements as names of gretl variables in general: the name
can be no longer than 31 characters, must start with a letter, and must be composed of nothing but letters,
numbers and the underscore character.
15.2 Empty matrices
The syntax
matrix A = {}
creates an empty matrixa matrix with zero rows and zero columns.
The main purpose of the concept of an empty matrix is to enable the user to dene a starting point
for subsequent concatenation operations. For instance, if X is an already dened matrix of any size,
the commands
matrix A = {}
matrix B = A ~ X
result in a matrix B identical to X.
From an algebraic point of view, one can make sense of the idea of an empty matrix in terms of
vector spaces: if a matrix is an ordered set of vectors, then A={} is the empty set. As a consequence,
operations involving addition and multiplications dont have any clear meaning (arguably, they have
none at all), but operations involving the cardinality of this set (that is, the dimension of the space
spanned by A) are meaningful.
Legal operations on empty matrices are listed in Table 15.1. (All other matrix operations gener-
ate an error when an empty matrix is given as an argument.) In line with the above interpreta-
tion, some matrix functions return an empty matrix under certain conditions: the functions diag,
vec, vech, unvech when the arguments is an empty matrix; the functions I, ones, zeros,
mnormal, muniform when one or more of the arguments is 0; and the function nullspace when
its argument has full column rank.
Function Return value
A, transp(A) A
rows(A) 0
cols(A) 0
rank(A) 0
det(A) NA
ldet(A) NA
tr(A) NA
onenorm(A) NA
infnorm(A) NA
rcond(A) NA
Table 15.1: Valid functions on an empty matrix, A
15.3 Selecting sub-matrices
You can select sub-matrices of a given matrix using the syntax
A[rows,cols]
where rows can take any of these forms:
1. empty selects all rows
2. a single integer selects the single specied row
3. two integers separated by a colon selects a range of rows
4. the name of a matrix selects the specied rows
With regard to option 2, the integer value can be given numerically, as the name of an existing
scalar variable, or as an expression that evaluates to a scalar. With option 4, the index matrix given
in the rows eld must be either p 1 or 1 p, and should contain integer values in the range 1 to
n, where n is the number of rows in the matrix from which the selection is to be made.
The cols specication works in the same way, mutatis mutandis. Here are some examples.
matrix B = A[1,]
matrix B = A[2:3,3:5]
matrix B = A[2,2]
matrix idx = { 1, 2, 6 }
matrix B = A[idx,]
The rst example selects row 1 from matrix A; the second selects a 23 submatrix; the third selects
a scalar; and the fourth selects rows 1, 2, and 6 from matrix A.
If the matrix in question is n 1 or 1 m, it is OK to give just one index specier and omit the
comma. For example, A[2] selects the second element of A if A is a vector. Otherwise the comma
is mandatory.
In addition there is a pre-dened index specication, diag, which selects the principal diagonal of
a square matrix, as in B[diag], where B is square.
You can use selections of this sort on either the right-hand side of a matrix-generating formula or
the left. Here is an example of use of a selection on the right, to extract a 2 2 submatrix B from a
3 3 matrix A:
matrix A = { 1, 2, 3; 4, 5, 6; 7, 8, 9 }
matrix B = A[1:2,2:3]
And here are examples of selection on the left. The second line below writes a 22 identity matrix
into the bottom right corner of the 3 3 matrix A. The fourth line replaces the diagonal of A with
1s.
matrix A = { 1, 2, 3; 4, 5, 6; 7, 8, 9 }
matrix A[2:3,2:3] = I(2)
matrix d = { 1, 1, 1 }
matrix A[diag] = d
15.4 Matrix operators
The following binary operators are available for matrices:
+ addition
- subtraction
* ordinary matrix multiplication
pre-multiplication by transpose
\ matrix left division (see below)
/ matrix right division (see below)
~ column-wise concatenation
| row-wise concatenation
** Kronecker product
= test for equality
In addition, the following operators (dot operators) apply on an element-by-element basis:
.+ .- .* ./ .^ .= .> .< .>= .<=
Here are explanations of the less obvious cases.
For matrix addition and subtraction, in general the two matrices have to be of the same dimensions
but an exception to this rule is granted if one of the operands is a 11 matrix or scalar. The scalar
is implicitly promoted to the status of a matrix of the correct dimensions, all of whose elements
are equal to the given scalar value. For example, if A is an m n matrix and k a scalar, then the
commands
matrix C = A + k
matrix D = A - k
both produce mn matrices, with elements c
ij
= a
ij
+k and d
ij
= a
ij
k respectively.
By pre-multiplication by transpose we mean, for example, that
matrix C = XY
produces the product of X-transpose and Y. In eect, the expression XY is shorthand for X*Y
(which is also valid).
In matrix left division, the statement
matrix X = A \ B
is interpreted as a request to nd the matrix X that solves AX = B. If B is a square matrix, this is
in principle equivalent to A
1
B, which fails if A is singular; the numerical method employed here
is the LU decomposition. If A is a T k matrix with T > k, then X is the least-squares solution,
X = (A
A)
1
A
B, which fails if A
A is singular; the numerical method employed here is the QR

decomposition. Otherwise, the operation necessarily fails.
For matrix right division, as in X = A / B, X is the matrix that solves XB = A, in principle
equivalent to AB
1
.
In dot operations a binary operation is applied element by element; the result of this operation
is obvious if the matrices are of the same size. However, there are several other cases where such
operators may be applied. For example, if we write
matrix C = A .- B
then the result C depends on the dimensions of A and B. Let A be an m n matrix and let B be
p q; the result is as follows:
Case Result
Dimensions match (m= p and n = q) c
ij
= a
ij
b
ij
A is a column vector; rows match (m= p; n = 1) c
ij
= a
i
b
ij
B is a column vector; rows match (m= p; q = 1) c
ij
= a
ij
b
i
A is a row vector; columns match (m= 1; n = q) c
ij
= a
j
b
ij
B is a row vector; columns match (m= p; q = 1) c
ij
= a
ij
b
j
A is a column vector; B is a row vector (n = 1; p = 1) c
ij
= a
i
b
j
A is a row vector; B is a column vector (m= 1; q = 1) c
ij
= a
j
b
i
A is a scalar (m= 1 and n = 1) c
ij
= a b
ij
B is a scalar (p = 1 and q = 1) c
ij
= a
ij
b
If none of the above conditions are satised the result is undened and an error is agged.
Note that this convention makes it unnecessary, in most cases, to use diagonal matrices to perform
transformations by means of ordinary matrix multiplication: if Y = XV, where V is diagonal, it is
computationally much more convenient to obtain Y via the instruction
matrix Y = X .* v
where v is a row vector containing the diagonal of V.
In column-wise concatenation of an mn matrix A and an mp matrix B, the result is an m(n+p)
matrix. That is,
matrix C = A ~ B
produces C =
_
A B
_
.
Row-wise concatenation of an m n matrix A and an p n matrix B produces an (m+ p) n
matrix. That is,
matrix C = A | B
produces C =
_
A
B
_
.
15.5 Matrixscalar operators
For matrix A and scalar k, the operators shown in Table 15.2 are available. (Addition and subtrac-
tion were discussed in section 15.4 but we include them in the table for completeness.) In addition,
for square A and integer k 0, B = A^k produces a matrix B which is A raised to the power k.
Expression Eect
matrix B = A * k b
ij
= ka
ij
matrix B = A / k b
ij
= a
ij
/k
matrix B = k / A b
ij
= k/a
ij
matrix B = A + k b
ij
= a
ij
+k
matrix B = A - k b
ij
= a
ij
k
matrix B = k - A b
ij
= k a
ij
matrix B = A % k b
ij
= a
ij
modulo k
Table 15.2: Matrixscalar operators
15.6 Matrix functions
Most of the gretl functions available for scalars and series also apply to matrices in an element-by-
element fashion, and as such their behavior should be pretty obvious. This is the case for functions
such as log, exp, sin, etc. These functions have the eects documented in relation to the genr
command. For example, if a matrix A is already dened, then
matrix B = sqrt(A)
generates a matrix such that b
ij
=
a
ij
. All such functions require a single matrix as argument, or
an expression which evaluates to a single matrix.
1
In this section, we review some aspects of genr functions that apply specically to matrices. A full
account of each function is available in the Gretl Command Reference.
Matrix reshaping
In addition to the methods discussed in sections 15.1 and 15.3, a matrix can also be created by
re-arranging the elements of a pre-existing matrix. This is accomplished via the mshape function.
It takes three arguments: the input matrix, A, and the rows and columns of the target matrix, r
and c respectively. Elements are read from A and written to the target in column-major order. If A
contains fewer elements than n = r c, they are repeated cyclically; if A has more elements, only
the rst n are used.
For example:
matrix a = mnormal(2,3)
a
matrix b = mshape(a,3,1)
b
1
Note that to nd the matrix square root you need the cholesky function (see below); moreover, the exp function
computes the exponential element by element, and therefore does not return the matrix exponential unless the matrix is
diagonal to get the matrix exponential, use mexp.
Creation and I/O
colnames diag diagcat I lower mnormal
mread muniform ones rownames seq unvech
upper vec vech zeros
Shape/size/arrangement
cols dsort mreverse mshape msortby rows
selifc selifr sort trimr
Matrix algebra
cdiv cholesky cmult det eigengen eigensym
eigsolve fft ffti ginv hdprod infnorm
inv invpd ldet mexp nullspace onenorm
polroots psdroot qform qrdecomp rank rcond
svd toepsolv tr transp varsimul
Statistics/transformations
aggregate cdemean corr corrgm cov fcstats
ghk halton imaxc imaxr iminc iminr
irf iwishart kdensity maxc maxr mcorr
mcov mcovg meanc meanr minc minr
mlag mols mpols mrls mxtab pergm
princomp prodc prodr quadtable quantile ranking
resample sdc sumc sumr uniq values
Data utilities
isconst mwrite ok pshrink replace
Filters
filter kfilter ksimul ksmooth lrvar
Numerical methods
BFGSmax fdjac NRmax simann
Strings
colname
Transformations
chowlin cum lincomb
Table 15.3: Matrix functions by category
matrix b = mshape(a,5,2)
b
produces
? a
a
1.2323 0.99714 -0.39078
0.54363 0.43928 -0.48467
? matrix b = mshape(a,3,1)
Generated matrix b
? b
b
1.2323
0.54363
0.99714
? matrix b = mshape(a,5,2)
Replaced matrix b
? b
b
1.2323 -0.48467
0.54363 1.2323
0.99714 0.54363
0.43928 0.99714
-0.39078 0.43928
Complex multiplication and division
Gretl has no native provision for complex numbers. However, basic operations can be performed
on vectors of complex numbers by using the convention that a vector of n complex numbers is
represented as a n 2 matrix, where the rst column contains the real part and the second the
imaginary part.
Addition and subtraction are trivial; the functions cmult and cdiv compute the complex product
and division, respectively, of two input matrices, A and B, representing complex numbers. These
matrices must have the same number of rows, n, and either one or two columns. The rst column
contains the real part and the second (if present) the imaginary part. The return value is an n 2
matrix, or, if the result has no imaginary part, an n-vector.
For example, suppose you have z
1
= [1 +2i, 3 +4i]
and z
2
= [1, i]
:
? z1 = {1,2;3,4}
z1 = {1,2;3,4}
Generated matrix z1
? z2 = I(2)
z2 = I(2)
Generated matrix z2
? conj_z1 = z1 .* {1,-1}
conj_z1 = z1 .* {1,-1}
Generated matrix conj_z1
? eval cmult(z1,z2)
eval cmult(z1,z2)
1 2
-4 3
? eval cmult(z1,conj_z1)
eval cmult(z1,conj_z1)
5
25
Multiple returns and the null keyword
Some functions take one or more matrices as arguments and compute one or more matrices; these
are:
eigensym Eigen-analysis of symmetric matrix
eigengen Eigen-analysis of general matrix
mols Matrix OLS
qrdecomp QR decomposition
svd Singular value decomposition (SVD)
The general rule is: the main result of the function is always returned as the result proper.
Auxiliary returns, if needed, are retrieved using pre-existing matrices, which are passed to the
function as pointers (see 13.4). If such values are not needed, the pointer may be substituted with
the keyword null.
The syntax for qrdecomp, eigensym and eigengen is of the form
matrix B = func(A, &C)
The rst argument, A, represents the input data, that is, the matrix whose decomposition or analysis
is required. The second argument must be either the name of an existing matrix preceded by & (to
indicate the address of the matrix in question), in which case an auxiliary result is written to that
matrix, or the keyword null, in which case the auxiliary result is not produced, or is discarded.
In case a non-null second argument is given, the specied matrix will be over-written with the
auxiliary result. (It is not required that the existing matrix be of the right dimensions to receive the
result.)
The function eigensym computes the eigenvalues, and optionally the right eigenvectors, of a sym-
metric n n matrix. The eigenvalues are returned directly in a column vector of length n; if the
eigenvectors are required, they are returned in an nn matrix. For example:
matrix V
matrix E = eigensym(M, &V)
matrix E = eigensym(M, null)
In the rst case E holds the eigenvalues of M and V holds the eigenvectors. In the second, E holds
the eigenvalues but the eigenvectors are not computed.
The function eigengen computes the eigenvalues, and optionally the eigenvectors, of a general
nn matrix. The eigenvalues are returned directly in an n2 matrix, the rst column holding the
real components and the second column the imaginary components.
If the eigenvectors are required (that is, if the second argument to eigengen is not null), they
are returned in an n n matrix. The column arrangement of this matrix is somewhat non-trivial:
the eigenvectors are stored in the same order as the eigenvalues, but the real eigenvectors occupy
one column, whereas complex eigenvectors take two (the real part comes rst); the total num-
ber of columns is still n, because the conjugate eigenvector is skipped. Example 15.1 provides a
(hopefully) clarifying example (see also subsection 15.6).
Example 15.1: Complex eigenvalues and eigenvectors
set seed 34756
matrix v
A = mnormal(3,3)
/* do the eigen-analysis */
l = eigengen(A,&v)
/* eigenvalue 1 is real, 2 and 3 are complex conjugates */
print l
print v
/*
column 1 contains the first eigenvector (real)
*/
B = A*v[,1]
c = l[1,1] * v[,1]
/* B should equal c */
print B
print c
/*
columns 2:3 contain the real and imaginary parts
of eigenvector 2
*/
B = A*v[,2:3]
c = cmult(ones(3,1)*(l[2,]),v[,2:3])
/* B should equal c */
print B
print c
The qrdecomp function computes the QR decomposition of an mn matrix A: A = QR, where Q
is an mn orthogonal matrix and R is an nn upper triangular matrix. The matrix Q is returned
directly, while R can be retrieved via the second argument. Here are two examples:
matrix R
matrix Q = qrdecomp(M, &R)
matrix Q = qrdecomp(M, null)
In the rst example, the triangular R is saved as R; in the second, R is discarded. The rst line
above shows an example of a simple declaration of a matrix: R is declared to be a matrix variable
but is not given any explicit value. In this case the variable is initialized as a 1 1 matrix whose
single element equals zero.
The syntax for svd is
matrix B = func(A, &C, &D)
The function svd computes all or part of the singular value decomposition of the real mn matrix
A. Let k = min(m, n). The decomposition is
A = UV
where U is an mk orthogonal matrix, is an kk diagonal matrix, and V is an kn orthogonal

matrix.
2
The diagonal elements of are the singular values of A; they are real and non-negative,
and are returned in descending order. The rst k columns of U and V are the left and right singular
vectors of A.
The svd function returns the singular values, in a vector of length k. The left and/or right singu-
lar vectors may be obtained by supplying non-null values for the second and/or third arguments
respectively. For example:
matrix s = svd(A, &U, &V)
matrix s = svd(A, null, null)
matrix s = svd(A, null, &V)
In the rst case both sets of singular vectors are obtained, in the second case only the singular
values are obtained; and in the third, the right singular vectors are obtained but U is not computed.
Please note: when the third argument is non-null, it is actually V
that is provided. To reconstitute

the original matrix from its SVD, one can do:
matrix s = svd(A, &U, &V)
matrix B = (U.*s)*V
Finally, the syntax for mols is
matrix B = mols(Y, X, &U)
This function returns the OLS estimates obtained by regressing the T n matrix Y on the T k
matrix X, that is, a k n matrix holding (X
X)
1
X
Y. The Cholesky decomposition is used. The

matrix U, if not null, is used to store the residuals.
Reading and writing matrices from/to text les
The two functions mread and mwrite can be used for basic matrix input/output. This can be useful
to enable gretl to exchange data with other programs.
2
This is not the only denition of the SVD: some writers dene U as m m, as m n (with k non-zero diagonal
elements) and V as nn.
The mread function accepts one string parameter: the name of the (plain text) le from which the
matrix is to be read. The le in question may start with any number of comment lines, dened
as lines that start with the hash mark, #; such lines are ignored. Beyond that, the content must
conform to the following rules:
1. The rst non-comment line must contain two integers, separated by a space or a tab, indicat-
ing the number of rows and columns, respectively.
2. The columns must be separated by spaces or tab characters.
3. The decimal separator must be the dot . character.
Should an error occur (such as the le being badly formatted or inaccessible), an empty matrix (see
section 15.2) is returned.
The complementary function mwrite produces text les formatted as described above. The column
separator is the tab character, so import into spreadsheets should be straightforward. Usage is
illustrated in example 15.2. Matrices stored via the mwrite command can be easily read by other
programs; the following table summarizes the appropriate commands for reading a matrix A from
a le called a.mat in some widely-used programs.
3
Note that the Python example requires that the
numpy module is loaded.
Program Sample code
GAUSS tmp[] = load a.mat;
A = reshape(tmp[3:rows(tmp)],tmp[1],tmp[2]);
Octave fd = fopen("a.mat");
[r,c] = fscanf(fd, "%d %d", "C");
A = reshape(fscanf(fd, "%g", r*c),c,r);
fclose(fd);
Ox decl A = loadmat("a.mat");
R A <- as.matrix(read.table("a.mat", skip=1))
Python A = numpy.loadtxt(a.mat, skiprows=1)
Optionally, the mwrite and mread functions can use gzip compression: this is invoked if the name
of the matrix le has the sux .gz. In this case the elements of the matrix are written in a single
column. Note, however, that compression should not be applied when writing matrices for reading
by third-party software unless you are sure that the software can handle compressed data.
15.7 Matrix accessors
In addition to the matrix functions discussed above, various accessor strings allow you to create
copies of internal matrices associated with models previously estimated. These are set out in
Table 15.4.
Many of the accessors in Table 15.4 behave somewhat dierently depending on the sort of model
that is referenced, as follows:
Single-equation models: $sigma gets a scalar (the standard error of the regression); $coeff
and $stderr get column vectors; $uhat and $yhat get series.
System estimators: $sigma gets the cross-equation residual covariance matrix; $uhat and
$yhat get matrices with one column per equation. The format of $coeff and $stderr de-
pends on the nature of the system: for VARs and VECMs (where the matrix of regressors is
3
Matlab users may nd the Octave example helpful, since the two programs are mostly compatible with one another.
Example 15.2: Matrix input/output via text les
nulldata 64
scalar n = 3
string f1 = "a.csv"
string f2 = "b.csv"
matrix a = mnormal(n,n)
matrix b = inv(a)
err = mwrite(a, f1)
if err != 0
fprintf "Failed to write %s\n", f1
else
err = mwrite(b, f2)
endif
if err != 0
fprintf "Failed to write %s\n", f2
else
c = mread(f1)
d = mread(f2)
a = c*d
printf "The following matrix should be an identity matrix\n"
print a
endif
$coeff matrix of estimated coecients
$compan companion matrix (after VAR or VECM estimation)
$jalpha matrix (loadings) from Johansens procedure
$jbeta matrix (cointegration vectors) from Johansens procedure
$jvbeta covariance matrix for the unrestricted elements of from Johansens procedure
$rho autoregressive coecients for error process
$sigma residual covariance matrix
$stderr matrix of estimated standard errors
$uhat matrix of residuals
$vcv covariance matrix of parameter estimates
$vma VMA matrices in stacked form (see section 26.2)
$yhat matrix of tted values
Table 15.4: Matrix accessors for model data
the same for all equations) these return matrices with one column per equation, but for other
system estimators they return a big column vector.
VARs and VECMs: $vcv is not available, but X
X
1
(where X is the common matrix of regres-
sors) is available as $xtxinv.
If the accessors are given without any prex, they retrieve results from the last model estimated, if
any. Alternatively, they may be prexed with the name of a saved model plus a period (.), in which
case they retrieve results from the specied model. Here are some examples:
matrix u = $uhat
matrix b = m1.$coeff
matrix v2 = m1.$vcv[1:2,1:2]
The rst command grabs the residuals from the last model; the second grabs the coecient vector
from model m1; and the third (which uses the mechanism of sub-matrix selection described above)
grabs a portion of the covariance matrix from model m1.
If the model in question a VAR or VECM (only) $compan and $vma return the companion matrix and
the VMA matrices in stacked form, respectively (see section 26.2 for details). After a vector error
correction model is estimated via Johansens procedure, the matrices $jalpha and $jbeta are also
available. These have a number of columns equal to the chosen cointegration rank; therefore, the
product
matrix Pi = $jalpha * $jbeta
returns the reduced-rank estimate of A(1). Since is automatically identied via the Phillips nor-
malization (see section 27.5), its unrestricted elements do have a proper covariance matrix, which
can be retrieved through the $jvbeta accessor.
15.8 Namespace issues
Matrices share a common namespace with data series and scalar variables. In other words, no two
objects of any of these types can have the same name. It is an error to attempt to change the type
of an existing variable, for example:
scalar x = 3
matrix x = ones(2,2) # wrong!
It is possible, however, to delete or rename an existing variable then reuse the name for a variable
of a dierent type:
scalar x = 3
delete x
matrix x = ones(2,2) # OK
15.9 Creating a data series from a matrix
Section 15.1 above describes how to create a matrix from a data series or set of series. You may
sometimes wish to go in the opposite direction, that is, to copy values from a matrix into a regular
data series. The syntax for this operation is
series sname = mspec
where sname is the name of the series to create and mspec is the name of the matrix to copy from,
possibly followed by a matrix selection expression. Here are two examples.
series s = x
series u1 = U[,1]
It is assumed that x and U are pre-existing matrices. In the second example the series u1 is formed
from the rst column of the matrix U.
For this operation to work, the matrix (or matrix selection) must be a vector with length equal to
either the full length of the current dataset, n, or the length of the current sample range, n
. If
n
< n then only n
elements are drawn from the matrix; if the matrix or selection comprises n
elements, the n
values starting at element t

1
are used, where t
1
represents the starting observation
of the sample range. Any values in the series that are not assigned from the matrix are set to the
missing code.
15.10 Matrices and lists
To facilitate the manipulation of named lists of variables (see Chapter 14), it is possible to convert
between matrices and lists. In section 15.1 above we mentioned the facility for creating a matrix
from a list of variables, as in
matrix M = { listname }
That formulation, with the name of the list enclosed in braces, builds a matrix whose columns hold
the variables referenced in the list. What we are now describing is a dierent matter: if we say
matrix M = listname
(without the braces), we get a row vector whose elements are the ID numbers of the variables in the
list. This special case of matrix generation cannot be embedded in a compound expression. The
syntax must be as shown above, namely simple assignment of a list to a matrix.
To go in the other direction, you can include a matrix on the right-hand side of an expression that
denes a list, as in
list Xl = M
where M is a matrix. The matrix must be suitable for conversion; that is, it must be a row or column
vector containing non-negative whole-number values, none of which exceeds the highest ID number
of a variable (series or scalar) in the current dataset.
Example 15.3 illustrates the use of this sort of conversion to normalize a list, moving the constant
(variable 0) to rst position.
15.11 Deleting a matrix
To delete a matrix, just write
delete M
where M is the name of the matrix to be deleted.
15.12 Printing a matrix
To print a matrix, the easiest way is to give the name of the matrix in question on a line by itself,
which is equivalent to using the print command:
Example 15.3: Manipulating a list
function void normalize_list (matrix *x)
# If the matrix (representing a list) contains var 0,
# but not in first position, move it to first position
if (x[1] != 0)
scalar k = cols(x)
loop for (i=2; i<=k; i++) --quiet
if (x[i] = 0)
x[i] = x[1]
x[1] = 0
break
endif
endloop
endif
end function
open data9-7
list Xl = 2 3 0 4
matrix x = Xl
normalize_list(&x)
list Xl = x
matrix M = mnormal(100,2)
M
print M
You can get ner control on the formatting of output by using the printf command, as illustrated
in the interactive session below:
? matrix Id = I(2)
matrix Id = I(2)
Generated matrix Id
? print Id
print Id
Id (2 x 2)
1 0
0 1
? printf "%10.3f", Id
1.000 0.000
0.000 1.000
For presentation purposes you may wish to give titles to the columns of a matrix. For this you can
use the colnames function: the rst argument is a matrix and the second is either a named list of
variables, whose names will be used as headings, or a string that contains as many space-separated
substrings as the matrix has columns. For example,
? matrix M = mnormal(3,3)
? colnames(M, "foo bar baz")
? print M
M (3 x 3)
foo bar baz
1.7102 -0.76072 0.089406
-0.99780 -1.9003 -0.25123
-0.91762 -0.39237 -1.6114
15.13 Example: OLS using matrices
Example 15.4 shows how matrix methods can be used to replicate gretls built-in OLS functionality.
Example 15.4: OLS via matrix methods
open data4-1
matrix X = { const, sqft }
matrix y = { price }
matrix b = invpd(XX) * Xy
print "estimated coefficient vector"
b
matrix u = y - X*b
scalar SSR = uu
scalar s2 = SSR / (rows(X) - rows(b))
matrix V = s2 * inv(XX)
V
matrix se = sqrt(diag(V))
print "estimated standard errors"
se
# compare with built-in function
ols price const sqft --vcv
Chapter 16
Cheat sheet
This chapter explains how to perform some commonand some not so commontasks in gretls
scripting language, hansl. Some but not all of the techniques listed here are also available through
the graphical interface. Although the graphical interface may be more intuitive and less intimidat-
ing at rst, we encourage users to take advantage of the power of gretls scripting language as soon
as they feel comfortable with the program.
16.1 Dataset handling
Weird periodicities
Problem: You have data sampled each 3 minutes from 9am onwards; youll probably want to specify
the hour as 20 periods.
Solution:
setobs 20 9:1 --special
Comment: Now functions like sdiff() (seasonal dierence) or estimation methods like seasonal
ARIMA will work as expected.
Generating a panel dataset of given dimensions
Problem: You want to generate via nulldata a panel dataset and specify in advance the number of
units and the time length of your series via two scalar variables.
Solution:
scalar n_units = 100
scalar T = 12
scalar NT = T * n_units
nulldata NT --preserve
setobs T 1:1 --stacked-time-series
Comment: The essential ingredient that we use here is the --preserve option: it protects existing
scalars (and matrices, for that matter) from being trashed by nulldata, thus making it possible to
use the scalar T in the setobs command.
Help, my data are backwards!
Problem: Gretl expects time series data to be in chronological order (most recent observation last),
but you have imported third-party data that are in reverse order (most recent rst).
Solution:
setobs 1 1 --cross-section
genr sortkey = -obs
dataset sortby sortkey
setobs 1 1950 --time-series
135
Chapter 16. Cheat sheet 136
Comment: The rst line is required only if the data currently have a time series interpretation: it
removes that interpretation, because (for fairly obvious reasons) the dataset sortby operation is
not allowed for time series data. The following two lines reverse the data, using the negative of the
built-in index variable obs. The last line is just illustrative: it establishes the data as annual time
series, starting in 1950.
If you have a dataset that is mostly the right way round, but a particular variable is wrong, you can
reverse that variable as follows:
genr x = sortby(-obs, x)
Dropping missing observations selectively
Problem: You have a dataset with many variables and want to restrict the sample to those observa-
tions for which there are no missing observations for the variables x1, x2 and x3.
Solution:
list X = x1 x2 x3
smpl --no-missing X
Comment: You can now save the le via a store command to preserve a subsampled version of
the dataset. Alternative solutions based on the ok function, such as
list X = x1 x2 x3
genr sel = ok(X)
smpl sel --restrict
are perhaps less obvious, but more exible. Pick your poison.
By operations
Problem: You have a discrete variable d and you want to run some commands (for example, estimate
a model) by splitting the sample according to the values of d.
Solution:
matrix vd = values(d)
m = rows(vd)
loop i=1..m
scalar sel = vd[i]
smpl d==sel --restrict --replace
ols y const x
endloop
smpl --full
Comment: The main ingredient here is a loop. You can have gretl perform as many instructions as
you want for each value of d, as long as they are allowed inside a loop. Note, however, that if all
you want is descriptive statistics, the summary command does have a --by option.
Adding a time series to a panel
Problem: You have a panel dataset (comprising observations of n indidivuals in each of T periods)
and you want to add a variable which is available in straight time-series form. For example, you
want to add annual CPI data to a panel in order to deate nominal income gures.
In gretl a panel is represented in stacked time-series format, so in eect the task is to create a new
variable which holds n stacked copies of the original time series. Lets say the panel comprises 500
individuals observed in the years 1990, 1995 and 2000 (n = 500, T = 3), and we have these CPI
data in the ASCII le cpi.txt:
date cpi
1990 130.658
1995 152.383
2000 172.192
What we need is for the CPI variable in the panel to repeat these three values 500 times.
Solution: Simple! With the panel dataset open in gretl,
append cpi.txt
Comment: If the length of the time series is the same as the length of the time dimension in the
panel (3 in this example), gretl will perform the stacking automatically. Rather than using the
append command you could use the Append data item under the File menu in the GUI program.
For this to work, your main dataset must be recognized as a panel. This can be arranged via the
setobs command or the Dataset structure item under the Data menu.
16.2 Creating/modifying variables
Generating a dummy variable for a specic observation
Problem: Generate d
t
= 0 for all observation but one, for which d
t
= 1.
Solution:
genr d = (t=="1984:2")
Comment: The internal variable t is used to refer to observations in string form, so if you have a
cross-section sample you may just use d = (t=="123"). If the dataset has observation labels you
can use the corresponding label. For example, if you open the dataset mrw.gdt, supplied with gretl
among the examples, a dummy variable for Italy could be generated via
genr DIta = (t=="Italy")
Note that this method does not require scripting at all. In fact, you might as well use the GUI Menu
Add/Dene new variable for the same purpose, with the same syntax.
Generating an ARMA(1,1)
Problem: Generate y
t
= 0.9y
t1
+
t
0.5
t1
, with
t
NIID(0, 1).
Recommended solution:
alpha = 0.9
theta = -0.5
series y = filter(normal(), {1, theta}, alpha)
Bread and butter solution:
alpha = 0.9
theta = -0.5
series e = normal()
series y = 0
series y = alpha * y(-1) + e + theta * e(-1)
Comment: The filter function is specically designed for this purpose so in most cases youll
want to take advantage of its speed and exibility. That said, in some cases you may want to
generate the series in a manner which is more transparent (maybe for teaching purposes).
In the second solution, the statement series y = 0 is necessary because the next statement eval-
uates y recursively, so y[1] must be set. Note that you must use the keyword series here instead
of writing genr y = 0 or simply y = 0, to ensure that y is a series and not a scalar.
Recoding a variable
Problem: You want to perform a 1-to-1 recode on a variable. For example, consider tennis points:
you may have a variable x holding values 1 to 3 and you want to recode it to 15, 30, 40.
Solution 1:
series x = replace(x, 1, 15)
Solution 2:
matrix tennis = {15, 30, 40}
series x = replace(x, seq(1,3), tennis)
Comment: There are many equivalent ways to achieve the same eect, but for simple cases such
as this, the replace function is simple and transparent. If you dont mind using matrices, scripts
using replace can also be remarkably compact.
Recoding a variable by classes
Problem: You want to recode a variable by classes. For example, you have the age of a sample of
individuals (x
i
) and you need to compute age classes (y
i
) as
y
i
= 1 for x
i
< 18
y
i
= 2 for 18 x
i
< 65
y
i
= 3 for x
i
65
Solution:
series y = 1 + (x >= 18) + (x >= 65)
Comment: True and false expressions are evaluated as 1 and 0 respectively, so they can be ma-
nipulated algebraically as any other number. The same result could also be achieved by using the
conditional assignment operator (see below), but in most cases it would probably lead to more
convoluted constructs.
Conditional assignment
Problem: Generate y
t
via the following rule:
y
t
=
_
x
t
for d
t
> a
z
t
for d
t
a
Solution:
series y = (d > a) ? x : z
Comment: There are several alternatives to the one presented above. One is a brute force solution
using loops. Another one, more ecient but still suboptimal, would be
series y = (d>a)*x + (d<=a)*z
However, the ternary conditional assignment operator is not only the most numerically ecient
way to accomplish what we want, it is also remarkably transparent to read when one gets used to
it. Some readers may nd it helpful to note that the conditional assignment operator works exactly
the same way as the =IF() function in spreadsheets.
Generating a time index for panel datasets
Problem: gretl has a $unit accessor, but not the equivalent for time. What should I use?
Solution:
series x = time
Comment: The special construct genr time and its variants are aware of whether a dataset is a
panel.
16.3 Neat tricks
Interaction dummies
Problem: You want to estimate the model y
i
= x
i
1
+ z
i
2
+ d
i
3
+ (d
i
z
i
)
4
+
t
, where d
i
is a
dummy variable while x
i
and z
i
are vectors of explanatory variables.
Solution: As of version 1.9.12, gretl provides the ^ operator to make this operation easy. See
section 14.1 for details (especially example script 14.1). But back in my day, we used loops to do
that! Heres how:
list X = x1 x2 x3
list Z = z1 z2
list dZ = null
loop foreach i Z
series d$i = d * $i
list dZ = dZ d$i
endloop
ols y X Z d dZ
Comment: Its amazing what string substitution can do for you, isnt it?
Realized volatility
Problem: Given data by the minute, you want to compute the realized volatility for the hour as
RV
t
=
1
60
60
=1
y
2
t:
. Imagine your sample starts at time 1:1.
Solution:
smpl --full
genr time
series minute = int(time/60) + 1
series second = time % 60
setobs minute second --panel
genr rv = psd(y)^2
setobs 1 1
smpl second==1 --restrict
store foo rv
Comment: Here we trick gretl into thinking that our dataset is a panel dataset, where the minutes
are the units and the seconds are the time; this way, we can take advantage of the special
function psd(), panel standard deviation. Then we simply drop all observations but one per minute
and save the resulting data (store foo rv translates as store in the gretl datale foo.gdt the
series rv).
Looping over two paired lists
Problem: Suppose you have two lists with the same number of elements, and you want to apply
some command to corresponding elements over a loop.
Solution:
list L1 = a b c
list L2 = x y z
k1 = 1
loop foreach i L1 --quiet
k2 = 1
loop foreach j L2 --quiet
if k1 == k2
ols $i 0 $j
endif
k2++
endloop
k1++
endloop
Comment: The simplest way to achieve the result is to loop over all possible combinations and
lter out the unneeded ones via an if condition, as above. That said, in some cases variable names
can help. For example, if
list Lx = x1 x2 x3
list Ly = y1 y2 y3
then we could just loop over the integersquite intuitive and certainly more elegant:
loop i=1..3
ols y$i const x$i
endloop
Part II
Econometric methods
141
Chapter 17
Robust covariance matrix estimation
17.1 Introduction
Consider (once again) the linear regression model
y = X +u (17.1)
where y and u are T-vectors, X is a T k matrix of regressors, and is a k-vector of parameters.
As is well known, the estimator of given by Ordinary Least Squares (OLS) is
= (X
X)
1
X
y (17.2)
If the condition E(uX) = 0 is satised, this is an unbiased estimator; under somewhat weaker
conditions the estimator is biased but consistent. It is straightforward to show that when the OLS
estimator is unbiased (that is, when E(
) = 0), its variance is

Var(
) = E
_
(
)(
_
= (X
X)
1
X
X(X
X)
1
(17.3)
where = E(uu
) is the covariance matrix of the error terms.

Under the assumption that the error terms are independently and identically distributed (iid) we
can write =
2
I, where
2
is the (common) variance of the errors (and the covariances are zero).
In that case (17.3) simplies to the classical formula,
Var(
) =
2
(X
X)
1
(17.4)
If the iid assumption is not satised, two things follow. First, it is possible in principle to construct
a more ecient estimator than OLSfor instance some sort of Feasible Generalized Least Squares
(FGLS). Second, the simple classical formula for the variance of the least squares estimator is no
longer correct, and hence the conventional OLS standard errorswhich are just the square roots
of the diagonal elements of the matrix dened by (17.4) do not provide valid means of statistical
inference.
In the recent history of econometrics there are broadly two approaches to the problem of non-
iid errors. The traditional approach is to use an FGLS estimator. For example, if the departure
from the iid condition takes the form of time-series dependence, and if one believes that this
could be modeled as a case of rst-order autocorrelation, one might employ an AR(1) estimation
method such as CochraneOrcutt, HildrethLu, or PraisWinsten. If the problem is that the error
variance is non-constant across observations, one might estimate the variance as a function of the
independent variables and then perform weighted least squares, using as weights the reciprocals
of the estimated variances.
While these methods are still in use, an alternative approach has found increasing favor: that is,
use OLS but compute standard errors (or more generally, covariance matrices) that are robust with
respect to deviations from the iid assumption. This is typically combined with an emphasis on
using large datasetslarge enough that the researcher can place some reliance on the (asymptotic)
consistency property of OLS. This approach has been enabled by the availability of cheap computing
power. The computation of robust standard errors and the handling of very large datasets were
daunting tasks at one time, but now they are unproblematic. The other point favoring the newer
methodology is that while FGLS oers an eciency advantage in principle, it often involves making
142
Chapter 17. Robust covariance matrix estimation 143
additional statistical assumptions which may or may not be justied, which may not be easy to test
rigorously, and which may threaten the consistency of the estimatorfor example, the common
factor restriction that is implied by traditional FGLS corrections for autocorrelated errors.
James Stock and Mark Watsons Introduction to Econometrics illustrates this approach at the level of
undergraduate instruction: many of the datasets they use comprise thousands or tens of thousands
of observations; FGLS is downplayed; and robust standard errors are reported as a matter of course.
In fact, the discussion of the classical standard errors (labeled homoskedasticity-only) is conned
to an Appendix.
Against this background it may be useful to set out and discuss all the various options oered
by gretl in respect of robust covariance matrix estimation. The rst point to notice is that gretl
produces classical standard errors by default (in all cases apart from GMM estimation). In script
mode you can get robust standard errors by appending the --robust ag to estimation commands.
In the GUI program the model specication dialog usually contains a Robust standard errors
check box, along with a congure button that is activated when the box is checked. The congure
button takes you to a conguration dialog (which can also be reached from the main menu bar:
Tools Preferences General HCCME). There you can select from a set of possible robust
estimation variants, and can also choose to make robust estimation the default.
The specics of the available options depend on the nature of the data under consideration
cross-sectional, time series or panel and also to some extent the choice of estimator. (Although
we introduced robust standard errors in the context of OLS above, they may be used in conjunction
with other estimators too.) The following three sections of this chapter deal with matters that are
specic to the three sorts of data just mentioned. Note that additional details regarding covariance
matrix estimation in the context of GMM are given in chapter 22.
We close this introduction with a brief statement of what robust standard errors can and cannot
achieve. They can provide for asymptotically valid statistical inference in models that are basically
correctly specied, but in which the errors are not iid. The asymptotic part means that they
may be of little use in small samples. The correct specication part means that they are not a
magic bullet: if the error term is correlated with the regressors, so that the parameter estimates
themselves are biased and inconsistent, robust standard errors will not save the day.
17.2 Cross-sectional data and the HCCME
With cross-sectional data, the most likely departure from iid errors is heteroskedasticity (non-
constant variance).
1
In some cases one may be able to arrive at a judgment regarding the likely
form of the heteroskedasticity, and hence to apply a specic correction. The more common case,
however, is where the heteroskedasticity is of unknown form. We seek an estimator of the covari-
ance matrix of the parameter estimates that retains its validity, at least asymptotically, in face of
unspecied heteroskedasticity. It is not obvious, a priori, that this should be possible, but White
(1980) showed that
Var
h
(
) = (X
X)
1
X
X(X
X)
1
(17.5)
does the trick. (As usual in statistics, we need to say under certain conditions, but the conditions
are not very restrictive.)

is in this context a diagonal matrix, whose non-zero elements may be
estimated using squared OLS residuals. White referred to (17.5) as a heteroskedasticity-consistent
covariance matrix estimator (HCCME).
Davidson and MacKinnon (2004, chapter 5) oer a useful discussion of several variants on Whites
HCCME theme. They refer to the original variant of (17.5) in which the diagonal elements of

are estimated directly by the squared OLS residuals, u

2
t
as HC
0
. (The associated standard errors
are often called Whites standard errors.) The various renements of Whites proposal share a
common point of departure, namely the idea that the squared OLS residuals are likely to be too
1
In some specialized contexts spatial autocorrelation may be an issue. Gretl does not have any built-in methods to
handle this and we will not discuss it here.
small on average. This point is quite intuitive. The OLS parameter estimates,

, satisfy by design
the criterion that the sum of squared residuals,
_
u
2
t
=
_
_
y
t
X
t
_
2
is minimized for given X and y. Suppose that

. This is almost certain to be the case: even if
OLS is not biased, it would be a miracle if the

calculated from any nite sample were exactly equal
to . But in that case the sum of squares of the true, unobserved errors,

u
2
t
=

(y
t
X
t
)
2
is
bound to be greater than
u
2
t
. The elaborated variants on HC
0
take this point on board as follows:
HC
1
: Applies a degrees-of-freedom correction, multiplying the HC
0
matrix by T/(T k).
HC
2
: Instead of using u
2
t
for the diagonal elements of

, uses u
2
t
/(1 h
t
), where h
t
=
X
t
(X
X)
1
X
t
, the t
th
diagonal element of the projection matrix, P, which has the property
that P y = y. The relevance of h
t
is that if the variance of all the u
t
is
2
, the expectation
of u
2
t
is
2
(1 h
t
), or in other words, the ratio u
2
t
/(1 h
t
) has expectation
2
. As Davidson
and MacKinnon show, 0 h
t
< 1 for all t, so this adjustment cannot reduce the the diagonal
elements of

and in general revises them upward.
HC
3
: Uses u
2
t
/(1 h
t
)
2
. The additional factor of (1 h
t
) in the denominator, relative to
HC
2
, may be justied on the grounds that observations with large variances tend to exert a
lot of inuence on the OLS estimates, so that the corresponding residuals tend to be under-
estimated. See Davidson and MacKinnon for a fuller explanation.
The relative merits of these variants have been explored by means of both simulations and the-
oretical analysis. Unfortunately there is not a clear consensus on which is best. Davidson and
MacKinnon argue that the original HC
0
is likely to perform worse than the others; nonetheless,
Whites standard errors are reported more often than the more sophisticated variants and there-
fore, for reasons of comparability, HC
0
is the default HCCME in gretl.
If you wish to use HC
1
, HC
2
or HC
3
you can arrange for this in either of two ways. In script mode,
you can do, for example,
set hc_version 2
In the GUI program you can go to the HCCME conguration dialog, as noted above, and choose any
of these variants to be the default.
17.3 Time series data and HAC covariance matrices
Heteroskedasticity may be an issue with time series data too, but it is unlikely to be the only, or
even the primary, concern.
One form of heteroskedasticity is common in macroeconomic time series, but is fairly easily dealt
with. That is, in the case of strongly trending series such as Gross Domestic Product, aggregate
consumption, aggregate investment, and so on, higher levels of the variable in question are likely
to be associated with higher variability in absolute terms. The obvious x, employed in many
macroeconometric studies, is to use the logs of such series rather than the raw levels. Provided the
proportional variability of such series remains roughly constant over time, the log transformation
is eective.
Other forms of heteroskedasticity may resist the log transformation, but may demand a special
treatment distinct from the calculation of robust standard errors. We have in mind here autore-
gressive conditional heteroskedasticity, for example in the behavior of asset prices, where large
disturbances to the market may usher in periods of increased volatility. Such phenomena call for
specic estimation strategies, such as GARCH (see chapter 25).
Despite the points made above, some residual degree of heteroskedasticity may be present in time
series data: the key point is that in most cases it is likely to be combined with serial correlation
(autocorrelation), hence demanding a special treatment. In Whites approach,

, the estimated
covariance matrix of the u
t
, remains conveniently diagonal: the variances, E(u
2
t
), may dier by
t but the covariances, E(u
t
u
s
), are all zero. Autocorrelation in time series data means that at
least some of the the o-diagonal elements of

should be non-zero. This introduces a substantial
complication and requires another piece of terminology; estimates of the covariance matrix that
are asymptotically valid in face of both heteroskedasticity and autocorrelation of the error process
are termed HAC (heteroskedasticity and autocorrelation consistent).
The issue of HAC estimation is treated in more technical terms in chapter 22. Here we try to
convey some of the intuition at a more basic level. We begin with a general comment: residual
autocorrelation is not so much a property of the data, as a symptom of an inadequate model. Data
may be persistent though time, and if we t a model that does not take this aspect into account
properly, we end up with a model with autocorrelated disturbances. Conversely, it is often possible
to mitigate or even eliminate the problem of autocorrelation by including relevant lagged variables
in a time series model, or in other words, by specifying the dynamics of the model more fully. HAC
estimation should not be seen as the rst resort in dealing with an autocorrelated error process.
That said, the obvious extension of Whites HCCME to the case of autocorrelated errors would
seem to be this: estimate the o-diagonal elements of

(that is, the autocovariances, E(u
t
u
s
))
using, once again, the appropriate OLS residuals:
ts
= u
t
u
s
. This is basically right, but demands
an important amendment. We seek a consistent estimator, one that converges towards the true
as the sample size tends towards innity. This cant work if we allow unbounded serial depen-
dence. Bigger samples will enable us to estimate more of the true
ts
elements (that is, for t and
s more widely separated in time) but will not contribute ever-increasing information regarding the
maximally separated
ts
pairs, since the maximal separation itself grows with the sample size.
To ensure consistency, we have to conne our attention to processes exhibiting temporally limited
dependence, or in other words cut o the computation of the
ts
values at some maximum value
of p = t s (where p is treated as an increasing function of the sample size, T, although it cannot
increase in proportion to T).
The simplest variant of this idea is to truncate the computation at some nite lag order p, where
p grows as, say, T
1/4
. The trouble with this is that the resulting

may not be a positive denite
matrix. In practical terms, we may end up with negative estimated variances. One solution to this
problem is oered by The NeweyWest estimator (Newey and West, 1987), which assigns declining
weights to the sample autocovariances as the temporal separation increases.
To understand this point it is helpful to look more closely at the covariance matrix given in (17.5),
namely,
(X
X)
1
(X
X)(X
X)
1
This is known as a sandwich estimator. The bread, which appears on both sides, is (X
X)
1
.
This is a k k matrix, and is also the key ingredient in the computation of the classical covariance
matrix. The lling in the sandwich is
= X

X
(kk) (kT) (TT) (Tk)
Since = E(uu
), the matrix being estimated here can also be written as

= E(X
uu
X)
which expresses as the long-run covariance of the random k-vector X
u.
From a computational point of view, it is not necessary or desirable to store the (potentially very
large) T T matrix

as such. Rather, one computes the sandwich lling by summation as
(0) +
p
_
j=1
w
j
_
(j) +
(j)
_
where the k k sample autocovariance matrix

(j), for j 0, is given by
(j) =
1
T
T
_
t=j+1
u
t
u
tj
X
t
X
tj
and w
j
is the weight given to the autocovariance at lag j > 0.
This leaves two questions. How exactly do we determine the maximum lag length or bandwidth,
p, of the HAC estimator? And how exactly are the weights w
j
to be determined? We will return to
the (dicult) question of the bandwidth shortly. As regards the weights, gretl oers three variants.
The default is the Bartlett kernel, as used by Newey and West. This sets
w
j
=
_
_
_
1
j
p+1
j p
0 j > p
so the weights decline linearly as j increases. The other two options are the Parzen kernel and the
Quadratic Spectral (QS) kernel. For the Parzen kernel,
w
j
=
_
_
1 6a
2
j
+6a
3
j
0 a
j
0.5
2(1 a
j
)
3
0.5 < a
j
1
0 a
j
> 1
where a
j
= j/(p +1), and for the QS kernel,
w
j
=
25
12
2
d
2
j
_
sinm
j
m
j
cos m
j
_
where d
j
= j/p and m
j
= 6d
i
/5.
Figure 17.1 shows the weights generated by these kernels, for p = 4 and j = 1 to 9.
Figure 17.1: Three HAC kernels
Bartlett Parzen QS
In gretl you select the kernel using the set command with the hac_kernel parameter:
set hac_kernel parzen
set hac_kernel qs
set hac_kernel bartlett
Selecting the HAC bandwidth
The asymptotic theory developed by Newey, West and others tells us in general terms how the
HAC bandwidth, p, should grow with the sample size, T that is, p should grow in proportion
to some fractional power of T. Unfortunately this is of little help to the applied econometrician,
working with a given dataset of xed size. Various rules of thumb have been suggested, and gretl
implements two such. The default is p = 0.75T
1/3
, as recommended by Stock and Watson (2003).
An alternative is p = 4(T/100)
2/9
, as in Wooldridge (2002b). In each case one takes the integer
part of the result. These variants are labeled nw1 and nw2 respectively, in the context of the set
command with the hac_lag parameter. That is, you can switch to the version given by Wooldridge
with
set hac_lag nw2
As shown in Table 17.1 the choice between nw1 and nw2 does not make a great deal of dierence.
T p (nw1) p (nw2)
50 2 3
100 3 4
150 3 4
200 4 4
300 5 5
400 5 5
Table 17.1: HAC bandwidth: two rules of thumb
You also have the option of specifying a xed numerical value for p, as in
set hac_lag 6
In addition you can set a distinct bandwidth for use with the Quadratic Spectral kernel (since this
need not be an integer). For example,
set qs_bandwidth 3.5
Prewhitening and data-based bandwidth selection
An alternative approach is to deal with residual autocorrelation by attacking the problem from two
sides. The intuition behind the technique known as VAR prewhitening (Andrews and Monahan,
1992) can be illustrated by a simple example. Let x
t
be a sequence of rst-order autocorrelated
random variables
x
t
= x
t1
+u
t
The long-run variance of x
t
can be shown to be
V
LR
(x
t
) =
V
LR
(u
t
)
(1 )
2
In most cases, u
t
is likely to be less autocorrelated than x
t
, so a smaller bandwidth should suce.
Estimation of V
LR
(x
t
) can therefore proceed in three steps: (1) estimate ; (2) obtain a HAC estimate
of u
t
= x
t
x
t1
; and (3) divide the result by (1 )
2
.
The application of the above concept to our problem implies estimating a nite-order Vector Au-
toregression (VAR) on the vector variables
t
= X
t
u
t
. In general, the VAR can be of any order, but
in most cases 1 is sucient; the aim is not to build a watertight model for
t
, but just to mop up
a substantial part of the autocorrelation. Hence, the following VAR is estimated
t
= A
t1
+
t
Then an estimate of the matrix X
X can be recovered via

(I

A)
1
(I

A
)
1
where

is any HAC estimator, applied to the VAR residuals.

