Cobbledick John 1995

AN APPLIED TRANCENDENTAL LOGARITHMIC
COST FUNCTION: ECONOMIES OF SCALE

AND ELASTICITIES OF SUBSTITUTION IN
SELECTED SOUTH AFRICAN
MANUFACTURING SECTORS
(1972-1990)
JOHN COBBLEDICK
A dissertation submitted to the Department ofEconomics,

University ofNatal, Durban in partial fulfilment of the requirements
for the degree of Master of Commerce in Economics
University of Natal, Durban

1995
ACKNOWLEDGMENTS
I am deeply indebted to a number of people who supported me - both emotionally and

intellectually - during the preparation of this dissertation. I am grateful to both my
family, Patricia, Michael, and Daisy and Gail Bowyer who encouraged and understood
me and displayed remarkable patience throughout.
I will remain eternally indebted to my close friends and colleagues, Jason Probert and
Matthew Sharratt, without whose cajoling and encouragement this dissertation may
never have begun, let alone be complete.
Julian Hofineyr provided me with outstanding supervision. He was always willing to

consult obscure literature, read and edit unwieldy drafts, identify inconsistencies and
suggest alternative routes around apparent obstacles. His support went beyond mere
supervision, however, as he constantly encouraged me and focused my grandiose
tendencies.
ABSTRACT
Moll (1991) has criticised the proposal that demand restructuring should act as the
impetus for economic growth in a post-apartheid South Africa on the grounds of, a
lack of empirical support. The demand restructuring thesis is premised on two
empirically testable assertions: firstly that realisable economies of scale are greater in
labour-intensive wage goods sectors than in luxury goods and secondly that in
manufacturing as a whole labour can easily substitute for capital. While a number of
studies employing either the Cobb-Douglas (Cobb & Douglas, 1948) or Constant
Elasticity of Substitution (CES) ( Arrow, Chenery, Minhas & Solow, 1961) functions
have attempted to quantify these features of technology, their conclusions are
potentially invalid.
Both functions impose the maintained hypotheses ofhomotheticity, homogeneity and
seperability a priori. As primary hypothesis tests regarding the magnitude of
parameters depend on the validity of both the hypothesis being tested and the
underlying maintained hypotheses, the plausibility of maintained hypotheses is an
important consideration when choosing a functional form for econometric analysis.
Homotheticity and homogeneity constrain the theoretical determinants of economies
of scale and seperability. The theoretical determinants of substitution thus limit the
contexts in which functions which embody these hypotheses are likely to be
appropriate.
The mathematical concept of duality has permitted the development of flexible, general
functions, such as the Transcendental Logarithmic Cost Function (Christensen,
Jorgensen and Lau, 1971, 1973), which rather than imposing, permits the testing of the
most commonly imposed maintained hypotheses. By applying this function to three
sub-sectors of South Mrican manufacturing both the validity of the commonly imposed
maintained hypotheses and the empirical premises of the demand restructuring position
are assessed in this dissertation. This application indicates that not only are the
hypotheses ofhomotheticity, homogeneity and seperability invalid but that the
inappropriate imposition of homotheticity, homogeneity and seperability invalid but
that the inappropriate imposition ofhomotheticity biases estimates of scale
downwards. Evidence also emerges to challenge Moll's (1991) assertions regarding
the empirical validity of demand restructuring.
iv
TABLE OF CONTENTS
LIST OF APPENDICES VI
TABLES AND FIGURES Vlll
CHAPTER 1: INTRODUCTION 1
CHAPTER 1: CHOICE OF AN APPROPRIATE FUNCTIONAL FORM 5

2.1. Introduction 5
2.2. Economies of scale and homothetic functions 8
2.2.1 Economies of Scale: A graphical exposition 9
2.2.2 Economies of Scale: A mathematical exposition 13
2.2.3 Homotheticity 19 '
2.2.4 Homotheticity and Cobb-Douglas 27
and CES functions
2.3 Elasticities of substitution and functional separability 28 .•
2.3.1. Elasticities of substitution 29'
2.3.2 Seperability 34
2.4 Conclusion 36
CHAPTER 2: THE TRANSCENDENTAL LOGARITHMIC COST FUNCTION 38

3. 1 Introduction 38
3.1.1 Behavioural assumptions of duality 40
and the assumed nature of regressors
3.1.2 Choosing between different flexible forms 42
3.2 The Transcendental Logarithmic Cost Function 43 "
3.2.1 Components of the system 44
3.2.2 Constraints on the system 48 .
3.3 The Translog and Scale, Substitution 52
and Technical Change
3.3.1 Economies of Scale 53
3.3.2 Elasticities of Substitution and Concavity Testing 54 ..
3.3.3 Modelling technical change 57
v
CHAPTER 4: DATA, ESTIMATION AND TESTING PROCEDURES 64

4. 1 Introduction 64
4.2 Method of Estimation 65
4.2.1 The Zeller Efficient Estimator (ZEF) 66 ..
4.2.2 Three Stages Least Squares (3SLS) 68
4.2.3 Invariance of Estimators: Cluver (1981) Revisited 70
4.2.4 Testing endogeneity: 72
The Spencer and Berk (1981) Test
4.2.5 Autocorrelation 74
4.3 Methods of Statistical Inference 75
4.3.1 Nested Hypothesis tests 76
4.3.2 Goodness offit 79
4.4. Data Considerations 80
4.4.1 Capital 0 82
4.4.2 Materials 85
4.4.3 Labour , 88
4.4.4 Instrumental Variables 89
CHAPTER 5: ECONOMETRIC RESULTS 93

5.1 Introduction 93
5.2 Electrical Appliances and Household Goods (ISIC 3833) 96
5.2.1 Data and a priori analysis 96
5.2.2 Regression Analysis 99
5.2.3 Conclusions 107
5.3 Furniture (ISIC 3320) 107
5.3.1 Data and a priori analysis 107
5.4 Motor Vehicles, Parts and Accessories (ISIC 3840) ; 113
5.4.1 Data and a priori analysis , 113
5.5 Conclusion 120
CHAPTER 6: CONCLUSION 123
BIBLIOGRAPHY 127
APPENDICES 135
vi
APPENDICES
Appendix Page
3 CONCAVITY TESTING A3-1
4 DATA SERIES AND INSTRUMENTAL VARIABLES A4-1
4.1 Data series A4-1

4.1.1 Electrical Appliances and Household Goods (ISIC3 833) A4.1-1
4.1.2 Motor Vehicles, Part and Accessories (ISIC3840) A4.1-5
4.1.3 Furniture (ISIC3320) A4.1-7
4.2 Instrumental variables A4-2-1
5 ECONOMETRIC RESULTS A5-1
5.1 Forms of alternative specifications of the A5-1

Translog system
5.1.1. Model 1. Technical change augmented A5-1
non-homothetic form
5.1.2 Model 2. Non-homothetic form A5.1-2
(Constant technology)
5.1.3 Model 3. Homothetic form (Constant technology) A5.1-2
5.1.4 Model 4. Homogenous form (Constant technology) A5.1-3
5.1.5 Model 5. Variable returns to scale A5.1-3
Cobb-Douglas form
5.2. A priori data analysis A5.2-1

5.2.1 Electric appliances and household goods A5.2-1
(ISIC 3883)
5.2.2 Furniture (ISIC 3220) A5.2-4
5.2.3 Motor vehicle parts and accessories (ISIC 3840) A5.2-7
5.3 Motor vehicles, parts and accessories (ISIC 3840) A5.3-1

5.3.1 Hypothesis Testing. A5.3-1
5.3.2 Durbin-Watson test results c, A5.3-3
5.3.3 Parameter estimates. A5.3-4
5.3.4 Monotonicity and concavity test results· A5.3-8
5.3.5 Economies of scale A5.3-23
5.3.6 Graphs of economies of scale A5.3-25
5.4 Furniture (ISIC 3320) A5.4-1

5.4.1 Hypothesis test results A5.4-1
5.4.2 Durbin-Watson test results A5.4-3
5.4.3 Parameter estimates A5.4-4
5.4.5 Economies ofscale A5.4-22
Appendices are numbered according to the chapters to which they relate. There are thus no appendices I or 2
vii
5.5 Electrical appliances and household goods A5.5-1

(ISIC 3833)
5.5.2 Durbin-Watson test results A5.5-1
5.5.3 Parameter estimates A 5.5-3
5.5.4 Monotonicity and concavity test results A5.5-7
5.5.4 Economies of scale A5.5-22'
viii
TABLES AND FIGURES
Table Page
4.1.1.1 Cost of capital: electrical appliances and A4.1-1

household goods (ISIC 3833)
4.1.1.2 Cost of labour and price of labour: electrical A4.1-2
appliances and household goods (ISIC 3833)
4.1.1.3 Output, cost of materials and price of materials: A4.1-3
electrical appliances and household goods
(ISIC 3833)
4.1.1.4 Cost and cost shares of capital, labour and A4.1-4
materials:electrical appliances and household
goods (ISIC 3833)
4.1.2.1 Cost of capital: motor vehicles, parts and A4.1-5
accessories (ISIC 3840)
4.1.2.2 Cost of labour and price of labour: motor A4.1-6
vehicles, parts and accessories (ISIC 3840)
motor vehicles, parts and accessories (ISIC 3840)
4.1.2.4 Costs and cost shares of capital, labour & materials: A4.1-8
Motor vehicles, parts and accessories (ISIC 3840)
4.1.3.1 Cost of capital: furniture (ISIC 3320) A4.1-9
4.1.3.2 Cost oflabour and price oflabour: A4.1-1O
furniture (ISIC 3320)
furniture (ISIC 3320)
4.1.3.4 Cost and cost shares of capital, labour and A4.1-12
materials: furniture (ISIC 3320)
5.2.1.1 Input-output coefficients, relative input prices A5.2-1
and real average costs (ISIC 3833)
5.2.1.2 Descriptive statistics of relative factor prices A5.2-1
and input-output coefficients (ISIC 3833)
5.2.1.3 Input-output coefficients of materials (ISIC3833) A5.2-3
5.2.1.4 Relative prices of Capital, Labour and Materials A5.2-3
(ISIC 3833)
and input-output coefficients (ISIC 3840) .
5.2.2.3 Input-output coefficients of Materials (ISIC 3320) A5.2-6
5.2.2.4 Relative prices of Capital, Labour and Materials A5.2-6
(ISIC 3320)
and input-output coefficients (ISIC 3840)
ix
5.2.3.3 Input-output coefficients of materials (ISIC 3840) A5.2-9

5.2.3.4 Relative prices of capital, labour and materials A5.2-9
(ISIC3840)
5.3.1.1 I.R test statistics and appropriate chi-square A5.3-1
statistics for IZEF and 3SLS estimates of
ISIC (3840)
5.3.1.2 I.R test statistics for comparison ofIZEF. A5.3-1
and IZEF(AR) estimators ISIC 3840
5.3.1.3 I.R test statistics for comparison of 13 SLS A.5.3-1
and 13SLS(AR) estimators ISIC 3840
5.3.1.4 I.R. test of economies of scale - A5.3-2
IZEF estimates (ISIC 3840)
5.3.1.5 1.R. test of economies of scale - A5.3-2
13SLS estimates (ISIC 3840)
5.3.2.1 Computed and critical Durbin-Watson statistics A5.3-3
for IZEF estimatesoflSIC 3840
5.3.2.2 Computed and critical Durbin-Watson statistics A5.3-3
for 13SLS estimates ofISIC 3840
5.3.3.1 IZEF parameter estimates for motor vehicles A5.3-4
(ISIC 3840)
5.3.3.2 IZEF(AR) parameter estimates for motor vehicles A5.3-5
(ISIC 3840)
5.3.3.3 13 SLS parameter estimates for motor vehicles A5.3-6
(ISIC 3840)
5.3.3.4 13 SLS (AR) parameter estimates for motor A5.3-7
vehicles (ISIC 3840)
5.3.4.1 Fitted cost shares and own AESs for IZEF A5.3-8
estimates Model 1 (ISIC 3840)
5.3.4.2 Proper AESs and determinants of matrices A5.3-8
of AESs for IZEF estimates, Model 1 (ISIC 3840)
5.3.4.3 Fitted cost shares and own AESs for A5.3-9
IZEF estimates, Model 2 (ISIC 3840)
of AESs for IZEF estimates, Model 2
5.3.4.5 Fitted cost shares and Own AESs for IZEF A5.3-10
estimates, Model 3 (ISIC 3840)
5.3.4.8 Proper AESs and determinants of AESs A5.3-11
for IZEF estimates, Model 4 (ISIC 3840)
(AR) estimates, Model 1 (ISIC 3840)
of AESs for IZEF (AR) estimates, Model 1 (ISIC 3840)
5.3.4.11 Fitted cost shares and Own AESs for A5.3-13
IZEF (AR) estimates, Model 2 (ISIC 3840)
x

of AESs for IZEF (AR) estimates, Model 2
(ISIC 3840)
5.3.4.16 Proper AESs and determinants of matrices of A5.3-15
AESs for IZEF (AR) estimates, Model 4
(ISIC 3840)
13 SLS estimates, Model 1 (ISIC 3840)
AESs for 13 SLS estimates, Model 1 (ISIC 3840)
5.3.4.19 Fitted cost shares and Own AESs for 13 SLS A5.3-17
of AESs for 13 SLS estimates, Model 2
(ISIC 3840)
13SLS estimates, Model 3 (ISIC 3840)
5.3.4.22 Proper AESs and determinants of matrices A5.3.,18
of AESs for 13SLS estimates, Model 3
(ISIC 3840)
5.3.4.23 Fitted cost shares and Own AESs for estimates, A5.3-19
Model 4 (ISIC 3840)
(ISIC 3840)
AES for 13 SLS (AR) estimates, Model 2
(ISIC 3840)
of AESs for 13 SLS (AR) estimates,
Model 3 (ISIC 3840)
(AR) estimates Model 4 (ISIC 3840)
of AESs for 13 SLS (AR) estimates,
Model 4 (ISIC 3840)
5.3.5.1 IZEF estimates of Economies of Scale - A5.3-23
ISIC 3840
xi
5.3.5.2 13 SLS estimates of Economies of Scale- A5.3-24

ISIC 3840
5.3.6.1 Economies of Scale based on IZEF A5.3-25
(AR) estimates - ISIC 3840
5.3.6.2 Economies of Scale based on A5.3-25
IZEF (AR) estimates - ISIC 3840
13 SLS estimates - ISIC 3840
13 SLS (AR) estimates - ISIC 3840
5.3.6.5 Economies of Scale - Model 1 A5.3-27
IZEF, IZEF (AR) and 13 SLS
5.3.6.6 Economies of Scale - Model 2 IZEF, IZEF A5.3-27
(AR), 13 SLS and 13 SLS (AR)
5.3.6.7 Economies of Scale - Model 3 IZEF, IZEF A5.3-28
(AR), 13 SLS and 13 SLS (AR)
5.4.1.1 LR test statistics and appropriate chi-square A5.4-1
statistics for IZEF and 3SLS
5.4.1.2 LR test statistics for comparison of IZEF A5.4-1
(AR) estimates ISIC 3320
5.4.1.3 LR test statistics for comparison of 13 SLS A5.4-1
and 13 SLS (AR) estimators ISIC 3320
5.4.1.4 LR test of Economies of Scale - A5.4-2
IZEF estimates (ISIC 3320)
5.4.1.5 LR test of Economies of Scale- A5.4-2
13SLS estimates (ISIC 3320)
5.4.2.1 Computed and critical Durbin-Watson A5.4-3
statistics for IZEF estimates ofISIC 3320
statistics for 13SLS estimates ofISIC 3320
5.4.3.1 IZEF parameter estimates for Furniture A5.4-4
(ISIC 3320)
5.4.3.2 IZEF AR(l) parameter estimates for A5.4-5
Furniture (ISIC 3320)
5.4.3.3 13 SLS parameter estimates for Furniture A5.4-6
(ISIC 3320)
5.4.3.4 13 SLS AR( 1) parameter estimates for A5.4-7
Furniture (ISIC 3320)
of AESs for IZEF estimates, Model I
(ISIC 3320)
of AESs for IZEF estimates, Model 2 (ISIC 332
5.4.4.5 Fitted cosst shares and Own AESs for A5.4-10
xii

(ISIC 3320)
5.4.4.8 Proper AESs and determinants of matrices A. 5.4-11
of AESs for IZEF estimates, Model 4 (I~IC 3320)
5.4.4.9 Fitted cost shares and Own AESs for A.5.4-12
IZEF (AR) estimates, Model 1, (ISIC 3320)
5.4.4.10 Proper AESs and determinats of matrices A5.4-12
of AESs for IZEF (AR) estimates,
Model 1 (ISIC 3320)
of AESs for IZEF (AR) estimates,
Model 3 (ISIC 3320)
(ISIC 3320)
5.4.4:15 Fitted cost shares and Own AESs A5.4-15
for 13SLS estimates, Model 1 (ISIC 3320)
(ISIC 3320)
5.4.4.17 Fitted cost shares and Own AESs for A.5.4-16
AESs for 13 SLS estimates, Model 2
(ISIC 3320)
(ISIC 3320)
(ISIC 3320)
of AESs for 13SLS (AR) estimates, Model 2
(ISIC 3320)
xiii

AESs for 13SLS (AR) estimates,
Model 3 (ISIC 3320)
5.4.4.27 Fitted cost shares and Own AESs for 13SLS (AR) A5.4-21
of AESs for 13SLS (AR) estimates,
Model 4 (ISIC 3320)
5.4.5.1 IZEF estimates of Economies of Scale A5.4-22
(ISIC 3320)
5.4.5.2 13SLS estimates of Economies of Scale A5.4-23
(ISIC3320)
Fig5.4.6.1 Economies of Scale - IZEF estimates - A5.4-24

ISIC 3320
Fig5.4.6.2 Economies of Scale - IZEF (AR) A5.4-24
estimates - ISIC 3320
Fig5.4.6.3 Economies of Scale - 13 SLS estimates - A5.4-25
ISIC 3320
Fig5.4.6.4 Economies of Scale - 13SLS (AR) estimates - A5.4-25
ISIC 3320
Fig5.4.6.5 Economies of Scale - Model 1 - ISIC 3320 A5.4-26
Fig5.4.6.6 Economies of Scale - Model 2 - ISIC 3320 A5.4-26
Fig5.4.6.7 Economies of Scale - Model 3 -ISIC 3320 A5.4-27

statistics for IZEF estimates oflSIC 3833
statistics for 13SLS estimates ofISIC 3833
5.5.3.1 IZEF parameter estimates for Electrical A5.5-3
Appliances and Household Goods (ISIC 3833)
5.5.3.2 IZEF AR(1) parameter estimates for A5.5-4
Electrical Appliances and Household Goods
(ISIC 3833)
5.5.3.3 13 SLS parameter estimates for Electrical AS.S-S
Apliances and Household Goods (ISIC 3883)
5.5.3.4 13SLS (AR) parameter estimates for A5.5-6
Electrical Appliances and Household Goods
(ISIC 3883)
5.5.4 Monotonicity and Concavity test results A5.5-7
of AESs IZEF estimates, Model 1
(ISIC3833)
5.5.4.3 Fitted cost shares and Own AESs for IZEF A5.5-S
xiv
5.5.4.4 Proper AESs and Determinants of matrices A5.5-8

(ISIC 38~3)
of AESs IZEF estimates, Model 3 (ISIC 3833)
5.5.4.7 Fitted cost shares and Own AESs for IZEF A5.5-1O
(ISIC 3833)
of AESs IZEF (AR) estimates, Model 1
(ISIC 3833)
5.5.4.11 Fitted cost shares and Own AESs IZEF (AR) A5.5-12
of AESs IZEF (AR) estimates, Model 2
(ISIC 3833)
5.5.4.13 Fitted cost shares and Own AESs IZEF (AR) A5.5-13
(ISIC 3833)
5.5.4.15 Fitted cost shares and Own AESs for IZEF (AR) A5.5-14
5.5.4.16 Proper AESs and Determinants of matrices of A5.5-14
AESs IZEF (AR) estimates, Model 4
(ISIC 3833)
of AESs 13SLS estimates, Model 1
(ISIC 3833)
of AESs 13SLS estimates, Model 2 (ISIC 3833)
5.5.4.21 Fitted cost shares and Own AESs for 13SLS A5.5-17
(ISIC 3833)
xv

(ISIC 3833)
of AESs 13SLS (AR) estimates, Model 2
(ISIC 3833)
(ISIC 3833)
(ISIC 3833)
5.5.5.1 IZEF estimates of Economies of Scale A5.5-22
- ISIC 3833
5.5.5.2 13SLS estimates of Economies of Scale A5.5-22
- ISIC 3833
Fig 5.5.6.1 IZEF estimates of scale - ISIC 3833 A5.5-23

Fig 5.5.6.2 IZEF (AR) estimates of scale - ISIC 3833 A5.5-23
Fig 5.5.6.3 13SLS estimates of scale - ISIC 3833 A5.5-24
Fig 5.5.6.4 13SLS (AR) estimates of scale - ISIC 3833 A5.5-24
Fig 5.5.6.5 IZEF, IZEF (AR) and 13SLS estimates A5.5-25
of Model 1 - ISIC 3833
Fig 5.5.6.6 IZEF, IZEF (AR), 13SLS and 13SLS (AR) A5.5-25
estimates of Model 2 - ISIC 3833
Fig 5.5.6.7 IZEF, IZEF (AR), 13SLS and 13SLS (AR) A5.5-26
estimates of Model 3 - ISIC 3833
CHAPTER 1
INTRODUCTION
The fust broad economic policy proposal advanced by the African National Congress
(ANC) after its unbanning, and implicitly endorsed by the Congress of South African
Trade Unions (Cosatu), was ' ... a programme of Growth through Redistribution in
which redistribution acts as a spur to growth.' (McMenamin, 1992, p249 and Cosatu
1992, p14 ). Moll (1991, p314) has identified and criticised two variants of this growth
model: the 'spare capacity' approach and the demand restructuring approach. Neither
approach, Moll (1991, p325) contends, has sufficient empirical support. This
dissertation is explicitly concerned with addressing one, and partially another of the
three empirical issues Moll (1991, p325) argues need to be addressed in order for a
policy of demand restructuring to be successful; namely, the comparative extent of
economies of scale in wage (basic) goods industries as opposed to luxury goods
industries (Moll, 1991, p323) and the magnitude of elasticities of substitution in basic
goods industries (Moll, 1991, p325). The conclusions are not, however, limited to the
debate regarding demand restructuring, and could be useful in the wider context of the
debates and empirical analyses surrounding industrial restructuring and policy.
The demand restructuring approach to growth through redistribution is premised on

the argument that the historically unequal distribution of income in South Africa
distorted industrial demand (Moll, 1991, p320). The distribution of income, it is
argued, has led to a small high income market for durables and luxuries existing
parallel to a limited market for labour intensive basic goods which are subject to high
scale economies (Black, 1991, p165). Reducing the inequality in income distribution
would, it is argued, stimulate demand in the latter market and consequently stimulate
both employment and economic growth.
Moll (1991, p322) has argued that the demand restructuring approach could hold in
South Africa if four conditions, three of which are empirically testable, held. Firstly,
that the consumption basket of the poor is, indeed, more labour intensive than that of
the rich. Secondly, that basic goods industries do experience greater economies of
scale than do luxury goods industries. Thirdly, that manufacturing as a whole can
2
easily substitute between capital and labour, so that the costs of shifting to a new more
labour intensive productive structure are low. The fourth condition is that a suitable
method of redistribution, which ensures that demand changes in the desired manner,
exists. This paper is explicitly concerned with the second issue and partly the third (the
focus here is limited to selected manufacturing sub-sectors, rather than manufacturing
as a whole).
Empirical estimation of the characteristics of the technology underlying either the

whole, or parts, of the South African manufacturing industry is not new. With two
exceptions, all published work has involved estimation of either the Cobb-Douglas
(Cobb & Douglas, 1928) or Constant Elasticity of Substitution (CES) (Arrow,
Chenery, Minhas, & Solow, 1961) function. The choice of function being determined
primarily by the objective of the application.
Both the Cobb-Douglas and CES functions are, however, limited. Both contain
restrictive maintained hypotheses which may be undesirable representations of reality.
A priori they impose constraints of homogeneity, homotheticity and separability on the
underlying technology. A manifestation of these hypotheses is that the derived
estimates of economies of scale and elasticities of substitution are constrained to being
constant, irrespective of the level of output. This dissertation argues that not only are
these hypotheses theoretically undesirable but empirically untenable. Moreover, the
data suggest that imposing the hypothesis of homotheticty biases estimates of the
magnitude of economies scale downwards.
Constrained Cobb-Douglas functions, where economies of scale are held constant,

have been fitted by Browne (1943) and Enke (1962). Unconstrained applications
appear in van der Dussen (1970); Spandau (1973); Matsebula (1979) and Standish and
Galloway (1991). Notwithstanding the restrictions contained in the maintained
hypotheses (in particular the assumptions of homogeneity and homotheticity in output)
of the Cobb-Douglas form, applications of the unconstrained form which estimate
economies of scale for the disaggregated components of South African manufacturing
(van der Dussen (1970); Standish and Galloway (1991)) are perhaps useful for
answering the first question to be addressed in this paper. Van der Dussen , however,
concluded that the Cobb-Douglas form was not suitable for the manufacturing sectors
with which he was concerned (Cluver, 1981, p61). A similar conclusion emerges from
this study.
3
Cobb-Douglas estimates do not, however, contribute to an answer to the second

question. The Cobb-Douglas form limits the elasticity of substitution between inputs to
unity. This constraint on the elasticity of substitution between inputs was a motivating
factor for the development of the CES form which permits an arbitrary constant
elasticity of substitution between two inputs. In the South African context the CES has
been applied by van der Dussen (1970) (Cluver, 1981, p62).
The application of duality theory to economic problems has provided an alternative

approach to that afforded by production functions. Duality theory allows one to obtain
estimates of the features of technology by empirically estimating a,cost function which
is dual to the production function. One such cost function is Christensen, Jorgensen
and Lau's (1971, 1973) Transcendental Logarithmic Cost Function (Translog)
consisting of a cost function and derived factor cost share equations. The Translog
system constitutes a general, flexible functional form which enables the testing of the
validity of the maintained hypotheses imposed by the conventional forms. Further, the
derived estimates of economies of scale and elasticities of substitution vary with the
level of output. The variation of the scale coefficient is particularly valuable as it
provides an indication of the most efficient output level.
The Translog cost system has been used with South African data on one occasion.
Cluver (1981)1 applied a four input (capital, labour, energy and other inputs) Translog
function, to eight sectors of South African manufacturing for the period 1961-1972
(Cluver & Contogiannis, 1984, pI8). A Translog production function, which is in the
same class of function as the Translog cost system, has been used by Van der Walt and
Swanepoel (1987) who applied a two input (capital and labour) Translog function to
aggregate manufacturing data for the period 1946-1983 (van der Walt & Swanepoel,
1987, p39). Cluver's (1981) application is flawed.
The Translog system, unlike either the Cobb-Douglas or CES functions, is not globally
well-behaved. As a result, applications require testing for local' good behavior'. Badly
behaved results suggest that for the data employed, the assumptions used to derive the
dual Translog cost function are violated. Badly behaved results therefore imply that the
assumption of the dual relationship is unwarranted. Cluver's (1981) results suggest a
possible violation of the requirement of well-behaved results, although he fails to
1
Cluver's (1981) results have subsequently been presented in Cluver and Contogiaullis (1984).
4
comment on this. Furthermore the precise econometric technique employed2 is not

presented, and the method which appears to have been adopted is not appropriate.
This dissertation presents an application of the Translog system to three Standard

Industrial Classification (SIC) South African manufacturing industries at the 2, 3 and 4
digit levels for the period 1972-1990. Two of the industries (Electrical Appliances and
Household Goods (ISIC 3833) and Furniture (ISIC 3320)) are deemed to be wage
goods industries and the third (Motor Vehicles Parts and Accesssories (ISIC (3840)) a
luxury industry which is used to draw comparisons with the basic/wage industries.
The second chapter of this dissertation explains why the choice of a flexible non-
homothetic form, which does not impose separability a priori, is justified. The
modeling of two features of technology, namely scale and substitution effects, and the
impact and implications of different maintained hypotheses on the modeling of these
effects is addressed.
Chapter three describes the Translog model indicating the different constraints which
can be imposed on the model, and the constraints which are required for successful
estimation. In addition, the question of why the Translog is preferred to other flexible,
general forms is addressed.
Chapter four presents a discussion of appropriate econometric techniques the data set
which is employed and how different hypotheses will be tested.
Chapter five presents the econometric results while Chapter six is a concluding chapter
which presents the implications for future research emerging from the estimates.
2Estimation of Ihe Translog system involves Ihe dropping of one of Ihe share equations. Appropriate
econometric techniques are those which are invariant to which equation is dropped. Some techniques
generate different results depending on which share equation is dropped.
CHAPTER 2
CHOICE OF AN APPROPRIATE FUNCTIONAL FORM
2.1. INTRODUCTION
The empirical analysis of technology is undertaken for many reasons in different

contexts. No single 'first-best' functional form exists for all purposes, '... to the
contrary, many ... functional forms are well-suited for specific applications but poorly-
suited for use as general purpose characterisations of technology' (Fuss, McFadden & .
Mundlak, 1978, p220). Indeed, the evolution of different functional forms for the
analysis of technology has been influenced primarily by differing objectives of
production studies. Fuss et al (1978, pp220"'221) identify five main objectives of
empirical analyses of technology:
• Distribution (i.e. the share of different factors of production in income) which

provided the motivation for the development of possibly the most ubiquitous
function of all- the Cobb-Douglas function (Douglas, 1948)
• scale (i.e. the existence of constant, increasing or decreasing returns)
l
• the degree of substitutability between factors of production
• separability (i.e. whether or not the production process can be decomposed [i.e.
separated] into additive components)
• technical change where the thrust of analysis tends to focus on three broad areas:
whether technical change is embodied or disembodied2; whether technical change is
factor, scale, or substitution augmenting 3 ; or whether it is endogenous4 •
1Arrow, Cbenery, Minbas and Solow's (1961) development of the CES function, for example, was
motivated by a desire to overcome the a priori restriction on elasticity of substitution imposed by the
Cobb-Douglas form (Jorgensen, 1986, pI843).
2Embodied technical cbange is technical change which is embodied in a factor of production (usually
capital but possibly other factors, most likely skilled labour). Disembodied technical change occurs
when innovations require no specific capital (or presumably any other factor of production).
6
Apart from these five main thrusts of production studies, a number of auxiliary topics
have also been the focus of econometric analyses: technological flexibility (i.e. the
robustness of technology to adjust to changing environments); efficiency (i.e.
operation on or within the technological boundary); and homotheticity (i.e. where
factor shares are unchanged with changes in scale) (Fuss et al, 1978, p222).
While the objective of any empirical analysis will undoubtedly be a central concern in
the choice of functional form, a number of other considerations ought to inform the
practitioner's choice (Chambers, 1988, pI59). In particular, cognisance ought to be
given to the fact that the use of a specific functional form in econometric analyses
(irrespective of the context) requires the acceptance of a number of ' ... maintained
hypotheses which are not themselves tested as part of the analysis, but are assumed
true' (Fuss et al, 1978, p222). Maintained hypotheses could possibly be classified
according to the degree to which they may be regarded as universal truths. In
production analyses, the most fundamental maintained hypotheses are the basic axioms
of the nature of the technology which are widely held to be universal truths; namely
that the production possibilities set is non-empty and closed (Fuss et aI, 1978, p226
and Nadiri, 1982, p432). At the next level are both technological and behavioral
hypotheses [such as monotonicity (Nadiri, 1982, p422) and convexity (Fuss et al,
1978, p222 & p226), or that behaviour is cost minimising] which while not accepted as
universally true are regarded as plausible for the problem at hand. A third level of
hypotheses are those which are made to facilitate the analysis (such as a stochastic
structure of independent normal errors) and are deemed to be harmless approximations
to reality. The fmallevel hypothesis, which is the most restrictive, is the assumption
that a specific parametric functional form is valid. Such hypotheses are made for
convenience and are justified due to a perceived absence of negative consequences
rather than the plausibility of the assumption.
The constraint of an implausible maintained hypothesis concerning the validity of a