You can ask for prewhitening in gretl using
set hac_prewhiten on
There is at present no mechanism for specifying an order other than 1 for the initial VAR.
A further renement is available in this context, namely data-based bandwidth selection. It makes
intuitive sense that the HAC bandwidth should not simply be based on the size of the sample,
but should somehow take into account the time-series properties of the data (and also the kernel
chosen). A nonparametric method for doing this was proposed by Newey and West (1994); a good
concise account of the method is given in Hall (2005). This option can be invoked in gretl via
set hac_lag nw3
This option is the default when prewhitening is selected, but you can override it by giving a specic
numerical value for hac_lag.
Even the NeweyWest data-based method does not fully pin down the bandwidth for any particular
sample. The rst step involves calculating a series of residual covariances. The length of this series
is given as a function of the sample size, but only up to a scalar multiplefor example, it is given
as O(T
2/9
) for the Bartlett kernel. Gretl uses an implied multiple of 1.
VARs: a special case
A well-specied vector autoregression (VAR) will generally include enough lags of the dependent
variables to obviate the problem of residual autocorrelation, in which case HAC estimation is
redundantalthough there may still be a need to correct for heteroskedasticity. For that rea-
son plain HCCME, and not HAC, is the default when the --robust ag is given in the context of the
var command. However, if for some reason you need HAC you can force the issue by giving the
option --robust-hac.
17.4 Special issues with panel data
Since panel data have both a time-series and a cross-sectional dimension one might expect that, in
general, robust estimation of the covariance matrix would require handling both heteroskedasticity
and autocorrelation (the HAC approach). In addition, some special features of panel data require
attention.
The variance of the error term may dier across the cross-sectional units.
The covariance of the errors across the units may be non-zero in each time period.
If the between variation is not removed, the errors may exhibit autocorrelation, not in the
usual time-series sense but in the sense that the mean error for unit i may dier from that of
unit j. (This is particularly relevant when estimation is by pooled OLS.)
Gretl currently oers two robust covariance matrix estimators specically for panel data. These are
available for models estimated via xed eects, pooled OLS, and pooled two-stage least squares.
The default robust estimator is that suggested by Arellano (2003), which is HAC provided the panel
is of the large n, small T variety (that is, many units are observed in relatively few periods). The
Arellano estimator is
A
=
_
X
X
_
1
_
_
n
_
i=1
X
i
u
i
u
i
X
i
_
_
_
X
X
_
1
where X is the matrix of regressors (with the group means subtracted, in the case of xed eects) u
i
denotes the vector of residuals for unit i, and n is the number of cross-sectional units.
2
Cameron
and Trivedi (2005) make a strong case for using this estimator; they note that the ordinary White
HCCME can produce misleadingly small standard errors in the panel context because it fails to take
2
This variance estimator is also known as the clustered (over entities) estimator.
autocorrelation into account. In addition Stock and Watson (2008) show that the White HCCME is
inconsistent in the xed-eects panel context for xed T > 2.
In cases where autocorrelation is not an issue the estimator proposed by Beck and Katz (1995)
and discussed by Greene (2003, chapter 13) may be appropriate. This estimator, which takes into
account contemporaneous correlation across the units and heteroskedasticity by unit, is
BK
=
_
X
X
_
1
_
_
n
_
i=1
n
_
j=1

ij
X
i
X
j
_
_
_
X
X
_
1
The covariances
ij
are estimated via

ij
=
u
i
u
j
T
where T is the length of the time series for each unit. Beck and Katz call the associated standard
errors Panel-Corrected Standard Errors (PCSE). This estimator can be invoked in gretl via the
command
set pcse on
The Arellano default can be re-established via
set pcse off
(Note that regardless of the pcse setting, the robust estimator is not used unless the --robust ag
is given, or the Robust box is checked in the GUI program.)
17.5 The cluster-robust estimator
One further variance estimator is available in gretl, namely the cluster-robust estimator. This may
be appropriate (for cross-sectional data, mostly) when the observations naturally fall into groups or
clusters, and one suspects that the error term may exhibit dependency within the clusters and/or
have a variance that diers across clusters. Such clusters may be binary (e.g. employed versus
unemployed workers), categorical with several values (e.g. products grouped by manufacturer) or
ordinal (e.g. individuals with low, middle or high education levels).
For linear regression models estimated via least squares the cluster estimator is dened as
C
=
_
X
X
_
1
_
_
m
_
j=1
X
j
u
j
u
j
X
j
_
_
_
X
X
_
1
where m denotes the number of clusters, and X
j
and u
j
denote, respectively, the matrix of regres-
sors and the vector of residuals that fall within cluster j. As noted above, the Arellano variance
estimator for panel data models is a special case of this, where the clustering is by panel unit.
For models estimated by the method of Maximum Likelihood (in which case the standard variance
estimator is the inverse of the negative Hessian, H), the cluster estimator is
C
= H
1
_
_
m
_
j=1
G
j
G
j
_
_
H
1
where G
j
is the sum of the score (that is, the derivative of the loglikelihood with respect to the
parameter estimates) across the observations falling within cluster j.
It is common to apply a degrees of freedom adjustment to these estimators (otherwise the variance
may appear misleadingly small in comparison with other estimators, if the number of clusters is
small). In the least squares case the factor is (m/(m 1)) (n 1)/(n k), where n is the total
number of observations and k is the number of parameters estimated; in the case of ML estimation
the factor is just m/(m1).
Availability and syntax
The cluster-robust estimator is currently available for models estimated via OLS and TSLS, and also
for most ML estimators other than those specialized for time-series data: binary logit and probit,
ordered logit and probit, multinomial logit, Tobit, interval regression, biprobit, count models and
duration models. In all cases the syntax is the same: you give the option ag --cluster= followed
by the name of the series to be used to dene the clusters, as in
ols y 0 x1 x2 --cluster=cvar
The specied clustering variable must (a) be dened (not missing) at all observations used in esti-
mating the model and (b) take on at least two distinct values over the estimation range. The clusters
are dened as sets of observations having a common value for the clustering variable. It is generally
expected that the number of clusters is substantially less than the total number of observations.
Chapter 18
Panel data
18.1 Estimation of panel models
Pooled Ordinary Least Squares
The simplest estimator for panel data is pooled OLS. In most cases this is unlikely to be adequate,
but it provides a baseline for comparison with more complex estimators.
If you estimate a model on panel data using OLS an additional test item becomes available. In the
GUI model window this is the item panel diagnostics under the Tests menu; the script counterpart
is the hausman command.
To take advantage of this test, you should specify a model without any dummy variables represent-
ing cross-sectional units. The test compares pooled OLS against the principal alternatives, the xed
eects and random eects models. These alternatives are explained in the following section.
The xed and random eects models
In gretl version 1.6.0 and higher, the xed and random eects models for panel data can be es-
timated in their own right. In the graphical interface these options are found under the menu
item Model/Panel/Fixed and random eects. In the command-line interface one uses the panel
command, with or without the --random-effects option.
This section explains the nature of these models and comments on their estimation via gretl.
The pooled OLS specication may be written as
y
it
= X
it
+u
it
(18.1)
where y
it
is the observation on the dependent variable for cross-sectional unit i in period t, X
it
is a 1 k vector of independent variables observed for unit i in period t, is a k 1 vector of
parameters, and u
it
is an error or disturbance term specic to unit i in period t.
The xed and random eects models have in common that they decompose the unitary pooled
error term, u
it
. For the xed eects model we write u
it
=
i
+
it
, yielding
y
it
= X
it
+
i
+
it
(18.2)
That is, we decompose u
it
into a unit-specic and time-invariant component,
i
, and an observation-
specic error,
it
.
1
The
i
s are then treated as xed parameters (in eect, unit-specic y-intercepts),
which are to be estimated. This can be done by including a dummy variable for each cross-sectional
unit (and suppressing the global constant). This is sometimes called the Least Squares Dummy Vari-
ables (LSDV) method. Alternatively, one can subtract the group mean from each of variables and
estimate a model without a constant. In the latter case the dependent variable may be written as
y
it
= y
it
y
i
The group mean, y
i
, is dened as
y
i
=
1
T
i
T
i
_
t=1
y
it
1
It is possible to break a third component out of u
it
, namely w
t
, a shock that is time-specic but common to all the
units in a given period. In the interest of simplicity we do not pursue that option here.
151
Chapter 18. Panel data 152
where T
i
is the number of observations for unit i. An exactly analogous formulation applies to the
independent variables. Given parameter estimates,

, obtained using such de-meaned data we can
recover estimates of the
i
s using

i
=
1
T
i
T
i
_
t=1
_
y
it
X
it
_
These two methods (LSDV, and using de-meaned data) are numerically equivalent. gretl takes the
approach of de-meaning the data. If you have a small number of cross-sectional units, a large num-
ber of time-series observations per unit, and a large number of regressors, it is more economical
in terms of computer memory to use LSDV. If need be you can easily implement this manually. For
example,
genr unitdum
ols y x du_*
(See Chapter 9 for details on unitdum).
The
i
estimates are not printed as part of the standard model output in gretl (there may be a large
number of these, and typically they are not of much inherent interest). However you can retrieve
them after estimation of the xed eects model if you wish. In the graphical interface, go to the
Save menu in the model window and select per-unit constants. In command-line mode, you can
do genr newname = $ahat, where newname is the name you want to give the series.
For the random eects model we write u
it
= v
i
+
it
, so the model becomes
y
it
= X
it
+v
i
+
it
(18.3)
In contrast to the xed eects model, the v
i
s are not treated as xed parameters, but as random
drawings from a given probability distribution.
The celebrated GaussMarkov theorem, according to which OLS is the best linear unbiased esti-
mator (BLUE), depends on the assumption that the error term is independently and identically
distributed (IID). In the panel context, the IID assumption means that E(u
2
it
), in relation to equa-
tion 18.1, equals a constant,
2
u
, for all i and t, while the covariance E(u
is
u
it
) equals zero for all
s t and the covariance E(u
jt
u
it
) equals zero for all j i.
If these assumptions are not metand they are unlikely to be met in the context of panel data
OLS is not the most ecient estimator. Greater eciency may be gained using generalized least
squares (GLS), taking into account the covariance structure of the error term.
Consider observations on a given unit i at two dierent times s and t. From the hypotheses above
it can be worked out that Var(u
is
) = Var(u
it
) =
2
v
+
2
, while the covariance between u

is
and u
it
is given by E(u
is
u
it
) =
2
v
.
In matrix notation, we may group all the T
i
observations for unit i into the vector y
i
and write it as
y
i
= X
i
+u
i
(18.4)
The vector u
i
, which includes all the disturbances for individual i, has a variancecovariance matrix
given by
Var(u
i
) =
i
=
2
I +
2
v
J (18.5)
where J is a square matrix with all elements equal to 1. It can be shown that the matrix
K
i
= I

i
T
i
J,
where
i
= 1
_

2
2
+T
i
2
v
, has the property
K
i
K
i
=
2
I
It follows that the transformed system
K
i
y
i
= K
i
X
i
+K
i
u
i
(18.6)
satises the GaussMarkov conditions, and OLS estimation of (18.6) provides ecient inference.
But since
K
i
y
i
= y
i
i
y
i
GLS estimation is equivalent to OLS using quasi-demeaned variables; that is, variables from which
we subtract a fraction of their average.
2
Notice that for
2
0, 1, while for
2
v
0, 0.
This means that if all the variance is attributable to the individual eects, then the xed eects
estimator is optimal; if, on the other hand, individual eects are negligible, then pooled OLS turns
out, unsurprisingly, to be the optimal estimator.
To implement the GLS approach we need to calculate , which in turn requires estimates of the
variances
2
and
2
v
. (These are often referred to as the within and between variances re-
spectively, since the former refers to variation within each cross-sectional unit and the latter to
variation between the units). Several means of estimating these magnitudes have been suggested
in the literature (see Baltagi, 1995); by default gretl uses the method of Swamy and Arora (1972):
is estimated by the residual variance from the xed eects model, and the sum
2
+ T
i
2
v
is
estimated as T
i
times the residual variance from the between estimator,
y
i
=

X
i
+e
i
The latter regression is implemented by constructing a data set consisting of the group means
of all the relevant variables. Alternatively, if the --nerlove option is given, gretl uses the method
suggested by Nerlove (1971). In this case
2
v
is estimated as the sample variance of the xed eects,

2
v
=
1
n1
n
_
i=1
(
i
)
2
where n is the number of individuals and is the mean of the xed eects.
Choice of estimator
Which panel method should one use, xed eects or random eects?
One way of answering this question is in relation to the nature of the data set. If the panel comprises
observations on a xed and relatively small set of units of interest (say, the member states of the
European Union), there is a presumption in favor of xed eects. If it comprises observations on a
large number of randomly selected individuals (as in many epidemiological and other longitudinal
studies), there is a presumption in favor of random eects.
Besides this general heuristic, however, various statistical issues must be taken into account.
1. Some panel data sets contain variables whose values are specic to the cross-sectional unit
but which do not vary over time. If you want to include such variables in the model, the xed
eects option is simply not available. When the xed eects approach is implemented using
dummy variables, the problem is that the time-invariant variables are perfectly collinear with
the per-unit dummies. When using the approach of subtracting the group means, the issue is
that after de-meaning these variables are nothing but zeros.
2. A somewhat analogous prohibition applies to the random eects estimator. This estimator is
in eect a matrix-weighted average of pooled OLS and the between estimator. Suppose we
have observations on n units or individuals and there are k independent variables of interest.
If k > n, the between estimator is undenedsince we have only n eective observations
and hence so is the random eects estimator.
2
In a balanced panel, the value of is common to all individuals, otherwise it diers depending on the value of T
i
.
If one does not fall foul of one or other of the prohibitions mentioned above, the choice between
xed eects and random eects may be expressed in terms of the two econometric desiderata,
eciency and consistency.
From a purely statistical viewpoint, we could say that there is a tradeo between robustness and
eciency. In the xed eects approach, we do not make any hypotheses on the group eects (that
is, the time-invariant dierences in mean between the groups) beyond the fact that they exist
and that can be tested; see below. As a consequence, once these eects are swept out by taking
deviations from the group means, the remaining parameters can be estimated.
On the other hand, the random eects approach attempts to model the group eects as drawings
from a probability distribution instead of removing them. This requires that individual eects are
representable as a legitimate part of the disturbance term, that is, zero-mean random variables,
uncorrelated with the regressors.
As a consequence, the xed-eects estimator always works, but at the cost of not being able to
estimate the eect of time-invariant regressors. The richer hypothesis set of the random-eects
estimator ensures that parameters for time-invariant regressors can be estimated, and that esti-
mation of the parameters for time-varying regressors is carried out more eciently. These advan-
tages, though, are tied to the validity of the additional hypotheses. If, for example, there is reason
to think that individual eects may be correlated with some of the explanatory variables, then the
random-eects estimator would be inconsistent, while xed-eects estimates would still be valid.
It is precisely on this principle that the Hausman test is built (see below): if the xed- and random-
eects estimates agree, to within the usual statistical margin of error, there is no reason to think
the additional hypotheses invalid, and as a consequence, no reason not to use the more ecient RE
estimator.
Testing panel models
If you estimate a xed eects or random eects model in the graphical interface, you may notice
that the number of items available under the Tests menu in the model windowis relatively limited.
Panel models carry certain complications that make it dicult to implement all of the tests one
expects to see for models estimated on straight time-series or cross-sectional data.
Nonetheless, various panel-specic tests are printed along with the parameter estimates as a matter
of course, as follows.
When you estimate a model using xed eects, you automatically get an F-test for the null hy-
pothesis that the cross-sectional units all have a common intercept. That is to say that all the
i
s
are equal, in which case the pooled model (18.1), with a column of 1s included in the X matrix, is
adequate.
When you estimate using random eects, the BreuschPagan and Hausman tests are presented
automatically.
The BreuschPagan test is the counterpart to the F-test mentioned above. The null hypothesis is
that the variance of v
i
in equation (18.3) equals zero; if this hypothesis is not rejected, then again
we conclude that the simple pooled model is adequate.
The Hausman test probes the consistency of the GLS estimates. The null hypothesis is that these
estimates are consistentthat is, that the requirement of orthogonality of the v
i
and the X
i
is
satised. The test is based on a measure, H, of the distance between the xed-eects and random-
eects estimates, constructed such that under the null it follows the
2
distribution with degrees
of freedom equal to the number of time-varying regressors in the matrix X. If the value of H is
large this suggests that the random eects estimator is not consistent and the xed-eects model
is preferable.
There are two ways of calculating H, the matrix-dierence method and the regression method. The
procedure for the matrix-dierence method is this:
Collect the xed-eects estimates in a vector

and the corresponding random-eects esti-
mates in

, then form the dierence vector (

).
Form the covariance matrix of the dierence vector as Var(

) = Var(
) Var(
) = ,
where Var(
) and Var(
) are estimated by the sample variance matrices of the xed- and

random-eects models respectively.
3
Compute H =
_
1
_
_
.
Given the relative eciencies of

and

, the matrix should be positive denite, in which case
H is positive, but in nite samples this is not guaranteed and of course a negative
2
value is not
admissible. The regression method avoids this potential problem. The procedure is:
Treat the random-eects model as the restricted model, and record its sum of squared resid-
uals as SSR
r
.
Estimate via OLS an unrestricted model in which the dependent variable is quasi-demeaned y
and the regressors include both quasi-demeaned X (as in the RE model) and the de-meaned
variants of all the time-varying variables (i.e. the xed-eects regressors); record the sum of
squared residuals from this model as SSR
u
.
Compute H = n(SSR
r
SSR
u
) /SSR
u
, where n is the total number of observations used. On
this variant H cannot be negative, since adding additional regressors to the RE model cannot
raise the SSR.
By default gretl computes the Hausman test via the regression method, but it uses the matrix-
dierence method if you pass the option --matrix-diff to the panel command.
Robust standard errors
For most estimators, gretl oers the option of computing an estimate of the covariance matrix that
is robust with respect to heteroskedasticity and/or autocorrelation (and hence also robust standard
errors). In the case of panel data, robust covariance matrix estimators are available for the pooled
and xed eects model but not currently for random eects. Please see section 17.4 for details.
The constant in the xed eects model
Users are sometimes puzzled by the constant or intercept reported by gretl on estimation of the
xed eects model: how can a constant remain when the group means have been subtracted from
the data? The method of calculation of this term is a matter of convention, but the gretl authors
decided to follow the convention employed by Stata; this involves adding the global mean back into
the variables from which the group means have been removed.
4
The method that gretl uses internally is exemplied in Example 18.1. The coecients in the nal
OLS estimation, including the intercept, agree with those in the initial xed eects model, though
the standard errors dier due to a degrees of freedom correction in the xed-eects covariance
matrix. (Note that the pmean function returns the group mean of a series.)
18.2 Autoregressive panel models
Special problems arise when a lag of the dependent variable is included among the regressors in a
panel model. Consider a dynamic variant of the pooled model (eq. 18.1):
y
it
= X
it
+y
it1
+u
it
(18.7)
3
Hausman (1978) showed that the covariance of the dierence takes this simple form when

is an ecient estimator
and

is inecient.
4
See http://www.stata.com/support/faqs/statistics/intercept-in-fixed-effects-model/ for an expla-
nation from William Gould of StataCorp.
Example 18.1: Calculating the intercept in the xed eects model
open abdata.gdt
panel n const w k ys --fixed-effects
depvar = n - pmean(n) + mean(n)
list indepvars = const
loop foreach i w k ys --quiet
x_$i = $i - pmean($i) + mean($i)
indepvars += x_$i
endloop
ols depvar indepvars
First, if the error u
it
includes a group eect, v
i
, then y
it1
is bound to be correlated with the error,
since the value of v
i
aects y
i
at all t. That means that OLS applied to (18.7) will be inconsistent
as well as inecient. The xed-eects model sweeps out the group eects and so overcomes this
particular problem, but a subtler issue remains, which applies to both xed and random eects
estimation. Consider the de-meaned representation of xed eects, as applied to the dynamic
model,
y
it
=

X
it
+ y
i,t1
+
it
where y
it
= y
it
y
i
and
it
= u
it
u
i
(or u
it
i
, using the notation of equation 18.2). The trouble
is that y
i,t1
will be correlated with
it
via the group mean, y
i
. The disturbance
it
inuences y
it
directly, which inuences y
i
, which, by construction, aects the value of y
it
for all t. The same
issue arises in relation to the quasi-demeaning used for random eects. Estimators which ignore
this correlation will be consistent only as T (in which case the marginal eect of
it
on the
group mean of y tends to vanish).
One strategy for handling this problem, and producing consistent estimates of and , was pro-
posed by Anderson and Hsiao (1981). Instead of de-meaning the data, they suggest taking the rst
dierence of (18.7), an alternative tactic for sweeping out the group eects:
y
it
= X
it
+y
i,t1
+
it
(18.8)
where
it
= u
it
= (v
i
+
it
) =
it

i,t1
. Were not in the clear yet, given the structure of the
error
it
: the disturbance
i,t1
is an inuence on both
it
and y
i,t1
= y
it
y
i,t1
. The next step
is then to nd an instrument for the contaminated y
i,t1
. Anderson and Hsiao suggest using
either y
i,t2
or y
i,t2
, both of which will be uncorrelated with
it
provided that the underlying
errors,
it
, are not themselves serially correlated.
The AndersonHsiao estimator is not provided as a built-in function in gretl, since gretls sensible
handling of lags and dierences for panel data makes it a simple application of regression with
instrumental variablessee Example 18.2, which is based on a study of country growth rates by
Nerlove (1999).
5
Although the AndersonHsiao estimator is consistent, it is not most ecient: it does not make the
fullest use of the available instruments for y
i,t1
, nor does it take into account the dierenced
structure of the error
it
. It is improved upon by the methods of Arellano and Bond (1991) and
Blundell and Bond (1998). These methods are taken up in the next chapter.
5
Also see Clint Cummins benchmarks page, http://www.stanford.edu/~clint/bench/.
Example 18.2: The AndersonHsiao estimator for a dynamic panel model
# Penn World Table data as used by Nerlove
open penngrow.gdt
# Fixed effects (for comparison)
panel Y 0 Y(-1) X
# Random effects (for comparison)
panel Y 0 Y(-1) X --random-effects
# take differences of all variables
diff Y X
# Anderson-Hsiao, using Y(-2) as instrument
tsls d_Y d_Y(-1) d_X ; 0 d_X Y(-2)
# Anderson-Hsiao, using d_Y(-2) as instrument
tsls d_Y d_Y(-1) d_X ; 0 d_X d_Y(-2)
Chapter 19
Dynamic panel models
Since gretl version 1.9.2, the primary command for estimating dynamic panel models has been
dpanel. The closely related arbond command predated dpanel, and is still present, but whereas
arbond only supports the so-called dierence estimator (Arellano and Bond, 1991), dpanel is
addition oers the system estimator (Blundell and Bond, 1998), which has become the method of
choice in the applied literature.
19.1 Introduction
Notation
A dynamic linear panel data model can be represented as follows (in notation based on Arellano
(2003)):
y
it
= y
i,t1
+
x
it
+
i
+v
it
(19.1)
The main idea on which the dierence estimator is based is to get rid of the individual eect via
dierencing. First-dierencing eq. (19.1) yields
y
it
= y
i,t1
+
x
it
+v
it
=
W
it
+v
it
, (19.2)
in obvious notation. The error term of (19.2) is, by construction, autocorrelated and also correlated
with the lagged dependent variable, so an estimator that takes both issues into account is needed.
The endogeneity issue is solved by noting that all values of y
i,tk
, with k > 1 can be used as
instruments for y
i,t1
: unobserved values of y
i,tk
(because they could be missing, or pre-sample)
can safely be substituted with 0. In the language of GMM, this amounts to using the relation
E(v
it
y
i,tk
) = 0, k > 1 (19.3)
as an orthogonality condition.
Autocorrelation is dealt with by noting that if v
it
is white noise, the covariance matrix of the vector
whose typical element is v
it
is proportional to a matrix H that has 2 on the main diagonal, 1
on the rst subdiagonals and 0 elsewhere. One-step GMM estimation of equation (19.2) amounts to
computing
=
_
_
_
_
N
_
i=1
WWW
i
ZZZ
i
_
_
AAA
N
_
_
N
_
i=1
ZZZ
i
WWW
i
_
_
_
_
1
_
_
N
_
i=1
WWW
i
ZZZ
i
_
_
AAA
N
_
_
N
_
i=1
ZZZ
i
yyy
i
_
_
(19.4)
where
yyy
i
=
_
y
i,3
y
i,T
_
WWW
i
=
_
y
i,2
y
i,T1
x
i,3
x
i,T
_
ZZZ
i
=
_
_
_
_
_
_
_
y
i1
0 0 0 x
i3
0 y
i1
y
i2
0 x
i4
.
.
.
0 0 0 y
i,T2
x
iT
_
_
_
_
_
_
_
158
Chapter 19. Dynamic panel models 159
and
AAA
N
=
_
_
N
_
i=1
ZZZ
i
HZZZ
i
_
_
1
Once the 1-step estimator is computed, the sample covariance matrix of the estimated residuals
can be used instead of H to obtain 2-step estimates, which are not only consistent but asymp-
totically ecient. (In principle the process may be iterated, but nobody seems to be interested.)
Standard GMM theory applies, except for one thing: Windmeijer (2005) has computed nite-sample
corrections to the asymptotic covariance matrix of the parameters, which are nowadays almost
universally used.
The dierence estimator is consistent, but has been shown to have poor properties in nite samples
when is near one. People these days prefer the so-called system estimator, which complements
the dierenced data (with lagged levels used as instruments) with data in levels (using lagged
dierences as instruments). The system estimator relies on an extra orthogonality condition which
has to do with the earliest value of the dependent variable y
i,1
. The interested reader is referred
to Blundell and Bond (1998, pp. 124125) for details, but here it suces to say that this condition
is satised in mean-stationary models and brings about eciency that may be substantial in many
cases.
The set of orthogonality conditions exploited in the system approach is not very much larger than
with the dierence estimator, the reason being that most of the possible orthogonality conditions
associated with the equations in levels are redundant, given those already used for the equations
in dierences.
The key equations of the system estimator can be written as
=
_
_
_
_
N
_
i=1
WWW
ZZZ
_
_
AAA
N
_
_
N
_
i=1
ZZZ

WWW
_
_
_
_
1
_
_
N
_
i=1
WWW
ZZZ
_
_
AAA
N
_
_
N
_
i=1
ZZZ
yyy
i
_
_
(19.5)
where
yyy
i
=
_
y
i3
y
iT
y
i3
y
iT
_
WWW
i
=
_
y
i2
y
i,T1
y
i2
y
i,T1
x
i3
x
iT
x
i3
x
iT
_
ZZZ
i
=
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
y
i1
0 0 0 0 0 x
i,3
0 y
i1
y
i2
0 0 0 x
i,4
.
.
.
0 0 0 y
i,T2
0 0 x
iT
.
.
.
0 0 0 0 y
i2
0 x
i3
.
.
.
0 0 0 0 0 y
i,T1
x
iT
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
and
AAA
N
=
_
_
N
_
i=1
ZZZ
ZZZ
_
_
1
In this case choosing a precise form for the matrix H
for the rst step is no trivial matter. Its

north-west block should be as similar as possible to the covariance matrix of the vector v
it
, so
the same choice as the dierence estimator is appropriate. Ideally, the south-east block should
be proportional to the covariance matrix of the vector
i
+ vvv, that is
2
v
I +
2
; but since
2
is
unknown and any positive denite matrix renders the estimator consistent, people just use I. The
o-diagonal blocks should, in principle, contain the covariances between v
is
and v
it
, which would
be an identity matrix if v
it
is white noise. However, since the south-east block is typically given a
conventional value anyway, the benet in making this choice is not obvious. Some packages use I;
others use a zero matrix. Asymptotically, it should not matter, but on real datasets the dierence
between the resulting estimates can be noticeable.
Rank deciency
Both the dierence estimator (19.4) and the system estimator (19.5) depend, for their existence,
on the invertibility of AAA
N
. This matrix may turn out to be singular for several reasons. However,
this does not mean that the estimator is not computable: in some cases, adjustments are possible
such that the estimator does exist, but the user should be aware that in these cases not all software
packages use the same strategy and replication of results may prove dicult or even impossible.
A rst reason why AAA
N
may be singular could be the unavailability of instruments, chiey because
of missing observations. This case is easy to handle. If a particular row of

ZZZ
i
is zero for all
units, the corresponding orthogonality condition (or the corresponding instrument if you prefer) is
automatically dropped; of course, the overidentication rank is adjusted for testing purposes.
Even if no instruments are zero, however, AAA
N
could be rank decient. A trivial case occurs if there
are collinear instruments, but a less trivial case may arise when T (the total number of time periods
available) is not much smaller than N (the number of units), as, for example, in some macro datasets
where the units are countries. The total number of potentially usable orthogonality conditions is
O(T
2
), which may well exceed N in some cases. Of course AAA
N
is the sum of N matrices which have,
at most, rank 2T 3 and therefore it could well happen that the sum is singular.
In all these cases, what we consider the proper way to go is to substitute the pseudo-inverse of
AAA
N
(MoorePenrose) for its regular inverse. Again, our choice is shared by some software packages,
but not all, so replication may be hard.
Treatment of missing values
Textbooks seldom bother with missing values, but in some cases their treatment may be far from
obvious. This is especially true if missing values are interspersed between valid observations. For
example, consider the plain dierence estimator with one lag, so
y
t
= y
t1
+ +
t
where the i index is omitted for clarity. Suppose you have an individual with t = 1. . . 5, for which
y
3
is missing. It may seem that the data for this individual are unusable, because dierencing y
t
would produce something like
t 1 2 3 4 5
y
t

y
t

where = nonmissing and = missing. Estimation seems to be unfeasible, since there are no
periods in which y
t
and y
t1
are both observable.
However, we can use a k-dierence operator and get
k
y
t
=
k
y
t1
+
k
t
where
k
= 1 L
k
and past levels of y
t
are perfectly valid instruments. In this example, we can
choose k = 3 and use y
1
as an instrument, so this unit is in fact perfectly usable.
Not all software packages seem to be aware of this possibility, so replicating published results may
prove tricky if your dataset contains individuals with gaps between valid observations.
19.2 Usage
One of the concepts underlying the syntax of dpanel is that you get default values for several
choices you may want to make, so that in a standard situation the command itself is very short
to write (and read). The simplest case of the model (19.1) is a plain AR(1) process:
y
i,t
= y
i,t1
+
i
+v
it
. (19.6)
If you give the command
dpanel 1 ; y
gretl assumes that you want to estimate (19.6) via the dierence estimator (19.4), using as many
orthogonality conditions as possible. The scalar 1 between dpanel and the semicolon indicates that
only one lag of y is included as an explanatory variable; using 2 would give an AR(2) model. The
syntax that gretl uses for the non-seasonal AR and MA lags in an ARMA model is also supported in
this context.
1
For example, if you want the rst and third lags of y (but not the second) included as
explanatory variables you can say
dpanel {1 3} ; y
or you can use a pre-dened matrix for this purpose:
matrix ylags = {1, 3}
dpanel ylags ; y
To use a single lag of y other than the rst you need to employ this mechanism:
dpanel {3} ; y # only lag 3 is included
dpanel 3 ; y # compare: lags 1, 2 and 3 are used
To use the system estimator instead, you add the --system option, as in
dpanel 1 ; y --system
The level orthogonality conditions and the corresponding instrument are appended automatically
(see eq. 19.5).
Regressors
If we want to introduce additional regressors, we list them after the dependent variable in the same
way as other gretl commands, such as ols.
For the dierence orthogonality relations, dpanel takes care of transforming the regressors in par-
allel with the dependent variable. Note that this diers from gretls arbond command, where only
the dependent variable is dierenced automatically; it brings us more in line with other software.
One case of potential ambiguity is when an intercept is specied but the dierence-only estimator
is selected, as in
dpanel 1 ; y const
In this case the default dpanel behavior, which agrees with Statas xtabond2, is to drop the con-
stant (since dierencing reduces it to nothing but zeros). However, for compatibility with the
DPD package for Ox, you can give the option --dpdstyle, in which case the constant is retained
1
This represents an enhancement over the arbond command.
(equivalent to including a linear trend in equation 19.1). A similar point applies to the period-
specic dummy variables which can be added in dpanel via the --time-dummies option: in the
dierences-only case these dummies are entered in dierenced form by default, but when the
--dpdstyle switch is applied they are entered in levels.
The standard gretl syntax applies if you want to use lagged explanatory variables, so for example
the command
dpanel 1 ; y const x(0 to -1) --system
would result in estimation of the model
y
it
= y
i,t1
+
0
+
1
x
it
+
2
x
i,t1
+
i
+v
it
.
Instruments
The default rules for instruments are:
lags of the dependent variable are instrumented using all available orthogonality conditions;
and
additional regressors are considered exogenous, so they are used as their own instruments.
If a dierent policy is wanted, the instruments should be specied in an additional list, separated
from the regressors list by a semicolon. The syntax closely mirrors that for the tsls command,
but in this context it is necessary to distinguish between regular instruments and what are often
called GMM-style instruments (that, instruments that are handled in the same block-diagonal
manner as lags of the dependent variable, as described above).
Regular instruments are transformed in the same way as regressors, and the contemporaneous
value of the transformed variable is used to form an orthogonality condition. Since regressors are
treated as exogenous by default, it follows that these two commands estimate the same model:
dpanel 1 ; y z
dpanel 1 ; y z ; z
The instrument specication in the second case simply conrms what is implicit in the rst: that
z is exogenous. Note, though, that if you have some additional variable z2 which you want to add
as a regular instrument, it then becomes necessary to include z in the instrument list if it is to be
treated as exogenous:
dpanel 1 ; y z ; z2 # z is now implicitly endogenous
dpanel 1 ; y z ; z z2 # z is treated as exogenous
The specication of GMM-style instruments is handled by the special constructs GMM() and
GMMlevel(). The rst of these relates to instruments for the equations in dierences, and the
second to the equations in levels. The syntax for GMM() is
GMM(varname, minlag, maxlag)
where varname is replaced by the name of a series, and minlag and maxlag are replaced by the
minimum and maximum lags to be used as instruments. The same goes for GMMlevel().
One common use of GMM() is to limit the number of lagged levels of the dependent variable used
as instruments for the equations in dierences. Its well known that although exploiting all pos-
sible orthogonality conditions yields maximal asymptotic eciency, in nite samples it may be
preferable to use a smaller subset (but see also Okui (2009)). For example, the specication
dpanel 1 ; y ; GMM(y, 2, 4)
ensures that no lags of y
t
earlier than t 4 will be used as instruments.
A second use of GMM() is to exploit more fully the potential block-diagonal orthogonality conditions
oered by an exogenous regressor, or a related variable that does not appear as a regressor. For
example, in
dpanel 1 ; y x ; GMM(z, 2, 6)
the variable x is considered an endogenous regressor, and up to 5 lags of z are used as instruments.
Note that in the following script fragment
dpanel 1 ; y z
dpanel 1 ; y z ; GMM(z,0,0)
the two estimation commands should not be expected to give the same result, as the sets of orthog-
onality relationships are subtly dierent. In the latter case, you have T 2 separate orthogonality
relationships pertaining to z
it
, none of which has any implication for the other ones; in the former
case, you only have one. In terms of the ZZZ
i
matrix, the rst form adds a single row to the bottom
of the instruments matrix, while the second form adds a diagonal block with T 2 columns; that
is,
_
z
i3
z
i4
z
it
_
versus
_
_
_
_
_
_
_
z
i3
0 0
0 z
i4
0
.
.
.
.
.
.
0 0 z
it
_
_
_
_
_
_
_
19.3 Replication of DPD results
In this section we show how to replicate the results of some of the pioneering work with dynamic
panel-data estimators by Arellano, Bond and Blundell. As the DPD manual (Doornik, Arellano and
Bond, 2006) explains, it is dicult to replicate the original published results exactly, for two main
reasons: not all of the data used in those studies are publicly available; and some of the choices
made in the original software implementation of the estimators have been superseded. Here, there-
fore, our focus is on replicating the results obtained using the current DPD package and reported
in the DPD manual.
The examples are based on the program les abest1.ox, abest3.ox and bbest1.ox. These
are included in the DPD package, along with the ArellanoBond database les abdata.bn7 and
abdata.in7.
2
The ArellanoBond data are also provided with gretl, in the le abdata.gdt. In the
following we do not show the output from DPD or gretl; it is somewhat voluminous, and is easily
generated by the user. As of this writing the results from Ox/DPD and gretl are identical in all
relevant respects for all of the examples shown.
3
A complete Ox/DPD program to generate the results of interest takes this general form:
#include <oxstd.h>
#import <packages/dpd/dpd>
2
See http://www.doornik.com/download.html.
3
To be specic, this is using Ox Console version 5.10, version 1.24 of the DPD package, and gretl built from CVS as of
2010-10-23, all on Linux.
main()
{
decl dpd = new DPD();
dpd.Load("abdata.in7");
dpd.SetYear("YEAR");
// model-specific code here
delete dpd;
}
In the examples below we take this template for granted and show just the model-specic code.
Example 1
The following Ox/DPD codedrawn from abest1.oxreplicates column (b) of Table 4 in Arellano
and Bond (1991), an instance of the dierences-only or GMM-DIF estimator. The dependent variable
is the log of employment, n; the regressors include two lags of the dependent variable, current and
lagged values of the log real-product wage, w, the current value of the log of gross capital, k, and
current and lagged values of the log of industry output, ys. In addition the specication includes
a constant and ve year dummies; unlike the stochastic regressors, these deterministic terms are
not dierenced. In this specication the regressors w, k and ys are treated as exogenous and serve
as their own instruments. In DPD syntax this requires entering these variables twice, on the X_VAR
and I_VAR lines. The GMM-type (block-diagonal) instruments in this example are the second and
subsequent lags of the level of n. Both 1-step and 2-step estimates are computed.
dpd.SetOptions(FALSE); // dont use robust standard errors
dpd.Select(Y_VAR, {"n", 0, 2});
dpd.Select(X_VAR, {"w", 0, 1, "k", 0, 0, "ys", 0, 1});
dpd.Select(I_VAR, {"w", 0, 1, "k", 0, 0, "ys", 0, 1});
dpd.Gmm("n", 2, 99);
dpd.SetDummies(D_CONSTANT + D_TIME);
print("\n\n***** Arellano & Bond (1991), Table 4 (b)");
dpd.SetMethod(M_1STEP);
dpd.Estimate();
dpd.Estimate();
Here is gretl code to do the same job:
open abdata.gdt
list X = w w(-1) k ys ys(-1)
dpanel 2 ; n X const --time-dummies --asy --dpdstyle
dpanel 2 ; n X const --time-dummies --asy --two-step --dpdstyle
Note that in gretl the switch to suppress robust standard errors is --asymptotic, here abbreviated
to --asy.
4
The --dpdstyle ag species that the constant and dummies should not be dierenced,
in the context of a GMM-DIF model. With gretls dpanel command it is not necessary to specify the
exogenous regressors as their own instruments since this is the default; similarly, the use of the
second and all longer lags of the dependent variable as GMM-type instruments is the default and
need not be stated explicitly.
4
Option ags in gretl can always be truncated, down to the minimal unique abbreviation.
Example 2
The DPD le abest3.ox contains a variant of the above that diers with regard to the choice of
instruments: the variables w and k are now treated as predetermined, and are instrumented GMM-
style using the second and third lags of their levels. This approximates column (c) of Table 4 in
Arellano and Bond (1991). We have modied the code in abest3.ox slightly to allow the use of
robust (Windmeijer-corrected) standard errors, which are the default in both DPD and gretl with
2-step estimation:
dpd.Select(I_VAR, {"ys", 0, 1});
dpd.Gmm("n", 2, 99);
dpd.Gmm("w", 2, 3);
dpd.Gmm("k", 2, 3);
print("\n***** Arellano & Bond (1991), Table 4 (c)\n");
print(" (but using different instruments!!)\n");
dpd.Estimate();
The gretl code is as follows:
open abdata.gdt
list X = w w(-1) k ys ys(-1)
list Ivars = ys ys(-1)
dpanel 2 ; n X const ; GMM(w,2,3) GMM(k,2,3) Ivars --time --two-step --dpd
Note that since we are now calling for an instrument set other then the default (following the second
semicolon), it is necessary to include the Ivars specication for the variable ys. However, it is
not necessary to specify GMM(n,2,99) since this remains the default treatment of the dependent
variable.
Example 3
Our third example replicates the DPD output from bbest1.ox: this uses the same dataset as the
previous examples but the model specications are based on Blundell and Bond (1998), and involve
comparison of the GMM-DIF and GMM-SYS (system) estimators. The basic specication is slightly
simplied in that the variable ys is not used and only one lag of the dependent variable appears as
a regressor. The Ox/DPD code is:
dpd.Select(X_VAR, {"w", 0, 1, "k", 0, 1});
print("\n\n***** Blundell & Bond (1998), Table 4: 1976-86 GMM-DIF");
dpd.Gmm("n", 2, 99);
dpd.Gmm("w", 2, 99);
dpd.Gmm("k", 2, 99);
dpd.Estimate();
print("\n\n***** Blundell & Bond (1998), Table 4: 1976-86 GMM-SYS");
dpd.GmmLevel("n", 1, 1);
dpd.GmmLevel("w", 1, 1);
dpd.GmmLevel("k", 1, 1);
dpd.Estimate();
Here is the corresponding gretl code:
open abdata.gdt
list X = w w(-1) k k(-1)
# Blundell & Bond (1998), Table 4: 1976-86 GMM-DIF
dpanel 1 ; n X const ; GMM(w,2,99) GMM(k,2,99) --time --two-step --dpd
# Blundell & Bond (1998), Table 4: 1976-86 GMM-SYS
dpanel 1 ; n X const ; GMM(w,2,99) GMM(k,2,99) \
GMMlevel(w,1,1) GMMlevel(k,1,1) --time --two-step --dpd --system
Note the use of the --system option ag to specify GMM-SYS, including the default treatment of
the dependent variable, which corresponds to GMMlevel(n,1,1). In this case we also want to
use lagged dierences of the regressors w and k as instruments for the levels equations so we
need explicit GMMlevel entries for those variables. If you want something other than the default
treatment for the dependent variable as an instrument for the levels equations, you should give an
explicit GMMlevel specication for that variableand in that case the --system ag is redundant
(but harmless).
For the sake of completeness, note that if you specify at least one GMMlevel term, dpanel will then
include equations in levels, but it will not automatically add a default GMMlevel specication for
the dependent variable unless the --system option is given.
19.4 Cross-country growth example
The previous examples all used the ArellanoBond dataset; for this example we use the dataset
CEL.gdt, which is also included in the gretl distribution. As with the ArellanoBond data, there
are numerous missing values. Details of the provenance of the data can be found by opening the
dataset information window in the gretl GUI (Data menu, Dataset info item). This is a subset of the
BarroLee 138-country panel dataset, an approximation to which is used in Caselli, Esquivel and
Lefort (1996) and Bond, Hoeer and Temple (2001).
5
Both of these papers explore the dynamic
panel-data approach in relation to the issues of growth and convergence of per capita income across
countries.
The dependent variable is growth in real GDP per capita over successive ve-year periods; the
regressors are the log of the initial (ve years prior) value of GDP per capita, the log-ratio of in-
vestment to GDP, s, in the prior ve years, and the log of annual average population growth, n,
over the prior ve years plus 0.05 as stand-in for the rate of technical progress, g, plus the rate of
depreciation, (with the last two terms assumed to be constant across both countries and periods).
The original model is
5
y
it
= y
i,t5
+s
it
+(n
it
+g +) +
t
+
i
+
it
(19.7)
which allows for a time-specic disturbance
t
. The Solow model with CobbDouglas production
function implies that = , but this assumption is not imposed in estimation. The time-specic
disturbance is eliminated by subtracting the period mean from each of the series.
Equation (19.7) can be transformed to an AR(1) dynamic panel-data model by adding y
i,t5
to both
sides, which gives
y
it
= (1 +)y
i,t5
+s
it
+(n
it
+g +) +
i
+
it
(19.8)
5
We say an approximation because we have not been able to replicate exactly the OLS results reported in the papers
cited, though it seems from the description of the data in Caselli et al. (1996) that we ought to be able to do so. We note
that Bond et al. (2001) used data provided by Professor Caselli yet did not manage to reproduce the latters results.
where all variables are now assumed to be time-demeaned.
In (rough) replication of Bond et al. (2001) we now proceed to estimate the following two models:
(a) equation (19.8) via GMM-DIF, using as instruments the second and all longer lags of y
it
, s
it
and
n
it
+ g + ; and (b) equation (19.8) via GMM-SYS, using y
i,t1
, s
i,t1
and (n
i,t1
+ g + ) as
additional instruments in the levels equations. We report robust standard errors throughout. (As a
purely notational matter, we now use t 1 to refer to values ve years prior to t, as in Bond et al.
(2001)).
The gretl script to do this job is shown below. Note that the nal transformed versions of the
variables (logs, with time-means subtracted) are named ly (y
it
), linv (s
it
) and lngd (n
it
+g +).
open CEL.gdt
ngd = n + 0.05
ly = log(y)
linv = log(s)
lngd = log(ngd)
# take out time means
loop i=1..8 --quiet
smpl (time == i) --restrict --replace
ly -= mean(ly)
linv -= mean(linv)
lngd -= mean(lngd)
endloop
smpl --full
list X = linv lngd
# 1-step GMM-DIF
dpanel 1 ; ly X ; GMM(linv,2,99) GMM(lngd,2,99)
# 2-step GMM-DIF
dpanel 1 ; ly X ; GMM(linv,2,99) GMM(lngd,2,99) --two-step
# GMM-SYS
dpanel 1 ; ly X ; GMM(linv,2,99) GMM(lngd,2,99) \
GMMlevel(linv,1,1) GMMlevel(lngd,1,1) --two-step --sys
For comparison we estimated the same two models using Ox/DPD and the Stata command xtabond2.
(In each case we constructed a comma-separated values dataset containing the data as transformed
in the gretl script shown above, using a missing-value code appropriate to the target program.) For
reference, the commands used with Stata are reproduced below:
insheet using CEL.csv
tsset unit time
xtabond2 ly L.ly linv lngd, gmm(L.ly, lag(1 99)) gmm(linv, lag(2 99))
gmm(lngd, lag(2 99)) rob nolev
gmm(lngd, lag(2 99)) rob nolev twostep
gmm(lngd, lag(2 99)) rob nocons twostep
For the GMM-DIF model all three programs nd 382 usable observations and 30 instruments, and
yield identical parameter estimates and robust standard errors (up to the number of digits printed,
or more); see Table 19.1.
6
Results for GMM-SYS estimation are shown in Table 19.2. In this case we show two sets of gretl
results: those labeled gretl(1) were obtained using gretls --dpdstyle option, while those labeled
6
The coecient shown for ly(-1) in the Tables is that reported directly by the software; for comparability with the
original model (eq. 19.7) it is necesary to subtract 1, which produces the expected negative value indicating conditional
convergence in per capita income.
1-step 2-step
coe std. error coe std. error
ly(-1) 0.577564 0.1292 0.610056 0.1562
linv 0.0565469 0.07082 0.100952 0.07772
lngd 0.143950 0.2753 0.310041 0.2980
Table 19.1: GMM-DIF: BarroLee data
gretl(2) did not use that optionthe intent being to reproduce the H matrices used by Ox/DPD
and xtabond2 respectively.
gretl(1) Ox/DPD gretl(2) xtabond2
ly(-1) 0.9237 (0.0385) 0.9167 (0.0373) 0.9073 (0.0370) 0.9073 (0.0370)
linv 0.1592 (0.0449) 0.1636 (0.0441) 0.1856 (0.0411) 0.1856 (0.0411)
lngd 0.2370 (0.1485) 0.2178 (0.1433) 0.2355 (0.1501) 0.2355 (0.1501)
Table 19.2: 2-step GMM-SYS: BarroLee data (standard errors in parentheses)
In this case all three programs use 479 observations; gretl and xtabond2 use 41 instruments and
produce the same estimates (when using the same H matrix) while Ox/DPD nominally uses 66.
7
It is noteworthy that with GMM-SYS plus messy missing observations, the results depend on the
precise array of instruments used, which in turn depends on the details of the implementation of
the estimator.
19.5 Auxiliary test statistics
We have concentrated above on the parameter estimates and standard errors. It may be worth
adding a few words on the additional test statistics that typically accompany both GMM-DIF and
GMM-SYS estimation. These include the Sargan test for overidentication, one or more Wald tests
for the joint signicance of the regressors (and time dummies, if applicable) and tests for rst- and
second-order autocorrelation of the residuals from the equations in dierences.
As in Ox/DPD, the Sargan test statistic reported by gretl is
S =
_
_
N
_
i=1
vvv
i
ZZZ
i
_
_
AAA
N
_
_
N
_
i=1
ZZZ
i
vvv
i
_
_
where the vvv
i
are the transformed (e.g. dierenced) residuals for unit i. Under the null hypothesis
that the instruments are valid, S is asymptotically distributed as chi-square with degrees of freedom
equal to the number of overidentifying restrictions.
In general we see a good level of agreement between gretl, DPD and xtabond2 with regard to these
statistics, with a few relatively minor exceptions. Specically, xtabond2 computes both a Sargan
test and a Hansen test for overidentication, but what it calls the Hansen test is, apparently, what
DPD calls the Sargan test. (We have had diculty determining from the xtabond2 documentation
(Roodman, 2006) exactly how its Sargan test is computed.) In addition there are cases where the
degrees of freedom for the Sargan test dier between DPD and gretl; this occurs when the AAA
N
matrix is singular (section 19.1). In concept the df equals the number of instruments minus the
number of parameters estimated; for the rst of these terms gretl uses the rank of AAA
N
, while DPD
appears to use the full dimension of this matrix.
7
This is a case of the issue described in section 19.1: the full AAA
N
matrix turns out to be singular and special measures
must be taken to produce estimates.
19.6 Memo: dpanel options
ag eect
--asymptotic Suppresses the use of robust standard errors
--two-step Calls for 2-step estimation (the default being 1-step)
--system Calls for GMM-SYS, with default treatment of the dependent variable,
as in GMMlevel(y,1,1)
--time-dummies Includes period-specic dummy variables
--dpdstyle Compute the H matrix as in DPD; also suppresses dierencing of
automatic time dummies and omission of intercept in the GMM-DIF
case
--verbose When --two-step is selected, prints the 1-step estimates rst
--vcv Calls for printing of the covariance matrix
--quiet Suppresses the printing of results
The time dummies option supports the qualier noprint, as in
--time-dummies=noprint
This means that although the dummies are included in the specication their coecients, standard
errors and so on are not printed.
Chapter 20
Nonlinear least squares
20.1 Introduction and examples
Gretl supports nonlinear least squares (NLS) using a variant of the LevenbergMarquardt algorithm.
The user must supply a specication of the regression function; prior to giving this specication the
parameters to be estimated must be declared and given initial values. Optionally, the user may
supply analytical derivatives of the regression function with respect to each of the parameters.
If derivatives are not given, the user must instead give a list of the parameters to be estimated
(separated by spaces or commas), preceded by the keyword params. The tolerance (criterion for
terminating the iterative estimation procedure) can be adjusted using the set command.
The syntax for specifying the function to be estimated is the same as for the genr command. Here
are two examples, with accompanying derivatives.
# Consumption function from Greene
nls C = alpha + beta * Y^gamma
deriv alpha = 1
deriv beta = Y^gamma
deriv gamma = beta * Y^gamma * log(Y)
end nls
# Nonlinear function from Russell Davidson
nls y = alpha + beta * x1 + (1/beta) * x2
deriv alpha = 1
deriv beta = x1 - x2/(beta*beta)
end nls --vcv
Note the command words nls (which introduces the regression function), deriv (which introduces
the specication of a derivative), and end nls, which terminates the specication and calls for
estimation. If the --vcv ag is appended to the last line the covariance matrix of the parameter
estimates is printed.
20.2 Initializing the parameters
The parameters of the regression function must be given initial values prior to the nls command.
This can be done using the genr command (or, in the GUI program, via the menu item Variable,
Dene new variable).
In some cases, where the nonlinear function is a generalization of (or a restricted form of) a linear
model, it may be convenient to run an ols and initialize the parameters from the OLS coecient
estimates. In relation to the rst example above, one might do:
ols C 0 Y
genr alpha = $coeff(0)
genr beta = $coeff(Y)
genr gamma = 1
And in relation to the second example one might do:
170
Chapter 20. Nonlinear least squares 171
ols y 0 x1 x2
genr alpha = $coeff(0)
genr beta = $coeff(x1)
20.3 NLS dialog window
It is probably most convenient to compose the commands for NLS estimation in the form of a
gretl script but you can also do so interactively, by selecting the item Nonlinear Least Squares
under the Model, Nonlinear models menu. This opens a dialog box where you can type the
function specication (possibly prefaced by genr lines to set the initial parameter values) and the
derivatives, if available. An example of this is shown in Figure 20.1. Note that in this context you
do not have to supply the nls and end nls tags.
Figure 20.1: NLS dialog box
20.4 Analytical and numerical derivatives
If you are able to gure out the derivatives of the regression function with respect to the para-
meters, it is advisable to supply those derivatives as shown in the examples above. If that is not
possible, gretl will compute approximate numerical derivatives. However, the properties of the NLS
algorithm may not be so good in this case (see section 20.7).
This is done by using the params statement, which should be followed by a list of identiers
containing the parameters to be estimated. In this case, the examples above would read as follows:
# Greene
nls C = alpha + beta * Y^gamma
params alpha beta gamma
end nls
# Davidson
nls y = alpha + beta * x1 + (1/beta) * x2
params alpha beta
end nls
If analytical derivatives are supplied, they are checked for consistency with the given nonlinear
function. If the derivatives are clearly incorrect estimation is aborted with an error message. If the
derivatives are suspicious a warning message is issued but estimation proceeds. This warning
may sometimes be triggered by incorrect derivatives, but it may also be triggered by a high degree
of collinearity among the derivatives.
Note that you cannot mix analytical and numerical derivatives: you should supply expressions for
all of the derivatives or none.
20.5 Controlling termination
The NLS estimation procedure is an iterative process. Iteration is terminated when the criterion for
convergence is met or when the maximum number of iterations is reached, whichever comes rst.
Let k denote the number of parameters being estimated. The maximum number of iterations is
100 (k + 1) when analytical derivatives are given, and 200 (k + 1) when numerical derivatives
are used.
Let denote a small number. The iteration is deemed to have converged if at least one of the
following conditions is satised:
Both the actual and predicted relative reductions in the error sum of squares are at most .
The relative error between two consecutive iterates is at most .
This default value of is the machine precision to the power 3/4,
1
but it can be adjusted using the
set command with the parameter nls_toler. For example
set nls_toler .0001
will relax the value of to 0.0001.
20.6 Details on the code
The underlying engine for NLS estimation is based on the minpack suite of functions, available
from netlib.org. Specically, the following minpack functions are called:
lmder LevenbergMarquardt algorithm with analytical derivatives
chkder Check the supplied analytical derivatives
lmdif LevenbergMarquardt algorithm with numerical derivatives
fdjac2 Compute nal approximate Jacobian when using numerical derivatives
dpmpar Determine the machine precision
On successful completion of the LevenbergMarquardt iteration, a GaussNewton regression is used
to calculate the covariance matrix for the parameter estimates. If the --robust ag is given a
robust variant is computed. The documentation for the set command explains the specic options
available in this regard.
Since NLS results are asymptotic, there is room for debate over whether or not a correction for
degrees of freedom should be applied when calculating the standard error of the regression (and
the standard errors of the parameter estimates). For comparability with OLS, and in light of the
reasoning given in Davidson and MacKinnon (1993), the estimates shown in gretl do use a degrees
of freedom correction.
1
On a 32-bit Intel Pentium machine a likely value for this parameter is 1.82 10
12
.
20.7 Numerical accuracy
Table 20.1 shows the results of running the gretl NLS procedure on the 27 Statistical Reference
Datasets made available by the U.S. National Institute of Standards and Technology (NIST) for test-
ing nonlinear regression software.
2
For each dataset, two sets of starting values for the parameters
are given in the test les, so the full test comprises 54 runs. Two full tests were performed, one
using all analytical derivatives and one using all numerical approximations. In each case the default
tolerance was used.
3
Out of the 54 runs, gretl failed to produce a solution in 4 cases when using analytical derivatives,
and in 5 cases when using numeric approximation. Of the four failures in analytical derivatives
mode, two were due to non-convergence of the LevenbergMarquardt algorithm after the maximum
number of iterations (on MGH09 and Bennett5, both described by NIST as of Higher diculty) and
two were due to generation of range errors (out-of-bounds oating point values) when computing
the Jacobian (on BoxBOD and MGH17, described as of Higher diculty and Average diculty
respectively). The additional failure in numerical approximation mode was on MGH10 (Higher di-
culty, maximum number of iterations reached).
The table gives information on several aspects of the tests: the number of outright failures, the
average number of iterations taken to produce a solution and two sorts of measure of the accuracy
of the estimates for both the parameters and the standard errors of the parameters.
For each of the 54 runs in each mode, if the run produced a solution the parameter estimates
obtained by gretl were compared with the NIST certied values. We dene the minimum correct
gures for a given run as the number of signicant gures to which the least accurate gretl esti-
mate agreed with the certied value, for that run. The table shows both the average and the worst
case value of this variable across all the runs that produced a solution. The same information is
shown for the estimated standard errors.
4
The second measure of accuracy shown is the percentage of cases, taking into account all parame-
ters from all successful runs, in which the gretl estimate agreed with the certied value to at least
the 6 signicant gures which are printed by default in the gretl regression output.
Using analytical derivatives, the worst case values for both parameters and standard errors were
improved to 6 correct gures on the test machine when the tolerance was tightened to 1.0e14.
Using numerical derivatives, the same tightening of the tolerance raised the worst values to 5
correct gures for the parameters and 3 gures for standard errors, at a cost of one additional
failure of convergence.
Note the overall superiority of analytical derivatives: on average solutions to the test problems
were obtained with substantially fewer iterations and the results were more accurate (most notably
for the estimated standard errors). Note also that the six-digit results printed by gretl are not 100
percent reliable for dicult nonlinear problems (in particular when using numerical derivatives).
Having registered this caveat, the percentage of cases where the results were good to six digits or
better seems high enough to justify their printing in this form.
2
For a discussion of gretls accuracy in the estimation of linear models, see Appendix D.
3
The data shown in the table were gathered from a pre-release build of gretl version 1.0.9, compiled with gcc 3.3,
linked against glibc 2.3.2, and run under Linux on an i686 PC (IBM ThinkPad A21m).
4
For the standard errors, I excluded one outlier from the statistics shown in the table, namely Lanczos1. This is an
odd case, using generated data with an almost-exact t: the standard errors are 9 or 10 orders of magnitude smaller
than the coecients. In this instance gretl could reproduce the certied standard errors to only 3 gures (analytical
derivatives) and 2 gures (numerical derivatives).
Table 20.1: Nonlinear regression: the NIST tests
Analytical derivatives Numerical derivatives
Failures in 54 tests 4 5
Average iterations 32 127
Mean of min. correct gures, 8.120 6.980
parameters
Worst of min. correct gures, 4 3
parameters
Mean of min. correct gures, 8.000 5.673
standard errors
Worst of min. correct gures, 5 2
standard errors
Percent correct to at least 6 gures, 96.5 91.9
parameters
Percent correct to at least 6 gures, 97.7 77.3
standard errors
Chapter 21
Maximum likelihood estimation
21.1 Generic ML estimation with gretl
Maximum likelihood estimation is a cornerstone of modern inferential procedures. Gretl provides
a way to implement this method for a wide range of estimation problems, by use of the mle com-
mand. We give here a few examples.
To give a foundation for the examples that follow, we start from a brief reminder on the basics
of ML estimation. Given a sample of size T, it is possible to dene the density function
1
for the
whole sample, namely the joint distribution of all the observations f(Y; ), where Y =
_
y
1
, . . . , y
T
_
.
Its shape is determined by a k-vector of unknown parameters , which we assume is contained in
a set , and which can be used to evaluate the probability of observing a sample with any given
characteristics.
After observing the data, the values Y are given, and this function can be evaluated for any legiti-
mate value of . In this case, we prefer to call it the likelihood function; the need for another name
stems from the fact that this function works as a density when we use the y
t
s as arguments and
as parameters, whereas in this context is taken as the functions argument, and the data Y only
have the role of determining its shape.
In standard cases, this function has a unique maximum. The location of the maximum is unaected
if we consider the logarithm of the likelihood (or log-likelihood for short): this function will be
denoted as
() = logf(Y; )
The log-likelihood functions that gretl can handle are those where () can be written as
() =
T
_
t=1
t
()
which is true in most cases of interest. The functions
t
() are called the log-likelihood contribu-
tions.
Moreover, the location of the maximum is obviously determined by the data Y. This means that the
value
(Y) =Argmax
() (21.1)
is some function of the observed data (a statistic), which has the property, under mild conditions,
of being a consistent, asymptotically normal and asymptotically ecient estimator of .
Sometimes it is possible to write down explicitly the function

(Y); in general, it need not be so. In
these circumstances, the maximum can be found by means of numerical techniques. These often
rely on the fact that the log-likelihood is a smooth function of , and therefore on the maximum
its partial derivatives should all be 0. The gradient vector, or score vector, is a function that enjoys
many interesting statistical properties in its own right; it will be denoted here as g(). It is a
1
We are supposing here that our data are a realization of continuous random variables. For discrete random variables,
everything continues to apply by referring to the probability function instead of the density. In both cases, the distribution
may be conditional on some exogenous variables.
175
Chapter 21. Maximum likelihood estimation 176
k-vector with typical element
g
i
() =
()
i
=
T
_
t=1
t
()
i
Gradient-based methods can be shortly illustrated as follows:
1. pick a point
0
;
2. evaluate g(
0
);
3. if g(
0
) is small, stop. Otherwise, compute a direction vector d(g(
0
));
4. evaluate
1
=
0
+d(g(
0
));
5. substitute
0
with
1
;
6. restart from 2.
Many algorithms of this kind exist; they basically dier from one another in the way they compute
the direction vector d(g(
0
)), to ensure that (
1
) > (
0
) (so that we eventually end up on the
maximum).
The default method gretl uses to maximize the log-likelihood is a gradient-based algorithm known
as the BFGS (Broyden, Fletcher, Goldfarb and Shanno) method. This technique is used in most
econometric and statistical packages, as it is well-established and remarkably powerful. Clearly,
in order to make this technique operational, it must be possible to compute the vector g() for
any value of . In some cases this vector can be written explicitly as a function of Y. If this is
not possible or too dicult the gradient may be evaluated numerically. The alternative Newton-
Raphson algorithm is also available, which is more eective under some circumstances but is also
more fragile; see section 21.8 and chapter 31 for details.
The choice of the starting value,
0
, is crucial in some contexts and inconsequential in others. In
general, however, it is advisable to start the algorithm from sensible values whenever possible. If
a consistent estimator is available, this is usually a safe and ecient choice: this ensures that in
large samples the starting point will be likely close to

and convergence can be achieved in few
iterations.
The maximum number of iterations allowed for the BFGS procedure, and the relative tolerance
for assessing convergence, can be adjusted using the set command: the relevant variables are
bfgs_maxiter (default value 500) and bfgs_toler (default value, the machine precision to the
power 3/4).
Covariance matrix and standard errors
By default the covariance matrix of the parameter estimates is based on the Outer Product of the
Gradient. That is,
Var
OPG
(
) =
_
G
)G(
)
_
1
(21.2)
where G(
) is the T k matrix of contributions to the gradient. Two other options are available. If
the --hessian ag is given, the covariance matrix is computed from a numerical approximation to
the Hessian at convergence. If the --robust option is selected, the quasi-ML sandwich estimator
is used:
Var
QML
(
) = H(
)
1
G
)G(
)H(
)
1
where H denotes the numerical approximation to the Hessian.
21.2 Gamma estimation
Suppose we have a sample of T independent and identically distributed observations froma Gamma
distribution. The density function for each observation x
t
is
f(x
t
) =

p
(p)
x
p1
t
exp(x
t
) (21.3)
The log-likelihood for the entire sample can be written as the logarithm of the joint density of all
the observations. Since these are independent and identical, the joint density is the product of the
individual densities, and hence its log is
(, p) =
T
_
t=1
log
_

p
(p)
x
p1
t
exp(x
t
)
_
=
T
_
t=1
t
(21.4)
where
t
= p log(x
t
) (p) logx
t
x
t
and () is the log of the gamma function. In order to estimate the parameters and p via ML, we
need to maximize (21.4) with respect to them. The corresponding gretl code snippet is
scalar alpha = 1
scalar p = 1
mle logl = p*ln(alpha * x) - lngamma(p) - ln(x) - alpha * x
params alpha p
end mle
The rst two statements
alpha = 1
p = 1
are necessary to ensure that the variables alpha and p exist before the computation of logl is
attempted. Inside the mle block these variables (which could be either scalars, vectors or a com-
bination of the two see below for an example) are identied as the parameters that should be
adjusted to maximize the likelihood via the params keyword. Their values will be changed by the
execution of the mle command; upon successful completion, they will be replaced by the ML esti-
mates. The starting value is 1 for both; this is arbitrary and does not matter much in this example
(more on this later).
The above code can be made more readable, and marginally more ecient, by dening a variable
to hold x
t
. This command can be embedded in the mle block as follows:
mle logl = p*ln(ax) - lngamma(p) - ln(x) - ax
series ax = alpha*x
params alpha p
end mle
The variable ax is not added to the params list, of course, since it is just an auxiliary variable to
facilitate the calculations. You can insert as many such auxiliary lines as you require before the
params line, with the restriction that they must contain either (a) commands to generate series,
scalars or matrices or (b) print commands (which may be used to aid in debugging).
In a simple example like this, the choice of the starting values is almost inconsequential; the algo-
rithm is likely to converge no matter what the starting values are. However, consistent method-of-
moments estimators of p and can be simply recovered from the sample mean m and variance V:
since it can be shown that
E(x
t
) = p/ V(x
t
) = p/
2
it follows that the following estimators
= m/V
p = m
are consistent, and therefore suitable to be used as starting point for the algorithm. The gretl script
code then becomes
scalar m = mean(x)
scalar alpha = m/var(x)
scalar p = m*alpha
mle logl = p*ln(ax) - lngamma(p) - ln(x) - ax
series ax = alpha*x
params alpha p
end mle
Another thing to note is that sometimes parameters are constrained within certain boundaries: in
this case, for example, both and p must be positive numbers. Gretl does not check for this: it
is the users responsibility to ensure that the function is always evaluated at an admissible point
in the parameter space during the iterative search for the maximum. An eective technique is to
dene a variable for checking that the parameters are admissible and setting the log-likelihood as
undened if the check fails. An example, which uses the conditional assignment operator, follows:
scalar m = mean(x)
scalar alpha = m/var(x)
scalar p = m*alpha
mle logl = check ? p*ln(ax) - lngamma(p) - ln(x) - ax : NA
series ax = alpha*x
scalar check = (alpha>0) && (p>0)
params alpha p
end mle
21.3 Stochastic frontier cost function
When modeling a cost function, it is sometimes worthwhile to incorporate explicitly into the sta-
tistical model the notion that rms may be inecient, so that the observed cost deviates from the
theoretical gure not only because of unobserved heterogeneity between rms, but also because
two rms could be operating at a dierent eciency level, despite being identical under all other
respects. In this case we may write
C
i
= C
i
+u
i
+v
i
where C
i
is some variable cost indicator, C
i
is its theoretical value, u
i
is a zero-mean disturbance
term and v
i
is the ineciency term, which is supposed to be nonnegative by its very nature.
A linear specication for C
i
is often chosen. For example, the CobbDouglas cost function arises
when C
i
is a linear function of the logarithms of the input prices and the output quantities.
The stochastic frontier model is a linear model of the form y
i
= x
i
+
i
in which the error term
i
is the sum of u
i
and v
i
. A common postulate is that u
i
N(0,
2
u
) and v
i

N(0,
2
v
)
. If
independence between u
i
and v
i
is also assumed, then it is possible to show that the density
function of
i
has the form:
f(
i
) =
_
1
_
(21.5)
where () and () are, respectively, the distribution and density function of the standard normal,
=
_
2
u
+
2
v
and =

u
v
.
As a consequence, the log-likelihood for one observation takes the form (apart form an irrelevant
constant)
t
= log
_
_
log() +

2
i
2
2
_
Therefore, a CobbDouglas cost function with stochastic frontier is the model described by the
following equations:
logC
i
= logC
i
+
i
logC
i
= c +
m
_
j=1
j
logy
ij
+
n
_
j=1
j
logp
ij
i
= u
i
+v
i
u
i
N(0,
2
u
)
v
i

N(0,
2
v
)
In most cases, one wants to ensure that the homogeneity of the cost function with respect to
the prices holds by construction. Since this requirement is equivalent to

n
j=1
j
= 1, the above
equation for C
i
can be rewritten as
logC
i
logp
in
= c +
m
_
j=1
j
logy
ij
+
n
_
j=2
j
(logp
ij
logp
in
) +
i
(21.6)
The above equation could be estimated by OLS, but it would suer from two drawbacks: rst,
the OLS estimator for the intercept c is inconsistent because the disturbance term has a non-zero
expected value; second, the OLS estimators for the other parameters are consistent, but inecient
in view of the non-normality of
i
. Both issues can be addressed by estimating (21.6) by maximum
likelihood. Nevertheless, OLS estimation is a quick and convenient way to provide starting values
for the MLE algorithm.
Example 21.1 shows how to implement the model described so far. The banks91 le contains part
of the data used in Lucchetti, Papi and Zazzaro (2001).
The script in example 21.1 is relatively easy to modify to show how one can use vectors (that is,
1-dimensional matrices) for storing the parameters to optimize: example 21.2 holds essentially the
same script in which the parameters of the cost function are stored together in a vector. Of course,
this makes also possible to use variable lists and other renements which make the code more
compact and readable.
21.4 GARCH models
GARCH models are handled by gretl via a native function. However, it is instructive to see how they
can be estimated through the mle command.
2
The following equations provide the simplest example of a GARCH(1,1) model:
y
t
= +
t
t
= u
t

t
u
t
N(0, 1)
h
t
= +
2
t1
+h
t1
.
Since the variance of y
t
depends on past values, writing down the log-likelihood function is not
simply a matter of summing the log densities for individual observations. As is common in time
2
The gig addon, which handles other variants of conditionally heteroskedastic models, uses mle as its internal engine.
Example 21.1: Estimation of stochastic frontier cost function (with scalar parameters)
open banks91
# Cobb-Douglas cost function
ols cost const y p1 p2 p3
# Cobb-Douglas cost function with homogeneity restrictions
genr rcost = cost - p3
genr rp1 = p1 - p3
genr rp2 = p2 - p3
ols rcost const y rp1 rp2
# and inefficiency
scalar b0 = $coeff(const)
scalar b1 = $coeff(y)
scalar b2 = $coeff(rp1)
scalar b3 = $coeff(rp2)
scalar su = 0.1
scalar sv = 0.1
mle logl = ln(cnorm(e*lambda/ss)) - (ln(ss) + 0.5*(e/ss)^2)
scalar ss = sqrt(su^2 + sv^2)
scalar lambda = su/sv
series e = rcost - b0*const - b1*y - b2*rp1 - b3*rp2
params b0 b1 b2 b3 su sv
end mle
Example 21.2: Estimation of stochastic frontier cost function (with matrix parameters)
open banks91
# Cobb-Douglas cost function
ols cost const y p1 p2 p3
genr rcost = cost - p3
genr rp1 = p1 - p3
genr rp2 = p2 - p3
list X = const y rp1 rp2
ols rcost X
# and inefficiency
matrix b = $coeff
scalar su = 0.1
scalar sv = 0.1
mle logl = ln(cnorm(e*lambda/ss)) - (ln(ss) + 0.5*(e/ss)^2)
scalar ss = sqrt(su^2 + sv^2)
scalar lambda = su/sv
series e = rcost - lincomb(X, b)
params b su sv
end mle
series models, y
t
cannot be considered independent of the other observations in our sample, and
consequently the density function for the whole sample (the joint density for all observations) is
not just the product of the marginal densities.
Maximum likelihood estimation, in these cases, is achieved by considering conditional densities, so
what we maximize is a conditional likelihood function. If we dene the information set at time t as
F
t
=
_
y
t
, y
t1
, . . .
_
,
then the density of y
t
conditional on F
t1
is normal:
y
t
F
t1
N[, h
t
] .
By means of the properties of conditional distributions, the joint density can be factorized as
follows
f(y
t
, y
t1
, . . .) =
_
_
T
t=1
f(y
t
F
t1
)
_
_
f(y
0
)
If we treat y
0
as xed, then the termf(y
0
) does not depend on the unknown parameters, and there-
fore the conditional log-likelihood can then be written as the sum of the individual contributions
as
(, , , ) =
T
_
t=1
t
(21.7)
where
t
= log
_
1
_
h
t
_
y
t
_
h
t
__
=
1
2
_
log(h
t
) +
(y
t
)
2
h
t
_
The following script shows a simple application of this technique, which uses the data le djclose;
it is one of the example dataset supplied with gretl and contains daily data from the Dow Jones
stock index.
open djclose
series y = 100*ldiff(djclose)
scalar mu = 0.0
scalar omega = 1
scalar alpha = 0.4
scalar beta = 0.0
mle ll = -0.5*(log(h) + (e^2)/h)
series e = y - mu
series h = var(y)
series h = omega + alpha*(e(-1))^2 + beta*h(-1)
params mu omega alpha beta
end mle
21.5 Analytical derivatives
Computation of the score vector is essential for the working of the BFGS method. In all the previous
examples, no explicit formula for the computation of the score was given, so the algorithm was fed
numerically evaluated gradients. Numerical computation of the score for the i-th parameter is
performed via a nite approximation of the derivative, namely
(
1
, . . . ,
n
)
i
=
(
1
, . . . ,
i
+h, . . . ,
n
) (
1
, . . . ,
i
h, . . . ,
n
)
2h
where h is a small number.
In many situations, this is rather ecient and accurate. A better approximation to the true deriva-
tive may be obtained by forcing mle to use a technique known as Richardson Extrapolation, which
gives extremely precise results, but is considerably more CPU-intensive. This feature may be turned
on by using the set command as in
set bfgs_richardson on
However, one might want to avoid the approximation and specify an exact function for the deriva-
tives. As an example, consider the following script:
nulldata 1000
genr x1 = normal()
genr x2 = normal()
genr x3 = normal()
genr ystar = x1 + x2 + x3 + normal()
genr y = (ystar > 0)
scalar b0 = 0
scalar b1 = 0
scalar b2 = 0
scalar b3 = 0
mle logl = y*ln(P) + (1-y)*ln(1-P)
series ndx = b0 + b1*x1 + b2*x2 + b3*x3
series P = cnorm(ndx)
params b0 b1 b2 b3
end mle --verbose
Here, 1000 data points are articially generated for an ordinary probit model:
3
y
t
is a binary
variable, which takes the value 1 if y
t
=
1
x
1t
+
2
x
2t
+
3
x
3t
+
t
> 0 and 0 otherwise. Therefore,
y
t
= 1 with probability (
1
x
1t
+
2
x
2t
+
3
x
3t
) =
t
. The probability function for one observation
can be written as
P(y
t
) =
y
t
t
(1
t
)
1y
t
Since the observations are independent and identically distributed, the log-likelihood is simply the
sum of the individual contributions. Hence
=
T
_
t=1
y
t
log(
t
) +(1 y
t
) log(1
t
)
The --verbose switch at the end of the end mle statement produces a detailed account of the
iterations done by the BFGS algorithm.
In this case, numerical dierentiation works rather well; nevertheless, computation of the analytical
score is straightforward, since the derivative

i
can be written as
i
=

i
via the chain rule, and it is easy to see that
t
=
y
t
1 y
t
1
t
i
= (
1
x
1t
+
2
x
2t
+
3
x
3t
) x
it
3
Again, gretl does provide a native probit command (see section 32.1), but a probit model makes for a nice example
here.
The mle block in the above script can therefore be modied as follows:
series ndx = b0 + b1*x1 + b2*x2 + b3*x3
series m = dnorm(ndx)*(y/P - (1-y)/(1-P))
deriv b0 = m
deriv b1 = m*x1
deriv b2 = m*x2
deriv b3 = m*x3
end mle --verbose
Note that the params statement has been replaced by a series of deriv statements; these have the
double function of identifying the parameters over which to optimize and providing an analytical
expression for their respective score elements.
21.6 Debugging ML scripts
We have discussed above the main sorts of statements that are permitted within an mle block,
namely
auxiliary commands to generate helper variables;
deriv statements to specify the gradient with respect to each of the parameters; and
a params statement to identify the parameters in case analytical derivatives are not given.
For the purpose of debugging ML estimators one additional sort of statement is allowed: you can
print the value of a relevant variable at each step of the iteration. This facility is more restricted
then the regular print command. The command word print should be followed by the name of
just one variable (a scalar, series or matrix).
In the last example above a key variable named m was generated, forming the basis for the analytical
derivatives. To track the progress of this variable one could add a print statement within the ML
block, as in
print m
21.7 Using functions
The mle command allows you to estimate models that gretl does not provide natively: in some
cases, it may be a good idea to wrap up the mle block in a user-dened function (see Chapter 13),
so as to extend gretls capabilities in a modular and exible way.
As an example, we will take a simple case of a model that gretl does not yet provide natively:
the zero-inated Poisson model, or ZIP for short.
4
In this model, we assume that we observe a
mixed population: for some individuals, the variable y
t
is (conditionally on a vector of exogenous
covariates x
t
) distributed as a Poisson random variate; for some others, y
t
is identically 0. The
trouble is, we dont know which category a given individual belongs to.
For instance, suppose we have a sample of women, and the variable y
t
represents the number of
children that woman t has. There may be a certain proportion, , of women for whom y
t
= 0 with
certainty (maybe out of a personal choice, or due to physical impossibility). But there may be other
4
The actual ZIP model is in fact a bit more general than the one presented here. The specialized version discussed in
this section was chosen for the sake of simplicity. For futher details, see Greene (2003).
women for whom y
t
= 0 just as a matter of chance they havent happened to have any children
at the time of observation.
In formulae:
P(y
t
= kx
t
) = d
t
+(1 )
_
e
y
t
t
y
t
!
_
t
= exp(x
t
)
d
t
=
_
1 for y
t
= 0
0 for y
t
> 0
Writing a mle block for this model is not dicult:
mle ll = logprob
series xb = exp(b0 + b1 * x)
series d = (y=0)
series poiprob = exp(-xb) * xb^y / gamma(y+1)
series logprob = (alpha>0) && (alpha<1) ? \
log(alpha*d + (1-alpha)*poiprob) : NA
params alpha b0 b1
end mle -v
However, the code above has to be modied each time we change our specication by, say, adding
an explanatory variable. Using functions, we can simplify this task considerably and eventually be
able to write something easy like
list X = const x
zip(y, X)
Example 21.3: Zero-inated Poisson Model user-level function
/*
user-level function: estimate the model and print out
the results
*/
function void zip(series y, list X)
matrix coef_stde = zip_estimate(y, X)
printf "\nZero-inflated Poisson model:\n"
string parnames = "alpha,"
string parnames += varname(X)
modprint coef_stde parnames
end function
Lets see how this can be done. First we need to dene a function called zip() that will take two ar-
guments: a dependent variable y and a list of explanatory variables X. An example of such function
can be seen in script 21.3. By inspecting the function code, you can see that the actual estimation
does not happen here: rather, the zip() function merely uses the built-in modprint command to
print out the results coming from another user-written function, namely zip_estimate().
The function zip_estimate() is not meant to be executed directly; it just contains the number-
crunching part of the job, whose results are then picked up by the end function zip(). In turn,
zip_estimate() calls other user-written functions to perform other tasks. The whole set of in-
ternal functions is shown in the panel 21.4.
Example 21.4: Zero-inated Poisson Model internal functions
/* compute log probabilities for the plain Poisson model */
function series ln_poi_prob(series y, list X, matrix beta)
series xb = lincomb(X, beta)
return -exp(xb) + y*xb - lngamma(y+1)
end function
/* compute log probabilities for the zero-inflated Poisson model */
function series ln_zip_prob(series y, list X, matrix beta, scalar p0)
# check if the probability is in [0,1]; otherwise, return NA
if (p0>1) || (p0<0)
series ret = NA
else
series ret = ln_poi_prob(y, X, beta) + ln(1-p0)
series ret = (y=0) ? ln(p0 + exp(ret)) : ret
endif
return ret
end function
/* do the actual estimation (silently) */
function matrix zip_estimate(series y, list X)
# initialize alpha to a "sensible" value: half the frequency
# of zeros in the sample
scalar alpha = mean(y=0)/2
# initialize the coeffs (we assume the first explanatory
# variable is the constant here)
matrix coef = zeros(nelem(X), 1)
coef[1] = mean(y) / (1-alpha)
# do the actual ML estimation
mle ll = ln_zip_prob(y, X, coef, alpha)
params alpha coef
end mle --hessian --quiet
return $coeff ~ $stderr
end function
All the functions shown in 21.3 and 21.4 can be stored in a separate inp le and executed once, at
the beginning of our job, by means of the include command. Assuming the name of this script le
is zip_est.inp, the following is an example script which (a) includes the script le, (b) generates a
simulated dataset, and (c) performs the estimation of a ZIP model on the articial data.
set echo off
set messages off
# include the user-written functions
include zip_est.inp
# generate the artificial data
nulldata 1000
set seed 732237
scalar truep = 0.2
scalar b0 = 0.2
scalar b1 = 0.5
series x = normal()
series y = (uniform()<truep) ? 0 : genpois(exp(b0 + b1*x))
list X = const x
# estimate the zero-inflated Poisson model
zip(y, X)
The results are as follows:
Zero-inflated Poisson model:
coefficient std. error z-stat p-value
-------------------------------------------------------
alpha 0.203069 0.0238035 8.531 1.45e-17 ***
const 0.257014 0.0417129 6.161 7.21e-10 ***
x 0.466657 0.0321235 14.53 8.17e-48 ***
A further step may then be creating a function package for accessing your new zip() function via
gretls graphical interface. For details on how to do this, see section 13.5.
21.8 Advanced use of mle: functions, analytical derivatives, algorithm choice
All the techniques decribed in the previous sections may be combined, and mle can be used for
solving non-standard estimation problems (provided, of course, that one chooses maximum likeli-
hood as the preferred inference method).
The strategy that, as of this writing, has proven most successful in designing scripts for this pur-
pose is:
Modularize your code as much as possible.
Use analytical derivatives whenever possible.
Choose your optimization method wisely.
In the rest of this section, we will expand on the probit example of section 21.5 to give the reader
an idea of what a heavy-duty application of mle looks like. Most of the code fragments come from
mle-advanced.inp, which is one of the sample scripts supplied with the standard installation of
gretl (see under File > Script les > Practice File).
BFGS with and without analytical derivatives
The example in section 21.5 can be made more general by using matrices and user-written func-
tions. Consider the following code fragment:
list X = const x1 x2 x3
matrix b = zeros(nelem(X),1)
series ndx = lincomb(X, b)
params b
end mle
In this context, the fact that the model we are estimating has four explanatory variables is totally
incidental: the code is written in such a way that we could change the content of the list X without
having to make any other modication. This was made possible by:
1. gathering the parameters to estimate into a single vector b rather than using separate scalars;
2. using the nelem() function to initialize b, so that its dimension is kept track of automatically;
3. using the lincomb() function to compute the index function.
A parallel enhancement could be achieved in the case of analytically computed derivatives: since
b is now a vector, mle expects the argument to the deriv keyword to be a matrix, in which each
column is the partial derivative to the corresponding element of b. It is useful to re-write the score
for the i-th observation as
= m
i
x
i
(21.8)
where m
i
is the signed Mills ratio, that is
m
i
= y
i
(x
i
)
(x
i
)
(1 y
i
)
(x
i
)
1 (x
i
)
,
which was computed in section 21.5 via
Here, we will code it in a somewhat terser way as
series m = y ? invmills(-ndx) : -invmills(ndx)
and make use of the conditional assignment operator and of the specialized function invmills()
for eciency. Building the score matrix is now easily achieved via
matrix mX = {X}
deriv b = mX .* {m}
end mle
in which the {} operator was used to turn series and lists into matrices (see chapter 15). However,
proceeding in this way for more complex models than probit may imply inserting into the mle
block a long series of instructions; the example above merely happens to be short because the
score matrix for the probit model is very easy to write in matrix form.
A better solution is writing a user-level function to compute the score and using that inside the mle
block, as in
function matrix score(matrix b, series y, list X)
return {m} .* {X}
end function
[...]
deriv b = score(b, y, X)
end mle
In this way, no matter how complex the computation of the score is, the mle block remains nicely
compact.
Newtons method and the analytical Hessian
Since version 1.9.7, gretl oers the user the option of using Newtons method for maximizing the
log-likelihood. In terms of the notation used in section 21.1, the direction for updating the inital
parameter vector
0
is given by
d[g(
0
)] = H(
0
)
1
g(
0
), (21.9)
where H() is the Hessian of the total loglikelihood computed at and 0 < < 1 is a scalar called
the step length.
The above expression makes a few points clear:
1. At each step, it must be possible to compute not only the score g(), but also its derivative
H();
2. the matrix H() should be nonsingular;
3. it is assumed that for some positive value of , (
1
) > (
0
); in other words, that going in
the direction d[g(
0
)] leads upwards for some step length.
The strength of Newtons method lies in the fact that, if the loglikelihood is globally concave,
then (21.9) enjoys certain optimality properties and the number of iterations required to reach the
maximum is often much smaller than it would be with other methods, such as BFGS. However, it
may have some disadvantages: for a start, the Hessian H() may be dicult or very expensive to
compute; moreover, the loglikelihood may not be globally concave, so for some values of , the
matrix H() is not negative denite or perhaps even singular. Those cases are handled by gretls
implementation of Newtons algorithm by means of several heuristic techniques
5
, but a number of
adverse consequences may occur, which range from longer computation time for optimization to
non-convergence of the algorithm.
As a consequence, using Newtons method is advisable only when the computation of the Hessian
is not too CPU-intensive and the nature of the estimator is such that it is known in advance that
5
The gist to it is that, if H is not negative denite, it is substituted by k dg(H) + (1 k) H, where k is a suitable
scalar; however, if youre interested in the precise details, youll be much better o looking at the source code: the le
youll want to look at is lib/src/gretl_bfgs.c.
the loglikelihood is globally concave. The probit models satises both requisites, so we will expand
the preceding example to illustrate how to use Newtons method in gretl.
A rst example may be given simply by issuing the command
set optimizer newton
before the mle block.
6
This will instruct gretl to use Newtons method instead of BFGS. If the deriv
keyword is used, gretl will dierentiate the score function numerically; otherwise, if the score
has to be computed itself numerically, gretl will calculate H() by dierentiating the loglikelihood
numerically twice. The latter solution, though, is generally to be avoided, as may be extremely
time-consuming and may yield imprecise results.
A much better option is to calculate the Hessian analytically and have gretl use its true value rather
than a numerical approximation. In most cases, this is both much faster and numerically stable,
but of course comes at the price of having to dierentiate the loglikelihood twice to respect with
the parameters and translate the resulting expressions into ecient hansl code.
Luckily, both tasks are relatively easy in the probit case: the matrix of second derivatives of
i
may
be written as
= m
i
_
m
i
+x
_
x
i
x
i
so the total Hessian is
n
_
i=1
= X
_
_
_
_
_
_
_
w
1
w
2
.
.
.
w
n
_
_
_
_
_
_
_
X (21.10)
where w
i
= m
i
_
m
i
+x
_
. It can be shown that w
i
> 0, so the Hessian is guaranteed to be negative
denite in all sensible cases and the conditions are ideal for applying Newtons method.
A hansl translation of equation (21.10) may look like
function void Hess(matrix *H, matrix b, series y, list X)
/* computes the negative Hessian for a Probit model */
series w = m*(m+ndx)
matrix mX = {X}
H = (mX .* {w})mX
end function
There are two characteristics worth noting of the function above. For a start, it doesnt return
anything: the result of the computation is simply stored in the matrix pointed at by the rst
argument of the function. Second, the result is not the Hessian proper, but rather its negative. This
function becomes usable from within an mle block by the keyword hessian. The syntax is
mle ...
...
hessian funcname(&mat_addr, ...)
end mle
In other words, the hessian keyword must be followed by the call to a function whose rst argu-
ment is a matrix pointer which is supposed to be lled with the negative of the Hessian at .
Another feature worth noting is that gretl does not perform any numerical check on whether the
function computes the Hessian correctly or not. On the one hand, this means that you can trick
6
To go back to BFGS, you use set optimizer bfgs.
mle into using alternatives to the Hessian and thereby implement other optimization methods. For
example, if you substitute in equation 21.9 the Hessian H with the negative of the OPG matrix G
G,
as dened in (21.2), you get the so-called BHHH optimization method (see Berndt et al. (1974)).
Again, the sample le mle-advanced.inp provides an example. On the other hand, you may want
to perform a check of your analytically-computed H matrix versus a numerical approximation.
If you have a function that computes the score, this is relatively simple to do by using the fdjac
function, briey described in section 31.4, which computes a numerical approximation to a deriv-
ative. In practice, you need a function computing g() as a row vector and then use fdjac to
dierentiate it numerically with respect to . The result can then be compared to your analytically-
computed Hessian. The code fragment below shows an example of how this can be done in the
probit case:
function matrix totalscore(matrix *b, series y, list X)
/* computes the total score */
return sumc(score(b, y, X))
end function
function void check(matrix b, series y, list X)
/* compares the analytical Hessian to its numerical
approximation obtained via fdjac */
matrix aH
Hess(&aH, b, y, X) # stores the analytical Hessian into aH
matrix nH = fdjac(b, "totalscore(&b, y, X)")
nH = 0.5*(nH + nH) # force symmetry
printf "Numerical Hessian\n%16.6f\n", nH
printf "Analytical Hessian (negative)\n%16.6f\n", aH
printf "Check (should be zero)\n%16.6f\n", nH + aH
end function
Chapter 22
GMM estimation
22.1 Introduction and terminology
The Generalized Method of Moments (GMM) is a very powerful and general estimation method,
which encompasses practically all the parametric estimation techniques used in econometrics. It
was introduced in Hansen (1982) and Hansen and Singleton (1982); an excellent and thorough
treatment is given in chapter 17 of Davidson and MacKinnon (1993).
The basic principle on which GMM is built is rather straightforward. Suppose we wish to estimate
a scalar parameter based on a sample x
1
, x
2
, . . . , x
T
. Let
0
indicate the true value of . Theo-
retical considerations (either of statistical or economic nature) may suggest that a relationship like
the following holds:
E
_
x
t
g()
_
= 0 =
0
, (22.1)
with g() a continuous and invertible function. That is to say, there exists a function of the data
and the parameter, with the property that it has expectation zero if and only if it is evaluated at the
true parameter value. For example, economic models with rational expectations lead to expressions
like (22.1) quite naturally.
If the sampling model for the x
t
s is such that some version of the Law of Large Numbers holds,
then
X =
1
T
T
_
t=1
x
t
p
g(
0
);
hence, since g() is invertible, the statistic
= g
1
(
X)
p
0
,
so

is a consistent estimator of . A dierent way to obtain the same outcome is to choose, as an
estimator of , the value that minimizes the objective function
F() =
_
_
1
T
T
_
t=1
(x
t
g())
_
_
2
=
_
X g()
_
2
; (22.2)
the minimum is trivially reached at

= g
1
(
X), since the expression in square brackets equals 0.

The above reasoning can be generalized as follows: suppose is an n-vector and we have m
relations like
E [f
i
(x
t
, )] = 0 for i = 1. . . m, (22.3)
where E[] is a conditional expectation on a set of p variables z
t
, called the instruments. In the
above simple example, m = 1 and f(x
t
, ) = x
t
g(), and the only instrument used is z
t
= 1.
Then, it must also be true that
E
_
f
i
(x
t
, ) z
j,t
_
= E
_
f
i,j,t
()
_
= 0 for i = 1. . . m and j = 1. . . p; (22.4)
equation (22.4) is known as an orthogonality condition, or moment condition. The GMM estimator is
dened as the minimum of the quadratic form
F(, W) =
f, (22.5)
192
Chapter 22. GMM estimation 193
where
f is a (1 m p) vector holding the average of the orthogonality conditions and W is some

symmetric, positive denite matrix, known as the weights matrix. A necessary condition for the
minimum to exist is the order condition n m p.
The statistic
=Argmin
F(, W) (22.6)
is a consistent estimator of whatever the choice of W. However, to achieve maximum asymp-
totic eciency W must be proportional to the inverse of the long-run covariance matrix of the
orthogonality conditions; if W is not known, a consistent estimator will suce.
These considerations lead to the following empirical strategy:
1. Choose a positive denite W and compute the one-step GMM estimator

1
. Customary choices
for W are I
mp
or I
m
(Z
Z)
1
.
2. Use

1
to estimate V(f
i,j,t
()) and use its inverse as the weights matrix. The resulting esti-
mator

2
is called the two-step estimator.
3. Re-estimate V(f
i,j,t
()) by means of

2
and obtain

3
; iterate until convergence. Asymp-
totically, these extra steps are unnecessary, since the two-step estimator is consistent and
ecient; however, the iterated estimator often has better small-sample properties and should
be independent of the choice of W made at step 1.
In the special case when the number of parameters n is equal to the total number of orthogonality
conditions m p, the GMM estimator

is the same for any choice of the weights matrix W, so the
rst step is sucient; in this case, the objective function is 0 at the minimum.
If, on the contrary, n < m p, the second step (or successive iterations) is needed to achieve
eciency, and the estimator so obtained can be very dierent, in nite samples, from the one-
step estimator. Moreover, the value of the objective function at the minimum, suitably scaled by
the number of observations, yields Hansens J statistic; this statistic can be interpreted as a test
statistic that has a
2
distribution with m p n degrees of freedom under the null hypothesis of
correct specication. See Davidson and MacKinnon (1993, section 17.6) for details.
In the following sections we will show how these ideas are implemented in gretl through some
examples.
22.2 OLS as GMM
It is instructive to start with a somewhat contrived example: consider the linear model y
t
= x
t
+
u
t
. Although most of us are used to read it as the sum of a hazily dened systematic part plus an
equally hazy disturbance, a more rigorous interpretation of this familiar expression comes from
the hypothesis that the conditional mean E(y
t
x
t
) is linear and the denition of u
t
as y
t
E(y
t
x
t
).
From the denition of u
t
, it follows that E(u
t
x
t
) = 0. The following orthogonality condition is
therefore available:
E [f()] = 0, (22.7)
where f() = (y
t
x
t
)x
t
. The denitions given in the previous section therefore specialize here
to:
is ;
the instrument is x
t
;
f
i,j,t
() is (y
t
x
t
)x
t
= u
t
x
t
; the orthogonality condition is interpretable as the requirement
that the regressors should be uncorrelated with the disturbances;
W can be any symmetric positive denite matrix, since the number of parameters equals the
number of orthogonality conditions. Lets say we choose I.
The function F(, W) is in this case
F(, W) =
_
_
1
T
T
_
t=1
( u
t
x
t
)
_
_
2
and it is easy to see why OLS and GMM coincide here: the GMM objective function has the
same minimizer as the objective function of OLS, the residual sum of squares. Note, however,
that the two functions are not equal to one another: at the minimum, F(, W) = 0 while the
minimized sum of squared residuals is zero only in the special case of a perfect linear t.
The code snippet contained in Example 22.1 uses gretls gmm command to make the above opera-
tional.
Example 22.1: OLS via GMM
/* initialize stuff */
series e = 0
scalar beta = 0
matrix W = I(1)
/* proceed with estimation */
gmm
series e = y - x*beta
orthog e ; x
weights W
params beta
end gmm
We feed gretl the necessary ingredients for GMM estimation in a command block, starting with gmm
and ending with end gmm. After the end gmm statement two mutually exclusive options can be
specied: --two-step or --iterate, whose meaning should be obvious.
Three elements are compulsory within a gmm block:
1. one or more orthog statements
2. one weights statement
3. one params statement
The three elements should be given in the stated order.
The orthog statements are used to specify the orthogonality conditions. They must follow the
syntax
orthog x ; Z
where x may be a series, matrix or list of series and Z may also be a series, matrix or list. In
example 22.1, the series e holds the residuals and the series x holds the regressor. If x had been
a list (a matrix), the orthog statement would have generated one orthogonality condition for each
element (column) of x. Note the structure of the orthogonality condition: it is assumed that the
term to the left of the semicolon represents a quantity that depends on the estimated parameters
(and so must be updated in the process of iterative estimation), while the term on the right is a
constant function of the data.
The weights statement is used to specify the initial weighting matrix and its syntax is straightfor-
ward. But please note. . .
In cases other than one-step estimation the specied weights matrix will be overwritten with the nal
weights on completion of the gmm command. If you wish to execute more than one GMM block with a
common starting-point it is therefore necessary to reinitialize the weights matrix between runs.
The params statement species the parameters with respect to which the GMM criterion should be
minimized; it follows the same logic and rules as in the mle and nls commands.
The minimum is found through numerical minimization via BFGS (see chapters 31 and 21). The
progress of the optimization procedure can be observed by appending the --verbose switch to
the end gmm line.
22.3 TSLS as GMM
Moving closer to the proper domain of GMM, we now consider two-stage least squares (TSLS) as a
case of GMM.
TSLS is employed in the case where one wishes to estimate a linear model of the formy
t
= X
t
+u
t
,
but where one or more of the variables in the matrix X are potentially endogenouscorrelated with
the error term, u. We proceed by identifying a set of instruments, Z
t
, which are explanatory for
the endogenous variables in X but which are plausibly uncorrelated with u. The classic two-stage
procedure is (1) regress the endogenous elements of X on Z; then (2) estimate the equation of
interest, with the endogenous elements of X replaced by their tted values from (1).
An alternative perspective is given by GMM. We dene the residual u
t
as y
t
X
t
, as usual. But
instead of relying on E(uX) = 0 as in OLS, we base estimation on the condition E(uZ) = 0. In this
case it is natural to base the initial weighting matrix on the covariance matrix of the instruments.
Example 22.2 presents a model from Stock and Watsons Introduction to Econometrics. The demand
for cigarettes is modeled as a linear function of the logs of price and income; income is treated as
exogenous while price is taken to be endogenous and two measures of tax are used as instruments.
Since we have two instruments and one endogenous variable the model is over-identied and there-
fore the weights matrix will inuence the solution. Partial output from this script is shown in 22.3.
The estimated standard errors from GMM are robust by default; if we supply the --robust option
to the tsls command we get identical results.
1
22.4 Covariance matrix options
The covariance matrix of the estimated parameters depends on the choice of W through
= (J
WJ)
1
J
WWJ(J
WJ)
1
(22.8)
where J is a Jacobian term
J
ij
=

f
i
j
and is the long-run covariance matrix of the orthogonality conditions.
Gretl computes J by numeric dierentiation (there is no provision for specifying a user-supplied
analytical expression for J at the moment). As for , a consistent estimate is needed. The simplest
choice is the sample covariance matrix of the f
t
s:
0
() =
1
T
T
_
t=1
f
t
()f
t
()
(22.9)
1
The data le used in this example is available in the Stock and Watson package for gretl. See http://gretl.
sourceforge.net/gretl_data.html.
Example 22.2: TSLS via GMM
open cig_ch10.gdt
# real avg price including sales tax
genr ravgprs = avgprs / cpi
# real avg cig-specific tax
genr rtax = tax / cpi
# real average total tax
genr rtaxs = taxs / cpi
# real average sales tax
genr rtaxso = rtaxs - rtax
# logs of consumption, price, income
genr lpackpc = log(packpc)
genr lravgprs = log(ravgprs)
genr perinc = income / (pop*cpi)
genr lperinc = log(perinc)
# restrict sample to 1995 observations
smpl --restrict year=1995
# Equation (10.16) by tsls
list xlist = const lravgprs lperinc
list zlist = const rtaxso rtax lperinc
tsls lpackpc xlist ; zlist --robust
# setup for gmm
matrix Z = { zlist }
matrix W = inv(ZZ)
series e = 0
scalar b0 = 1
scalar b1 = 1
scalar b2 = 1
gmm e = lpackpc - b0 - b1*lravgprs - b2*lperinc
orthog e ; Z
weights W
params b0 b1 b2
end gmm
Example 22.3: TSLS via GMM: partial output
Model 1: TSLS estimates using the 48 observations 1-48
Dependent variable: lpackpc
Instruments: rtaxso rtax
Heteroskedasticity-robust standard errors, variant HC0
VARIABLE COEFFICIENT STDERROR T STAT P-VALUE
const 9.89496 0.928758 10.654 <0.00001 ***
lravgprs -1.27742 0.241684 -5.286 <0.00001 ***
lperinc 0.280405 0.245828 1.141 0.25401
Model 2: 1-step GMM estimates using the 48 observations 1-48
e = lpackpc - b0 - b1*lravgprs - b2*lperinc
PARAMETER ESTIMATE STDERROR T STAT P-VALUE
b0 9.89496 0.928758 10.654 <0.00001 ***
b1 -1.27742 0.241684 -5.286 <0.00001 ***
b2 0.280405 0.245828 1.141 0.25401
GMM criterion = 0.0110046
This estimator is robust with respect to heteroskedasticity, but not with respect to autocorrela-
tion. A heteroskedasticity- and autocorrelation-consistent (HAC) variant can be obtained using
the Bartlett kernel or similar. A univariate version of this is used in the context of the lrvar()
functionsee equation (9.1). The multivariate version is set out in equation (22.10).
k
() =
1
T
Tk
_
t=k
_
_
k
_
i=k
w
i
f
t
()f
ti
()
_
_
, (22.10)
Gretl computes the HAC covariance matrix by default when a GMM model is estimated on time
series data. You can control the kernel and the bandwidth (that is, the value of k in 22.10) using
the set command. See chapter 17 for further discussion of HAC estimation. You can also ask gretl
not to use the HAC version by saying
set force_hc on
22.5 A real example: the Consumption Based Asset Pricing Model
To illustrate gretls implementation of GMM, we will replicate the example given in chapter 3 of
Hall (2005). The model to estimate is a classic application of GMM, and provides an example of a
case when orthogonality conditions do not stem from statistical considerations, but rather from
economic theory.
A rational individual who must allocate his income between consumption and investment in a
nancial asset must in fact choose the consumption path of his whole lifetime, since investment
translates into future consumption. It can be shown that an optimal consumption path should
satisfy the following condition:
pU
(c
t
) =
k
E
_
r
t+k
U
(c
t+k
)
t
_
, (22.11)
where p is the asset price, U() is the individuals utility function, is the individuals subjective
discount rate and r
t+k
is the assets rate of return between time t and time t + k.
t
is the infor-
mation set at time t; equation (22.11) says that the utility lost at time t by purchasing the asset
instead of consumption goods must be matched by a corresponding increase in the (discounted)
future utility of the consumption nanced by the assets return. Since the future is uncertain, the
individual considers his expectation, conditional on what is known at the time when the choice is
made.
We have said nothing about the nature of the asset, so equation (22.11) should hold whatever asset
we consider; hence, it is possible to build a system of equations like (22.11) for each asset whose
price we observe.
If we are willing to believe that
the economy as a whole can be represented as a single gigantic and immortal representative
individual, and
the function U(x) =
x
is a faithful representation of the individuals preferences,

then, setting k = 1, equation (22.11) implies the following for any asset j:
E
_
r
j,t+1
p
j,t
_
C
t+1
C
t
_
1
t
_
= 1, (22.12)
where C
t
is aggregate consumption and and are the risk aversion and discount rate of the
representative individual. In this case, it is easy to see that the deep parameters and can be
estimated via GMM by using
e
t
=
r
j,t+1
p
j,t
_
C
t+1
C
t
_
1
1
as the moment condition, while any variable known at time t may serve as an instrument.
In the example code given in 22.4, we replicate selected portions of table 3.7 in Hall (2005). The
variable consrat is dened as the ratio of monthly consecutive real per capita consumption (ser-
vices and nondurables) for the US, and ewr is the returnprice ratio of a ctitious asset constructed
by averaging all the stocks in the NYSE. The instrument set contains the constant and two lags of
each variable.
The command set force_hc on on the second line of the script has the sole purpose of replicating
the given example: as mentioned above, it forces gretl to compute the long-run variance of the
orthogonality conditions according to equation (22.9) rather than (22.10).
We run gmm four times: one-step estimation for each of two initial weights matrices, then iterative
estimation starting from each set of initial weights. Since the number of orthogonality conditions
(5) is greater than the number of estimated parameters (2), the choice of initial weights should
make a dierence, and indeed we see fairly substantial dierences between the one-step estimates
(Models 1 and 2). On the other hand, iteration reduces these dierences almost to the vanishing
point (Models 3 and 4).
Part of the output is given in 22.5. It should be noted that the J test leads to a rejection of the
hypothesis of correct specication. This is perhaps not surprising given the heroic assumptions
required to move from the microeconomic principle in equation (22.11) to the aggregate system
that is actually estimated.
22.6 Caveats
A few words of warning are in order: despite its ingenuity, GMM is possibly the most fragile esti-
mation method in econometrics. The number of non-obvious choices one has to make when using
GMM is high, and in nite samples each of these can have dramatic consequences on the eventual
output. Some of the factors that may aect the results are:
1. Orthogonality conditions can be written in more than one way: for example, if E(x
t
) = 0,
then E(x
t
/ 1) = 0 holds too. It is possible that a dierent specication of the moment
conditions leads to dierent results.
Example 22.4: Estimation of the Consumption Based Asset Pricing Model
open hall.gdt
set force_hc on
scalar alpha = 0.5
scalar delta = 0.5
series e = 0
list inst = const consrat(-1) consrat(-2) ewr(-1) ewr(-2)
matrix V0 = 100000*I(nelem(inst))
matrix Z = { inst }
matrix V1 = $nobs*inv(ZZ)
gmm e = delta*ewr*consrat^(alpha-1) - 1
orthog e ; inst
weights V0
params alpha delta
end gmm
orthog e ; inst
weights V1
params alpha delta
end gmm
orthog e ; inst
weights V0
params alpha delta
end gmm --iterate
orthog e ; inst
weights V1
params alpha delta
end gmm --iterate
Example 22.5: Estimation of the Consumption Based Asset Pricing Model output
Model 1: 1-step GMM estimates using the 465 observations 1959:04-1997:12
e = d*ewr*consrat^(alpha-1) - 1
alpha -3.14475 6.84439 -0.459 0.64590
d 0.999215 0.0121044 82.549 <0.00001 ***
Model 2: 1-step GMM estimates using the 465 observations 1959:04-1997:12
alpha 0.398194 2.26359 0.176 0.86036
d 0.993180 0.00439367 226.048 <0.00001 ***
Model 3: Iterated GMM estimates using the 465 observations 1959:04-1997:12
alpha -0.344325 2.21458 -0.155 0.87644
d 0.991566 0.00423620 234.070 <0.00001 ***
J test: Chi-square(3) = 11.8103 (p-value 0.0081)
Model 4: Iterated GMM estimates using the 465 observations 1959:04-1997:12
alpha -0.344315 2.21359 -0.156 0.87639
d 0.991566 0.00423469 234.153 <0.00001 ***
J test: Chi-square(3) = 11.8103 (p-value 0.0081)
2. As with all other numerical optimization algorithms, weird things may happen when the ob-
jective function is nearly at in some directions or has multiple minima. BFGS is usually quite
good, but there is no guarantee that it always delivers a sensible solution, if one at all.
3. The 1-step and, to a lesser extent, the 2-step estimators may be sensitive to apparently trivial
details, like the re-scaling of the instruments. Dierent choices for the initial weights matrix
can also have noticeable consequences.
4. With time-series data, there is no hard rule on the appropriate number of lags to use when
computing the long-run covariance matrix (see section 22.4). Our advice is to go by trial and
error, since results may be greatly inuenced by a poor choice. Future versions of gretl will
include more options on covariance matrix estimation.
One of the consequences of this state of things is that replicating various well-known published
studies may be extremely dicult. Any non-trivial result is virtually impossible to reproduce unless
all details of the estimation procedure are carefully recorded.
Chapter 23
Model selection criteria
23.1 Introduction
In some contexts the econometrician chooses between alternative models based on a formal hy-
pothesis test. For example, one might choose a more general model over a more restricted one if
the restriction in question can be formulated as a testable null hypothesis, and the null is rejected
on an appropriate test.
In other contexts one sometimes seeks a criterion for model selection that somehow measures the
balance between goodness of t or likelihood, on the one hand, and parsimony on the other. The
balancing is necessary because the addition of extra variables to a model cannot reduce the degree
of t or likelihood, and is very likely to increase it somewhat even if the additional variables are
not truly relevant to the data-generating process.
The best known such criterion, for linear models estimated via least squares, is the adjusted R
2
,
R
2
= 1
SSR/(nk)
TSS/(n1)
where n is the number of observations in the sample, k denotes the number of parameters esti-
mated, and SSR and TSS denote the sum of squared residuals and the total sum of squares for
the dependent variable, respectively. Compared to the ordinary coecient of determination or
unadjusted R
2
,
R
2
= 1
SSR
TSS
the adjusted calculation penalizes the inclusion of additional parameters, other things equal.
23.2 Information criteria
A more general criterion in a similar spirit is Akaikes (1974) Information Criterion (AIC). The
original formulation of this measure is
AIC = 2(
) +2k (23.1)
where (
) represents the maximum loglikelihood as a function of the vector of parameter esti-

mates,

, and k (as above) denotes the number of independently adjusted parameters within the
model. In this formulation, with AIC negatively related to the likelihood and positively related to
the number of parameters, the researcher seeks the minimum AIC.
The AIC can be confusing, in that several variants of the calculation are in circulation. For exam-
ple, Davidson and MacKinnon (2004) present a simplied version,
AIC = (
) k
which is just 2 times the original: in this case, obviously, one wants to maximize AIC.
In the case of models estimated by least squares, the loglikelihood can be written as
(
) =
n
2
(1 +log2 logn)
n
2
logSSR (23.2)
202
Chapter 23. Model selection criteria 203
Substituting (23.2) into (23.1) we get
AIC = n(1 +log2 logn) +nlogSSR +2k
which can also be written as
AIC = nlog
_
SSR
n
_
+2k +n(1 +log2) (23.3)
Some authors simplify the formula for the case of models estimated via least squares. For instance,
William Greene writes
AIC = log
_
SSR
n
_
+
2k
n
(23.4)
This variant can be derived from (23.3) by dividing through by n and subtracting the constant
1 +log2. That is, writing AIC
G
for the version given by Greene, we have
AIC
G
=
1
n
AIC (1 +log2)
Finally, Ramanathan gives a further variant:
AIC
R
=
_
SSR
n
_
e
2k/n
which is the exponential of the one given by Greene.
Gretl began by using the Ramanathan variant, but since version 1.3.1 the program has used the
original Akaike formula (23.1), and more specically (23.3) for models estimated via least squares.
Although the Akaike criterion is designed to favor parsimony, arguably it does not go far enough
in that direction. For instance, if we have two nested models with k 1 and k parameters respec-
tively, and if the null hypothesis that parameter k equals 0 is true, in large samples the AIC will
nonetheless tend to select the less parsimonious model about 16 percent of the time (see Davidson
and MacKinnon, 2004, chapter 15).
An alternative to the AIC which avoids this problem is the Schwarz (1978) Bayesian information
criterion (BIC). The BIC can be written (in line with Akaikes formulation of the AIC) as
BIC = 2(
) +klogn
The multiplication of k by logn in the BIC means that the penalty for adding extra parameters
grows with the sample size. This ensures that, asymptotically, one will not select a larger model
over a correctly specied parsimonious model.
A further alternative to AIC, which again tends to select more parsimonious models than AIC,
is the HannanQuinn criterion or HQC (Hannan and Quinn, 1979). Written consistently with the
formulations above, this is
HQC = 2(
) +2kloglogn
The HannanQuinn calculation is based on the law of the iterated logarithm (note that the last term
is the log of the log of the sample size). The authors argue that their procedure provides a strongly
consistent estimation procedure for the order of an autoregression, and that compared to other
strongly consistent procedures this procedure will underestimate the order to a lesser degree.
Gretl reports the AIC, BIC and HQC (calculated as explained above) for most sorts of models. The
key point in interpreting these values is to know whether they are calculated such that smaller
values are better, or such that larger values are better. In gretl, smaller values are better: one wants
to minimize the chosen criterion.
Chapter 24
Time series lters
In addition to the usual application of lags and dierences, gretl provides fractional dierencing
and various lters commonly used in macroeconomics for trend-cycle decomposition: notably the
HodrickPrescott lter (Hodrick and Prescott, 1997), the BaxterKing bandpass lter (Baxter and
King, 1999) and the Butterworth lter (Butterworth, 1930).
24.1 Fractional dierencing
The concept of dierencing a time series d times is pretty obvious when d is an integer; it may seem
odd when d is fractional. However, this idea has a well-dened mathematical content: consider the
function
f(z) = (1 z)
d
,
where z and d are real numbers. By taking a Taylor series expansion around z = 0, we see that
f(z) = 1 +dz +
d(d +1)
2
z
2
+
or, more compactly,
f(z) = 1 +
_
i=1
i
z
i
with
k
=
k
i=1
(d +i 1)
k!
=
k1
d +k 1
k
The same expansion can be used with the lag operator, so that if we dened
Y
t
= (1 L)
0.5
X
t
this could be considered shorthand for
Y
t
= X
t
0.5X
t1
0.125X
t2
0.0625X
t3