specific functional form, is manifested in the testing of specific primary hypotheses
regarding the magnitude of estimated parameters and the overall fit of a modeL The
outcome of the testing of a primary hypothesis will depend on both the validity of the
hypothesis in question and/or the validity of the underlying maintained hypotheses.
'This suggests a general principle that one should not attempt to test a hypothesis in
3Factor augmenting technical change improves the effective quality of inputs; scale augmenting
change expands the scale level where decreasing retums set in, while substitution augmenting change
improves the substitutability of inputs.
4Technical change is endogenous if it occurs by leaming-by-doing. for example.
7
the presence of maintained hypotheses that have less commonly accepted validity'
[emphasis in the original] (Fuss et al, 1978, p223). Chambers (1988, p159) offers the
following example: should one wish to test the hypothesis that a certain elasticity of
substitution were 2, an obvious choice of model would be the CES form. A rejection
of the null hypothesis that the substitution elasticity under consideration were 2 would
not imply that that elasticity is never 2 because by employing the CES the
econometrician has a priori restricted him/herself to considering only constant
elasticities of substitution. A direct implication of the above principle is that (at least)
for tests of the fundamental hypotheses of production theory in general t1exible
5
functional forms, embodying few maintained hypotheses, ought to be used .
Clearly a central, and perhaps primary, concern in the choice among different
functional forms which are able to model the economic effect of interest ought to be
the restrictiveness of maintained hypotheses. However, where a number of alternative
forms are suited to the objective of the analysis at hand, and are compatible with the
same distinct set of maintained hypotheses, additional criteria ought to be used to
choose between competing forms. Fuss et al (1978, p224 & p225), suggest, inter alia,
parsimony in parameters to avoid multicollinearity problems and to preserve degrees of
freedom, ease of interpretation of the economic effects of interest, computational ease,
and interpolative robustness (i.e. within the range of the observed data the chosen form
should be well-behaved, displaying consistency with maintained hypotheses, such as
convexity or concavity). Nevertheless, irrespective of the criteria adopted to choose a
functional form, cognisance ought to be given to the impact that imposed maintained
hypotheses have on the interpretation of estimates of the features of technology and
the testing of statistical hypotheses.
By imposing specific restrictions upon the different economic effects (such as scale and
substitution) modeled by production functions, different functional forms are obtained
(Nadiri, 1982, p439). The maintained hypotheses implicit in any particular functional
form will obviously be manifest in how these effects are modeled by different
functions. An obvious corollary is that the presence and implication of different
maintained hypotheses is often best described by their impact on the different effects
modeled by production functions. How different effects are modeled by different
functional forms, and the relationships between different effects when modeled by
5A further important implication, not explored here, is that given the qualitative, non-parametric
nature of the fundamental axioms of production theory which make extensive use of implicit rather
than explicit functions the more relevant tests will be non-parametric rather than based on parametric
(even general) functional fonus (Fuss et al, 1978, p223).
8
different forms, ought to provide important insights into the implicit maintained
hypotheses.
Although the presence of homotheticity and separability are empirical questions in

there own right, both the Cobb-Douglas and CES functions impose these
characteristics as maintained hypotheses a priori. The Cobb-Douglas and CES
functions are the two forms which have been used to model those features of South
African manufacturing technology under investigation in this dissertation scale and
substitution effects6 • This chapter is concerned with assessing the impact of the
hypotheses of homotheticity and separability on the modeling of scale and substitution
effects respectively, and thus highlighting perceived deficiencies of past attempts to
quantify these effects in South African manufacturing. How scale effects are described
and presented in neoclassical production theory and the impact of homotheticity 7 on
the modeling of scale is the concern of section 2.2. Modeling of substitution and the
impact of separability on substitution is discussed in section 2.3.
2.2. ECONOMIES OF SCALE AND HOMOTHETIC FUNCTIONS
The existence of economies of scale8 refers to all circumstances where the unit cost
decreases with increasing output (Zamagni, 1987, p281). At the broadest level two
types of economies can be identified according to their influence on the long-run
average cost (LRAC) curve of the firm: economies which are internal to the firm and
economies which are external to it. While internal economies determine the shape of
the LRAC curve, it's position is int1uenced by external economies - technological and
factor price changes exogenous to the firm's behaviour (Koutsoyiannis, 1979, p126).
Internal economies are usually classified according to the phenomena which lead to
their emergence, and two types are distinguished: real and pecuniary economies.
Pecuniary economies which emerge as a result of a fmn paying lower prices for factor
inputs and distribution of output are derived from the degree of monopsony power
enjoyed by the fmn in labour markets (Koutsoyiannis, 1979, p128) and the fmn's
contractual strength, due to its size, in other markets (such as capital, materials and
6Meaningful estimates of these two effects required the introduction of technological change into the
model. A discussion of technical change in the context of the particular model used in this study is
provided in Chapter 3.
7Homotheticity is closely related to homogeneity, another common maintained hypothesis. The
discussion of homotheticity includes a discussion of homogeneity and the effects of homogeneity on
the modeling of scale effects.
8While economies of scale refer to all sources of unit cost decreases from producing larger levels of
output the term retums to scale refers to cost decreases due only to technical considerations
(Koutsoyiannis, 1979, p77). Using this distinction, returns to scale would be associated with real
economies as described below.
9
distribution markets) (Koutsoyiannis, 1979, P126 and Zamagani, 1987, p281). Real
economies, on the other hand, are those associated with a reduction in the quantity of
inputs employed by the firm per unit of output as output levels increase. Koutsoyiannis
(1979, pp128-136) identifies four broad groupings of real economies of scale:
production (which includes labour, technical and inventory economies); selling or
marketing (which is associated with the distribution of the fIrm's product) managerial,
and transport and storage. Internal economies can be analysed using either graphs or
mathematical techniques.
2.2.1 ECONOMIES OF SCALE: A GRAPHICAL ANALYSIS
The LRAC curve is the locus of points of the lowest cost of producing different levels
of output when all factors of production are variable (Lipsey, 1963, p227). While the
long-run is conventionally defined as the period of time which is long enough for all
factors to be variable, it is an analytical construct rather than an operating period as
such. Indeed, all economic activity takes place in the short-run (Gould & Ferguson,
1980, p179). An entrepreneur can be regarded as operating in the long-run when
he/she is about to make an investment and is able to choose among different short-run
situations in which he/she will operate in the future. Hence, the long-run should rather .
be viewed as a planning horizon, encapsulating the fact that economic agents can plan
and choose different aspects of future short-runs in which they will operate. This
conceptualisation of the long-run is central to the derivation of the LRAC curve from
short-run average cost (SAC) curves.
The shape and derivation of the LRAC curve is best described by initially employing
the simplifying assumption that at a particular point in time available technology is
such that only three methods of production, corresponding to three different plant
sizes, are available. The smallest plant operates with costs given by SAC 1 in Figure
2.1, the medium size plant with costs given by SAC2 and the largest firm with costs
given by SAC3.
10
Figure 2.1 Short-Run Average Cost Curves for plants of different sizes.
\C, ,
, SAC2
............
\
,, ,.~
/
, I SAC3
\
\
\
'''' "
/
o x'1
Quantity of output
Source: Gould and Ferguson, 1980, p190.
In the long-run the entrepreneur is able to choose between three investment

alternatives represented by the three different SAC curves. Choice of one of the three
alternatives will be determined by expected demand for output. If, for example, the
firm plans (on the basis of expected demand) to produce output of OX 1, it will invest
in the smallest plant size. If it plans to produce OX2, it will choose the medium plant
size, and if it plans to produce OX3, the largest plant size represented by SAC3 will be
chosen. Choice between alternative plant sizes is obviously informed by the lowest unit
cost of producing ditIerent levels of output. At levels of output such as OX I' and OX 2'
two different plant sizes have the same average cost for producing a particular level of
output and the firm can either continue to produce using the present plant size or it can
move to a larger plant size. The decision here would depend on expectations about
future demand. If demand were expected to expand in the future, the larger plant
would be chosen by the cost minimising producer.
If when plant size was chosen planned output was OX 1, the plant represented by SAC 1
would be built. Sinlilarly, if planned output was OX2, the plant associated with SAC2
would be built. The solid line in Figure 2.1 thus indicates which plant size would be
chosen to produce different levels of output, with the choice of plant size being
11
determined by cost minimisation considerations. This line, which is the locus of lowest
cost points when all factors, and hence plant size, can be varied, is thus the planning
curve of the cost-minimising fIrm.
Relaxing the assumption of only three available technologies to now assume that an
inftnite number of technologies and hence plant sizes exist; each suitable for the
production of a certain level of output, a continuous smooth LRAC curve emerges
(see Figure 2.2). Each point on this curve shows the least cost of producing different
levels of output, when all factors are variable. Traditional classical theory of the ftrm
assumes that the LRAC curve is U shaped9 . This is based on the presumption that
returns to scale will initially be increasing, become constant at some level of output,
and thereafter decrease. Economies of scale are thus assumed to exist only up to a
certain plant size, the optimum plant size. Larger plants invoke diseconomies of scale
due to managerial ineffIciencies.
Figure 2.2 Long -Run Average Cost Curve
Quantity of output
9The traditional (classical) assumption of a U shaped LRAC which is based on rhe assumption of laws
of returns to scale has been questioned both on theoretical and empirical grounds (Kotitsoyiannis,
1979, p114).
12
How costs change as output changes, and therefore, whether returns to scale are
increasing, constant or decreasing, can be analysed using an alternative construct the
fIrm's expansion path (EP)lO. This is the locus of tangency points between isocosts and
isoquants indicating the 'rational,ll input-combination for producing different levels of
output. That long-run total costs are directly related to the expansion path can be seen
from Figure 2.3, where panel A presents an EP and panel B a long-run total cost
(LRTC) schedule.
Figure 2.3 The Expansion Path and Long-Run Total Costs schedule.
Labor
(A)
LRTC
'--:-=--~~----'-----0
02 03
Total output
(8)
Point A on the expansion path indicates that output level Ql can be produced by the
ftnn for a minimum total cost of Cl- Similarly, points Band C indicate that the
I1ne concept of an expansion path, in particular its derivation, is discussed in more detail below.
llThat is, consistent with the postulate that the entrepreneur is a cost-minimiser.
13
minimum total costs at which output levels Q2 and Q3 can be produced are C2 and
C3, respectively. Greater levels of output are associated with higher costs. Repeating
this procedure for all levels of output would yield a locus of output-cost combinations
- the LRTC schedule. Such an exercise has been undertaken to derive the LRTC
schedule in panel B, where the points corresponding to the EP of panel A are
indicated. The LRTC curve is merely cost-output equivalent of the EP (Gould &
Ferguson, 1980, pI76). Where the slope of the LRTC curve is increasing, the
implication is that expansion of output is becoming increasingly costly, implying that
LRACs are increasing and that returns to scale are decreasing. Hence, the change in
returns to scale along the expansion path (i.e. output responses to changes in costs)
determines the shape of the LRAC curve.
2.2.2 ECONOMIES OF SCALE: A MATHEMATICAL EXPOSITION
Mathematically, two different concepts are used to describe the phenomenon of

returns to scale, and derive measures of scale effects from different functions
(production and cost) ~ the elasticity of scale and the elasticity of size (Chambers,
1988, p72 and Hanoch, 1975, p492). The two concepts are often mistakenly used
interchangeably in the literature. While the two measures coincide at cost-minimising
points (such as A, B, and C in panel A of Figure 2.3), they are different phenomena
and unless the production function is homothetic the two measures will differ in
response to a change in output. In this regard homothetic functions are unique. The
elasticity of size is, however, the more relevant of the two measures for the fum.
2.2.2.1 ELASTICITY OF SCALE
l2
For a regular production function f (x), where x is a vector of factor inputs
(XI' x 2 "'" x" ), the elasticity of scale 13 , which is the more commonly used defInition of
returns to scale, is defined as:
e(x) = dIn !(kx)1 2.1

dIn k k=l
where: k is some scalar and kx is a formula for a ray through the origin in the input
space. e(x) is, thus, an elasticity coefficient measuring the relative increase in output
12A production function is deemed regular if that function is: positive; finite; continuously twice
differentiable; strictly monotonic; and strongly quasi-concave.
13First identified by Johansen (1913), and known variously as the "elasticity of production"; "passus
coefficient" and the" function coefficient".
14
as all input quantities are increased proportionally, i.e. along a ray from the origin in
input space (Chambers, 1988, p72; Hanoch, 1975, p492 and Jehle, 1991, p227). The
elasticity of scale delineates three types of returns to scale: decreasing, constant or
increasing as e(x) is less than, equal to or greater than unity respectively. The
delineation of returns to scale is often erroneously applied. The elasticity of scale,
because it involves the use of a derivative, is a local measure, not a global one. For
production to be characterised by either decreasing, constant or increasing returns to
scale over the entire input space, output must always responds to proportional changes
in inputs in the same quantitative manner, irrespective of the initial level of output
(Jehle, 1991, p227). Most technologies exhibit increasing, constant or decreasing
returns to scale over different ranges of output (Jehle, 1991, p227 and Chambers,
1988, p24). Assuming e(x) is constant, as in homogenous production functions, is
obviously highly restrictive.
Marginal productivity may be measured by a unit-free measure, the elasticity of output,

which for a production function y = f(x) is defined as:
df(x) Xi
e=--·- 2.2
dXi Y
I
The elasticity of output, is thus merely the ratio of the marginal product of the ith input
4
to the average produce , and provides a measure of the percentage change in output in
response to a one percent change in the ith input (Chambers, 1988, p18 and Nadiri,
1982, p439). In other words, the elasticity of output provides a "normalized" measure
of the relative importance of a particular input (Griliches & Ringstad, 1971, p6). The
elasticity of scale can be shown to be mathematically equivalent to the sum of factor-
output elasticities l5 :
14The first term in the formula is marginal product while the second is the reciprocal of average
product. Multiplying marginal product by the reciprocal of average product is equivalent to dividing
marginal product by average product.
15.
Given e (X) = dind /(Iex)1 .
and lettmg U
' " tlIat U IS a vector) and y =/ (u).
=lex (Implymg
In k k=1
.) dln/(u)
q e( X =--=----
dlnk
qe(x) =~. d/(U) [Given tllat y = feu)]
y dk
q,,(x)
<;..
=~ £..J
~ df(u). dUi [Using tlIe Chain Rule given tllat u is a vector]
y i dUi dk
k
qe(x) =-Y Lfi.Xj
I
15
2.3
An implication of this relationship is that changes in the relative importance of different

inputs impacts on the magnitude of the elasticity of scale.
2.2.2.2 ELASTICITY OF SIZE
The second measure of returns to scale is the elasticity of size, which measures the
increase in output relative to costs for variations along the expansion path (i.e. the
locus of cost minimising points in input space). Mathematically the elasticity of size is
expressed as:
\!I(X) = aln Y[ 2.4

alnc w
The elasticity of scale and the elasticity of size are equal at cost minimising points.
Establishing the equality of the two measures at cost minimising points rests on two
results, both of which stem from a manipulation of the [lIst order conditions of a
constrained cost minimisation problem (Henderson and Quandt, 1980, p77 and
Hanoch, 1975, p493).
The Lagrangian for a constrained optlmlsation problem where the entrepreneur

attempts to minimise the costs of producing a given output l = j(x), where x is a
vector of factor inputs (Xl' X 2 ' ... , x1/) and the total cost function has the specific form
1/
C= I WiXi ' where wj is the price of factor i, would be:

i=1
TI
Z =I WjXi + ~[l- j(x)] 2.5

i=1
At constant prices the first order conditions for a minimum would be given by the
following set of simultaneous equations: .
[By definition (see equation 2.2)]

16
az
-a =w ~!i =0, i=I,2, ... ,n
j -
Xi 2.6
az
-=y 0
- !(X)= 0
a~
Manipulating of this set of simultaneous equations yields the result that the Lagrangian
multiplier ~ equals marginal cost: Keeping wi (for i =1,2, ... ,n) constant the
differential of the cost function employed here is:
n
~ wdx
de= L... I I
2.7
i=1
The fIrst equation of 2.6 yields wi =~J:. Substituting this into 2.7 gives;
n
de =~ L J:dx j 2.8
i=1
Given that the differential of the production function is
n
dy = LJ:dxj 2.9
i=1
dividing 2.8 by 2.9 will yield marginal cost:
11
~L!idxi
ae
- = i=1 2.10
=~
ay 11
L!idxi
i=!
The equality of the Lagrangian multiplier with marginal cost is the fIrst result needed
to prove the equality between elasticities of scale and size at cost minimising points.
The second result needed to prove the equality emerges from a manipulation of the
fust equation in 2.6. Rearranging that equation, and using the result of 2.10 yields:
ae a!(x)
-·--=w· 2.11
ay aXi I
which provides the result that the [urn will minimise costs if it hires inputs, and
produces the level of output where the product of marginal cost and marginal product
17
of ~ach input equals the price of each input. Substituting the defmition of output
elasticity (2.2) into 2.3 yields:
2.12
Rearranging equation 2.11 by making marginal product the subject of the formula and
substituting this into 2.12 allows one to express the elasticity of scale as:
e(x) = '"
£... w-x.
1 1 2.13
~f(x)
The numerator of equation 2.13 is total costs, hence 2.13 can be expressed as:
c AC
e(x)=--=- 2.14
~f(x) ~
where AC is average costs (formally AC =c/y). Now, because (by virtue of 2.10)
/-.l = dC/dy,
CldC
e(x) =- -
y dy
and, therefore;
dc C dln y
e(x) = - . - = - - = ",(x) 2.15
dy y dlne
An important issue is which of the two measures is the more relevant for the fInn. The
reciprocal of the elasticity of size is the ratio of marginal to average cost or the cost
tlexibility ratio. The cost tlexibility ratio is a central detemlinant of the shape of the
total cost function. Graphically, marginal cost is the slope of the total cost function
while average cost for any level of output is the slope of ray from origin to the point
on the total cost curve corresponding to that level of output.
18
Figure 2.4 Total cost, average cost, marginal cost and cost flexibility.
c.(w,y)
Source: Chambers, 1988, p78.
In Figure 2.4, when costs are given by point A on c( w, y) marginal cost and average
costs are equal and the cost flexibility ratio will be 1. For levels of output less than that
corresponding to point A on the cost curve, a ray from the origin will cut a tangent to
the curve from below implying that the slope of the ray from the origin is greater than
the slope of the tangent to the curve at that point, indicating that average cost is
greater than marginal cost and that the cost flexibility ratio is less than 1. Thus when
cost flexibility is less than 1 economies of size are greater than 1 (economies of size
being the reciprocal of cost flexibility). The converse will occur for levels of output
greater than the level corresponding with point A on the cost function. Clearly, the
magnitude of economies of size, which is inversely related to the cost t1exibility ratio,
is central to the determination of the shape of the average cost curve and leads Hanoch
(1975, p492) to argue that the elasticity of size is the more relevant measure for the
analysis of finn and industry behaviour. Just as the elasticity of scale delineates three
types of returns to scale (usually referred to as economies of scale), the elasticity of
size delineates three types of returns to size, depending on whether 'V(x) is greater
than, equal to or less than unity.
19
Although the two measures of returns to scale described above generate equivalent
measures at cost minimising points, how each changes with changes in output is
different. Only if a production function is homothetic will the two measures change in
l6
the same manner as output changes. While no proof of this result is provided here an
intuitive explanation of this result emerges from the fact that homothetic functions are
the only class of functions where the expansion path is a straight line (ray) through the
origin. It is to a discussion of the features of homothetic functions that I now turn.
2.2.3 HOMOTHETICITY
The concept of homotheticity is closely related to, albeit more general than, the
concept of homogeneity. While every homogenous function is homothetic, homothetic
functions are not necessarily homogenous (Chiang, 1984, p423 and Madden, 1986,
p240). A function f (XI"'" XII) is said to be homogenous of degree p if multiplication
of each of its independent variables by some constant k alters the value of the function
by the proportion e, in other words if the following relationship holds:
2.16
In production theory wide use is made of homogenous functions in general and

functions which are homogenous of the fIrst degree (i.e. linearly homogenous, where
p = 1) in particular. The assumption of linear homogeneity imposes an assumption of
global constant returns to scale on the underlying technology as raising all inputs
(independent variables) k-fold will raise output (the value of the function) exactly k-
fold. Similarly, for a function which is homogenous of a degree greater than (less than)
unity, returns to scale will be globally increasing (decreasing).
That returns to scale are modeled as a global phenomenon for homogenous functions
is readily apparent from an application of the defInition of the elasticity of scale
(equation 2.1). Homogeneity (defIned by equation 2.16) can be expressed in
logarithmic form, as:
In f (lex l , ... ,kxlI ) = pln k + In f (Xl , ... ,XII) 2.17
Substituting 2.17 into the defInition of elasticity of scale (2.1) yields:
16See, for example Hanoch (1975) or Chambers (1988) for proofs of this result.
20
d[Plnk+ln/(Xl, ... ,Xn)]1 2.18

() =
eX =P
din k k=l
implying that for homogenous functions returns to scale are always given by a constant
(p) and therefore globally defined. This feature of homogenous functions is an
important limitation. While economies of scale are in general a function of the input
bundle and the level of output, for homogenous functions they are invariant to any of
these features being modeled as global, rather than as local phenomena. The
implications of this have already been discussed.
A function is classified as homothetic if it can be regarded as a monotonically

increasing transformation of a homogenous function. Mathematically a function I (x)
is homothetic if it can be represented as: I(x) = F[f'(x)] (Clemhout, 1968, p91),
where F is a monotonically increasing function of I' (x), which is regular and
17
homogenous • Because a homothetic function is a transform of a homogenous
function, an intuitive explanation of a homothetic function requires an intuitive
explanation of a transform.
A transform of a production function, as suggested by the definition of a homothetic

function, is a function of a production function. Because output is simply an amalgam
of inputs and technology, one can consider a production function as merely an
'aggregate input' and consequently a transform can be viewed as a single input
production function (Shepherd, 1970, cited in Chambers, 1988, p37).
An intuitive explanation of a homothetic function, requires a consideration of

homogeneity and points to how scale effects are modeled by these functions. If a
production function is homogenous an equi-proportional change in all inputs will lead
to a proportional change in the value of the function. Hence homothetic functions
(which are transforms of homogenous functions) are that class of transforms where
proportionate changes in all inputs are accurately expressed by a proportionate change
in the aggregate input (Chambers, 1988, p38). Thus for homothetic functions
increasing the scale of operation for each of the actual inputs is equivalent to an
17Different authors disagree as to the degree of homogeneity of I' (x). Chambers (1988, p37),
Lancaster (1968, p334), Clemhout (1968, p9l) and Madden (1986, p240) for example, indicate that
I' (x) is linearly homogenous; while Hanoch (1971, p697 and 1975. p492) argues that f' (x) may
be homogenous of any degree and the definition provided by Lancaster (1968) is erroneous. The
confusion may however stem from the fact that Shepherd (1953 and 1970), who has been
instrumental in introducing the concept into economic analyses, used a definition of homotheticity
where f' (x) was linearly homogenous (Denny & May. 1978. p65).
21
increase in the scale of operation for the aggregate input, and where scale-type
decisions must be made, no generality is lost in dealing only with the aggregate input
(Chambers, 1988, p38). This intuition is obvious when one considers that an important
feature of homothetic (and homogenous) functions is that they display straight line
expansion paths. A straight line expansion path implies that when output is increased
the ratio in which inputs are used remains fIxed. An implication of this is that
economies of scale for homothetic functions are not influenced by changes in relative
input utilisation (which is fixed) but are a function of the level of output alone. Formal
proofs of both the linearity of the expansion path and the result that scale effects are a
function of output alone are presented below.
2.2.3.1 EXPANSION PATHS OF HOMOTHETIC FUNCTIONS
Proof of the linearity of the expansion path of homothetic functions is an extension of

the proof that the expansion path of a homogenous functions is a straight line.
Assuming only two factor inputs, the first equation of the fIrst order condition of the
cost minimising problem will provide the following equality 18:
2.19
that is, at the point of optimal input combination, the input price-marginal product
ratio for each input must be equal for both inputs (and equal to marginal cost).
Equation 2.19 can be rearranged and expressed in the form:
2.20
18 For the two input case, the cost function would be: c = w\x\ + W 2 X 2 and the production function;
yO = f (Xl' x 2 ), yielding a Lagrangian for the constrained cost minirnsation problem having the
form Z = w\x\ + W 2 X 2 + fl.[/- f (XI' x2 )]. First order conditions for the optimsation problem are:
az
-=w\-fl.f\=O
ax\
-a
az
= W 2 -fl.f2 =0
X 2
az
afl. =y - f (X) =0
Manipulation of the fIrst two equations of these fIrst-order conditions will yield equation 2.19.
22
9
The fll fz ratio is the negative of the slope of the isoquane ; hence it is a measure of
the marginal rate of technical substitution of input I for input 2. The price ratio in
equation 2.20 represents the negative of the slope of an isocost line - the locus of input
combinations yielding the same total cost. Total costs for the two input case are given
by:
2.21
Making X z the subject of 2.21 provides the result that the price ratio wllwz (the term
on the left-hand side of equation 2.20) is indeed the negative of the slope of the isocost
line:
2.22
When plotted in the xlXZ plane, equation 2.22 will yield a family of straight lines whose
slopes, given by the negative of price ratio, are equal while prices are fixed. The
vertical intercept, given by the ratio clwz ' is proportional to the level of costs (c). The
equality implied by the first order condition of the cost minimisation problem and
expressed in equation 2.20 is equivalent to the graphical condition of equality of slope
between isoquant and isocost lines. Given that the isoquants are strictly convex to the
originZO the requirement is one of tangency between the isocost and isoquant curves.
Graphically:
19 Alongan isoquant output is constant, hence in the 2-input case dy = fldx l + fzdx z = o.
Rearranging yields: dxjdXl = - ft! fz . Given that an isoquant is drawn in the input spac~, this
equation provides an expression for the slope of an isoquant.
20The second order conditions of the cost minimization problem require the isoquants be strictly
convex at the chosen input combination (Chiang, 1984, p362).
23
Figure 2.5 Expansion path of a homothetic function
Source: Adapted from Chambers, 1988, p 72.
The least cost combination for producing a level of output l is the input combination
(XI' x 2 ). For successively higher levels of output, the least cost input combination will
correspond to the tangency of both a higher isoquant and a higher isocost. As
discussed earlier, the locus of points of tangency (i.e. the locus of cost-minimising
points in input space describing the least cost combinations required to produce
varying levels of output) is known as the expansion path of the firm.
The linearity of the expansion path of a homogenous function rests on the following
argument: if a function (y = f(x\, Xl» is homogenous of degree p, then its marginal
products fl and f2 will both be homogenous of degree p -1 in the inputs XI and x 221.
Hence multiplying both inputs by a constant k will raise the value of both marginal
products by k p- I , and leave their ratio unchanged. Thus if the fIrst order condition 2.20
is satisfIed for an input combination (xl' x 2 ) then for a given fIxed price ratio it will
also be satisfied for a combination (k.xl'k.x 2 ). Whenever the initial cost minimising
quantities of inputs are multiplied by the same scalar the cost-minimising condition will
remain fulfilled. Graphically this implies that the cost minimisation problem will be
21This follows from the general result that the nth order partial derivatives of a function homogenous
of degree p are themselves homogenous of degree p- n (Lancaster, 1968, p335).
24
satisfied for any multiple of the initial input combination, implying that the locus of
cost minimising points (the EP) will be linear for homogenous functions.
Turning to the case of a homothetic function: y = F[f* (Xl' x z )]' where 1 (x" x 2 ) is
homogenous, the slope of its isoquant will be:
a yla Xl _ F (l )..rt = _.rt

I
2.23
a yla Xz -- F'(f*).1; 12*
where the first equality emerges form the application of the chain rule. 2.23 indicates
that the cost minimising condition will be given by:
a yla Xl _ WI _ it
a yla z -X W
z - 12*
Because 1 * (xl' X z ) is homogenous multiplying both inputs by the same constant will
leave the ratio 11* /1z* unchanged. The implication is that for a given price vector the
locus of cost minimising points will be a straight line through the origin.
2.2.3.2 SCALE EFFECTS UNDER HOMOTHETICITY
The coincidence of the EP with a ray has important implications for the modeling of
the elasticity of scale. In general the elasticity of scale, which is concerned with
measuring how output varies along a ray from the origin, is a function of the input
vector and the level of output. This can be seen from substituting equation 2.2 into
equation 2.3:
c(X) = L Xi. aj(x)

i y dXi
=g(x, y) 2.24
This result does not, however, hold for either homogenous or homothetic functions,
and is a feature which distinguishes those classes of functions from others.
For homogenous functions the elasticity of scale is a global constant and therefore
independent of the input vector and the level of output (see equation 2.18 above). For
a function which is homothetic, the elasticity of scale is merely a function of the level
of output. To prove this result, an alternative, but equivalent, defmition of a
homothetic function needs to be employed (Chambers, 1988, p38). The original
25
definition of homotheticity used above was that a function f Cx) is homothetic if it can
be represented as: fCx)=F[f"Cx)], where r(x) is regular, and therefore strictly
monotonic, and homogeneous. An alternative expression would be:
y = F[f"(x)] 2.25
where y is output. Because FO is, by definition monotonic, the inverse of FO will

be defined and exist CChiang, 1984, p172). An implication of the existence of the
inverse of FO is that a homothetic function can alternatively be expressed by:
Specifying hCy) = F-1Cy) a homothetic function may be described as:
hCy) = f"Cx) 2.26
where f" (x) is homogenous. Using 2.3, which shows that the elasticity of scale is the
sum of output elasticities of different inputs, the elasticity of scale can be expressed as;
2.27
Using the chain rule of differentiation and applying 2.27 to 2.25 yields:
I dy df'Cx)
e= I dy . df"Cx). Xi = i drCx)' dXj 'X
i
2.28
j dr(x) dX j y y
Given that f" (x) is homogeneous 2.28 can be simplified by applying Euler's theorem 22
dy "
e=dr(x)'Pf (x)
2.29
y
22Eulers theorem states that if some function y = f (x) is linearly homogenous, then
L :; .Xi = Y (Chiang, 1984, p413). If the function is homogenous of degree p then the Euler's
I I
theorem implies that L ~y . x =py, where p is tlle degree of homogeneity of t1le function.
j
I oXi
26
where p is a constant indicating the degree of homogeneity of f'(x). By virtue of

2.26, equation 2.29 can be written as:
_d_y_.ph(y)
e = _dh~(=y..:..-)_ _ 2.30
y
23
Applying the inverse function rule , 2.30 yields:
e = ph(y) 2.31
h'(y)y
where h'(y) is the derivative of hey) with respect to y.
Equation 2.31 shows that for a homothetic production function, the elasticity of scale
is a function of the level of output y only. Moreover, given the coincidence of the
expansion path with a ray through the origin under homotheticity (and the consequent
equality of the elasticity of scale and the elasticity of size) the elasticity of size is also a
function of the level of output alone, responding in the same manner as the elasticity of
scale to output changes. This equality of the two measures of scale also holds for
homogenous functions which by definition are also homothetic.
The result that the elasticity of scale is a function of the level of output alone for
homothetic functions is intuitively appealing. A straight line expansion path implies
that the isoquants are parallel and their shape is independent of the scale of production
(Clemhout, 1968, p94; Griliches & Ringstad, 1971, p7 and McFadden ,1978, p77),
and therefore when output is increased the ratio in which inputs are used remains
fixed.
The characteristic of a straight line expansion path under homotheticity (and the
implied equality of the elasticities of scale and size) will also have an impact on the
shape of the cost functions associated with homothetic production functions. The
shape of the total cost function, as shown above, is determined by the cost flexibility
and hence the elasticity of size. For homothetic functions the cost flexibility cannot be
23If a function f (x) is monotonic it has an inverse and the inverse function rule of differentiation
dx 1
=---/- holds (Chiang, 1984, pI71-173). Given that hey) is by definition monotonic
df(x) df(x) dx
~= 1 = 1
dh(y) dh(y)/dy h'(y)
27
influenced by changes in input use and hence the shape of the cost functions will be a
function of the level of output only for a given input-price ratio. Changing relative
input utilization as output expands (which is manifested in an expansion path deviating
from a ray through the origin) will obviously impact directly on average costs and total
costs. Explicitly preventing such chamges is obviously a restrictive feature of
homothetic functions.
2.2.4 HOMOTHETICITY AND COBB-DOUGLAS AND CES FUNCTIONS
Both functional forms used, most often, to analyse features of the technology
employed in South African manufacturing, the Cobb-Douglas and CES functions, are
homogenous and thus homothetic: Consider an n-input Cobb-Douglas function:
n
y = ArrX~j 2.32
;=1
where A is an efficiency parameter and the Xi are the inputs. Input elasticities are given
by;
2.33
The elasticity of scale is therefore:
"
E(X) =Ia i 2.34
i=1
This result indicates that for constant a i s, E(X) will be constant and will not vary with
different levels of output, implying that the function is homogenous. Homogeneity of
the Cobb-Douglas in turn implies homotheticity of the function, and the properties of
homothetic functions described above apply to the Cobb-Douglas function.
The CES function can be expressed as:
2.35
where 'Y is an efficiency parameter; 8 (0(8(1) is the input intensity, p (00 ~ p ~ -1) is
the substitution parameter and v represents the degree of homogeneity of the function
and is the returns to scale parameter (Brown, 1968, p45). Clearly, the CES models
28
economies of scale as a global phenomenon, independent of the level of output.