In gretl this transformation can be accomplished by the syntax
genr Y = fracdiff(X,0.5)
24.2 The HodrickPrescott lter
This lter is accessed using the hpfilt() function, which takes as its rst argument the name of
the variable to be processed. (A further optional argument is explained below.)
A time series y
t
may be decomposed into a trend or growth component g
t
and a cyclical component
c
t
.
y
t
= g
t
+c
t
, t = 1, 2, . . . , T
204
Chapter 24. Time series lters 205
The HodrickPrescott lter eects such a decomposition by minimizing the following:
T
_
t=1
(y
t
g
t
)
2
+
T1
_
t=2
_
(g
t+1
g
t
) (g
t
g
t1
)
_
2
.
The rst term above is the sum of squared cyclical components c
t
= y
t
g
t
. The second term is a
multiple of the sum of squares of the trend components second dierences. This second term
penalizes variations in the growth rate of the trend component: the larger the value of , the higher
is the penalty and hence the smoother the trend series.
Note that the hpfilt function in gretl produces the cyclical component, c
t
, of the original series.
If you want the smoothed trend you can subtract the cycle from the original:
genr ct = hpfilt(yt)
genr gt = yt - ct
Hodrick and Prescott (1997) suggest that a value of = 1600 is reasonable for quarterly data. The
default value in gretl is 100 times the square of the data frequency (which, of course, yields 1600
for quarterly data). The value can be adjusted using an optional second argument to hpfilt(), as
in
genr ct = hpfilt(yt, 1300)
24.3 The Baxter and King lter
This lter is accessed using the bkfilt() function, which again takes the name of the variable to
be processed as its rst argument. The operation of the lter can be controlled via three further
optional argument.
Consider the spectral representation of a time series y
t
:
y
t
=
_
e
i
dZ()
To extract the component of y
t
that lies between the frequencies and one could apply a
bandpass lter:
c
t
=
_
()e
i
dZ()
where F
() = 1 for < < and 0 elsewhere. This would imply, in the time domain,
applying to the series a lter with an innite number of coecients, which is undesirable. The
Baxter and King bandpass lter applies to y
t
a nite polynomial in the lag operator A(L):
c
t
= A(L)y
t
where A(L) is dened as
A(L) =
k
_
i=k
a
i
L
i
The coecients a
i
are chosen such that F() = A(e
i
)A(e
i
) is the best approximation to F
()
for a given k. Clearly, the higher k the better the approximation is, but since 2k observations have
to be discarded, a compromise is usually sought. Moreover, the lter has also other appealing
theoretical properties, among which the property that A(1) = 0, so a series with a single unit root
is made stationary by application of the lter.
In practice, the lter is normally used with monthly or quarterly data to extract the business
cycle component, namely the component between 6 and 36 quarters. Usual choices for k are 8 or
12 (maybe higher for monthly series). The default values for the frequency bounds are 8 and 32,
and the default value for the approximation order, k, is 8. You can adjust these values using the
full form of bkfilt(), which is
bkfilt(seriesname, f1, f2, k)
where f1 and f2 represent the lower and upper frequency bounds respectively.
24.4 The Butterworth lter
The Butterworth lter (Butterworth, 1930) is an approximation to an ideal square-wave lter.
The ideal lter divides the spectrum of a time series into a pass-band (frequencies less than some
chosen
for a low-pass lter, or frequencies greater than
for high-pass) and a stop-band; the

gain is 1 for the pass-band and 0 for the stop-band. The ideal lter is unattainable in practice since
it would require an innite number of coecients, but the Butterworth lter oers a remarkably
good approximation. This lter is derived and persuasively advocated by Pollock (2000).
For data y, the ltered sequence x is given by
x = y Q(M +Q
Q)
1
Q
y (24.1)
where
= 2I
T
(L
T
+L
1
T
)
T2
and M = 2I
T
+(L
T
+L
1
T
)
T
I
T
denotes the identity matrix of order T; L
T
= [e
1
, e
2
, . . . , e
T1
, 0] is the nite-sample matrix
version of the lag operator; and Q is dened such that pre-multiplication of a T-vector of data by
Q
of order (T 2) T produces the second dierences of the data. The matrix product
Q
Q = 2I
T
(L
T
+L
1
T
)
T
is a Toeplitz matrix.
The behavior of the Butterworth lter is governed by two parameters: the frequency cuto
and
an integer order, n, which determines the number of coecients used. The that appears in (24.1)
is tan(
/2)
2n
. Higher values of n produce a better approximation to the ideal lter in principle
(i.e. a sharper cut between the pass-band and the stop-band) but there is a downside: with a greater
number of coecients numerical instability may be an issue, and the inuence of the initial values
in the sample may be exaggerated.
In gretl the Butterworth lter is implemented by the bwfilt() function,
1
which takes three argu-
ments: the series to lter, the order n and the frequency cuto,
, expressed in degrees. The

cuto value must be greater than 0 and less than 180. This function operates as a low-pass lter;
for the high-pass variant, subtract the ltered series from the original, as in
series bwcycle = y - bwfilt(y, 8, 67)
Pollock recommends that the parameters of the Butterworth lter be tuned to the data: one should
examine the periodogram of the series in question (possibly after removal of a polynomial trend)
in search of a dead spot of low power between the frequencies one wishes to exclude and the
frequencies one wishes to retain. If
is placed in such a dead spot then the job of separation

can be done with a relatively small n, hence avoiding numerical problems. By way of illustration,
consider the periodogram for quarterly observations on new cars sales in the US,
2
1975:1 to 1990:4
(the upper panel in Figure 24.1).
A seasonal pattern is clearly visible in the periodogram, centered at an angle of 90
or 4 periods.
If we set
= 68
(or thereabouts) we should be able to excise the seasonality quite cleanly using
1
The code for this lter is based on D. S. G. Pollocks programs IDEOLOG and DETREND. The Pascal source code for
the former is available from http://www.le.ac.uk/users/dsgp1 and the C sources for the latter were kindly made
available to us by the author.
2
This is the variable QNC from the Ramanathan data le data9-7.
0
50000
100000
150000
200000
250000
300000
0 20 40 60 80 100 120 140 160 180
64.0 10.7 5.8 4.0 3.0 2.5 2.1
degrees
periods
1600
1800
2000
2200
2400
2600
2800
3000
3200
3400
1976 1978 1980 1982 1984 1986 1988 1990
QNC (original data)
QNC (smoothed)
0
0.2
0.4
0.6
0.8
1
0 /4 /2 3/4
Figure 24.1: The Butterworth lter applied
n = 8. The result is shown in the lower panel of the Figure, along with the frequency response or
gain plot for the chosen lter. Note the smooth and reasonably steep drop-o in gain centered on
the nominal cuto of 68
3/8.
The apparatus that supports this sort of analysis in the gretl GUI can be found under the Variable
menu in the main window: the items Periodogram and Filter. In the periodogram dialog box you
have the option of expressing the frequency axis in degrees, which is helpful when selecting a
Butterworth lter; and in the Butterworth lter dialog you have the option of plotting the frequency
response as well as the smoothed series and/or the residual or cycle.
Chapter 25
Univariate time series models
25.1 Introduction
Time series models are discussed in this chapter and the next two. Here we concentrate on ARIMA
models, unit root tests, and GARCH. The following chapter deals with VARs, and chapter 27 with
cointegration and error correction.
25.2 ARIMA models
Representation and syntax
The arma command performs estimation of AutoRegressive, Integrated, Moving Average (ARIMA)
models. These are models that can be written in the form
(L)y
t
= (L)
t
(25.1)
where (L), and (L) are polynomials in the lag operator, L, dened such that L
n
x
t
= x
tn
, and
t
is a white noise process. The exact content of y
t
, of the AR polynomial (), and of the MA
polynomial (), will be explained in the following.
Mean terms
The process y
t
as written in equation (25.1) has, without further qualications, mean zero. If the
model is to be applied to real data, it is necessary to include some term to handle the possibility
that y
t
has non-zero mean. There are two possible ways to represent processes with nonzero
mean: one is to dene
t
as the unconditional mean of y
t
, namely the central value of its marginal
distribution. Therefore, the series y
t
= y
t

t
has mean 0, and the model (25.1) applies to y
t
. In
practice, assuming that
t
is a linear function of some observable variables x
t
, the model becomes
(L)(y
t
x
t
) = (L)
t
(25.2)
This is sometimes known as a regression model with ARMA errors; its structure may be more
apparent if we represent it using two equations:
y
t
= x
t
+u
t
(L)u
t
= (L)
t
The model just presented is also sometimes known as ARMAX (ARMA + eXogenous variables). It
seems to us, however, that this label is more appropriately applied to a dierent model: another
way to include a mean term in (25.1) is to base the representation on the conditional mean of y
t
,
that is the central value of the distribution of y
t
given its own past. Assuming, again, that this can
be represented as a linear combination of some observable variables z
t
, the model would expand
to
(L)y
t
= z
t
+(L)
t
(25.3)
The formulation (25.3) has the advantage that can be immediately interpreted as the vector of
marginal eects of the z
t
variables on the conditional mean of y
t
. And by adding lags of z
t
to
208
Chapter 25. Univariate time series models 209
this specication one can estimate Transfer Function models (which generalize ARMA by adding
the eects of exogenous variable distributed across time).
Gretl provides a way to estimate both forms. Models written as in (25.2) are estimated by maximum
likelihood; models written as in (25.3) are estimated by conditional maximum likelihood. (For more
on these options see the section on Estimation below.)
In the special case when x
t
= z
t
= 1 (that is, the models include a constant but no exogenous
variables) the two specications discussed above reduce to
(L)(y
t
) = (L)
t
(25.4)
and
(L)y
t
= +(L)
t
(25.5)
respectively. These formulations are essentially equivalent, but if they represent one and the same
process and are, fairly obviously, not numerically identical; rather
=
_
1
1
. . .
p
_
The gretl syntax for estimating (25.4) is simply

arma p q ; y
The AR and MA lag orders, p and q, can be given either as numbers or as pre-dened scalars.
The parameter can be dropped if necessary by appending the option --nc (no constant) to the
command. If estimation of (25.5) is needed, the switch --conditional must be appended to the
command, as in
arma p q ; y --conditional
Generalizing this principle to the estimation of (25.2) or (25.3), you get that
arma p q ; y const x1 x2
would estimate the following model:
y
t
x
t
=
1
_
y
t1
x
t1
_
+. . . +
p
_
y
tp
x
tp
_
+
t
+
1
t1
+. . . +
q
tq
where in this instance x
t
=
0
+x
t,1
1
+x
t,2
2
. Appending the --conditional switch, as in
arma p q ; y const x1 x2 --conditional
would estimate the following model:
y
t
= x
t
+
1
y
t1
+. . . +
p
y
tp
+
t
+
1
t1
+. . . +
q
tq
Ideally, the issue broached above could be made moot by writing a more general specication that
nests the alternatives; that is
(L)
_
y
t
x
t
_
= z
t
+(L)
t
; (25.6)
we would like to generalize the arma command so that the user could specify, for any estimation
method, whether certain exogenous variables should be treated as x
t
s or z
t
s, but were not yet at
that point (and neither are most other software packages).
Seasonal models
A more exible lag structure is desirable when analyzing time series that display strong seasonal
patterns. Model (25.1) can be expanded to
(L)(L
s
)y
t
= (L)(L
s
)
t
. (25.7)
For such cases, a fuller form of the syntax is available, namely,
arma p q ; P Q ; y
where p and q represent the non-seasonal AR and MA orders, and P and Q the seasonal orders. For
example,
arma 1 1 ; 1 1 ; y
would be used to estimate the following model:
(1 L)(1 L
s
)(y
t
) = (1 +L)(1 +L
s
)
t
If y
t
is a quarterly series (and therefore s = 4), the above equation can be written more explicitly as
y
t
= (y
t1
) +(y
t4
) ( )(y
t5
) +
t
+
t1
+
t4
+( )
t5
Such a model is known as a multiplicative seasonal ARMA model.
Gaps in the lag structure
The standard way to specify an ARMA model in gretl is via the AR and MA orders, p and q respec-
tively. In this case all lags from 1 to the given order are included. In some cases one may wish to
include only certain specic AR and/or MA lags. This can be done in either of two ways.
One can construct a matrix containing the desired lags (positive integer values) and supply
the name of this matrix in place of p or q.
One can give a comma-separated list of lags, enclosed in braces, in place of p or q.
The following code illustrates these options:
matrix pvec = {1,4}
arma pvec 1 ; y
arma {1,4} 1 ; y
Both forms above specify an ARMA model in which AR lags 1 and 4 are used (but not 2 and 3).
This facility is available only for the non-seasonal component of the ARMA specication.
Dierencing and ARIMA
The above discussion presupposes that the time series y
t
has already been subjected to all the
transformations deemed necessary for ensuring stationarity (see also section 25.3). Dierencing is
the most common of these transformations, and gretl provides a mechanism to include this step
into the arma command: the syntax
arma p d q ; y
would estimate an ARMA(p, q) model on
d
y
t
. It is functionally equivalent to
series tmp = y
loop i=1..d
tmp = diff(tmp)
endloop
arma p q ; tmp
except with regard to forecasting after estimation (see below).
When the series y
t
is dierenced before performing the analysis the model is known as ARIMA (I
for Integrated); for this reason, gretl provides the arima command as an alias for arma.
Seasonal dierencing is handled similarly, with the syntax
arma p d q ; P D Q ; y
where D is the order for seasonal dierencing. Thus, the command
arma 1 0 0 ; 1 1 1 ; y
would produce the same parameter estimates as
genr dsy = sdiff(y)
arma 1 0 ; 1 1 ; dsy
where we use the sdiff function to create a seasonal dierence (e.g. for quarterly data, y
t
y
t4
).
In specifying an ARIMA model with exogenous regressors we face a choice which relates back to the
discussion of the variant models (25.2) and (25.3) above. If we choose model (25.2), the regression
model with ARMA errors, how should this be extended to the case of ARIMA? The issue is whether
or not the dierencing that is applied to the dependent variable should also be applied to the
regressors. Consider the simplest case, ARIMA with non-seasonal dierencing of order 1. We may
estimate either
(L)(1 L)(y
t
X
t
) = (L)
t
(25.8)
or
(L)
_
(1 L)y
t
X
t
_
= (L)
t
(25.9)
The rst of these formulations can be described as a regression model with ARIMA errors, while the
second preserves the levels of the X variables. As of gretl version 1.8.6, the default model is (25.8),
in which dierencing is applied to both y
t
and X
t
. However, when using the default estimation
method (native exact ML, see below), the option --y-diff-only may be given, in which case gretl
estimates (25.9).
1
Estimation
The default estimation method for ARMA models is exact maximum likelihood estimation (under
the assumption that the error term is normally distributed), using the Kalman lter in conjunc-
tion with the BFGS maximization algorithm. The gradient of the log-likelihood with respect to the
parameter estimates is approximated numerically. This method produces results that are directly
comparable with many other software packages. The constant, and any exogenous variables, are
treated as in equation (25.2). The covariance matrix for the parameters is computed using a nu-
merical approximation to the Hessian at convergence.
The alternative method, invoked with the --conditional switch, is conditional maximum likeli-
hood (CML), also known as conditional sum of squares (see Hamilton, 1994, p. 132). This method
was exemplied in the script 12.3, and only a brief description will be given here. Given a sample of
size T, the CML method minimizes the sum of squared one-step-ahead prediction errors generated
1
Prior to gretl 1.8.6, the default model was (25.9). We changed this for the sake of consistency with other software.
by the model for the observations t
0
, . . . , T. The starting point t
0
depends on the orders of the AR
polynomials in the model. The numerical maximization method used is BHHH, and the covariance
matrix is computed using a GaussNewton regression.
The CML method is nearly equivalent to maximum likelihood under the hypothesis of normality;
the dierence is that the rst (t
0
1) observations are considered xed and only enter the like-
lihood function as conditioning variables. As a consequence, the two methods are asymptotically
equivalent under standard conditionsexcept for the fact, discussed above, that our CML imple-
mentation treats the constant and exogenous variables as per equation (25.3).
The two methods can be compared as in the following example
open data10-1
arma 1 1 ; r
arma 1 1 ; r --conditional
which produces the estimates shown in Table 25.1. As you can see, the estimates of and
are quite similar. The reported constants dier widely, as expectedsee the discussion following
equations (25.4) and (25.5). However, dividing the CML constant by 1 we get 7.38, which is not
far from the ML estimate of 6.93.
Table 25.1: ML and CML estimates
Parameter ML CML
6.93042 (0.923882) 1.07322 (0.488661)
0.855360 (0.0511842) 0.852772 (0.0450252)
0.588056 (0.0986096) 0.591838 (0.0456662)
Convergence and initialization
The numerical methods used to maximize the likelihood for ARMA models are not guaranteed
to converge. Whether or not convergence is achieved, and whether or not the true maximum of
the likelihood function is attained, may depend on the starting values for the parameters. Gretl
employs one of the following two initialization mechanisms, depending on the specication of the
model and the estimation method chosen.
1. Estimate a pure AR model by Least Squares (nonlinear least squares if the model requires
it, otherwise OLS). Set the AR parameter values based on this regression and set the MA
parameters to a small positive value (0.0001).
2. The HannanRissanen method: First estimate an autoregressive model by OLS and save the
residuals. Then in a second OLS pass add appropriate lags of the rst-round residuals to the
model, to obtain estimates of the MA parameters.
To see the details of the ARMA estimation procedure, add the --verbose option to the command.
This prints a notice of the initialization method used, as well as the parameter values and log-
likelihood at each iteration.
Besides the built-in initialization mechanisms, the user has the option of specifying a set of starting
values manually. This is done via the set command: the rst argument should be the keyword
initvals and the second should be the name of a pre-specied matrix containing starting values.
For example
matrix start = { 0, 0.85, 0.34 }
set initvals start
arma 1 1 ; y
The specied matrix should have just as many parameters as the model: in the example above
there are three parameters, since the model implicitly includes a constant. The constant, if present,
is always given rst; otherwise the order in which the parameters are expected is the same as the
order of specication in the arma or arima command. In the example the constant is set to zero,
1
to 0.85, and
1
to 0.34.
You can get gretl to revert to automatic initialization via the command set initvals auto.
Two variants of the BFGS algorithm are available in gretl. In general we recommend the default vari-
ant, which is based on an implementation by Nash (1990), but for some problems the alternative,
limited-memory version (L-BFGS-B, see Byrd et al., 1995) may increase the chances of convergence
on the ML solution. This can be selected via the --lbfgs option to the arma command.
Estimation via X-12-ARIMA
As an alternative to estimating ARMA models using native code, gretl oers the option of using
the external program X-12-ARIMA. This is the seasonal adjustment software produced and main-
tained by the U.S. Census Bureau; it is used for all ocial seasonal adjustments at the Bureau.
Gretl includes a module which interfaces with X-12-ARIMA: it translates arma commands using the
syntax outlined above into a form recognized by X-12-ARIMA, executes the program, and retrieves
the results for viewing and further analysis within gretl. To use this facility you have to install
X-12-ARIMA separately. Packages for both MS Windows and GNU/Linux are available from the gretl
website, http://gretl.sourceforge.net/.
To invoke X-12-ARIMA as the estimation engine, append the ag --x-12-arima, as in
arma p q ; y --x-12-arima
As with native estimation, the default is to use exact ML but there is the option of using conditional
ML with the --conditional ag. However, please note that when X-12-ARIMA is used in conditional
ML mode, the comments above regarding the variant treatments of the mean of the process y
t
do
not apply. That is, when you use X-12-ARIMA the model that is estimated is (25.2), regardless
of whether estimation is by exact ML or conditional ML. In addition, the treatment of exogenous
regressors in the context of ARIMA dierencing is always that shown in equation (25.8).
Forecasting
ARMA models are often used for forecasting purposes. The autoregressive component, in particu-
lar, oers the possibility of forecasting a process out of sample over a substantial time horizon.
Gretl supports forecasting on the basis of ARMA models using the method set out by Box and
Jenkins (1976).
2
The Box and Jenkins algorithm produces a set of integrated AR coecients which
take into account any dierencing of the dependent variable (seasonal and/or non-seasonal) in the
ARIMA context, thus making it possible to generate a forecast for the level of the original variable.
By contrast, if you rst dierence a series manually and then apply ARMA to the dierenced series,
forecasts will be for the dierenced series, not the level. This point is illustrated in Example 25.1.
The parameter estimates are identical for the two models. The forecasts dier but are mutually
consistent: the variable fcdiff emulates the ARMA forecast (static, one step ahead within the
sample range, and dynamic out of sample).
2
See in particular their Program 4 on p. 505.
Example 25.1: ARIMA forecasting
open greene18_2.gdt
# log of quarterly U.S. nominal GNP, 1950:1 to 1983:4
genr y = log(Y)
# and its first difference
genr dy = diff(y)
# reserve 2 years for out-of-sample forecast
smpl ; 1981:4
# Estimate using ARIMA
arima 1 1 1 ; y
# forecast over full period
smpl --full
fcast fc1
# Return to sub-sample and run ARMA on the first difference of y
smpl ; 1981:4
arma 1 1 ; dy
smpl --full
fcast fc2
genr fcdiff = (t<=1982:1)? (fc1 - y(-1)) : (fc1 - fc1(-1))
# compare the forecasts over the later period
smpl 1981:1 1983:4
print y fc1 fc2 fcdiff --byobs
The output from the last command is:
y fc1 fc2 fcdiff
1981:1 7.964086 7.940930 0.02668 0.02668
1981:2 7.978654 7.997576 0.03349 0.03349
1981:3 8.009463 7.997503 0.01885 0.01885
1981:4 8.015625 8.033695 0.02423 0.02423
1982:1 8.014997 8.029698 0.01407 0.01407
1982:2 8.026562 8.046037 0.01634 0.01634
1982:3 8.032717 8.063636 0.01760 0.01760
1982:4 8.042249 8.081935 0.01830 0.01830
1983:1 8.062685 8.100623 0.01869 0.01869
1983:2 8.091627 8.119528 0.01891 0.01891
1983:3 8.115700 8.138554 0.01903 0.01903
1983:4 8.140811 8.157646 0.01909 0.01909
25.3 Unit root tests
The ADF test
The Augmented DickeyFuller (ADF) test is, as implemented in gretl, the t-statistic on in the
following regression:
y
t
=
t
+y
t1
+
p
_
i=1
i
y
ti
+
t
. (25.10)
This test statistic is probably the best-known and most widely used unit root test. It is a one-sided
test whose null hypothesis is = 0 versus the alternative < 0 (and hence large negative values
of the test statistic lead to the rejection of the null). Under the null, y
t
must be dierenced at least
once to achieve stationarity; under the alternative, y
t
is already stationary and no dierencing is
required.
One peculiar aspect of this test is that its limit distribution is non-standard under the null hy-
pothesis: moreover, the shape of the distribution, and consequently the critical values for the test,
depends on the form of the
t
term. A full analysis of the various cases is inappropriate here:
Hamilton (1994) contains an excellent discussion, but any recent time series textbook covers this
topic. Suce it to say that gretl allows the user to choose the specication for
t
among four
dierent alternatives:
t
command option
0 --nc
0
--c
0
+
1
t --ct
0
+
1
t +
1
t
2
--ctt
These option ags are not mutually exclusive; when they are used together the statistic will be
reported separately for each selected case. By default, gretl uses the combination --c --ct. For
each case, approximate p-values are calculated by means of the algorithm developed in MacKinnon
(1996).
The gretl command used to perform the test is adf; for example
adf 4 x1
would compute the test statistic as the t-statistic for in equation 25.10 with p = 4 in the two
cases
t
=
0
and
t
=
0
+
1
t.
The number of lags (p in equation 25.10) should be chosen as to ensure that (25.10) is a para-
metrization exible enough to represent adequately the short-run persistence of y
t
. Setting p
too low results in size distortions in the test, whereas setting p too high leads to low power. As
a convenience to the user, the parameter p can be automatically determined. Setting p to a neg-
ative number triggers a sequential procedure that starts with p lags and decrements p until the
t-statistic for the parameter
p
exceeds 1.645 in absolute value.
The ADF-GLS test
Elliott, Rothenberg and Stock (1996) proposed a variant of the ADF test which involves an alterna-
tive method of handling the parameters pertaining to the deterministic term
t
: these are estimated
rst via Generalized Least Squares, and in a second stage an ADF regression is performed using the
GLS residuals. This variant oers greater power than the regular ADF test for the cases
t
=
0
and
t
=
0
+
1
t.
The ADF-GLS test is available in gretl via the --gls option to the adf command. When this option
is selected the --nc and --ctt options become unavailable, and only one case can be selected at
a time; by default the constant-only model is used but a trend can be added using the --ct ag.
When a trend is present in this test MacKinnon-type p-values are not available; instead we show
critical values from Table 1 in Elliott et al. (1996).
The KPSS test
The KPSS test (Kwiatkowski, Phillips, Schmidt and Shin, 1992) is a unit root test in which the null
hypothesis is opposite to that in the ADF test: under the null, the series in question is stationary;
the alternative is that the series is I(1).
The basic intuition behind this test statistic is very simple: if y
t
can be written as y
t
= + u
t
,
where u
t
is some zero-mean stationary process, then not only does the sample average of the y
t
s
provide a consistent estimator of , but the long-run variance of u
t
is a well-dened, nite number.
Neither of these properties hold under the alternative.
The test itself is based on the following statistic:
=
T
i=1
S
2
t
T
2

2
(25.11)
where S
t
=
t
s=1
e
s
and
2
is an estimate of the long-run variance of e
t
= (y
t
y). Under the null,
this statistic has a well-dened (nonstandard) asymptotic distribution, which is free of nuisance
parameters and has been tabulated by simulation. Under the alternative, the statistic diverges.
As a consequence, it is possible to construct a one-sided test based on , where H
0
is rejected if
is bigger than the appropriate critical value; gretl provides the 90, 95 and 99 percent quantiles.
The critical values are computed via the method presented by Sephton (1995), which oers greater
accuracy than the values tabulated in Kwiatkowski et al. (1992).
Usage example:
kpss m y
where m is an integer representing the bandwidth or window size used in the formula for estimating
the long run variance:

2
=
m
_
i=m
_
1
i
m+1
_

i
The
i
terms denote the empirical autocovariances of e
t
from order m through m. For this
estimator to be consistent, m must be large enough to accommodate the short-run persistence of
e
t
, but not too large compared to the sample size T. If the supplied m is non-positive a default value
is computed, namely the integer part of 4
_
T
100
_
1/4
.
The above concept can be generalized to the case where y
t
is thought to be stationary around a
deterministic trend. In this case, formula (25.11) remains unchanged, but the series e
t
is dened as
the residuals from an OLS regression of y
t
on a constant and a linear trend. This second form of
the test is obtained by appending the --trend option to the kpss command:
kpss n y --trend
Note that in this case the asymptotic distribution of the test is dierent and the critical values
reported by gretl dier accordingly.
Panel unit root tests
The most commonly used unit root tests for panel data involve a generalization of the ADF pro-
cedure, in which the joint null hypothesis is that a given times series is non-stationary for all
individuals in the panel.
In this context the ADF regression (25.10) can be rewritten as
y
it
=
it
+
i
y
i,t1
+
p
i
_
j=1
ij
y
i,tj
+
it
(25.12)
The model (25.12) allows for maximal heterogeneity across the individuals in the panel: the pa-
rameters of the deterministic term, the autoregressive coecient , and the lag order p are all
specic to the individual, indexed by i.
One possible modication of this model is to impose the assumption that
i
= for all i; that is,
the individual time series share a common autoregressive root (although they may dier in respect
of other statistical properties). The choice of whether or not to impose this assumption has an
important bearing on the hypotheses under test. Under model (25.12) the joint null is
i
= 0 for
all i, meaning that all the individual time series are non-stationary, and the alternative (simply the
negation of the null) is that at least one individual time series is stationary. When a common is
assumed, the null is that = 0 and the alternative is that < 0. The null still says that all the
individual series are non-stationary, but the alternative now says that they are all stationary. The
choice of model should take this point into account, as well as the gain in power from forming a
pooled estimate of and, of course, the plausibility of assuming a common AR(1) coecient.
3
In gretl, the formulation (25.12) is used automatically when the adf command is used on panel
data. The joint test statistic is formed using the method of Im, Pesaran and Shin (2003). In this
context the behavior of adf diers from regular time-series data: only one case of the deterministic
term is handled per invocation of the command; the default is that
it
includes just a constant but
the --nc and --ct ags can be used to suppress the constant or to include a trend, respectively;
and the quadratic trend option --ctt is not available.
The alternative that imposes a common value of is implemented via the levinlin command.
The test statistic is computed as per Levin, Lin and Chu (2002). As with the adf command, the rst
argument is the lag order and the second is the name of the series to test; and the default case for
the deterministic component is a constant only. The options --nc and --ct have the same eect
as with adf. One renement is that the lag order may be given in either of two forms: if a scalar
is given, this is taken to represent a common value of p for all individuals, but you may instead
provide a vector holding a set of p
i
values, hence allowing the order of autocorrelation of the series
to dier by individual. So, for example, given
levinlin 2 y
levinlin {2,2,3,3,4,4} y
the rst command runs a joint ADF test with a common lag order of 2, while the second (which
assumes a panel with six individuals) allows for diering short-run dynamics. The rst argument
to levinlin can be given as a set of comma-separated integers enclosed in braces, as shown above,
or as the name of an appropriately dimensioned pre-dened matrix (see chapter 15).
Besides variants of the ADF test, the KPSS test also can be used with panel data via the kpss
command. In this case the test (of the null hypothesis that the given time series is stationary for
all individuals) is implemented using the method of Choi (2001). This is an application of meta-
analysis, the statistical technique whereby an overall or composite p-value for the test of a given
null hypothesis can be computed from the p-values of a set of separate tests. Unfortunately, in
the case of the KPSS test we are limited by the unavailability of precise p-values, although if an
individual test statistic falls between the 10 percent and 1 percent critical values we are able to
interpolate with a fair degree of condence. This gives rise to four cases.
1. All the individual KPSS test statistics fall between the 10 percent and 1 percent critical values:
the Choi method gives us a plausible composite p-value.
3
If the assumption of a common seems excessively restrictive, bear in mind that we routinely assume common
slope coecients when estimating panel models, even if this is unlikely to be literally true.
2. Some of the KPSS test statistics exceed the 1 percent value and none fall short of the 10
percent value: we can give an upper bound for the composite p-value by setting the unknown
p-values to 0.01.
3. Some of the KPSS test statistics fall short of the 10 percent critical value but none exceed the
1 percent value: we can give a lower bound to the composite p-value by setting the unknown
p-values to 0.10.
4. None of the above conditions are satised: the Choi method fails to produce any result for
the composite KPSS test.
25.4 Cointegration tests
The generally recommended test for cointegration is the Johansen test, which is discussed in detail
in chapter 27. In this context we oer a few remarks on the cointegration test of Engle and Granger
(1987), which builds on the ADF test discussed above (section 25.3).
For the EngleGranger test, the procedure is:
1. Test each series for a unit root using an ADF test.
2. Run a cointegrating regression via OLS. For this we select one of the potentially cointegrated
variables as dependent, and include the other potentially cointegrated variables as regressors.
3. Perform an ADF test on the residuals from the cointegrating regression.
The idea is that cointegration is supported if (a) the null of non-stationarity is not rejected for each
of the series individually, in step 1, while (b) the null is rejected for the residuals at step 3. That is,
each of the individual series is I(1) but some linear combination of the series is I(0).
This test is implemented in gretl by the coint command, which requires an integer lag order
(for the ADF tests) followed by a list of variables to be tested, the rst of which will be taken
as dependent in the cointegrating regression. Please see the online help for coint, or the Gretl
Command Reference, for further details.
25.5 ARCH and GARCH
Heteroskedasticity means a non-constant variance of the error term in a regression model. Autore-
gressive Conditional Heteroskedasticity (ARCH) is a phenomenon specic to time series models,
whereby the variance of the error displays autoregressive behavior; for instance, the time series ex-
hibits successive periods where the error variance is relatively large, and successive periods where
it is relatively small. This sort of behavior is reckoned to be common in asset markets: an unsettling
piece of news can lead to a period of increased volatility in the market.
An ARCH error process of order q can be represented as
u
t
=
t
t
;
2
t
E(u
2
t
t1
) =
0
+
q
_
i=1
i
u
2
ti
where the
t
s are independently and identically distributed (iid) with mean zero and variance 1,
and where
t
is taken to be the positive square root of
2
t
.
t1
denotes the information set as of
time t 1 and
2
t
is the conditional variance: that is, the variance conditional on information dated
t 1 and earlier.
It is important to notice the dierence between ARCH and an ordinary autoregressive error process.
The simplest (rst-order) case of the latter can be written as
u
t
= u
t1
+
t
; 1 < < 1
where the
t
s are independently and identically distributed with mean zero and variance
2
. With
an AR(1) error, if is positive then a positive value of u
t
will tend to be followed by a positive
u
t+1
. With an ARCH error process, a disturbance u
t
of large absolute value will tend to be followed
by further large absolute values, but with no presumption that the successive values will be of the
same sign. ARCH in asset prices is a stylized fact and is consistent with market eciency; on the
other hand autoregressive behavior of asset prices would violate market eciency.
One can test for ARCH of order q in the following way:
1. Estimate the model of interest via OLS and save the squared residuals, u
2
t
.
2. Perform an auxiliary regression in which the current squared residual is regressed on a con-
stant and q lags of itself.
3. Find the TR
2
value (sample size times unadjusted R
2
) for the auxiliary regression.
4. Refer the TR
2
value to the
2
distribution with q degrees of freedom, and if the p-value is
small enough reject the null hypothesis of homoskedasticity in favor of the alternative of
ARCH(q).
This test is implemented in gretl via the modtest command with the --arch option, which must
follow estimation of a time-series model by OLS (either a single-equation model or a VAR). For
example,
ols y 0 x
modtest 4 --arch
This example species an ARCH order of q = 4; if the order argument is omitted, q is set equal to
the periodicity of the data. In the graphical interface, the ARCH test is accessible from the Tests
menu in the model window (again, for single-equation OLS or VARs).
GARCH
The simple ARCH(q) process is useful for introducing the general concept of conditional het-
eroskedasticity in time series, but it has been found to be insucient in empirical work. The
dynamics of the error variance permitted by ARCH(q) are not rich enough to represent the patterns
found in nancial data. The generalized ARCH or GARCH model is now more widely used.
The representation of the variance of a process in the GARCH model is somewhat (but not exactly)
analogous to the ARMA representation of the level of a time series. The variance at time t is allowed
to depend on both past values of the variance and past values of the realized squared disturbance,
as shown in the following system of equations:
y
t
= X
t
+u
t
(25.13)
u
t
=
t
t
(25.14)
2
t
=
0
+
q
_
i=1
i
u
2
ti
+
p
_
j=1
2
tj
(25.15)
As above,
t
is an iid sequence with unit variance. X
t
is a matrix of regressors (or in the simplest
case, just a vector of 1s allowing for a non-zero mean of y
t
). Note that if p = 0, GARCH collapses
to ARCH(q): the generalization is embodied in the
j
terms that multiply previous values of the
error variance.
In principle the underlying innovation,
t
, could follow any suitable probability distribution, and
besides the obvious candidate of the normal or Gaussian distribution the Students t distribution
has been used in this context. Currently gretl only handles the case where
t
is assumed to be
Gaussian. However, when the --robust option to the garch command is given, the estimator gretl
uses for the covariance matrix can be considered Quasi-Maximum Likelihood even with non-normal
disturbances. See below for more on the options regarding the GARCH covariance matrix.
Example:
garch p q ; y const x
where p 0 and q > 0 denote the respective lag orders as shown in equation (25.15). These values
can be supplied in numerical form or as the names of pre-dened scalar variables.
GARCH estimation
Estimation of the parameters of a GARCH model is by no means a straightforward task. (Consider
equation 25.15: the conditional variance at any point in time,
2
t
, depends on the conditional
variance in earlier periods, but
2
t
is not observed, and must be inferred by some sort of Maximum
Likelihood procedure.) By default gretl uses native code that employs the BFGS maximizer; you
also have the option (activated by the --fcp command-line switch) of using the method proposed
by Fiorentini et al. (1996),
4
which was adopted as a benchmark in the study of GARCH results
by McCullough and Renfro (1998). It employs analytical rst and second derivatives of the log-
likelihood, and uses a mixed-gradient algorithm, exploiting the information matrix in the early
iterations and then switching to the Hessian in the neighborhood of the maximum likelihood. (This
progress can be observed if you append the --verbose option to gretls garch command.)
Several options are available for computing the covariance matrix of the parameter estimates in
connection with the garch command. At a rst level, one can choose between a standard and a
robust estimator. By default, the Hessian is used unless the --robust option is given, in which
case the QML estimator is used. A ner choice is available via the set command, as shown in
Table 25.2.
Table 25.2: Options for the GARCH covariance matrix
command eect
set garch_vcv hessian Use the Hessian
set garch_vcv im Use the Information Matrix
set garch_vcv op Use the Outer Product of the Gradient
set garch_vcv qml QML estimator
set garch_vcv bw BollerslevWooldridge sandwich estimator
It is not uncommon, when one estimates a GARCH model for an arbitrary time series, to nd that
the iterative calculation of the estimates fails to converge. For the GARCH model to make sense,
there are strong restrictions on the admissible parameter values, and it is not always the case
that there exists a set of values inside the admissible parameter space for which the likelihood is
maximized.
The restrictions in question can be explained by reference to the simplest (and much the most
common) instance of the GARCH model, where p = q = 1. In the GARCH(1, 1) model the conditional
variance is
2
t
=
0
+
1
u
2
t1
+
1
2
t1
(25.16)
Taking the unconditional expectation of (25.16) we get
2
=
0
+
1
2
+
1
2
4
The algorithm is based on Fortran code deposited in the archive of the Journal of Applied Econometrics by the
authors, and is used by kind permission of Professor Fiorentini.
so that
2
=

0
1
1
1
For this unconditional variance to exist, we require that
1
+
1
< 1, and for it to be positive we
require that
0
> 0.
A common reason for non-convergence of GARCH estimates (that is, a common reason for the non-
existence of
i
and
i
values that satisfy the above requirements and at the same time maximize
the likelihood of the data) is misspecication of the model. It is important to realize that GARCH, in
itself, allows only for time-varying volatility in the data. If the mean of the series in question is not
constant, or if the error process is not only heteroskedastic but also autoregressive, it is necessary
to take this into account when formulating an appropriate model. For example, it may be necessary
to take the rst dierence of the variable in question and/or to add suitable regressors, X
t
, as in
(25.13).
Chapter 26
Vector Autoregressions
Gretl provides a standard set of procedures for dealing with the multivariate time-series models
known as VARs (Vector AutoRegression). More general modelssuch as VARMAs, nonlinear models
or multivariate GARCH modelsare not provided as of now, although it is entirely possible to
estimate them by writing custom procedures in the gretl scripting language. In this chapter, we will
briey review gretls VAR toolbox.
26.1 Notation
A VAR is a structure whose aim is to model the time persistence of a vector of n time series, y
t
,
via a multivariate autoregression, as in
y
t
= A
1
y
t1
+A
2
y
t2
+ +A
p
y
tp
+Bx
t
+
t
(26.1)
The number of lags p is called the order of the VAR. The vector x
t
, if present, contains a set of
exogenous variables, often including a constant, possibly with a time trend and seasonal dummies.
The vector
t
is typically assumed to be a vector white noise, with covariance matrix .
Equation (26.1) can be written more compactly as
A(L)y
t
= Bx
t
+
t
(26.2)
where A(L) is a matrix polynomial in the lag operator, or as
_
_
_
_
_
_
y
t
y
t1

y
tp1
_
_
_
_
_
_
= A
_
_
_
_
_
_
y
t1
y
t2

y
tp
_
_
_
_
_
_
+
_
_
_
_
_
_
B
0

0
_
_
_
_
_
_
x
t
+
_
_
_
_
_
_
t
0

0
_
_
_
_
_
_
(26.3)
The matrix A is known as the companion matrix and equals
A =
_
_
_
_
_
_
_
A
1
A
2
A
p
I 0 0
0 I 0
.
.
.
.
.
.
.
.
.
.
.
.
_
_
_
_
_
_
_
Equation (26.3) is known as the companion form of the VAR.
Another representation of interest is the so-called VMA representation, which is written in terms
of an innite series of matrices
i
dened as
i
=
y
t
ti
(26.4)
The
i
matrices may be derived by recursive substitution in equation (26.1): for example, assuming
for simplicity that B = 0 and p = 1, equation (26.1) would become
y
t
= Ay
t1
+
t
222
Chapter 26. Vector Autoregressions 223
which could be rewritten as
y
t
= A
n+1
y
tn1
+
t
+A
t1
+A
2
t2
+ +A
n
tn
In this case
i
= A
i
. In general, it is possible to compute
i
as the n n north-west block of the
i-th power of the companion matrix A (so
0
is always an identity matrix).
The VAR is said to be stable if all the eigenvalues of the companion matrix A are smaller than 1
in absolute value, or equivalently, if the matrix polynomial A(L) in equation (26.2) is such that
A(z) = 0 implies z > 1. If this is the case, lim
n
n
= 0 and the vector y
t
is stationary; as a
consequence, the equation
y
t
E(y
t
) =
_
i=0
ti
(26.5)
is a legitimate Wold representation.
If the VAR is not stable, then the inferential procedures that are called for become somewhat more
specialized, except for some simple cases. In particular, if the number of eigenvalues of A with
modulus 1 is between 1 and n1, the canonical tool to deal with these models is the cointegrated
VAR model, discussed in chapter 27.
26.2 Estimation
The gretl command for estimating a VAR is var which, in the command line interface, is invoked
in the following manner:
[ modelname <- ] var p Ylist [; Xlist]
where p is a scalar (the VAR order) and Ylist is a list of variables specifying the content of y
t
.
The optional Xlist argument can be used to specify a set of exogenous variables. If this argument
is omitted, the vector x
t
is taken to contain a constant (only); if present, it must be separated
from Ylist by a semicolon. Note, however, that a few common choices can be obtained in a
simpler way: the options --trend and --seasonals call for inclusion of a linear trend and a set of
seasonal dummies respectively. In addition the --nc option (no constant) can be used to suppress
the standard inclusion of a constant.
The <- construct can be used to store the model under a name (see section 3.2), if so desired. To
estimate a VAR using the graphical interface, choose Time Series, Vector Autoregression, under
the Model menu.
The parameters in eq. (26.1) are typically free from restrictions, which implies that multivariate
OLS provides a consistent and asymptotically ecient estimator of all the parameters.
1
Given
the simplicity of OLS, this is what every software package, including gretl, uses; example script
26.1 exemplies the fact that the var command gives you exactly the output you would have
from a battery of OLS regressions. The advantage of using the dedicated command is that, after
estimation is done, it makes it much easier to access certain quantities and manage certain tasks.
For example, the $coeff accessor returns the estimated coecients as a matrix with n columns
and $sigma returns an estimate of the matrix , the covariance matrix of
t
.
Moreover, for each variable in the system an F test is automatically performed, in which the null hy-
pothesis is that no lags of variable j are signicant in the equation for variable i. This is commonly
known as a Granger causality test.
In addition, two accessors become available for the companion matrix ($compan) and the VMA rep-
resentation ($vma). The latter deserves a detailed description: since the VMA representation (26.5)
is of innite order, gretl denes a horizon up to which the
i
matrices are computed automatically.
1
In fact, under normality of
t
OLS is indeed the conditional ML estimator. You may want to use other methods if you
need to estimate a VAR in which some parameters are constrained.
Example 26.1: Estimation of a VAR via OLS
Input:
open sw_ch14.gdt
genr infl = 400*sdiff(log(PUNEW))
scalar p = 2
list X = LHUR infl
list Xlag = lags(p,X)
loop foreach i X
ols $i const Xlag
end loop
var p X
Model 1: OLS, using observations 1960:3-1999:4 (T = 158)
Dependent variable: LHUR
coefficient std. error t-ratio p-value
--------------------------------------------------------
const 0.113673 0.0875210 1.299 0.1960
LHUR_1 1.54297 0.0680518 22.67 8.78e-51 ***
LHUR_2 -0.583104 0.0645879 -9.028 7.00e-16 ***
infl_1 0.0219040 0.00874581 2.505 0.0133 **
infl_2 -0.0148408 0.00920536 -1.612 0.1090
Mean dependent var 6.019198 S.D. dependent var 1.502549
Sum squared resid 8.654176 S.E. of regression 0.237830
...
VAR system, lag order 2
OLS estimates, observations 1960:3-1999:4 (T = 158)
Log-likelihood = -322.73663
Determinant of covariance matrix = 0.20382769
AIC = 4.2119
BIC = 4.4057
HQC = 4.2906
Portmanteau test: LB(39) = 226.984, df = 148 [0.0000]
Equation 1: LHUR
--------------------------------------------------------
const 0.113673 0.0875210 1.299 0.1960
LHUR_1 1.54297 0.0680518 22.67 8.78e-51 ***
LHUR_2 -0.583104 0.0645879 -9.028 7.00e-16 ***
infl_1 0.0219040 0.00874581 2.505 0.0133 **
infl_2 -0.0148408 0.00920536 -1.612 0.1090
Sum squared resid 8.654176 S.E. of regression 0.237830
Periodicity horizon
Quarterly 20 (5 years)
Monthly 24 (2 years)
Daily 3 weeks
All other cases 10
Table 26.1: VMA horizon as a function of the dataset periodicity
By default, this is a function of the periodicity of the data (see table 26.1), but it can be set by the
user to any desired value via the set command with the horizon parameter, as in
set horizon 30
Calling the horizon h, the $vma accessor returns an (h+1) n
2
matrix, in which the (i +1)-th row
is the vectorized form of
i
.
VAR lag-order selection
In order to help the user choose the most appropriate VAR order, gretl provides a special variant
of the var command:
var p Ylist [; Xlist] --lagselect
When the --lagselect option is given, estimation is performed for all lags up to p and a table
is printed: it displays, for each order, a Likelihood Ratio test for the order p versus p 1, plus
an array of information criteria (see chapter 23). For each information criterion in the table, a
star indicates what appears to be the best choice. The same output can be obtained through the
graphical interface via the Time Series, VAR lag selection entry under the Model menu.
Warning: in nite samples the choice of the maximum lag, p, may aect the outcome of the proce-
dure. This is not a bug, but rather an unavoidable side eect of the way these comparisons should
be made. If your sample contains T observations and you invoke the lag selection procedure with
maximum order p, gretl examines all VARs of order ranging form 1 to p, estimated on a uniform
sample of T p observations. In other words, the comparison procedure does not use all the avail-
able data when estimating VARs of order less than p, so as to ensure that all the models in the
comparison are estimated on the same data range. Choosing a dierent value of p may therefore
alter the results, although this is unlikely to happen if your sample size is reasonably large.
An example of this unpleasant phenomenon is given in example script 26.2. As can be seen, ac-
cording to the Hannan-Quinn criterion, order 2 seems preferable to order 1 if the maximum tested
order is 4, but the situation is reversed if the maximum tested order is 6.
26.3 Structural VARs
Gretl does not currently provide a native implementation for the general class of models known as
Structural VARs; however, it provides an implementation of the Cholesky decomposition-based
approach, the classic and most popular SVAR variant.
IRF and FEVD
Assume that the disturbance in equation (26.1) can be thought of as a linear function of a vector
of structural shocks u
t
, which are assumed to have unit variance and to be mutually unncorrelated,
so V(u
t
) = I. If
t
= Ku
t
, it follows that = V(
t
) = KK
.
Example 26.2: VAR lag selection via Information Criteria
Input:
open denmark
list Y = 1 2 3 4
var 4 Y --lagselect
var 6 Y --lagselect
VAR system, maximum lag order 4
The asterisks below indicate the best (that is, minimized) values
of the respective information criteria, AIC = Akaike criterion,
BIC = Schwarz Bayesian criterion and HQC = Hannan-Quinn criterion.
lags loglik p(LR) AIC BIC HQC
1 609.15315 -23.104045 -22.346466* -22.814552
2 631.70153 0.00013 -23.360844* -21.997203 -22.839757*
3 642.38574 0.16478 -23.152382 -21.182677 -22.399699
4 653.22564 0.15383 -22.950025 -20.374257 -21.965748
VAR system, maximum lag order 6
The asterisks below indicate the best (that is, minimized) values
of the respective information criteria, AIC = Akaike criterion,
BIC = Schwarz Bayesian criterion and HQC = Hannan-Quinn criterion.
lags loglik p(LR) AIC BIC HQC
1 594.38410 -23.444249 -22.672078* -23.151288*
2 615.43480 0.00038 -23.650400* -22.260491 -23.123070
3 624.97613 0.26440 -23.386781 -21.379135 -22.625083
4 636.03766 0.13926 -23.185210 -20.559827 -22.189144
5 658.36014 0.00016 -23.443271 -20.200150 -22.212836
6 669.88472 0.11243 -23.260601 -19.399743 -21.795797
The main object of interest in this setting is the sequence of matrices
C
k
=
y
t
u
ti
=
k
K, (26.6)
known as the structural VMA representation. From the C
k
matrices dened in equation (26.6) two
quantities of interest may be derived: the Impulse Response Function (IRF) and the Forecast Error
Variance Decomposition (FEVD).
The IRF of variable i to shock j is simply the sequence of the elements in row i and column j of
the C
k
matrices. In symbols:
J
i,j,k
=
y
i,t
u
j,tk
As a rule, Impulse Response Functions are plotted as a function of k, and are interpreted as the
eect that a shock has on an observable variable through time. Of course, what we observe are
the estimated IRFs, so it is natural to endow them with condence intervals: following common
practice, gretl computes the condence intervals by using the bootstrap;
2
details are given later in
this section.
Another quantity of interest that may be computed from the structural VMA representation is the
Forecast Error Variance Decomposition (FEVD). The forecast error variance after h steps is given by
h
=
h
_
k=0
C
k
C
k
hence the variance for variable i is
2
i
= [
h
]
i,i
=
h
_
k=0
diag(C
k
C
k
)
i
=
h
_
k=0
n
_
l=1
(
k
c
i.l
)
2
where
k
c
i.l
is, trivially, the i, l element of C
k
. As a consequence, the share of uncertainty on variable
i that can be attributed to the j-th shock after h periods equals
`1
i,j,h
=
h
k=0
(
k
c
i.j
)
2
h
k=0
n
l=1
(
k
c
i.l
)
2
.
This makes it possible to quantify which shocks are most important to determine a certain variable
in the short and/or in the long run.
Triangularization
The formula 26.6 takes K as known, while of course it has to be estimated. The estimation problem
has been the subject of an enormous body of literature we will not even attempt to summarize
here: see for example (Ltkepohl, 2005, chapter 9).
Suce it to say that the most popular choice dates back to Sims (1980), and consists in assuming
that K is lower triangular, so its estimate is simply the Cholesky decomposition of the estimate of .
The main consequence of this choice is that the ordering of variables within the vector y
t
becomes
meaningful: since K is also the matrix of Impulse Response Functions at lag 0, the triangularity
assumption means that the rst variable in the ordering responds instantaneously only to shock
number 1, the second one only to shocks 1 and 2, and so forth. For this reason, each variable is
thought to own one shock: variable 1 owns shock number 1, and so on.
In this sort of exercise, therefore, the ordering of the y variables is important, and the applied
literature has developed the most exogenous rst mantrawhere, in this setting, exogenous
2
It is possible, in principle, to compute analytical condence intervals via an asymptotic approximation, but this is
not a very popular choice: asymptotic formulae are known to often give a very poor approximation of the nite-sample
properties.
really means instantaneously insensitive to structural shocks.
3
To put it dierently, if variable
foo comes before variable bar in the Y list, it follows that the shock owned by foo aects bar
instantaneously, but not vice versa.
Impulse Response Functions and the FEVD can be printed out via the command line interface by us-
ing the --impulse-response and --variance-decomp options, respectively. If you need to store
them into matrices, you can compute the structural VMA and proceed from there. For example, the
following code snippet shows you how to compute a matrix containing the IRFs:
open denmark
list Y = 1 2 3 4
scalar n = nelem(Y)
var 2 Y --quiet --impulse
matrix K = cholesky($sigma)
matrix V = $vma
matrix IRF = V * (K ** I(n))
print IRF
in which the equality
vec(C
k
) = vec(
k
K) = (K
I)vec(
k
)
was used.
FIXME: show all the nice stu we have under the GUI.
IRF bootstrap
FIXME: todo
3
The word exogenous has caught on in this context, but its a rather unfortunate choice: for a start, each shock
impacts on every variable after one lag, so nothing is really exogenous here. A better choice of words would probably
have been something like sturdy, but its too late now.
Chapter 27
Cointegration and Vector Error Correction Models
27.1 Introduction
The twin concepts of cointegration and error correction have drawn a good deal of attention in
macroeconometrics over recent years. The attraction of the Vector Error Correction Model (VECM)
is that it allows the researcher to embed a representation of economic equilibrium relationships
within a relatively rich time-series specication. This approach overcomes the old dichotomy be-
tween (a) structural models that faithfully represented macroeconomic theory but failed to t the
data, and (b) time-series models that were accurately tailored to the data but dicult if not impos-
sible to interpret in economic terms.
The basic idea of cointegration relates closely to the concept of unit roots (see section 25.3). Sup-
pose we have a set of macroeconomic variables of interest, and we nd we cannot reject the hypoth-
esis that some of these variables, considered individually, are non-stationary. Specically, suppose
we judge that a subset of the variables are individually integrated of order 1, or I(1). That is, while
they are non-stationary in their levels, their rst dierences are stationary. Given the statistical
problems associated with the analysis of non-stationary data (for example, the threat of spurious
regression), the traditional approach in this case was to take rst dierences of all the variables
before proceeding with the analysis.
But this can result in the loss of important information. It may be that while the variables in
question are I(1) when taken individually, there exists a linear combination of the variables that
is stationary without dierencing, or I(0). (There could be more than one such linear combina-
tion.) That is, while the ensemble of variables may be free to wander over time, nonetheless the
variables are tied together in certain ways. And it may be possible to interpret these ties, or
cointegrating vectors, as representing equilibrium conditions.
For example, suppose we nd some or all of the following variables are I(1): money stock, M, the
price level, P, the nominal interest rate, R, and output, Y. According to standard theories of the
demand for money, we would nonetheless expect there to be an equilibrium relationship between
real balances, interest rate and output; for example
mp =
0
+
1
y +
2
r
1
> 0,
2
< 0
where lower-case variable names denote logs. In equilibrium, then,
mp
1
y
2
r =
0
Realistically, we should not expect this condition to be satised each period. We need to allow for
the possibility of short-run disequilibrium. But if the system moves back towards equilibrium fol-
lowing a disturbance, it follows that the vector x = (m, p, y, r)
is bound by a cointegrating vector
= (
1
,
2
,
3
,
4
), such that
x is stationary (with a mean of

0
). Furthermore, if equilibrium is
correctly characterized by the simple model above, we have
2
=
1
,
3
< 0 and
4
> 0. These
things are testable within the context of cointegration analysis.
There are typically three steps in this sort of analysis:
1. Test to determine the number of cointegrating vectors, the cointegrating rank of the system.
2. Estimate a VECM with the appropriate rank, but subject to no further restrictions.
229
Chapter 27. Cointegration and Vector Error Correction Models 230
3. Probe the interpretation of the cointegrating vectors as equilibrium conditions by means of
restrictions on the elements of these vectors.
The following sections expand on each of these points, giving further econometric details and
explaining how to implement the analysis using gretl.
27.2 Vector Error Correction Models as representation of a cointegrated system
Consider a VAR of order p with a deterministic part given by
t
(typically, a polynomial in time).
One can write the n-variate process y
t
as
y
t
=
t
+A
1
y
t1
+A
2
y
t2
+ +A
p
y
tp
+
t
(27.1)
But since y
ti
y
t1
(y
t1
+y
t2
+ +y
ti+1
), we can re-write the above as
y
t
=
t
+y
t1
+
p1
_
i=1
i
y
ti
+
t
, (27.2)
where =
p
i=1
A
i
I and
i
=
p
j=i+1
A
j
. This is the VECM representation of (27.1).
The interpretation of (27.2) depends crucially on r, the rank of the matrix .
If r = 0, the processes are all I(1) and not cointegrated.
If r = n, then is invertible and the processes are all I(0).
Cointegration occurs in between, when 0 < r < n and can be written as
. In this case,
y
t
is I(1), but the combination z
t
=
y
t
is I(0). If, for example, r = 1 and the rst element
of was 1, then one could write z
t
= y
1,t
+
2
y
2,t
+ +
n
y
n,t
, which is equivalent to
saying that
y
1
t
=
2
y
2,t
+ +
n
y
n,t
z
t
is a long-run equilibrium relationship: the deviations z
t
may not be 0 but they are stationary.
In this case, (27.2) can be written as
y
t
=
t
+
y
t1
+
p1
_
i=1
i
y
ti
+
t
. (27.3)
If were known, then z
t
would be observable and all the remaining parameters could be
estimated via OLS. In practice, the procedure estimates rst and then the rest.
The rank of is investigated by computing the eigenvalues of a closely related matrix whose rank
is the same as : however, this matrix is by construction symmetric and positive semidenite. As a
consequence, all its eigenvalues are real and non-negative, and tests on the rank of can therefore
be carried out by testing how many eigenvalues are 0.
If all the eigenvalues are signicantly dierent from 0, then all the processes are stationary. If,
on the contrary, there is at least one zero eigenvalue, then the y
t
process is integrated, although
some linear combination
y
t
might be stationary. At the other extreme, if no eigenvalues are
signicantly dierent from 0, then not only is the process y
t
non-stationary, but the same holds
for any linear combination
y
t
; in other words, no cointegration occurs.
Estimation typically proceeds in two stages: rst, a sequence of tests is run to determine r, the
cointegration rank. Then, for a given rank the parameters in equation (27.3) are estimated. The two
commands that gretl oers for estimating these systems are coint2 and vecm, respectively.
The syntax for coint2 is
coint2 p ylist [ ; xlist [ ; zlist ] ]
where p is the number of lags in (27.1); ylist is a list containing the y
t
variables; xlist is an
optional list of exogenous variables; and zlist is another optional list of exogenous variables
whose eects are assumed to be conned to the cointegrating relationships.
The syntax for vecm is
vecm p r ylist [ ; xlist [ ; zlist ] ]
where p is the number of lags in (27.1); r is the cointegration rank; and the lists ylist, xlist and
zlist have the same interpretation as in coint2.
Both commands can be given specic options to handle the treatment of the deterministic compo-
nent
t
. These are discussed in the following section.
27.3 Interpretation of the deterministic components
Statistical inference in the context of a cointegrated system depends on the hypotheses one is
willing to make on the deterministic terms, which leads to the famous ve cases.
In equation (27.2), the term
t
is usually understood to take the form
t
=
0
+
1
t.
In order to have the model mimic as closely as possible the features of the observed data, there is a
preliminary question to settle. Do the data appear to follow a deterministic trend? If so, is it linear
or quadratic?
Once this is established, one should impose restrictions on
0
and
1
that are consistent with this
judgement. For example, suppose that the data do not exhibit a discernible trend. This means that
y
t
is on average zero, so it is reasonable to assume that its expected value is also zero. Write
equation (27.2) as
(L)y
t
=
0
+
1
t +z
t1
+
t
, (27.4)
where z
t
=
y
t
is assumed to be stationary and therefore to possess nite moments. Taking
unconditional expectations, we get
0 =
0
+
1
t +m
z
.
Since the left-hand side does not depend on t, the restriction
1
= 0 is a safe bet. As for
0
, there are
just two ways to make the above expression true: either
0
= 0 with m
z
= 0, or
0
equals m
z
.
The latter possibility is less restrictive in that the vector
0
may be non-zero, but is constrained to
be a linear combination of the columns of . In that case,
0
can be written as c, and one may
write (27.4) as
(L)y
t
=
_
c
_
_
y
t1
1
_
+
t
.
The long-run relationship therefore contains an intercept. This type of restriction is usually written
0
= 0,
where
is the left null space of the matrix .

An intuitive understanding of the issue can be gained by means of a simple example. Consider a
series x
t
which behaves as follows
x
t
= m+x
t1
+
t
where m is a real number and
t
is a white noise process: x
t
is then a random walk with drift m.
In the special case m = 0, the drift disappears and x
t
is a pure random walk.
Consider now another process y
t
, dened by
y
t
= k +x
t
+u
t
where, again, k is a real number and u
t
is a white noise process. Since u
t
is stationary by denition,
x
t
and y
t
cointegrate: that is, their dierence
z
t
= y
t
x
t
= k +u
t
is a stationary process. For k = 0, z
t
is simple zero-mean white noise, whereas for k }= 0 the process
z
t
is white noise with a non-zero mean.
After some simple substitutions, the two equations above can be represented jointly as a VAR(1)
system
_
y
t
x
t
_
=
_
k +m
m
_
+
_
0 1
0 1
__
y
t1
x
t1
_
+
_
u
t
+
t
t
_
or in VECM form
_
y
t
x
t
_
=
_
k +m
m
_
+
_
1 1
0 0
__
y
t1
x
t1
_
+
_
u
t
+
t
t
_
=
=
_
k +m
m
_
+
_
1
0
_
_
1 1
_
_
y
t1
x
t1
_
+
_
u
t
+
t
t
_
=
=
0
+
_
y
t1
x
t1
_
+
t
=
0
+z
t1
+
t
,
where is the cointegration vector and is the loadings or adjustments vector.
We are now ready to consider three possible cases:
1. m}= 0: In this case x
t
is trended, as we just saw; it follows that y
t
also follows a linear trend
because on average it keeps at a xed distance k from x
t
. The vector
0
is unrestricted.
2. m = 0 and k }= 0: In this case, x
t
is not trended and as a consequence neither is y
t
. However,
the mean distance between y
t
and x
t
is non-zero. The vector
0
is given by
0
=
_
k
0
_
which is not null and therefore the VECM shown above does have a constant term. The
constant, however, is subject to the restriction that its second element must be 0. More
generally,
0
is a multiple of the vector . Note that the VECM could also be written as
_
y
t
x
t
_
=
_
1
0
_
_
1 1 k
_
_
_
_
_
y
t1
x
t1
1
_
_
_
_
+
_
u
t
+
t
t
_
which incorporates the intercept into the cointegration vector. This is known as the restricted
constant case.
3. m = 0 and k = 0: This case is the most restrictive: clearly, neither x
t
nor y
t
are trended, and
the mean distance between them is zero. The vector
0
is also 0, which explains why this case
is referred to as no constant.
In most cases, the choice between these three possibilities is based on a mix of empirical obser-
vation and economic reasoning. If the variables under consideration seem to follow a linear trend
then we should not place any restriction on the intercept. Otherwise, the question arises of whether
it makes sense to specify a cointegration relationship which includes a non-zero intercept. One ex-
ample where this is appropriate is the relationship between two interest rates: generally these are
not trended, but the VAR might still have an intercept because the dierence between the two (the
interest rate spread) might be stationary around a non-zero mean (for example, because of a risk
or liquidity premium).
The previous example can be generalized in three directions:
1. If a VAR of order greater than 1 is considered, the algebra gets more convoluted but the
conclusions are identical.
2. If the VAR includes more than two endogenous variables the cointegration rank r can be
greater than 1. In this case, is a matrix with r columns, and the case with restricted constant
entails the restriction that
0
should be some linear combination of the columns of .
3. If a linear trend is included in the model, the deterministic part of the VAR becomes
0
+
1
t.
The reasoning is practically the same as above except that the focus now centers on
1
rather
than
0
. The counterpart to the restricted constant case discussed above is a restricted
trend case, such that the cointegration relationships include a trend but the rst dierences
of the variables in question do not. In the case of an unrestricted trend, the trend appears
in both the cointegration relationships and the rst dierences, which corresponds to the
presence of a quadratic trend in the variables themselves (in levels).
In order to accommodate the ve cases, gretl provides the following options to the coint2 and
vecm commands:
t
option ag description
0 --nc no constant
0
,
0
= 0 --rc restricted constant
0
--uc unrestricted constant
0
+
1
t,
1
= 0 --crt constant + restricted trend
0
+
1
t --ct constant + unrestricted trend
Note that for this command the above options are mutually exclusive. In addition, you have the
option of using the --seasonal options, for augmenting
t
with centered seasonal dummies. In
each case, p-values are computed via the approximations devised by Doornik (1998).
27.4 The Johansen cointegration tests
The two Johansen tests for cointegration are used to establish the rank of , or in other words
the number of cointegrating vectors. These are the -max test, for hypotheses on individual
eigenvalues, and the trace test, for joint hypotheses. Suppose that the eigenvalues
i
are sorted
from largest to smallest. The null hypothesis for the -max test on the i-th eigenvalue is that
i
= 0.
The corresponding trace test, instead, considers the hypothesis
j
= 0 for all j i.
The gretl command coint2 performs these two tests. The corresponding menu entry in the GUI is
Model, Time Series, Cointegration Test, Johansen.
As in the ADF test, the asymptotic distribution of the tests varies with the deterministic component
t
one includes in the VAR (see section 27.3 above). The following code uses the denmark data le,
supplied with gretl, to replicate Johansens example found in his 1995 book.
open denmark
coint2 2 LRM LRY IBO IDE --rc --seasonal
In this case, the vector y
t
in equation (27.2) comprises the four variables LRM, LRY, IBO, IDE. The
number of lags equals p in (27.2) (that is, the number of lags of the model written in VAR form).
Part of the output is reported below:
Johansen test:
Number of equations = 4
Lag order = 2
Estimation period: 1974:3 - 1987:3 (T = 53)
Case 2: Restricted constant
Rank Eigenvalue Trace test p-value Lmax test p-value
0 0.43317 49.144 [0.1284] 30.087 [0.0286]
1 0.17758 19.057 [0.7833] 10.362 [0.8017]
2 0.11279 8.6950 [0.7645] 6.3427 [0.7483]
3 0.043411 2.3522 [0.7088] 2.3522 [0.7076]
Both the trace and -max tests accept the null hypothesis that the smallest eigenvalue is 0 (see the
last row of the table), so we may conclude that the series are in fact non-stationary. However, some
linear combination may be I(0), since the -max test rejects the hypothesis that the rank of is 0
(though the trace test gives less clear-cut evidence for this, with a p-value of 0.1284).
27.5 Identication of the cointegration vectors
The core problem in the estimation of equation (27.2) is to nd an estimate of that has by con-
struction rank r, so it can be written as =
, where is the matrix containing the cointegration

vectors and contains the adjustment or loading coecients whereby the endogenous vari-
ables respond to deviation from equilibrium in the previous period.
Without further specication, the problem has multiple solutions (in fact, innitely many). The
parameters and are under-identied: if all columns of are cointegration vectors, then any
arbitrary linear combinations of those columns is a cointegration vector too. To put it dierently,
if =
0
0
for specic matrices
0
and
0
, then also equals (
0
Q)(Q
1
0
) for any conformable
non-singular matrix Q. In order to nd a unique solution, it is therefore necessary to impose
some restrictions on and/or . It can be shown that the minimum number of restrictions that
is necessary to guarantee identication is r
2
. Normalizing one coecient per column to 1 (or 1,
according to taste) is a trivial rst step, which also helps in that the remaining coecients can be
interpreted as the parameters in the equilibrium relations, but this only suces when r = 1.
The method that gretl uses by default is known as the Phillips normalization, or triangular
representation.
1
The starting point is writing in partitioned form as in
=
_

1
2
_
,
where
1
is an r r matrix and
2
is (n r) r. Assuming that
1
has full rank, can be
post-multiplied by
1
1
, giving
=
_
I
1
1
_
=
_
I
B
_
,
The coecients that gretl produces are

, with B known as the matrix of unrestricted coecients.
In terms of the underlying equilibriumrelationship, the Phillips normalization expresses the system
1
For comparison with other studies, you may wish to normalize dierently. Using the set command you
can do set vecm_norm diag to select a normalization that simply scales the columns of the original such that
ij
= 1 for i = j and i r, as used in the empirical section of Boswijk and Doornik (2004). Another alternative is
set vecm_norm first, which scales such that the elements on the rst row equal 1. To suppress normalization
altogether, use set vecm_norm none. (To return to the default: set vecm_norm phillips.)
of r equilibrium relations as
y
1,t
= b
1,r+1
y
r+1,t
+. . . +b
1,n
y
n,t
y
2,t
= b
2,r+1
y
r+1,t
+. . . +b
2,n
y
n,t
.
.
.
y
r,t
= b
r,r+1
y
r+1,t
+. . . +b
r,n
y
r,t
where the rst r variables are expressed as functions of the remaining nr.
Although the triangular representation ensures that the statistical problem of estimating is
solved, the resulting equilibrium relationships may be dicult to interpret. In this case, the user
may want to achieve identication by specifying manually the system of r
2
constraints that gretl
will use to produce an estimate of .
As an example, consider the money demand system presented in section 9.6 of Verbeek (2004). The
variables used are m (the log of real money stock M1), infl (ination), cpr (the commercial paper
rate), y (log of real GDP) and tbr (the Treasury bill rate).
2
Estimation of can be performed via the commands
open money.gdt
smpl 1954:1 1994:4
vecm 6 2 m infl cpr y tbr --rc
and the relevant portion of the output reads
Maximum likelihood estimates, observations 1954:1-1994:4 (T = 164)
Cointegration rank = 2
Case 2: Restricted constant
beta (cointegrating vectors, standard errors in parentheses)
m 1.0000 0.0000
(0.0000) (0.0000)
infl 0.0000 1.0000
(0.0000) (0.0000)
cpr 0.56108 -24.367
(0.10638) (4.2113)
y -0.40446 -0.91166
(0.10277) (4.0683)
tbr -0.54293 24.786
(0.10962) (4.3394)
const -3.7483 16.751
(0.78082) (30.909)
Interpretation of the coecients of the cointegration matrix would be easier if a meaning could
be attached to each of its columns. This is possible by hypothesizing the existence of two long-run
relationships: a money demand equation
m = c
1
+
1
infl +
2
y +
3
tbr
and a risk premium equation
cpr = c
2
+
4
infl +
5
y +
6
tbr
2
This data set is available in the verbeek data package; see http://gretl.sourceforge.net/gretl_data.html.
which imply that the cointegration matrix can be normalized as
=
_
_
_
_
_
_
_
_
_
_
_
_
1 0
1

4
0 1
2

5
3

6
c
1
c
2
_
_
_
_
_
_
_
_
_
_
_
_
This renormalization can be accomplished by means of the restrict command, to be given after
the vecm command or, in the graphical interface, by selecting the Test, Linear Restrictions menu
entry. The syntax for entering the restrictions should be fairly obvious:
3
restrict
b[1,1] = -1
b[1,3] = 0
b[2,1] = 0
b[2,3] = -1
end restrict
which produces
Cointegrating vectors (standard errors in parentheses)
m -1.0000 0.0000
(0.0000) (0.0000)
infl -0.023026 0.041039
(0.0054666) (0.027790)
cpr 0.0000 -1.0000
(0.0000) (0.0000)
y 0.42545 -0.037414
(0.033718) (0.17140)
tbr -0.027790 1.0172
(0.0045445) (0.023102)
const 3.3625 0.68744
(0.25318) (1.2870)
27.6 Over-identifying restrictions
One purpose of imposing restrictions on a VECM system is simply to achieve identication. If these
restrictions are simply normalizations, they are not testable and should have no eect on the max-
imized likelihood. In addition, however, one may wish to formulate constraints on and/or that
derive from the economic theory underlying the equilibrium relationships; substantive restrictions
of this sort are then testable via a likelihood-ratio statistic.
Gretl is capable of testing general linear restrictions of the form
R
b
vec() = q (27.5)
and/or
R
a
vec() = 0 (27.6)
Note that the restriction may be non-homogeneous (q 0) but the restriction must be homo-
geneous. Nonlinear restrictions are not supported, and neither are restrictions that cross between
3
Note that in this context we are bending the usual matrix indexation convention, using the leading index to refer to
the column of (the particular cointegrating vector). This is standard practice in the literature, and defensible insofar as
it is the columns of (the cointegrating relations or equilibrium errors) that are of primary interest.
and . In the case where r > 1 such restrictions may be in common across all the columns of
(or ) or may be specic to certain columns of these matrices. This is the case discussed in Boswijk
(1995) and Boswijk and Doornik (2004), section 4.4.
The restrictions (27.5) and (27.6) may be written in explicit form as
vec() = H+h
0
(27.7)
and
vec(
) = G (27.8)
respectively, where and are the free parameter vectors associated with and respectively.
We may refer to the free parameters collectively as (the column vector formed by concatenating
and ). Gretl uses this representation internally when testing the restrictions.
If the list of restrictions that is passed to the restrict command contains more constraints than
necessary to achieve identication, then an LR test is performed; moreover, the restrict com-
mand can be given the --full switch, in which case full estimates for the restricted system are
printed (including the
i
terms), and the system thus restricted becomes the current model for
the purposes of further tests. Thus you are able to carry out cumulative tests, as in Chapter 7 of
Johansen (1995).
Syntax
The full syntax for specifying the restriction is an extension of that exemplied in the previous
section. Inside a restrict. . . end restrict block, valid statements are of the form
parameter linear combination = scalar
where a parameter linear combination involves a weighted sum of individual elements of or
(but not both in the same combination); the scalar on the right-hand side must be 0 for combina-
tions involving , but can be any real number for combinations involving . Below, we give a few
examples of valid restrictions:
b[1,1] = 1.618
b[1,4] + 2*b[2,5] = 0
a[1,3] = 0
a[1,1] - a[1,2] = 0
Special syntax is used when a certain constraint should be applied to all columns of : in this case,
one index is given for each b term, and the square brackets are dropped. Hence, the following
syntax
restrict
b1 + b2 = 0
end restrict
corresponds to
=
_
_
_
_
_
_
11

21
11

21
13

23
14

24
_
_
_
_
_
_
The same convention is used for : when only one index is given for an a term the restriction is
presumed to apply to all r columns of , or in other words the variable associated with the given
row of is weakly exogenous. For instance, the formulation
restrict
a3 = 0
a4 = 0
end restrict
species that variables 3 and 4 do not respond to the deviation from equilibrium in the previous
period.
4
A variant on the single-index syntax for common restrictions on and is available: you can
replace the index number with the name of the corresponding variable, in square brackets. For
example, instead of a3 = 0 one could write a[cpr] = 0, if the third variable in the system is
named cpr.
Finally, a short-cut is available for setting up complex restrictions (but currently only in relation
to ): you can specify R
b
and q, as in R
b
vec() = q, by giving the names of previously dened
matrices. For example,
matrix I4 = I(4)
matrix vR = I4**(I4~zeros(4,1))
matrix vq = mshape(I4,16,1)
restrict
R = vR
q = vq
end restrict
which manually imposes Phillips normalization on the estimates for a system with cointegrating
rank 4.
An example
Brand and Cassola (2004) propose a money demand system for the Euro area, in which they postu-
late three long-run equilibrium relationships:
money demand m=
l
l +
y
y
Fisher equation = l
Expectation theory of l = s
interest rates
where m is real money demand, l and s are long- and short-term interest rates, y is output and
is ination.
5
(The names for these variables in the gretl data le are m_p, rl, rs, y and infl,
respectively.)
The cointegration rank assumed by the authors is 3 and there are 5 variables, giving 15 elements
in the matrix. 3 3 = 9 restrictions are required for identication, and a just-identied system
would have 15 9 = 6 free parameters. However, the postulated long-run relationships feature
only three free parameters, so the over-identication rank is 3.
Example 27.1 replicates Table 4 on page 824 of the Brand and Cassola article.
6
Note that we use
the $lnl accessor after the vecm command to store the unrestricted log-likelihood and the $rlnl
accessor after restrict for its restricted counterpart.
4
Note that when two indices are given in a restriction on the indexation is consistent with that for restrictions:
the leading index denotes the cointegrating vector and the trailing index the equation number.
5
A traditional formulation of the Fisher equation would reverse the roles of the variables in the second equation,
but this detail is immaterial in the present context; moreover, the expectation theory of interest rates implies that the
third equilibrium relationship should include a constant for the liquidity premium. However, since in this example the
system is estimated with the constant term unrestricted, the liquidity premium gets merged in the system intercept and
disappears from z
t
.
6
Modulo what appear to be a few typos in the article.
Example 27.1: Estimation of a money demand system with constraints on
Input:
open brand_cassola.gdt
# perform a few transformations
m_p = m_p*100
y = y*100
infl = infl/4
rs = rs/4
rl = rl/4
# replicate table 4, page 824
vecm 2 3 m_p infl rl rs y -q
genr ll0 = $lnl
restrict --full
b[1,1] = 1
b[1,2] = 0
b[1,4] = 0
b[2,1] = 0
b[2,2] = 1
b[2,4] = 0
b[2,5] = 0
b[3,1] = 0
b[3,2] = 0
b[3,3] = 1
b[3,4] = -1
b[3,5] = 0
end restrict
genr ll1 = $rlnl
Partial output:
Unrestricted loglikelihood (lu) = 116.60268
Restricted loglikelihood (lr) = 115.86451
2 * (lu - lr) = 1.47635
P(Chi-Square(3) > 1.47635) = 0.68774
beta (cointegrating vectors, standard errors in parentheses)
m_p 1.0000 0.0000 0.0000
(0.0000) (0.0000) (0.0000)
infl 0.0000 1.0000 0.0000
(0.0000) (0.0000) (0.0000)
rl 1.6108 -0.67100 1.0000
(0.62752) (0.049482) (0.0000)
rs 0.0000 0.0000 -1.0000
(0.0000) (0.0000) (0.0000)
y -1.3304 0.0000 0.0000
(0.030533) (0.0000) (0.0000)
The example continues in script 27.2, where we perform further testing to check whether (a) the
income elasticity in the money demand equation is 1 (
y
= 1) and (b) the Fisher relation is homo-
geneous ( = 1). Since the --full switch was given to the initial restrict command, additional
restrictions can be applied without having to repeat the previous ones. (The second script contains
a few printf commands, which are not strictly necessary, to format the output nicely.) It turns out
that both of the additional hypotheses are rejected by the data, with p-values of 0.002 and 0.004.
Example 27.2: Further testing of money demand system
Input:
restrict
b[1,5] = -1
end restrict
genr ll_uie = $rlnl
restrict
b[2,3] = -1
end restrict
genr ll_hfh = $rlnl
# replicate table 5, page 824
printf "Testing zero restrictions in cointegration space:\n"
printf " LR-test, rank = 3: chi^2(3) = %6.4f [%6.4f]\n", 2*(ll0-ll1), \
pvalue(X, 3, 2*(ll0-ll1))
printf "Unit income elasticity: LR-test, rank = 3:\n"
printf " chi^2(4) = %g [%6.4f]\n", 2*(ll0-ll_uie), \
pvalue(X, 4, 2*(ll0-ll_uie))
printf "Homogeneity in the Fisher hypothesis:\n"
printf " LR-test, rank = 3: chi^2(4) = %6.3f [%6.4f]\n", 2*(ll0-ll_hfh), \
pvalue(X, 4, 2*(ll0-ll_hfh))
Output:
Testing zero restrictions in cointegration space:
LR-test, rank = 3: chi^2(3) = 1.4763 [0.6877]
Unit income elasticity: LR-test, rank = 3:
chi^2(4) = 17.2071 [0.0018]
Homogeneity in the Fisher hypothesis:
LR-test, rank = 3: chi^2(4) = 15.547 [0.0037]
Another type of test that is commonly performed is the weak exogeneity test. In this context, a
variable is said to be weakly exogenous if all coecients on the corresponding row in the matrix
are zero. If this is the case, that variable does not adjust to deviations from any of the long-run
equilibria and can be considered an autonomous driving force of the whole system.
The code in Example 27.3 performs this test for each variable in turn, thus replicating the rst
column of Table 6 on page 825 of Brand and Cassola (2004). The results show that weak exogeneity
might perhaps be accepted for the long-term interest rate and real GDP (p-values 0.07 and 0.08
respectively).
Identication and testability
One point regarding VECM restrictions that can be confusing at rst is that identication (does
the restriction identify the system?) and testability (is the restriction testable?) are quite separate
matters. Restrictions can be identifying but not testable; less obviously, they can be testable but
not identifying.
Example 27.3: Testing for weak exogeneity
Input:
restrict
a1 = 0
end restrict
ts_m = 2*(ll0 - $rlnl)
restrict
a2 = 0
end restrict
ts_p = 2*(ll0 - $rlnl)
restrict
a3 = 0
end restrict
ts_l = 2*(ll0 - $rlnl)
restrict
a4 = 0
end restrict
ts_s = 2*(ll0 - $rlnl)
restrict
a5 = 0
end restrict
ts_y = 2*(ll0 - $rlnl)
loop foreach i m p l s y --quiet
printf "\Delta $i\t%6.3f [%6.4f]\n", ts_$i, pvalue(X, 6, ts_$i)
endloop
Output (variable, LR test, p-value):
\Delta m 18.111 [0.0060]
\Delta p 21.067 [0.0018]
\Delta l 11.819 [0.0661]
\Delta s 16.000 [0.0138]
\Delta y 11.335 [0.0786]
This can be seen quite easily in relation to a rank-1 system. The restriction
1
= 1 is identifying
(it pins down the scale of ) but, being a pure scaling, it is not testable. On the other hand, the
restriction
1
+
2
= 0 is testablethe system with this requirement imposed will almost certainly
have a lower maximized likelihoodbut it is not identifying; it still leaves open the scale of .
We said above that the number of restrictions must equal at least r
2
, where r is the cointegrating
rank, for identication. This is a necessary and not a sucient condition. In fact, when r > 1 it can
be quite tricky to assess whether a given set of restrictions is identifying. Gretl uses the method
suggested by Doornik (1995), where identication is assessed via the rank of the information ma-
trix.
It can be shown that for restrictions of the sort (27.7) and (27.8) the information matrix has the
same rank as the Jacobian matrix
() =
_
(I
p
)G : (I
p
1
)H
_
A sucient condition for identication is that the rank of () equals the number of free para-
meters. The rank of this matrix is evaluated by examination of its singular values at a randomly
selected point in the parameter space. For practical purposes we treat this condition as if it were
both necessary and sucient; that is, we disregard the special cases where identication could be
achieved without this condition being met.
7
27.7 Numerical solution methods
In general, the ML estimator for the restricted VECM problem has no closed form solution, hence
the maximum must be found via numerical methods.
8
In some cases convergence may be dicult,
and gretl provides several choices to solve the problem.
Switching and LBFGS
Two maximization methods are available in gretl. The default is the switching algorithm set out
in Boswijk and Doornik (2004). The alternative is a limited-memory variant of the BFGS algorithm
(LBFGS), using analytical derivatives. This is invoked using the --lbfgs ag with the restrict
command.
The switching algorithm works by explicitly maximizing the likelihood at each iteration, with re-
spect to

, and

(the covariance matrix of the residuals) in turn. This method shares a feature
with the basic Johansen eigenvalues procedure, namely, it can handle a set of restrictions that does
not fully identify the parameters.
LBFGS, on the other hand, requires that the model be fully identied. When using LBFGS, therefore,
you may have to supplement the restrictions of interest with normalizations that serve to identify
the parameters. For example, one might use all or part of the Phillips normalization (see section
27.5).
Neither the switching algorithm nor LBFGS is guaranteed to nd the global ML solution.
9
The
optimizer may end up at a local maximum (or, in the case of the switching algorithm, at a saddle
point).
The solution (or lack thereof) may be sensitive to the initial value selected for . By default, gretl
selects a starting point using a deterministic method based on Boswijk (1995), but two further
7
See Boswijk and Doornik (2004), pp. 4478 for discussion of this point.
8
The exception is restrictions that are homogeneous, common to all or all (in case r > 1), and involve either
only or only. Such restrictions are handled via the modied eigenvalues method set out by Johansen (1995). We solve
directly for the ML estimator, without any need for iterative methods.
9
In developing gretls VECM-testing facilities we have considered a fair number of tricky cases from various sources.
Wed like to thank Luca Fanelli of the University of Bologna and Sven Schreiber of Goethe University Frankfurt for their
help in devising torture-tests for gretls VECM code.
options are available: the initialization may be adjusted using simulated annealing, or the user may
supply an explicit initial value for .
The default initialization method is:
1. Calculate the unrestricted ML

using the Johansen procedure.
2. If the restriction on is non-homogeneous, use the method proposed by Boswijk:
0
= [(I
r

H]
+
(I
r

h
0
(27.9)
where

= 0 and A
+
denotes the MoorePenrose inverse of A. Otherwise
0
= (H
H)
1
H
vec(
) (27.10)
3. vec(
0
) = H
0
+h
0
.
4. Calculate the unrestricted ML conditional on
0
, as per Johansen:
= S
01
0
(
0
S
11
0
)
1
(27.11)
5. If is restricted by vec(
) = G, then
0
= (G
G)
1
G
vec(
) and vec(
0
) = G
0
.
Alternative initialization methods
As mentioned above, gretl oers the option of adjusting the initialization using simulated anneal-
ing. This is invoked by adding the --jitter option to the restrict command.
The basic idea is this: we start at a certain point in the parameter space, and for each of n iterations
(currently n = 4096) we randomly select a new point within a certain radius of the previous one,
and determine the likelihood at the new point. If the likelihood is higher, we jump to the new point;
otherwise, we jump with probability P (and remain at the previous point with probability 1P). As
the iterations proceed, the system gradually coolsthat is, the radius of the random perturbation
is reduced, as is the probability of making a jump when the likelihood fails to increase.
In the course of this procedure many points in the parameter space are evaluated, starting with the
point arrived at by the deterministic method, which well call
0
. One of these points will be best
in the sense of yielding the highest likelihood: call it
. This point may or may not have a greater

likelihood than
0
. And the procedure has an end point,
n
, which may or may not be best.
The rule followed by gretl in selecting an initial value for based on simulated annealing is this: use
if
>
0
, otherwise use
n
. That is, if we get an improvement in the likelihood via annealing,
we make full use of this; on the other hand, if we fail to get an improvement we nonetheless allow
the annealing to randomize the starting point. Experiments indicated that the latter eect can be
helpful.
Besides annealing, a further alternative is manual initialization. This is done by passing a prede-
ned vector to the set command with parameter initvals, as in
set initvals myvec
The details depend on whether the switching algorithm or LBFGS is used. For the switching algo-
rithm, there are two options for specifying the initial values. The more user-friendly one (for most
people, we suppose) is to specify a matrix that contains vec() followed by vec(). For example:
open denmark.gdt
vecm 2 1 LRM LRY IBO IDE --rc --seasonals
matrix BA = {1, -1, 6, -6, -6, -0.2, 0.1, 0.02, 0.03}
set initvals BA
restrict
b[1] = 1
b[1] + b[2] = 0
b[3] + b[4] = 0
end restrict
In this examplefrom Johansen (1995) the cointegration rank is 1 and there are 4 variables.
However, the model includes a restricted constant (the --rc ag) so that has 5 elements. The
matrix has 4 elements, one per equation. So the matrix BA may be read as
(
1
,
2
,
3
,
4
,
5
,
1
,
2
,
3
,
4
)
The other option, which is compulsory when using LBFGS, is to specify the initial values in terms
of the free parameters, and . Getting this right is somewhat less obvious. As mentioned above,
the implicit-form restriction Rvec() = q has explicit form vec() = H+ h
0
, where H = R
, the
right nullspace of R. The vector is shorter, by the number of restrictions, than vec(). The
savvy user will then see what needs to be done. The other point to take into account is that if is
unrestricted, the eective length of is 0, since it is then optimal to compute using Johansens
formula, conditional on (equation 27.11 above). The example above could be rewritten as:
open denmark.gdt
vecm 2 1 LRM LRY IBO IDE --rc --seasonals
matrix phi = {-8, -6}
set initvals phi
restrict --lbfgs
b[1] = 1
b[1] + b[2] = 0
b[3] + b[4] = 0
end restrict
In this more economical formulation the initializer species only the two free parameters in (5
elements in minus 3 restrictions). There is no call to give values for since is unrestricted.
Scale removal
Consider a simpler version of the restriction discussed in the previous section, namely,
restrict
b[1] = 1
b[1] + b[2] = 0
end restrict
This restriction comprises a substantive, testable requirementthat
1
and
2
sum to zeroand
a normalization or scaling,
1
= 1. The question arises, might it be easier and more reliable to
maximize the likelihood without imposing
1
= 1?
10
If so, we could record this normalization,
remove it for the purpose of maximizing the likelihood, then reimpose it by scaling the result.
Unfortunately it is not possible to say in advance whether scale removal of this sort will give
better results, for any particular estimation problem. However, this does seem to be the case more
often than not. Gretl therefore performs scale removal where feasible, unless you
explicitly forbid this, by giving the --no-scaling option ag to the restrict command; or
provide a specic vector of initial values; or
10
As a numerical matter, that is. In principle this should make no dierence.
select the LBFGS algorithm for maximization.
Scale removal is deemed infeasible if there are any cross-column restrictions on , or any non-
homogeneous restrictions involving more than one element of .
In addition, experimentation has suggested to us that scale removal is inadvisable if the system is
just identied with the normalization(s) included, so we do not do it in that case. By just identied
we mean that the system would not be identied if any of the restrictions were removed. On that
criterion the above example is not just identied, since the removal of the second restriction would
not aect identication; and gretl would in fact perform scale removal in this case unless the user
specied otherwise.
Chapter 28
Multivariate models
By a multivariate model we mean one that includes more than one dependent variable. Certain
specic types of multivariate model for time-series data are discussed elsewhere: chapter 26 deals
with VARs and chapter 27 with VECMs. Here we discuss two general sorts of multivariate model,
implemented in gretl via the system command: SUR systems (Seemingly Unrelated Regressions),
in which all the regressors are taken to be exogenous and interest centers on the covariance of the
error term across equations; and simultaneous systems, in which some regressors are assumed to
be endogenous.
In this chapter we give an account of the syntax and use of the system command and its compan-
ions, restrict and estimate; we also explain the options and accessors available in connection
with multivariate models.
28.1 The system command
The specication of a multivariate system takes the form of a block of statements, starting with
system and ending with end system. Once a system is specied it can estimated via various
methods, using the estimate command, with or without restrictions, which may be imposed via
the restrict command.
Starting a system block
The rst line of a system block may be augmented in either (or both) of two ways:
An estimation method is specied for the system. This is done by following system with
an expression of the form method=estimator, where estimator must be one of ols (Ordinary
Least Squares), tsls (Two-Stage Least Squares), sur (Seemingly Unrelated Regressions), 3sls
(Three-Stage Least Squares), liml (Limited Information Maximum Likelihood) or fiml (Full
Information Maximum Likelihood). Two examples:
system method=sur
system method=fiml
OLS, TSLS and LIML are, of course, single-equation methods rather than true system estima-
tors; they are included to facilitate comparisons.
The system is assigned a name. This is done by giving the name rst, followed by a back-
arrow, <-, followed by system. If the name contains spaces it must be enclosed in double-
quotes. Here are two examples:
sys1 <- system
"System 1" <- system
Note, however, that this naming method is not available within a user-dened function, only
in the main body of a gretl script.
If the initial system line is augmented in the rst way, the eect is that the system is estimated as
soon as its denition is completed, using the specied method. The eect of the second option is
246
Chapter 28. Multivariate models 247
that the system can then be referenced by the assigned name for the purposes of the restrict and
estimate commands; in the gretl GUI an additional eect is that an icon for the system is added
to the Session view.
These two possibilities can be combined, as in
mysys <- system method=3sls
In this example the system is estimated immediately via Three-Stage Least Squares, and is also
available for subsequent use under the name mysys.
If the system is not named via the back-arrow mechanism, it is still available for subsequent use
via restrict and estimate; in this case you should use the generic name $system to refer to the
last-dened multivariate system.
The body of a system block
The most basic element in the body of a system block is the equation statement, which is used
to specify each equation within the system. This takes the same form as the regression specica-
tion for single-equation estimators, namely a list of series with the dependent variable given rst,
followed by the regressors, with the series given either by name or by ID number (order in the
dataset). A system block must contain at least two equation statements, and for systems without
endogenous regressors these statements are all that is required. So, for example, a minimal SUR
specication might look like this:
system method=sur
equation y1 const x1
equation y2 const x2
end system
For simultaneous systems it is necessary to determine which regressors are endogenous and which
exogenous. By default all regressors are treated as exogenous, except that any variable that appears
as the dependent variable in one equation is automatically treated as endogeous if it appears as a
regressor elsewhere. However, an explicit list of endogenous regressors may be supplied follow-
ing the equations lines: this takes the form of the keyword endog followed by the names or ID
numbers of the relevant regressors.
When estimation is via TSLS or 3SLS it is possible to specify a particular set of instruments for each
equation. This is done by giving the equation lists in the format used with the tsls command:
rst the dependent variable, then the regressors, then a semicolon, then the instruments, as in
system method=3sls
equation y1 const x11 x12 ; const x11 z1
equation y2 const x21 x22 ; const x21 z2
end system
In this case, as in tsls, any regressors that are not also listed as instruments are treated as en-
dogenous, so in the example above x11 and x21 are treated as exogenous while x21 and x22 are
endogenous, and instrumented by z1 and z2 respectively.
One more sort of statement is allowed in a system block: that is, the keyword identity followed by
an equation that denes an accounting relationship, rather then a stochastic one, between variables.
For example,
identity Y = C + I + G + X
There can be more than one identity in a system block. But note that these statements are specic
to estimation via FIML; they are ignored for other estimators.
Equation systems within functions
It is also possible to dene a multivariate systemin a programmatic way. This is useful if the precise
specication of the system depends on some input parameters that are not known in advance, but
are given when the script is actually run.
1
The relevant syntax is given by the equations keyword (note the plural), which replaces the block
of equation lines in the standard form. An equations line requires two list arguments. The rst
list must contain all series on the left-hand side of the system; thus the number of elements in this
rst list determines the number of equations in the system. The second list is a list of lists, which
is a special variant of the list data type. That is, for each equation of the system you must provide a
list of right-hand side variables, and the lists for all equations must be joined by assigning them to
another list object; in that assignment, they must be separated by a semicolon. Here is an example
for a two-equation system:
list syslist = xlist1 ; xlist2
Therefore, specifying a system generically in this way just involves building the necessary list
arguments, as shown in the following example:
open denmark
list LHS = LRM LRY
list RHS1 = const LRM(-1) IBO(-1) IDE(-1)
list RHS2 = const LRY(-1) IBO(-1)
list RHS = RHS1 ; RHS2
system method=ols
equations LHS RHS
end system
As mentioned above, the option of assigning a specic name to a system is not available within
functions, but the generic identier $system can be used to similar eect. The following example
shows how one can dene a system, estimate it via two methods, apply a restriction, then re-
estimate it subject to the restriction.
function void anonsys(series x, series y)
system
equation x const
equation y const
end system
estimate $system method=ols
estimate $system method=sur
restrict $system
b[1,1] - b[2,1] = 0
end restrict
estimate $system method=ols
end function
28.2 Restriction and estimation
The behavior of the restrict command is a little dierent for multivariate systems as compared
with single-equation models.
In the single-equation case, restrict refers to the last-estimated model, and once the command is
completed the restriction is tested. In the multivariate case, you must give the name of the system
to which the restriction is to be applied (or $system to refer to the last-dened system), and the
1
This feature was added in version 1.9.7 of gretl.
eect of the command is just to attach the restriction to the system; testing is not done until the
next estimate command is given. In addition, in the system case the default is to produce full
estimates of the restricted model; if you are not interested in the full estimates and just want the
test statistic you can append the --quiet option to estimate.
A given system restriction remains in force until it is replaced or removed. To return a system to
its unrestricted state you can give an empty restrict block, as in
restrict sysname
end restrict
As illustrated above, you can use the method tag to specify an estimation method with the estimate
command. If the system has already been estimated you can omit this tag and the previous method
is used again.
The estimate command is the main locus for options regarding the details of estimation. The
available options are as follows:
If the estimation method is SUR or 3SLS and the --iterate ag is given, the estimator will be
iterated. In the case of SUR, if the procedure converges the results are maximum likelihood
estimates. Iteration of three-stage least squares, however, does not in general converge on the
full-information maximum likelihood results. This ag is ignored for other estimators.
If the equation-by-equation estimators OLS or TSLS are chosen, the default is to apply a de-
grees of freedom correction when calculating standard errors. This can be suppressed using
the --no-df-corr ag. This ag has no eect with the other estimators; no degrees of free-
dom correction is applied in any case.
By default, the formula used in calculating the elements of the cross-equation covariance
matrix is

ij
=
u
i
u
j
T
where T is the sample size and u
i
is the vector of residuals from equation i. But if the
--geomean ag is given, a degrees of freedom correction is applied: the formula is

ij
=
u
i
u
j
_
(T k
i
)(T k
j
)
where k
i
denotes the number of independent parameters in equation i.
If an iterative method is specied, the --verbose option calls for printing of the details of
the iterations.
When the system estimator is SUR or 3SLS the cross-equation covariance matrix is initially
estimated via OLS or TSLS, respectively. In the case of a system subject to restrictions the
question arises: should the initial single-equation estimator be restricted or unrestricted?
The default is the former, but the --unrestrict-init ag can be used to select unrestricted
initialization. (Note that this is unlikely to make much dierence if the --iterate option is
given.)
28.3 System accessors
After system estimation various matrices may be retrieved for further analysis. Let g denote the
number of equations in the system and let K denote the total number of estimated parameters
(K =

i
k
i
). The accessors $uhat and $yhat get T g matrices holding the residuals and tted
values respectively. The accessor $coeff gets the stacked K-vector of parameter estimates; $vcv
gets the KK variance matrix of the parameter estimates; and $sigma gets the gg cross-equation
covariance matrix,

.
A test statistic for the hypothesis that is diagonal can be retrieved as $diagtest and its p-value
as $diagpval. This is the BreuschPagan test except when the estimator is (unrestricted) iterated
SUR, in which case its a Likelihood Ratio test. The BreuschPagan test is computed as
LM = T
g
_
i=2
i1
_
j=1
r
2
ij
where r
ij
=
ij
/
_

ii

jj
; the LR test is
LR = T
_
_
g
_
i=1
log
2
i
log
_
_
where
2
i
is u
i
u
i
/T from the individual OLS regressions. In both cases the test statistic is distrib-
uted asymptotically as
2
with g(g 1)/2 degrees of freedom.
Structural and reduced forms
Systems of simultaneous systems can be represented in structural form as
y
t
= A
1
y
t1
+A
2
y
t2
+ +A
p
y
tp
+Bx
t
+
t
where y
t
represents the vector of endogenous variables in period t, x
t
denotes the vector of ex-
ogenous variables, and p is the maximum lag of the endogenous regressors. The structural-form
matrices can be retrieved as $sysGamma, $sysA and $sysB respectively. If y
t
is m 1 and x
t
is
n1, then is mm and B is mn. If the system contains no lags of the endogenous variables
then the A matrix is not dened, otherwise A is the horizontal concatenation of A
1
, . . . , A
p
, and is
therefore mmp.
From the structural form it is straightforward to obtain the reduced form, namely,
y
t
=
1
_
_
p
_
i=1
A
i
y
ti
_
_
+
1
Bx
t
+v
t
where v
t

1
t
. The reduced form is used by gretl to generate forecasts in response to the fcast
command. This means thatin contrast to single-equation estimationthe values produced via
fcast for a static, within-sample forecast will in general dier from the tted values retrieved via
$yhat. The tted values for equation i represent the expectation of y
ti
conditional on the contem-
poraneous values of all the regressors, while the fcast values are conditional on the exogenous
and predetermined variables only.
The above account has to be qualied for the case where a system is set up for estimation via TSLS
or 3SLS using a specic list of instruments per equation, as described in section 28.1. In that case
it is possible to include more endogenous regressors than explicit equations (although, of course,
there must be sucient instruments to achieve identication). In such systems endogenous re-
gressors that have no associated explicit equation are treated as if exogenous when constructing
the structural-form matrices. This means that forecasts are conditional on the observed values of
the extra endogenous regressors rather than solely on the values of the exogenous and predeter-
mined variables.
Chapter 29
Forecasting
29.1 Introduction
In some econometric contexts forecasting is the prime objective: one wants estimates of the future
values of certain variables to reduce the uncertainty attaching to current decision making. In other
contexts where real-time forecasting is not the focus prediction may nonetheless be an important
moment in the analysis. For example, out-of-sample prediction can provide a useful check on
the validity of an econometric model. In other cases we are interested in questions of what if:
for example, how might macroeconomic outcomes have diered over a certain period if a dierent
policy had been pursued? In the latter cases prediction need not be a matter of actually projecting
into the future but in any case it involves generating tted values from a given model. The term
postdiction might be more accurate but it is not commonly used; we tend to talk of prediction
even when there is no true forecast in view.
This chapter oers an overview of the methods available within gretl for forecasting or prediction
(whether forward in time or not) and explicates some of the ner points of the relevant commands.
29.2 Saving and inspecting tted values
In the simplest case, the predictions of interest are just the (within sample) tted values from an
econometric model. For the single-equation linear model, y
t
= X
t
+u
t
, these are y
t
= X
t
.
In command-line mode, the y series can be retrieved, after estimating a model, using the accessor
$yhat, as in
series yh = $yhat
If the model in question takes the form of a system of equations, $yhat returns a matrix, each
column of which contains the tted values for a particular dependent variable. To extract the tted
series for, e.g., the dependent variable in the second equation, do
matrix Yh = $yhat
series yh2 = Yh[,2]
Having obtained a series of tted values, you can use the fcstats function to produce a vector of
statistics that characterize the accuracy of the predictions (see section 29.4 below).
The gretl GUI oers several ways of accessing and examining within-sample predictions. In the
model display window the Save menu contains an item for saving tted values, the Graphs menu
allows plotting of tted versus actual values, and the Analysis menu oers a display of actual, tted
and residual values.
29.3 The fcast command
The fcast command generates predictions based on the last estimated model. Several questions
arise here: How to control the range over which predictions are generated? How to control the
forecasting method (where a choice is available)? How to control the printing and/or saving of the
results? Basic answers can be found in the Gretl Command Reference; we add some more details
here.
251
Chapter 29. Forecasting 252
The forecast range
The range defaults to the currently dened sample range. If this remains unchanged following esti-
mation of the model in question, the forecast will be within sample and (with some qualications
noted below) it will essentially duplicate the information available via the retrieval of tted values
(see section 29.2 above).
A common situation is that a model is estimated over a given sample and then forecasts are
wanted for a subsequent out-of-sample range. The simplest way to accomplish this is via the
--out-of-sample option to fcast. For example, assuming we have a quarterly time-series dataset
containing observations from 1980:1 to 2008:4, four of which are to be reserved for forecasting:
# reserve the last 4 observations
smpl 1980:1 2007:4
ols y 0 xlist
fcast --out-of-sample
This will generate a forecast from 2008:1 to 2008:4.
There are two other ways of adjusting the forecast range, oering ner control:
Use the smpl command to adjust the sample range prior to invoking fcast.
Use the optional startobs and endobs arguments to fcast (which should come right after the
command word). These values set the forecast range independently of the sample range.
What if one wants to generate a true forecast that goes beyond the available data? In that case
one can use the dataset command with the addobs parameter to add extra observations before
forecasting. For example:
# use the entire dataset, which ends in 2008:4
ols y 0 xlist
dataset addobs 4
fcast 2009:1 2009:4
But this will work as stated only if the set of regressors in xlist does not contain any stochastic
regressors other than lags of y. The dataset addobs command attempts to detect and extrapolate
certain common deterministic variables (e.g., time trend, periodic dummy variables). In addition,
lagged values of the dependent variable can be supported via a dynamic forecast (see below for
discussion of the static/dynamic distinction). But future values of any other included regressors
must be supplied before such a forecast is possible. Note that specic values in a series can be
set directly by date, for example: x1[2009:1] = 120.5. Or, if the assumption of no change in the
regressors is warranted, one can do something like this:
loop t=2009:1..2009:4
loop foreach i xlist
$i[t] = $i[2008:4]
endloop
endloop
Static, dynamic and rolling forecasts
The distinction between static and dynamic forecasts applies only to dynamic models, i.e., those
that feature one or more lags of the dependent variable. The simplest case is the AR(1) model,
y
t
=
0
+
1
y
t1
+
t
(29.1)
In some cases the presence of a lagged dependent variable is implicit in the dynamics of the error
term, for example
y
t
= +u
t
u
t
= u
t1
+
t
which implies that
y
t
= (1 ) +y
t1
+
t
Suppose we want to forecast y for period s using a dynamic model, say (29.1) for example. If
we have data on y available for period s 1 we could form a tted value in the usual way: y
s
=

0
+
1
y
s1
. But suppose that data are available only up to s 2. In that case we can apply the
chain rule of forecasting:
y
s1
=
0
+
1
y
s2
y
s
=
0
+
1
y
s1
This is what is called a dynamic forecast. A static forecast, on the other hand, is simply a tted
value (even if it happens to be computed out-of-sample).
Printing and saving forecasts
To be written.
29.4 Univariate forecast evaluation statistics
Let y
t
be the value of a variable of interest at time t and let f
t
be a forecast of y
t
. We dene the
forecast error as e
t
= y
t
f
t
. Given a series of T observations and associated forecasts we can
construct several measures of the overall accuracy of the forecasts. Some commonly used measures
are the Mean Error (ME), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute
Error (MAE), Mean Percentage Error (MPE) and Mean Absolute Percentage Error (MAPE). These are
dened as follows.
ME =
1
T
T
_
t=1
e
t
MSE =
1
T
T
_
t=1
e
2
t
RMSE =
_
1
T
T
_
t=1
e
2
t
MAE =
1
T
T
_
t=1
e
t
MPE =
1
T
T
_
t=1
100
e
t
y
t
MAPE =
1
T
T
_
t=1
100
e
t
y
t
A further relevant statistic is Theils U (Theil, 1966), dened as the positive square root of
U
2
=
1
T
T1
_
t=1
_
f
t+1
y
t+1
y
t
_
2
_
_
1
T
T1
_
t=1
_
y
t+1
y
t
y
t
_
2
_
_
1
The more accurate the forecasts, the lower the value of Theils U, which has a minimum of 0.
1
This measure can be interpreted as the ratio of the RMSE of the proposed forecasting model to the
RMSE of a nave model which simply predicts y
t+1
= y
t
for all t. The nave model yields U = 1;
values less than 1 indicate an improvement relative to this benchmark and values greater than 1 a
deterioration.
1
This statistic is sometimes called U
2
, to distinguish it from a related but dierent U dened in an earlier work by
Theil (1961). It seems to be generally accepted that the later version of Theils U is a superior statistic, so we ignore the
earlier version here.
In addition, Theil (1966, pp. 3336) proposed a decomposition of the MSE which can be useful in
evaluating a set of forecasts. He showed that the MSE could be broken down into three non-negative
components as follows
MSE =
_
f y
_
2
+
_
s
f
rs
y
_
2
+
_
1 r
2
_
s
2
y
where

f and y are the sample means of the forecasts and the observations, s
f
and s
y
are the re-
spective standard deviations (using T in the denominator), and r is the sample correlation between
y and f. Dividing through by MSE we get
_
f y
_
2
MSE
+
_
s
f
rs
y
_
2
MSE
+
_
1 r
2
_
s
2
y
MSE
= 1 (29.2)
Theil labeled the three terms on the left-hand side of (29.2) the bias proportion (U
M
), regression
proportion (U
R
) and disturbance proportion (U
D
), respectively. If y and f represent the in-sample
observations of the dependent variable and the tted values from a linear regression then the rst
two components, U
M
and U
R
, will be zero (apart from rounding error), and the entire MSE will be
accounted for by the unsystematic part, U
D
. In the case of out-of-sample prediction, however (or
prediction over a sub-sample of the data used in the regression), U
M
and U
R
are not necessarily
close to zero, although this is a desirable property for a forecast to have. U
M
diers from zero if
and only if the mean of the forecasts diers from the mean of the realizations, and U
R
is non-zero
if and only if the slope of a simple regression of the realizations on the forecasts diers from 1.
The above-mentioned statistics are printed as part of the output of the fcast command. They can
also be retrieved in the form of a column vector using the function fcstats, which takes two series
arguments corresponding to y and f. The vector returned is
_
ME MSE MAE MPE MAPE U U
M
U
R
U
D
_
(Note that the RMSE is not included since it can easily be obtained given the MSE.) The series given
as arguments to fcstats must not contain any missing values in the currently dened sample
range; use the smpl command to adjust the range if needed.
29.5 Forecasts based on VAR models
To be written.
29.6 Forecasting from simultaneous systems
To be written.
Chapter 30
The Kalman Filter
30.1 Preamble
The Kalman lter has been used behind the scenes in gretl for quite some time, in computing
ARMA estimates. But user access to the Kalman lter is new and it has not yet been tested to any
great extent. We have run some tests of relatively simple cases against the benchmark of SsfPack
Basic. This is state-space software written by Koopman, Shephard and Doornik and documented in
Koopman, Shephard and Doornik (1999). It requires Doorniks ox program. Both ox and SsfPack
are available as free downloads for academic use but neither is open-source; see http://www.
ssfpack.com. Since Koopman is one of the leading researchers in this area, presumably the results
from SsfPack are generally reliable. To date we have been able to replicate the SsfPack results in
gretl with a high degree of precision.
We welcome both success reports and bug reports.
30.2 Notation
It seems that in econometrics everyone is happy with y = X + u, but we cant, as a community,
make up our minds on a standard notation for state-space models. Harvey (1989), Hamilton (1994),
Harvey and Proietti (2005) and Pollock (1999) all use dierent conventions. The notation used here
is based on James Hamiltons, with slight variations.
A state-space model can be written as
t+1
= F
t
t
+v
t
(30.1)
y
t
= A
t
x
t
+H
t
+w
t
(30.2)
where (30.1) is the state transition equation and (30.2) is the observation or measurement equation.
The state vector,
t
, is (r 1) and the vector of observables, y
t
, is (n 1); x
t
is a (k 1) vector of
exogenous variables. The (r 1) vector v
t
and the (n1) vector w
t
are assumed to be vector white
noise:
E(v
t
v
s
) = Q
t
for t = s, otherwise 0
E(w
t
w
s
) = R
t
for t = s, otherwise 0
The number of time-series observations will be denoted by T. In the special case when F
t
= F,
H
t
= H, A
t
= A, Q
t
= Q and R
t
= R, the model is said to be time-invariant.
The Kalman recursions
Using this notation, and assuming for the moment that v
t
and w
t
are mutually independent, the
Kalman recursions can be written as follows.
Initialization is via the unconditional mean and variance of
1
:
10
= E(
1
)
P
10
= E
_
_
1
E(
1
)
_ _
1
E(
1
)
_
_
255
Chapter 30. The Kalman Filter 256
Usually these are given by

10
= 0 and
vec(P
10
) = [I
r
2 F F]
1
vec(Q) (30.3)
but see below for further discussion of the initial variance.
Iteration then proceeds in two steps.
1
First we update the estimate of the state
t+1t
= F
t
tt1
+K
t
e
t
(30.4)
where e
t
is the prediction error for the observable:
e
t
= y
t
A
t
x
t
H
tt1
and K
t
is the gain matrix, given by
K
t
= F
t
P
tt1
H
t
1
t
(30.5)
with
t
= H
t
P
tt1
H
t
+R
t
The second step then updates the estimate of the variance of the state using
P
t+1t
= F
t
P
tt1
F
t
K
t
t
K
t
+Q
t
(30.6)
Cross-correlated disturbances
The formulation given above assumes mutual independence of the disturbances in the state and
observation equations, v
t
and w
t
. This assumption holds good in many practical applications, but
a more general formulation allows for cross-correlation. In place of (30.1)(30.2) we may write
t+1
= F
t
t
+B
t
t
y
t
= A
t
x
t
+H
t
+C
t
t
where
t
is a (p 1) disturbance vector, all the elements of which have unit variance, B
t
is (r p)
and C
t
is (np).
The no-correlation case is nested thus: dene v
t
and w
t
as modied versions of v
t
and w
t
, scaled
such that each element has unit variance, and let
t
=
_
v
t
w
t
_
so that p = r +n. Then (suppressing time subscripts for simplicity) let
B =
_

rr
.
.
.
. 0
rn
_
C =
_
0
nr
.
.
.
.
nn
_
where and are lower triangular matrices satisfying Q =
and R =
respectively. The zero

sub-matrices in the above expressions for B and C produce the case of mutual independence; this
corresponds to the condition BC
= 0.
In the general case p is not necessarily equal to r +n, and BC
may be non-zero. This means that

the Kalman gain equation (30.5) must be modied as
K
t
= (F
t
P
tt1
H
t
+B
t
C
t
)
1
t
(30.7)
Otherwise, the equations given earlier hold good, if we write BB
in place of Q and CC
in place of
R.
In the account of gretls Kalman facility below we take the uncorrelated case as the baseline, but
add remarks on how to handle the correlated case where applicable.
1
For a justication of the following formulae see the classic book by Anderson and Moore (1979) or, for a more
modern treatment, Pollock (1999) or Hamilton (1994). A transcription of R. E. Kalmans original paper (Kalman, 1960) is
available at http://www.cs.unc.edu/~welch/kalman/kalmanPaper.html.
30.3 Intended usage
The Kalman lter can be used in three ways: two of these are the classic forward and backward
pass, or ltering and smoothing respectively; the third use is simulation. In the ltering/smoothing
case you have the data y
t
and you want to reconstruct the states
t
(and the forecast errors as a by-
product), but we may also have a computational apparatus that does the reverse: given articially-
generated series w
t
and v
t
, generate the states
t
(and the observables y
t
as a by-product).
The usefulness of the classical lter is well known; the usefulness of the Kalman lter as a sim-
ulation tool may be huge too. Think for instance of Monte Carlo experiments, simulation-based
inferencesee Gourieroux and Monfort (1996) or Bayesian methods, especially in the context of
the estimation of DSGE models.
30.4 Overview of syntax
Using the Kalman lter in gretl is a two-step process. First you set up your lter, using a block
of commands starting with kalman and ending with end kalmanmuch like the gmm command.
Then you invoke the functions kfilter, ksmooth or ksimul to do the actual work. The next two
sections expand on these points.
30.5 Dening the lter
Each line within the kalman . . . end kalman block takes the form
keyword value
where keyword represents a matrix, as shown below. (An additional matrix which may be useful in
some cases is introduced later under the heading Constant term in the state transition.)
Keyword Symbol Dimensions
obsy y T n
obsymat H r n
obsx x T k
obsxmat A k n
obsvar R nn
statemat F r r
statevar Q r r
inistate

10
r 1
inivar P
10
r r
For the data matrices y and x the corresponding value may be the name of a predened matrix, the
name of a data series, or the name of a list of series.
2
For the other inputs, value may be the name of a predened matrix or, if the input in question
happens to be (11), the name of a scalar variable or a numerical constant. If the value of a
coecient matrix is given as the name of a matrix or scalar variable, the input is not hard-wired
into the Kalman structure, rather a record is made of the name of the variable and on each run
of a Kalman function (as described below) its value is re-read. It is therefore possible to write one
kalman block and then do several ltering or smoothing passes using dierent sets of coecients.
3
2
Note that the data matrices obsy and obsx have T rows. That is, the column vectors y
t
and x
t
in (30.1) and (30.2)
are in fact the transposes of the t-dated rows of the full matrices.
3
Note, however, that the dimensions of the various input matrices are dened via the initial kalman set-up and it is
an error if any of the matrices are changed in size.
An example of this technique is provided later, in the example scripts 30.1 and 30.2. This facility
to alter the values of the coecients between runs of the lter is to be distinguished from the case
of time-varying matrices, which is discussed below.
Not all of the above-mentioned inputs need be specied in every case; some are optional. (In
addition, you can specify the matrices in any order.) The mandatory elements are y, H, F and Q, so
the minimal kalman block looks like this:
kalman
obsy y
obsymat H
statemat F
statevar Q
end kalman
The optional matrices are listed below, along with the implication of omitting the given matrix.
Keyword If omitted. . .
obsx no exogenous variables in observation equation
obsxmat no exogenous variables in observation equation
obsvar no disturbance term in observation equation
inistate

10
is set to a zero vector
inivar P
10
is set automatically
It might appear that the obsx (x) and obsxmat (A) matrices must go togethereither both are
given or neither is given. But an exception is granted for convenience. If the observation equation
includes a constant but no additional exogenous variables, you can give a (1n) value for A without
having to specify obsx. More generally, if the row dimension of A is 1 greater than the column
dimension of x, it is assumed that the rst element of A is associated with an implicit column of
1s.
Regarding the automatic initialization of P
10
(in case no inivar input is given): by default this
is done as in equation (30.3). However, this method is applicable only if all the eigenvalues of F
lie inside the unit circle. If this condition is not satised we instead apply a diuse prior, setting
P
10
= I
r
with = 10
7
. If you wish to impose this diuse prior from the outset, append the option
ag --diffuse to the end kalman statement.
4
Time-varying matrices
Any or all of the matrices obsymat, obsxmat, obsvar, statemat and statevar may be time-
varying. In that case the value corresponding to the matrix keyword should be given in a special
form: the name of an existing matrix plus a function call which modies that matrix, separated by
a semicolon. Note that in this case you must use a matrix variable, even if the matrix in question
happens to be 1 1.
For example, suppose the matrix H is time-varying. Then we might write
obsymat H ; modify_H(&H, theta)
where modify_H is a user-dened function which modies matrix H (and theta is a suitable addi-
tional argument to that function, if required).
4
Initialization of the Kalman lter outside of the case where equation (30.3) applies has been the subject of much
discussion in the literaturesee for example de Jong (1991), Koopman (1997). At present gretl does not implement any
of the more elaborate proposals that have been made.
The above is just an illustration: the matrix argument does not have to come rst, and the function
can have as many arguments as you like. The essential point is that the function must modify the
specied matrix, which requires that it be given as an argument in pointer form (preceded by &).
The function need not return any value directly; if it does, that value is ignored.
Such matrix-modifying functions will be called at each time-step of the lter operation, prior to
performing any calculations. They have access to the current time-step of the Kalman lter via the
internal variable $kalman_t, which has value 1 on the rst step, 2 on the second, and so on, up
to step T. They also have access to the previous n-vector of forecast errors, e
t1
, under the name
$kalman_uhat. When t = 1 this will be a zero vector.
Correlated disturbances
Dening a lter in which the disturbances v
t
and w
t
are correlated involves one modication to the
account given above. If you append the --cross option ag to the end kalman statement, then
the matrices corresponding to the keywords statevar and obsvar are interpreted not as Q and R
but rather as B and C as discussed in section 30.2. Gretl then computes Q = BB
and R = CC
as
well as the cross-product BC
and utilizes the modied expression for the gain as given in equation
(30.7). As mentioned above, B should be (r p) and C should be (np), where p is the number of
elements in the combined disturbance vector
t
.
Constant term in the state transition
In some applications it is useful to be able to represent a constant term in the state transition
equation explicitly; that is, equation (30.1) becomes
t+1
= +F
t
t
+v
t
(30.8)
This is never strictly necessary; the system (30.1) and (30.2) is general enough to accommodate
such a term, by absorbing it as an extra (unvarying) element in the state vector. But this comes
at the cost of expanding all the matrices that touch the state (, F, v, Q, H), making the model
relatively awkward to formulate and forecasts relatively expensive to compute.
As a simple illustration, consider a univariate model in which the state, s
t
, is just a random walk
with drift and the observed variable, y
t
, is the state plus white noise:
s
t+1
= +s
t
+v
t
(30.9)
y
t
= s
t
+w
t
(30.10)
Putting this into the standard form of (30.1) and (30.2) we get:
_
s
t+1
_
=
_
1 1
0 1
__
s
t
_
+
_
v
t
0
_
, Q =
_

2
v
0
0 0
_
y
t
=
_
1 0
_
_
s
t
_
+w
t
In such a simple case the notational and computational burden is not very great; nonetheless it is
clearly more natural to express this system in the form of (30.9) and (30.10) and in a multivariate
model the gain in parsimony could be substantial.
For this reason we support the use of an additional named matrix in the kalman setup, namely
stconst. This corresponds to in equation (30.8); it should be an r 1 vector (or if r = 1 may be
given as the name of a scalar variable). The use of stconst in setting up a lter corresponding to
(30.9) and (30.10) is shown below.
matrix H = {1}
matrix R = {1}
matrix F = {1}
matrix Q = {1}
matrix mu = {0.05}
kalman
obsy y
obsymat H
obsvar R
statemat F
statevar Q
stconst mu
end kalman
Handling of missing values
It is acceptable for the data matrices, obsy and obsx, to contain missing values. In this case the
ltering operation will work around the missing values, and the ksmooth function can be used to
obtain estimates of these values. However, there are two points to note.
First, gretls default behavior is to skip missing observations when constructing matrices from data
series. To change this, use the set command thus:
set skip_missing off
Second, the handling of missing values is not yet quite right for the case where the observable
vector y
t
contains more than one element. At present, if any of the elements of y
t
are missing
the entire observation is ignored. Clearly it should be possible to make use of any non-missing
elements, and this is not very dicult in principle, its just awkward and is not implemented yet.
Persistence and identity of the lter
At present there is no facility to create a named lter. Only one lter can exist at any point
in time, namely the one created by the last kalman block.
5
If a lter is already dened, and you
give a new kalman block, the old lter is over-written. Otherwise the existing lter persists (and
remains available for the kfilter, ksmooth and ksimul functions) until either (a) the gretl session
is terminated or (b) the command delete kalman is given.
30.6 The kfilter function
Once a lter is established, as discussed in the previous section, kfilter can be used to run a
forward, forecasting pass. This function returns a scalar code: 0 for successful completion, or 1
if numerical problems were encountered. On successful completion, two scalar accessor variables
become available: $kalman_lnl, which gives the overall log-likelihood under the joint normality
assumption,
=
1
2
_
_
nT log(2) +
T
_
t=1
log
t
+
T
_
t=1
e
1
t
e
t
_
_
and $kalman_s2, which gives the estimated variance,

2
=
1
nT
T
_
t=1
e
1
t
e
t
5
This is not quite true: more precisely, there can be no more than one Kalman lter at each level of function execution.
That is, if a gretl script creates a Kalman lter, a user-dened function called from that script may also create a lter,
without interfering with the original one.
(but see below for modications to these formulae for the case of a diuse prior). In addition the
accessor $kalman_llt gives a (T 1) vector, element t of which is
t
=
1
2
_
nlog(2) +log
t
+e
1
t
e
t
_
The kfilter function does not require any arguments, but up to ve matrix quantities may be
retrieved via optional pointer arguments. Each of these matrices has T rows, one for each time-
step; the contents of the rows are shown in the following listing.
1. Forecast errors for the observable variables: e
t
, n columns.
2. Variance matrix for the forecast errors: vech(
t
)
, n(n+1)/2 columns.
3. Estimate of the state vector:

tt1
, r columns.
4. MSE of estimate of the state vector: vech(P
tt1
)
, r(r +1)/2 columns.