Homogeneity of the CES implies homotheticity of the function, and the properties of
homothetic functions described above apply to the CES function.
As argued in the introduction to this chapter the presence of restrictive or implausible

maintained hypotheses will be manifested in the testing of specific primary hypotheses.
Using either the Cobb-Douglas or CES forms to test the hypothesis that economies of
scale are, for example, of a particular magnitude could rest on the validity of the
underlying hypotheses of homotheticity and homogeneity which impose constraints on
how scale is modeled. If one is to use either the Cobb-Douglas or CES forms with
confidence the validity of the hypotheses of homotheticity and homogeneity ought to
be established. The introduction of the mathematical concept of duality to the analysis
of economic phenomena has led to the development of functional forms which are
sufficiently flexible to allow for the testing of homotheticity and homogeneity, rather
than imposing these hypotheses a priori. The Transcendental Logarithmic cost
function (Christensen, Jorgensen and Lau, 1971 and 1973) (Trans10g) is an example of
a flexible function emerging from the use of duality theory which allows the explicit
testing of the validity of the hypotheses of homotheticity and homogeneity. Using this
function to analyse scale is clearly superior to adopting either a Cobb-Douglas or CES
form. Both the concept of duality and the specific form of the Translog are developed
in Chapter 3.
2.3 ELASTICITIES OF SUBSTITUTION AND FUNCTIONAL

SEPARABILITY
This dissertation is explicitly concerned with attempting to quantify two features of the
technology employed in specific South African manufacturing subsectors: returns to
scale and elasticities of substitution. The impact of the maintained hypotheses
embodied in the most popular functional forms on the modeling of scale effects has
been addressed above. How elasticities of substitution are modeled in theory and the
impact that different maintained hypotheses have on elasticities of substitution is the
concern of this part of this chapter.
Berndt and Christensen (1973a, p403-409) have shown that for production functions
employing more than two factor inputs, the internal structure of a function, and in
particular whether a function of several arguments can be separated into sub functions,
is closely related to equality constraints on one particular measure of the partial
elasticity of substitution between inputs, namely the Allen partiaL eLasticity of
29
substitution (AES) (AlIen, 1938, p504). The corollary to their result is that, where
separability is a maintained hypothesis, the modeling of AES will be restrictive.
Although a number of different measures of the elasticity of substitution exist when
more than two inputs are considered the AES is of particular interest. AESs can be
obtained from simple algebraic manipulations of the parameters of the function
employed here. Moreover AESs play an important role in the estimation of the
function used in this analysis.
2.3.1. ELASTICITIES OF SUBSTITUTION
The possibility of producing a constant output level with a variety of input

combinations and the degree to which inputs can be substituted for one another is an
important question for economic decision making (Chambers, 1988, p28). The most
common measure of substitutability is the marginal rate of technical substitution
(MRTS) which measures the rate at which one factor can be substituted for another
factor while the level of output remains constant. Graphically it is a measure of the
slope of an isoquant. Mathematically, for a two-factor production function
y =f (Xl' x 2 ), it can be represented by:
2.36
The assumption that the production function is monotonically increasing in each of the
inputs (if x; ~ Xl' then f (x;) ~ f (Xl) and if X~ ~ x2 then f (x;) ~ f (x2 » implies that
the MRTS must be everywhere negative, because output can only remain constant if
when the use of one factor is increased, the use of another is decreased. The
assumption that isoquants are convex to the origin means that the MRTS is
everywhere diminishing in value (Jehle, 1991, p221). The existence of diminishing
MRTS, which implies some friction in substitutability, is intuitively appealing as it is
likely that it will become increasingly difficult to substitute one factor for another and
still maintain output at a fixed level.
Elasticities of substitution are essentially unit-free measures of the MRTS. For the
two-input case, the elasticity of substitution has been defined by Hicks (1963, cited in
Chambers, 1988, p29 and Jorgensen, 1986, p1844) as:
() = d1o(X2/X1) = d(xZ/x1) .fJf2

2.37
lZ d 10 (fJ j~ ) d U; / j~ ) Xz/ Xl
30
Hence, 0 is the elasticity of the input ratio with respect to the MRTS. In other words,
it is a measure of the proportional change in the factor ratio (x21Xl) resulting from a
proportional change in the MRTS of input I for input 2. When factor prices are ftxed
as quantities of factors demanded change (and the tinn is a cost minimiser) 0 may be
interpreted as the percentage change in the ratio of inputs given a small percentage
24
change in the ratio of marginal products or of the price rati0 (Nadiri, 1982, p442).
An alternative explanation of the concept can be obtained from the following graph:
Figure 2.6 Elasticity of Substitution
J!::..... ----.:~ __::::."____ Xl
Source: Chambers, 1988, p 31
If the initial input ratio was given by the ray OAC, the corresponding (initial) MRTS
would be given by the slope of the tangent to the isoquant at point C. Should the input
ratio change to one given by the ray OBD, the MRTS would now be given by the
tangent to the isoquant at point D. 0 can hence be regarded as the ratio of the two
angles d k I d( , and can be regarded as a measure of the curvature of the isoquant. This
can be seen from an interpretation of possible values for 0. The value of 0 ranges form
zero to inftnity. The closer 0 is to zero, the more difficult substitution between factors
becomes. Because the ratio d k I d( becomes smaller as 0 approaches zero, the
24The first order conditions of tlle cost minimising problem can be manipulated to obtain tlle equality:
wJw2 = fJ f2 (see equation 2.20).
31
implication is that the isoquants will be more convex the smaller a is. In the limit, when
a = 0 and there is no substitutability between factors, isoquants will be right-angles to
the origin. Conversely the larger is cr, the Hatter the isoquants and the easier
substitution between factors.
Two important features of the elasticity of substitution need to be noted: fIrst that it is
always positive in the two-input case and secondly that it is symmetric i.e. a 12 = a 21 •
The fIrst of these results can be explained using the above graph. A movement from
point C to point D would imply an increase in both the x2 1XI ratio, implying that
25
d(x2 Ixl ) would be positive, and in the illi 2 rati0 , implying that d(iJi 2 ) would be
positive. Given the defInition of a (equation 2.37) it is clear that moving from C to D
would yield a positive a. Movement the other way generates a negative numerator and
denominator in the formula and hence a positive cr. In the two input case, then, factors
can only be characterised as substitutes (Chambers, 1988, p32).
The second feature of the elasticity of substitution which warrants some explanation is
the symmetry of the measure. Symmetry is intuitively appealing but can also be
explained by considering an alternative formulation of the defInition. a can equivalently
be expressed as (Chambers, 1988, p32):
F is the determinant of the bordered hession of the production function:
o il i 2
F= i l ill i l2
j~ 121 122
and F;2 is the cofactor of i 12 · Finding cr21 would merely require the replacement of F;2
with F;l' which will be equal given that if 1 (X) is assumed to be regular, it will be
twice continuously differentiable and Young's theorem U;2 = j~l) will hold.
The discussion up to this point has been limited to the case of only two inputs in the
production process. In the more general n-factor case only the concept of partial
25Due to the assumption of diminishing marginal productivity using less of factor Xl implies that its
iI
marginal product i l will rise and vice versa, hence itS Xl is substituted for x 2 the ratio 1 j~ will
increase.
32
elasticity can be defined. Elasticity of substitution for the two input case is defined
subject to the constraint that the level of output remains constant. Generalisation of the
elasticity concept to the n-input case requires in addition to the constraint of constant
output, the imposition of further constraints on the variables. Depending on the
constraints imposed a number of different elasticities exist (Griliches & Ringstad,
1971, p6). Of the available alternatives, two will be developed here: the direct
elasticity of substitution (DES), and the Allen partial elasticity of substitution
(AES)26. Only the AES is of direct relevance to the function used here. It is not only
directly computable from parameter estimates but also imperative for generating
27
meaningful results • The DES is developed here because it would appear to be the
most natural definition in the n-input case, and provides a useful indicator of how the
AES differs from the concept of elasticity of substitution as defined for the two-input
case.
The DES is defined as:
2.38
where all Xk (k:#= i, j) are held constant. Because this measure assumes that inputs
other than those with which it is directly concerned are held constant, it can be
regarded as a short-run measure. The interpretation of the DES is essentially the same
as that for the two-input elasticity: it is a measure of the proportional change in the
factor ratio (xjx j ) resulting from a proportional change in the MRTS of input Xi for
input x j holding all other factors and the level of output constant. Moreover, as in the
two input context, (Jif is symmetrical. Goods are regarded as complements when the
DES is negative and substitutes when it is positive.
Possibly the most used definition of elasticity of substitution for the n-input case is the
AES (AlIen, 1938, p504). This is a measure of the change in a firm's derived demand
for factor j given a change in the price of factor i , all other factor prices and output
26Further examples of elasticitys in the n-input case are the Shadow elasticity of substitution
(McFadden, 1978, p80) and the Morishima elasticity of substitution (MES) (Chambers, 1988, p35).
Different definitions of the elasticity of substitution in the n-input case display vastly differing
characteristics. While both the DES and AES are symmetrical, the MES is not symmetrical.
Furtllermore, while Allen substitutes are always Morishima substitutes, Allen complements are not
necessarily Morishima complements (Chambers, 1988, p35).
27The reasoning is developed in Chapters 3 and 4. .
33
remaining constant (Bemdt & Christensen, 1973a, p405). Formally, the AES is defmed
as 28 :
2.39
where:
11
0 111
F=
11 III 1111
In 1111 I11n
and F;j is the cofactor associated with the element lij' Stability conditions29 are
fulfilled if F',.nl F(O, (i.e. if F',.n and F are opposite in sign) (AlIen, 1938 pp502-504).
Although the values of the AES can be positive or negative (a negative AES indicating
complementarity and a positive AES, substitutability), the stability conditions require
that, when weighted, the positive elasticities must counter balance the negative
elasticities (AlIen, 1938, p505 and Nadiri, 1982, p443). 'In particular, the (11-1)
partial elasticities of substitution between anyone factor and the others cannot all be
negative' (Allen, 1938, p505). In other words, in terms of the AES a factor of
production cannot be a complement to all other factors of production. Chambers
(1988, p35) argues that this restriction is intuitively appealing, given that in the two-
factor case all inputs are substitutes. As in the case of the DES the AESs are
symmetrical.
While the interpretation of the DES is relatively straightforward the interpretation of

the AES is more complex. The AES provides information on the cross-demand
elasticities of inputs (McFadden, 1978, p80). When production is efficient and when
the supply of inputs is perfectly elastic (ftrms are perfect competitors in input markets)
the AES and cross-demand elasticities of inputs are related in the following manner
(Allen, 1938, p508):
28 . . . .
If 1 = } then the AES is termed an own AES. If 1 ~ } then !lIe AES is regarded as a proper AES.
29 Stability conditions refer to the stability of demand for factors of production. In the 2-input case, for
example, demands for factors of production are stable if !lIe isoquants derived from the production
function are convex to the origin at all relevant points (AlIen, 1938, p502).
34
A Tlij 2.40
cr·· = ---"--
IJ W.x.jC
I I
where, Tlij is the partial (cross) elasticity of demand for factor j with respect to the
price of another factor i (Tlij =aln xJaln wi ), WjXj is the expenditure on good Xi and
C is total costs. The denominator of the term on the right-hand side will always be a
positive fraction (O(PiXJC(l), hence the sign of cr~ will determined by the sign of Tlij'
Should Tlij be positive, in terms of cross elasticity of demand factor j would be
regarded as a substitute for factor i, cr~ would also be positive and factor j would be
deemed to be competitive with factor i for the grouping of factors considered. On the
other hand, if Tlij were negative, cr~ would also be negative and factors j and i would
be considered complementary for the grouping of factors considered (Allen, 1938,
p509).
2.3.2 SEPARABILITY
Both analytical and econometric considerations have influenced the tendency for
production analysis in general, and applied production work in particular, to assume a
relatively small number of input types. Analytical considerations arise from the
difficulties associated with geometric expressions of dimensions higher than two, while
econometric analyses have been constrained by the need for data with sufficient
independent observations. Both considerations have resulted in economists generally
adopting the classical categorization of inputs as land, capital and labour (Chambers,
1988, p4l). The classification of inputs into different types is usually accompanied by
the assumption that the degree of substitutability between inputs of one type differ
from the degree of substitutability between inputs of other types (Jehle, 1991, p22l).
Production functions whose form embodies this type of assumption are deemed
separable (Jehle, 1991, p221)
A more general explanation of separability is provided by Chambers (1988, pp42-45).

Technology is separable if technology can be regarded as occurring in two distinct
stages. In the first stage different inputs are combined (via what could be considered
micro-production functions) to produce aggregate inputs30 which are then combined
(in the macro production function) to produce output. The existence of two stages is
necessary, but not, sufficient to characterise a technology as being separable. Two
types of separability can be identified by the conditions which are sufficient for their
30The aggregation through the creation of an index of differelll types of labor or capital is analogous
to the existence of micro-production functions.
35
existence. A technology can be regarded as weakly separable if, in addition to

production occurring in two logical stages, the micro-production functions are
independent of one another. A strongly separable technology is one which, in addition
to fulfilling the conditions of weak separability, has aggregate inputs which are perfect
substitutes in the production of output. In general a function of several arguments is
regarded as separable if it can be separated into sub-functions (Berndt and Christensen,
1973a, p403).
The imposition of a maintained hypothesis of separability on technology may be

regarded as untenable in many contexts. The existence of separability is, however,
crucial for justifying the aggregation of diverse heterogeneous inputs (Berndt &
Christensen 1973b, p82 and Nadiri, 1982, p447). Disaggregation of indexes of factor
inputs (such as labour into skilled and unskilled) can not be pursued in generalisations
of the Cobb-Douglas or CES functions, for example. Those functions assume strong
separability which is '... equivalent to assuming that the conditions for consistent
aggregate capital and labour indexes are satisfied' (Bemdt & Christensen, 1973b, p82).
Formally, separability can be explained in the following manner. If we let

N = {1,2, ... ,n} denote the set of n inputs, and if we assume that this set of inputs can
be partitioned into r mutually exclusive and exhaustive subsets [N1, ... ,Nr ], the
production function is said to be weakly separable if the MRTS between any two
inputs Xi and x j from any subset N s (s = 1, ... ,r) is independent of all inputs which are
not elements of that subset, i.e.:
2.41
The production function is regarded as strongly separable if the MRTS between any
two inputs from subsets N s and NI is independent of the use of all inputs which are
not elements of either subset, i.e.:
d df (X)/dXi _ • •
~ ~ /~ - 0, for aUi ENs' ) E NI' k e N s U NI 2.42
aXk df(x) aX j
Berndt and Christensen (l973a, pp406-408» show that separability restrictions on

production functions are equivalent to certain equality restrictions on the AESs.
Berndt and Christensen prove, inter alia, the following theorems:
36
• Weak separability of a production function at any point in input space is

necessary and sufficient for all proper AESs cr1, cr7k (i,j E Ns,k eNs> to be equal at
that point
• Strong separability of a production function at any point in input space is
necessary and sufficient for all proper AESs cr1, (J~ (i ENs' j E N" k e N s u N,) to be
equal at that point.
• Complete strong separability of a production functions at every point in input
space is equivalent to equality and constancy of all proper AESs.
Although the assumption of strong separability can reduce the number of parameters
needed to be estimated in applied analysis, it does impose an important cost in that it
constrains proper AESs to being constant over the entire input space (Chambers,
1988, p48). Chambers (1988, p46) shows that a completely strongly separable
production function is also homothetic in aggregate inputs. Given that both the Cobb-
Douglas and CES functions are homothetic in the aggregate inputs, both are also
completely strongly separable and hence their modeling of substitution is restrictive.
Not only is the elasticity of substitution constrained to be constant irrespective of the
level of output, a constraint which is difficult to justify technologically (Fuss et al,
1978, p240), but when extended to include more than two inputs all proper AESs are
equal (Diewert, 1971; Fuss et aI, 1978 and Jorgensen, 1986). The potential problems
associated with using a function which is a priori completely strongly separable have
already been discussed in the introduction to this chapter. Flexible functions emerging
from the application of duality theory are not a priori separable and hence allow for a
more superior investigation of the possibilities of substitution than that offered by
either the Cobb-Douglas or CES forms.
2.4 CONCLUSION
Econometric applications are often confronted with a trade-off between the generality
of competing models and the analytical tractability of those models. Production
analysis is no different. Prior to the extension of duality principles to production
problems most applications employed either the Cobb-Douglas or CES functions or
some generalisation or extension of these. Both these functions sacrifice generality for
analytic tractability. The sacrifice of generality and the implicit assumptions, in the
form of maintained hypotheses, which accompany it may have hidden costs. Indeed,
the quantitative results emerging from the inappropriate adoption of the hypotheses of
homotheticity, homogeneity and separability are potentially misleading. The main
attraction of the dual approach is that it greatly mitigates the trade-off between
37
generality and tractability, although by no means eliminating it (Chambers, 1988, p37).

Using functions which exploit the duality between cost and production allows for
testing of the hypotheses of homotheticity, homogeneity and separability and hence
potentially more valid estimates of the characteristics of technology. The use of duality
in production analysis and the specific form of the function employed here is the topic
of Chapter 3.
CHAPTER 3
THE TRANSCENDENTAL LOGARITHMIC COST FUNCTION
3.1 INTRODUCTION
Recognition of the constraints implicit in the traditional production function approach

to econometric modeling of technology (see Chapter 2) motivated the development of
a number of more general, tlexible functional forms which make use of the existence of
duality between cost and production. Such forms are capable of representing non-
homothetic production technologies where, inter alia, patterns of elasticities of
substitution are not constrained a priori. Diewert (1971) developed the Generalised
Leontiefproduction function (OL). The OL is a quadratic function in the square roots
of an arbitrary number of input prices, reducing to the Leontief (fIxed input ratios)
form as a special case (Diewert, 1971, p48l). Christensen, Jorgensen and Lau (1971,
1973) developed the Transcendental Logarithmic cost function (Translog). This
function is both linear and quadratic in the logarithms of the prices of an arbitrary
number of inputs, and the level of output (if returns to scale are not restricted to being
constant) and may be augmented to include an index of the level of technology (usually
time in different forms) (Christensen et aI, 1973, p28 and Jorgensen, 1986, p1848).
The Translog reduces to the multi-input Cobb-Douglas function as a special case. Both
the OL and the Translog can be regarded as flexible in the sense that they embody few
maintained hypotheses l . Hypotheses of homogeneity, homotheticity and separability
(for example) are not imposed on the underlying technology but rather are testable and
can be adopted if compatible with the data being employed2• Furthermore, both
functions are general in that they are easily adaptable to include not only multiple
inputs but also multiple outputs.
The development of tlexible functional forms is a direct result of the application of

duality theory to economic problems in general and the theory of production in
particular (Christensen and Oreene, 1976, p658). The basic idea of the dual approach
lDuality theory has also been employed to develop int1exible dual cost functions. Nerlove (1963), for
example, developed the dual Cobb-Douglas cost function which, because it embodies the same
maintainedbypotheses as the Cobb-Douglas production function, cannot be regarded as either t1exible
or general (Bemdt, 1991, p457).
2Not only are the most common maintained hypotheses statistically testable, when t1exible functional
fomls are used, but the validity of the underlying theory of production is also test'lble Oorgensen,
1986, pI847). How one would test maintained hypotheses is discussed in detail in Chapter 4.
39
to analysing technology is that because technology conditions the responses of

producers to market phenomena, examining these conditioned responses of producers
should provide insights into the structure of technology (Chambers, 1988, p49).
Shepherd (1953) developed the duality between cost and production functions by
showing that given certain regularity conditions3 a cost function may be used to defme
a production function (Le. there exist cost and production functions dual to each other)
(Baumol, 1972 and Diewert, 1971 & 1982). An important conclusion of duality theory
is that the structure of the technology underlying production may be analysed
empirically using either a cost or a production function. The fundamental advantage of
the dual approach is that the implications of optimizing behaviour as presented in the
classic treatise of production theory (such as Hicks' Value and Capital (1946)), which
used general functions, can be obtained without imposing a priori arbitrary constraints
on the underlying technology, such as homogeneity or homotheticty (Jorgensen, 1986,
p1843-1844).
Before turning to a presentation of -the Translog model two issues need to be

addressed, both of which are concerned with the validity of the choice of the Translog
function in the present context. The theoretical advantages of a functional form which
imposes few a priori maintained hypotheses on the underlying technology are obvious.
Whether employing a dual cost function in a particular context is appropriate is not,
however, determined merely by the perceived theoretical advantages of these forms.
The existence of the dual function is premised on the presence of specific behavioural
traits of economic agents, in this case the firm. Application of a dual cost function,
should occur only where those traits exist. The first issue to be addressed below is the
question: What behavioural traits is the firm assumed to display in order for the dual
relationship between cost and production to exist? An ancillary consideration is the
implications of the violation of that behaviour in a particular context. The first issue is
then broadly concerned with the appropriateness 4 of cost functions, as opposed to
production functions, for characterising the features of technology of subsectors of
South African manufacturing. Because a number of dual cost functions existS, the
second issue dealt with below is which of the available cost functions is most
appropriate for this study.
3Regularity conditions required to establish the duality between cost and production functions are that
the cost function be: positive for positive input prices and a positive level of output; linearly
homogenous in input prices; strictly monotonically increasing in outputs; monotonically increasing
and concave in input prices and differentiable (and therefore continuous) with respect to both input
prices and output quantities (Baumol. 1977. p366-367; Diewert. 1982. p554-555 and Jorgensen, 1986,
pU85).
4Notwithstanding tlle obvious theoretical advantages of the flexibility of dual cost functions.
sIndeed the literature has produced a competition in tile development of exotic functional fomls
(Greene, 1993. p504).
40
3.1.1 BEHAVIOURAL ASSUMPTIONS OF DUALITY AND THE ASSUMED

NATURE OF REGRESSORS
Prima facie it would appear that using a cost function approach, such as the Translog,
in applied work would be more desirable, given it's greater flexibility and generality6.
Theoretically the choice of approach should, however, be determined by the nature of
the data and is essentially a question to be decided on statistical grounds (Christensen
& Greene, 1976, p658 and Fuss et aI, 1978, p266). Indeed, while the dual cost
function imposes few maintained hypotheses on the underlying technology, the
construction of these functions does require maintained hypotheses on both market
structure and flrm behaviour, and as a consequence on the statistical nature of the data
(Fuss et al, 1978, p266). In particular, the duality between cost and production
functions is premised on the assumption of cost minimising behaviour and thus from
the perspective of the producing unit, output is regarded as fIxed (and therefore
exogenous) and competitive markets deemed to exist for all inputs, implying that input
prices are exogenous (Baumol, 1977, p364; Henderson and Quandt, 1980, p77 and
Jorgensen, 1986, p1884).
Fuss et al (1978, p266) argue that violation of ' ... one of these maintained hypotheses
may result in a model which does not have the postulated structural relationship to the
underlying technological parameters' (own emphasis). None of the literature 7 (applied
or theoretical) is, however, explicit regarding precisely when the dual relationship will
be violated. It does, however, appear from the literature that where either output or
prices (or both) are endogenous a cost function may still be estimated as long as an
appropriate systems estimator is used.
Berndt (1991, P 474) and Fuss et al (1978, p276) explicitly argue that violation of the
assumption of exogenous input prices is not a serious problem for the estimation of
cost functions dual to the production function, and may be overcome by using an
appropriate instrumental variable estimator8 . Their argument may be vindicated by the
many applications of the Translog in situations where the level of aggregation is high,
6 A further obvious feature of using a dual cost function is tlle fact that the regressors required for
empirical estimation are all economic observables (input prices and costs). Estimating a production
function directly would require data on quantities of inputs, which are likely to be unobservable in
many instances.
7See for exanlple BaumOl (1977), Bemdt (1991), Bemdt & Christensen (1973b), Bemdt & Wood
(1975), Chambers (1988), Christensen & Greene (1976), Denny & May (1978), Diewert (1982),
Jorgensen (1986), and Nadiri (1982).
8Why an instrumental variable estimator would be appropriate and which instrumental variables
ought to be used when estimating tlle Translog is discussed in detail in Chapter 4.
41
such as US. and Canadian manufacturing sectors, and prices are clearly not exogenous
(see for example Berndt & Christensen (1973b); Berndt & Wood (1975); Denny &
May (1978); and Fuss (1977)). Unfortunately, none of these applications of the
Translog deal with the question of the nature of output. In all of them constant returns
to scale are imposed a priori, and as a consequence output does not appear as a
regressor in the system9 • The applications are nevertheless premised on the existence
of a duality between cost and production functions. It is reasonable to assume that
despite the level of aggregation and the endogeneity of output the conditions for the
existence of that duality were not violated in any of those applications. An implication
of these applications, then, is that despite output being endogenous a dual cost
function may be estimated where an appropriate estimator is available lo .
The argument above that the endogeneity of either (or both) price or (and) output does
not violate the duality which exists between cost and production gains further currency
from a review of econometric texts which survey appropriate estimators of seemingly
unrelated (SUR) systems such as the Translog and attendant share equations ll . Berndt
(1991), Greene (1993) and Kmenta (1986) all discuss suitable instrumental variable
techniques for estimating SUR systems, which are necessary when independent
variables (in this case input prices and output) cannot be assumed to be exogenous.
While the duality between cost and production functions is premised on cost
minimising behaviour, it would appear from the literature that violation of the two
assumptions commonly used to depict cost minimising behaviour, exogeneity of prices
and output, does not render the assumed duality invalid. For the present analysis, the
assumption of exogeneity of output is violated. The present analysis is concerned with
estimating economies of scale in specific 2, 3 and 4 digit standard industrial
classification (SIC) industries. Following Berndt (1991, p460), it is argued that the
9Why this is tlle case is explained in section 3.4.1 and3.4.2.

100f the applications of the Translog consulted, only one is explicitly concerned with returns to scale
and hence the nature of output: Christensen and Greene (1976). While they (Christensen and Greene,
1976, p658) argue that estimating a cost function is more attractive than estimating a production
function when output is exogenous, they do not provide reasons for their arguement. nor do they
discuss the implications of attempting to estimate a cost function when output is assumed exogenous
but is in fact endogenous. They applied tlle Translog to the D.S. electric power industry, an industry
where output prices are regulated and hence output (which is a function of the output price) is
exogenous to the firm (p658-659).
llWhile the discussion up to this point has referred to cost functions alone, empirical estimation
typically involves estimating a system consisting of a cost function and associated share equations.
How share equations are derived is discussed in detail below.
42
level of analysis is sufficiently disaggregated to assume that input prices are likely to be
exogenous to the industry. Output cannot, however, be assumed exogenous 12 .
3.1.2 CHOOSING BETWEEN DIFFERENT FLEXmLE FORMS
A final issue which needs to be addressed before presenting the Translog model is the
issue of why the Translog form has been chosen over another popular flexible form,
the GL. Both forms can be regarded as second-order Taylor series approximations to
an arbitrary cost function (Berndt, 1991, p469 and Chambers, 1988, pI81). Hence, on
theoretical grounds, one is not able to distinguish between them (Berndt, Darrough &
Diewert, 1977, p661). Moreover, given that both forms have the same dependent
variables, maximize similar likelihood functions and that maximization of the likelihood
functions' is invariant to scaling of normalised prices, a priori one is unable to chose
between the two forms on econometric grounds (Berndt et aI, 1977, p662). Hence,
some other criterion needs to be used. Berndt et al (1977, p668) evaluated these forms
on the basis of how well they fitted a set of observed data and found the Translog form
to be preferred on Bayesian grounds a posteriori.
Guilkey, Lovell and Sickles (1983, p59l) argue that the type of approach adopted by
Berndt et al (1977) is not sufficiently general, and is only useful if interest is focused
on the data set used for the Bayesian testing. They argue that a better approach would
be to use a Monte Carlo technique, where one begins with a known technology and
then examines the ability of different functional forms to track that technology.
Adopting this approach Guilkey et al (1983) found that while all the forms they
considered estimate economies of scale well, their results indicated a clear preference
for the Translog form. The superiority of the Translog form is most apparent where
inputs are complements (Guilkey et al, 1983, pp599-560). Based on Guilkey et al's
(1983) results and the emphasis of this study on the magnitude of economies of scale,
the Translog was chosen over the GL (or other tlexible forms) for the present analysis.
Before turning to the Translog, an important caveat regarding flexible forms in general
needs to be addressed. While these forms have the nexibility to model sophisticated
technologies, that flexibility is only achieved at the expense of not displaying globally
good behaviour. In neo-classical production theory production functions are assumed,
inter alia, to be monotonic and either concave or quasi-concave in inputs (Lau, 1978,
p409). Functions which automatically satisfy these requirements are well-behaved.
12Given the potential endogeneity of output an instrumental variable estimator was employed here.
TIle choice of an appropriate estimator and instrumental variables is discussed in Chapter 4.
43
While the functional fonn of both the Cobb-Douglas and CES fonns ensure that these
requirements are globally satisfied, for an arbitrary set of parameters, the Translog cost
function will not necessarily ensure the satisfaction of either requirement for the dual
production function, either locally or globally (Lau, 1978, p411). In order for a cost
function to completely describe a well-behaved production function, it will have to be,
inter alia 13 , monotonically increasing and concave in input prices. Neither condition is
fufilled for the Translog cost function a priori, being detennined by the nature of the
data being analysed. Hence, there is a need to test these hypotheses over the range of
data employed.
Violation of the concavity requirement may be particularly problematic. Duality

between cost and production functions is, as argued above, premised on a maintained
hypothesis of the existence of cost minimizing behaviour. Violation of this maintained
hypothesis may undennine the postulated structural relationship between the cost
function and the underlying technology. Concavity in input prices of the cost function
is not only crucial for ensuring that the production function is well-behaved but for the
fullfillment of the hypothesis of cost minimising behaviour. Parameter estimates
derived in situations where concavity is violated should, therefore, be treated with
circumspection.
3.2 THE TRANSCENDENTAL LOGARITHMIC COST FUNCTION
The discussion of the Translog so far has tended to refer only to a cost function. Any
mention of a system of equations has been in passing. One of the broad fundamental
advantages of employing a dual function (such as a cost function) is its
computationally simple relation to the derived demand functions which impose few
arbitrary maintained hypotheses on the underlying technology14. Indeed, by employing
Shepherd's lemma (1953) and merely taking partial derivatives of any specification of
the Translog, an appropriate set of share equations, which are closely related to
derived demand equations, can be obtained. Those share equations together with the
cost function constitute a system (of seemingly unrelated (SUR)15) equations. The
fonn of the Translog cost function, the mechanics of Shepherd's lemma, the fonn of
13Further conditions of duality are: that the cost function is linearly homogenous for produceable
outputs and strictly positive input prices; is strictly monotonically increasing in outputs; and is
differentiable (and therefore continuous) with respect to both input prices and output quantities
(Baumol, 1977, pp366-367 and Diewert, 1982, pp554-555).
140btaining demand functions is usually desirable given tllat many features of technology may be
characterised by the derivatives of those functions.
15The differences between a SUR system and a conventional economic model, ,md appropriate
methods for estimating SUR systems are discussed in greater detail in Chapter four.
44
the derived cost shares, the relation of those cost shares to derived demand equations
and why the cost function together with cost shares constitutes a system of equations,
is explored below. A further issue addressed below is the different constraints which
must be imposed, due to either duality or production theory, on the system and those
which may be imposed on the system by the econometrician wishing to test different
hypotheses.
3.2.1 COMPONENTS OF THE SYSTEM
The Translog cost function can be interpreted as a second-order Taylor series

approximation to an arbitrary twice differentiable cost function which is linearly
homogenous in factor prices (Berndt, 1991, p469; Chambers, 1988, p180 and Greene,
1993, p504). The non-homothetic form of the function augmented for analysis of
returns to scale can be expressed as:
In C = lna o +a y In y+ ~ Yy/ln y)2 + Liai In Pi+ ~ LiL j Yij In Pi In Pj + LiYiY lnyln Pi

3.1
where C is total costs, y is output and Pi is the price of input i (i = 1, ... , n). Two
common variations of the Translog cost function involve the imposition of constant
returns to scale a priori; and the incorporation of regressors reflecting technical
change. Should returns to scale be assumed to be constant a priori, the third term and
the last term in equation 3.1 would be dropped and the parameter on the second term
(ay) would be constrained to unity 16. The function can be augmented to
simultaneously model scale effects and fairly general types of technical change
(Christensen (1977) (cited in Greene, 1983, p128) and Nadiri (1982)17):
3.2
16As shown in, Chapter two, economies of scale are measured by the reciprocal of tlle cost flexibility
dlne
ratio: £ = 1 --:\--. Differentiating equation 3.1 partially with respect to tlle log of output would
( oln y
yield ay + Yyy In Y + Li Yiy In Pi . If tlle conditions for constant returns to scale were imposed
(Y iy = 0'ifi = 1,... ,11; Yyy =0 and ay = 1) tllen cost flexibility ratio would equal one and hence
economies of scale would be constant.
17Nadiri's specification differs slightly from Christensen. Instead of including time in its simple form,
Nadiri uses a log specification.
45
This specification: 18 has been criticised by Greene (1983, P 128) on the grounds that it
implicitly assumes that the efficient scale grows at a constant rate over time 19. The
restriction is accepted here on the grounds that the modeling of technical change (and
changes in the efficient scale) is ancilliary to the task at hand: estimating the magnitude
of economies of scale and elasticities of substitution. Technical change was modelled
only after persistent violation of the concavity requirements for the specification given
by equation 3.1, which implicitly assumes a constant state of technology, suggested the
possibility that assuming constant technology may amount to a model misspecification.
'Patterns of producer behaviour can be described most usefully in terms of the

behaviour of the derivatives of demand and supply functions' (Jorgensen, 1986,
p1844). Following Hicks (1963) substitution effects, for example, can be specified in
terms of the response of patterns of demand to changes in input prices (Jorgensen,
1986, p1844). Possibly the most important advantage of the dual approach is that
demand and supply functions may be generated without imposing arbitrary constraints,
which characterise the traditional approach, on production patterns. A further
fundamental advantage of the use of a cost function in empirical analysis ' ... lies in its
computationally simple relation to the cost minimising input demand functions'
(McFadden, 1978, p3). This advantage is a product of Shepherd's lemma (1953)
(Diewert, 1982, p574 and Fuss et aI, 1978, p229). According to this lemma if a cost
function satisfies the regularity conditions outlined earlier and in addition is
differentiable with respect to input prices, then the derived demand for each input can
be obtained by partially differentiating the cost function with respect to the factor price
of that input (Nadiri, 1982, p467). Two important consequences flow from this lemma.
Firstly, systematic investigations of cost minimising [trIllS can be undertaken without
having to establish the corresponding production function (Chambers, 1988, p66).
Secondly, it is no longer necessary to derive input demand functions from the
production function using Lagrangian techniques (Diewert, 1982, p547).
Following Diewert (1982, p576) and expressing Shepherd's lemma in logarithmic form
(a In Cia In Pi) and applying it to either of the above specifications of the Translog (3.1
or 3.2) generates cost shares of the different factors, which are linear in the parameters
(see below). For the specification of the cost function given by equation 3.1, the cost
share functions will assume the following form:
18111is specification is modified slightly from ilie specification appearing in Greene (1988). In Greene
(1983, P127) tlle tenn Yyt t In y appears as Yye In yt . Analysis of tlle rest of that paper suggests tlmt
ilie tenn Yyt In yt should in fact have been written as eiilier Yye In y. t or Yytt In y to avoid it being
confused foryyt In (yt).
19This result is developed in section 3.4.3.
46
-aIne
- = a · +y. Iny+ I y.Inp. 3.3
aIn Pi I 'Y j I) }
For the second specification incorporating technical change, the cost share functions
for the different factors of production are given by20:
aIne
--=a.+y.
aIn Pi 'Y
Iny+ ~
L. j
I
y.Inp +y 1/ t I) }
3.4
The reason why the logarithmic version of Shepherd's lemma yields cost share
functions, as opposed to derived demand functions, and the relationship between
derived demands and cost shares is premised on the fact that the logarithmic version of
Shepherd's lenuna, being a logarithmic derivative, can be alternatively expressed as
aIne Pi ae
- -=-.- 3.5
aIn Pi e api
The cost of input i is given by Pi Xi ' and therefore the cost share is given by:
s = Pi Xi 3.6
I e
11
where C = L PiXi' Now, Xi is also the derived demand for input i, and therefore, by
i=\
Shepherd's le/nlna
3.7
Substituting 3.7 into 3.6 yields
3.8
As is the case with a number of other issues regarding flexible dual cost functions in
general and the Translog cost function in particular, the literature is annoyingly
ambiguous on the reasons why either the cost share equations or the cost share
2oShould Nadiri's (1982) specification, referred to in footnote 17 above, which uses the logarithim of
time instead of time, be used, then the last tenn of this equation would involve In t instead of t.
47
equations together with the cost function constitute a system of equations21 which
need to be estimated simultaneously. All applications consulted either implictly assume,
or explicitly state without adequate explanation, that the cost share equations, either
with or without the cost equation appended, constitute a system of equations which
need to be estimated simultaneously. This assumption is employed despite the fact that
for all applications of the Translog, the cost equation contains all the parameters that
appear in the share equations and the computational burden of estimating the cost
function (a single equation) is considerably less than that of estimating either of the
systems.
The reason why the cost share equations taken together constitute a system of
equations would, nonetheless, appear to be intuitively obvious. The features of
technology can be described using a number of different analytical devices, and
production functions, dual cost, profit or revenue functions or derived demand
functions are examples (Fuss et al. 1978, p266). Were one to use a factor-demand
approach, a comprehensive analysis of the nature of production could only be achieved
by using full sets of factor-demand equations. Single factor-demand equations would
provide information only on the features of technology relating to each factor
independently. Considering all share equations simultaneously would therefore provide
more complete information. Not only do different share equations contain different
information regarding each input individually22 but regarding all factor demand
equations as a system 23 would provide information relevant to the use of several inputs
simultaneously and would thus provide additional information on the characteristics of
technology.
The reason why in certain contexts the cost equation together with the derived demand
equations constitute a system of equations is also intuitively reasonable. The magnitude
of economies of scale is explicitly determined by the relationship between output and
costs. That information is contained in the cost function. While the set of derived
demand equations constitute a complete set of infoffilation in the case where scale is
constrained to being constant they will not constitute a complete set when economies
21Whicb system is employed in a specific context is deternlined by a priori expectations regarding

economies of scale. Which system would be appropriate in different contexts concerning scale is
discussed in section 3.4.1 and 3.4.2.
22As witnessed by the fact that in the case of the Translog different share equations contain unique
parameters: eacb share equation contains the parameterslX i and "fiY' where i refers to the ith input.
23Not a truely simultaneous system but a seemingly unrelated system (SUR) where the component
equations of the system are related by virtue of the fact that an exogenous shock to one of the
equations of the system will also affect other equations in the system.
48
of scale are variable. In that case the cost function appended to the set of derived
demand equations will constitute a complete system.
While it is intuitively clear why, when economies of scale are constant, a set of derived
demand equations; and when economies of scale are variable, the set of derived
demand equations together with the cost equation, constitute a SUR model, why
estimation of such a set as opposed to merely the cost equation (which contains the
same set of parameters) is advocated in the literature, is not obvious. Indeed, given
that all the parameters which appear in the cost share equations appear also in the cost
function, direct estimation of the cost function would appear to be desirable,
particularly from the persepective of computational cost. Berndt (1991, p470) and
Christensen and Greene (1976, p622) argue that such anapproach would, however,
not be efficient as it neglects the additional information contained in the cost share
equations. What the source of the additional information is is not, however, clear from
either Berndt (1991) or Christensen and Greene (1976) nor is it prima jacie obvious.
Intuition would suggest that the additional information emerging from the system
stems from the fact that a number of parameters appear in more than one of the
equations of the system. By obtaining estimates from the cost equation alone, other
influences on parameters would be ignored.
3.2.2 CONSTRAINTS ON THE SYSTEM.
The Translog system is a product of duality theory which imposes a number of a

priori constraints on the parameters of the Translog system. While these constraints
amount to untestable maintained hypotheses they do not compromise the flexibility
which makes the Translog an attractive form to estimate. Indeed, these maintained
hypotheses are either universally acceptable axioms or are regarded as plausible for the
problem at hand. A further set of constraints may be imposed at the discretion of the
econometrician, and as a consequence are testable. This second set consists of less
tenable hypotheses concerning the nature of technology, such as homotheticity and
homogeneity, which are often embodied in less t1exible representations of technology.
That such hypotheses can be imposed and tested when the Translog is employed
constitutes a fundamental advantage of this functional form. Not only does it allow the
econometrician to test the consistency of these hypotheses with the data being
employed but it also allows an analysis of the implications of these restrictions on the
various features of technology, such as economies of scale (Christensen & Greene,
1976, p661). Both sets of constraints are described here.
49
3.2.2.1 A PRIORI IMPOSED CONSTRAINTS
The duality between cost and production functions is premised on a cost function
fulfiling certain regularity conditions24 . Some of these regularity conditions can be
imposed on the Translog model via parameter constraints, while the remainder are
testable using parameter estimates (Jorgensen, 1986, p1889). Three parameter
constraints need to be imposed a priori: cost exhaustion (i.e. that the cost shares of
different factors sum to unity); linear homogeneity of the cost function in input prices;
and parameter symmetry. Monotonicity and concavity of the cost function are testable
using parameter estimates. Testing for these two conditions is discussed below when
AESs are discussed.
An important consequence of employing Shepherd's lemma is that cost shares of the

different factors will always sum to unity (LiSi = 1), i.e. costs are exhausted
(Jorgensen, 1986, p1890). This 'adding up condition', which is intuitively appealing,
has important implications for the econometric estimation of the model. In the case of
a three-input specification, for example, only two of the factor share equations can be
regarded as statistically independent (Diewert, 1982, p576). Estimation of the system
(that is, either of the cost functions 3.1 or 3.2 and the corresponding set of share
equations 3.3 or 3.4) requires the dropping of one of the share equations. The
requirement that one share equation needs to be dropped is important for choice of
econometric technique: the method of estimation needs to be invariant to which share
equation is dropped 25 .
The second constraint which needs to be inlposed a priori is linear homogeneity in

input prices. Thus if all input prices increase in the same proportion while output
remains constant, total costs should increase in the same proportion. Given that the
cost share equations are ftrst-order partial derivatives of the cost function the
requirement of linear homogeneity of the cost function is equivalent to the requirement
that the cost share equations be homogenous of degree zero in input prices 26 . Linear
homogeneity of the cost function is fultilled when the following relationships among
parameters is imposed on the system (Berndt, 1991, p469 and Christensen & Greene,
1976, p660):
24See footnote 3.
25Estimators which are not invariant to which share equation is deleted are open to abuse by the
econometrician seeking to support a particular hypothesis. TIle issue of invariance of estimators is
discussed in greater detail in Chapter 4.
lh
26This follows from the general result that the I1 order partial derivatives of a function homogenous
of degree p are themselves homogenous of degree p -11 (Lancaster, 1968, p335).
50
3.9
3.10
3.11
When the Translog system is estimated, these constraints are fulfIlled by normalising
the prices of the factors represented by the share equations which remain in the model
to be estimated in terms of the price of the factor whose share equation is dropped
from the system (Greene, 1993, p505).
The third set of constraints which need to be imposed a priori are the following
symmetry constraints:
3.12
The imposition of these cross-equation parameter constraints is crucial for ensuring

that the model corresponds to the underlying theory (Jorgensen, 1986, pI890). While
these theoretical restrictions are testable, the meaning of the model may be ambiguous
should these constraints be violated (Greene, 1999, p499). The imposition of cross-
equation symmetry constraints does however considerably reduce the number of
parameters, hence conserving degrees of freedom and possibly eliminating problems of
multicollinearity (Fuss et aI, 1978, p229).
The need to impose certain constraints a priori has two important implications for
estimating the Translog system and hence choice of estimator. Linear homogeneity in
input prices and cost exhaustion together mean that one should estimate the Translog
system by dropping one of the cost share equations and then normalising both total
costs and the prices of the factor inputs in the price of the variable represented by the
dropped share equation. The symmetry requirements will be fulfilled if in addition to
the above the parameter constraints implied by 3.12 are imposed.
3.2.2.2 CONSTRAINTS FOR TESTABLE HYPOTHESES
A number of additional, empirically testable, constraints may be imposed on the cost

function. These constraints, which are manifestations of a series of restrictive
hypotheses concerning the underlying technology, highlight a fundamental advantage
of employing a flexible functional form. Because different hypotheses can be imposed
by parameter constraints their validity for a particular set of data can be tested using
51
nested hypotheses. Rather than imposing a priori restrictions (as would be the case if
either a Cobb-Douglas or CES function were employed) the econometrician is able to
test the validity of different hypotheses and impose those which are appropriate. In
addition, the imposition of constraints also allows the econometrician to analyse the
implications of different maintained hypotheses for various aspects of technology, such
as economies of scale (Christensen & Greene, 1976, p66l). Hypotheses which are
testable in the case of the Translog are: homotheticity, homogeneity in output, linear
homogeneity in output, and Cobb-Douglas technology [that is, constant unitary
elasticity of substitution and returns to scale which are either constant (the function is
linearly homogenous in output) or variable (and the function is homogenous of a
degree greater than one)27].
'A cost function corresponds to a homothetic production structure if and only if the
cost function can be written as a separable function in output and factor prices'
(Greene & Christensen, 1976, p66l). Mathematically this implies that the Translog
cost function will be homothetic if, and only if (Bemdt, 1991, p470):
YIy· =0 3.13
A homothetic production structure is further restricted to being homogenous in output

if, and only if, the elasticity of cost with respect to output is constant. Mathematically,
this implies that the underlying production function will be homogenous in output if in
addition to the homotheticity constraint 3.13, the following holds 28 :
3.14
The reasoning behind this constraint is clear. The elasticity of the Translog cost
function with respect to output is:
3.15
This expression will be a constant if, and only if y.ly =0 and y y=

y'
0 y.ly =0 is the
condition for homotheticity.Hence, linear homogeneity will be imposed if in addition to
the homotheticity condition y yy =O. Testing for homogeneity will be nested within a
27Being homogenous the Cobb-Douglas function is also automatically homothetic. Indeed every
homogenous function is also homothetic (Chiang, 1984, p423 and Madden, 1986, p240).
281l1e arguements developed below supporting the homogeneity constraints are based on the Translog
cost specification given by 3.1. What the constraints would be, and why this is the case for the
Translog specification given by 3.2 are presented in footnote 29.
52
test for homotheticity (Nadiri, 1982, p467) which follows from the general result that
any homogenous function is also homothetic.
A more restrictive form of homogeneity, linear homogeneity (or constant returns to

scale), may be imposed on the Translog system by imposing, in addition to the
homotheticity and general homogeneity constraints, the constraint
a y =1 3.16
The reason why this constraint is required for the imposition of constant returns to
scale is clear. A function is linearly homogenous if the elasticity of cost with respect to
output equals one29 .
A variable returns to scale Cobb-Douglas technology (that is where the cost function is
homogenous in output and elasticities of substitution are unitary and constant) may be
imposed on the system if in addition to constraints 3.13 and 3.14 Yij = 0 is imposed
(Berndt, 1991, p470). A constant returns to scale Cobb-Douglas technology may be
imposed if in addition ay = I (Berndt, 1991, p470).
While the the flexibility of the Translog is possibly its most desirable feature, the fact
that a number of different restrictions can be imposed with relative ease enhances the
usefulness of this form considerably allowing the statistical testing of the most common
maintained hypotheses and analysis of the impact of different hypotheses on the
features of technology. How different features of. technology are modeled by the
Translog, are discussed below.
3.3 THE TRANSLOG AND SCALE, SUBSTITUTION AND TECHNICAL

CHANGE
As argued above, possibly the most attractive feature of employing cost functions such
as the Translog to analyse production, is the more sophisticated estimation of
production characteristics afforded by these models. Indeed both partial elasticities of
29Were one to employ a Translog function augmented for technical change (such as the one given by
3.2) elasticity of cost with respect to output would be:
dIn Cid In y =a y +Y yy In y+ Li Yiy In Pi +Y ylt
Homogeneity in output would be imposed on the system if, in addition to Yiy =0 and Yyy =0 being
imposed, Y•vI = O. Linear homogeneity in this case would be achieved if in addition to y.ry =0;
Yyy = 0 and ay = 1, Yyl = O.
53
substitution and economies of scale are not constrained to the same magnitude
throughout the sample. Rather they are permitted to vary with the level of output.
3.3.1 ECONOMIES OF SCALE
Hanoch (1975, pp492-493) has shown that at cost minimising points:
e(x) = Li fxi = /alnc(y,p) =l/ alnC 3.17

lex) 7 alny 7 alny
which if applied to the first specificatlon of the Translog (3.1) yields the following
formula for economies of scale:
3.18
If the Translog is augmented to be able to model technological change, then economies

of scale will become 30
3.19
Using either of these formulae clearly yield measures of economies of scale which vary
as the level of output varies. Should homogeneity (and therefore homotheticity) be
imposed, the scale would not vary with output. In this case, Yiy and Yyy =0 and scale
e(x) would be constant at 1/ ay. When the Translog is augmented to model technical
change and homogeneity is imposed, economies of scale while invariant to output will
nevertheless change over time. In that case scale will be the reciprocal of the sum of
two terms: a y and YyJ •
Calculation of scale in the general non-homogenous case which ignores technical

change 31 requires values for three parameters: ay; Yyy and Yiy· Of these, only Yiy
appears in the cost share equations (3.3). ay and Yyy appear in the cost equation (3.1).
Estimation of the magnitude of scale would, therefore, require estimation of the cost
function. While in instances where economies of scale are assumed constant (which is
usually the case where elasticity of substitution is the objective of the analysis 32 ) the
appropriate system to estimate is the system of cost share equations, where the
30Using the log of time as an index of t.echnology, as suggested by Nadiri (1982), would obviously
alter the lasttenn of this fonnula to In t .
31111e argument developed here is easily extended to the case where technical change is included.
32See for example Bemdt & Wood (1975) and Bemdt & Christensen (1973b).
54
objective of the analysis is the magnitude of scale the appropriate system should
include the cost function as well (Berndt, 1991, p476).
3.3.2 ELASTICITIES OF SUBSTITUTION AND CONCAVITY TESTING
In the n-factor case only the concept of partial elasticities can be defined, for which no
single definition exists. Elasticity of substitution for the two input case is defined
subject to the constraint that the level of output remains constant. Generalistion of the
elasticity concept to the n-input case requires in addition the imposition of further
constraints. Depending on the constraints imposed a number of different elasticities
exist (Griliches and Ringstad, 1971, p6). Computationally the obvious choice here is
the AES. Not only is it relatively simple to obtain estimates directly from the cost share
estimates and fitted values, but the AES is also involved directly in testing of local
'good behaviour' and hence their calculation is imperative. Uzawa (1962) has shown
that own and proper AESs can be obtained using the following formulae:
,...~
v"
= Yii +S?2 -Sj for i =1, ... ,n 3.20
S·I
A Yij +SjSj
cr··
IJ
= S.S.
for i,J" =1, ... ,n and i *" J' 3.21
I }
where Sj is the cost share of the i th factor. These formula highlight an important
advantage of estimating a flexible functional form such as the Translog: elasticities of
substitution (like economies of scale) are not constant but rather vary with output. This
is because they depend directly on the fitted values of cost shares, which in turn are a
function of the level of output.
Where estimates of AESs are the only feature of technology of interest and economies
of scale are assumed to be constant33 , all the information required for the compuation
of AES (Yii; Yij; Si and S) can be obtained form the share equations (the Yij
parameter appears in both the cost function and the share equations). Hence, if one
were interested only in elasticities of substitution, a system of share equations is all that
one need to estimate (see for example Berndt & Wood (1975) and Cluver (1981)).
33This is likely to be the case in two instances: when elasticity of substitution is of direct concern (see
for example Berndt & Wood (1975» or when the issue of separability of inputs and the question of
consistent aggregation of inputs is of interest (see for example Bemdt & Christensen (1973b). Bemdt
and Christensen (1973a, pp403-409) show that separability (and hence the question of consistent
aggregation) is closely related to equality constraints on AESs (Alien. 1938, p504).
55
Apart from providing measures of the degree of substitutability of inputs AESs have
two other important uses. Firstly, a close relationship between AES and price
elasticities exist which means that a slight modification of the own and proper AES
formulae will yield formulae for own and proper price elasticities. The second
important use of AESs is in testing whether the regularity conditions of monotonicity
and concavity of the cost function are fulfilled for the data set being used.
AESs are related to price elasticities and cost shares in the following manner (AlIen,
1938, p508):
cr~ = 11 I)
3.22
') Pix;/C
where llij is the partial elasticity of factor j with respect to the price of factor i and
the denominator is the share in costs of the ith factor Sj' Rearranging to make llij the
subject of the formula, yiedls:
n .. -_
"I)
cr IJAS
...1 3.23
Using this result (3.23) and equations 3.20 and 3.21, one is able to calculate partial
price e1asticities directly from estimators generated from the Translog function and
associated cost shares:
'11 .. + S.S.
llij= /') ') ,orlJ=
f .. l,... b ' .
,n, utl~J 3.24
Sj
_ Yjj + Sj2 - Sj . ._ I
lljj - , tor l - , ... .)1 3.25
S. 1
The second ancillary use of AESs is the most important. The derivation of the
Trans10g cost function (and other general flexible cost functions) is based on a
mathematical theorem concerning the duaP4 of a concave program (Madden, 1986,
p265). As argued above (Section 3.3.1.1) Diewert (1982, pp554-555) and Baumol
(1977, pp366-367) show that, according to this theorem, in order for a cost function
34The tenn duality is often erroneously interchanged with Shepherd's lemma (Shepherd 1953), which
is used to derive input demand equations from the cost function. The lemma is actually a corrollary to
tlle tlleorem upon which dual cost functions are based (Diewert, 1982, p555 and Madden, 1986,
p265).
56
to be a sufficient statistic for a production function 35 , the cost function will have to be,
inter alia36 , monotonically increasing and concave 37 in input prices. Violation of the
concavity requirement is particularly problematic. Not only does it amount to a
violation of one of the basic behavioural postulates of neo-classical production theory:
that the entrepreneur is a cost minimiser, but, more importantly, the cost function may
not have the postulated structural relationship to the underlying technological
parameters (Fuss et al, 1978, p266 and Lau, 1978, p411). The parameter estimates
derived in situations where concavity is violated should, therefore, be treated with
considerable circumspection.
Unfortunately, for an arbitrary set of parameters the Translog cost function will not
necessarily satisfy the monotonicity or concavity requirements, either locally or
globally (Lau, 1978, p411). There are two simple tests for establishing the
monotonicity and concavity requirements: the Translog cost function is monotonically
increasing in input prices if the fitted factor shares are all positive and it is strictly
quasi-concave if the nXn matrix of AESs is negative semi-definite 38 at each
observation (Berndt, 1991, p477).
The only previous application of the Translog cost function found in the South African
literature, Cluver (1981) (and Cluver and Contogiannis (1984) which is based on
Cluver (1981», would appear to violate the requirements that the cost function be
monotonically increasing and concave (or quasi-concave) in input prices. No explicit
testing of these two requirements appears. While Cluver (1981, p104) does present
AESs for each of the sectors he analyses, he presents only a single AES39 for each
sector. Whether the AESs presented are for a particular point in time or are a mean of
those for different years is unclear. Given these AESs one is unable to conclude
whether the concavity requirements have been fulfilled or not. As a consequence, one
35That is to completely describe a production function which is consistent with the 'minimum'
assumptions required for neo-classical production theory, i.e. that the function be real-valued,
continuous, increasing and quasi-concave (McFadden, 1966, cited in Diewert, 1982, p553-554).
36Further conditions of duality are: that the cost function is linearly homogenous for produceable
outputs and strictly positive input prices; is strictly monotonically increasing in outputs; and is
differentiable (and therefore continuous) with respect to both input prices and output quantities
(Baumol, 1977, pp366-367 and Diewert, 1982, pp554-555).
37Bemdt (1991, p477) argues that theory requires that the cost function merely be quasi concave in
input prices.
38A matrix [B] is negative semi-definite if IBll:::; IB"I:2:
O~B21:2: O,... jB"I:::; 0 if n is odd and 0 if
n is even.The matrix will be negative definite if the weak inequalitites are replaced by stomg
inequalities (Chiang, 1984, p394). The precise form of the test of negative semi-defmiteness of tbe
matrix of AESs for the three input case is given in Appendix 3.
39 AESs, as argued above, vary with the level of output. For time series applications a different AES
for each observation in the sample should be generated.
57
is unable to determine whether the conditions required for the application of the duality
theorem, upon which the validity of the Translog as a sufficient statistic for modelling
the dual production function is based, are fulfilled. Given that the violation of
concavity requirements may lead to the collapse of the theoretical relationship between
the dual cost function and the underlying technological parameters, Cluver's (1981)
results should be treated with due care.
Although the results presented by Cluver (1981) allow for calculation of fitted shares
and thus all own and proper AESs, which would enable testing the monotonicity and
concavity requirements, this has not been undertaken here. That the issue of concavity
is not addressed in Cluver's (1981) application is nevertheless a serious flaw in the
thesis. A further potential flaw in that thesis, the possible lack of invariance of the
estimator employed, is addressed in Chapter 4.
3.3.3 MODELLING TECHNICAL CHANGE
The specification of the Translog given by equation 3.1 implicitly assumes that the
state of technology is constant over the period under analysis. Persistent violation of
the concavity requirements for estimators of that specification suggested the possibility·
that the assumption of constant technology may amount to a model misspecification.
Rather than being non-concave, it is possible that the violation of the concavity
requirement is due to the cost function shifting over time. To test this hypothesis an
alternative form of the Translog, which does not constrain technical change to being
constant, was also estimated. This alternative specification (3.2) which was developed
by Christensen (1977) includes time (as a proxy for technology) in various forms.
Christensen's (1977) specification of the Translog is quite general in the types of
technical change it can accomodate. Indeed, simple parameter restrictions can be
employed to impose a number of special cases of technical change on the underlying
production structure (Greene, 1983, pl27).
Because the modelling of technical change is ancillary to the main purpose of this
dissertation, serving to assist the modelling of scale and substitution effects rather than
being an area of interest in its own right, alternative approaches to modelling technical
change and the impact of different maintained hypotheses on the modelling of technical
change were not discussed in Chapter 2. 'Technical change', although referring to the
general effects that technological advances have on the production process, is
measured in a plethora of different ways. Before interpreting the additional terms
appended to the original Translog in the specification given by equation 3.2, a brief
58
synopsis of some of the relevant alternative measures of technical change is provided

below.
3.3.3.1 MEASURING TECHNICAL CHANGE
The theoretical analysis of technical change tends to focus on two broad areas: its
nature and source (Fuss et al, 1978, p221). Attempts to measure technical change
employing econometric techniques tend to focus primarily on the nature of technical
change. Analysis of the source of technical change would appear to be largely limited
to the theoretical literature. A common feature of the analysis of technical change is
the extensive use of taxonomies to describe different aspects of the two focus areas:
source and nature. Regarding the source of technical change, analysis is concerned
with two broad areas: whether technical advances emerge from within or outside of the
fmn; and whether technical advance involves changes in the nature of inputs used in
the process of production and hence whether the source of technical advance is new
inputs. Analysis of the nature of technical change is concerned with measuring whether
it may be deemed progressive or regressive, which aspect of technology may have been
altered due to technical advances, and whether or not there is bias.
3.3.3.1.1 SOURCES OF TECHNICAL CHANGE
Regarding the sources of technical change, a broad distinction may be drawn between
endogenous and exogenous technical change. Technical change is endogenous if it
originates within the fIrm, for example as a result of leaming-by-doing or innovation
(Nadiri, 1982, p445). Of ancillary interest, in this regard, is what motivates
endogenous technical change. Hicks (1963), for example, has proposed an induced-
invention hypothesis: that technical change is a response to market phenomena such as
relative price changes (Chambers, 1988, p204).
Closely related to the endogeneity-exogeneity dichotomy is the embodiment hypothesis

which in a sense straddles the two focus areas of the theoretical literature: source and
nature. Technical change may be embodied in a factor of production (usually capital40
but possibly other factors, most likely skilled labour). When technical change is not
embodied in any particular input or group of inputs, it is referred to as disembodied.
Where technical change is embodied the basic form of the production function will
change over time.
40 Aclassical example of embodied technical change is Eli Witney's cotton gin. The technical
innovation was embodied in the gin in the sense that the gin had to be acquired to have access to the
new technology (Chambers. 1988. p205).
59
3.3.3.1.2.THE NA TURE OF TECHNICAL CHANGE
Technical change is deemed progressive if it expands the input requirement set, that is,
if it allows input bundles formerly incapable of producing a certain level of output to
produce that level of output. Graphically, progressive technical change would be
depicted by an inward shifting of isoquants. Technical change is regressive if it reduces
the input requirement set by eliminating bundles capable of producing a given level of
output. Regressive technical change would be depicted graphically by an outward shift
of isoquants. Such change may not be intuitively appealing. A more restricted version,
locally regressive technical change, is however more appealing. Technical change is
locally regressive if the new technology involves very intensive committal of certain
inputs but not others. Graphically this would involve a rotation of isoquants toward the
axis measuring the input being used more extensively as a consequence of technical
advance, rather than a shift of isoquants (Chambers, 1988, p206-207).
Interest in which aspect of technology has been altered by technical change usually
results in considering whether technical change is factor -, scale - or substitution -
augmenting (Fuss et al, 1978, p221). Technical change is deemed factor-augmenting or
input-augmenting if it improves input efficiency and therefore the effective quality of
inputs (Chambers, 1988, p210 and Fuss et aI, 1978, p22l). Scale augmenting change
expands the level where decreasing returns set in, while substitution augmenting
change improves the substitutability of inputs.
The tInal taxonomy which exists regarding the nature of technical change is that
associated with its bias. That technical advance may lead to the displacement of
resources is widely conceded (Chambers, 1988, p203). When new production
techniques are employed they will either have a neutral effect on the production
process or alter the input-output relationship and thus be regarded as biased (Nadiri,
1982, p444). Identifying the nature of technical change is often regarded as being
synonymous with characterising its bias. Indeed, technical progress is often defined
according to its input bias. In this regard a number of different defmitions of technical
change exist, the most familiar being the Hicks, Harrod and Solow characterisations of
technical progress (Nadiri, 1982, p444). All three are concerned with how the marginal
rate of technical substitution among inputs changes over time. The differences emerge
from which aspect of technology is held constant while changes in the marginal rate of
technical substitution over time are analysed. In the Hicksian definition, the capital-
labour ratio is held constant and change in the marginal rate of technical susbstitution
60
over time is used as a measure of technical bias (Chambers, 1988, p20?). Formally, for
the two factor case Hicksian technical change is defined as:
labour - saving
aCfK/fL)1 > 0, Hicks neutral
at K/LcolUtUt
< {
capital- saving
The Harrod definition of technical change holds the capital-output ratio constant and
can be formally expressed as (Nadiri, 1982, p444) :
labour - saving
aCfK/fL)1 > 0, Harroa neutral
at K/Y
cOll1tanl
< {
capital- saving
The Solow definition holds the labour-output ratio constant and can be fOffi1ally
expressed as (Nadiri, 1982, p444) :
labour - saving
aCfK/fL)1
at L/Y
> °
< '
Soldw
{
neutral
constUll capital- saving
Given the above lexicon of characterisations of technical change it is clear that the
empirical analysis of technical change will in all likelihood be constrained in the number
of effects which can be modelled by the need to adopt a specific functional form. The
most common approach to extending an analysis to include the effects of technical
change is to append a time term to the function being used to model the characteristics
of technology41. While the simple inclusion of a time term would appear to be
relatively innocuous it entails the adoption of a number of limiting assumptions. Not
only does it ignore the source or motivation for technical change but it also involves
the tacit assumption that technical change is disembodied, that is that technical change
is not embodied in new inputs. The approach which is adopted here is more
sophisticated than merely appending a sinlply time term to the function. The precise
implications of the terms appended to the Translog cost function are, however, not
clear from the literature.
41111e obvious motivation being iliat technical advances usually require ilie passage of time
(Chambers, 1988, p204).
61
3.3.3.2 THE TECHNOLOGY AUGMENTED TRANSLOG
Following Christensen (1977) technical change has been modelled here by appending
to the cost function (3.1.) a simple time variable, a time-squared variable, the product
of time and the log of output and the sum of the product of the log of the prices of the
different factors and time: [a t·
t 2 ,. y yt tiny and '"
t,
.1 y
2 yt L.Ji y ti.tlnp.] Because the cost-
I
share equations are obtained by employing Shepherd's lemma and differentiating the
Translog cost function with respect to the log of the price of the different factor inputs,
the cost share equations will also be modified somewhat. In particular, each of the cost
share equations will have an additional term, time: Yi/ . By modifying the Translog
system in this manner, a number of different types of technical change can be
accommodated. Moreover a number of special cases of technical change can be
imposed on the system via simple parameter restrictions (Greene, 1983, p127).
Unfortunately, which types of technical change can be modelled by the modified
Translog is not clear from the literature. The model was developed in an unpublished
paper (Christensen, 1977) and while it is presented in other sources (Berndt, 1991;
Berndt and Wood, 1982 and Greene, 1983 ) the precise types of technical change
which can be modelled, when scale is not constrained to being constant, are not
clear4 2 • The literature does, nevertheless highlight three aspects of technical change
which can be analysed using the model: whether or not technical change is regressive
or progressive; the factor bias of technical change and the impact of technical change
on efficient scale. Furthermore, it is apparent from the literature that when using the
Translog cost function, one is unable to test for the possible endogeneity of technical
change such as that suggested by Hicks's induced-innovation hypothesis.
For ease of exposition these different features the specification of the Translog which
incorporates technical change and the associated cost share equations are repeated
here as equations 3.24 and 3.25, respectively:
3.26
42Bemdt & Wood (1982, p206) do include a fairly detailed analysis on the types of technical chnage
which can be modelled if economies of scale are constrained to being constant. TIlat analysis is not
particularly useful here given that an express objective of this paper is estimation of economies of
scale. Nevertheless wher scale is constrained to being constant the specification of the Translog
provided by 3.2 is able to model, inter alia, Hicks, Harrod, Solow and Leontief neutral technical
change.
62
aInC
--=a+y. Iny+ I y Inp +y t 3.27
a In Pi 'Y I j IJ J It
Including time as a variable in any cost function 43 allows the easy categorisation of
technical change as either progressive or regressive. If technical change is progressive
c( Pi' y, t) is nonincreasing in t; if technical change is regressive, c( Pi, y, t) is
nondecreasing in t. This result rests on the fact that progressive technical change
expands the input requirement set. The mimimum cost of producing a given level of
output using the expanded input requirement set can be no larger than the minimum
cost using the original input requirement set since the original cost-minimizing bundle
remains feasible. A broad classit1cation of technical change as progressive or
regressive in any particular context can be achievedby analysing the rate of cost
dimunition (Chambers, 1988, p 214):
A= am c(Pj, y,t) 3.28