5. Kalman gain: vec(K
t
)
, rn columns.
Unwanted trailing arguments can be omitted, otherwise unwanted arguments can be skipped by
using the keyword null. For example, the following call retrieves the forecast errors in the matrix
E and the estimate of the state vector in S:
matrix E S
kfilter(&E, null, &S)
Matrices given as pointer arguments do not have to be correctly dimensioned in advance; they will
be resized to receive the specied content.
Further note: in general, the arguments to kfilter should all be matrix-pointers, but under two
conditions you can give a pointer to a series variable instead. The conditions are: (i) the matrix
in question has just one column in context (for example, the rst two matrices will have a single
column if the length of the observables vector, n, equals 1) and (ii) the time-series length of the
lter is equal to the current gretl sample size.
Likelihood under the diuse prior
There seems to be general agreement in the literature that the log-likelihood calculation should
be modied in the case of a diuse prior for P
10
. However, it is not clear to us that there is a
well-dened correct method for this. At present we emulate SsfPack (see Koopman et al. (1999)
and section 30.1). In case P
10
= I
r
, we set d = r and calculate
=
1
2
_
_
(nT d) log(2) +
T
_
t=1
log
t
+
T
_
t=1
e
1
t
e
t
dlog()
_
_
and

2
=
1
nT d
T
_
t=1
e
1
t
e
t
30.7 The ksmooth function
This function returns the (T r) matrix of smoothed estimates of the state vectorthat is, esti-
mates based on all T observations: row t of this matrix holds

tT
. This function has no required
arguments but it oers one optional matrix-pointer argument, which retrieves the variance of the
smoothed state estimate, P
tT
. The latter matrix is (T r(r +1)/2); each row is in transposed vech
form. Examples:
matrix S = ksmooth() # smoothed state only
matrix P
S = ksmooth(&P) # the variance is wanted
These values are computed via a backward pass of the lter, from t = T to t = 1, as follows:
L
t
= F
t
K
t
H
t
u
t1
= H
t
1
t
e
t
+L
t
u
t
U
t1
= H
t
1
t
H
t
+L
t
U
t
L
t
tT
=

tt1
+P
tt1
u
t1
P
tT
= P
tt1
P
tt1
U
t1
P
tt1
with initial values u
T
= 0 and U
T
= 0.
6
This iteration is preceded by a special forward pass in which the matrices K
t
,
1
t
,

tt1
and P
tt1
are stored for all t. If F is time-varying, its values for all t are stored on the forward pass, and
similarly for H.
30.8 The ksimul function
This simulation function takes up to three arguments. The rst, mandatory, argument is a (T r)
matrix containing articial disturbances for the state transition equation: row t of this matrix
represents v
t
. If the current lter has a non-null R (obsvar) matrix, then the second argument
should be a (T n) matrix containing articial disturbances for the observation equation, on the
same pattern. Otherwise the second argument should be given as null. If r = 1 you may give a
series for the rst argument, and if n = 1 a series is acceptable for the second argument.
Provided that the current lter does not include exogenous variables in the observation equation
(obsx), the T for simulation need not equal that dened by the original obsy data matrix: in eect T
is temporarily redened by the row dimension of the rst argument to ksimul. Once the simulation
is completed, the T value associated with the original data is restored.
The value returned by ksimul is a (T n) matrix holding simulated values for the observables at
each time step. A third optional matrix-pointer argument allows you to retrieve a (T r) matrix
holding the simulated state vector. Examples:
matrix Y = ksimul(V) # obsvar is null
Y = ksimul(V, W) # obsvar is non-null
matrix S
Y = ksimul(V, null, &S) # the simulated state is wanted
The initial value
1
is calculated thus: we nd the matrix T such that TT
= P
10
(as given by the
inivar element in the kalman block), multiply it into v
1
, and add the result to
10
(as given by
inistate).
If the disturbances are correlated across the two equations the arguments to ksimul must be
revised: the rst argument should be a (T p) matrix, each row of which represents
t
(see sec-
tion 30.2), and the second argument should be given as null.
30.9 Example 1: ARMA estimation
As is well known, the Kalman lter provides a very ecient way to compute the likelihood of ARMA
models; as an example, take an ARMA(1,1) model
y
t
= y
t1
+
t
+
t1
6
See I. Karibzhanovs exposition at http://www.econ.umn.edu/~karib003/help/kalcvs.htm.
One of the ways the above equation can be cast in state-space form is by dening a latent process
t
= (1 L)
1
t
. The observation equation corresponding to (30.2) is then
y
t
=
t
+
t1
(30.11)
and the state transition equation corresponding to (30.1) is
_

t
t1
_
=
_
0
1 0
__

t1
t2
_
+
_

t
0
_
The gretl syntax for a corresponding kalman block would be
matrix H = {1; theta}
matrix F = {phi, 0; 1, 0}
matrix Q = {s^2, 0; 0, 0}
kalman
obsy y
obsymat H
statemat F
statevar Q
end kalman
Note that the observation equation (30.11) does not include an error term; this is equivalent
to saying that V(w
t
) = 0 and, as a consequence, the kalman block does not include an obsvar
keyword.
Once the lter is set up, all it takes to compute the log-likelihood for given values of , and
2
is to execute the kfilter() function and use the $kalman_lnl accessor (which returns the
total log-likelihood) or, more appropriately if the likelihood has to be maximized through mle, the
$kalman_llt accessor, which returns the series of individual contribution to the log-likelihood for
each observation. An example is shown in script 30.1.
30.10 Example 2: local level model
Suppose we have a series y
t
=
t
+
t
, where
t
is a random walk with normal increments of
variance
2
1
and
t
is a normal white noise with variance
2
2
, independent of
t
. This is known as
the local level model in Harveys (1989) terminology, and it can be cast in state-space form as
equations (30.1)-(30.2) with F = 1, v
t
N(0,
2
1
), H = 1 and w
t
N(0,
2
2
). The translation to a
kalman block is
kalman
obsy y
obsymat 1
statemat 1
statevar s2
obsvar s1
end kalman --diffuse
The two unknown parameters
2
1
and
2
2
can be estimated via maximum likelihood. Script 30.2
provides an example of simulation and estimation of such a model. For the sake of brevity, simu-
lation is carried out via ordinary gretl commands, rather than the state-space apparatus described
above.
The example contains two functions: the rst one carries out the estimation of the unknown pa-
rameters
2
1
and
2
2
via maximum likelihood; the second one uses these estimates to compute a
smoothed estimate of the unobservable series
t
calles muhat. A plot of
t
and its estimate is
presented in Figure 30.1.
Example 30.1: ARMA estimation
function void arma11_via_kalman(series y)
/* parameter initalization */
phi = 0
theta = 0
sigma = 1
/* Kalman filter setup */
matrix H = {1; theta}
matrix F = {phi, 0; 1, 0}
matrix Q = {sigma^2, 0; 0, 0}
kalman
obsy y
obsymat H
statemat F
statevar Q
end kalman
/* maximum likelihood estimation */
mle logl = ERR ? NA : $kalman_llt
H[2] = theta
F[1,1] = phi
Q[1,1] = sigma^2
ERR = kfilter()
params phi theta sigma
end mle -h
end function
# ------------------------ main ---------------------------
open arma.gdt # open the "arma" example dataset
arma11_via_kalman(y) # estimate an arma(1,1) model
arma 1 1 ; y --nc # check via native command
Example 30.2: Local level model
function matrix local_level (series y)
/* starting values */
scalar s1 = 1
scalar s2 = 1
/* Kalman filter set-up */
kalman
obsy y
obsymat 1
statemat 1
statevar s2
obsvar s1
/* ML estimation */
mle ll = ERR ? NA : $kalman_llt
ERR = kfilter()
params s1 s2
end mle
return s1 ~ s2
end function
function series loclev_sm (series y, scalar s1, scalar s2)
/* return the smoothed estimate of \mu_t */
kalman
obsy y
obsymat 1
statemat 1
statevar s2
obsvar s1
series ret = ksmooth()
return ret
end function
/* -------------------- main script -------------------- */
nulldata 200
set seed 202020
setobs 1 1 --special
true_s1 = 0.25
true_s2 = 0.5
v = normal() * sqrt(true_s1)
w = normal() * sqrt(true_s2)
mu = 2 + cum(w)
y = mu + v
matrix Vars = local_level(y) # estimate the variances
muhat = loclev_sm(y, Vars[1], Vars[2]) # compute the smoothed state
-8
-6
-4
-2
0
2
4
6
8
10
0 50 100 150 200
mu
muhat
Figure 30.1: Local level model:
t
and its smoothed estimate
By appending the following code snippet to the example in Table 30.2, one may check the results
against the R command StructTS.
foreign language=R --send-data
y <- gretldata[,"y"]
a <- StructTS(y, type="level")
a
StateFromR <- as.ts(tsSmooth(a))
gretl.export(StateFromR)
end foreign
append @dotdir/StateFromR.csv
ols muhat 0 StateFromR --simple
Chapter 31
Numerical methods
Several functions are available to aid in the construction of special-purpose estimators: one group
of functions are used to maximize user-supplied functions via numerical methods: BFGS, Newton
Raphson and Simulated Annealing. Another relevant function is fdjac, which produces a forward-
dierence approximation to the Jacobian.
31.1 BFGS
The BFGSmax function has two required arguments: a vector holding the initial values of a set of
parameters, and a call to a function that calculates the (scalar) criterion to be maximized, given
the current parameter values and any other relevant data. If the object is in fact minimization, this
function should return the negative of the criterion. On successful completion, BFGSmax returns the
maximized value of the criterion and the matrix given via the rst argument holds the parameter
values which produce the maximum. It is assumed here that the objective function is a user-dened
function (see Chapter 13) with the following general set-up:
function scalar ObjFunc (matrix *theta, matrix *X)
scalar val = ... # do some computation
return val
end function
The rst argument contains the arguments upon which the maximization has to take place, while
the second argument may be used to hold extra values that are necessary to compute the objective
function, but are not the variables of the optimization problem. For example, if the objective
function were a loglikelihood, the rst argument would contain the parameters and the second one
the data. Or, for more economic-theory inclined readers, if the objective function were the utility
of a consumer, the rst argument would contain the quantities of goods and the second one their
prices and disposable income.
Example 31.1: Finding the minimum of the Rosenbrock function
function scalar Rosenbrock(matrix *param)
scalar x = param[1]
scalar y = param[2]
return -(1-x)^2 - 100 * (y - x^2)^2
end function
matrix theta = { 0, 0 }
set max_verbose 1
M = BFGSmax(theta, Rosenbrock(&theta))
print theta
267
Chapter 31. Numerical methods 268
The operation of BFGS can be adjusted using the set variables bfgs_maxiter and bfgs_toler
(see Chapter 21). In addition you can provoke verbose output from the maximizer by assigning a
positive value to max_verbose, again via the set command.
The Rosenbrock function is often used as a test problem for optimization algorithms. It is also
known as Rosenbrocks Valley or Rosenbrocks Banana Function, on account of the fact that its
contour lines are banana-shaped. It is dened by:
f(x, y) = (1 x)
2
+100(y x
2
)
2
The function has a global minimum at (x, y) = (1, 1) where f(x, y) = 0. Example 31.1 shows a
gretl script that discovers the minimum using BFGSmax (giving a verbose account of progress). Note
that, in this particular case, the function to be maximized only depends on the parameters, so the
second parameter is omitted from the denition of the function Rosenbrock.
Limited-memory variant
See Byrd et al. (1995) (. . . FIXME: expand a little here . . . )
Supplying analytical derivatives for BFGS
An optional third argument to the BFGSmax function enables the user to supply analytical deriva-
tives of the criterion function with respect to the parameters (without which a numerical approxi-
mation to the gradient is computed). This argument is similar to the second one in that it species
a function call. In this case the function that is called must have the following signature.
Its rst argument should be a pre-dened matrix correctly dimensioned to hold the gradient; that
is, if the parameter vector contains k elements, the gradient matrix must also be a k-vector. This
matrix argument must be given in pointer form so that its content can be modied by the func-
tion. (Note that unlike the parameter vector, where the choice of initial values can be important,
the initial values given to the gradient are immaterial and do not aect the results.)
In addition the gradient function must have as one of its argument the parameter vector. This may
be given in pointer form (which enhances eciency) but that is not required. Additional arguments
may be specied if necessary.
Given the current parameter values, the function call must ll out the gradient vector appropriately.
It is not required that the gradient function returns any value directly; if it does, that value is
ignored.
Example 31.2 illustrates, showing how the Rosenbrock script can be modied to use analytical
derivatives. (Note that since this is a minimization problem the values written into g[1] and g[2]
in the function Rosen_grad are in fact the derivatives of the negative of the Rosenbrock function.)
31.2 NewtonRaphson
BFGS, discussed above, is an excellent all-purpose maximizer, and about as robust as possible given
the limitations of digital computer arithmetic. The NewtonRaphson maximizer is not as robust,
but may converge much faster than BFGS for problems where the maximand is reasonably well
behavedin particular, where it is anything like quadratic (see below). The case for using Newton
Raphson is enhanced if it is possible to supply a function to calculate the Hessian analytically.
The gretl function NRmax, which implements the NewtonRaphson method, has a maximum of four
arguments. The rst two (required) arguments are exactly as for BFGS: an initial parameter vector,
and a function call which returns the maximand given the parameters. The (optional) third argu-
ment is again as in BFGS: a function call that calculates the gradient. Specic to NRmax is an optional
fourth argument, namely a function call to calculate the (negative) Hessian. The rst argument of
this function must be a pre-dened matrix of the right dimension to hold the Hessianthat is, a
k k matrix, where k is the length of the parameter vectorgiven in pointer form. The second
Example 31.2: Rosenbrock function with analytical gradient
function scalar Rosenbrock (matrix *param)
scalar x = param[1]
scalar y = param[2]
return -(1-x)^2 - 100 * (y - x^2)^2
end function
function void Rosen_grad (matrix *g, matrix *param)
scalar x = param[1]
scalar y = param[2]
g[1] = 2*(1-x) + 2*x*(200*(y-x^2))
g[2] = -200*(y - x^2)
end function
matrix grad = { 0, 0 }
set max_verbose 1
M = BFGSmax(theta, Rosenbrock(&theta), Rosen_grad(&grad, &theta))
print theta
print grad
argument should be the parameter vector (optionally in pointer form). Other data may be passed
as additional arguments as needed. Similarly to the case with the gradient, if the fourth argument
to NRmax is omitted then a numerical approximation to the Hessian is constructed.
What is ultimately required in NewtonRaphson is the negative inverse of the Hessian. Note that
if you give the optional fourth argument, your function should compute the negative Hessian, but
should not invert it; NRmax takes care of inversion, with special handling for the case where the
matrix is not negative denite, which can happen far from the maximum.
Script 31.3 extends the Rosenbrock example, using NRmax with a function Rosen_hess to compute
the Hessian. The functions Rosenbrock and Rosen_grad are just the same as in Example 31.2 and
are omitted for brevity.
The idea behind NewtonRaphson is to exploit a quadratic approximation to the maximand, under
the assumption that it is concave. If this is true, the method is very eective. However, if the
algorithm happens to evaluate the function at a point where the Hessian is not negative denite,
things may go wrong. Script 31.4 exemplies this by using a normal density, which is concave in the
interval (1, 1) and convex elsewhere. If the algorithmis started fromwithin the interval everything
goes well and NR is (slightly) more eective than BFGS. If, however, the Hessian is positive at the
starting point BFGS converges with only little more diculty, while NewtonRaphson fails.
31.3 Simulated Annealing
Simulated annealingas implemented by the gretl function simannis not a full-blown maxi-
mization method in its own right, but can be a useful auxiliary tool in problems where convergence
depends sensitively on the initial values of the parameters. The idea is that you supply initial values
and the simulated annealing mechanism tries to improve on them via controlled randomization.
The simann function takes up to three arguments. The rst two (required) are the same as for
BFGSmax and NRmax: an initial parameter vector and a function that computes the maximand. The
Example 31.3: Rosenbrock function via NewtonRaphson
function void Rosen_hess (matrix *H, matrix *param)
scalar x = param[1]
scalar y = param[2]
H[1,1] = 2 - 400*y + 1200*x^2
H[1,2] = -400*x
H[2,1] = -400*x
H[2,2] = 200
end function
matrix g = { 0, 0 }
matrix H = zeros(2, 2)
set max_verbose 1
M = NRmax(theta, Rosenbrock(&theta), Rosen_grad(&g, &theta),
Rosen_hess(&H, &theta))
print theta
print grad
Example 31.4: Maximization of a Gaussian density
function scalar ND(matrix x)
scalar z = x[1]
return exp(-0.5*z*z)
end function
set max_verbose 1
x = {0.75}
A = BFGSmax(x, ND(x))
x = {0.75}
A = NRmax(x, ND(x))
x = {1.5}
A = BFGSmax(x, ND(x))
x = {1.5}
A = NRmax(x, ND(x))
optional third argument is a positive integer giving the maximum number of iterations, n, which
defaults to 1024.
Starting from the specied point in the parameter space, for each of n iterations we select at
random a new point within a certain radius of the previous one and determine the value of the
criterion at the new point. If the criterion is higher we jump to the new point; otherwise, we jump
with probability P (and remain at the previous point with probability 1 P). As the iterations
proceed, the system gradually coolsthat is, the radius of the random perturbation is reduced,
as is the probability of making a jump when the criterion fails to increase.
In the course of this procedure n+1 points in the parameter space are evaluated: call them
i
, i =
0, . . . , n, where
0
is the initial value given by the user. Let
denote the best point among
1
, . . . ,
n
(highest criterion value). The value written into the parameter vector on completion is
then
if
is better than
0
, otherwise
n
. In other words, failing an actual improvement in
the criterion, simann randomizes the starting point, which may be helpful in tricky optimization
problems.
Example 31.5 shows simann at work as a helper for BFGSmax in nding the maximum of a bimodal
function. Unaided, BFGSmax requires 60 function evaluations and 55 evaluations of the gradient,
while after simulated annealing the maximum is found with 7 function evaluations and 6 evalua-
tions of the gradient.
1
Example 31.5: BFGS with initialization via Simulated Annealing
function scalar bimodal (matrix x, matrix A)
scalar ret = exp(-qform((x-1), A))
ret += 2*exp(-qform((x+4), A))
return ret
end function
set seed 12334
set max_verbose on
scalar k = 2
matrix A = 0.1 * I(k)
matrix x0 = {3; -5}
x = x0
u = BFGSmax(x, bimodal(x, A))
print x
x = x0
u = simann(x, bimodal(x, A), 1000)
print x
u = BFGSmax(x, bimodal(x, A))
print x
31.4 Computing a Jacobian
Gretl oers the possibility of dierentiating numerically a user-dened function via the fdjac
function.
This function takes two arguments: an n1 matrix holding initial parameter values and a function
1
Your mileage may vary: these gures are somewhat compiler- and machine-dependent.
call that calculates and returns an m1 matrix, given the current parameter values and any other
relevant data. On successful completion it returns an m n matrix holding the Jacobian. For
example,
matrix Jac = fdjac(theta, SumOC(&theta, &X))
where we assume that SumOC is a user-dened function with the following structure:
function matrix SumOC (matrix *theta, matrix *X)
matrix V = ... # do some computation
return V
end function
This may come in handy in several cases: for example, if you use BFGSmax to estimate a model, you
may wish to calculate a numerical approximation to the relevant Jacobian to construct a covariance
matrix for your estimates.
Another example is the delta method: if you have a consistent estimator of a vector of parameters
, and a consistent estimate of its covariance matrix , you may need to compute estimates for a
nonlinear continuous transformation = g(). In this case, a standard result in asymptotic theory
is that
_
_
_
T
_

_
d
N(0, )
_
_
_
=
_
_
_
= g(
)
p
= g()
T
_

_
d
N(0, JJ
)
_
_
_
where T is the sample size and J is the Jacobian
g(x)
x
x=
.
Script 31.6 exemplies such a case: the example is taken from Greene (2003), section 9.3.1. The
slight dierences between the results reported in the original source and what gretl returns are due
to the fact that the Jacobian is computed numerically, rather than analytically as in the book.
On the subject of numerical versus analytical derivatives, one may wonder what dierence it makes
to use one method or another. Simply put, the answer is: analytical derivatives may be painful to
derive and to translate into code, but in most cases they are much faster than using fdjac; as a
consequence, if you need to use derivatives as part of an algorithm that requires iteration (such
as numerical optimization, or a Monte Carlo experiment), youll denitely want to use analytical
derivatives.
Analytical derivatives are also, in most cases, more precise than numerical ones, but this advantage
may or may not be negligible in practice depending on the practical details: the two fundamental
aspects to take in consideration are nonlinearity and machine precision.
As an example, consider the derivative of a highly nonlinear function such as the matrix inverse. In
order to keep the example simple, lets focus on 2 2 matrices and dene the function
function matrix vecinv(matrix x)
A = mshape(x,2,2)
return vec(inv(A))
end function
which, given vec(A), returns vec(A
1
). As is well known (see for example Magnus and Neudecker
(1988)),
vec(A
1
)
vec(A)
= (A
1
)
(A
1
),
which is rather easy to code in hansl as
function matrix grad(matrix x)
iA = inv(mshape(x,2,2))
return -iA ** iA
end function
Example 31.6: Delta Method
function matrix MPC(matrix *param, matrix *Y)
beta = param[2]
gamma = param[3]
y = Y[1]
return beta*gamma*y^(gamma-1)
end function
# William Greene, Econometric Analysis, 5e, Chapter 9
set echo off
set messages off
open greene5_1.gdt
# Use OLS to initialize the parameters
ols realcons 0 realdpi --quiet
genr a = $coeff(0)
genr b = $coeff(realdpi)
genr g = 1.0
# Run NLS with analytical derivatives
nls realcons = a + b * (realdpi^g)
deriv a = 1
deriv b = realdpi^g
deriv g = b * realdpi^g * log(realdpi)
end nls
matrix Y = realdpi[2000:4]
matrix theta = $coeff
matrix V = $vcv
mpc = MPC(&theta, &Y)
matrix Jac = fdjac(theta, MPC(&theta, &Y))
Sigma = qform(Jac, V)
printf "\nmpc = %g, std.err = %g\n", mpc, sqrt(Sigma)
scalar teststat = (mpc-1)/sqrt(Sigma)
printf "\nTest for MPC = 1: %g (p-value = %g)\n", \
teststat, pvalue(n,abs(teststat))
Using the fdjac function to obtain the same result is even easier: you just invoke it like
fdjac(a, "vecinv(a)")
In order to see what the dierence is, in terms of precision, between analytical and numerical
Jacobians, start from A =
_
2 1
1 1
_
; the following code
a = {2; 1; 1; 1}
ia = vecinv(a)
ag = grad(a)
ng = fdjac(a, "vecinv(a)")
dg = ag - ng
print ag ng dg
gives
ag (4 x 4)
-1 1 1 -1
1 -2 -1 2
1 -1 -2 2
-1 2 2 -4
ng (4 x 4)
-1 1 1 -1
1 -2 -1 2
1 -1 -2 2
-1 2 2 -4
dg (4 x 4)
-3.3530e-08 -3.7251e-08 -3.7251e-08 -3.7255e-08
2.6079e-08 5.2150e-08 3.7251e-08 6.7060e-08
2.6079e-08 3.7251e-08 5.2150e-08 6.7060e-08
-2.2354e-08 -5.9600e-08 -5.9600e-08 -1.4902e-07
in which the analytically-computed derivative and its numerical approximation are essentially the
same. If, however, you set A =
_
1.0001 1
1 1
_
you end up evaluating the function at a point in
which the function itself is considerably more nonlinear since the matrix is much closer to being
singular. As a consequence, the numerical approximation becomes much less satisfactory:
ag (4 x 4)
-1.0000e+08 1.0000e+08 1.0000e+08 -1.0000e+08
1.0000e+08 -1.0001e+08 -1.0000e+08 1.0001e+08
1.0000e+08 -1.0000e+08 -1.0001e+08 1.0001e+08
-1.0000e+08 1.0001e+08 1.0001e+08 -1.0002e+08
ng (4 x 4)
-9.9985e+07 1.0001e+08 1.0001e+08 -9.9985e+07
9.9985e+07 -1.0002e+08 -1.0001e+08 9.9995e+07
9.9985e+07 -1.0001e+08 -1.0002e+08 9.9995e+07
-9.9985e+07 1.0002e+08 1.0002e+08 -1.0001e+08
dg (4 x 4)
-14899. -14901. -14901. -14900.
14899. 14903. 14901. 14902.
14899. 14901. 14903. 14902.
-14899. -14903. -14903. -14903.
Moreover, machine precision may have its impact: if you take A = 0.00001
_
2 1
1 1
_
, the matrix
itself is not singular at all, but the order of magnitude of its elements is close enough to machine
precision to provoke problems:
ag (4 x 4)
-1.0000e+10 1.0000e+10 1.0000e+10 -1.0000e+10
1.0000e+10 -2.0000e+10 -1.0000e+10 2.0000e+10
1.0000e+10 -1.0000e+10 -2.0000e+10 2.0000e+10
-1.0000e+10 2.0000e+10 2.0000e+10 -4.0000e+10
ng (4 x 4)
-1.0000e+10 1.0000e+10 1.0000e+10 -1.0000e+10
1.0000e+10 -2.0000e+10 -1.0000e+10 2.0000e+10
1.0000e+10 -1.0000e+10 -2.0000e+10 2.0000e+10
-1.0000e+10 2.0000e+10 2.0000e+10 -4.0000e+10
dg (4 x 4)
-488.30 -390.60 -390.60 -195.33
634.79 781.21 390.60 585.98
634.79 488.26 683.55 585.98
-781.27 -976.52 -781.21 -1367.3
Chapter 32
Discrete and censored dependent variables
This chapter deals with models for dependent variables that are discrete or censored or otherwise
limited (as in event counts or durations, which must be positive) and that therefore call for estima-
tion methods other than the classical linear model. We discuss several estimators (mostly based on
the Maximum Likelihood principle), adding some details and examples to complement the material
on these methods in the Gretl Command Reference.
32.1 Logit and probit models
It often happens that one wants to specify and estimate a model in which the dependent variable
is not continuous, but discrete. A typical example is a model in which the dependent variable is
the occupational status of an individual (1 = employed, 0 = unemployed). A convenient way of
formalizing this situation is to consider the variable y
i
as a Bernoulli random variable and analyze
its distribution conditional on the explanatory variables x
i
. That is,
y
i
=
_
1 P
i
0 1 P
i
(32.1)
where P
i
= P(y
i
= 1x
i
) is a given function of the explanatory variables x
i
.
In most cases, the function P
i
is a cumulative distribution function F, applied to a linear combi-
nation of the x
i
s. In the probit model, the normal cdf is used, while the logit model employs the
logistic function (). Therefore, we have
probit P
i
= F(z
i
) = (z
i
) (32.2)
logit P
i
= F(z
i
) = (z
i
) =
1
1 +e
z
i
(32.3)
z
i
=
k
_
j=1
x
ij
j
(32.4)
where z
i
is commonly known as the index function. Note that in this case the coecients
j
cannot
be interpreted as the partial derivatives of E(y
i
x
i
) with respect to x
ij
. However, for a given value
of x
i
it is possible to compute the vector of slopes, that is
slope
j
( x) =
F(z)
x
j
z= z
Gretl automatically computes the slopes, setting each explanatory variable at its sample mean.
Another, equivalent way of thinking about this model is in terms of an unobserved variable y
i
which can be described thus:
y
i
=
k
_
j=1
x
ij
j
+
i
= z
i
+
i
(32.5)
We observe y
i
= 1 whenever y
i
> 0 and y
i
= 0 otherwise. If
i
is assumed to be normal, then we
have the probit model. The logit model arises if we assume that the density function of
i
is
(
i
) =
(
i
)
i
=
e
i
(1 +e
i
)
2
276
Chapter 32. Discrete and censored dependent variables 277
Both the probit and logit model are estimated in gretl via maximum likelihood, where the log-
likelihood can be written as
L() =
_
y
i
=0
ln[1 F(z
i
)] +
_
y
i
=1
lnF(z
i
), (32.6)
which is always negative, since 0 < F() < 1. Since the score equations do not have a closed form
solution, numerical optimization is used. However, in most cases this is totally transparent to the
user, since usually only a few iterations are needed to ensure convergence. The --verbose switch
can be used to track the maximization algorithm.
Example 32.1: Estimation of simple logit and probit models
open greene19_1
logit GRADE const GPA TUCE PSI
probit GRADE const GPA TUCE PSI
As an example, we reproduce the results given in chapter 21 of Greene (2000), where the eective-
ness of a program for teaching economics is evaluated by the improvements of students grades.
Running the code in example 32.1 gives the output reported in Table 32.1; note that, for the probit
model, a conditional moment test on skewness and kurtosis is printed out automatically as a test
for normality.
In this context, the $uhat accessor function takes a special meaning: it returns generalized resid-
uals as dened in Gourieroux, Monfort, Renault and Trognon (1987), which can be interpreted as
unbiased estimators of the latent disturbances
i
. These are dened as
u
i
=
_
_
_
y
i

P
i
for the logit model
y
i
( z
i
)
( z
i
)
(1 y
i
)
( z
i
)
1( z
i
)
for the probit model
(32.7)
Among other uses, generalized residuals are often used for diagnostic purposes. For example, it is
very easy to set up an omitted variables test equivalent to the familiar LM test in the context of a
linear regression; example 32.2 shows how to perform a variable addition test.
Example 32.2: Variable addition test in a probit model
open greene19_1
probit GRADE const GPA PSI
series u = $uhat
%$
ols u const GPA PSI TUCE -q
printf "Variable addition test for TUCE:\n"
printf "Rsq * T = %g (p. val. = %g)\n", $trsq, pvalue(X,1,$trsq)
The perfect prediction problem
One curious characteristic of logit and probit models is that (quite paradoxically) estimation is not
feasible if a model ts the data perfectly; this is called the perfect prediction problem. The reason
Model 1: Logit estimates using the 32 observations 1-32
Dependent variable: GRADE
VARIABLE COEFFICIENT STDERROR T STAT SLOPE
(at mean)
const -13.0213 4.93132 -2.641
GPA 2.82611 1.26294 2.238 0.533859
TUCE 0.0951577 0.141554 0.672 0.0179755
PSI 2.37869 1.06456 2.234 0.449339
Mean of GRADE = 0.344
Number of cases correctly predicted = 26 (81.2%)
f(betax) at mean of independent vars = 0.189
McFaddens pseudo-R-squared = 0.374038
Likelihood ratio test: Chi-square(3) = 15.4042 (p-value 0.001502)
Akaike information criterion (AIC) = 33.7793
Schwarz Bayesian criterion (BIC) = 39.6422
Hannan-Quinn criterion (HQC) = 35.7227
Predicted
0 1
Actual 0 18 3
1 3 8
Model 2: Probit estimates using the 32 observations 1-32
Dependent variable: GRADE
VARIABLE COEFFICIENT STDERROR T STAT SLOPE
(at mean)
const -7.45232 2.54247 -2.931
GPA 1.62581 0.693883 2.343 0.533347
TUCE 0.0517288 0.0838903 0.617 0.0169697
PSI 1.42633 0.595038 2.397 0.467908
Mean of GRADE = 0.344
f(betax) at mean of independent vars = 0.328
McFaddens pseudo-R-squared = 0.377478
Likelihood ratio test: Chi-square(3) = 15.5459 (p-value 0.001405)
Akaike information criterion (AIC) = 33.6376
Schwarz Bayesian criterion (BIC) = 39.5006
Hannan-Quinn criterion (HQC) = 35.581
Predicted
0 1
Actual 0 18 3
1 3 8
Test for normality of residual -
Null hypothesis: error is normally distributed
Test statistic: Chi-square(2) = 3.61059
with p-value = 0.164426
Table 32.1: Example logit and probit
why this problem arises is easy to see by considering equation (32.6): if for some vector and scalar
k its the case that z
i
< k whenever y
i
= 0 and z
i
> k whenever y
i
= 1, the same thing is true
for any multiple of . Hence, L() can be made arbitrarily close to 0 simply by choosing enormous
values for . As a consequence, the log-likelihood has no maximum, despite being bounded.
Gretl has a mechanism for preventing the algorithm from iterating endlessly in search of a non-
existent maximum. One sub-case of interest is when the perfect prediction problem arises because
of a single binary explanatory variable. In this case, the oending variable is dropped from the
model and estimation proceeds with the reduced specication. Nevertheless, it may happen that
no single perfect classier exists among the regressors, in which case estimation is simply impos-
sible and the algorithm stops with an error. This behavior is triggered during the iteration process
if
maxz
i
i:y
i
=0
<minz
i
i:y
i
=1
If this happens, unless your model is trivially mis-specied (like predicting if a country is an oil
exporter on the basis of oil revenues), it is normally a small-sample problem: you probably just
dont have enough data to estimate your model. You may want to drop some of your explanatory
variables.
This problem is well analyzed in Stokes (2004); the results therein are replicated in the example
script murder_rates.inp.
32.2 Ordered response models
These models constitute a simple variation on ordinary logit/probit models, and are usually applied
when the dependent variable is a discrete and ordered measurementnot simply binary, but on
an ordinal rather than an interval scale. For example, this sort of model may be applied when the
dependent variable is a qualitative assessment such as Good, Average and Bad.
In the general case, consider an ordered response variable, y, that can take on any of the J+1 values
0, 1, 2, . . . , J. We suppose, as before, that underlying the observed response is a latent variable,
y
= X + = z +
Now dene cut points,
1
<
2
< <
J
, such that
y = 0 if y

1
y = 1 if
1
< y

2
.
.
.
y = J if y
>
J
For example, if the response takes on three values there will be two such cut points,
1
and
2
.
The probability that individual i exhibits response j, conditional on the characteristics x
i
, is then
given by
P(y
i
= j x
i
) =
_
_
P(y

1
x
i
) = F(
1
z
i
) for j = 0
P(
j
< y

j+1
x
i
) = F(
j+1
z
i
) F(
j
z
i
) for 0 < j < J
P(y
>
J
x
i
) = 1 F(
J
z
i
) for j = J
(32.8)
The unknown parameters
j
are estimated jointly with the s via maximum likelihood. The
j
estimates are reported by gretl as cut1, cut2 and so on. For the probit variant, a conditional
moment test for normality constructed in the spirit of Chesher and Irish (1987) is also included.
Note that the
j
parameters can be shifted arbitrarily by adding a constant to z
i
, so the model is
under-identied if there is some linear combination of the explanatory variables which is constant.
The most obvious case in which this occurs is when the model contains a constant term; for this
reason, gretl drops automatically the intercept if present. However, it may happen that the user in-
adventently species a list of regressors that may be combined in such a way to produce a constant
(for example, by using a full set of dummy variables for a discrete factor). If this happens, gretl will
also drop any oending regressors.
In order to apply these models in gretl, the dependent variable must either take on only non-
negative integer values, or be explicitly marked as discrete. (In case the variable has non-integer
values, it will be recoded internally.) Note that gretl does not provide a separate command for
ordered models: the logit and probit commands automatically estimate the ordered version if
the dependent variable is acceptable, but not binary.
Example 32.3 reproduces the results presented in section 15.10 of Wooldridge (2002a). The ques-
tion of interest in this analysis is what dierence it makes, to the allocation of assets in pension
funds, whether individual plan participants have a choice in the matter. The response variable is
an ordinal measure of the weight of stocks in the pension portfolio. Having reported the results
of estimation of the ordered model, Wooldridge illustrates the eect of the choice variable by ref-
erence to an average participant. The example script shows how one can compute this eect in
gretl.
After estimating ordered models, the $uhat accessor yields generalized residuals as in binary mod-
els; additionally, the $yhat accessor function returns z
i
, so it is possible to compute an unbiased
estimator of the latent variable y
i
simply by adding the two together.
32.3 Multinomial logit
When the dependent variable is not binary and does not have a natural ordering, multinomial
models are used. Multinomial logit is supported in gretl via the --multinomial option to the
logit command. Simple models can also be handled via the mle command (see chapter 21). We
give here an example of such a model. Let the dependent variable, y
i
, take on integer values
0, 1, . . . p. The probability that y
i
= k is given by
P(y
i
= kx
i
) =
exp(x
i
k
)
p
j=0
exp(x
i
j
)
For the purpose of identication one of the outcomes must be taken as the baseline; it is usually
assumed that
0
= 0, in which case
P(y
i
= kx
i
) =
exp(x
i
k
)
1 +
p
j=1
exp(x
i
j
)
and
P(y
i
= 0x
i
) =
1
1 +
p
j=1
exp(x
i
j
)
.
Example 32.4 reproduces Table 15.2 in Wooldridge (2002a), based on data on career choice from
Keane and Wolpin (1997). The dependent variable is the occupational status of an individual (0 = in
school; 1 = not in school and not working; 2 = working), and the explanatory variables are education
and work experience (linear and square) plus a black binary variable. The full data set is a panel;
here the analysis is conned to a cross-section for 1987.
32.4 Bivariate probit
The bivariate probit model is simply a two-equation system in which each equation is a probit
model, but the two disturbance terms may not be independent. In formulae,
y
1,i
=
k
1
_
j=1
x
ij
j
+
1,i
y
1,i
= 1 y
1,i
> 0 (32.9)
Example 32.3: Ordered probit model
/*
Replicate the results in Wooldridge, Econometric Analysis of Cross
Section and Panel Data, section 15.10, using pension-plan data from
Papke (AER, 1998).
The dependent variable, pctstck (percent stocks), codes the asset
allocation responses of "mostly bonds", "mixed" and "mostly stocks"
as {0, 50, 100}.
The independent variable of interest is "choice", a dummy indicating
whether individuals are able to choose their own asset allocations.
*/
open pension.gdt
# demographic characteristics of participant
list DEMOG = age educ female black married
# dummies coding for income level
list INCOME = finc25 finc35 finc50 finc75 finc100 finc101
# Papkes OLS approach
ols pctstck const choice DEMOG INCOME wealth89 prftshr
# save the OLS choice coefficient
choice_ols = $coeff(choice)
# estimate ordered probit
probit pctstck choice DEMOG INCOME wealth89 prftshr
k = $ncoeff
matrix b = $coeff[1:k-2]
a1 = $coeff[k-1]
a2 = $coeff[k]
/*
Wooldridge illustrates the choice effect in the ordered probit
by reference to a single, non-black male aged 60, with 13.5 years
of education, income in the range $50K - $75K and wealth of $200K,
participating in a plan with profit sharing.
*/
matrix X = {60, 13.5, 0, 0, 0, 0, 0, 0, 1, 0, 0, 200, 1}
# with choice = 0
scalar Xb = (0 ~ X) * b
P0 = cdf(N, a1 - Xb)
P50 = cdf(N, a2 - Xb) - P0
P100 = 1 - cdf(N, a2 - Xb)
E0 = 50 * P50 + 100 * P100
# with choice = 1
Xb = (1 ~ X) * b
P0 = cdf(N, a1 - Xb)
P50 = cdf(N, a2 - Xb) - P0
P100 = 1 - cdf(N, a2 - Xb)
E1 = 50 * P50 + 100 * P100
printf "\nWith choice, E(y) = %.2f, without E(y) = %.2f\n", E1, E0
printf "Estimated choice effect via ML = %.2f (OLS = %.2f)\n", E1 - E0,
choice_ols
Example 32.4: Multinomial logit
Input:
open keane.gdt
smpl (year=87) --restrict
logit status 0 educ exper expersq black --multinomial
Model 1: Multinomial Logit, using observations 1-1738 (n = 1717)
Missing or incomplete observations dropped: 21
Dependent variable: status
Standard errors based on Hessian
coefficient std. error z p-value
--------------------------------------------------------
status = 2
const 10.2779 1.13334 9.069 1.20e-19 ***
educ -0.673631 0.0698999 -9.637 5.57e-22 ***
exper -0.106215 0.173282 -0.6130 0.5399
expersq -0.0125152 0.0252291 -0.4961 0.6199
black 0.813017 0.302723 2.686 0.0072 ***
status = 3
const 5.54380 1.08641 5.103 3.35e-07 ***
educ -0.314657 0.0651096 -4.833 1.35e-06 ***
exper 0.848737 0.156986 5.406 6.43e-08 ***
expersq -0.0773003 0.0229217 -3.372 0.0007 ***
black 0.311361 0.281534 1.106 0.2687
Log-likelihood -907.8572 Akaike criterion 1835.714
Schwarz criterion 1890.198 Hannan-Quinn 1855.874
Likelihood ratio test: Chi-square(8) = 583.722 [0.0000]
y
2,i
=
k
2
_
j=1
z
ij
j
+
2,i
y
2,i
= 1 y
2,i
> 0 (32.10)
_

2,i
2,i
_
N
_
0,
_
1
1
__
(32.11)
The explanatory variables for the rst equation x and for the second equation z may overlap
example contained in biprobit.inp
$uhat and $yhat are matrices
FIXME: expand.
32.5 The Tobit model
The Tobit model is used when the dependent variable of a model is censored. Assume a latent
variable y
i
can be described as
y
i
=
k
_
j=1
x
ij
j
+
i
,
where
i
N(0,
2
). If y
i
were observable, the models parameters could be estimated via ordinary
least squares. On the contrary, suppose that we observe y
i
, dened as
y
i
=
_
_
a for y
i
a
y
i
for a < y
i
< b
b for y
i
b
(32.12)
In most cases found in the applied literature, a = 0 and b = , so in practice negative values of y
i
are not observed and are replaced by zeros.
In this case, regressing y
i
on the x
i
s does not yield consistent estimates of the parameters ,
because the conditional mean E(y
i
x
i
) is not equal to

k
j=1
x
ij
j
. It can be shown that restricting
the sample to non-zero observations would not yield consistent estimates either. The solution is to
estimate the parameters via maximum likelihood. The syntax is simply
tobit depvar indvars
As usual, progress of the maximization algorithm can be tracked via the --verbose switch, while
$uhat returns the generalized residuals. Note that in this case the generalized residual is dened
as u
i
= E(
i
y
i
= 0) for censored observations, so the familiar equality u
i
= y
i
y
i
only holds for
uncensored observations, that is, when y
i
> 0.
An important dierence between the Tobit estimator and OLS is that the consequences of non-
normality of the disturbance term are much more severe: non-normality implies inconsistency for
the Tobit estimator. For this reason, the output for the Tobit model includes the Chesher and Irish
(1987) normality test by default.
The general case in which a is nonzero and/or b is nite can be handled by using the options
--llimit and --rlimit. So, for example,
tobit depvar indvars --llimit=10
would tell gretl that the left bound a is set to 10.
32.6 Interval regression
The interval regression model arises when the dependent variable is unobserved for some (possibly
all) observations; what we observe instead is an interval in which the dependent variable lies. In
other words, the data generating process is assumed to be
y
i
= x
i
+
i
but we only know that m
i
y
i
M
i
, where the interval may be left- or right-unbounded (but
not both). If m
i
= M
i
, we eectively observe y
i
and no information loss occurs. In practice, each
observation belongs to one of four categories:
1. left-unbounded, when m
i
= ,
2. right-unbounded, when M
i
= ,
3. bounded, when < m
i
< M
i
< and
4. point observations when m
i
= M
i
.
It is interesting to note that this model bears similarities to other models in several special cases:
When all observations are point observations the model trivially reduces to the ordinary linear
regression model.
The interval model could be thought of an ordered probit model (see 32.2) in which the cut
points (the
j
coecients in eq. 32.8) are observed and dont need to be estimated.
The Tobit model (see 32.5) is a special case of the interval model in which m
i
and M
i
do not
depend on i, that is, the censoring limits are the same for all observations. As a matter of
fact, gretls tobit commands is handled internally as a special case of the interval model.
The gretl command intreg estimates interval models by maximum likelihood, assuming normality
of the disturbance term
i
. Its syntax is
intreg minvar maxvar X
where minvar contains the m
i
series, with NAs for left-unbounded observations, and maxvar con-
tains M
i
, with NAs for right-unbounded observations. By default, standard errors are computed
using the negative inverse of the Hessian. If the --robust ag is given, then QML or HuberWhite
standard errors are calculated instead. In this case the estimated covariance matrix is a sandwich
of the inverse of the estimated Hessian and the outer product of the gradient.
If the model specication contains regressors other than just a constant, the output includes a
chi-square statistic for testing the joint null hypothesis that none of these regressors has any
eect on the outcome. This is a Wald statistic based on the estimated covariance matrix. If you
wish to construct a likelihood ratio test, this is easily done by estimating both the full model
and the null model (containing only the constant), saving the log-likelihood in both cases via the
$lnl accessor, and then referring twice the dierence between the two log-likelihoods to the chi-
square distribution with k degrees of freedom, where k is the number of additional regressors (see
the pvalue command in the Gretl Command Reference). Also included is a conditional moment
normality test, similar to those provided for the probit, ordered probit and Tobit models (see
above). An example is contained in the sample script wtp.inp, provided with the gretl distribution.
As with the probit and Tobit models, after a model has been estimated the $uhat accessor returns
the generalized residual, which is an estimate of
i
: more precisely, it equals y
i
x
i
for point
observations and E(
i
m
i
, M
i
, x
i
) otherwise. Note that it is possible to compute an unbiased pre-
dictor of y
i
by summing this estimate to x
i
. Script 32.5 shows an example. As a further similarity

with Tobit, the interval regression model may deliver inconsistent estimates if the disturbances are
non-normal; hence, the Chesher and Irish (1987) test for normality is included by default here too.
Example 32.5: Interval model on articial data
Input:
nulldata 100
# generate artificial data
set seed 201449
x = normal()
epsilon = 0.2*normal()
ystar = 1 + x + epsilon
lo_bound = floor(ystar)
hi_bound = ceil(ystar)
# run the interval model
intreg lo_bound hi_bound const x
# estimate ystar
gen_resid = $uhat
yhat = $yhat + gen_resid
corr ystar yhat
Model 1: Interval estimates using the 100 observations 1-100
Lower limit: lo_bound, Upper limit: hi_bound
---------------------------------------------------------
const 0.993762 0.0338325 29.37 1.22e-189 ***
x 0.986662 0.0319959 30.84 8.34e-209 ***
Chi-square(1) 950.9270 p-value 8.3e-209
Schwarz criterion 102.2407 Hannan-Quinn 97.58824
sigma = 0.223273
Left-unbounded observations: 0
Right-unbounded observations: 0
Bounded observations: 100
Point observations: 0
...
corr(ystar, yhat) = 0.98960092
Under the null hypothesis of no correlation:
t(98) = 68.1071, with two-tailed p-value 0.0000
32.7 Sample selection model
In the sample selection model (also known as Tobit II model), there are two latent variables:
y
i
=
k
_
j=1
x
ij
j
+
i
(32.13)
s
i
=
p
_
j=1
z
ij
j
+
i
(32.14)
and the observation rule is given by
y
i
=
_
y
i
for s
i
> 0
for s
i
0
(32.15)
In this context, the symbol indicates that for some observations we simply do not have data on
y: y
i
may be 0, or missing, or anything else. A dummy variable d
i
is normally used to set censored
observations apart.
One of the most popular applications of this model in econometrics is a wage equation coupled
with a labor force participation equation: we only observe the wage for the employed. If y
i
and s
i
were (conditionally) independent, there would be no reason not to use OLS for estimating equation
(32.13); otherwise, OLS does not yield consistent estimates of the parameters
j
.
Since conditional independence between y
i
and s
i
is equivalent to conditional independence be-
tween
i
and
i
, one may model the co-dependence between
i
and
i
as
i
=
i
+v
i
;
substituting the above expression in (32.13), you obtain the model that is actually estimated:
y
i
=
k
_
j=1
x
ij
j
+
i
+v
i
,
so the hypothesis that censoring does not matter is equivalent to the hypothesis H
0
: = 0, which
can be easily tested.
The parameters can be estimated via maximum likelihood under the assumption of joint normality
of
i
and
i
; however, a widely used alternative method yields the so-called Heckit estimator, named
after Heckman (1979). The procedure can be briey outlined as follows: rst, a probit model is t
on equation (32.14); next, the generalized residuals are inserted in equation (32.13) to correct for
the eect of sample selection.
Gretl provides the heckit command to carry out estimation; its syntax is
heckit y X ; d Z
where y is the dependent variable, X is a list of regressors, d is a dummy variable holding 1 for
uncensored observations and Z is a list of explanatory variables for the censoring equation.
Since in most cases maximum likelihood is the method of choice, by default gretl computes ML
estimates. The 2-step Heckit estimates can be obtained by using the --two-step option. After
estimation, the $uhat accessor contains the generalized residuals. As in the ordinary Tobit model,
the residuals equal the dierence between actual and tted y
i
only for uncensored observations
(those for which d
i
= 1).
Example 32.6 shows two estimates from the dataset used in Mroz (1987): the rst one replicates
Table 22.7 in Greene (2003),
1
while the second one replicates table 17.1 in Wooldridge (2002a).
32.8 Count data
Here the dependent variable is assumed to be a non-negative integer, so a probabilistic description
of y
i
x
i
must hinge on some discrete distribution. The most common model is the Poisson model,
1
Note that the estimates given by gretl do not coincide with those found in the printed volume. They do, however,
match those found on the errata web page for Greenes book: http://pages.stern.nyu.edu/~wgreene/Text/Errata/
ERRATA5.htm.
Example 32.6: Heckit model
open mroz87.gdt
genr EXP2 = AX^2
genr WA2 = WA^2
genr KIDS = (KL6+K618)>0
# Greenes specification
list X = const AX EXP2 WE CIT
list Z = const WA WA2 FAMINC KIDS WE
heckit WW X ; LFP Z --two-step
heckit WW X ; LFP Z
# Wooldridges specification
series NWINC = FAMINC - WW*WHRS
series lww = log(WW)
list X = const WE AX EXP2
list Z = X NWINC WA KL6 K618
heckit lww X ; LFP Z --two-step
in which
P(y
i
= Yx
i
) = e
Y
i
Y!
i
= exp
_
_
_
j
x
ij
j
_
_
In some cases, an oset variable is needed. The number of occurrences of y
i
in a given time is
assumed to be strictly proportional to the oset variable n
i
. In the epidemiology literature, the
oset is known as population at risk. In this case, the model becomes
i
= n
i
exp
_
_
_
j
x
ij
j
_
_
Another way to look at the oset variable is to consider its natural log as just another explanatory
variable whose coecient is constrained to be one.
Estimation is carried out by maximum likelihood and follows the syntax
poisson depvar indep
If an oset variable is needed, it has to be specied at the end of the command, separated from the
list of explanatory variables by a semicolon, as in
poisson depvar indep ; offset
It should be noted that the poisson command does not use, internally, the same optimization
engines as most other gretl command, such as arma or tobit. As a consequence, some details may
dier: the --verbose option will yield dierent output and settings such as bfgs_toler will not
work.
Overdispersion
In the Poisson model, E(y
i
x
i
) = V(y
i
x
i
) =
i
, that is, the conditional mean equals the conditional
variance by construction. In many cases, this feature is at odds with the data, as the conditional
variance is often larger than the mean; this phenomenon is called overdispersion. The output
from the poisson command includes a conditional moment test for overdispersion (as per David-
son and MacKinnon (2004), section 11.5), which is printed automatically after estimation.
Overdispersion can be attributed to unmodeled heterogeneity between individuals. Two data points
with the same observable characteristics x
i
= x
j
may dier because of some unobserved scale
factor s
i
}= s
j
so that
E(y
i
x
i
, s
i
) =
i
s
i
}=
j
s
j
= E(y
j
x
j
, s
j
)
even though
i
=
j
. In other words, y
i
is a Poisson random variable conditional on both x
i
and s
i
,
but since s
i
is unobservable, the only thing we can we can use, P(y
i
x
i
), will not follow the Poisson
distribution.
It is often assumed that s
i
can be represented as a gamma random variable with mean 1 and
variance : the parameter is estimated together with the vector , and measures the degree of
heterogeneity between individuals.
In this case, the conditional probability for y
i
given x
i
can be shown to be
P(y
i
x
i
) =
(y
i
+
1
)
(
1
) (y
i
+1)
_

i
i
+
1
_
y
i
_

1
i
+
1
_
1
(32.16)
which is known as the Negative Binomial Model. The conditional mean is still E(y
i
x
i
) =
i
, but
the variance equals V(y
i
x
i
) =
i
(1 +
i
). The gretl command for this model is negbin depvar
indep.
There is also a less used variant of the negative binomial model, in which the conditional vari-
ance is a scalar multiple of the conditional mean, that is V(y
i
x
i
) =
i
(1 +). To distinguish
between the two, the model (32.16) is termed Type 2. Gretl implements model 1 via the
option --model1.
A script which exemplies the above models is included among gretls sample scripts, under
the name camtriv.inp.
FIXME: expand.
32.9 Duration models
In some contexts we wish to apply econometric methods to measurements of the duration of certain
states. Classic examples include the following:
From engineering, the time to failure of electronic or mechanical components: how long do,
say, computer hard drives last until they malfunction?
From the medical realm: how does a new treatment aect the time from diagnosis of a certain
condition to exit from that condition (where exit might mean death or full recovery)?
From economics: the duration of strikes, or of spells of unemployment.
In each case we may be interested in how the durations are distributed, and how they are aected
by relevant covariates. There are several approaches to this problem; the one we discuss here
which is currently the only one supported by gretl is estimation of a parametric model by means
of Maximum Likelihood. In this approach we hypothesize that the durations follow some denite
probability law and we seek to estimate the parameters of that law, factoring in the inuence of
covariates.
We may express the density (PDF) of the durations as f(t, X, ), where t is the length of time in the
state in question, X is a matrix of covariates, and is a vector of parameters. The likelihood for a
sample of n observations indexed by i is then
L =
n
i=1
f(t
i
, x
i
, )
Rather than working with the density directly, however, it is standard practice to factor f() into
two components, namely a hazard function, , and a survivor function, S. The survivor function
gives the probability that a state lasts at least as long as t; it is therefore 1 F(t, X, ) where F
is the CDF corresponding to the density f(). The hazard function addresses this question: given
that a state has persisted as long as t, what is the likelihood that it ends within a short increment
of time beyond t that is, it ends between t and t +? Taking the limit as goes to zero, we end
up with the ratio of the density to the survivor function:
2
(t, X, ) =
f(t, X, )
S(t, X, )
(32.17)
so the log-likelihood can be written as
=
n
_
i=1
logf(t
i
, x
i
, ) =
n
_
i=1
log(t
i
, x
i
, ) +logS(t
i
, x
i
, ) (32.18)
One point of interest is the shape of the hazard function, in particular its dependence (or not) on
time since the state began. If does not depend on t we say the process in question exhibits du-
ration independence: the probability of exiting the state at any given moment neither increases nor
decreases based simply on how long the state has persisted to date. The alternatives are positive
duration dependence (the likelihood of exiting the state rises, the longer the state has persisted)
or negative duration dependence (exit becomes less likely, the longer it has persisted). Finally, the
behavior of the hazard with respect to time need not be monotonic; some parameterizations allow
for this possibility and some do not.
Since durations are inherently positive the probability distribution used in modeling must respect
this requirement, giving a density of zero for t 0. Four common candidates are the exponential,
Weibull, log-logistic and log-normal, the Weibull being the most common choice. The table below
displays the density and the hazard function for each of these distributions as they are commonly
parameterized, written as functions of t alone. ( and denote, respectively, the Gaussian PDF
and CDF.)
density, f(t) hazard, (t)
Exponential exp(t)
Weibull
t
1
exp[(t)
t
1
Log-logistic
(t)
1
[1 +(t)
]
2