at
If A::;; 0 technical change is progressive. Technical change is regressive if A;::: O.
Applying 3.28 and differentiating 3.26 with respect to time yields the rate of cost
dimunition for the Translog cost function as:
3.29
Greene (1983, P126) defines factor bias bj as:
. .
l-savmg
{ i.- ne~tral
l-usmg
3.30
From 3.27 it is clear that aSJat =Y jt • Hence the qualitative bias of technical change
can be determined directly from the sign of the coefficient on the time variable in each
share equation. Technical change is therefore input i-using when Yjt > 0, input i-saving
when Y it < 0 and input neutral when Yit =0. Berndt and Wood (1982, p203 -204)
argue that because the bias coefficients Yjt are constants, they do not vary over the
sample and remained fixed as relative input prices vary. As a consequence, they cannot
43Including time in a cost function will modify the general expression for a cost function from
c( y, p) to c( y , Pi ,t) .
63
be used to assess Hick's induced innovation hypothesis: that technical change is a

response to market phenomena such as relative price changes.
Berndt and Wood (1982, p204) raise a second caveat regarding bias coefficients. Bias
parameters, they argue, represent relative rather than absolute changes in factor
demands in response to technical change. This has important implications for the
interpretation of the bias coefficients. Yit > 0, which implies that technical change is
factor i-using, does not necessarily imply that technical change has increased the
amount of factor i which is demanded in production. Factor i-using technical change
could mean that while the demand for all factors was reduced through technical
progress the demand for factor i was not reduced by as much as the demand for other
factors.
The incorporation of technical change into the Translog in the manner described by
equation 3.27 impacts on the how scale is modelled in the system. As was indicated by
equations 3.18 and 3.19 above, when technical change is modelled the formula for
economies of scale employed in the case when technical change is assumed constant
will be altered and have a the temi Yytt added to the denominator. That is, economies
of scale will be measured by:
3.31
Following Greene (1983, p127) and solving this expression for the efficient scale, that
is the level of output where average costs reaches its minimum and e(x) = 1, yields:
~ y. in p. - Y t
in y* = 1- a. y
- Yyy in y - L..i ry J yt
3.32
Yyy
where y* denotes the level of output where efficient scale is achieved. Partially
differentiating this expression with respect to time:
3.33
An implication of this results is that the level of efficient scale changes over time by the
constant44 -Yyt!Yyy ' Greene (1983, p128) argues that it seems overly restrictive to
assume that efficient scale grows at a constant rate over time.
44_y yt!Y yy is a constant as it is a ratio of two parameters which are fixed throughout the sample.
CHAPTER 4
DATA, ESTIMATION AND TESTING PROCEDURES
4.1 INTRODUCTION
The Translog cost function when combined with the derived cost-share equations, as
argued in Chapter 3, constitutes what can be regarded as a seemingly unrelated
regression (SUR) model (Greene, 1990, p528). A model (or system) is deemed
seemingly unrelated if the component equations comprising the system are connected,
not because of direct interaction between equations, but rather as a result of their
disturbances being correlated (Kennedy, 1985, p137; Pindyck & Rubinfeld, 1991,
p308). While prima facie the component equations of the Translog system appear
unrelated in that the variables on the left hand side do not appear anywhere in the
system as regressors the reasoning why the equations comprise a 'simultaneous system'
is far more subtle, yet intuitively clear. An exogenous shock to the cost of capital, for
example, may impact on the demand for other factors and hence affect their cost shares
and as a result also total costs. SUR models are in this sense unique - while not 'truly'
simultaneous - the component equations are nevertheless related via their disturbances.
Because of the unique structure of a SUR model, the conventional methods for
estimating both a system and single equations will generally not be appropriate. A
primary concern of this chapter is establishing the appropriate method for estimating
SUR systems under a number of different conditions: when regressors are endogenous
rather than exogenous and when the disturbances of the component equations are
correlated. Two other broad areas, both of which pertain directly to estimating the
parameters of a Translog system, are addressed here: econometric testing procedures
and data treatment. Testing for the fulfillment of the duality conditions has already
been addressed in Chapter 31• Consideration is here to the appropriate methods of
testing in three other areas: determining which of a number of alternative estimators is
appropriate in a given context; whether the various restrictions (such as homogeneity
and homotheticity) which the researcher is able to impose on the underlying
technology are valid, and how good the overall fit of the model is. The fmal concern of
this chapter is the data which will be employed in the estimation of the model.
IDuality is valid only if the cost function Cell be regarded as concave in input prices. Testing for
concavity is discussed in Section 3.4.2 and Appendix 3.2.
65
4.2 METHOD OF ESTIMATION
A model is deemed a SUR model if the equations of the model are related through non
zero covariances of the errors across different equations at a given point in time
(Pindyck & Rubinfeld, 1991, p326). Should the assumptions of the classical linear
regression model be fulfilled for each of the component equations of the SUR model,
application of OLS, on an equation by equation basis, while yielding unbiased and
consistent parameter estimates will not, however, provide efficient estimates (Greene,
1993, p488; Kennedy, 1985, p137 and Kmenta, 1986, p637). Estimating each equation
seperately and independently would disregard the information contained in the cross-
equation correlation of disturbances. An improvement in efficiency will be gained by
explicitly taking into account the fact that cross-equation error correlations may not be
zero (Kennedy, 1985, p137, Kmenta, 1986, p637 and Zellner, 1962, p353).
The use of generalised least squares (GLS), which explicitly takes into account the fact
that cross-equation error correlations may not be zero, to estimate a SUR model will
bring gains in efficiency unless the equations comprising the system have identical
explanatory variables in which case GLS and OLS will be identica? (Greene, 1993,
p488 and Pindyck & Rubinfeld, 1991, p3l0). Gains ii1 efficiency from the use of the
GLS estimator will be positively related to the degree of correlation of the disturbances
among equations, and negatively to the correlation of explanatory variables (Kmenta,
1986, p642). The application of GLS is based on knowledge of the elements of the
variance-covariance matrix of disturbances. A lack of knowledge of this variance-
covariance matrix can be overcome by a consistent estimate of it. The application of
GLS using a consistent estimator of the matrix (estimated or feasible GLS (FGLS»
will yield parameter estimators which display the same asymptotic properties as the
GLS estimator (Kmenta, 1986, p643). Four alternative FGLS estimators are available
for estimation of the Translog modeL Zellner (1962, p348-368) has developed a two-
stage estimated generalised least squares (FGLS) method for estimating a SUR, often
referred to in the literature as the Zellner Efficient Estimator (ZEF). The second
estimator is Zellner and Theil's (1962, p54-78) three-stages least squares (3SLS), a
systems estimator in the general sense, which is the appropriate estimator of the
Translog model when at least one of the regressors is endogenous. Two further
estimators can be obtained by adjusting the ZEF and 3SLS estimators to take
autocorrelation into account - the ZEF(AR) and 3SLS(AR) estimators. Which of the
7
-A second, obvious situation where OLS and GLS will be identical is where cross-equation covariance
of disturbances is zero.
66
four is appropriate can, theoretically, be determined by testing for the presence of both
endogeneity and autocorrelation - issues dealt with in sections 4.2.4 and 4.2.5.
4.2.1 THE ZELLNER EFFICIENT ESTIMATOR (ZEF)
Being an FGLS estimator, the ZEF estimator emerges from a two-stage procedure.
The ftrst stage of the ZEF procedure involves estimating each of the equations of the
system by OLS. An explicit assumption of the t1rst stage is that the disturbances of
3
each equation are homoscedastic and uncorrelated with each other (Zellner, 1962,
p350). The residuals from this process provide a consistent estimate of the variance-
covariance matrix. The second stage of the ZEF involves applying Aitken's GLS
4
procedure to the system of equations expressed as one large equation and thus
generating an FGLS estimator which is BLUE and asymptotically equivalent to the
GLS estimator (Zellner, 1962, pp350 - 351). For small samples the ZEF is unbiased
and efficient relative to the OLS estimator (Kmenta, 1986, p644).
Prior to the application of any estimator to a model or system the stochastic structure
of the model needs to be specifted. Following Christensen and Greene (1976, p662)
and Jorgensen (1986, pI893), additive disturbances for each of the cost share
equations and the cost functions, which are homoscedastic and independently and
normally distributed, are assumed 6 . This speciftcation has important implications for
the Translog system. Because the cost share equations always sum to unity, where
there are n share equations, only n -1 will be linearly independent. An implication of
this is that for each observation, the sum of disturbances across the cost share
equations will always equal zero (Berndt, 1991, p472). This means that the OLS
estimated disturbance variance-covariance matrix which is generated in the ftrst stage
3Yiolation of this assumption will lead to inefflcielll estimates (Greene, 1993,1'498). The issues of
testing for autocorrelation and how the problem of aULOcorrelation may be overcome are discussed
below in section 4.2.5.
41f the system consisted of N equations of the foml: YI = X.a
I I
+ £ I where the subscript i refers to the
itb equation, writing the syste~ as qne_large equation involves expressing the system as:
1'; Xl 0 0 al El
1; 0 X2 a2 E2
+
Y" 0 0 Xn an En
As the cost share equations are obtained by differentiating the cost function, the disturbance term of
e cost functiOIl is not included in the share equations.
IIn addition, ~o~lowing Jorg.ens.en ("986, pI892) it is assumed that tile disturbance vector for the
ystem has aJomt normal dlstnbul.1on, With mean vector zero and a constant covariance matrix
67
of the ZEF procedure will be singular, rendering this procedure inoperational

(Christensen & Greene, 1976, p663). A possible solution to this problem would be to
drop one of the share equations and to apply the ZEF procedure to the remaining cost
function and share equations. Parameter estimates for the deleted share equation could
be found by employing the constraints imposed by the requirement that the cost
7
function be homogenous of degree one in input prices • Application of this procedure
is, however, only valid if the estimator used is invariant to which share equation is
dropped8 (Berndt, 1991, p473). The ZEF will be invariant to which share equation is
dropped only if no cross equation symmetry constraints are imposed (Berndt, 1991,
p474). Because the imposition of these cross-equation parameter constraints is crucial
for ensuring that the model corresponds to the underlying theory9 (Jorgensen, 1986,
pI890), the ZEF is obviously not appropriate.
Barten (1969) has shown that maximum likelihood estimates of a system of share
equations which has one equation deleted are invariant to which share equation is
dropped (Christensen & Bemdt, 1973b, p89 and Christensen & Greene, 1976, p663).
Christensen and Greene (1976, p663) argue that Barten's (1969) conclusion can be
extended to a system which includes the cost function. Maximum likelihood estimates
can be obtained by iterating the Zellner method until convergence (Greene, 1993,
p493). Iterative ZEF estimates (lZEF) will thus be invariant to which share equation
will be dropped, and is consequently an appropriate estimator for applications of the
Translog.
The application of the IZEF is premised on the assumption that all the component
equations of the system satisfy the assumptions of the classical linear regression model.
The presence of endogenous regressors in any of the component equations would
violate the assumption that regressors and the errors are uncorrelated lO • (Gujarati,
1988, p556). Violation of the assumption causes the least squares estimator to be
biased ll , even asymptotically (Kennedy, 1985, pI26). Hence, applying OLS to the
7 Specifically that the following equalities hold: Li (li = 1; Li Yij = 0 and Li Yiy =O.
SMethods which are not invariant to the share equation which is dropped are obviously open to
manipulation by the econometrician who may by dropping different equations be able to support a
priori assertions.
9Wllile cross-equation parameter constraints are testable, the meaning of the model may be ambiguous
should these constraints be violated (Greene, 1999, p499)
lOAn explicit assumption of the OLS method is that the explanatory variables are either non
stochastic, or if stochastic, are distributed independently of the stochastic disturbance term (Gujarati,
1988, p556).
lIThe source of tlle bias is intuitively obvious. An exogenous shock, manifested in a sudden change in
the error term will influence the dependent variable both directly and indirectly via its influence on
68
Translog model where the regressors are endogenous would yield estimates which
were not only inefficient but also inconsistent. While the use of the IZEF in this
context would improve efficiency it would not improve consistency. An estimator
which explicitly deals with both endogeneity of regressors and correlation between the
errors of the different equations of the model would need to be employed. 3SLS is an
example of such an estimator.
4.2.2 THREE STAGES LEAST SQUARES
The intuitively obvious approach to the resolution of the bias associated with using
endogenous regressors would be to remove the source of that bias: the correlation
between regressors and disturbances (Gujarati, 1988, p524). Replacing endogenous
regressors with instrumental variables (IVs), which are highly correlated with the
regressor being replaced but uncorrelated with the errors would achieve this end.
Obvious candidates as IVs are exogenous variables appearing in the model. The
problem of determining which of various exogenous variables would be the best IV in
a particular context prompted the development of the two stages least squares (2SLS)
approach, a single equation estimator which employs IVs which are combinations of all
exogenous variables (Kennedy, 1985, p134). In the first stage of the 2SLS procedure, .
each endogenous regressor is regressed on all the exogenous variables in the system
and the fitted values retained. In the second stage, the fitted values of the endogenous
regressors are employed in place of the original endogenous regressors and OLS
estimators are obtained in the usual manner (Gujarati, 1988, p608 and Kennedy, 1985,
p134). Being an IV estimator which purges the correlation between regressors and
errors, the 2SLS estimator will be consistent (Kennedy, 1985, pI34).
While in the context of the Translog model with endogenous regressors the 2SLS
estimator yields unbiased and consistent parameter estimates it will not provide
efficient estimates. The lack of efficiency of the 2SLS estimator stems from the fact
that using 2SLS amounts to estimating each equation seperately and independently and
thus information about the correlation of disturbances between equations would be
disregarded An improvement in efficiency will be gained by explicitly taking into
account the fact that cross-equation error correlations may not be zero, and estimating
all equations in the system simultaneously - in other words by employing a GLS
estimator (in combination with an instrumental variable estimator) (Kmenta, 1986,
tbe endogenous regressor witb which it is correlated. The OLS technique attributes botb influences on
the dependent variable to tbe regressor, ratber tban only the indirect effect. Hence tbe OLS estimator
will biased, even asymptotically (Kennedy, 1985, p127).
69
p695). The properties of the 2SLS estimator, and how these properties can be
improved (by employing GLS) in the context of endogenous regressors, mirrors the
properties of the OLS estimator, and how it's properties could be improved (by
employing a GLS estimator - ZEF) when the regressors of the Translog model are
exogenous.
Although the theoretical application of GLS requires knowledge of the elements of the
variance-covariance matrix of errors, absence of this information can be overcome by a
consistent estimate of the variance-covariance matrix. The application of GLS using a
consistent estimate of that matrix (FGLS) will yield parameter estimators which display
the same asymptotic properties as the GLS estimator (Kmenta, 1986, p643). Three
stages least squares (3SLS) is an example of a FGLS estimator which simultaneously
accounts for endogeneity of regressors and inter-equation disturbance correlation, and
where regressors of the Translog are endogenous would yield estimates which are
consistent and asymptotically efficient (Kennedy, 1985, p136).
The 3SLS estimator involves first applying the 2SLS procedure to all the equations in
the system from which the residuals are retrieved and used to derive a consistent
estimate of the covariance-variance matrix of errors of the system. The errors of each
of the structural equations estimated using 2SLS are assumed to be both
homoscedastic and independently distributed 12 (Zellner and Theil, 1962, p55). The
consistent estimator of the variance-covariance matrix is then employed in the second
stage where the Aitken GLS procedure is applied to a single large equation (the same
equation as that employed by the ZEF estimator) which includes all the component
equations of the system. Kmenta (1986, p697) argues that the 3SLS approach takes
the correlation between disturbances of the different equations into account by treating
the system of equations as a SUR system. The appropriateness of this estimator for
Translog systems where regressors are endogenous is obvious, improving both
consistency and efficiency. Improved efficiency derives from the explicit consideration
of cross equation error correlations, improved consistency from the fact that the 3SLS
estimator is an instrumental variable estimator (Greene, 1993, p612).
The 3SLS method can not, unfortunately, be applied to the Translog system which
includes an additive disturbance term with each of the equations. Because the cost
share equations always sum to unity should there be n share equations in the system,
only 11 -1 will be linearly independent. An implication of this is that the 2SLS
12Yiolation of this assumption would lead to inefficient estimaLOrs (Kmenta, 1986, p706)
70
estimated disturbance variance-covariance matrix will be singular, rendering 3SLS

inoperational. A possible solution to this problem, which also emerges if one attempts
to apply the ZEF procedure to the entire system, would be to drop one of the share
equations and to apply the 3SLS procedure to the remaining cost function and share
equations. Parameter estimates for the deleted share equation could be found by
employing the constraints imposed by the requirement that the cost function be
l3
homogenous of degree one in input prices .
As has been argued above, deleting a share equation is, however, only valid if the
estimator used is invariant to which share equation is dropped. Iterating the 3SLS
(I3SLS) procedure 14, does not, unlike iteration of the ZEF estimator, provide the
maximum likelihood estimator, nor does it improve efficiency (Greene, 1993, p6l2 and
Kennedy, 1985, p140). Despite not being the maximum likelihood estimator I3SLS
estimates are nevertheless invariant to which of the share equations are deleted (Berndt
& Wood, 1975, p261 and Berndt, 1991, p474). The I3SLS estimator is therefore an
appropriate estimator of the Translog model.
I5
4.2.3 INVARIANCE OF ESTIMATORS: CLUVER (1981) REVISITED
That an estimator which is invariant to which share equation is dropped needs to be

chosen highlights a further potential flaw in Cluver's (1981) application of the Translog
system 16. That application used 3SLS (Zellner & Theil, (1962) (Cluver, 1981, p103)).
As argued above, 3SLS would usually only be employed where regressors were
endogenous: a situation which is more likely to occur when highly aggregated data is
used 17 . Cluver's (1981) study is, however, concerned with relatively disaggregated data
and IZEF may also have been appropriate. Indeed were the regressors exogenous
IZEF would yield consistent and efficient estimates, while 3SLS, if properly applied
and iterated, would yield estimates which were inefficient compared to the IZEF
estimates (Berndt, 1991, p379). The potential lack of efficiency is not that serious a
problem, but the fact that 3SLS appears not to have been iterated is potentially
damaging.
13Specifically that the following equalities hold: Li U i =1; Li 'Y =0 and Li'Yiy =O.
ij
14That is, using residuals from the estimated 3SLS equations to obtain new estimates of the variance-
covariance matrix and then reapplying the second stage of the 3SLS process and repeating this
process until there is no change in the parameter estimates (Kmenta, 1986, p700).
15 A second potential flaw in Cluver's (1981) thesis, failure to adequately test for fulfillment of
concavity conditions, has already been discussed in Chapter 3, section 3.4.2
16The Haw applies also to Cluver and Contogiannis (1984) which is based directly on Cluver (1981).
17Whether a 3SLS estimator is appropriate for the present analysis is dealt with in more detail below.
71
As noted earlier, iteration is necessary to yield parameter estimates which are invariant
to which share equation is dropped. Whether I3SLS was employed by Cluver (1981) is
not clear, since reference is made only to the use of 3SLS as developed by Zellner and
Theil (1962) (Cluver, 1981, P103). Given that Zellner and Theil's (1962)development
of 3SLS merely suggests the possibility of iterating the procedure without specifying
the implications and method of iteration (Zellner and Theil, 1962, p78) it is probable
that Cluver (1981) did not apply iteration and his results are not invariant to which
share equation was dropped. This conclusion is supported by the fact that although
symmetry constraints, which require the use of iterative estimators, are explicitly
discussed, and presumably applied, there is no discussion of the issue of the invariance
of different estimators.
Clearly, whether the regressors of the component equations of the Translog model are
exogenous or not, by iterating either the ZEF or 3SLS estimators parameter estimates
can be obtained which are both consistent and efficient, and invariant to which share
equation is dropped. An important issue is determining which estimator would be most
appropriate as incorrect choice would lead to inefficiency at best (using I3SLS instead
of IZEF) or inconsistency at worst (using IZEF instead of I3SLS). The choice of·
which estimator (lZEF or I3SLS) is employed should be determined by the
characteristics of the regressors. The choice between either the IZEF or I3SLS
procedures in other applications (and 'textbook' expositions) is motivated by intuition
regarding the probable endogeneity of input prices l8 rather than statistical testing of
that intuition (see for example Berndt (1991); Berndt & Wood (1975); Berndt &
Christensen (1973), Cluver (1981) and Jorgensen (1986». While intuition may be used
to make assertions about the nature of input prices for this application of the Translog
a more reliable method ought to be used to decide upon the nature of output.
The present analysis is concerned with estimating economies of scale and elasticities of
substitution in specific 2, 3 and 4 digit standard industrial classification (SIC)
industries. Following Berndt (1991, p460) it is argued that the level of analysis here is
l8 The Translog as originally specified has both input prices and output (in various fonus) as
regressors. Most applications are explicitly concerned with estimates of elasticities of substitution and
hence impose constant returns to scale a priori. Imposing constant returns to scale reduces the
appropriate system to the cost share equations alone (see section 3.3.1), where those cost share
equations have only prices as regressors (output is deleted (see 3.3.1.2 for details)). Hence the nature
of input prices, and not output, would need to be addressed when choosing between IZEF or I3SLS. In
applications such as this, where the magnitude of scale is an explicit concern, output appears as a
regressor (in both the share equations and the cost function which is included in the system) and the
nature of output also needs to be addressed.
72
sufficiently disaggregated to assume that input prices are likely to be exogenous to the
different industries who are unable to assert monopsony power in input markets.
Output cannot, however, be assumed a priori to be exogenous. Although theoretical
intuition would suggest that output is likely to be endogenous, the absence of
precedents in the applied literature prompted the need to test rather than speculate
upon the statistical nature of output. The theoretically appropriate method for testing
the nature of output is, however, not suitable for the Translog in this application.
4.2.4 TESTING ENDOGENEITY: THE SPENCER AND BERK (1981) TEST.
Hausman (1978) has developed specitication tests concerned with testing whether the
OLS assumption of exogeneity of regressors (orthogona1ity) is violated, in a variety of
different contexts (Hausman, 1978, p1251). Spencer and Berk (1981) provide a
simpler version of the test which Hausman proposed for the simultaneous equation
context (Spencer & Berk, 1981, P1079). Spencer and Berk's (1981) version of
Hausmans test is concerned with testing the specitication of the component equations
of systems (Spencer & Berk, 1981, p1079), The proposed test comprises two stages.
In the ftrst stage, the suspected endogenous regressors in the equation under scrutiny
are regressed on all the exogenous variables in the system, and the titted values, which
constitute instrumental variables, are retrieved. In the second stage the instrumental
variables are added to the original equation and OLS is performed on this expanded
equation (Berndt, 1991, p566; Greene, 1993, p618 and Kmenta, 1986, p718). The null
hypothesis that the regressors under consideration are exogenous is equivalent to
testing the null hypothesis that the coefficients on the instrumental variables are
insignillcant. While the Spencer and Berk Test will provide an indication of whether
the ZEF procedure or the 3SLS procedure ought to be used, precisely which form of
l9
those estimators should be employed would require testing for the presence of
autocorrelation.
Application of Spencer and Berk's (1981) test to the Translog model using the samples
employed here is, unfortunately, not possible. The second stage of the test could not be
applied for one of two reasons: a lack of degrees of freedom or excessive
multicollinearity. The degrees of freedom problem emerged when the nature of output
and other regressors which are linear functions of output20 were tested in the cost
19
In the case of the Zellner procedure IZEF or IZEF adjusted for autocorrelation (IZEF(AR)) and in
the case of 3SLS, I3SLS or I3SLS adjusted for autocorrelation (l3SLS(AR»
20 . tl
0 utput appears III le system as output squared, the product of time and output and the product of
different input prices and output.
73
21
function of the technical change augmented cost function . The addition of the fitted
values, obtained in the first stage of the test, to the appropriate equation increased the
number of regressors in that equation beyond the number of observations in the sample
22
thus eliminating all available degrees of freedom and rendering OLS impossible .
The multicollineairty problem emerged in all other equations where output appears as a
regressor. Attempts to apply the second stage of Spencer and Berk's (1981) test to the
cost functions of the homothetic, homogenous and Cobb-Douglas variable returns to
scale versions of the Translog23 , and the cost share equations of the technical-change
augmented and non-homothetic forms of the Translog24 all failed due to the presence
of excessive multicollinearity. Why the problem of extreme multicollinearity emerges is
intuitively obvious: the IVs used in the second stage of Spencer and Berk's (1981) test
are merely linear combinations of the other variables appearing as regressors in that
second stage.
The failure of the Spencer and Berk (1981) test means that the most appropriate
estimator cannot be determined statistically. Should output be exogenous IZEF
estimators (either IZEF or IZEF(AR) depending on the nature of the errors of the
component equations) would be appropriate yielding consistent and efficient estimates.
Employing I3SLS estimators would yield consistent but inefficient estimators. Should
output be endogenous I3SLS estimators would be appropriate yielding consistent and
efficient estimators. IZEF estimates would in this case be both inefficient and
inconsistent relative to the I3SLS estimates. Given that I3SLS estimates are consistent
irrespective of the context in which they are applied and the only penalty for applying
the I3SLS process in the wrong context is a loss of efficiency (which merely imposes
the potentially beneficial result that statistical inference is more conservative) one
could be tempted to merely apply I3SLS. One problem with adopting that approach
here is that the sample used is very small and while I3SLS estimates will be consistent
21 See Appendix 5.1 for detailed functional fonus of the Translog when different maintained
hypotheses are imposed a priori.
22The sample being used here contains 19 observations. The cost function for the technical-change
augmented specification of the Translog contains 15 regressors (including the constant). 7 of those
regressors are some function of output. Were one to employ Spencer and Berk's (1981) test, fitted
values for each of these 7 regressors would need to be appended to the cost function generating an
equation with 22 regressors. No degrees of freedom would be available to perfonn OLS on that
equation and the test cannot be perfonned.
230utput does not appear as a regressor in the cost function of the linear homogenous fonn of the
Translog.
240utput does not appear as a regressor in the cost share equations of any of the other specifications of
the Translog.
74
25
there is no guarantee of their unbiasedness. For this reason, and other arguments ,
both classes of estimator (IZEF and I3SLS) are used here. The precise form of the two
classes of estimator which would be appropriate 26 was determined by testing for
autocorrelation.
4.2.5 AUTOCORRELATION
The asymptotic efficiency of the IZEF and I3SLS estimators is derived from the
assumption in both cases [Zellner, 1962, p350 (for IZEF) and Zellner and Theil, 1962,
p55 (for I3SLS)] that disturbances of the equations estimated in the first stage are
independently distributed 27 (Greene, 1993, p498 and Kmenta, 1986, p706). Berndt and
Christensen (1973b, p95) citing both Ourbin (1957) and Malinvaud (1970) suggest
employing a conventional single equation Ourbin-Watson statistic to test for the
presence of autocorrelation in the component equations of the Translog model. The
test, which is performed individually for the separate equations of the system, involves
the computation of the test statistic using the residuals from the [mal stage estimates
(Malinvaud, 1970, p509). The appropriate degrees of freedom for the test differ for the
IZEF and I3SLS estimators. When applying the Ourbin-Watson test to the IZEF
estimates the number of degrees of freedom are the· number of regressors in each
equation and the number of observations in the sample. For the I3SLS estimates, the
number of degrees of freedom are the number of exogenous variables employed in the
first stage of the procedure and the number of observations in the sample (Berndt and
Christensen, 1973b, p95 and Malinvaud, 1970, p509).
Should autocorrelation be detected, the estimator being employed ought to be

modified to take this phenomenon into account. For the ZEF estimator, Greene (1993,
p498) and Kmenta (1986, p646-647) show that by preceding the usual ZEF procedure
with a stage which first estimates the coefficient of correlation (p) for each equation of
the system and then transforms the data to remove any autocorrelation the resulting
ZEF estimator would be efficient. For the I3SLS procedure, autocorrelation can be
adjusted for in an analogous manner. Rather than employing 2SLS in the first stage of
the 3SLS a weighted 2SLS estimator which explicitly accounts for the serial
correlation of the errors of the component equations would be employed (Kmenta,
1986, p706).
25
See footnote 5 of Chapter 5.
26That is the autocorrelation-augmented version as opposed to the version which assumes that the
component equations of the Translog system fulfill the assumptions of the CLR model.
27 Because the data used here are time-series it is assumed that the errors of the component equations
are homoscedastic and no testing of this hypothesis was undertaken.
75
According to Greene (1993) iteration of the ZEF process modified in this manner
would not yield maximum likelihood estimates. Maximum likelihood estimates are
important for ensuring that parameter estimates are invariant to which share equation
is dropped. Given that the I3SLS estimates, although not maximum likelihood
estimates are nevertheless invariant to which equation is deleted from the system,
iteration of the ZEF process modified for autocorrelation ought to be invariant to
which share equation is dropped. The literature is, however, conspicuously silent on
this point. Despite giving considerable attention to estimating the Translog system
when autocorrelation is suspected and to the issue of invariant estimators. Berndt
(1991), for example, does not mention that the IZEF estimator modified for
autocorrelation is not the maximum likelihood estimator nor whether or not it is
invariant to which share equation is dropped. The same arguement can be applied to
the I3SLS estimator adjusted to account for autocorrelation.
Applying the Durbin-Watson Test, as described above, often yields test statisitcs which
fall within the region of indecision. While one can employ Theil and Nagar's (1961)
result that where the regressors are changing slowly, the upper distribution of the D-W
statistic is the appropriate distribution (Kennedy, 1985, p106) to reach a decision
regarding the presence of autocorrelation, it was decided that further testing may be
needed. Using an estimator which specifically accounts for autocorrelation, by
assuming a different error structure (Le. errors which are serially correlated), changes
the specification of the model. The specification assumed when an autocorrelation-
augmented estimator is used differs from the specification assumed when a
conventional estimator is employed in that for the former the coefficient of correlation
(p) is not zero. The validity of the specification assumed by an autocorrelation-
augmented estimator can be established by testing the null hypothesis that p =O. The
Likelihood Ratio Test, discussed below, is an exanlple of a test which may be
employed in the systems context to test such a nested hypothesis. The mechanics of
applying the test in this context are discussed further in section 4.3.1.
4.3 METHODS OF STATISTICAL INFERENCE
In this study statistical testing is required in three areas: determination of the

appropriate estimator; determination of which of the various restrictions (such as
homogeneity and homotheticity) which the researcher is able to impose on the
underlying technology are valid, and determination of how good the overall fit of the
model is. The choice of appropriate estimator rests on establishing the statistical nature
76
(i.e. whether endogenous or exogenous) of the regressors and the relationship among
the disturbances of the component equations of the system (i.e. whether component
equation disturbances are autocorrelated). Both these issues have been dealt with
above (see sections 4.2.4 and 4.2.5). Should autocorrelation be suspected on the basis
of the Durbin-Watson test described in section 4.2.4 and an autocorrelation-augmented
estimator be employed the validity of the derived estimates can be tested using a
nested-hypothesis test: the Likelihood ratio test. The same general procedure can be
employed for testing which of the various restrictions (such as homogeneity and
homotheticity) which the researcher is able to impose on the underlying technology are
valid. Nested hypothesis tests in general and the Likelihood ratio test in particular are
discussed in section 4.3.1. Section 4.3.2 is concerned with the final area in which
testing is required: the overall fit of the model.
4.3.1 NESTED HYPOTHESIS TESTS
Three alternative asymptotically equivalent tests are available for testing the validity of
parameter restrictions (nested hypotheses) imposed on systems of equations: the
Likelihood ratio test (LR test); the Wald test and the Lagrange multiplier test (Greene,
1993, p129 and Kennedy, 1985, p58).The three tests are asymptotically equivalent
generating chi-square statistics with degrees of freedom equal to the number of
restrictions imposed. As the small sample properties of the three tests are only known
for a few special cases28 , choice among the three tests is most often made on the basis
of ease of computation29 • Prirna facie either the Wald or Lagrange multiplier tests
would appear the most attractive. While the LR test requires the calculation of both
the constrained and unconstrained estimators, the Wald test requires only the
unconstrained estimator and the Lagrange multiplier test only the constrained
estimator. Computational consideration inform the choice here: not only is the LR test
simpler computationally when constrained and unconstrained estimators are easily
obtained 30 (Kennedy, 1985, p59) but the program used for the econometric work,
Micro-TSP, generates the determinant of the variance-covariance matrix of residuals,
which can be used to construct a LR test statistic directly when estinlating systems of
equations.
28 A weakness common to all three tests when the sample is small is the fact that critical values from
the chi-square distribution are used despite the fact that in small samples they are not distributed as
chi-square (Kennedy, 1985, p64).
29 Choice of test may lead to conflicting results in the small sample case. Berndt and Savin (1977)
show that in small samples the value of the three tests adopts the following inequality: Wald>
Likelihood ratio>Lagrange multiplier (Berndt, 1991, p467 and Kennedy, 1985, pM).
30The WaId or Lagrange multiplier test would be preferred where constraints impose or remove non-
Iinearities, respectively (Greene, 1993, pI29-130)
77
Intuitively, the LR test is based on the idea that any restriction on parameters would be
valid if there were no significant reduction in the value of the log-likelihood function
once the restriction is imposed (Kmenta, 1986, p491). The test is based on comparing
the difference between the log-likelihood function of the constrained and unconstrained
estimates. Formally, if we denote the maximum of the likelihood function for the
A
constrained estimator (which constitutes the null hypothesis) as Le ' and the maximum
of the likelihood function of the unconstrained estimator as iJ, the LR is defined as:
4.1
which will lie between 0 and 1 as both likelihood functions are positive and because a
restricted optimum can never be greater than an unrestricted optimum (i.e4 >ie ).
The test procedure is based on the result that the large sample distribution of -2ln A is
chi-squared with degrees of freedom equal to the number of restrictions imposed
(Greene, 1993, p130-131 and Kennedy, 1985, p66).
Two alternative formulae can be used to calculate the LR test statistic. The first
formula which is merely a mathematical manipulation of the definition of the LR as
defined by 4.1 31 (Berndt, 1991, p466):
4.2
Where m refers to the number of restrictions imposed in the constrained model. The
null hypothesis that a constraint is valid will be rejected if the value of the test statistic
is greater than the appropriate critical Chi-square value.
The alternative formula for the LR test statistic (for a proof see Cramer, 1986, p122)
is:
4.3
Where n is the size of the sample, IQllcl and IQcl are the determinants of the residual
variance-covariance matrices for the unconstrained and constrained models
31 TI1e equivalence between fonnulation 4.2 and that given by 4.1 is obvious when considering that the
log of a ratio is equal to the difference between tile log of tile numerator and tile log of tile
denominator.
78
respectively, and In refers to the number of restrictions imposed on the unconstrained

model in order to generate the constrained model. Interpretation of the test statistic
generated by the second formula is equivalent to the interpretation applied to the first
formula.
The LR test is used to test the validity of imposing restrictions (usually nonlinear
restrictions) on the parameters of a model, and as a consequence is appropriate for
testing nested as opposed to non-nested hypotheses (Kennedy, 1985, p58 and Kmenta,
1986, p491). Fortunately, a large amount of the statistical testing undertaken in the
context of estimating the Translog model involves the testing of nested hypotheses.
Indeed determining both the correct specification of the Translog model and whether a
chosen error specification is correct, both involve the testing of nested hypotheses.
Alternative maintained hypotheses, such as homotheticity, homogeneity, constant

returns to scale (homogeneity of degree one), Cobb-Douglas technology, can all be
specified, as described in section 3.3.1.2, by constraining the parameters of the non-
homothetic form of the Translog. Thus, determining which of the alternative
maintained hypotheses is valid for the different sectors, can be ascertained by using a
LR test of the validity of the parameter constraints associated with different maintained
hypotheses.
In addition to being able to test which specification of the Translog is appropriate, the
LR test may also be used to establish whether an estimator which accounts for
presumed autocorrelation of disturbances is valid (Berndt, 1991, p497 and Kmenta,
1986, 711). The test amounts to testing a null hypothesis that the disturbances are
nonautocorrelated against an alternate hypothesis that the disturbances are, in fact,
autocorrelated. The test involves estimation of the parameters of the system with and
without the null hypothesis imposed, and then a comparison of the values of the
maximised likelihood functions (or determinants of the variance-covariance matrices)
in the manner described above, with degrees of freedom equaling the number of
equations in the system 32 (Kmenta, 1986, p711).
32
As the I3SLS, I3SLS (AR) and the IZEF (AR) estimates are not the maximum likelihood estimates
they should not, theoretically be used for LR testing. However, given that the LR test is an asymptotic
test and given that I3SLS, I3SLS (AR) and lZEF (AR) have tile same asymptotic distribution as the
maximum likelihood estimator, the asymptotic validity of tile test should be unaffected (Kmenta,
1986, p711).
79
4.3.2 GOODNESS OF FIT
In the single equation context, the coefficient of detennination (R2) is usually

employed as a measure of the goodness of fit of a model. This coefficient is calculated
as one minus the ratio of unexplained variation of the dependent variable (i.e. the sum
of the squared residuals) to the total variation of the dependent variable (i.e. the sum of
the squared deviations of the estimated values of the dependent variable around their
mean) (Kennedy, 1985, pll; Gujarati, 1988, p176) i.e.:
4.4
where, ej denotes the residuals at each observation y the fitted value of the dependent
variable at each observation· and y the mean of the fitted values of the dependent
variable. Because the OLS estimator minimises the sum of the squared residuals (i.e.
L ej2 ), R2 will be maximised when least squares is employed.
This single-equation measure of goodness-of-fit may not, however, be appropriate in

the context of a system of equations. Least squares estimation ensures that the sum of
residuals is zero, implying that the mean of the residuals will also be zero. As a result
the numerator of the above expression for R2 is equivalent to the sum of the square of
deviation of the residuals around their mean. For system estimation in general, the sum
of the residuals for each equation is not necessarily zero, implying that the mean of the
residuals may not be zero and it is therefore possible that Le: could be greater than
L(Y- y)2 which would yield a negative R2 (Berndt, 1991, p468). A second problem
regarding the use of the single equation R2 emerges from the estimation technique
used to estimate systems of equations. The estimators employed here minimise the
detenninant of the variance-covariance matrix (rather than the sum of squared
residuals of each equation), and hence do not necessarily maximize the R2 of each
equation (Berndt, 1991, p468 and Kennedy, 1985, p144).
Given these two problems with the single equation coefficient of detennination, an
alternative measure is needed. McElroy (1977, p384) has developed a measure of
goodness of fit for SUR systems:
4.5
80
The numerator, lE' El, is the determinant of the variance-covariance matrix of the
errors of the system, and is therefore the sum of squared residuals (RSS) of the system
(Berndt, 1991, p468 and Gujarati, 1988, p176). The denominator, IY'YI,
is the total
sum of squares of the system. The measure is confined to the range 0 < R: < 1 and may
be related to either an (asymptotic) F test statistic (McElroy, 1977, p384) or a LR test
statistic (Berndt, 1991, p469):
4.6
Where T is the number of observations in the sample, and the number of degrees of
freedom is given by the number of independent slope coefficients in the system of
equations.
4.4 DATA CONSIDERATIONS
Estimating the Translog system here requires data on two types of variables: the
variables appearing in the system directly and the instrumental variables used in I3SLS.
Sections 4.4.1 to 4.4.3 are concerned with the variables appearing directly in the
system while section 4.4.4 is concerned with the choice of appropriate instruments and
their sources.
Four types of variables appear in the Translog system: the level of output Y for the
sector under analysis, the prices of the different factors of production ~, the share of
the different factors in total costs Si and total costs C. Cost shares of different factors
are calculated as the ratio of total expenditure on each input to total costs. Where total
expenditure on an input is not directly available, expenditure can be derived from the
product of the price of the factor P; and the quantity employed Q .While data on the
level of output produced in the different sub-sectors is readily available 33 , input price
and cost share data, particularly for capital, are less readily available, requiring the
manipulation of other data. The data employed in this analysis have been obtained from
two sources. The price of Capital (PK ) for the manufacturing sector34 was calculated
employing data appearing in Lombard and van den Heever (1990, p19) and data
33The IDC publish series of total production for different subsectors of manufacturing. They derive
total production figures using CSS input-output tables for the years when those tables are available.
Total production for the interim years is calculated using trends in the indices of physical volume of
manufacturing production per sector (IDC, 1992, p7).
34Separate prices of Capital could not be obtained for all the sectors under analysis. The problems
involved in calculating the price of Capital for different sectors are discussed in detail below.
81
published by the Industrial Development Corporation (IDC, 1992). All other variables
employed in this analysis were derived from the later source. The problems associated
with the development of price and cost share data differ with the inputs. Before
discussing the problems associated with the price and cost share series employed here
for the different factors of production, some comment on the choice of inputs is
required.
Applications of the Cobb-Douglas and CES generally employ only two inputs: capital
and labour35 • Applications of flexible forms such as the Translog generally employ
more than two inputs. The convention of using aggregated capital and labour as inputs
36
has, however, been adopted in most empirical applications of the Translog system •
Indeed, it would appear that it is only where the objective of analysis has been the
estimation of elasticities of substitution between components of an aggregate input (in
order to test whether the requirements for consistent aggregation have been fulfilled)
that aggregates have not been used 37 • Choice of other inputs has largely been informed
by the objective of the analysis at hand. For example, where the nature of the relation
of energy to other inputs has been concerned,' energy together with materials, capital
and labour have been used 38 •
Applications of the Translog which are concerned with the estimation of economies of
scale use, in addition to capital and labour, those inputs which are important in the
generation of the output of the industry with which they are concerned. For this reason
both Christensen and Greene (1976, p663) and Greene (1983, p13l), whose concern is
the estimation of scale economies for D.S electric power generation, use fuel as a third
input. While it would be desirable to include the most important factor, other than
capital or labour, for each of the industries analysed here, the difficulty of establishing
which 'other' factor is most important and obtaining appropriate data precludes
adopting Christensen and Greene (1976) and Greene's (1983) approach. Instead, a
third input, 'materials' - which is essentially a composite input of all factors of
production other than capital or labour - is used.
35 Some South African applications, for example Browne (1943) and Spandau (1973) have, however,
disaggregated labour into White and Black categories.
36
See for example Berndt and Wood (1975); Christensen and Greene (1976); Contogiannis and
Cluver (1984); Denny and May (1978); Denny and Pinto (978); and Green (1983).
37 See for example Berndt & Christensen (1973b).
38
See Berndt & Wood (1975) and Cluver & Contogiannis (984)
82
4.4.1 CAPITAL
The generation of prices and quantities of capital services is beset by problems which
are very different from those associated with either labour or materials. While active
markets exist for transacting the services of labour and materials, in general no market
exists for capital services since "... the supplier of the capital service and its ultimate
user are typically within the same economic unit" (Christensen and Jorgensen, 1969,
p293). As a result data on the value of transactions in capital services do not exist and
general procedures for constructing price· and quantities cannot be adopted.
Consequently, alternative procedures have been developed. These involve fairly
lengthy chains of indirect inference, beginning with data on the value of transactions in
investment goods and involving the imposition of fairly strong assumptions (Jorgensen
& Griliches, 1967, p255). As already stated, the quantity of different factors is required
for the generation of cost shares and is not, in itself, needed for the estimation of the
Translog model. In the case of capital, total expenditure on capital services in each
sub-sector can be directly calculated as the sum of depreciation and interest paid minus
interest received.
The absence of a market in capital services and hence an explicit price of capital
necessitates the construction of an 'implicit rental price' for capital services. Because
both supplier and consumer of capital are within the same economic entity, the implicit
rental value is conceptually the same as the user cost of capital (Mark & Waldorf,
1983, p5). Formulae for the user cost of capital emerge from manipulating the
neoclassical theory of optimal capital accumulation - in particular from a manipulation
of the conditions required for the fulfillment of the assumption of perfect capital
markets.
The precise form that the function describing perfect capital markets takes depends on
the assumptions which are adopted regarding the survival function 39 • The choice of a
specific form is constrained by Arrow (1964), Hall (1968) and Jorgensen's (1974),
establishment of the duality between gross capital stock and the user cost of capital (Bi
<pm, 1989, p5I). An important implication of this duality is that, for consistency, the
same assumption regarding age-efficiency (survival) ought to be adopted in the
generation of both price. and quantity series. Should the procedure used for creating
capital stocks assume a linear survival function, so to should the calculation of the user
39 Thatis the function expressing the relationship between the retirement of capital units over time
and the loss of efficiency of remaining units.
83
cost of capital adopt a linear survival function. In South Africa, fixed capital stock
figures are produced using a perpetual inventory method where provision for
depreciation is on a straight-line basis for most economic sectors (Mohr, 1988, p66
and Moll, 1990, p200). The use of a straight-line method of depreciation embodies the
assumption of a linear survival function.
Adopting the assumption of a linear survival function and manipulating the condition
expressing the requirements for the existence of a perfect capital market yields the
40
expression
q(t) = f' e-r(s-t)c(t)ey(s-t) (1- o(s - t»ds 4.7
where, q is the price of capital goods, r is the nominal rate of return on financial assets,
'Y is the investment price index and 8 is the rate of decline of the efficiency of capital
goods and s is the assumed life span of the investment good. The equation states that
the current purchase cost of an investment good at a particular time is equal to the
present value of its future service price, when allowance is made for retirement and a
linear decline in efficiency with age (Bi<j>rn, 1988, p53)
An expression for the user cost of capital (c) may be obtained by first differentiating
equation 4.7 with respect to the time of purchase41 ,
aq =q(r+8)-e(t) 4.8
at .
and then rearranging the result making e(t) the subject of the formula yields:
aq
e(t) =q(r+O)-- 4.9
at
Because the differential aq/at is concerned with instantaneous rates of change which
are not empirically tractable applications using this formula substitute aq/at with
qr - qr-Jql-l which is essentially the differential for discrete periods of time. The
formula employed by Lombard and van den Heever (1990, p19) to calculate the user
40 See Bi<jlm (1988) for Cl delailed proof.
41 1n general, given a funclion q(t) = li(t,s)ds, -oq Of s)ds- J(t t) '

= f.ootat
-(t
at ' ,
84
cost of capital for South African manufacturing is similar to that given by equation 4.9.
They employ the following:
c(t) =q(r + 8) -q[(l- 8)(q, -q,-Jq,-I)] 4.10
where: r which represents the interest rate is the annual average rate on long-term
company securities; 8 represents the depreciation rate and employs national accounts
assumptions; q represents the purchase price of capital which for the purposes of their
application, was assumed to be the deflator for gross fIxed investments. Although
42
aware of the impact of taxes on the user cost of capital , as manifested in investment
incentive schemes, Lombard and van den Heever (1990, p8) do not take these effects
into account "for practical reasons." They do, however, concede that such schemes
were likely to have decreased the user cost of capital as calculated using their formula.
What the source of the difference between the formula employed by Lombard and van
den Heever (1990) and the one presented in 4.9 is, is unclear. Lombard and van den
Heever (1990, p7) although not explicit, appear to cite Jorgensen43 (1963) as the
source of their formula. Jorgensen (1963) does not explicitly deal with the derivation
of a user cost formula although a modified version (incorporating the effects of tax)
employed by him (Jorgensen, 1963, p249) is consistent with the formulation given by
4.9. A possible explanation of the source of the difference between equation 4.9 and
equation 4.10 is that Lombard and van den Heever (1990) employed different
assumptions regarding the age-effIciency of assets and changes in the service price of
assets over time to those employed above. In particular, it is possible that the survival
function adopted by Lombard and van den Heever (1990) was non-linear44 •
Given the uncertainty regarding the assumptions implicit in equation 4.10, and as a
result the source/s of the difference between equations 4.9 and 4.10, equation 4.9,
which is consistent with the assumption employed in the perpetual inventory method of
45
deriving capital stock and hence fulftlls the consistency required by the duality
between the quantity and price of capital, has been used here. The application of
equation 4.9, requires data on a nominal rate of interest, a rate of depreciation, and the
42 How the user-cost of capital is effected by different types taxation, both direct and indirect, has been
explored by, inter alia, Christensen and Jorgensen (1969); Hall and Jorgensen (1967).
43Jorgensen (1963) is the only reference in Lombard and van den Heever (1993) which deals with the
capital theory.
44 Use of the simple exponential function would, however, yield equation 4.9 suggesting that a more
complex non-linear survival function may have been used.
45 That the efficiency of an asset declines in a linear fashion (Mohr, 1988, p66 and Moll, 1990, p200)
85
purchase price of capital. Ideally series of these variables which are specific to each
sector under analysis ought to be employed. While it is plausible to assume that the
rate of interest faced by different firms can be represented by the same composite rate
of return46 , it is possibly less plausible to assume that the price of investment goods
used by different fIrms and the rate of depreciation of that capital, is the same.
Unfortunately, the data set employed prevents the generation of prices and
47
depreciation rates specific to the different sectors . As a result, the same user cost of
capital series had to be used for all the sectors under analysis. The raw data used to
generate the user cost of capital and the user cost of capital are presented in Table 4.1
below.
The development of the user cost of capital presented here is further Hawed in one
crucial respect: the effect of taxation on the user cost of capital is ignored. While
theoretically incorporating the effects of taxation is relatively straightforward (see for
example Christensen & Jorgensen (1969) or Jorgensen (1963», the practical constraint
of data unavailability has prevented the inclusion of the effects of taxation in the
calculation of the user cost of capital for South Mrican manufacturing. That the user
cost of capital is likely to have been effected by taxation is without doubt.
4.4.2 MATERIALS
Although markets exist, and hence transactions occur, in material inputs, data on the
value (either in nominal or real terms) of expenditure on materials appears only to be
available via indirect inference using (or rather reversing) the conventions of national
income accounting. Where value-added is available, reversing the method used to
generate value-added enables one to construct the price of, and total expenditure on,
materials. There is, fortunately, no need to construct material prices for the different
sectors analysed here. The IDC publish data on material prices48 for the various sub-
46The interest rate used in equation 4.9 represents the opportunity cost which firms incur as a
consequence of them holding real capital assets as opposed to interest bearing financial assets. As
fimls are able to invest in a spectrum of different financial assets the opportunity cost incurred by the
firm investing in real assets can best be represented by a composite rate of interest.
47While the data set contains series on the conSL:'lnt Rand value of capital, no data on the current Rand
value of capital is available, and as a consequence a price of capital series cannot be created. Further,
altllough a series of the current Rand value of depreciation does appear, in order to generate a rate of
depreciation rate (Le. tlle percentage of capital depreciated each time period) specific to each sector,
eitller a constant Rand value of depreciation, or a current Rand value of capital would need to be
generated.
48 The IDC publish indexes of tlle prices of inputs, excluding labour or capital, specific to different
manufacturing sub-sectors. Local prices and input prices are published seperately and combined into a
single input price index which has 1990 as 100 (IDC 1992, p9). The price index used here is the
composite index modified so that 1970 = 1.
86
sectors of manufacturing. Data on the total expenditure on materials is not, however,

directly available and needs to be created. An indirect method for calculating total
expenditure on materials, analogous to that used by Denny and May (1978), which is
based on the conventions of national income accounting, has been employed here.
..
Ta bl e 41 User cost 0 f C apt'
'tal
Year Purchase Interest Depreciation Capital User Cost User Cost User Cost
price of Rate (r) Rate (8) Gains Uj) of Capital: of Capital: of Capital:
capital (q) Value c(t) Index (Pk) Log of
Index
(lnP!c).
1972 1.575 0.095 0.124 0.094 0.251 1.000 0.000
1973 1.752 0.095 0.118 0.112 0.261 1.038 0.038
1974 2.001 0.116 0.115 0.142 0.320 1.274 0.243
1975 2.406 0.131 0.113 0.202 0.385 1.531 0.426
1976 2.809 0.135 0.115 0.167 0.535 2.129 0.756
1977 3.075 0.136 0.114 0.095 0.674 2.684 0.987
1978 3.425 0.120 0.111 0.114 0.677 2.697 0.992
1979 3.917 0.109 0.101 0.144 0.679 2.703 0.994
1980 4.466 0.115 0.093 0.140 0.789 3.140 1.144
1981 5.044 0.141 0.092 0.129 1.046 4.164 1.426
1982 5.866 0.156 0.093 0.163 1.298 5.166 1.64~
1983 6.633 0.152 0.095 0.131 1.508 6.002 1.792
1984 7.204 0.177 0.097 0.086 1.888 7.516 2.017
1985 8.489 0.189 0.103 0.178 2.300 9.159 2.215
1986 10.373 0.177 0.106 0.222 2.714 10.804 2.380
1987 11.661 0.167 0.108 0.124 3.083 12.273 2.507
1988 13.384 0.171 0.109 0.148 3.600 14.332 2.662
1989 15.682 0.182 0.106 0.172 4.345 17.298 2.851
1990 17.301 0.181 0.106 0.103 4.862 19.358 2.963
Note: 1. The user cost of capital was calculated using the following formula c(t) = q(r + 8) - q:,
where capital gains (q) is calculated as q= qt - qt-ll qt~l
2.For 1972, the lagged value of the price of capital needed for calculating Ct was 1.44
(Lombard and van den Heever, 1990, pI9).
3.The purchase price of capital (q) used here is a price index of the deflator for gross fixed
investment which has 1960 as it's base Year.
Gross Domestic Product (GDP) can be calculated using three different methods which
are equivalent ex post: the production, income and expenditure methods (Mohr et al,
1988, p38). Manipulation of the production and income methods has been used here to
generate data on expenditure on materials. The production method provides an
estimate of GDP based on the sum of the contribution of each industry to GDP.
Double counting is avoided by measuring the contribution of each industry in terms of
value added (net output) rather than gross output. Value-added is merely the
difference between the value of any industry's output and its purchases of intermediate
products (Mohr et aI, 1988, p39). An alternative definition of value-added emerges
from the income method of calculating GDP and rests on the equality between the
87
alternative methods for calculating GDP. The income method of calculating GDP
involves summing all income received by factors of production. The equivalence of the
income and production methods of calculating GDP rests on the equality of the value-
added of any industry, as defined, above and the incomes received by factors of
production in that industry. Hence value-added for any industry can be calculated
either as the sum of wages and salaries, profits, rent, depreciation allowances, interest
and dividends or the difference between the value of the industryies output and its
purchases of intermediate products (Mohr, et aI, 1988, p39).
Denny and May (1978, p59) whose analysis required data on the use of materials for
Canadian manufacturing as a whole reversed the procedure employed by Statistics
Canada to derive real domestic product. Statistics Canada calculate real domestic
product, using the production method for calculating GDP, as the difference between
deflated gross output and denated materials (Denny & May, 1978, p59). Denny and
May (1978), reversing the production method of calculating GDP, calculated current
and constant dollar materials as the difference between current and constant dollar
gross output and domestic product. They then derived an implicit price of materials by
dividing the· current dollar value of materials by the constant dollar value (Denny &
May, 1978, p59).
Reversing the production method of calculating GDP and finding the difference
between the gross output of any industry and the contribution of that industry to
domestic product (Le. value-added or net-output) would yield the value of the
intermediate products employed by that industry. Finding this difference, both in real
and nominal terms, for the Canadian manufacturing sector as a whole is what Denny
and May (1978) did to derive a price series for materials. As the IDC publish data on
both the value of output of an industry and the value-added; data on expenditure on
materials can be computed directly for the different subsectors of South African
manufacturing. There is, unfortunately, a flaw in this approach which Denny and May
(1978) appear not to have considered, which leads to an upward bias in the value of
expenditure on materials derived in this manner. The IDC (1990, p6) point out that
rather than merely being the difference between gross output and intermediate
products used in an industry, value-added is the difference between the gross output
and intermediate products plus indirect taxes less subsidies. Reversing the production
method of finding value-added, and subtracting value-added from gross output will not
yield data on expenditure on materials alone but will generate data on expenditure on
materials and indirect taxes and subsidies. Unless separate data on indirect taxes and
88
subsidies are available it would appear that it is not possible to obtain a 'true' measure
of expenditure on materials.
The IDC has calculated value-added for the sub-sectors of South African
manufacturing for the period 1972-1990 using the income method described above.
The series were adjusted so as to equal the contribution of each sub-sector to GDP
(IDC, 1992, p7), and are thus equivalent to value-added calculated using the
production method. Real total expenditure on materials (using 1990 as the base) was
calculated here by subtracting real value-added from the real value of output (in 1990
prices). In order to be able to employ expenditure on materials in the calculation of
cost shares, a nominal value had to be employed. This was achieved by multiplying the
real cost of materials (i.e. in 1990 prices) by a price index which had 1990 =1. That
price index was obtained by dividing each observation of the composite price of
49
materials index published by the IDC, which has 1990 = 100 •
4.4.3 LABOUR
The IDC (1992) publishes time-series of both the total number of labourers employed
in each sector of manufacturing and the nominal rand value of expenditure on labour.
The nominal rand value of expenditure on labour can be used in the computation of
total costs and thus the cost share of expenditure on labour. This precludes the need to
derive a labour quantity variable for the different sectors. A price of labour series was
derived by dividing nominal total expenditure on labour by the total number of
labourers employed in each sector. This price of labour was then converted into a price
index by dividing each observation by the fIrst observation 50 • Deliving the cost share of
labour and the price of labour in this manner, embodies a number of potentially
untenable assumptions and is thus problematic in a number of respects.
Total expenditure on labour, while ret1ecting the total amount expended on wages and
salaries, does not reflect total expenditure on labour services. The approach adopted
here, by regarding the quantity of labour as merely the sum of the number of labourers
implicitly assumes that labour services are proportional to the stock of labour. Such an
assumption is obviously naive' (Jorgensen and Griliches, 1967, p266). The most
obvious solution to this problem would be to measure labour in terms of man-hours.
49 Raw dam and the final expenditure on materials series, together with the relevant price of materials
data, used in this analysis are found in Appendix 4.
50The raw data required to generate both the total expenditure on labour and the price index of labour
for the different sectors, together with the derived series, are presented in Appendix 4.
89
However, assuming that one could measure services of labour in terms of man-hours
would be equally incorrect, for it fails to account for variations in the intensity of effort
and the impact that qualitative differences between workers make on labour inputs.
'[T]he intensity of effort varies with the number of hours worked per week, so that the
labour input can be measured accurately only if data on man-hours are corrected for
variations in the number of hours per man on labour intensity' (Jorgensen and
Griliches, 1967, p266). Although ratios have been developed elsewhere to adjust man-
hours for intensity effects (Denison (1962) cited in Jorgensen & Grilliches (1967,
p266)), none exist for the present study. While in principle, a fairly accurate measure
of the flow of labour services can be constructed and have been employed in
applications of the Translog system (see for example Denny & May (1978) and Denny
& Pinto (1978)) lack of the required information precludes such accuracy in this study.
A further problem in this regard is that in the data set employed, data on labour
remuneration reflects remuneration to all employed in a a particular sub-sector during a
specific period of time (including casual and seasonal employees (IDC, 1992, p3)).
Hence multiplying the number of workers by the average number of hours (even if
adjusted for hours worked and intensity) would not provide an accurate indication of
the flow of labour services. Indeed, the fact that in the data set used the number of
employees includes casual and seasonal employees introduces a bias into the price of
labour. The price of per unit of labour is derived by dividing total expenditure on
labour by the total number of employees. Because the total number of employees
includes casual labourers the price of labour will be biased downwards.
4.4.4. INSTRUMENTAL VARIABLES
The 2SLS procedure, which is the first stage of the 3SLS procedure, uses all the
exogenous variables in a system as instruments (Kennedy, 1985, p134). The Translog
cost function and cost share equations can be regarded as the 'supply side' of a broader
market modeL As a consequence exogenous variables appearing in both the Translog
system and the demand side of the model would be appropriate instruments for 3SLS.
Variables in the Translog model which are definitely exogenous are those which do not
include output in any form and are therefore appropriate instruments and are used
here.
While no demand model has been specified to determine appropriate instruments the
variables which are likely to be employed in such a model are intuitively obvious.
90
Conventionally, demand for a commodity is modeled as a function of, inter alia, the
price of the commodity concerned, the prices of other related commodities, disposable
income, the size of the population, and the level of expenditure. Following this
convention, in addition to the exogenous variables in the Translog, the following
variables have been used as instruments in the different sectors analysed here: an
output price index, a consumer price index (as a proxy for the prices of related
commodities), an index of the level of manufacturing employment (as a proxy for the
population), an index of real personal disposable income, and indexes of the level of
real private and government consumption expenditure. These instruments together
with details of how they were generated are presented in Appendix 4.2.
CHAPTER 5
ECONOMETRIC RESULTS
5.1 INTRODUCTION
Given that the central concern of this dissertation is a comparison of the technology
used in wage and luxury goods sectors examples of both types of sectors have been
analysed. Motor Vehicles, Parts and Accessories (ISIC 3840) is analysed as a
representative luxury goods sector. The Electrical Appliances and Household Goods
(ISIC 3833) and Furniture (Isle 3320) sectors are analysed as representative wage
goods sectors. While the output of the Motor Vehicles I sector can quite reasonably be
deemed a luxury the output of the Appliance2 and Furniture3 sectors is not necessarily
exclusively limited to commodities which could be regarded as 'basic' or 'wage' goods.
Indeed, given that electricity is a prerequisite for the purchase of an electric appliance
and in addition that 23 million South Mricans (approximately 60% of the population)
(I.D.T, 1993, plO) do not have electricity, it is likely that consumers of appliances
would tend to fall into middle or upper income brackets rather than lowest income
brackets (however these may be defined). Similarly furniture would appear to be prima
facie a luxury commodity. Nevertheless, both these sectors were chosen for analysis on
the grounds that not only is any growth path which improves the distribution of income
likely to increase demand for these commodities in the future but also that the
proposed national housing and electrification campaigns (ANC, 1994, p22 and p33)
are likely to effect demand for these goods in the short-term.
Using the Translog system to obtain estimates of economies of scale and elasticity's of
substitution involves use of both statistical and theoretical criteria to choose between
I The Motor Vehicles sector covers the "...specialised manufacture of mOtor vehicles, caravans,
trailers, vehicle bodies, motor vehicle parts and accessories such as engines, brakes, radiators,
transmissions, frames etc." (IDC, 1992, p22).
2 The electrical appliances and household goods sector is involved in the manufacture of "...smaller
electrical appliances and housewares, such as electric space heaters; blankets and heating pads; hot
plates, boilers, roasters, toasters and food mixers; irons and mangles; fans, vacuum cleaners and floor
. waxers and polishers; hair driers, toothbrushes, hair clippers, shavers and water heaters." (SIC, 1988,
p61-62).
3 The Furniture sector involves" ... the manufacture of household, office, public building, professional
and restaurant furniture and fixtures which are made mainly of wood or other materials other than
metal." (SIC, 1988, p49).
93
competing estimates. Theoretical criteria ought, however, to be afforded a superior