(t)
1
[1 +(t)
]
Log-normal
1
t
_
(logt )/
_
1
t
_
(logt )/
_
_
(logt )/
_
The hazard is constant for the exponential distribution. For the Weibull, it is monotone increasing
in t if > 1, or monotone decreasing for < 1. (If = 1 the Weibull collapses to the exponential.)
2
For a fuller discussion see, for example, Davidson and MacKinnon (2004).
The log-logistic and log-normal distributions allow the hazard to vary with t in a non-monotonic
fashion.
Covariates are brought into the picture by allowing them to govern one of the parameters of the
density, so that the durations are not identically distributed across cases. For example, when using
the log-normal distribution it is natural to make , the expected value of logt, depend on the
covariates, X. This is typically done via a linear index function: = X.
Note that the expressions for the log-normal density and hazard contain the term (logt )/.
Replacing with X this becomes (logt X)/. It turns out that this constitutes a useful simpli-
fying change of variables for all of the distributions discussed here. As in Kalbeisch and Prentice
(2002), we dene
w
i
(logt
i
x
i
)/
The interpretation of the scale factor, , in this expression depends on the distribution. For the
log-normal, represents the standard deviation of logt; for the Weibull and the log-logistic it
corresponds to 1/; and for the exponential it is xed at unity. For distributions other than the
log-normal, X corresponds to log, or in other words = exp(X).
With this change of variables, the density and survivor functions may be written compactly as
follows (the exponential is the same as the Weibull).
density, f(w
i
) survivor, S(w
i
)
Weibull exp(w
i
e
w
i
) exp(e
w
i
)
Log-logistic e
w
i
(1 +e
w
i
)
2
(1 +e
w
i
)
1
Log-normal (w
i
) (w
i
)
In light of the above we may think of the generic parameter vector , as in f(t, X, ), as composed
of the coecients on the covariates, , plus (in all cases apart from the exponential) the additional
parameter .
A complication in estimation of is posed by incomplete spells. That is, in some cases the state
in question may not have ended at the time the observation is made (e.g. some workers remain
unemployed, some components have not yet failed). If we use t
i
to denote the time from entering
the state to either (a) exiting the state or (b) the observation window closing, whichever comes rst,
then all we know of the right-censored cases (b) is that the duration was at least as long as t
i
.
This can be handled by rewriting the the log-likelihood (compare 32.18) as
i
=
n
_
i=1
i
logS (w
i
) +(1
i
)
_
log +logf (w
i
)
_
(32.19)
where
i
equals 1 for censored cases (incomplete spells), and 0 for complete observations. The
rationale for this is that the log-density equals the sum of the log hazard and the log survivor
function, but for the incomplete spells only the survivor function contributes to the likelihood. So
in (32.19) we are adding up the log survivor function alone for the incomplete cases, plus the full
log density for the completed cases.
Implementation in gretl and illustration
The duration command accepts a list of series on the usual pattern: dependent variable followed
by covariates. If right-censoring is present in the data this should be represented by a dummy
variable corresponding to
i
above, separated from the covariates by a semicolon. For example,
duration durat 0 X ; cens
where durat measures durations, 0 represents the constant (which is required for such models), X
is a named list of regressors, and cens is the censoring dummy.
By default the Weibull distribution is used; you can substitute any of the other three distribu-
tions discussed here by appending one of the option ags --exponential, --loglogistic or
--lognormal.
Interpreting the coecients in a duration model requires some care, and we will work through an
illustrative case. The example comes from section 20.3 of Wooldridge (2002a), and it concerns
criminal recidivism.
3
The data (lename recid.gdt) pertain to a sample of 1,445 convicts released
from prison between July 1, 1977 and June 30, 1978. The dependent variable is the time in months
until they are again arrested. The information was gathered retrospectively by examining records
in April 1984; the maximum possible length of observation is 81 months. Right-censoring is impor-
tant: when the date were compiled about 62 percent had not been arrested. The dataset contains
several covariates, which are described in the data le; we will focus below on interpretation of the
married variable, a dummy which equals 1 if the respondent was married when imprisoned.
Example 32.7 shows the gretl commands for a Weibull model along with most of the output. Con-
sider rst the scale factor, . The estimate is 1.241 with a standard error of 0.048. (We dont print
a z score and p-value for this term since H
0
: = 0 is not of interest.) Recall that corresponds
to 1/; we can be condent that is less than 1, so recidivism displays negative duration depen-
dence. This makes sense: it is plausible that if a past oender manages to stay out of trouble for
an extended period his risk of engaging in crime again diminishes. (The exponential model would
therefore not be appropriate in this case.)
On a priori grounds, however, we may doubt the monotonic decline in hazard that is implied by
the Weibull specication. Even if a person is liable to return to crime, it seems relatively unlikely
that he would do so straight out of prison. In the data, we nd that only 2.6 percent of those
followed were rearrested within 3 months. The log-normal specication, which allows the hazard
to rise and then fall, may be more appropriate. Using the duration command again with the same
covariates but the --lognormal ag, we get a log-likelihood of 1597 as against 1633 for the
Weibull, conrming that the log-normal gives a better t.
Let us now focus on the married coecient, which is positive in both specications but larger and
more sharply estimated in the log-normal variant. The rst thing is to get the interpretation of the
sign right. Recall that X enters negatively into the intermediate variable w. The Weibull hazard is
(w
i
) = e
w
i
, so being married reduces the hazard of re-oending, or in other words lengthens the
expected duration out of prison. The same qualitative interpretation applies for the log-normal.
To get a better sense of the married eect, it is useful to show its impact on the hazard across time.
We can do this by plotting the hazard for two values of the index function X: in each case the
values of all the covariates other than married are set to their means (or some chosen values) while
married is set rst to 0 then to 1. Example 32.8 provides a script that does this, and the resulting
plots are shown in Figure 32.1. Note that when computing the hazards we need to multiply by the
Jacobian of the transformation from t
i
to w
i
= log(t
i
x
i
)/, namely 1/t. Note also that the
estimate of is available via the accessor $sigma, but it is also present as the last element in the
coecient vector obtained via $coeff.
A further dierence between the Weibull and log-normal specications is illustrated in the plots.
The Weibull is an instance of a proportional hazard model. This means that for any sets of values of
the covariates, x
i
and x
j
, the ratio of the associated hazards is invariant with respect to duration. In
this example the Weibull hazard for unmarried individuals is always 1.1637 times that for married.
In the log-normal variant, on the other hand, this ratio gradually declines from 1.6703 at one month
to 1.1766 at 100 months.
3
Germn Rodrguez of Princeton University has a page discussing this example and displaying estimates from Stata
at http://data.princeton.edu/pop509a/recid1.html.
Example 32.7: Weibull model for recidivism data
Input:
open recid.gdt
list X = workprg priors tserved felon alcohol drugs \
black married educ age
duration durat 0 X ; cens
duration durat 0 X ; cens --lognormal
Partial output:
Model 1: Duration (Weibull), using observations 1-1445
Dependent variable: durat
--------------------------------------------------------
const 4.22167 0.341311 12.37 3.85e-35 ***
workprg -0.112785 0.112535 -1.002 0.3162
priors -0.110176 0.0170675 -6.455 1.08e-10 ***
tserved -0.0168297 0.00213029 -7.900 2.78e-15 ***
felon 0.371623 0.131995 2.815 0.0049 ***
alcohol -0.555132 0.132243 -4.198 2.69e-05 ***
drugs -0.349265 0.121880 -2.866 0.0042 ***
black -0.563016 0.110817 -5.081 3.76e-07 ***
married 0.188104 0.135752 1.386 0.1659
educ 0.0289111 0.0241153 1.199 0.2306
age 0.00462188 0.000664820 6.952 3.60e-12 ***
sigma 1.24090 0.0482896
Model 2: Duration (log-normal), using observations 1-1445
Dependent variable: durat
---------------------------------------------------------
const 4.09939 0.347535 11.80 4.11e-32 ***
workprg -0.0625693 0.120037 -0.5213 0.6022
priors -0.137253 0.0214587 -6.396 1.59e-10 ***
tserved -0.0193306 0.00297792 -6.491 8.51e-11 ***
felon 0.443995 0.145087 3.060 0.0022 ***
alcohol -0.634909 0.144217 -4.402 1.07e-05 ***
drugs -0.298159 0.132736 -2.246 0.0247 **
black -0.542719 0.117443 -4.621 3.82e-06 ***
married 0.340682 0.139843 2.436 0.0148 **
educ 0.0229194 0.0253974 0.9024 0.3668
age 0.00391028 0.000606205 6.450 1.12e-10 ***
sigma 1.81047 0.0623022
Example 32.8: Create plots showing conditional hazards
open recid.gdt -q
# leave married separate for analysis
list X = workprg priors tserved felon alcohol drugs \
black educ age
# Weibull variant
duration durat 0 X married ; cens
# coefficients on all Xs apart from married
matrix beta_w = $coeff[1:$ncoeff-2]
# married coefficient
scalar mc_w = $coeff[$ncoeff-1]
scalar s_w = $sigma
# Log-normal variant
duration durat 0 X married ; cens --lognormal
matrix beta_n = $coeff[1:$ncoeff-2]
scalar mc_n = $coeff[$ncoeff-1]
scalar s_n = $sigma
list allX = 0 X
# evaluate X\beta at means of all variables except marriage
scalar Xb_w = meanc({allX}) * beta_w
scalar Xb_n = meanc({allX}) * beta_n
# construct two plot matrices
matrix mat_w = zeros(100, 3)
matrix mat_n = zeros(100, 3)
loop t=1..100 -q
# first column, duration
mat_w[t, 1] = t
mat_n[t, 1] = t
wi_w = (log(t) - Xb_w)/s_w
wi_n = (log(t) - Xb_n)/s_n
# second col: hazard with married = 0
mat_w[t, 2] = (1/t) * exp(wi_w)
mat_n[t, 2] = (1/t) * pdf(z, wi_n) / cdf(z, -wi_n)
wi_w = (log(t) - (Xb_w + mc_w))/s_w
wi_n = (log(t) - (Xb_n + mc_n))/s_n
# third col: hazard with married = 1
mat_w[t, 3] = (1/t) * exp(wi_w)
mat_n[t, 3] = (1/t) * pdf(z, wi_n) / cdf(z, -wi_n)
endloop
colnames(mat_w, "months unmarried married")
colnames(mat_n, "months unmarried married")
gnuplot 2 3 1 --with-lines --supp --matrix=mat_w --output=weibull.plt
gnuplot 2 3 1 --with-lines --supp --matrix=mat_n --output=lognorm.plt
0.006
0.008
0.010
0.012
0.014
0.016
0.018
0.020
0 20 40 60 80 100
months
Weibull
unmarried
married
0.006
0.008
0.010
0.012
0.014
0.016
0.018
0.020
0 20 40 60 80 100
months
Log-normal
unmarried
married
Figure 32.1: Recidivism hazard estimates for married and unmarried ex-convicts
Alternative representations of the Weibull model
One point to watch out for with the Weibull duration model is that the estimates may be represented
in dierent ways. The representation given by gretl is sometimes called the accelerated failure-time
(AFT) metric. An alternative that one sometimes sees is the log relative-hazard metric; in fact this is
the metric used in Wooldridges presentation of the recidivism example. To get from AFT estimates
to log relative-hazard form it is necessary to multiply the coecients by
1
. For example, the
married coecient in the Weibull specication as shown here is 0.188104 and is 1.24090, so the
alternative value is 0.152, which is what Wooldridge shows (2002a, Table 20.1).
Fitted values and residuals
By default, gretl computes tted values (accessible via $yhat) as the conditional mean of duration.
The formulae are shown below (where denotes the gamma function, and the exponential variant
is just Weibull with = 1).
Weibull Log-logistic Log-normal
exp(X) (1 +) exp(X)

sin()
exp(X +
2
/2)
The expression given for the log-logistic mean, however, is valid only for < 1; otherwise the
expectation is undened, a point that is not noted in all software.
4
Alternatively, if the --medians option is given, gretls duration command will produce conditional
medians as the content of $yhat. For the Weibull the median is exp(X)(log2)
; for the log-logistic

and log-normal it is just exp(X).
The values we give for the accessor $uhat are generalized (CoxSnell) residuals, computed as the
integrated hazard function, which equals the negative of the log of the survivor function:
u
i
= (t
i
, x
i
, ) = logS(t
i
, x
i
, )
Under the null of correct specication of the model these generalized residuals should follow the
unit exponential distribution, which has mean and variance both equal to 1 and density e
1
. See
Cameron and Trivedi (2005) for further discussion.
4
The predict adjunct to the streg command in Stata 10, for example, gaily produces large negative values for the
log-logistic mean in duration models with > 1.
Chapter 33
Quantile regression
33.1 Introduction
In Ordinary Least Squares (OLS) regression, the tted values, y
i
= X
i
, represent the conditional

mean of the dependent variableconditional, that is, on the regression function and the values
of the independent variables. In median regression, by contrast and as the name implies, tted
values represent the conditional median of the dependent variable. It turns out that the principle of
estimation for median regression is easily stated (though not so easily computed), namely, choose
so as to minimize the sum of absolute residuals. Hence the method is known as Least Absolute
Deviations or LAD. While the OLS problem has a straightforward analytical solution, LAD is a linear
programming problem.
Quantile regression is a generalization of median regression: the regression function predicts the
conditional -quantile of the dependent variablefor example the rst quartile ( = .25) or the
ninth decile ( = .90).
If the classical conditions for the validity of OLS are satisedthat is, if the error term is indepen-
dently and identically distributed, conditional on Xthen quantile regression is redundant: all the
conditional quantiles of the dependent variable will march in lockstep with the conditional mean.
Conversely, if quantile regression reveals that the conditional quantiles behave in a manner quite
distinct from the conditional mean, this suggests that OLS estimation is problematic.
As of version 1.7.5, gretl oers quantile regression functionality (in addition to basic LAD regres-
sion, which has been available since early in gretls history via the lad command).
1
33.2 Basic syntax
The basic invocation of quantile regression is
quantreg tau reglist
where
reglist is a standard gretl regression list (dependent variable followed by regressors, including
the constant if an intercept is wanted); and
tau is the desired conditional quantile, in the range 0.01 to 0.99, given either as a numerical
value or the name of a pre-dened scalar variable (but see below for a further option).
Estimation is via the FrischNewton interior point solver (Portnoy and Koenker, 1997), which is sub-
stantially faster than the traditional BarrodaleRoberts (1974) simplex approach for large prob-
lems.
1
We gratefully acknowledge our borrowing from the quantreg package for GNU R (version 4.17). The core of the
quantreg package is composed of Fortran code written by Roger Koenker; this is accompanied by various driver and
auxiliary functions written in the R language by Koenker and Martin Mchler. The latter functions have been re-worked
in C for gretl. We have added some guards against potential numerical problems in small samples.
296
Chapter 33. Quantile regression 297
By default, standard errors are computed according to the asymptotic formula given by Koenker
and Bassett (1978). Alternatively, if the --robust option is given, we use the sandwich estimator
developed in Koenker and Zhao (1994).
2
33.3 Condence intervals
An option --intervals is available. When this is given we print condence intervals for the para-
meter estimates instead of standard errors. These intervals are computed using the rank inversion
method and in general they are asymmetrical about the point estimatesthat is, they are not sim-
ply plus or minus so many standard errors. The specics of the calculation are inected by
the --robust option: without this, the intervals are computed on the assumption of IID errors
(Koenker, 1994); with it, they use the heteroskedasticity-robust estimator developed by Koenker
and Machado (1999).
By default, 90 percent intervals are produced. You can change this by appending a condence value
(expressed as a decimal fraction) to the intervals option, as in
quantreg tau reglist --intervals=.95
When the condence intervals option is selected, the parameter estimates are calculated using
the BarrodaleRoberts method. This is simply because the FrischNewton code does not currently
support the calculation of condence intervals.
Two further details. First, the mechanisms for generating condence intervals for quantile esti-
mates require that the model has at least two regressors (including the constant). If the --intervals
option is given for a model containing only one regressor, an error is agged. Second, when a model
is estimated in this mode, you can retrieve the condence intervals using the accessor $coeff_ci.
This produces a k 2 matrix, where k is the number of regressors. The lower bounds are in the
rst column, the upper bounds in the second. See also section 33.5 below.
33.4 Multiple quantiles
As a further option, you can give tau as a matrixeither the name of a predened matrix or in
numerical form, as in {.05, .25, .5, .75, .95}. The given model is estimated for all the
values and the results are printed in a special form, as shown below (in this case the --intervals
option was also given).
Model 1: Quantile estimates using the 235 observations 1-235
Dependent variable: foodexp
With 90 percent confidence intervals
VARIABLE TAU COEFFICIENT LOWER UPPER
const 0.05 124.880 98.3021 130.517
0.25 95.4835 73.7861 120.098
0.50 81.4822 53.2592 114.012
0.75 62.3966 32.7449 107.314
0.95 64.1040 46.2649 83.5790
income 0.05 0.343361 0.343327 0.389750
0.25 0.474103 0.420330 0.494329
0.50 0.560181 0.487022 0.601989
0.75 0.644014 0.580155 0.690413
0.95 0.709069 0.673900 0.734441
2
These correspond to the iid and nid options in Rs quantreg package, respectively.
0.3
0.35
0.4
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0 0.2 0.4 0.6 0.8 1
tau
Coefficient on income
Quantile estimates with 90% band
OLS estimate with 90% band
Figure 33.1: Regression of food expenditure on income; Engels data
The gretl GUI has an entry for Quantile Regression (under /Model/Robust estimation), and you can
select multiple quantiles there too. In that context, just give space-separated numerical values (as
per the predened options, shown in a drop-down list).
When you estimate a model in this way most of the standard menu items in the model window
are disabled, but one extra item is availablegraphs showing the sequence for a given coe-
cient in comparison with the OLS coecient. An example is shown in Figure 33.1. This sort of
graph provides a simple means of judging whether quantile regression is redundant (OLS is ne) or
informative.
In the example shownbased on data on household income and food expenditure gathered by
Ernst Engel (18211896) it seems clear that simple OLS regression is potentially misleading. The
crossing of the OLS estimate by the quantile estimates is very marked.
However, it is not always clear what implications should be drawn from this sort of conict. With
the Engel data there are two issues to consider. First, Engels famous law claims an income-
elasticity of food consumption that is less than one, and talk of elasticities suggests a logarithmic
formulation of the model. Second, there are two apparently anomalous observations in the data
set: household 105 has the third-highest income but unexpectedly low expenditure on food (as
judged from a simple scatter plot), while household 138 (which also has unexpectedly low food
consumption) has much the highest income, almost twice that of the next highest.
With n = 235 it seems reasonable to consider dropping these observations. If we do so, and adopt
a loglog formulation, we get the plot shown in Figure 33.2. The quantile estimates still cross the
OLS estimate, but the evidence against OLS is much less compelling: the 90 percent condence
bands of the respective estimates overlap at all the quantiles considered.
33.5 Large datasets
As noted above, when you give the --intervals option with the quantreg command, which calls
for estimation of condence intervals via rank inversion, gretl switches from the default Frisch
Newton algorithm to the BarrodaleRoberts simplex method.
0.76
0.78
0.8
0.82
0.84
0.86
0.88
0.9
0.92
0.94
0.96
0 0.2 0.4 0.6 0.8 1
tau
Coefficient on log(income)
Quantile estimates with 90% band
OLS estimate with 90% band
Figure 33.2: Loglog regression; 2 observations dropped from full Engel data set.
This is OK for moderately large datasets (up to, say, a few thousand observations) but on very large
problems the simplex algorithm may become seriously bogged down. For example, Koenker and
Hallock (2001) present an analysis of the determinants of birth weights, using 198377 observations
and with 15 regressors. Generating condence intervals via BarrodaleRoberts for a single value of
took about half an hour on a Lenovo Thinkpad T60p with 1.83GHz Intel Core 2 processor.
If you want condence intervals in such cases, you are advised not to use the --intervals option,
but to compute them using the method of plus or minus so many standard errors. (One Frisch
Newton run took about 8 seconds on the same machine, showing the superiority of the interior
point method.) The script below illustrates:
quantreg .10 y 0 xlist
scalar crit = qnorm(.95)
matrix ci = $coeff - crit * $stderr
ci = ci~($coeff + crit * $stderr)
print ci
The matrix ci will contain the lower and upper bounds of the (symmetrical) 90 percent condence
intervals.
To avoid a situation where gretl becomes unresponsive for a very long time we have set the maxi-
mum number of iterations for the BorrodaleRoberts algorithm to the (somewhat arbitrary) value
of 1000. We will experiment further with this, but for the meantime if you really want to use this
method on a large dataset, and dont mind waiting for the results, you can increase the limit using
the set command with parameter rq_maxiter, as in
set rq_maxiter 5000
Chapter 34
Nonparametric methods
The main focus of gretl is on parametric estimation, but we oer a selection of nonparametric
methods. The most basic of these
various tests for dierence in distribution (Sign test, Wilcoxon rank-sumtest, Wilcoxon signed-
rank test);
the Runs test for randomness; and
nonparametric measures of association: Spearmans rho and Kendalls tau.
Details on the above can be found by consulting the help for the commands difftest, runs, corr
and spearman. In the GUI program these items are found under the Tools menu and the Robust
estimation item under the Model menu.
In this chapter we concentrate on two relatively complex methods for nonparametric curve-tting
and prediction, namely William Clevelands loess (also known as lowess) and the Nadaraya
Watson estimator.
34.1 Locally weighted regression (loess)
Loess (Cleveland, 1979) is a nonparametric smoother employing locally weighted polynomial re-
gression. It is intended to yield an approximation to g() when the dependent variable, y, can be
expressed as
y
i
= g(x
i
) +
i
for some smooth function g().
Given a sample of n observations on the variables y and x, the procedure is to run a weighted least
squares regression (a polynomial of order d = 0, 1 or 2 in x) localized to each data point, i. In each
such regression the sample consists of the r nearest neighbors (in the x dimension) to the point i,
with weights that are inversely related to the distance x
i
x
k
, k = 1, . . . , r. The predicted value
y
i
is then obtained by evaluating the estimated polynomial at x
i
. The most commonly used order
is d = 1.
A bandwidth parameter 0 < q 1 controls the proportion of the total number of data points used
in each regression; thus r = qn (rounded up to an integer). Larger values of q lead to a smoother
tted series, smaller values to a series that tracks the actual data more closely; 0.25 q 0.5 is
often a suitable range.
In gretls implementation of loess the weighting scheme is that given by Cleveland, namely,
w
k
(x
i
) = W(h
1
i
(x
k
x
i
))
where h
i
is the distance between x
i
and its r
th
nearest neighbor, and W() is the tricube function,
W(x) =
_
(1 x
3
)
3
for x < 1
0 for x 1
300
Chapter 34. Nonparametric methods 301
The local regression can be made robust via an adjustment based on the residuals, e
i
= y
i
y
i
.
Robustness weights,
k
, are dened by
k
= B(e
k
/6s)
where s is the median of the e
i
and B() is the bisquare function,
B(x) =
_
(1 x
2
)
2
for x < 1
0 for x 1
The polynomial regression is then re-run using weight
k
w
k
(x
i
) at (x
k
, y
k
).
The loess() function in gretl takes up to ve arguments as follows: the y series, the x series, the
order d, the bandwidth q, and a boolean switch to turn on the robust adjustment. The last three
arguments are optional: if they are omitted the default values are d = 1, q = 0.5 and no robust
adjustment. An example of a full call to loess() is shown below; in this case a quadratic in x is
specied, three quarters of the data points will be used in each local regression, and robustness is
turned on:
series yh = loess(y, x, 2, 0.75, 1)
An illustration of loess is provided in Example 34.1: we generate a series that has a deterministic
sine wave component overlaid with noise uniformly distributed on (1, 1). Loess is then used to
retrieve a good approximation to the sine function. The resulting graph is shown in Figure 34.1.
Example 34.1: Loess script
nulldata 120
series x = index
scalar n = $nobs
series y = sin(2*pi*x/n) + uniform(-1, 1)
series yh = loess(y, x, 2, 0.75, 0)
gnuplot y yh x --output=display --with-lines=yh
-2
-1.5
-1
-0.5
0
0.5
1
1.5
2
0 20 40 60 80 100 120
x
loess fit
Figure 34.1: Loess: retrieving a sine wave
34.2 The NadarayaWatson estimator
The NadarayaWatson nonparametric estimator (Nadaraya, 1964; Watson, 1964) is an estimator
for the conditional mean of a variable Y, available in a sample of size n, for a given value of a
conditioning variable X, and is dened as
m(X) =
n
j=1
y
j
K
h
(X x
j
)
n
j=1
K
h
(X x
j
)
where K
h
() is the so-called kernel function, which is usually some simple transform of a density
function that depends on a scalar called the bandwidth. The one gretl uses is given by
K
h
(x) = exp
_
x
2
2h
_
for x < and zero otherwise. The scalar is used to prevent numerical problems when the
kernel function is evaluated too far away from zero and is called the trim parameter.
Example 34.2: NadarayaWatson example
# Nonparametric regression example: husbands age on wifes age
open mroz87.gdt
# initial value for the bandwidth
scalar h = $nobs^(-0.2)
# three increasingly smoother estimates
series m0 = nadarwat(HA, WA, h)
series m1 = nadarwat(HA, WA, h * 5)
series m2 = nadarwat(HA, WA, h * 10)
# produce the graph
dataset sortby WA
gnuplot m0 m1 m2 HA WA --output=display --with-lines
Example 34.2 produces the graph shown in Figure 34.2 (after some slight editing).
The choice of the bandwidth is up to the user: larger values of h lead to a smoother m() function;
smaller values make the m() function follow the y
i
values more closely, so that the function
appears more jagged. In fact, as h , m(x
i
)

Y; on the contrary, if h 0, observations for
which x
i
}= X are not taken into account at all when computing m(X).
Also, the statistical properties of m() vary with h: its variance can be shown to be decreasing in
h, while its squared bias is increasing in h. It can be shown that choosing h n
1/5
minimizes the
RMSE, so that value is customarily taken as a reference point.
Note that the kernel function has its tails trimmed. The scalar , which controls the level at
which trimming occurs is set by default at 4 h; this setting, however, may be changed via the set
command. For example,
set nadarwat_trim 10
sets = 10 h. This may at times produce more sensible results in regions of X with sparse
support; however, you should be aware that in those same cases machine precision (division by
numerical zero) may render your results spurious. The default is relatively safe, but experimenting
with larger values may be a sensible strategy in some cases.
30
35
40
45
50
55
60
30 35 40 45 50 55 60
H
A
WA
m0
m1
m2
Figure 34.2: NadarayaWatson example for several choices of the bandwidth parameter
A common variant of the NadarayaWatson estimator is the so-called leave-one-out estimator:
this is a variant of the estimator that does not use the i-th observation for evaluating m(x
i
). This
makes the estimator more robust numerically and its usage is often advised for inference purposes.
In formulae, the leave-one-out estimator is
m(x
i
) =
j}=i
y
j
K
h
(x
i
x
j
)
j}=i
K
h
(x
i
x
j
)
In order to have gretl compute the leave-one-out estimator, just reverse the sign of h: if we changed
example 34.2 by substituting
scalar h = $nobs^(-0.2)
with
scalar h = -($nobs^(-0.2))
the rest of the example would have stayed unchanged, the only dierence being the usage of the
leave-one-out estimator.
Although X could be, in principle, any value, in the typical usage of this estimator you want to
compute m(X) for X equal to one or more values actually observed in your sample, that is m(x
i
).
If you need a point estimate of m(X) for some value of X which is not present among the valid
observations of your dependent variable, you may want to add some fake observations to your
dataset in which y is missing and x contains the values you want m(x) evaluated at. For example,
the following script evaluates m(x) at regular intervals between -2.0 and 2.0:
nulldata 120
set seed 120496
# first part of the sample: actual data
smpl 1 100
x = normal()
y = x^2 + sin(x) + normal()
# second part of the sample: fake x data
smpl 101 120
x = (obs-110) / 5
# compute the Nadaraya-Watson estimate
# with bandwidth equal to 0.4 (note that
# 100^(-0.2) = 0.398)
smpl full
m = nadarwat(y, x, 0.4)
# show m(x) for the fake x values only
smpl 101 120
print x m -o
and running it produces
x m
101 -1.8 1.165934
102 -1.6 0.730221
103 -1.4 0.314705
104 -1.2 0.026057
105 -1.0 -0.131999
106 -0.8 -0.215445
107 -0.6 -0.269257
108 -0.4 -0.304451
109 -0.2 -0.306448
110 0.0 -0.238766
111 0.2 -0.038837
112 0.4 0.354660
113 0.6 0.908178
114 0.8 1.485178
115 1.0 2.000003
116 1.2 2.460100
117 1.4 2.905176
118 1.6 3.380874
119 1.8 3.927682
120 2.0 4.538364
Part III
Technical details
305
Chapter 35
Gretl and T
E
X
35.1 Introduction
T
E
X initially developed by Donald Knuth of Stanford University and since enhanced by hundreds
of contributors around the world is the gold standard of scientic typesetting. Gretl provides
various hooks that enable you to preview and print econometric results using the T
E
X engine, and
to save output in a form suitable for further processing with T
E
X.
This chapter explains the ner points of gretls T
E
X-related functionality. The next section describes
the relevant menu items; section 35.3 discusses ways of ne-tuning T
E
X output; and section 35.4
gives some pointers on installing (and learning) T
E
X if you do not already have it on your computer.
(Just to be clear: T
E
X is not included with the gretl distribution; it is a separate package, including
several programs and a large number of supporting les.)
Before proceeding, however, it may be useful to set out briey the stages of production of a nal
document using T
E
X. For the most part you dont have to worry about these details, since, in regard
to previewing at any rate, gretl handles them for you. But having some grasp of what is going on
behind the scences will enable you to understand your options better.
The rst step is the creation of a plain text source le, containing the text or mathematics to be
typset, interspersed with mark-up that denes how it should be formatted. The second step is to
run the source through a processing engine that does the actual formatting. Typically this is either:
a program called latex that generates so-called DVI (device-independent) output, or
(more commonly nowadays) a program called pdatex that generates PDF output.
1
For previewing, one uses either a DVI viewer (typically xdvi on GNU/Linux systems) or a PDF viewer
(for example, Adobes Acrobat Reader or xpdf), depending on how the source was processed. If
the DVI route is taken, theres then a third step to produce printable output, typically using the
program dvips to generate a PostScript le. If the PDF route is taken, the output is ready for
printing without any further processing.
On MS Windows and Mac OS X, gretl calls pdatex to process the source le, and expects the
operating system to be able to nd the default viewer for PDF output; DVI is not supported. On
GNU/Linux the default is also to produce PDF, but if you prefer the DVI/PostScript route you can do
the following: select the menu item Tools, Preferences, General then the Programs tab. Find the
item titled Command to compile TeX les, and set this to latex. In the same window, Make sure
the commands to view DVI and PostScript les are set to something appropriate for your system.
35.2 T
E
X-related menu items
The model window
The fullest T
E
X support in gretl is found in the GUI model window. This has a menu item titled
LaTeX with sub-items View, Copy, Save and Equation options (see Figure 35.1).
1
Experts will be aware of something called plain T
E
X, which is processed using the program tex. The great majority
of T
E
X users, however, use the L
A
T
E
X macros, initially developed by Leslie Lamport. gretl does not support plain T
E
X.
306
Chapter 35. Gretl and T
E
X 307
Figure 35.1: L
A
T
E
X menu in model window
The rst three sub-items have branches titled Tabular and Equation. By Tabular we mean that
the model is represented in the form of a table; this is the fullest and most explicit presentation of
the results. See Table 35.1 for an example; this was pasted into the manual after using the Copy,
Tabular item in gretl (a few lines were edited out for brevity).
Table 35.1: Example of L
A
T
E
X tabular output
Model 1: OLS estimates using the 51 observations 151
Dependent variable: ENROLL
Variable Coecient Std. Error t-statistic p-value
const 0.241105 0.0660225 3.6519 0.0007
CATHOL 0.223530 0.0459701 4.8625 0.0000
PUPIL 0.00338200 0.00271962 1.2436 0.2198
WHITE 0.152643 0.0407064 3.7499 0.0005
Mean of dependent variable 0.0955686
S.D. of dependent variable 0.0522150
Sum of squared residuals 0.0709594
Standard error of residuals ( ) 0.0388558
Unadjusted R
2
0.479466
Adjusted

R
2
0.446241
F(3, 47) 14.4306
The Equation option is fairly self-explanatorythe results are written across the page in equation
format, as below:
ENROLL = 0.241105
(0.066022)
+0.223530
(0.04597)
CATHOL 0.00338200
(0.0027196)
PUPIL 0.152643
(0.040706)
WHITE
T = 51

R
2
= 0.4462 F(3, 47) = 14.431 = 0.038856
(standard errors in parentheses)
The distinction between the Copy and Save options (for both tabular and equation) is twofold.
First, Copy puts the T
E
X source on the clipboard while with Save you are prompted for the name
of a le into which the source should be saved. Second, with Copy the material is copied as a
E
X 308
fragment while with Save it is written as a complete le. The point is that a well-formed T
E
X
source le must have a header that denes the documentclass (article, report, book or whatever)
and tags that say \begin{document} and \end{document}. This material is included when you do
Save but not when you do Copy, since in the latter case the expectation is that you will paste
the data into an existing T
E
X source le that already has the relevant apparatus in place.
The items under Equation options should be self-explanatory: when printing the model in equa-
tion form, do you want standard errors or t-ratios displayed in parentheses under the parameter
estimates? The default is to show standard errors; if you want t-ratios, select that item.
Other windows
Several other sorts of output windows also have T
E
X preview, copy and save enabled. In the case of
windows having a graphical toolbar, look for the T
E
X button. Figure 35.2 shows this icon (second
from the right on the toolbar) along with the dialog that appears when you press the button.
Figure 35.2: T
E
X icon and dialog
One aspect of gretls T
E
X support that is likely to be particularly useful for publication purposes is
the ability to produce a typeset version of the model table (see section 3.4). An example of this is
shown in Table 35.2.
35.3 Fine-tuning typeset output
There are three aspects to this: adjusting the appearance of the output produced by gretl in
L
A
T
E
X preview mode; adjusting the formatting of gretls tabular output for models when using the
tabprint command; and incorporating gretls output into your own T
E
X les.
Previewing in the GUI
As regards preview mode, you can control the appearance of gretls output using a le named
gretlpre.tex, which should be placed in your gretl user directory (see the Gretl Command Ref-
erence). If such a le is found, its contents will be used as the preamble to the T
E
X source. The
default value of the preamble is as follows:
\documentclass[11pt]{article}
\usepackage[utf8]{inputenc}
\usepackage{amsmath}
\usepackage{dcolumn,longtable}
\begin{document}
\thispagestyle{empty}
E
X 309
Table 35.2: Example of model table output
OLS estimates
Dependent variable: ENROLL
Model 1 Model 2 Model 3
const 0.2907
0.2411
0.08557
(0.07853) (0.06602) (0.05794)
CATHOL 0.2216
0.2235
0.2065
(0.04584) (0.04597) (0.05160)

PUPIL 0.003035 0.003382 0.001697
(0.002727) (0.002720) (0.003025)
WHITE 0.1482
0.1526
(0.04074) (0.04071)
ADMEXP 0.1551
(0.1342)
n 51 51 51
R
2
0.4502 0.4462 0.2956
96.09 95.36 88.69
Standard errors in parentheses
* indicates signicance at the 10 percent level
** indicates signicance at the 5 percent level
E
X 310
Note that the amsmath and dcolumn packages are required. (For some sorts of output the longtable
package is also needed.) Beyond that you can, for instance, change the type size or the font by al-
tering the documentclass declaration or including an alternative font package.
In addition, if you wish to typeset gretl output in more than one language, you can set up per-
language preamble les. A localized preamble le is identied by a name of the formgretlpre_xx.tex,
where xx is replaced by the rst two letters of the current setting of the LANG environment vari-
able. For example, if you are running the program in Polish, using LANG=pl_PL, then gretl will do
the following when writing the preamble for a T
E
X source le.
1. Look for a le named gretlpre_pl.tex in the gretl user directory. If this is not found, then
2. look for a le named gretlpre.tex in the gretl user directory. If this is not found, then
3. use the default preamble.
Conversely, suppose you usually run gretl in a language other than English, and have a suitable
gretlpre.tex le in place for your native language. If on some occasions you want to produce T
E
X
output in English, then you could create an additional le gretlpre_en.tex: this le will be used
for the preamble when gretl is run with a language setting of, say, en_US.
Command-line options
After estimating a model via a scriptor interactively via the gretl console or using the command-
line program gretlcli you can use the commands tabprint or eqnprint to print the model to
le in tabular format or equation format respectively. These options are explained in the Gretl
Command Reference.
If you wish alter the appearance of gretls tabular output for models in the context of the tabprint
command, you can specify a custom row format using the --format ag. The format string must
be enclosed in double quotes and must be tied to the ag with an equals sign. The pattern for the
format string is as follows. There are four elds, representing the coecient, standard error, t-
ratio and p-value respectively. These elds should be separated by vertical bars; they may contain
a printf-type specication for the formatting of the numeric value in question, or may be left
blank to suppress the printing of that column (subject to the constraint that you cant leave all the
columns blank). Here are a few examples:
--format="%.4f|%.4f|%.4f|%.4f"
--format="%.4f|%.4f|%.3f|"
--format="%.5f|%.4f||%.4f"
--format="%.8g|%.8g||%.4f"
The rst of these specications prints the values in all columns using 4 decimal places. The second
suppresses the p-value and prints the t-ratio to 3 places. The third omits the t-ratio. The last one
again omits the t, and prints both coecient and standard error to 8 signicant gures.
Once you set a custom format in this way, it is remembered and used for the duration of the gretl
session. To revert to the default formatting you can use the special variant --format=default.
Further editing
Once you have pasted gretls T
E
X output into your own document, or saved it to le and opened it
in an editor, you can of course modify the material in any wish you wish. In some cases, machine-
generated T
E
X is hard to understand, but gretls output is intended to be human-readable and
-editable. In addition, it does not use any non-standard style packages. Besides the standard L
A
T
E
X
document classes, the only les needed are, as noted above, the amsmath, dcolumn and longtable
packages. These should be included in any reasonably full T
E
X implementation.
E
X 311
35.4 Installing and learning T
E
X
This is not the place for a detailed exposition of these matters, but here are a few pointers.
So far as we know, every GNU/Linux distribution has a package or set of packages for T
E
X, and in
fact these are likely to be installed by default. Check the documentation for your distribution. For
MS Windows, several packaged versions of T
E
X are available: one of the most popular is MiKT
E
X at
http://www.miktex.org/. For Mac OS X a nice implementation is iT
E
XMac, at http://itexmac.
sourceforge.net/. An essential starting point for online T
E
X resources is the Comprehensive T
E
X
Archive Network (CTAN) at http://www.ctan.org/.
As for learning T
E
X, many useful resources are available both online and in print. Among online
guides, Tony Roberts L
A
T
E
X: from quick and dirty to style and nesse is very helpful, at
http://www.sci.usq.edu.au/staff/robertsa/LaTeX/latexintro.html
An excellent source for advanced material is The L
A
T
E
X Companion (Goossens et al., 2004).
Chapter 36
Gretl and R
36.1 Introduction
R is, by far, the largest free statistical project.
1
Like gretl, it is a GNU project and the two have a
lot in common; however, gretls approach focuses on ease of use much more than R, which instead
aims to encompass the widest possible range of statistical procedures.
As is natural in the free software ecosystem, we dont view ourselves as competitors to R,
2
but
rather as projects sharing a common goal who should support each other whenever possible. For
this reason, gretl provides a way to interact with R and thus enable users to pool the capabilities of
the two packages.
In this chapter, we will explain how to exploit Rs power from within gretl. We assume that the
reader has a working installation of R available and a basic grasp of Rs syntax.
3
Despite several valiant attempts, no graphical shell has gained wide acceptance in the R community:
by and large, the standard method of working with R is by writing scripts, or by typing commands
at the R prompt, much in the same way as one would write gretl scripts or work with the gretl
console. In this chapter, the focus will be on the methods available to execute R commands without
leaving gretl.
36.2 Starting an interactive R session
The easiest way to use R from gretl is in interactive mode. Once you have your data loaded in gretl,
you can select the menu item Tools, Start GNU R and an interactive R session will be started, with
your dataset automatically pre-loaded.
A simple example: OLS on cross-section data
For this example we use Ramanathans dataset data4-1, one of the sample les supplied with gretl.
We rst run, in gretl, an OLS regression of price on sqft, bedrms and baths. The basic results are
shown in Table 36.1.
Table 36.1: OLS house price regression via gretl
const 129.062 88.3033 1.4616 0.1746
sqft 0.154800 0.0319404 4.8465 0.0007
bedrms 21.587 27.0293 0.7987 0.4430
baths 12.192 43.2500 0.2819 0.7838
1
Rs homepage is at http://www.r-project.org/.
2
OK, who are we kidding? But its friendly competition!
3
The main reference for R documentation is http://cran.r-project.org/manuals.html. In addition, R tutorials
abound on the Net; as always, Google is your friend.
312
Chapter 36. Gretl and R 313
We will now replicate the above results using R. Select the menu item Tools, Start GNU R. A
window similar to the one shown in gure 36.1 should appear.
Figure 36.1: R window
The actual look of the R window may be somewhat dierent from what you see in Figure 36.1
(especially for Windows users), but this is immaterial. The important point is that you have a
window where you can type commands to R. If the above procedure doesnt work and no R window
opens, it means that gretl was unable to launch R. You should ensure that R is installed and working
on your system and that gretl knows where it is. The relevant settings can be found by selecting
the Tools, Preferences, General menu entry, under the Programs tab.
Assuming R was launched successfully, you will see notication that the data from gretl are avail-
able. In the background, gretl has arranged for two R commands to be executed, one to load the
gretl dataset in the form of a data frame (one of several forms in which R can store data) and one
to attach the data so that the variable names dened in the gretl workspace are available as valid
identiers within R.
In order to replicate gretls OLS estimation, go into the R window and type at the prompt
model <- lm(price ~ sqft + bedrms + baths)
summary(model)
You should see something similar to Figure 36.2. Surprisethe estimates coincide! To get out, just
close the R window or type q() at the R prompt.
Time series data
We now turn to an example which uses time series data: we will compare gretls and Rs estimates
of Box and Jenkins immortal airline model. The data are contained in the bjg sample dataset.
The following gretl code
open bjg
arima 0 1 1 ; 0 1 1 ; lg --nc
produces the estimates shown in Table 36.2.
Figure 36.2: OLS regression on house prices via R
Table 36.2: Airline model from Box and Jenkins (1976) selected portion of gretls estimates
1
0.401824 0.0896421 4.4825 0.0000
1
0.556936 0.0731044 7.6184 0.0000
Variance of innovations 0.00134810
Log-likelihood 244.696
Akaike information criterion 483.39
If we now open an R session as described in the previous subsection, the data-passing mechanism
is slightly dierent. Since our data were dened in gretl as time series, we use an R time-series
object (ts for short) for the transfer. In this way we can retain in R useful information such as the
periodicity of the data and the sample limits. The downside is that the names of individual series,
as dened in gretl, are not valid identiers. In order to extract the variable lg, one needs to use the
syntax lg <- gretldata[, "lg"].
ARIMA estimation can be carried out by issuing the following two R commands:
lg <- gretldata[, "lg"]
arima(lg, c(0,1,1), seasonal=c(0,1,1))
which yield
Coefficients:
ma1 sma1
-0.4018 -0.5569
s.e. 0.0896 0.0731
sigma^2 estimated as 0.001348: log likelihood = 244.7, aic = -483.4
Happily, the estimates again coincide.
36.3 Running an R script
Opening an R window and keying in commands is a convenient method when the job is small. In
some cases, however, it would be preferable to have R execute a script prepared in advance. One
way to do this is via the source() command in R. Alternatively, gretl oers the facility to edit an R
script and run it, having the current dataset pre-loaded automatically. This feature can be accessed
via the File, Script Files menu entry. By selecting User le, one can load a pre-existing R script;
if you want to create a new script instead, select the New script, R script menu entry.
Figure 36.3: Editing window for R scripts
In either case, you are presented with a window very similar to the editor window used for ordinary
gretl scripts, as in Figure 36.3.
There are two main dierences. First, you get syntax highlighting for Rs syntax instead of gretls.
Second, clicking on the Execute button (the gears icon), launches an instance of R in which your
commands are executed. Before R is actually run, you are asked if you want to run R interactively
or not (see Figure 36.4).
An interactive run opens an R instance similar to the one seen in the previous section: your data
will be pre-loaded (if the pre-load data box is checked) and your commands will be executed.
Once this is done, you will nd yourself at the R prompt, where you can enter more commands.
Figure 36.4: Editing window for R scripts
A non-interactive run, on the other hand, will execute your script, collect the output from R and
present it to you in an output window; R will be run in the background. If, for example, the script
in Figure 36.3 is run non-interactively, a window similar to Figure 36.5 will appear.
Figure 36.5: Output from a non-interactive R run
36.4 Taking stu back and forth
As regards the passing of data between the two programs, so far we have only considered passing
series from gretl to R. In order to achieve a satisfactory degree of interoperability, more is needed.
In the following sub-sections we see how matrices can be exchanged, and how data can be passed
from R back to gretl.
Passing matrices from gretl to R
For passing matrices from gretl to R, you can use the mwrite matrix function described in section
15.6. For example, the following gretl code fragment generates the matrix
A =
_
_
_
_
_
_
3 7 11
4 8 12
5 9 13
6 10 14
_
_
_
_
_
_
and stores it into the le mymatfile.mat.
matrix A = mshape(seq(3,14),4,3)
err = mwrite(A, "mymatfile.mat")
The R code to import such a matrix is
A <- as.matrix(read.table("mymatfile.mat", skip=1))
Although in principle you can give your matrix le any valid lename, a couple of conventions may
prove useful. First, you may want to use an informative le sux such as .mat, but this is a
matter of taste. More importantly, the exact location of the le created by mwrite could be an
issue. By default, if no path is specied in the le name, gretl stores matrix les in the current
work directory. However, it may be wise for the purpose at hand to use the directory in which gretl
stores all its temporary les, whose name is stored in the built-in string dotdir (see section 14.2).
The value of this string is automatically passed to R as the string variable gretl.dotdir, so the
above example may be rewritten as
Gretl side:
matrix A = mshape(seq(3,14),4,3)
err = mwrite(A, "@dotdir/mymatfile.mat")
R side:
fname <- paste(gretl.dotdir, "mymatfile.mat", sep="")
A <- as.matrix(read.table(fname, skip=1))
As a convenience, you can trac in matrix les via gretls dotdir in a more abbreviated form.
First, mwrite has an optional third (integer) argument: if this is non-zero then the matrix le is
automatically written into dotdir. Second, the function gretl.loadmat, which is dened when
you run R via gretl, provides a shortcut for the last two lines of R code above. The simplied version
of the transaction is then
Gretl side:
mwrite(A, "mymatfile.mat", 1)
R side:
A <- gretl.loadmat("mymatfile.mat")
Passing data from R to gretl
For passing data in the opposite direction, gretl denes a special function that can be used in the R
environment. An R object will be written as a temporary le in gretls dotdir directory, from where
it can be easily retrieved from within gretl.
The name of this function is gretl.export(), and it accepts one argument, the object to be ex-
ported. At present, the objects that can be exported with this method are matrices, data frames
and time-series objects. The function creates a text le, with the same name as the exported object,
in gretls temporary directory. Data frames and time-series objects are stored as CSV les, and can
be retrieved by using gretls append command. Matrices are stored in a special text format that is
understood by gretl (see section 15.6); the le sux is in this case .mat, and to read the matrix in
gretl you must use the mread() function.
As an example, we take the airline data and use them to estimate a structural time series model
la Harvey (1989). The model we will use is the Basic Structural Model (BSM), in which a time series
is decomposed into three terms:
y
t
=
t
+
t
+
t
where
t
is a trend component,
t
is a seasonal component and
t
is a noise term. In turn, the
following is assumed to hold:
t
=
t1
+
t
t
=
t
t
=
t
where
s
is the seasonal dierencing operator, (1 L
s
), and
t
,
t
and
t
are mutually uncorre-
lated white noise processes. The object of the analysis is to estimate the variances of the noise
components (which may be zero) and to recover estimates of the latent processes
t
(the level),
t
(the slope) and
t
.
Gretl does not provide (yet) a command for estimating this class of models, so we will use Rs
StructTS command and import the results back into gretl. Once the bjg dataset is loaded in gretl,
we pass the data to R and execute the following script:
# extract the log series
y <- gretldata[, "lg"]
# estimate the model
strmod <- StructTS(y)
# save the fitted components (smoothed)
compon <- as.ts(tsSmooth(strmod))
# save the estimated variances
vars <- as.matrix(strmod$coef)
# export into gretls temp dir
gretl.export(compon)
gretl.export(vars)
Running this script via gretl produces minimal output:
current data loaded as ts object "gretldata"
wrote /home/cottrell/.gretl/compon.csv
wrote /home/cottrell/.gretl/vars.mat
However, we are now able to pull the results back into gretl by executing the following commands,
either from the console or by creating a small script:
4
append @dotdir/compon.csv
vars = mread("@dotdir/vars.mat")
The rst command reads the estimated time-series components from a CSV le, which is the format
that the passing mechanism employs for series. The matrix vars is read from the le vars.mat.
4
This example will work on Linux and presumably on OSX without modications. On the Windows platform, you may
have to substitute the / character with \.
4.6
4.8
5
5.2
5.4
5.6
5.8
6
6.2
6.4
6.6
1949 1955 1961
lg
4.6
4.8
5
5.2
5.4
5.6
5.8
6
6.2
1949 1955 1961
level
0.01
0.01005
0.0101
0.01015
0.0102
0.01025
1949 1955 1961
slope
-0.25
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
1949 1955 1961
sea
Figure 36.6: Estimated components from BSM
After the above commands have been executed, three new series will have appeared in the gretl
workspace, namely the estimates of the three components; by plotting them together with the
original data, you should get a graph similar to Figure 36.6. The estimates of the variances can be
seen by printing the vars matrix, as in
? print vars
vars (4 x 1)
0.00077185
0.0000
0.0013969
0.0000
That is,

2
= 0.00077185,
2
= 0,
2
= 0.0013969,
2
= 0
Notice that, since
2
= 0, the estimate for