status. Translog estimates are valid only if a dual relationship between production and
cost holds for the sample under analysis. Two conditions of duality, that the cost
function be monotonically increasing and concave in input prices, are not imposed a
priori and need to be tested. Violation of concavity is particularly problematic as it
may imply that the cost function does not have the structural relationship postulated by
duality theory to the underlying technological parameters of production (Fuss et aI,
1978, p266). Parameter estimates, and the associated estimates of scale and
substitution, which satisfy rigid statistical criteria but which violate .the concavity
condition should, therefore, be treated with considerable circumspection. Clearly
emphasis ought to be afforded theoretical, rather than statistical, criteria.
Four alternative econometric methods are available for estimating the parameters of
the Translog: the iterative Zellner efficient estimator (IZEF) which is applicable for
seemingly unrelated regression (SUR) models in general (Zellner, 1962); the IZEF
modified to take serial correlation in the component equations of the system into
consideration (IZEF(AR» (Greene, 1993, p498); the iterative three-stages least
squares (13 SLS) estimator which is appropriate jf regressors of the SUR system are
endogenous (Zellner and Theil, 1962) and 13SLS modified to take serial correlation in
the component equations into account (13SLS (AR»(Kmenta, 1986, p708). Which of
the four alternative regressors will yield the statistically best estimates requires
knowledge of the nature of regressors (whether they are exogenous or endogenous)
and the relationship among errors of the component equations.
Determining the nature of the regressors is required for determining which broad class
of estimator, ZEF or 3SLS, is most appropriate. Should any of the regressors4 be
endogenous the appropriate class of estimaptors, yielding consistent and asymptotically
efficient estimates (Kennedy, 1985, p136) are 3SLS estimators. Employing a ZEF
estimator in this context, while efficient, would not, however, be consistent (Kmenta,
1986, p718). Should all regressors be exogenous ZEF estimators would be
appropriate, yielding consistent and efficient parameter estimates. Although consistent,
3SLS estimators would, in this case, be inefficient relative to ZEF estimators (Berndt,
1991, p379). While tests do exist for establishing the nature of regressors in systems of
equations, and therefore, which broad class of regressors would be appropriate as
argued in Chapter 4 these tests are not suitable in this context. As a consequence it is
4 Regressors in the Translog model are combinations of input prices, output and time. Time is always
exogenous, while the relatively disaggregated level of this analysis permits the reasonable assumption
that input prices are exogenous to the industries under analysis. The nature of output, and therefore all
regressors which are functions of output, is, ·however, debatable.
94
not possible to determine statistically which of the two broad classes of estimator is
appropriate. Both classes were therefore used5 and the results generated compared.
While it is not possible to determine which class of estimator is most appropriate the
6
related question of the appropriate error structure can be established. Conventional
Durbin-Watson diagnostic testing for the presence of first-order autocorrelation among
the errors of the component equations is available to determine whether the
autocorrelation modified version (IZEF (AR) and I3SLS (AR» of the two classes of
estimator would be appropriate. Post estimation LR comparisons of estimates
emerging from different assumptions regarding error structure are available to
determine whether autocorrelation specifications (implied in the adoption of either the
IZEF(AR) or the I3SLS(AR) estimators) are statistically significantly different from
the nonautocorrelated error specifications.
LR testing is not only useful for determining which error structure is valid, it is also
useful for deriving the most appropriate estimates of technology. The modelling
flexibility of the Translog provides the opportunity to determine statistically whether
the commonly imposed restrictive maintained hypotheses of homotheticity or
homogeneity, and associated estimates of scale and substitution, are appropriate.
Homotheticity and homogeneity can be imposed by constraining parameters and hence
their validity determined by LR comparisons of the constrained and unconstrained
specifications.
For each estimator used here five different specifications of the Translog system were
estimated7 . The first specification (Model 1) which was estimated is the specification
augmented to simultaneously estimate scale, substitution and technical change effects,
as proposed by Christensen (1977, cited in Greene, 1983, pI27). The second
specification (Model 2) is the specification employed by Christensen and Greene
(1976) which assumes variable economies of scale in a production structure which is
non-homothetic. The third specification (Model 3) allows economies of scale to vary
5 3SLS estimators are consistent wherever they are applied. It could be argued that they alone ought to
be employed given that the only loss associated with their inappropriate use is a loss of efficiency.
Indeed, efficiency losses give rise to more conservative statistical hypothesis testing which could be
deemed a virtue rather than a cost. Despite this argument, and the additional computational burden,
both classes of estimator were employed here. The principle motivation for using both classes of
estimator emerges from two areas of interest associated with concavity: Firstly, whether concavity is
invariant to different assumptions regarding the nature of regressors, and therefore the class of
estimator used. Secondly, to establish empirically the theoretical assertion that using estimates which
are non-concave are potentially misleading.
6 As the data used here are time-series it is assumed that error variances are homoscedastic.
7 The form (for three inputs) of the different specifications analysed here are presented in Appendix
5.1.
95
with output but assumes that the dual production function is homothetic. The fourth
specification (Model 4) imposes homogeneity (and hence homotheticity) on the
underlying technology thus preventing economies of scale from varying with output.
The fifth specification, (Model 5) assumes a Cobb-Douglas technology with variable
returns to scale.
The maintained hypotheses embodied in Model's 2 through 5 are introduced into the
Translog system by imposing parameter restrictions on Model 1. The validity of these
maintained hypotheses can therefore be tested using nested hypothesis tests, such as
the LR test, of the validity of parameter constraints. This approach was adopted here.
In addition to testing the appropriate specification of the Model, each of the
specifications estimated was tested (using a LR test) to establish whether the emerging
estimate of economies of scale was significantly different from unitl·
The above testing procedures were employed for both classes of estimator using the
same general procedure. Parameter estimates were first obtained assuming a priori that
the component equations of the SUR system fulfilled the assumptions of the CLR
model. Those estimates were then tested to determine which of the five alternative
specifications of the model (see Appendix 5. 1) was statistically valid, and whether the
assumption that the errors of the component equations were uncorrelated across
observations was valid. Where autocorrelation was detected, autocorrellation-
augmented estimators were employed. Parameter estimates derived from the
autocorrelation augmented estimators were analysed to determine the correct
specification of the model: both in terms of comparisons among the different
specifications which assume autocorrelated errors and between specifications which
assume non-autocorrelated and autocorrelated errors. Testing of the hypothesis that
economies of scale are constant were then performed on all estimated specifications.
Finally testing was undertaken to determine whether or not the duality conditions of
monotonicity and concavity in input prices were fulfilled for different parameter
estimates for the sample under analysis.
A persistent feature of the parameter estimates generated here is the violation of the
concavity requirement. Given that non-concavity threatens the validity of the duality
between cost and production functions few theoretically sound estimates of technology
emerged from this analysis. A number of useful conclusions do, nevertheless, emerge.
The data clearly suggest that flexible, non-homothetic specifications such as the
8 Constant returns to scale can be imposed on each of the models by imposing constraints on a
number of parameters.
96
Translog are more appropriate than functions which are homothetic, such as the Cobb-
Douglas. Moreover, not only is homotheticity found to be statistically inappropriate
but it clearly biases estimates of scale downwards. Finally the data provide some
support for the argument that economies of scale are greater in wage goods industries
than in luxury industries.
The data used in this analysis9 are presented in Appendix 4. The 'raw data lO' and
certain manipulations of the data were used here as part of an a priori attempt to form
expectations regarding the magnitude and sign of elasticity's of substitution and
economies of scale. A similar regime of a priori analysis was conducted for each of the
sectors. Four 'devices' were used to form expectations about elasticity's of
substitution: changes in the cost shares of inputs; both plots and descriptive statistics
(means and standard deviations) of input-output coefficients11 and relative prices of the
three inputs l2 ; and the correlation matrix of the different input-output coefficients and
relative input prices. In order to form expectations about economies of scale the trend
of real average costsl 3 over the period was analysed.
•
5.2 ELECTRICAL APPLIANCES AND HOUSEHOLD GOODS (ISIC 3833)
5.2.1 INTRODUCTION and A PR/OR/DATA ANALYSIS
Of the three sectors analysed here the estimates emerging for the appliances sector are
most useful. Firstly, these estimates fulfil the c()!lcayj!y condition ITlost()ften - allowing
for the most complete set of results. The data indicate that all three inputs substitute
each other in the production of Appliances, with the degree of s.!lbstitlJ13.bil.ily.gr~31e.st
between c~ital ~l1dlabour. Conclusions regarding the magnitude and trend of scale
depend on assumptions made about the nature of output. The data do, however,
suggest that not only is imposing homotheticity statistically invalid, but in addition
9 Total Costs C; Cost shares of Capital SK ; Labour SL and Materials SM; Input prices of Capital
PK ; Labour PL and Materials PM and output y. Estimation of the model requires manipulation of
these series. In particular homogeneity in prices requires that total costs and the input prices retained
in the system are divided by the price of the factor whose cost-share is dropped from the system; and
the natural logarithm of the variables needs to be found in accordance with the specification of the
model.
10 That is, the non-logarithmic and non-normalised form.
" Input-output coefficients, which are the ratio of the quantity of different factors to the quantity of
output, were calculated by defining the quantities of factors as the constant Rand expenditure on the
different factors and the quantity of output as the constant Rand value of output.
J2 Relative prices are defined as the ratio of the price index of each input to the price index of output
as calculated by the I.D.C (1992).
13 Real average costs is nominal average costs deflated using a price index of output, where nominal
average costs are calculated as the ratio of the current value of total costs to the constant rand value of
output.
97
imposing this hypothesis a priori biases estimates of the magnitude of scale downward.
The second useful feature of these estimates also concerns concavity: most IZEF and
I3SLS results are concave while most IZEF (AR) and I3SLS (AR) estimates non-
concave. A consequence of this phenomenon is that for most specifications of the
system a concave and a non-concave set of estimates are available for comparison.
While differences are not systematic concave es~imates of scale and substitution do
differ from non-concave estimates. Before turning to a more detailed consideration of
these conclusions the a priori analysis is presented.
The appliance sector is characterised by a cost structure where the relative importance
of the three inputs (capital, labour and materials) in total costs has remained the same
(materials contributing the most, and capital the least, to total cost) over the period
1972-1990. A slight fluctuation in relative shares between 1978 and 1990 when
relative expenditure on materials increased at the expense of labour (and after 1986,
capital) appears to indicate that materials may have substituted both capital and labour,
which complement one another (see Figure 5.2.1).
Figure 5.2.1 Cost sbares of Capital, Labour and Materials for ISIC 3833 (1972-1990)
IiJ Cost share

• Cost share of Labour
1.2 • Cost share of Capital
0.8
0.6
0.4
0.2
o
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
This conclusion is supported by trends of input-output coefficients (see Figures A5.2.1

- A5.2.3, Appendix 5.2.1) and simple correlation's between input-output coefficients
and relative input prices. While the input-output coefficient of materials fluctuates in a
small band around an overall upward trend the input-output coefficients of capital and
labour both display general downward trends. The similarity in the capital and labour
98
trends 14 suggests that the two inputs complement one another in the sector. Probable
complementarity and susbtitutability between inputs can also be inferred from the sign
of the correlation between input-output coefficients of inputs and the sign of cross
1S
correlation's between input-output coefficients and relative prices . The data
presented in Table 5.2.1 suggest that materials is a substitute for both capital and
labour and that capital and labour are complements in the production of appliances.
The positive correlation between the input-output coefficient of materials with the
relative prices of both capital and labour is a possible indication that materials are a
substitute for both labour and capital. This conclusion is supported by the negative
correlation's between the input-output coefficient of materials and those of capital and
labour respectively1617 . The substitutability relationship would, however, appear to be
asymmetrical. The input-output coefficients of both labour and capital are negatively
correlated with the relative price of materials.
Complementarity between capital and labour can be inferred from the negative
correlation's between the input-output coefficients of capital and the relative prices of
labour and capital respectively. The possibility of such complementarity is supported
by the strong (0.94) positive correlation between the capital and labour input-output
coefficients.
Table 5.2.1 Correlation matrix of Relative orices and Inout-outout coefficients (ISIC 3833)
Relative Relative Relative Input- Input- Input-
price of price of price of output output output
Capital Labour Materials coefficient coefficient coefficient
of Capital of Labour of
Materials
Relative price of 1.0000
Capital
Relative price of 0.40547 1.0000
Labour
14 This assertion of similarity in the trends is supported by the ratio of standard deviation to mean of
the two input-output coefficients. For capital the ratio is 0.385 and for labour 0.333 (see Table A5.2.2
Appendix 5.2).
15 A positive correlation between the input-output coefficient of an input x with the relative price of
another input y suggests, ceteris paribus, that as the relative price of y rose more of variable x was
employed. Were that to occur x would be regarded as a substitute for y. A corollary to this conclusion
is that complementarity between inputs would be reflected in a negative correlation between input-
output coefficients and relative input prices.
16 A negative correlation between two input-output coefficients would imply that when the relative use
of one input increases the relative use of the other decreases.
17 The suggestion that materials are a substitute for both capital and labour may be explained by
evidence that materials is an essential input. The correlation between the relative price of materials
and the input-output coefficient of materials is positive. This suggests that demand for materials is
invariant to the price of materials and, therefore, that materials are an essential input. This conclusion
would be strengthened if the correlation was near to zero, which it is not.
99
Relative price of 0.84800 0.71051 1.0000

Materials
Input-output -0.83442 -0.43509 -0.73563 1.0000
coefficient of
Capital
Input-output -0.71333 -0.36033 -0.57128 0.94253 1.0000
coefficient of
Labour
Input-output 0.55160 0.28571 0.38985 -0.42224 -0.50724 1.0000
coefficient of
Materials
The final objective of undertaking an a priori analysis is to establish the relationship

between real average costs and output and hence obtain some indication of the
magnitude of economies of scale. Figure 5.2.2, which plots real average cost against
time, indicates that prior to 1978 the industry experienced a period of increasing real
average costs (suggesting diseconomies of scale), while between 1978 and 1986 the
industry experienced, in general, decreasing average costs (suggesting economies of
scale).
Figure 5.2.2 Real Average Costs over the period 1972-1990 (lSIC 3833).
0.18
0.16
0.14
0.12
L------
0.1
0.08
0.06
0.04
0.02
O+---+--+----t--t----I;--.........- -+----+--+----+_-+---+_-+----t_--+---f
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
5.2.2 REGRESSION ANALYSIS
An important consideration for establishing the properties of estimates of the Translog

is the relationship among errors. While both the IZEF and 13 SLS 18 procedures
explicitly take inter equation error correlation's into account, providing more efficient
estimators than OLS or 2SLS respectively, both procedures are premised on the
assumption that the errors of each individual equation are normally and independently
18 Parameter estimates are presented in Tables A5.5.3.1 and A5.5.3.2 in Appendix 5.5
100
distributed. Should the errors of each equation not be independently distributed, these
estimators would not be efficient. The presence of serial correlation can be detected by
calculating the single equation Durbin-Watson (D-W) statistic for the errors
corresponding to the component equations of the SUR system l9 . Computed and
critical D-W statistics are presented in Tables A5.5.2.1 and Table A5.5.2.2 - Appendix
5.5. 20 . Following Theil and Nagar21 (1961) for the case where the test statistic falls in
the region of indecision, and the usual Durbin-Watson conclusion for cases where the
test statistic falls below the lower limit of the critical value, first-order autocorrelation
would appear to exist in all the component equations of the different specifications of
the model, irrespective of whether IZEF or 13 SLS is used.
The implication emerging from the Durbin-Watson testing is that IZEF (AR) and
I3SLS (AR) 22 estimates are more appropriate than IZEF and I3SLS respectively.
Estimates for Model 1 using I3SLS(AR) could not, however, be obtained as estimates
did not converge on iteration23 . Whether the results generated by adopting this
alternative error specification are statistically different to those obtained when the
errors of the component equations are specified to be non-correlated can be tested
using LR tests. Table 5.2.2 and 5.2.3 present computed LR test statistics, for the five
specifications of the model, of the null hypothesis that the error specification assumed
for the IZEF/I3SLS estimators is valid. For Models I through 4 the number of degrees
of freedom is 3, while for the 5th model the number of degrees of freedom is 124 • For
both classes of estimator for all specifications apart from Model 5, the computed LR
19The different methods which need to be adopted when testing for autocorrelation ofIZEF and
I3SLS estimators is discussed in Chapter 4. An important difference between the two tests is the
calculation of degrees of freedom.
20 No tests are performed on the cost share equations of the Cobb-Douglas form of the Translog
(ModelS) as that function has no regressors, only a constrained constant.
21 Theil and Nagar (1961) have shown that where regressors are changing slowly the upper
distribution of the D-W statistic is the appropriate distribution. Kennedy (1985, p106) argues that it is
likely that in economic time series that the regressors would be changing slowly.
22 Parameter estimates of the different specifications using IZEF (AR) and I3SLS (AR) are presented
in Tables A5.5.3.3 and A5.5.3.4, respectively.
23 By imposing an autoregressive error structure on this Model, the number of parameters being
estimated rises from 14 to 17 (each equation of the system being estimated was presumed to posses a
different coefficient of autocorrelation which meant that three different autocorrelation coefficients
needed to be estimated). As one observation is lost when employing the autoregressive estimator, the
number of degrees of freedom for estimating this specification is only one - hence the instability of the
parameter estimates emerging. Although 3SLS(AR) estimates could be obtained, the lack of
invariance of such estimates to which share equation was dropped from the system precluded there use
here.
24 The number of degrees of freedom is determined by the number of constraints imposed on the
unrestricted versioQ of the model to generate the restricted form of the model. Where all three
equations of the Translog system were modified to take autocorrelation into account, the number of
restrictions imposed on the unconstrained model (IZEF(AR» to generate the constrained model
th
(IZEF) would be three. For the 5 specification, only the cost equation is modified and hence only one
constraint is imposed to generate the constrained model.
101
statistic is greater than the critical chi-squared statistic leading to a rejection of the null
hypothesis that the non-autocorrelation augmented estimator is correct. Clearly, aside
from specification five, more efficient estimates will emerge if an autocorrelated error
structure is adopted, irrespective of the nature of the regressors.
Table 5.2.2 LR test statistics for comoarison of IZEF and IZEF(AR) estimators (lSIC 3833)
Model 1 2 3 4 5
Likelihood Ratio Statistic 36.707 17.645 31.129 30.919 -11.711
Critical Chi-squared statistic 7.81 7.81 7.81 7.81 3.84
Table 5.2.3 LR test statistics for comparison of I3SLS and I3SLS(AR) estimators (ISIC 3833)
Model 2 3 4 5
Likelihood Ratio Statistic 16.815 30.783 30.623 -12.591
Critical Chi-squared statistic 7.81 7.81 7.81 3.841
An interesting conclusion emerging from the all four sets of parameter estimates is a
statistical preference for the most general functional specifications and hence a
rejection of the hypotheses of homotheticity and homogeneity. The validity of the
alternative specifications relative to the most general model estimated by each
25
estimator was tested by performing LR tests - results of which are presented in Table
5.2.4. For the IZEF estimates testing the null hypothesis that the Translog model
should not include time, in any form, as a regressor or as a multiple of another
regressor (Model 2) against the alternative hypothesis that time ought to be included in
the Translog as specified by Model 1 the LR test statistic is 17. 1226 . The appropriate
chi-squared statistic for 5 degrees of freedom 27 at the 0.05 level of significance is
11.07 (Gujarati, 1988, p685). As the test statistic is greater than the appropriate chi-
squared statistic the null hypothesis, that Model 2 is a valid specification, cannot be
accepted with 95% confidence. Similarly the null-hypotheses implied by any of the
alternative models cannot be accepted for any of the estimators suggesting that the
most flexible, general representation of technology is appropriate for these data. This
result potentially damages analyses of these data using functions, such as the Cobb-
Douglas or CES, which impose the hypotheses of homotheticity and homogeneity a
priori.
25 In the cases ofIZEF, IZEF (AR) and I3SLS Model 1, while in the case ofI3SLS (AR) Model 2.
26 LR = -N(lnISucl-InIScl)= -19 [1n(2.49E-11)-ln(6. 13E-11»)=17. 12. See Chapter 4 for more
detail regarding LR testing.

27 The number of degrees of freedom equals the number of restrictions being tested. Five parameters
appearing in Model 1 are constrained to zero when estimating Model 2. A sixth appears to be
.missing' from the results table rMt. That parameter is, due to the linear homogeneity in input prices
constraints discussed in Chapter 3, a linear combination of the other parameters and is not estimated
directly.
102
Table 5.2.4. LR test statistics and appropriate chi-square statistics for ZEF and 3SLS estimates
of ISIC (3833)
Null Hypothesis Critical Computed Computed Computed Critical Computed
i L.R. L.R. L.R. i L.R.
statistic
statistic statistic statistic statistic statistic
and dffor IZEF IZEF I3SLS and dffor I3SLS
IZEF, estimates (AR) estimates I3SLS (AR)
IZEF estimates (AR) estimates
(AR) and estimates
I3SLS
estimates
Non-homothetic 11.07 (5) 17.12 37.24 17.14
Homotheticity 14.07 (7) 46.78 52.66 46.22 5.99 (2) 13.59
Homogeneity in Output 15.51 (8) 47.73 53.84 47.19 7.81 (3) 14.66
V.R.T.S. Cobb-Douglas 19.7 (11) 78.62 129.72 78.51 12.59 (6) 87.55
Note: I. For IZEF, IZEF (AR) and I3SLS the alternate hypothesis is that the technology augmented
form is valid. For I3SLS (AR) estimates the alternate hypothesis is that the non-homothetic
form is valid.
2. Numbers in parenthesis appearing after i
indicate Degrees of Freedom (df) (at the 5% level
of significance) which are determined by the number of parameter constraints required to
. obtain the specification of the different null hypothesis from the alternative specification.
As argued in the introduction to this chapter a crucial consideration in deriving

estimates of scale and substitution is the theoretical tractability of the estimates. The
Translog is neither monotonically increasing nor concave in input prices a priori. Non-
concave estimates are potentially meaningless as the assumptions upon which the dual
relationship is premised are violated. The appropriate procedures for testing
monotonicity and concavity have been discussed in Chapter 328 , and the data relevant
to the testing of both conditions are presented in Appendix 5.5 (Tables AS.S.4.1 -
A5.S.4.30). Although the monotonicity requirements are fulfilled for all sets of
parameter estimates an interesting anomaly regarding the fulfilment of the concavity
requirement for different estimators emerges. The concavity condition is fulfilled for all
29
specifications , over the whole sample, when the IZEF and 13SLS estimators are
30
used , but only intermittently when the auto-correlation augmented estimators are
used. Concavity is only fulfilled for a few observations in two of the specifications
28Monotonicity is fulfilled if the fitted cost shares are positive at every observation. Concavity is
fulfilled if the matrix [ A] of AESs is negative semi-definite at every observation. Negative semi-
definiteness emerges when IA,I::; OJAzl ~ O, ... JAnl::; 0 if n is odd and IAnl ~ 0 if n is even.
Violation of one of these conditions will mean that the matrix of AESs is not negative semi-definite
and that concavity has been violated. The precise form of the concavity test in the three input context
is presented in Appendix 3.1.
29 There is no need to test the concavity condition for Model 5 as it fulfilled a priori.
30 A requirement of concavity is that the determinant of the 3 x 3 matrix of AESs be negative or zero.
If the value of that determinant is positive but very small it will still be acceptable as deviations from
zero could be due to rounding errors
103
when the IZEF (AR) is used 31 . When I3SLS (AR) is used concavity is fulfilled for
Model 2 over the whole sample but not at all for any of the other specifications.
These concavity violations present an important choice regarding the ZEF estimates.
Statistical considerations reveal a preference for the IZEF (AR) estimates in general.
The concavity violation suggests that these estimates may, however, be theoretically
meaningless. A choice, therefore, exists between the efficiency of the IZEF (AR)
estimates and the theoretical validity of the IZEF estimates. Given that the cost of
inefficiency is merely more conservative hypothesis testing (which may even be
deemed desirable), while non-concave estimates are potentially meaningless, the IZEF
estimates are preferred. Should the ZEF class of estimator be valid, the appropriate
estimates of substitution and scale would, therefore, be those emerging from the IZEF
estimates of Model 132 . The same dilemma does not exist for the 3SLS set of
estimates. Combining specification and autocorrelation tests the I3SLS (AR) estimates
of Model 2 are preferred33 , and these estimates fulfil concavity over the whole sample.
Having determined which sets of parameter estimates are appropriate derived estimates
of substitution, as embodied in proper AESs, and economies of scale can be obtained.
The analysis of both features is concerned with both the magnitude of estimates and
the impact of concavity violation on estimates. The analysis of scale has a third
objective: assessing the impact of different maintained hypotheses on modelling scale.
5.2.2.1 ELASTICITY'S OF SUBSTITUTION - ISIC 3833
Whether the ZEF or 3SLS class of estimator is appropriate, the same conclusion
34
regarding the relationship among inputs emerges . Contrary to the a priori
expectation that cCWital. and labou~are complementary inputs proper AESs indicate
that all factors are S,UJ2g!tutes, in the production of appliances35 . This relationship
among inputs is, however, quantitatively different if different assumptions about the
nature of output are adopted. If output is assumed to be endogenous the degree of
substitutability between capital and labour is considerably greater than if output is
assumed to be exogenous. Such comparisons may, however, be invalid given that
31 For Model 2 concavity is fulfilled for 1972 and 1973 and 1977 and 1978. For Model 3 concavity is
fulfilled for 1988.
32 LR testing reveals a preference for this specification.
33 LR tests indicate a preference for more general specifications and estimators which assume an
autocorrelated error structure. Because no I3SLS (AR) estimates of Model I emerge the statistically
most desirable estimates are the I3SLS (AR) estimates of Model 2.
~: Deriv~d estimates of proper ~Ss for.all model~ are presented in Tables. A5.5.4.1 to A5.5.4.30.
There IS, however, one exception to this conclUSIOn. The I3SLS (AR) estimates of Model 2 indicate
that in 1990 labour and materials are complements.
104
different model (and less importantly error) specifications are valid for the two
6
assumptions regarding the nature of outpue .
Patterns of AESs emerging here provide empirical support for the theoretical assertion
that non-concavity may effect estimates of technology. Indeed, a comparison of the
proper AESs emerging from concave and non-concave sets of estimates of both classes
of estimator suggests that reliance on generally non-concave sets of parameters for
estimates of technology is potentially misleading. While for all concave parameter
estimates all inputs are (for the most part) substitutes37 the same does not hold for the
generally non-concave parameter estimates38 . No systematic relationship exists
between the violation of concavity and the pattern of AESs emerging39 .
5.2.2.1 ECONOMIES OF SCALE - ISIC 3833
As argued in Chapter 4 the statistical significance of estimates of scale emerging from

Translog estimates can be determined using LR tests of the validity of appropriate
parameter constraints. Results, presented in Table 5.2.5, reveal that for theoretically
valid estimates only Model 2 and Model 5 generate estimates of scale which suggest
non-constant returns to scale.
. Table 5.2.5 Statistical Si2l1ificance of Estimates of Scale (I SIC 3833)

Critical Computed Computed Computed Computed
Ch- LR-IZEF LR-IZEF LR- LR-
Squared estimates (AR) I3SLS I3SLS
Statistic estimates estimates (AR)
estimates
Model 1 (5) 11.071 10.876 305.555 11.01 --
Model 2 (4) 9.488 67.475 23.232 68.53 46.08
Model 3 (2) 5.991 0.114 2.301 0.391 2.582
Model 4 (1) 3.841 1.236 0.112 0.563 0.117
Model 5 (1) 3.841 9.270 12.029 10.787 18.975
Note: 1. Numbers in parenthesis after the model number are the number of degrees offreedom
2. Source of critical chi-squared values: Guiarati (1988, 0685)
36 If output is exogenous IZEF estimates of Model I are valid, however is output is endogenous 13 SLS
(AR) estimates of Model 2 are appropriate.
37 The exception are the I3SLS (AR) estimates of Model 2, which show that the relationship between
labour and materials may have changed to one of complementarity in 1990.
38 For the IZEF (AR) estimates of Model I capital and labour are everywhere complements, capital
and materials are substitutes only between 1972 and 1978 and in 1984 and 1985. The signs of the
proper AESs for the other non-concave estimates (IZEF (AR) estimates of Models 2, 3 and 4 and the
I3SLS (AR) estimates of Models 3 and 4) follows the same pattern - capital and labour and labour and
materials are everywhere substitutes while capital and materials are everywhere complements.
39 Compare, for example, the IZEF (AR) and IZEF estimates of Model 2. The IZEF estimates are
everywhere concave and produce a positive capita1:materials AS over the whole sample. The IZEF
(AR) estimates are concave in 1972, 1973, 1977 and 1978 yet produce proper capital:materials AESs
which are everywhere negative.
105
A comparison of the mean 40 magnitude of economies of scale suggests that imposing

the assumption of homotheticity as a maintained hypothesis produces lower estimates
of scale. Table 5.2.6 reveals that models 1 and 2 which are non-homothetic, produce
larger mean estimates of scale than the homothetic specifications suggesting the
conclusion that imposing homotheicity biases estimates of scale downwards. This
conclusion is, however, contradicted by the tests of the constant returns to scale
hypothesis which reveal that for model 1 estimates of scale are statistically not different
to unity.
Table 5.2.6 Mean economies of scale ISIC (3833)

IZEF IZEF (AR) I3SLS I3SLS (AR)
Model 1 1.331 1.620 1.332 --
Model 2 2.046 1.537 2.133 2.104
Model 3 1.005 1.011 0.988 1.017
Model 4 1.019 1.010 1.013 1.010
Model 5 1.083 1.36 1.073 1.504
Note: I.For Models 4 and 5 economies of scale are calculated as &(x) = 1/ay .
An analysis of the trends of the estimates of scale which vary with output (Models 1, 2
and 3) reveals two interesting results. Firstly, the data suggest that non-concave
estimates differ from concave estimates - albeit in a non-systematic manner. The
second result is that depending on the assumption made about the nature of regressors
different conclusions emerge regarding the trend of scale over the period.
A visual inspection of the trends of estimates of scale which vary with output41
indicate differences in the volatility of concave and non-concave estimates of the same
model. The observation regarding volatility is supported by comparisons of standard
deviations. For models 1 and 3, concave estimates42 produce estimates of scale which
are more stable over the sample, while in the case of model 2, the concave estimates43
are more volatile. An implication of this is that estimates of scale are not robust to the
fulfilment of concavity. Indeed, given that both features of technology analysed display
dissimilarities between the concave and non-concave sets of results care should be
40 Estimates of scale emerging from the nonhomogenous specifications (Models 1, 2 and 3) vary with
output. In order for comparisons to be drawn between these estimates and the estimates emerging
from the homogenous specifications where scale is invariant to the level of output means of the
nonhomogenous estimates were calculated. Trends are analysed below.
41 See figures A5.5.6.5 to A5.5.6.7.
42 For both models the IZEF and I3SLS estimates are concave over the whole sample.
43 For this specification the IZEF (AR) estimates are non-eoncave.
106
taken in drawing inferences from estimates of features of technology emerging where

concavity is violated.
Table 5t,2• 7 Standard deviations of scale estimates ISle 3833

Model 1 0.0855 0.4446 0.1247 --
Model 2 0.1064 0.0747 0.1387 0.3670
Model 3 0.0001 0.0498 0.0046 0.0532
Depending on the assumptions made about the nature of regressors, different

conclusions emerge regarding trends in scale. Assuming that all regressors are
exogenous, the ZEF class of estimates are valid and statistical and theoretical
considerations suggest that the most appropriate estimates of scale are those emerging
from Model 1. Not only are these estimates statistically not significantly different from
unity but they fail to display any obvious trend 44 .
Should output be endogenous and the 3SLS class valid, as argued above, the
appropriate estimates of the characteristics of technology are those emerging from the
I3SLS (AR) estimates of Model 2. These estimates indicate that while economies of
scale do prevail, as witnessed by a mean greater than 2, they have decreased in
magnitude over the sample (see figure 5.2.3 below), yielding the conclusion that
should output be endogenous, the statistically and theoretically valid conclusion
emerging is that economies of scale exist but are decreasing as output increases in the
appliances sector. A disturbing feature of the results is that should one discount
efficiency and employ the 13 SLS estimates of either model 1 or 2 the conclusion that
economies of scale are diminishing is rejected (see Figures A5. 5.6.5 and A5. 5.6.6).
Figure 5.2.3 Economies of Scale - ISle 3833
44 The IZEF results display an interesting feature: Model 2 generates estimates of scale which are
greater than the estimates emerging from Model 1. This result is, however, intuitively reasonable as
changes in costs due to technical change are likely to appear as scale effects where technical change is
not explicitly modelled.
107
3.5
2.5
1.5
0.5
0+---+-+--+-+--+-+--+-1---+---+-+---+-+--+-+--+-+--1
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
5.2.3. CONCLUSION - ISIC 3833
A number of important conclusions can be drawn from the preceding analysis. The
data clearly suggest, contrary to a priori expectations, that all three inputs are
substitutes for each other in the production' of appliances, with the degree of
substitutability greatest between capital and labour. Should output be exogenous, the
sector appears not to experience scale advantages. If, however, output is endogenous
the data suggest the presence of economies which appear to have decreased in size
over the sample. A further result regarding scale regards the impact of different
maintained hypotheses on the magnitude of estimates of scale: there is evidence to
suggest that not only is imposing homotheticity statistically invalid, but in addition
imposing this hypothesis a priori biases estimates of the magnitude of scale downward.
The final conclusion emerging from an analysis of the Appliances sector regards the
effect that concavity violations have on estimates of the features of technology. While
no systematic differences appear between estimates which are concave and those which
are non-concave - the features of technology emerging from non-concave estimates
differ from those emerging from the concave results.
5.3 FURNITURE (ISIC 3320)
5.3.1 INTRODUCTION and A PRIORI DATA ANALYSIS- ISIC 3320
Of the three sectors analysed in this dissertation the furniture sector estimates are least
useful. All four estimators generate parameter estimates which violate the concavity
condition at every point in the sample. Non-concavity, as argued above, may imply that
the cost function does not have the structural relationship postulated by duality theory
108
to the underlying technological parameters of production thus rendering parameter

estimates untenable (Fuss et ai, 1978, p266). Indeed, the appliance sector results
provide empirical support for this theoretical assertion that non-concave estimates are
unreliable. Clearly, the results for the furniture sector ought to be treated with caution.
Prior to performing the regression analysis the furniture data was analysed in the same
manner as the appliance sector data in an attempt to illicit expectations about scale and
substitution. That analysis suggests that the sector experienced slightly diseconomies
over the period under analysis. Furthermore, the data suggest complementarity
between capital and both materials and labour and that labour may be an essential
input.
The relative cost structure faced by the furniture sector is similar to that of the
appliances sector: materials share in total costs is greatest while capital's share is the
smallest (see Figure 5.3.1). While shifts in relative costs over the period under analysis,
do not reveal any obvious patterns of complementarity/substitutability there is some
4s
evidence to suggest complementarity between capital and labour. This suggestion is
supported by trends in input-output coefficients.
Figure 5.3.1 Cost shares of Capital, Labour and Materials for ISIC 3320 (1972-1990)
Ii1J Cost share of Materials