t
is constant and the level component is simply a
random walk with a drift.
36.5 Interacting with R from the command line
Up to this point we have spoken only of interaction with R via the GUI program. In order to do the
same from the command line interface, gretl provides the foreign command. This enables you to
embed non-native commands within a gretl script.
A foreign block takes the form
foreign language=R [--send-data] [--quiet]
... R commands ...
end foreign
and achieves the same eect as submitting the enclosed R commands via the GUI in the non-
interactive mode (see section 36.3 above). The --send-data option arranges for auto-loading of
the data present in the gretl session. The --quiet option prevents the output from R from being
echoed in the gretl output.
Example 36.1: Estimation of the Basic Structural Model simple
open bjg.gdt
foreign language=R --send-data
y <- gretldata[, "lg"]
strmod <- StructTS(y)
vars <- as.matrix(strmod$coef)
gretl.export(vars)
end foreign
rename level lg_level
rename slope lg_slope
rename sea lg_seas
vars = mread("@dotdir/vars.mat")
Using this method, replicating the example in the previous subsection is rather easy: basically, all it
takes is encapsulating the content of the R script in a foreign. . . end foreign block; see example
36.1.
The above syntax, despite being already quite useful by itself, shows its full power when it is used
in conjunction with user-written functions. Example 36.2 shows how to dene a gretl function that
calls R internally.
36.6 Performance issues with R
R is a large and complex program, which takes an appreciable time to initialize itself.
5
In interactive
use this not a signicant problem, but if you have a gretl script that calls R repeatedly the cumulated
start-up costs can become bothersome. To get around this, gretl calls the R shared library by
preference; in this case the start-up cost is borne only once, on the rst invocation of R code from
within gretl.
Support for the R shared library is built into the gretl packages for MS Windows and OS Xbut
the advantage is realized only if the library is in fact available at run time. If you are building gretl
yourself on Linux and wish to make use of the R library, you should ensure (a) that R has been
built with the shared library enabled (specify --enable-R-shlib when conguring your build of
R), and (b) that the pkg-config program is able to detect your R installation. We do not link to the
R library at build time, rather we open it dynamically on demand. The gretl GUI has an item under
the Tools/Preferences menu which enables you to select the path to the library, if it is not detected
automatically.
If you have the R shared library installed but want to force gretl to call the R executable instead,
you can do
5
About one third of a second on an Intel Core Duo machine of 2009 vintage.
Example 36.2: Estimation of the Basic Structural Model via a function
function list RStructTS(series myseries)
smpl ok(myseries) --restrict
sx = argname(myseries)
foreign language=R --send-data --quiet
@sx <- gretldata[, "myseries"]
strmod <- StructTS(@sx)
end foreign
rename level @sx_level
rename slope @sx_slope
rename sea @sx_seas
list ret = @sx_level @sx_slope @sx_seas
return ret
end function
# ------------ main -------------------------
open bjg.gdt
list X = RStructTS(lg)
set R_lib off
36.7 Further use of the R library
Besides improving performance, as noted above, use of the R shared library makes possible a
further renement. That is, you can dene functions in R, within a foreign block, then call those
functions later in your script much as if they were gretl functions. This is illustrated below.
set R_functions on
foreign language=R
plus_one <- function(q) {
z = q+1
invisible(z)
}
end foreign
scalar b=R.plus_one(2)
The R function plus_one is obviously trivial in itself, but the example shows a couple of points.
First, for this mechanism to work you need to enable R_functions via the set command. Second,
to avoid collision with the gretl function namespace, calls to functions dened in this way must be
prexed with R., as in R.plus_one.
Built-in R functions may also be called in this way, once R_functions is set on. For example one
can invoke Rs choose function, which computes binomial coecients:
set R_functions on
scalar b=R.choose(10,4)
Note, however, that the possibilities for use of built-in R functions are limited; only functions whose
arguments and return values are suciently generic (basically scalars or matrices) will work.
Chapter 37
Gretl and Ox
37.1 Introduction
Ox, written by Jurgen A. Doornik (see Doornik, 2007), is described by its author as an object-
oriented statistical system. At its core is a powerful matrix language, which is complemented by
a comprehensive statistical library. Among the special features of Ox are its speed [and] well-
designed syntax. . . . Ox comes in two versions: Ox Professional and Ox Console. Ox is available for
Windows, Linux, Mac (OS X), and several Unix platforms. (www.doornik.com)
Ox is proprietary, closed-source software. The command-line version of the program is, however,
available free of change for academic users. Quoting again from Doorniks website: The Console
(command line) versions may be used freely for academic research and teaching purposes only. . . .
The Ox syntax is public, and, of course, you may do with your own Ox code whatever you wish.
If you wish to use Ox in conjunction with gretl please refer to doornik.com for further details on
licensing.
As the reader will no doubt have noticed, all the other software that we discuss in this Guide is
open-source and freely available for all users. We make an exception for Ox on the grounds that it
is indeed fast and well designed, and that its statistical libraryalong with various add-on packages
that are also availablehas exceptional coverage of cutting-edge techniques in econometrics. The
gretl authors have used Ox for benchmarking some of gretls more advanced features such as
dynamic panel models and state space models.
1
37.2 Ox support in gretl
The support oered for Ox in gretl is similar to that oered for R, as discussed in chapter 36, but
with a few dierences. The rst dierence to note is that Ox support is not on by default; it must
be enabled explicitly.
To enable support for Ox, go to the Tools/Preferences/General menu item and check the box labeled
Enable Ox support. Click OK in the preferences dialog, then quit and restart gretl. You will now nd,
under the Programs tab in the Tools/Preferences/General dialog, an entry for specifying the path to the oxl
executable, that is, the program that runs Ox les (on MS Windows it is called oxl.exe). Make sure that path
is right, and youre ready to go.
With support enabled, you can open and edit Ox programs in the gretl GUI. Clicking the execute
icon in the editor window will send your code to Ox for execution. Figures 37.1 and Figure 37.2
show an Ox program and part of its output.
In addition you can embed Ox code within a gretl script using a foreign block, as described in
connection with R. A trivial example, which simply prints the gretl data matrix within Ox, is shown
below:
open data4-1
matrix m = { dataset }
mwrite(m, "@dotdir/gretl.mat")
1
For a review of Ox, see Cribari-Neto and Zarkos (2003) and for a (somewhat dated) comparison of Ox with other
matrix-oriented packages such as GAUSS, see Steinhaus (1999).
323
Chapter 37. Gretl and Ox 324
Figure 37.1: Ox editing window
Figure 37.2: Output from Ox
foreign language=Ox
#include <oxstd.h>
main()
{
decl gmat = gretl_loadmat("gretl.mat");
print(gmat);
}
end foreign
The above example illustrates how a matrix can be passed from gretl to Ox. We use the mwrite
function to write a matrix into the users dotdir (see section 14.2), then in Ox we use the function
gretl_loadmat to retrieve the matrix.
How does gretl_loadmat come to be dened? When gretl writes out the Ox program correspond-
ing to your foreign block it does two things in addition. First, it writes a small utility le named
gretl_io.ox into your dotdir. This contains a denition for gretl_loadmat and also for the
function gretl_export (see below). Second, gretl interpolates into your Ox code a line which in-
cludes this utility le (it is inserted right after the inclusion of oxstd.h, which is needed in all Ox
programs). Note that gretl_loadmat expects to nd the named le in the users dotdir.
37.3 Illustration: replication of DPD model
Example 37.1 shows a more ambitious case. This script replicates one of the dynamic panel data
models in Arellano and Bond (1991), rst using gretl and then using Ox; we then check the relative
dierences between the parameter estimates produced by the two programs (which turn out to be
reassuringly small).
Unlike the previous example, in this case we pass the dataset from gretl to Ox as a CSV le in
order to preserve the variable names. Note the use of the internal variable csv_na to get the right
representation of missing values for use with Oxand also note that the --send-data option for
the foreign command is not available in connection with Ox.
We get the parameter estimates back from Ox using gretl_export on the Ox side and mread on
the gretl side. The gretl_export function takes two arguments, a matrix and a le name. The le
is written into the users dotdir, from where it can be picked up using mread. The nal portion of
the output from Example 37.1 is shown below:
? matrix oxparm = mread("/home/cottrell/.gretl/oxparm.mat")
Generated matrix oxparm
? eval abs((parm - oxparm) ./ oxparm)
1.4578e-13
3.5642e-13
5.0672e-15
1.6091e-13
8.9808e-15
2.0450e-14
1.0218e-13
2.1048e-13
9.5898e-15
1.8658e-14
2.1852e-14
2.9451e-13
1.9398e-13
Example 37.1: Estimation of dynamic panel data model via gretl and Ox
open abdata.gdt
# Take first differences of the independent variables
genr Dw = diff(w)
genr Dk = diff(k)
genr Dys = diff(ys)
# 1-step GMM estimation
arbond 2 ; n Dw Dw(-1) Dk Dys Dys(-1) 0 --time-dummies
matrix parm = $coeff
# Write CSV file for Ox
set csv_na .NaN
store @dotdir/abdata.csv
# Replicate using the Ox DPD package
foreign language=Ox
#include <oxstd.h>
#import <packages/dpd/dpd>
main ()
{
decl dpd = new DPD();
dpd.Load("@dotdir/abdata.csv");
dpd.SetYear("YEAR");
dpd.Select(I_VAR, {"w", 0, 1, "k", 0, 0, "ys", 0, 1});
dpd.Gmm("n", 2, 99); // GMM-type instrument
dpd.SetTest(2, 2); // Sargan, AR 1-2 tests
dpd.Estimate(); // 1-step estimation
decl parm = dpd.GetPar();
gretl_export(parm, "oxparm.mat");
delete dpd;
}
end foreign
# Compare the results
matrix oxparm = mread("@dotdir/oxparm.mat")
eval abs((parm - oxparm) ./ oxparm)
Chapter 38
Gretl and Octave
38.1 Introduction
GNU Octave, written by John W. Eaton and others, is described as a high-level language, primar-
ily intended for numerical computations. The program is oriented towards solving linear and
nonlinear problems numerically and performing other numerical experiments using a language
that is mostly compatible with Matlab. (www.gnu.org/software/octave) Octave is available in
source-code form (naturally, for GNU software) and also in the form of binary packages for MS Win-
dows and Mac OS X. Numerous contributed packages that extend Octaves functionality in various
ways can be found at octave.sf.net.
38.2 Octave support in gretl
The support oered for Octave in gretl is similar to that oered for R (chapter 36). For example,
you can open and edit Octave scripts in the gretl GUI. Clicking the execute icon in the editor
window will send your code to Octave for execution. Figures 38.1 and Figure 38.2 show an Octave
script and its output; in this example we use the function logistic_regression to replicate some
results from Greene (2000).
In addition you can embed Octave code within a gretl script using a foreign block, as described
in connection with R. A trivial example, which simply loads and prints the gretl data matrix within
Octave, is shown below. (Note that in Octave, appending ; to a line suppresses verbose output;
leaving o the semicolon results in printing of the object that is produced, if any.)
open data4-1
matrix m = { dataset }
mwrite(m, "@dotdir/gretl.mat")
foreign language=Octave
gmat = gretl_loadmat("gretl.mat")
end foreign
We use the mwrite function to write a matrix into the users dotdir (see section 14.2), then in Oc-
tave we use the function gretl_loadmat to retrieve the matrix. The magic behind gretl_loadmat
works in essentially the same way as for Ox (chapter 37).
38.3 Illustration: spectral methods
We now present a more ambitious example which exploits Octaves handling of the frequency
domain (and also its ability to use code written for MATLAB), namely estimation of the spec-
tral coherence of two time series. For this illustration we require two extra Octave packages
from octave.sf.net, namely those supporting spectral functions (specfun) and signal process-
ing (signal). After downloading the packages you can install them from within Octave as follows
(using version numbers as of March 2010):
pkg install specfun-1.0.8.tar.gz
pkg install signal-1.0.10.tar.gz
327
Chapter 38. Gretl and Octave 328
Figure 38.1: Octave editing window
Figure 38.2: Output from Octave
In addition we need some specialized MATLAB les made available by Mario Forni of the Univer-
sity of Modena, at http://morgana.unimore.it/forni_mario/matlab.htm. The les needed are
coheren2.m, coheren.m, coher.m, cospec.m, crosscov.m, crosspec.m, crosspe.m and spec.m.
These are in a form appropriate for MS Windows. On Linux you could run the following shell script
to get the les and remove the Windows end-of-le character (which prevents the les from running
under Octave):
SITE=http://morgana.unimore.it/forni_mario/MYPROG
# download files and delete trailing Ctrl-Z
for f in \
coheren2.m \
coheren.m \
coher.m \
cospec.m \
crosscov.m \
crosspec.m \
crosspe.m \
spec.m ; do
wget $SITE/$f && \
cat $f | tr -d \\032 > tmp.m && mv tmp.m $f
done
The Forni les should be placed in some appropriate directory, and you should tell Octave where
to nd them by adding that directory to Octaves loadpath. On Linux this can be done via an entry
in ones ~/.octaverc le. For example
addpath("~/stats/octave/forni");
Alternatively, an addpath directive can be written into the Octave script that calls on these les.
With everything set up on the Octave side we now write a gretl script (see Example 38.1) which
opens a time-series dataset, constructs and writes a matrix containing two series, and denes a
foreign block containing the Octave statements needed to produce the spectral coherence matrix.
The matrix is exported via the gretl_export function, which is automatically dened for you; this
function takes two arguments, a matrix and a le name. The le is written into the users dotdir,
from where it can be picked up using mread. Finally, we produce a graph from the matrix in gretl.
In the script this is sent to the screen; Figure 38.3 shows the same graph in PDF format.
Example 38.1: Estimation of spectral coherence via Octave
open data9-7
matrix xy = { PRIME, UNEMP }
mwrite(xy, "@dotdir/xy.mat")
foreign language=Octave
# uncomment and modify the following if necessary
# addpath("~/stats/octave/forni");
xy = gretl_loadmat("xy.mat");
x = xy(:,1);
y = xy(:,2);
# note: the last parameter is the Bartlett window size
h = coher(x, y, 8);
gretl_export(h, "h.mat");
end foreign
h = mread("@dotdir/h.mat")
colnames(h, "coherence")
gnuplot 1 --time-series --with-lines --matrix=h --output=display
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
0 10 20 30 40 50 60
c
o
h
e
r
e
n
c
e
Figure 38.3: Spectral coherence estimated via Octave
Chapter 39
Gretl and Stata
Stata (www.stata.com) is closed-source, proprietary (and expensive) software and as such is not a
natural companion to gretl. Nonetheless, given Statas popularity it is desirable to have a convenient
way of comparing results across the two programs, and to that end we provide limited support for
Stata code under the foreign command.
The following example illustrates whats available. You can send the current gretl dataset to Stata
using the --send-data ag. And having dened a matrix within Stata you can export it for use with
gretl via the gretl_export command: this takes two arguments, the name of the matrix to export
and the lename to use; the le is written to the users dotdir, from where it can be retrieved
using the mread() function.
1
To suppress printed output from Stata you can add the --quiet ag
to the foreign block.
Example 39.1: Comparison of clustered standard errors with Stata
function matrix stata_reorder (matrix se)
# stata puts the intercept last, but gretl puts it first
scalar n = rows(se)
return se[n] | se[1:n-1]
end function
open data4-1
ols 1 0 2 3 --cluster=bedrms
matrix se = $stderr
foreign language=stata --send-data
regress price sqft bedrms, vce(cluster bedrms)
matrix vcv = e(V)
gretl_export vcv "vcv.mat"
end foreign
matrix stata_vcv = mread("@dotdir/vcv.mat")
stata_se = stata_reorder(sqrt(diag(stata_vcv)))
matrix check = se - stata_se
print check
Note that there is no support for editing Stata scripts via the gretl GUI. Also note that Stata coerces
all variable names to lower-case on data input, so even if series names in gretl are upper-case, or of
mixed case, its necessary to use all lower-case in Stata.
1
We do not currently oer the complementary functionality of gretl_loadmat, which enables reading of matrices
written by gretls mwrite() function in Ox and Octave. This is not at all easy to implement in Stata code.
331
Chapter 40
Gretl and Python
40.1 Introduction
According to www.python.org, Python is an easy to learn, powerful programming language. It has
ecient high-level data structures and a simple but eective approach to object-oriented program-
ming. Pythons elegant syntax and dynamic typing, together with its interpreted nature, make it an
ideal language for scripting and rapid application development in many areas on most platforms.
Indeed, Python is widely used in a great variety of contexts. Numerous add-on modules are avail-
able; the ones likely to be of greatest interest to econometricians include NumPy (the fundamen-
tal package for scientic computing with Pythonsee www.numpy.org); SciPy (which builds on
NumPysee www.scipy.org); and Statsmodels (http://statsmodels.sourceforge.net/).
40.2 Python support in gretl
The support oered for Python in gretl is similar to that oered for Octave (chapter 38). You can
open and edit Python scripts in the gretl GUI. Clicking the execute icon in the editor window will
send your code to Python for execution. In addition you can embed Python code within a gretl
script using a foreign block, as described in connection with R.
When you launch Python from within gretl one variable and two convenience functions are pre-
dened, as follows.
gretl_dotdir
gretl_loadmat(filename, autodot=1)
gretl_export(M, filename, autodot=1)
The variable gretl_dotdir holds the path to the users dot directory. The rst function loads a
matrix of the given filename as written by gretls mwrite function, and the second writes matrix
M, under the given filename, in the format wanted by gretl.
By default the trac in matrices goes via the dot directory on the Python side; that is, the name of
this directory is prepended to filename for both reading and writing. (This is complementary to
use of the export and import parameters with gretls mwrite and mread functions, respectively.)
However, if you wish to take control over the reading and writing locations you can supply a zero
value for autodot when calling gretl_loadmat and gretl_export: in that case the filename
argument is used as is.
Note that gretl_loadmat and gretl_export depend on NumPy; they make use of the functions
loadtxt and savetxt respectively. Nonetheless, the presence of NumPy is not an absolute require-
ment if you dont need to use these two functions.
40.3 Illustration: linear regression with multicollinearity
Example 40.1 compares the numerical accuracy of gretls ols command with that of NumPys
linalg.lstsq, using the notorious Longley test data which exhibit extreme multicollinearity. Un-
like some econometrics packages, NumPy does a good job on these data. The script computes and
332
Chapter 40. Gretl and Python 333
prints the log-relative error in estimation of the regression coecients, using the NIST-certied val-
ues as a benchmark;
1
the error values correspond to the number of correct digits (with a maximum
of 15). The results will dier somewhat by computer architecture; the output shown was obtained
on a 32-bit Linux Intel i5 system.
1
See http://www.itl.nist.gov/div898/strd/lls/data/Longley.shtml.
Chapter 40. Gretl and Python 334
Example 40.1: Comparing regression results with Python
set echo off
set messages off
function matrix logrel_error (matrix est, matrix true)
return -log10(abs(est - true) ./ abs(true))
end function
open longley.gdt -q
list LX = prdefl .. year
# gretls regular OLS
ols employ 0 LX -q
matrix b = $coeff
mwrite({employ}, "y.mat", 1)
mwrite({const} ~ {LX}, "X.mat", 1)
foreign language=python
import numpy as np
y = gretl_loadmat(y.mat, 1)
X = gretl_loadmat(X.mat, 1)
# NumPys OLS
b = np.linalg.lstsq(X, y)[0]
gretl_export(np.transpose(np.matrix(b)), py_b.mat, 1)
end foreign
# NISTs certified coefficient values
matrix nist_b = {-3482258.63459582, 15.0618722713733,
-0.358191792925910E-01, -2.02022980381683,
-1.03322686717359, -0.511041056535807E-01,
1829.15146461355}
matrix py_b = mread("py_b.mat", 1)
matrix errs = logrel_error(b, nist_b) ~ logrel_error(py_b, nist_b)
colnames(errs, "gretl python")
printf "Log-relative errors, Longley coefficients:\n\n%12.5g\n", errs
printf "Column means\n%12.5g\n", meanc(errs)
Output:
Log-relative errors, Longley coefficients:
gretl python
12.844 12.85
11.528 11.414
12.393 12.401
13.135 13.121
13.738 13.318
12.587 12.363
12.848 12.852
Column means
12.725 12.617
Chapter 41
Troubleshooting gretl
41.1 Bug reports
Bug reports are welcomewell, if not exactly welcome then useful and appreciated. Hopefully, you
are unlikely to nd bugs in the actual calculations done by gretl (although this statement does not
constitute any sort of warranty). You may, however, come across bugs or oddities in the behavior
of the graphical interface. Please remember that the usefulness of bug reports is greatly enhanced
if you can be as specic as possible: what exactly went wrong, under what conditions, and on what
operating system? If you saw an error message, what precisely did it say?
One way of making a bug report more useful is to run the program in such a way that you can see
(and copy) any additional information that gets printed to the stderr output stream. On Linux and
Mac OS X thats just a matter of launching gretl from the command prompt in a terminal window.
On MS Windows its a bit more complicated since stderr is by default invisble. However, you can
quite easily set up a special gretl shortcut that does the job. On the Windows desktop, right-click
and select New shortcut. In the dialog box that appears, browse to nd gretl.exe and append
the --debug ag, as shown in Figure 41.1. Note that there are two dashes before debug.
Figure 41.1: Creating a debugging shortcut
335
Chapter 41. Troubleshooting gretl 336
When you start gretl in this mode, a console window appears as well as the gretl window, and
stderr output goes to the console. To copy this output, click at the top left of the console window
for a menu (Figure 41.2): rst do Select all, then Copy. You can paste the results into Notepad
or similar.
Figure 41.2: The program with console window
41.2 Auxiliary programs
As mentioned above, gretl calls some other programs to accomplish certain tasks (gnuplot for
graphing, L
A
T
E
X for high-quality typesetting of regression output, GNU R). If something goes wrong
with such external links, it is not always easy for gretl to produce an informative error message.
If such a link fails when accessed from the gretl graphical interface, you may be able to get more
information by starting gretl from the command prompt rather than via a desktop menu entry or
icon. On the X window system, start gretl from the shell prompt in an xterm; on MS Windows, start
the program gretl.exe from a console window or DOS box using the -g or --debug option ag.
Additional error messages may be displayed on the terminal window.
Also please note that for most external calls, gretl assumes that the programs in question are
available in your paththat is, that they can be invoked simply via the name of the program,
without supplying the programs full location.
1
Thus if a given program fails, try the experiment of
typing the program name at the command prompt, as shown below.
Graphing Typesetting GNU R
X window system gnuplot pdatex R
MS Windows wgnuplot.exe pdatex RGui.exe
If the program fails to start from the prompt, its not a gretl issue but rather that the programs
home directory is not in your path, or the program is not installed (properly). For details on
modifying your path please see the documentation or online help for your operating system or
shell.
1
The exception to this rule is the invocation of gnuplot under MS Windows, where a full path to the program is given.
Chapter 42
The command line interface
The gretl package includes the command-line program gretlcli. On Linux it can be run from a
terminal window (xterm, rxvt, or similar), or at the text console. Under MS Windows it can be run in
a console window (sometimes inaccurately called a DOS box). gretlcli has its own help le, which
may be accessed by typing help at the prompt. It can be run in batch mode, sending output
directly to a le (see also the Gretl Command Reference).
If gretlcli is linked to the readline library (this is automatically the case in the MS Windows version;
also see Appendix C), the command line is recallable and editable, and oers command completion.
You can use the Up and Down arrow keys to cycle through previously typed commands. On a given
command line, you can use the arrow keys to move around, in conjunction with Emacs editing
keystrokes.
1
The most common of these are:
Keystroke Eect
Ctrl-a go to start of line
Ctrl-e go to end of line
Ctrl-d delete character to right
where Ctrl-a means press the a key while the Ctrl key is also depressed. Thus if you want
to change something at the beginning of a command, you dont have to backspace over the whole
line, erasing as you go. Just hop to the start and add or delete characters. If you type the rst
letters of a command name then press the Tab key, readline will attempt to complete the command
name for you. If theres a unique completion it will be put in place automatically. If theres more
than one completion, pressing Tab a second time brings up a list.
Probably the most useful mode for heavy-duty work with gretlcli is batch (non-interactive) mode,
in which the program reads and processes a script, and sends the output to le. For example
gretlcli -b scriptfile > outputfile
Note that scriptle is treated as a program argument; only the output le requires redirection (>).
Dont forget the -b (batch) switch, otherwise the program will wait for user input after executing
the script (and if output is redirected, the program will appear to hang).
1
Actually, the key bindings shown below are only the defaults; they can be customized. See the readline manual.
337
Part IV
Appendices
338
Appendix A
Data le details
A.1 Basic native format
In gretls native data format, a data set is stored in XML (extensible mark-up language). Data
les correspond to the simple DTD (document type denition) given in gretldata.dtd, which is
supplied with the gretl distribution and is installed in the system data directory (e.g. /usr/share/
gretl/data on Linux.) Data les may be plain text or gzipped. They contain the actual data values
plus additional information such as the names and descriptions of variables, the frequency of the
data, and so on.
Most users will probably not have need to read or write such les other than via gretl itself, but
if you want to manipulate them using other software tools you should examine the DTD and also
take a look at a few of the supplied practice data les: data4-1.gdt gives a simple example;
data4-10.gdt is an example where observation labels are included.
A.2 Traditional ESL format
For backward compatibility, gretl can also handle data les in the traditional format inherited
from Ramu Ramanathans ESL program. In this format a data set is represented by two text les;
one contains the actual data and the other information on how the data should be read. To be more
specic:
1. Actual data: A rectangular matrix of white-space separated numbers. By default, each column
represents a variable, each row an observation. The data columns can be separated by spaces
or tabs. Traditionally the data lename has no extension/sux.
2. Header: The data le must be accompanied by a header le which has the same basename as
the data le plus the sux .hdr. This le contains, in order:
(Optional) comments on the data, set o by the opening string (* and the closing string
*), each of these strings to occur on lines by themselves.
(Required) a list of the names of the variables in the data le, separated by white space.
Names are limited to 8 characters, must start with a letter, and are limited to alphanu-
meric characters plus the underscore. The list may continue over more than one line; it
should be terminated with a semicolon.
(Required) an observations line of the form 1 1 85. The rst element gives the data
frequency (1 for undated or annual data, 4 for quarterly, 12 for monthly). The second
and third elements give the starting and ending observations. These should be 1 and the
number of observations, respectively, for undated data. For time-series data one can use
dates of the form 1959.1 (quarterly, one digit after the point) or 1967.03 (monthly, two
digits after the point).
The keyword BYOBS (but see below).
Here is an example of a well-formed data header le; the corresponding data le contains three
columns of data, each having 90 entries.
339
Appendix A. Data le details 340
(*
DATA9-6:
Data on log(money), log(income) and interest rate from US.
Source: Stock and Watson (1993) Econometrica
(unsmoothed data) Period is 1900-1989 (annual data).
Data compiled by Graham Elliott.
*)
lmoney lincome intrate ;
1 1900 1989 BYOBS
Three further features of the traditional ESL data format may be noted.
1. If the BYOBS keyword is replaced by BYVAR this indicates that in the corresponding data le
the data are written out by variable rather than by observation.
2. If BYOBS is followed by the keyword MARKERS, gretl expects a data le in which the rst column
contains strings (8 characters maximum) used to identify the observations. This may be useful
in the case of cross-sectional data where the units of observation are identiable: countries,
states, cities or whatever. It can also be useful for irregular time series data, such as daily
stock price data where some days are not trading daysin this case the observations can be
marked with a date string such as 10/01/98. (Remember the 8-character maximum.) Note
that BYVAR and MARKERS are mutually exclusive ags. Also note that the markers are not
considered to be a variable: this column does not have a corresponding entry in the list of
variable names in the header le.
3. If a le with the same base name as the data le and header les, but with the sux .lbl, is
found, it is read to ll out the descriptive labels for the data series. The format of the label le
is simple: each line contains the name of one variable (as found in the header le), followed
by one or more spaces, followed by the descriptive label. Here is an example, giving a label
for a variable named price: price New car price index, 1982 base year
If you want to save data in traditional format, use the --traditional ag with the store com-
mand, either in the command-line program or in the console window of the GUI program.
A.3 Binary database details
A gretl database consists of two parts: an ASCII index le (with lename sux .idx) containing
information on the series, and a binary le (sux .bin) containing the actual data. Two examples
of the format for an entry in the idx le are shown below:
G0M910 Composite index of 11 leading indicators (1987=100)
M 1948.01 - 1995.11 n = 575
currbal Balance of Payments: Balance on Current Account; SA
Q 1960.1 - 1999.4 n = 160
The rst eld is the series name. The second is a description of the series (maximum 128 charac-
ters). On the second line the rst eld is a frequency code: M for monthly, Q for quarterly, A for
annual, B for business-daily (daily with ve days per week) and D for daily (seven days per week).
No other frequencies are accepted at present. Then comes the starting date (N.B. with two digits
following the point for monthly data, one for quarterly data, none for annual), a space, a hyphen,
another space, the ending date, the string n = and the integer number of observations. In the
case of daily data the starting and ending dates should be given in the form YYYY/MM/DD. This
format must be respected exactly.
Optionally, the rst line of the index le may contain a short comment (up to 64 characters) on the
source and nature of the data, following a hash mark. For example:
Appendix A. Data le details 341
# Federal Reserve Board (interest rates)
The corresponding binary database le holds the data values, represented as oats, that is, single-
precision oating-point numbers, typically taking four bytes apiece. The numbers are packed by
variable, so that the rst n numbers are the observations of variable 1, the next m the observations
on variable 2, and so on.
Appendix B
Data import via ODBC
Since version 1.7.5, gretl provides a method for retrieving data from databases which support the
Open Database Connectivity (ODBC) standard. Most users wont be interested in this, but there may
be some for whom this feature matters a lot typically, those who work in an environment where
huge data collections are accessible via a Data Base Management System (DBMS).
In the following section we explain what is needed for ODBC support in gretl. We provide some
background information on how ODBC works in section B.2, and explain the details of getting gretl
to retrieve data from a database in section B.3.
B.1 ODBC support
The piece of software that bridges between gretl and the ODBC system is a dynamically loaded
plugin. This is included in the gretl packages for MS Windows and Mac OS X (on OS X support
was added in gretl 1.9.0). On other unix-type platforms (notably Linux) you will have to build gretl
from source to get ODBC support. This is because the gretl plugin depends on having unixODBC
installed, which we cannot assume to be the case on typical Linux systems. To enable the ODBC
plugin when building gretl, you must pass the option --with-odbc to gretls configure script. In
addition, if unixODBC is installed in a non-standard location you will have to specify its installation
prex using --with-ODBC-prefix, as in (for example)
./configure --with-odbc --with-ODBC-prefix=/opt/ODBC
B.2 ODBC base concepts
ODBC is short for Open DataBase Connectivity, a group of software methods that enable a client to
interact with a database server. The most common operation is when the client fetches some data
from the server. ODBC acts as an intermediate layer between client and server, so the client talks
to ODBC rather than accessing the server directly (see Figure B.1).
ODBC
query
data
Figure B.1: Retrieving data via ODBC
For the above mechanism to work, it is necessary that the relevant ODBC software is installed
and working on the client machine (contact your DB administrator for details). At this point, the
database (or databases) that the server provides will be accessible to the client as a data source
with a specic identier (a Data Source Name or DSN); in most cases, a username and a password
are required to connect to the data source.
Once the connection is established, the user sends a query to ODBC, which contacts the database
manager, collects the results and sends them back to the user. The query is almost invariably
342
Appendix B. Data import via ODBC 343
formulated in a special language used for the purpose, namely SQL.
1
We will not provide here an
SQL tutorial: there are many such tutorials on the Net; besides, each database manager tends to
support its own SQL dialect so the precise form of an SQL query may vary slightly if the DBMS on
the other end is Oracle, MySQL, PostgreSQL or something else.
Suce it to say that the main statement for retrieving data is the SELECT statement. Within a DBMS,
data are organized in tables, which are roughly equivalent to spreadsheets. The SELECT statement
returns a subset of a table, which is itself a table. For example, imagine that the database holds a
table called NatAccounts, containing the data shown in Table B.1.
year qtr gdp consump tradebal
1970 1 584763 344746.9 5891.01
1970 2 597746 350176.9 7068.71
1970 3 604270 355249.7 8379.27
1970 4 609706 361794.7 7917.61
1971 1 609597 362490 6274.3
1971 2 617002 368313.6 6658.76
1971 3 625536 372605 4795.89
1971 4 630047 377033.9 6498.13
Table B.1: The NatAccounts table
The SQL statement
SELECT qtr, tradebal, gdp FROM NatAccounts WHERE year=1970;
produces the subset of the original data shown in Table B.2.
qtr tradebal gdp
1 5891.01 584763
2 7068.71 597746
3 8379.27 604270
4 7917.61 609706
Table B.2: Result of a SELECT statement
Gretl provides a mechanism for forwarding your query to the DBMS via ODBC and including the
results in your currently open dataset.
B.3 Syntax
At present gretl does not oer a graphical interface for ODBC import; this must be done via the
command line interface. The two commands used for fetching data via an ODBC connection are
open and data.
The open command is used for connecting to a DBMS: its syntax is
open dsn=database [user=username] [password=password] --odbc
The user and password items are optional; the eect of this command is to initiate an ODBC
connection. It is assumed that the machine gretl runs on has a working ODBC client installed.
In order to actually retrieve the data, the data command is used. Its syntax is:
1
See http://en.wikipedia.org/wiki/SQL.
data series [obs-format=format-string] query=query-string --odbc
where:
series is a list of names of gretl series to contain the incoming data, separated by spaces. Note that
these series need not exist pior to the ODBC import.
format-string is an optional parameter, used to handle cases when a rectangular organisation of
the database cannot be assumed (more on this later);
query-string is a string containing the SQL statement used to extract the data.
2
There should be no spaces around the equals signs in the obs-format and query elds in the data
command.
The query-string can, in principle, contain any valid SQL statement which results in a table. This
string may be specied directly within the command, as in
data x query="SELECT foo FROM bar" --odbc
which will store into the gretl variable x the content of the column foo from the table bar. However,
since in a real-life situation the string containing the SQL statement may be rather long, it may be
best to store it in a string variable. For example:
string SqlQry = "SELECT foo1, foo2 FROM bar"
data x y query=SqlQry --odbc
The observation format specier
If the optional parameter obs-format is absent, as in the above example, the SQL query should
return k columns of data, where k is the number of series names listed in the data command.
It may be necessary to include a smpl command before the data command to set up the right
window for the incoming data. In addition, if one cannot assume that the data will be delivered
in the correct order (typically, chronological order), the SQL query should contain an appropriate
ORDER BY clause.
The optional format string is used for those cases when there is no certainty that the data from the
query will arrive in the same order as the gretl dataset. This may happen when missing values are
interspersed within a column, or with data that do not have a natural ordering, e.g. cross-sectional
data. In this case, the SQL statement should return a table with m+k columns, where the rst m
columns are used to identify the observation or row in the gretl dataset into which the actual data
values in the nal k columns should be placed. The obs-format string is used to translate the rst
m elds into a string which matches the string gretl uses to identify observations in the currently
open dataset. Up to three columns can be used for this purpose (m 3).
Note that the strings gretl uses to identify observations can be seen by printing any variable by
observation, as in
print index --byobs
(The series named index is automatically added to a dataset created via the nulldata command.)
The format speciers available for use with obs-format are as follows:
%d print an integer value
%s print an string value
%g print a oating-point value
2
Prior to gretl 1.8.8, the tag query= was not required (or accepted) before the query string, and only one series
could be imported at a time. This variant is still accepted for the sake of backward compatibility.
In addition the format can include literal characters to be passed through, such as slashes or colons,
to make the resulting string compatible with gretls observation identiers.
For example, consider the following ctitious case: we have a 5-days-per-week dataset, to which we
want to add the stock index for the Verdurian market;
3
it so happens that in Verduria Saturdays
are working days but Wednesdays are not. We want a column which does not contain data on
Saturdays, because we wouldnt know where to put them, but at the same time we want to place
missing values on all the Wednesdays.
In this case, the following syntax could be used
string QRY="SELECT year,month,day,VerdSE FROM AlmeaIndexes"
data y obs-format="%d/%d/%d" query=QRY --odbc
The column VerdSE holds the data to be fetched, which will go into the gretl series y. The rst
three columns are used to construct a string which identies the day. Daily dates take the form
YYYY/MM/DD in gretl. If a row from the DBMS produces the observation string 2008/04/01 this will
match OK (its a Tuesday), but 2008/04/05 will not match since it is a Saturday; the corresponding
row will therefore be discarded. On the other hand, since no string 2008/04/23 will be found in
the data coming from the DBMS (its a Wednesday), that entry is left blank in our series y.
B.4 Examples
Table Consump Table DATA
Field Type
time decimal(7,2)
income decimal(16,6)
consump decimal(16,6)
Field Type
year decimal(4,0)
qtr decimal(1,0)
varname varchar(16)
xval decimal(20,10)
Table B.3: Example AWM database structure
Table Consump Table DATA
1970.00 424278.975500 344746.944000
1970.25 433218.709400 350176.890400
1970.50 440954.219100 355249.672300
1970.75 446278.664700 361794.719900
1971.00 447752.681800 362489.970500
1971.25 453553.860100 368313.558500
1971.50 460115.133100 372605.015300
. . .
1970 1 CAN 517.9085000000
1970 2 CAN 662.5996000000
1970 3 CAN 1130.4155000000
1970 4 CAN 467.2508000000
1970 1 COMPR 18.4000000000
1970 2 COMPR 18.6341000000
1970 3 COMPR 18.3000000000
1970 4 COMPR 18.2663000000
1970 1 D1 1.0000000000
1970 2 D1 0.0000000000
. . .
Table B.4: Example AWM database data
In the following examples, we will assume that access is available to a database known to ODBC
with the data source name AWM, with username Otto and password Bingo. The database
AWM contains quarterly data in two tables (see B.3 and B.4):
3
See http://www.almeopedia.com/index.php/Verduria.
The table Consump is the classic rectangular dataset; that is, its internal organization is the same
as in a spreadsheet or econometrics package: each row is a data point and each column is a variable.
The structure of the DATA table is dierent: each record is one gure, stored in the column xval,
and the other elds keep track of which variable it belongs to, for which date.
Example B.1: Simple query from a rectangular table
nulldata 160
setobs 4 1970:1 --time
open dsn=AWM user=Otto password=Bingo --odbc
string Qry = "SELECT consump, income FROM Consump"
data cons inc query=Qry --odbc
Example B.1 shows a query for two series: rst we set up an empty quarterly dataset. Then we
connect to the database using the open statement. Once the connection is established we retrieve
two columns from the Consump table. No observation string is required because the data already
have a suitable structure; we need only import the relevant columns.
Example B.2: Simple query from a non-rectangular table
string S = "select year, qtr, xval from DATA \
where varname=WLN ORDER BY year, qtr"
data wln obs-format="%d:%d" query=S --odbc
In example B.2, by contrast, we make use of the observation string since we are drawing from the
DATA table, which is not rectangular. The SQL statement stored in the string S produces a table with
three columns. The ORDER BY clause ensures that the rows will be in chronological order, although
this is not strictly necessary in this case.
Example B.3: Handling of missing values for a non-rectangular table
string foo = "select year, qtr, xval from DATA \
where varname=STN AND qtr>1"
data bar obs-format="%d:%d" query=foo --odbc
print bar --byobs
Example B.3 shows what happens if the rows in the outcome from the SELECT statement do not
match the observations in the currently open gretl dataset. The query includes a condition which
lters out all the data from the rst quarter. The query result (invisible to the user) would be
something like
+------+------+---------------+
| year | qtr | xval |
+------+------+---------------+
| 1970 | 2 | 7.8705000000 |
| 1970 | 3 | 7.5600000000 |
| 1970 | 4 | 7.1892000000 |
| 1971 | 2 | 5.8679000000 |
| 1971 | 3 | 6.2442000000 |
| 1971 | 4 | 5.9811000000 |
| 1972 | 2 | 4.6883000000 |
| 1972 | 3 | 4.6302000000 |
...
Internally, gretl lls the variable bar with the corresponding value if it nds a match; otherwise, NA
is used. Printing out the variable bar thus produces
Obs bar
1970:1
1970:2 7.8705
1970:3 7.5600
1970:4 7.1892
1971:1
1971:2 5.8679
1971:3 6.2442
1971:4 5.9811
1972:1
1972:2 4.6883
1972:3 4.6302
...
Appendix C
Building gretl
Here we give instructions detailed enough to allow a user with only a basic knowledge of a Unix-
type system to build gretl. These steps were tested on a fresh installation of Debian Etch. For other
Linux distributions (especially Debian-based ones, like Ubuntu and its derivatives) little should
change. Other Unix-like operating systems such as Mac OS X and BSD would probably require more
substantial adjustments.
In this guided example, we will build gretl complete with documentation. This introduces a few
more requirements, but gives you the ability to modify the documentation les as well, like the
help les or the manuals.
C.1 Installing the prerequisites
We assume that the basic GNU utilities are already installed on the system, together with these
other programs:
some T
E
X/L
A
T
E
Xsystem (texlive will do beautifully)
Gnuplot
ImageMagick
We also assume that the user has administrative privileges and knows how to install packages. The
examples below are carried out using the apt-get shell command, but they can be performed with
menu-based utilities like aptitude, dselect or the GUI-based program synaptic. Users of Linux
distributions which employ rpm packages (e.g. Red Hat/Fedora, Mandriva, SuSE) may want to refer
to the dependencies page on the gretl website.
The rst step is installing the C compiler and related basic utilities, if these are not already in
place. On a Debian system, these are contained in a bunch of packages that can be installed via the
command
apt-get install gcc autoconf automake1.9 libtool flex bison gcc-doc \
libc6-dev libc-dev gfortran gettext pkgconfig
Then it is necessary to install the development (dev) packages for the libraries that gretl uses:
348
Appendix C. Building gretl 349
Library command
GLIB apt-get install libglib2.0-dev
GTK 2.0 apt-get install libgtk2.0-dev
PNG apt-get install libpng12-dev
XSLT apt-get install libxslt1-dev
LAPACK apt-get install liblapack-dev
FFTW apt-get install libfftw3-dev
READLINE apt-get install libreadline-dev
ZLIB apt-get install zlib1g-dev
XML apt-get install libxml2-dev
GMP apt-get install libgmp3-dev
CURL apt-get install libcurl-dev
MPFR apt-get install libmpfr-dev
MPFR is optional, but recommended. It is possible to substitute GTK 3.0 for GTK 2.0. The dev
packages for these libraries are necessary to compile gretl youll also need the plain, non-dev
library packages to run gretl, but most of these should already be part of a standard installation. In
order to enable other optional features, like audio support, you may need to install more libraries.
The above steps can be much simplied on Linux systems that provide deb-based package managers,
such as Debian and its derivatives (Ubuntu, Knoppix and other distributions). The command
apt-get build-dep gretl
will download and install all the necessary packages for building the version of gretl that is currently present
in your APT sources. Techincally, this does not guarantee that all the software necessary to build the CVS
version is included, because the version of gretl on your repository may be quite old and build requirements
may have changed in the meantime. However, the chances of a mismatch are rather remote for a reasonably
up-to-date system, so in most cases the above command should take care of everything correctly.
C.2 Getting the source: release or CVS
At this point, it is possible to build from the source. You have two options here: obtain the latest
released source package, or retrieve the current CVS version of gretl (CVS = Concurrent Versions
System). The usual caveat applies to the CVS version, namely, that it may not build correctly and
may contain experimental code; on the other hand, CVS often contains bug-xes relative to the
released version. If you want to help with testing and to contribute bug reports, we recommend
using CVS gretl.
To work with the released source:
1. Download the gretl source package from gretl.sourceforge.net.
2. Unzip and untar the package. On a system with the GNU utilities available, the command
would be tar xvfz gretl-N.tar.gz (replace N with the specic version number of the le
you downloaded at step 1).
3. Change directory to the gretl source directory created at step 2 (e.g. gretl-1.9.11).
4. Proceed to the next section, Congure and make.
To work with CVS youll rst need to install the cvs client program if its not already on your system.
Relevant resources you may wish to consult include the CVS website at www.nongnu.org/cvs and
instructions specic to gretl at the SF gretl CVS page.
When grabbing the CVS sources for the rst time, you should rst decide where you want to store
the code. For example, you might create a directory called cvs under your home directory. Open a
terminal window, cd into this directory, and type the following commands:
cvs -d:pserver:[email protected]:/cvsroot/gretl login
cvs -z3 -d:pserver:[email protected]:/cvsroot/gretl co -P gretl
After the rst command you will be prompted for a password: just hit the Enter key. After the
second command, cvs should create a subdirectory named gretl and ll it with the current sources.
When you want to update the source, this is very simple: just move into the gretl directory and
type
cvs update -d -P
Assuming youre now in the CVS gretl directory, you can proceed in the same manner as with the
released source package.
C.3 Congure the source
The next command you need is ./configure; this is a complex script that detects which tools you
have on your system and sets things up. The configure command accepts many options; you may
want to run
./configure --help
rst to see what options are available. One option you way wish to tweak is --prefix. By default
the installation goes under /usr/local but you can change this. For example
./configure --prefix=/usr
will put everything under the /usr tree.
If you have a multi-core machine you may want to activate support for OpenMP, which permits the
parallelization of matrix multiplication and some other tasks. This requires adding the configure
ag
--enable-openmp
By default the gretl GUI is built using version 2.0 of the GTK libraries. However, another configure
option enables use of version 3.0:
--enable-gtk3
In order to have the documentation built, we need to pass the relevant option to configure, as in
--enable-build-doc
But please note that this option will work only if you are using the CVS source.
This, for example, if you want to install under /usr, with OpenMP and GTK 3.0 support, and also
build the documentation, you would do
./configure --prefix=/usr \
--enable-openmp \
--enable-gtk3 \
--enable-build-doc
Example C.1: Sample output from ./configure
Configuration:
Installation path: /usr
Use readline library: yes
Use gnuplot for graphs: yes
Use LaTeX for typesetting output: yes
MPFR support: yes
sse2 support for RNG: yes
openMP support: yes
Build with GTK version: 3.0
Build gretl documentation: yes
Use Lucida fonts: no
Build message catalogs: yes
X-12-ARIMA support: yes
TRAMO/SEATS support: yes
libR support: yes
ODBC support: yes
Experimental audio support: no
Use xdg-utils in installation: if DESTDIR not set
LAPACK libraries:
-llapack -lblas
Now type make to build gretl.
You can also do make pdfdocs to build the PDF documentation.
You will see a number of checks being run, and if everything goes according to plan, you should
see a summary similar to that displayed in Example C.1.
If youre using CVS, its a good idea to re-run the configure script after doing an update. This is not
always necessary, but sometimes it is, and it never does any harm. For this purpose, you may want to write
a little shell script that calls configure with any options you want to use.
C.4 Build and install
We are now ready to undertake the compilation proper: this is done by running the make command,
which takes care of compiling all the necessary source les in the correct order. All you need to do
is type
make
This step will likely take several minutes to complete; a lot of output will be produced on screen.
Once this is done, you can install your freshly baked copy of gretl on your system via
make install
On most systems, the make install command requires you to have administrative privileges.
Hence, either you log in as root before launching make install or you may want to use the sudo
utility, as in:
sudo make install
Now try if everything works: go back to your home directory and run gretl
cd ~
gretl &
If all is well, you ought to see gretl start, at which point just exit the program in the usual way. On
the other hand, there is the possibility that gretl doesnt start and instead you see a message like
/usr/local/bin/gretl_x11: error while loading shared libraries:
libgretl-1.0.so.0: cannot open shared object file: No such file or directory
In this case, just run
sudo ldconfig
The problem should be xed once and for all.
Appendix D
Numerical accuracy
Gretl uses double-precision arithmetic throughoutexcept for the multiple-precision plugin in-
voked by the menu itemModel, Other linear models, High precision OLS which represents oating-
point values using a number of bits given by the environment variable GRETL_MP_BITS (default
value 256).
The normal equations of Least Squares are by default solved via Cholesky decomposition, which
is highly accurate provided the matrix of cross-products of the regressors, X
X, is not very ill

conditioned. If this problem is detected, gretl automatically switches to use QR decomposition.
The program has been tested rather thoroughly on the statistical reference datasets provided by
NIST (the U.S. National Institute of Standards and Technology) and a full account of the results may
be found on the gretl website (follow the link Numerical accuracy).
To date, two published reviews have discussed gretls accuracy: Giovanni Baiocchi and Walter Dis-
taso (2003), and Talha Yalta and Yasemin Yalta (2007). We are grateful to these authors for their
careful examination of the program. Their comments have prompted several modications includ-
ing the use of Stephen Moshiers cephes code for computing p-values and other quantities relating
to probability distributions (see netlib.org), changes to the formatting of regression output to en-
sure that the program displays a consistent number of signicant digits, and attention to compiler
issues in producing the MS Windows version of gretl (which at one time was slighly less accurate
than the Linux version).
Gretl now includes a plugin that runs the NIST linear regression test suite. You can nd this
under the Tools menu in the main window. When you run this test, the introductory text explains
the expected result. If you run this test and see anything other than the expected result, please
send a bug report to [email protected].
All regression statistics are printed to 6 signicant gures in the current version of gretl (except
when the multiple-precision plugin is used, in which case results are given to 12 gures). If you
want to examine a particular value more closely, rst save it (for example, using the genr command)
then print it using printf, to as many digits as you like (see the Gretl Command Reference).
353
Appendix E
Related free software
Gretls capabilities are substantial, and are expanding. Nonetheless you may nd there are some
things you cant do in gretl, or you may wish to compare results with other programs. If you are
looking for complementary functionality in the realm of free, open-source software we recommend
the following programs. The self-description of each program is taken from its website.
GNU R r-project.org: R is a system for statistical computation and graphics. It consists of
a language plus a run-time environment with graphics, a debugger, access to certain system
functions, and the ability to run programs stored in script les. . . It compiles and runs on a
wide variety of UNIX platforms, Windows and MacOS. Comment: There are numerous add-on
packages for R covering most areas of statistical work.
GNU Octave www.octave.org: GNU Octave is a high-level language, primarily intended for
numerical computations. It provides a convenient command line interface for solving linear
and nonlinear problems numerically, and for performing other numerical experiments using
a language that is mostly compatible with Matlab. It may also be used as a batch-oriented
language.
JMulTi www.jmulti.de: JMulTi was originally designed as a tool for certain econometric pro-
cedures in time series analysis that are especially dicult to use and that are not available
in other packages, like Impulse Response Analysis with bootstrapped condence intervals for
VAR/VEC modelling. Now many other features have been integrated as well to make it possi-
ble to convey a comprehensive analysis. Comment: JMulTi is a java GUI program: you need
a java run-time environment to make use of it.
As mentioned above, gretl oers the facility of exporting data in the formats of both Octave and
R. In the case of Octave, the gretl data set is saved as a single matrix, X. You can pull the X matrix
apart if you wish, once the data are loaded in Octave; see the Octave manual for details. As for R,
the exported data le preserves any time series structure that is apparent to gretl. The series are
saved as individual structures. The data should be brought into R using the source() command.
In addition, gretl has a convenience function for moving data quickly into R. Under gretls Tools
menu, you will nd the entry Start GNU R. This writes out an R version of the current gretl
data set (in the users gretl directory), and sources it into a new R session. The particular way
R is invoked depends on the internal gretl variable Rcommand, whose value may be set under the
Tools, Preferences menu. The default command is RGui.exe under MS Windows. Under X it is
xterm -e R. Please note that at most three space-separated elements in this command string will
be processed; any extra elements are ignored.
354
Appendix F
Listing of URLs
Below is a listing of the full URLs of websites mentioned in the text.
Estima (RATS) http://www.estima.com/
FFTW3 http://www.fftw.org/
Gnome desktop homepage http://www.gnome.org/
GNU Multiple Precision (GMP) library http://gmplib.org/
CURL library http://curl.haxx.se/libcurl/
GNU Octave homepage http://www.octave.org/
GNU R homepage http://www.r-project.org/
GNU R manual http://cran.r-project.org/doc/manuals/R-intro.pdf
Gnuplot homepage http://www.gnuplot.info/
Gnuplot manual http://ricardo.ecn.wfu.edu/gnuplot.html
Gretl data page http://gretl.sourceforge.net/gretl_data.html
Gretl homepage http://gretl.sourceforge.net/
GTK+ homepage http://www.gtk.org/
GTK+ port for win32 http://www.gimp.org/~tml/gimp/win32/
InfoZip homepage http://www.info-zip.org/pub/infozip/zlib/
JMulTi homepage http://www.jmulti.de/
JRSoftware http://www.jrsoftware.org/
Mingw (gcc for win32) homepage http://www.mingw.org/
Minpack http://www.netlib.org/minpack/
Penn World Table http://pwt.econ.upenn.edu/
Readline homepage http://cnswww.cns.cwru.edu/~chet/readline/rltop.html
Readline manual http://cnswww.cns.cwru.edu/~chet/readline/readline.html
Xmlsoft homepage http://xmlsoft.org/
355
Bibliography
Akaike, H. (1974) A new look at the statistical model identication, IEEE Transactions on Auto-
matic Control AC-19: 716723.
Anderson, B. and J. Moore (1979) Optimal Filtering, Upper Saddle River, NJ: Prentice-Hall.
Anderson, T. W. and C. Hsiao (1981) Estimation of dynamic models with error components, Jour-
nal of the American Statistical Association 76: 598606.
Andrews, D. W. K. and J. C. Monahan (1992) An improved heteroskedasticity and autocorrelation
consistent covariance matrix estimator, Econometrica 60: 953966.
Arellano, M. (2003) Panel Data Econometrics, Oxford: Oxford University Press.
Arellano, M. and S. Bond (1991) Some tests of specication for panel data: Monte carlo evidence
and an application to employment equations, The Review of Economic Studies 58: 277297.
Baiocchi, G. and W. Distaso (2003) GRETL: Econometric software for the GNU generation, Journal
of Applied Econometrics 18: 105110.
Baltagi, B. H. (1995) Econometric Analysis of Panel Data, New York: Wiley.
Barrodale, I. and F. D. K. Roberts (1974) Solution of an overdetermined system of equations in the
l
norm, Communications of the ACM 17: 319320.
Baxter, M. and R. G. King (1999) Measuring business cycles: Approximate band-pass lters for
economic time series, The Review of Economics and Statistics 81(4): 575593.
Beck, N. and J. N. Katz (1995) What to do (and not to do) with time-series cross-section data, The
American Political Science Review 89: 634647.
Berndt, E., B. Hall, R. Hall and J. Hausman (1974) Estimation and inference in nonlinear structural
models, Annals of Economic and Social Measurement 3(4): 653665.
Blundell, R. and S. Bond (1998) Initial conditions and moment restrictions in dynamic panel data
models, Journal of Econometrics 87: 115143.
Bond, S., A. Hoeer and J. Temple (2001) GMMestimation of empirical growth models. Economics
Papers from Economics Group, Nueld College, University of Oxford, No 2001-W21.
Boswijk, H. P. (1995) Identiability of cointegrated systems. Tinbergen Institute Discussion Paper
95-78. http://www.ase.uva.nl/pp/bin/258fulltext.pdf.
Boswijk, H. P. and J. A. Doornik (2004) Identifying, estimating and testing restricted cointegrated
systems: An overview, Statistica Neerlandica 58(4): 440465.
Box, G. E. P. and G. Jenkins (1976) Time Series Analysis: Forecasting and Control, San Franciso:
Holden-Day.
Brand, C. and N. Cassola (2004) A money demand system for euro area M3, Applied Economics
36(8): 817838.
Butterworth, S. (1930) On the theory of lter ampliers, Experimental Wireless & The Wireless
Engineer 7: 536541.
356
Bibliography 357
Byrd, R. H., P. Lu, J. Nocedal and C. Zhu (1995) A limited memory algorithm for bound constrained
optimization, SIAM Journal on Scientic Computing 16: 11901208.
Cameron, A. C. and P. K. Trivedi (2005) Microeconometrics, Methods and Applications, Cambridge:
Cambridge University Press.
Caselli, F., G. Esquivel and F. Lefort (1996) Reopening the convergence debate: A new look at
cross-country growth empirics, Journal of Economic Growth 1(3): 363389.
Chesher, A. and M. Irish (1987) Residual analysis in the grouped and censored normal linear
model, Journal of Econometrics 34: 3361.
Choi, I. (2001) Unit root tests for panel data, Journal of International Money and Finance 20(2):
249272.
Cleveland, W. S. (1979) Robust locally weighted regression and smoothing scatterplots, Journal
of the American Statistical Association 74(368): 829836.
Cribari-Neto, F. and S. G. Zarkos (2003) Econometric and statistical computing using Ox, Compu-
tational Economics 21: 277295.
Davidson, R. and J. G. MacKinnon (1993) Estimation and Inference in Econometrics, New York:
Oxford University Press.
(2004) Econometric Theory and Methods, New York: Oxford University Press.
Doornik, J. A. (1995) Testing general restrictions on the cointegrating space. Discussion Paper,
Nueld College. http://www.doornik.com/research/coigen.pdf.
(1998) Approximations to the asymptotic distribution of cointegration tests, Journal of
Economic Surveys 12: 573593. Reprinted with corrections in McAleer and Oxley (1999).
(2007) Object-Oriented Matrix Programming Using Ox, London: Timberlake Consultants
Press, third edn. www.doornik.com.
Doornik, J. A., M. Arellano and S. Bond (2006) Panel Data estimation using DPD for Ox.
Elliott, G., T. J. Rothenberg and J. H. Stock (1996) Ecient tests for an autoregressive unit root,
Econometrica 64: 813836.
Engle, R. F. and C. W. J. Granger (1987) Co-integration and error correction: Representation, esti-
mation, and testing, Econometrica 55: 251276.
Fiorentini, G., G. Calzolari and L. Panattoni (1996) Analytic derivatives and the computation of
GARCH estimates, Journal of Applied Econometrics 11: 399417.
Frigo, M. and S. G. Johnson (2005) The design and implementation of FFTW3, Proceedings of the
IEEE 93 2: 216231.
Goossens, M., F. Mittelbach and A. Samarin (2004) The L
A
T
E
X Companion, Boston: Addison-Wesley,
second edn.
Gourieroux, C. and A. Monfort (1996) Simulation-Based Econometric Methods, Oxford: Oxford Uni-
versity Press.
Gourieroux, C., A. Monfort, E. Renault and A. Trognon (1987) Generalized residuals, Journal of
Econometrics 34: 532.
Greene, W. H. (2000) Econometric Analysis, Upper Saddle River, NJ: Prentice-Hall, fourth edn.
(2003) Econometric Analysis, Upper Saddle River, NJ: Prentice-Hall, fth edn.
Hall, A. D. (2005) Generalized Method of Moments, Oxford: Oxford University Press.
Bibliography 358
Hamilton, J. D. (1994) Time Series Analysis, Princeton, NJ: Princeton University Press.
Hannan, E. J. and B. G. Quinn (1979) The determination of the order of an autoregression, Journal
of the Royal Statistical Society, B 41: 190195.
Hansen, L. P. (1982) Large sample properties of generalized method of moments estimation,
Econometrica 50: 10291054.
Hansen, L. P. and K. J. Singleton (1982) Generalized instrumental variables estimation of nonlinear
rational expectations models, Econometrica 50: 12691286.
Harvey, A. C. (1989) Forecasting, structural time series models and the Kalman lter, Cambridge:
Cambridge University Press.
Harvey, A. C. and T. Proietti (2005) Readings in Unobserved Component Models, Oxford: Oxford
University Press.
Hausman, J. A. (1978) Specication tests in econometrics, Econometrica 46: 12511271.
Heckman, J. (1979) Sample selection bias as a specication error, Econometrica 47: 153161.
Hodrick, R. and E. C. Prescott (1997) Postwar U.S. business cycles: An empirical investigation,
Journal of Money, Credit and Banking 29: 116.
Im, K. S., M. H. Pesaran and Y. Shin (2003) Testing for unit roots in heterogeneous panels, Journal
of Econometrics 115: 5374.
Johansen, S. (1995) Likelihood-Based Inference in Cointegrated Vector Autoregressive Models, Ox-
ford: Oxford University Press.
de Jong, P. (1991) The diuse Kalman lter, The Annals of Statistics 19: 10731083.
Kalbeisch, J. D. and R. L. Prentice (2002) The Statistical Analysis of Failure Time Data, New York:
Wiley, second edn.
Kalman, R. E. (1960) A new approach to linear ltering and prediction problems, Transactions of
the ASMEJournal of Basic Engineering 82(Series D): 3545.
Keane, M. P. and K. I. Wolpin (1997) The career decisions of young men, Journal of Political
Economy 105: 473522.
Koenker, R. (1994) Condence intervals for regression quantiles. In P. Mandl and M. Huskova
(eds.), Asymptotic Statistics, pp. 349359. New York: Springer-Verlag.
Koenker, R. and G. Bassett (1978) Regression quantiles, Econometrica 46: 3350.
Koenker, R. and K. Hallock (2001) Quantile regression, Journal of Economic Perspectives 15(4):
143156.
Koenker, R. and J. Machado (1999) Goodness of t and related inference processes for quantile
regression, Journal of the American Statistical Association 94: 12961310.
Koenker, R. and Q. Zhao (1994) L-estimation for linear heteroscedastic models, Journal of Non-
parametric Statistics 3: 223235.
Koopman, S. J. (1997) Exact initial Kalman ltering and smoothing for nonstationary time series
models, Journal of the American Statistical Association 92: 16301638.
Koopman, S. J., N. Shephard and J. A. Doornik (1999) Statistical algorithms for models in state
space using SsfPack 2.2, Econometrics Journal 2: 113166.
Bibliography 359
Kwiatkowski, D., P. C. B. Phillips, P. Schmidt and Y. Shin (1992) Testing the null of stationarity
against the alternative of a unit root: How sure are we that economic time series have a unit
root?, Journal of Econometrics 54: 159178.
Levin, A., C.-F. Lin and J. Chu (2002) Unit root tests in panel data: asymptotic and nite-sample
properties, Journal of Econometrics 108: 124.
Lucchetti, R., L. Papi and A. Zazzaro (2001) Banks ineciency and economic growth: A micro
macro approach, Scottish Journal of Political Economy 48: 400424.
Ltkepohl, H. (2005) Applied Time Series Econometrics, Springer.
MacKinnon, J. G. (1996) Numerical distribution functions for unit root and cointegration tests,
Journal of Applied Econometrics 11: 601618.
Magnus, J. R. and H. Neudecker (1988) Matrix Dierential Calculus with Applications in Statistics
and Econometrics, John Wiley & Sons.
McAleer, M. and L. Oxley (1999) Practical Issues in Cointegration Analysis, Oxford: Blackwell.
McCullough, B. D. and C. G. Renfro (1998) Benchmarks and software standards: A case study of
GARCH procedures, Journal of Economic and Social Measurement 25: 5971.
Mroz, T. (1987) The sensitivity of an empirical model of married womens hours of work to eco-
nomic and statistical assumptions, Econometrica 5: 765799.
Nadaraya, E. A. (1964) On estimating regression, Theory of Probability and its Applications 9:
141142.
Nash, J. C. (1990) Compact Numerical Methods for Computers: Linear Algebra and Function Min-
imisation, Bristol: Adam Hilger, second edn.
Nerlove, M. (1971) Further evidence on the estimation of dynamic economic relations from a time
series of cross sections, Econometrica 39: 359382.
(1999) Properties of alternative estimators of dynamic panel models: An empirical analy-
sis of cross-country data for the study of economic growth. In C. Hsiao, K. Lahiri, L.-F. Lee and
M. H. Pesaran (eds.), Analysis of Panels and Limited Dependent Variable Models. Cambridge: Cam-
bridge University Press.
Newey, W. K. and K. D. West (1987) A simple, positive semi-denite, heteroskedasticity and auto-
correlation consistent covariance matrix, Econometrica 55: 703708.
(1994) Automatic lag selection in covariance matrix estimation, Review of Economic Stud-
ies 61: 631653.
Okui, R. (2009) The optimal choice of moments in dynamic panel data models, Journal of Econo-
metrics 151(1): 116.
Pollock, D. S. G. (1999) A Handbook of Time-Series Analysis, Signal Processing and Dynamics, New
York: Academic Press.
(2000) Trend estimation and de-trending via rational square-wave lters, Journal of
Econometrics 99(2): 317334.
Portnoy, S. and R. Koenker (1997) The Gaussian hare and the Laplacian tortoise: computability of
squared-error versus absolute-error estimators, Statistical Science 12(4): 279300.
Ramanathan, R. (2002) Introductory Econometrics with Applications, Fort Worth: Harcourt, fth
edn.
Bibliography 360
Roodman, D. (2006) How to do xtabond2: An introduction to dierence and system GMM in
Stata. Center for Global Development, Working Paper Number 103.
Schwarz, G. (1978) Estimating the dimension of a model, Annals of Statistics 6: 461464.
Sephton, P. S. (1995) Response surface estimates of the KPSS stationarity test, Economics Letters
47: 255261.
Sims, C. A. (1980) Macroeconomics and reality, Econometrica 48: 148.
Steinhaus, S. (1999) Comparison of mathematical programs for data analysis (edition 3). Univer-
sity of Frankfurt. http://www.informatik.uni-frankfurt.de/~stst/ncrunch/.
Stock, J. H. and M. W. Watson (2003) Introduction to Econometrics, Boston: Addison-Wesley.
(2008) Heteroskedasticity-robust standard errors for xed eects panel data regression,
Econometrica 76(1): 155174.
Stokes, H. H. (2004) On the advantage of using two or more econometric software systems to
solve the same problem, Journal of Economic and Social Measurement 29: 307320.
Swamy, P. A. V. B. and S. S. Arora (1972) The exact nite sample properties of the estimators of
coecients in the error components regression models, Econometrica 40: 261275.
Theil, H. (1961) Economic Forecasting and Policy, Amsterdam: North-Holland.
(1966) Applied Economic Forecasting, Amsterdam: North-Holland.
Verbeek, M. (2004) A Guide to Modern Econometrics, New York: Wiley, second edn.
Watson, G. S. (1964) Smooth regression analysis, Shankya Series A 26: 359372.
White, H. (1980) A heteroskedasticity-consistent covariance matrix astimator and a direct test for
heteroskedasticity, Econometrica 48: 817838.
Windmeijer, F. (2005) A nite sample correction for the variance of linear ecient two-step GMM
estimators, Journal of Econometrics 126: 2551.
Wooldridge, J. M. (2002a) Econometric Analysis of Cross Section and Panel Data, Cambridge, MA:
MIT Press.
(2002b) Introductory Econometrics, A Modern Approach, Mason, OH: South-Western, sec-
ond edn.
Yalta, A. T. and A. Y. Yalta (2007) GRETL 1.6.0 and its numerical accuracy, Journal of Applied
Econometrics 22: 849854.

Gretl Guide

Uploaded by

Copyright:

Available Formats

Gretl Guide

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Gretl Guide

Uploaded by

Copyright:

Available Formats

Gretl Users Guide

Gnu Regression, Econometrics and Time-series Library

A is singular; the numerical method employed here is the QR

where U is an mk orthogonal matrix, is an kk diagonal matrix, and V is an kn orthogonal

that is provided. To reconstitute

Y. The Cholesky decomposition is used. The

< n then only n

values starting at element t

) = 0), its variance is

) is the covariance matrix of the error terms.

are estimated directly by the squared OLS residuals, u

), the matrix being estimated here can also be written as

X can be recovered via

is any HAC estimator, applied to the VAR residuals.

, while the covariance between u

) are estimated by the sample variance matrices of the xed- and

for the rst step is no trivial matter. Its

X), since the expression in square brackets equals 0.

f is a (1 m p) vector holding the average of the orthogonality conditions and W is some

is a faithful representation of the individuals preferences,

) represents the maximum loglikelihood as a function of the vector of parameter esti-

for a low-pass lter, or frequencies greater than

for high-pass) and a stop-band; the

, expressed in degrees. The

is placed in such a dead spot then the job of separation

The gretl syntax for estimating (25.4) is simply

is bound by a cointegrating vector

x is stationary (with a mean of

is the left null space of the matrix .

, where is the matrix containing the cointegration

. This point may or may not have a greater

respectively. The zero

may be non-zero. This means that

, r(r +1)/2 columns.

denote the best point among

. Script 32.5 shows an example. As a further similarity

; for the log-logistic

, represent the conditional

(0.04584) (0.04597) (0.05160)

= 0, the estimate for

X, is not very ill

You might also like