1.2
• Cost share of Capital
1
0.8
0.6
0.4
0.2
o
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
The input-output coefficients of both capital and labour have followed the general
trend of their respective cost shares over the period under analysis (Figures A.5.2.2.1
and A5.2.2.2 - Appendix 5.2), decreasing between 1974 and 1979 and increasing after
1980. More relevant for this analysis though, is the distinct similarity in the trend of
45 Between 1974 and 1981 and again between 1982 and 1986 the cost shares of capital and labour
changed in the same manner suggesting complementarity.
109
these coefficients which suggests the complementarity of capital and labour. The
similarity in the trends is underlined by the fact that for both series the ratio of standard
deviation to mean is 0.137 (see Table A5.2.2.2)
The above conclusion is supported by the correlation matrix of relative prices and
input-output coefficients (Table 5.3.1) which shows a positive correlation (0.638)
between the capital and labour input-output coefficients. While this is evidence of
complementarity any such conclusion is potentially damaged by the positive
relationship between the labour input-output coefficient and the relative price of
capital. The matrix does, nevertheless, provide a clear conclusion regarding the
relationship between capital and materials. The negative correlation between the input-
output coefficient of capital (materials) and the relative price of materials (capital)
together with the positive correlation between the capital and materials input-output
coefficients indicate complementarity between capital and materials. The correlation
matrix suggests a further conclusion: the positive correlation between the relative price
of labour and its input-output coefficients suggests that labour is an essential input.
This inference is strengthened by the fact that the correlation is near to zero (0.0269).
Table 5.3.1 Correlation matrix of Relative orices and Input-output coefficients (ISIC 3320)
Materials
Capital
Labour
Materials
Input-output -0.68IE-OI -0.30511 -0.47790 1.0000
coefficient of
Capital
Input-output 0.65548 0.269E-Ol 0.31411 0.63804 1.0000
coefficient of
Labour
Input-output -0.33967 -0.64542 -0.32032 0.26850 0.16lE-Ol 1.0000
coefficient of
Materials
Turning to the possible presence of scale, a plot of real average costs over time (Figure
5.3.2) indicates a general increase in real average costs over the period under analysis,
suggesting that diseconomies prevail in the sector.
Figure 5.3.2 Real Average Costs over the period 1972-1990 (ISIC 3320)
110
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
O+--+-----1I---+---+-+--+-----1I---+---+-+--1---f-+--+--f---1---f--i
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
A disturbing feature of the regression results emerging from the furniture sector data is
the persistent violation of concavity. Indeed for all models and the different estimators
labour and materials both have positive own AES 46 (see Tables A5.4.4.1 to
A5.4.4.28). A consequence of these results is that no confidence can be placed on
estimates of the different features of technology emerging from these sets of results.
LR tests of the appropriate functional form for the different estimators generate the
same results for the IZEF, IZEF (AR) and I3SLS estimators as those obtained for the
appliance sector - that the more general specification is preferred (see Table A5.4. 1.1 -
Appendix 5.4). For both the IZEF and I3SLS estimates, the data indicate that the
technology augmented specification (Model 1) is preferred. Despite manipulation of
the convergence criterion, no estimates for Model 2 could be generated using the IZEF
(AR) estimator. However, for that estimator LR tests reveal a preference for Model 1
over Models 3,4 and 5. The I3SLS (AR) results contradict the usual preference for a
more general specification. As with the Appliances sector no estimates could be
generated for Model 1 using I3SLS (AR). A comparison of the estimates emerging
from Models 2, 3, 4 and 5 revealed that Model 2 was preferred only to Model 5.
Although Durbin-Watson testing of the errors emerging from the IZEF and 13 SLS
estimates suggested the presence of autocorrelation in all specifications estimated (see
Table A5.4.2.1 and Table A5.4.2.2), prompting the use of the IZEF (AR) and I3SLS
(AR) estimators respectively, a comparison of the estimates generated when
46 This is not the only violation of the concavity requirements, in most case at least two of the
determinants of the 2 x 2 matrixes of AESs are negative while concavity requires that they be positive.
For details on the requirements of concavity see Appendix 3.1.
III
autocorrelation was not accounted for to those estimates accounting for

autocorrelation contradict this finding. LR testing of the null hypothesis that IZEF
estimates were preferred to IZEF (AR) estimates (see Table A5.4.1.2) showed a
rejection of the null hypothesis for Models 3 and 4. A similar test comparing I3SLS
and I3SLS (AR) estimators (see Table A5.4.1.3) indicate that the I3SLS (AR)
estimates are preferred only for Model 4. Combining the results of this set of LR tests
with results of the model specification tests indicates that if output is exogenous
47
(endogenous) the IZEF (I3SLS) estimates of Model 1 are preferred .
A comparison of proper AESs reveals a pattern which contradicts that emerging from
the Appliances sector. Patterns of proper AESs differ according to model specification
rather than estimator - which was the case in the Appliances sector. All four estimates
of Models 3 and 4 reveal the same signs on the three proper AESs: capital:labour is
everywhere positive; while both capital:labour and labour materials are everywhere
negative. Moreover, all three estimates ofModel 1 generate the same pattern of proper
AESs: capital:labour and capital:materials are everywhere positive, while
labour:materials is everywhere negative. The only specification where the pattern is
inconsistent across all estimators is Model 2.
Using the earlier assertion that whether output is exogenous or endogenous the non-
autocorrelation augmented estimates of Model 1 are most appropriate suggests that
irrespective of the assumption made about output the same conclusion holds regarding
the relationship between inputs: that capital can substitute for both labour and
materials, and that materials and labour are complements in the production of furniture.
This conclusion should, nevertheless, be qualified by the fact that it emerges from sets
of non-concave estimates.
Tests of the hypothesis that economies of scale are constant for the different estimates
of the various models reveal an interesting difference between the autocorrelation
augmented estimates of the different specifications and those estimates which do not
take autocorrelation into consideration. For both the IZEF and I3SLS estimates of the
five models, the hypothesis of constant returns to scale are rejected (see Table
A5.4.1.4 and Table A5.4.1.5). For the IZEF (AR) estimates, the hypothesis of constant
47 The choice of appropriate estimates when output is exogenous is clear. When output is endogenous
LR tests are less revealing about a preferred model. LR tests comparing I3SLS estimates show Model
1 to be preferred. No I3SLS (AR) estimates of this model emerge preventing a test of the D-W
indication that errors are autocorrelated. Given that for most specifications LR testing revealed a
preference for estimates which assume errors are uncorrelated it is submitted that if output is
endogenous that 13SLS estimates of Model 1 are most appropriate.
112
returns can be rejected only for Model 1, while for the I3SLS (AR) estimates constant
returns to scale is valid for the Cobb-Douglas specification.
Comparing mean economies of scale of the different estimators of the five models (see
Table 5.3.2) reveals some startling differences in the estimates of scale. While IZEF
and IZEF (AR) estimates show a similar pattern to that emerging from the Appliance
sector - models which are nonhomothetic (Models 1 and 2) producing larger estimates
of scale than non-homothetic specifications (Models 3, 4 and 5), the same does not
hold for I3SLS and I3SLS (AR) estimates. The data suggest that should output be
exogenous and the ZEF class of estimators is preferred then non-homothetic
specifications generate larger estimates of scale than homothetic specifications. If,
output is endogenous and the 3SLS class of estimator is appropriate then homothetic
specifications yield larger estimates of scale. Despite no clear pattern of influence the
results indicate that homotheticity does effect the magnitude of estimates of scale.
Table 5.3.2 Mean estimates of Economies of Scale - ISIC 3320

Model 1 1.374-' 1.358 1.290· --
Model 2 1.361· -- 0.674'" -0.278
Model 3 1.191 1.287'" 1.190· 0.183
Model 4 1.176 1.193· 1.188 1.253'"
ModelS 1.263· 1.136 1.263· 1.116
Note: Asterixs indicate estimates which are statistically preferred on the basis ofLR tests comparing
autocorrelated augmented and non - autocorrelated augmented estimators.
As is the case with elasticity's of substitution economies of scale appear to be similar

irrespective of the assumption made about output. Both the IZEF and 13 SLS estimates
of Model 1 yield similar mean estimates of scale (1.374 and 1.290 respectively). The
trends of these scale estimates do, however, differ. A plot of the IZEF estimates of (see
Figure 5.3.3) suggest that economies of scale have prevailed over the entire period
under analysis and are increasing. The 13 SLS estimates, on the other hand, indicate
that since 1981 the magnitude of scale has decreased and any available economies are
already exhausted.
Figure 5.3.3 Economies of Scale - IZEF and I3SLS estimates of Model 1 - (ISle
3320)
113
1.8
1.6
1.4
1.2
0.8
0.6
0.4 ~EOS.IZEF
0.2 - 0 - EOS -13SLS
0
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
5.3. 3 CONCLUSION - ISIC 3320
While the extensive violation of concavity in this sector prevents any confidence being
placed on the estimates emerging here the data do provide some interesting results.
The data suggest, contrary to a priori expectations, that capital and labour are
substitutes in the production of furniture. Furthermore, patterns of proper AESs in this
sector appear to be sensitive to model specification rather than to the estimator used,
which was the case in the Appliances sector. The data also suggest that homotheticity
does, indeed, influence estimates of economies of scale. The impact of homotheticity
on scale estimates depends, however, on the nature of output. If output is exogenous
then non-homothetic specifications generate larger estimates of scale than homothetic
specifications. If, output is endogenous the converse holds. Furthermore the data
suggest that if output is exogenous, further economies of scale may be realised but if
output is endogenous economies of scale may already have been exhausted.
5.4 MOTOR VEHICLES, PARTS AND ACCESSORIES (ISIC 3840)
5.4.1 INTRODUCTION and A PRIORI ANALYSIS
As argued in the introduction to this dissertation the research questions posed by Moll
(1991) can only be answered by comparing the features of technology of wage goods
and luxury goods sectors. Motor Vehicles and Accessories (Isle 3840) is used here as
a suitable example of a luxury good industry which is believed to experience increasing
returns to scale (see for example Standish and Galloway (1991». An attempt to
address Moll's assertions by comparing the technology of this industry with that of the
two representative wage good industries (Appliances and Furniture) using the Translog
system was, unfortunately, not entirely successful. As was the case with the Furniture
114
sector the Motor Vehicles estimates are plagued by non-concavity. Indeed, none of the
sets of parameter estimates fulfilled the requirements of quasi-concavity over the entire
sample. A consequence of non-concavity is that very few theoretically valid estimates
of the features of technology emerged. The data do, nevertheless, yield both
quantitative and theoretical conclusions regarding Motor Vehicle technology and in
addition they allow valid quantitative comparisons of the Motor vehicles and
Appliances technologies. That comparison is deferred until section 5.5.
As was the case with the other two sectors manipulations of the data were analysed in
an attempt to form a priori expectations about substitution possibilities and economies
of scale. That analysis indicates that capital and labour may be complements and
materials a substitute for both capital and materials in the production of Motor
Vehicles. No obvious trends in real average costs prevent any expectations being
formed about economies of scale.
The composition of total costs faced by the Motor Vehicles sector is similar to that
faced by both the Appliances and Furniture sectors with materials contributing the
most and capital the least to total costs. The contribution of materials to total costs in
this sector, is however, considerably greater than in either of the other sectors. More
relevant for this dissertation, m particular for identifying possible
complementarity/supplementarity, are patterns in the relative cost shares of the three
inputs. The cost shares of capital and labour display similar trends over the period both
moving in a counter-cyclical manner to materials' cost share. These movements
suggest complementarity of capital and labour and the substitutability between
materials and both capital and labour.
Figure 5.4.1 Cost shares of Capital, Labour and Materials for ISIC 3833 (1972-1990)
El Cost share of Materials

1.2
• Cost share of Capital
1
0.8
0.6
0.4
0.2
o
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
115
The above inferences are supported by plots of the input-output coefficients of the
three inputs (see figures A5.2.3.1 to A5.2.3.4 - Appendix 5.2). which reveal that
between 1981 and 1985 the input-output coefficients of both capital and labour moved
together increasing considerably (decreasing again after 1986) while, despite
fluctuating, the material input-output coefficient generally declined over that period.
These conclusions are partially supported by the correlation matrix of input-output

coefficients and relative prices (see Table 5.2.3) The positive correlation between
capital and labours input-output ratios suggests that these two inputs are complements.
A conclusion which is supported by the negative corrrelation between the input-output
coefficient of capital and the relative price of labour. The conclusion is, however,
contradicted by the positive correlation between the input-output coefficient of labour
and the relative price of capital. The substitutability between materials and both labour
and capital is suggested by the negative correlation's between the input-output
coefficient of materials and those of labour and capital, respectively. It is, however,
contradicted by the negative correlation between the relative price of materials and the
input-output coefficients of capital and labour.
Table 5.2.3 Correlation matrix of Relative prices and Input-output coefficients (ISIC 3840)
Materials
Caoital
Labour
Materials
Input-output 0.55179 -0.249£-01 -0.45214 1.0000
coefficient of
CaDital
Input-output 0.41469 -0.30031 -0.40274 0.88435 1.0000
coefficient of
Labour
Input-output -0.120£-01 -0.42615 -0.880£-01 -0.794£-01 -0.4318£- 1.0000
coefficient of 01
Materials
A plot of real average total costs over the period under consideration does not reveal
nor discount the presence of scale effects. Real average total costs remained relatively
constant and stable until 1982 whence they began fluctuating until 1986. After 1986
average costs decline only to increase again in 1988.
116
Figure 5.2.6 Real Average Costs over the period 1972-1990 (lSIC 3840)
0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
0.01
O+--+-+--+-+--+-+--+-+--+-+--+-+--+--+---t--t----t----1
n n n n n n n n " ~ U ~ Un" u " " ~
The parameter estimates emerging for the Motor vehicles sector display a number of
parallels to those of the Appliances sector. In both sectors the errors of the component
equations are correlated48 and more flexible forms are statistically preferred
FOOTNOTE. Both these parallels are unremarkable. Autocorrelation is to be
anticipated given that the data are time series. The statistical desirability of more
flexible forms is reasonable given that the maintained hypotheses of the more
restrictive specifications are untenable. A third parallel between both sets results is,
however, remarkable. For both sectors, both non-autocorrelation augmented
estimators generated results which display concavity more frequently than the results
emerging from the autocorrelation augmented estimators. Moreover IZEF and I3 SLS
estimates display the same patterns of concavity49 as do the IZEF (AR) and I3SLS
(AR) estimates5051 (see Tables A5.3.4.1 to A5.3.4.30.). The source of the more
48 Durbin-Watson tests indicated the presence of autocorrelation in all specifications estimated using
both IZEF and I3SLS (see Tables A5.3.2.1 and A5.3.2.2) prompting the use ofIZEF (AR) and I3SLS
(AR) respectively. LR tests of the appropriateness ofthese autocorrelation augmented estimators
indicated their validity for all specifications (where comparisons were possible48 ) other than the
Cobb-Douglas form (see Tables A5.3.1.2 and A5.3.1.3).
49 For lZEF and I3SLS estimates, Model 1 was non-eoncave everywhere; Model 2 concave in 1972
and 1985 and 1986; Model 3 was concave in 1983, 1984, 1985, 1988 and 1989; and Model 4 was
concave between 1983 and 1989.
50 For IZEF (AR) and I3SLS (AR) estimates, Models 1,2 and 3 were everywhere non-eoncave, while
Model 4 was concave between 1983 and 1989.
51 These similarities between the two non-autocorrelation augmented estimators and the two
autcorrelation augmented estimators prevail to a large extent with respect to the sign and magnitude
of the proper AESs (see Tables A5.3.4.1 to A5.3.4.30). Consider for example estimates of Model 2.
The IZEF and I3SLS estimates produce the same pattern of AESs: both the capita1:labour and
capital:materials AESs are positive while the labour:materials AS is negative. The IZEF (AR) and
117
prevalent non-concavity among autocorrelation augmented estimates is not

unfortunately clear.
An implication of the widespread violation of concavity is that the theoretically sound

estimates of technology are not the most desirable statistically. LR and autocorrelation
testing revealed a preference for autocorrelation augmented estimators of the most
general specifications. All autocorrelation augmented estimators violated concavity at
all points in the sample as did the non-autocorrelation estimates of Model 1. Given that
53
priority ought to be afforded theoretical validitf2 the most appropriate estimates for
analysing scale and substitution are the IZEF and 13SLS estimates of Model 2 at three
points in the sample 1972, 1985, 1986. This limited number of valid observations
impairs an important advantage of using non-homothetic and non-homogenous models
such as the translog: that estimates of scale vary with output.
5.4.2.1 ELASTICITY'S OF SUBSTITUTION - ISIC 3840
Proper AESs emerging from the IZEF and I3SLS estimates of Model 2 (Table 5.4.2.)
yield two important results. Firstly the data indicate that whether output is exogenous
(and IZEF estimates are appropriate) or endogenous (and 13 SLS estimates are
appropriate) capital and labour are strong substitutes, capital and materials are weak
substitutes and labour and materials are weak complements. Secondly a comparison of
the generally concave estimates with the generally non-concave estimates indicates that
non-concavity does impact on estimates of substitution although no systematic impact
is obvious.
4 2 .S
T a hi e 5•• e iected eIasfICltv
· , S0f su bStltutlOD
. ISle 3840
Ca ital-Labour Proper AES
Year IZEF estimates 13SLS estimates IZEF (AR) I3SLS (AR)
estimates estimates
1972 3.7682 3.7266 4.4940 4.4709
1985 3.5247 3.4964 3.4955 3.4779
1986 3.5344 3.4997 3.5510 3.5266
Caoital-Materials Proper AES
estimates estimates
1972 0.2459 0.2609 -0.5428 -0.5381
I3SLS (AR) estimates produce similar patterns of AESs which differ significantly from the non-
autocorrelation estimates in that the capital:materials AES is everywhere negative.
52 The impact of non-concavity on parameter estimates is unknown.
53 These results are statistically flawed in that they are less efficient than estimates which take
account of autocorrelation. Furthermore this specification is less flexible than Model 1 which was
shown to be statistically the most desirable.
118
1985 0.2127 0.2263 I -0.2865 I -0.3382

1986 0.2367 0.2517 I -0.3451 I -0.3102
Labour-Materials Proper AES
estimates estimates
1972 -0.0894 -0.0880 -0.1628 -0.1778
1985 -0.0030 -0.0016 -0.0414 -0.0970
1986 -0.0283 -0.0275 -0.1353 -0.1260
5.4.2.2 ECONOMIES OF SCALE - ISIC 3840
The conclusion emerging from the AESs that concavity does influence estimates of the
different features of technology is supported by an analysis of estimates of economies
of scale. Trends in estimates of scale emerging from the four estimators suggests the
possibility of an inverse relationship between the stability of scale estimates and the
concavity of the estimates. For this sector non-autocorrelation augmented estimators
generated results which display concavity more frequently than the results emerging
from the autocorrelation augmented estimators. For all three specifications where scale
estimates vary with output (Models 1, 2 and 3) a comparison of trends indicates far
greater deviations from the mean in the IZEF (AR) and I3SLS (AR) estimates than the
IZEF and I3SLS estimates (see figures A5.3.6.5, A5.3.6.6 and A5.3.6.7.). This casual
observation regarding the volatility of non-concave results is supported by a
consideration of the standard deviations of scale estimates (see Table 5.4.3.). These
data indicate that for all specifications the autocorrelation augmented estimators
generate estimates of scale which vary more than the those estimates which ignore
autocorrelation. Given that the autocorrelation augmented estimators violate concavity
more generally it would appear that an inverse relationship exists between the volatility
of estimates of scale and the fulfilment of the concavity requirements.
..
Table 5 4 3 Means and standard deviations of estimates of Economies of Scale - ISle 3840
MEAN S.D MEAN S.D MEAN S.D. MEAN S.D.
Model 1
Model 2
1.245
1.234
0.049
0.027
1.465
0.774
0.102 1.251 0.061 -- -
0.235 1.238 0.028 1.317 0.049
Model 3 1.033 0.005 1.154 0.045 1.027 0.00001 1.121 0.054
As argued above the most appropriate estimates of scale are the IZEF and 13 SLS
estimates of Model 2 in 1972, 1985 and 1986. Those estimates indicate that
irrespective of the assumption made about the nature of output the same conclusion
emerges about trends of economies of scale: that economies of scale have improved
over the period under analysis but may be on the wane (see Table 5.4.4). An
119
interesting feature of the results is that while only a few points in the sample are valid
plots of the whole sample (see figure 5.4.3) reveal very similar trends, suggesting that
while economies of scale do prevail in the sector and increased considerably between
the early 1970's and 1985, they appear to be following a downward trend towards the
end of the sample.
Figure 5.4.3 Economies of Scale in Model 2 based on IZEF and I3SLS estimates -
ISle 3840
1.3
1.28
1.26
1.24
1.22
1.2
1.18
1.16
1.14 +--+-+--+-+--+-+--+-+--+--t--+--t--+--t---t--t---t----I
72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90
The conclusion emerging from the Appliances sector regarding the impact of
homotheticity on estimates of scale is vindicated by the results emerging from this
sector. The only non-homothetic estimates of scale which fulfil concavity are the IZEF
and I3SLS estimates of Model 2 for 1972, 1985 and 1986. Conclusions regarding the
impact of homotheticity emerge if one can compare these estimates of scale with
concave estimates of homothetic models at the same points in the sample. Although
the IZEF and I3SLS estimates of Model 3 are concave at five points in the sample only
one point, 1985, is common with the estimates of Model 2. For that observation both
the IZEF and 13 SLS estimates of scale emerging from Model 2 are greater than those
emerging from Model 3 (see Table 5.4.4. below). Furthermore, Model 5, the Cobb-
Douglas specification which is homogenous and thus homothetic, fulfils concavity a
priori and for both the IZEF (1.1056) and 13 SLS (1.0868) parameter estimates
produces estimates of scale which are smaller than the Model 2 estimates for 1976,
1985 and 1986.
.
Table 5.5 3 Economies of Scale, Models 2 and 3 - ISle 3840
Year Economies of Scale Economies of Scale Economies of Scale Economies of Scale
- Model 2 - IZEF - Model 3 - IZEF - Model 2 - 13 SLS - Model 3 - 13 SLS
estimates estimates estimates estimates
1972 1.2168 1.2126
120
1983 1.0254 1.0267

1984 1.0296 1.0267
1985 1.2842 1.0390 1.2833 1.0267
1986 1.2695 1.2665
1988 1.0307 1.0267
1989 1.0313 1.0267
5.4.3. CONCLUSION - ISIC 3840
Despite the widespread violation of concavity among the estimates generated using the
Motor Vehicles data a number of useful results emerge. Not only do these estimates
support the theoretical conclusions emerging from the Appliance data but also provide
estimates of the different features technology. These estimates support the conclusion
emerging from the Appliances sector that imposing homotheticty is invalid and that it
systematically biases estimates of scale downwards. Moreover the data indicate that
non-concavity impacts on estimates of technology. Indeed there is evidence to suggest
that an inverse relationship exists between non-concavity and stability of estimates of
scale.
Turning to estimates of different features of technology the data indicate that during
the period under analysis, and contrary to a priori expectations, that capital and labour
were strong substitutes while capital and materials were weak substitutes. Furthermore
as anticipated by the a priori analysis the estimates suggest that labour and materials
are complements, albeit weak complements. While no trends regarding scale are
obvious from an a priori analysis the regression analysis indicates that whether output
is exogenous or endogenous that economies of scale prevail in the sector and have
improved over the period under analysis but may be on the wane.
5.5 CONCLUSION
The application of the Translog system to the three sectors under analysis yields two
broad groups of conclusions - a theoretical set and a quantitative set. The theoretical
conclusions provide a useful contribution to attempts to quantify features of
technology, and hence the use of empirical estimates to support a position in debate on
industrial strategy. In this regard the data clearly indicate that the maintained
hypothesis embodied in the most common functional forms are not only statistically
inappropriate for the sectors analysed here but furthermore that imposing
homotheticity inappropriately biases estimates of scale downwards. Whether the
downward bias is always, in relative terms, of the same magnitude is an important
empirical question. If it is not, using estimates of scale which emerge from homothetic
121
functions such as the Cobb-Douglas to rank sectors according to the degree of scale
experienced would be misleading.
A second theoretical conclusion emerging from the preceding analysis concerns the
impact of non-concavity on estimates of the different features of technology. Duality
theory suggests that non-concave estimates ought to be treated with care. A
comparison of concave and non-concave estimates indicates that concavity violations
effect estimates of both AESs and economies of scale. In general no systematic
relationship exists between these features and concavity 54 .
The persistent violation of concavity among the furniture and motor vehicles estimates
considerably undermined the attempt of this dissertation to address the empirical
questions posed by Moll's (1991) critique of inward industrialisation as a post-
apartheid growth strategy for South Africa. That evidence which does emerge does,
however, suggests that inward industrialisation is empirically appropriate. This
conclusion, which emerges from a comparison of only two 'representative' sectors,
should nevertheless be treated with respect.
Concavity violations prevented the use of any of the estimates for the furniture sector
and prevented the use of the statistically most appropriate estimates for-the other two
sectors. For the appliance sector concavity violations merely meant that less efficient
estimates had to be used. For the motor vehicles sector the concavity problem was
more serious. Again no efficient estimates were available, but in addition the most
suitable specifications were precluded and the 'second best' estimates were only
appropriate at three points in the sample: 1972, 1985 and 1986. Consequently, in order
for a valid comparison to be drawn between the appliances and motor vehicles sectors
the 'second best' estimates of both sectors at the appropriate three points in the sample
were analysed.
A comparison of estimates of economies of scale at these points in the sample (see

Table 5.5.1) indicates that whether output was exogenous or endogenous both sectors
experienced economies of scale which followed the same trends: increasing between
1972 and 1985 and then decreasing slightly between 1985 and 198655 . In terms of
54 While in the case of the Motor Vehicles sector non-concave estimates produce more volatile
estimates of scale than concave estimates there does not appear to be any systematic relationship
between concavity and the volatility of scale estimates in the Appliance sector.
55 A disturbing feature of the conclusion emerging here is that the analysis of the Appliances sector
suggested that if output was exogenous that returns to scale were constant and that if output were
endogenous that scale was decreasing. Concavity problems prevented the use of the statistically most
desirable estimates of the Appliance sector for the comparison with the Motor Vehicles sector. A
122
relative magnitude of economies of scale, these estimates indicate that the appliance
sector enjoyed considerably larger economies of scale than the motor vehicle sector.
Should one argue that the appliance sector is a truly representative wage good sector -
there would appear to be some empirical evidence supporting the argument of inward
industrialisation. That argument is premised on wage goods sectors having large
economies of scale. A scale coefficient of 2 indicates that any percentage change in
output is double the associated percentage change in cost - indicating that there are
considerable benefits to be reaped from expanding production.
5.5.1 Comparative. estimates of economies ()f scale. Appliances and Motor vehicles, Model 2
Year IZEF estimates for 13SLS estimates for IZEF estimates for 13SLS estimates for
Motor Vehicles Motor Vehicles Appliances (ISIC Appliances (ISIC
(lSIC 3840) (lSIC 3840) 3833) 3833)
1972 1.2168 1.2126 1.9069 1. 9331
1985 1.2842 1.2833 2.0259 2.0955
1986 1.2695 1.2665 1.970 2.0388
The inward industrialisation argument is also premised on easy substitutability between

capital and labour. These data indicate that both sectors use technologies where capital.
and labour are substitutes, albeit the degree of substitutability is greater in the motor
vehicles sector.
..
552C omparatlve estimates 0 f proper AES s, Appliances and Motor vehicles, Model 2
Ca )ital - Labour proper AS
Year IZEF estimates - 13SLS estimates - IZEF estimates - I3SLS estimates -
ISIC 3840 ISIC 3840 ISIC 3833 ISIC 3833
1972 3.7682 3.7266 1.2263 1.3698
1985 3.5247 3.4964 1.3930 1.6402
1986 3.5344 3.4997 1.4245 1.6897
Capital - Materials AS
Year IZEF estimates - 13SLS estimates - IZEF estimates - 13SLS estimates -
ISIC 3840 ISIC 3840 ISIC 3833 ISIC 3833
1972 0.2459 0.2609 1.3293 1.3228
1985 0.2127 0.2263 1.2956 1.2823
. 1986 0.2367 0.2517 1.2836 1.2693
Labour· Materials
Year IZEF estimates - I3SLS estimates - IZEF estimates - 13SLS estimates -
ISIC 3840 ISle 3840 ISIC 3833 ISIC 3833
1972 -0.0894 -0.0880 0.2161 0.2216
1985 -0.0030 -0.0016 0.2150 0.2210
1986 -0.0283 -0.0275 0.1918 0.1969
consequence is that the conclusion of the comparison of these sectors rests on using 'second best'
estimates of the Appliances sector.
CHAPTER 6
CONCLUSION
The characteristics of the technology used in the sub-sectors of South African

manufacturing were central to the early debate concerning·the appropriate industrial
strategy for the post-apartheid era. The demand restructuring variant of the growth
through redistribution approach, for example, was premised on the validity of at least
two empirically testable conditions. Firstly, that realisable economies of scale are
greater in\labour-intensive wage goods production than in the production ofluxury
goods, and secondly that considerable possibilities for substituting labour for capital
exist in manufacturing as a whole.
Moll (1991) has questioned the validity of these two conditions and hence the viability
of the demand restructuring thesis. While a number of studies employing either the
Cobb-Douglas (Cobb & Douglas, 1948) or CES (Arrow, Chenery, Minhas & Solow,
1961) functions have attempted to quantify these features of techonology their
conclusions are potentially flawed. Both specifications impose the hypotheses of
homogeneity, homotheticity and seperability a priori. The most obvious manifestation
of these hypotheses is that they constrain estimates of the magnitude of scale and
substitutability to a constant over the sample - constraints which are intuitively and
theoretically unreasonable. Given that primary hypothesis tests regarding the
magnitude of parameters depend on the validity of both the hypothesis being tested and
the underlying maintained hypotheses, the presence of implausible maintained
hypotheses is potentially damaging to any econometric analysis. Indeed, establishing
the validity of imposed maintained hypothesis ought to be a central concern of any
analysis.
This dissertation - by employing a flexible non-homothetic Translog function

(Christensen, Jorgensen and Lau, 1971, 1973) - is explicity concerned with the issue of
124
This dissertation - by employing a flexible non-homothetic Translog function

(Christensen, Jorgensen and Lau, 1971, 1973) - is explicity concerned with the issue of
the validity of the most commonly imposed maintained hypotheses while simultaeously
addressing Moll's (1991) critique of the demand restructuring thesis.
The introduction to the study of economic phenomena of the mathematical tool of

duality has permitted the development of flexible functional forms which allow the
simultaneous modelling of a number of diflferent aspects of technology while imposing
few a priori maintained hypotheses. The Translog, an example of a function emerging
from the duality between cost and production, does not impose the commonly
maintained hypotheses ofhomtheticity, homogeneity or seperability a priori. Indeed
these hypotheses can be imposed on the function through parameter constraints and
thus are rendered testable.
The added flexibility afforded by duality is however, at a cost. Duality between cost
and production is premised on the validity of certain behavioural assumptions which
are manifested in the concavity of the cost function. Violation of these assumptions
renders the estimates of technology theoretically void. While non-concavity hampered
this application of the Translog the data do, nevertheless, yield some interesting and
potentially useful results and point to a number of areas of interest for future researcch.
Broadly, three sets of conclusions emerge from this dissertation. The first set concerns
the empirical questions posed by Moll (1991), the second the validity of homotheticity,
homogeneity and seperablity and the impact of homotheticity on estimates of
economies of scale while the third set concerns the impact of concavity on estimates of
scale and substitution. All three sets of conclusions suggesst that useful insights could
emerge form further applications of the Translog. Indeed all three conclusions would
be stregthened by both an analysis of a wider cross section of sub-sectors together with
an aattempt to improve the data.
In this regard an attempt could be made to create individual price of capital series for
the different sectors analysed and longer series of the other variables used here.
125
While concavity problems prevented a complete comparison of scale and substitution

in the sectors under analysis the data do sugggest a conclusion to both empirical
questions posed by Moll (1991). The data indicate that economies of scale are greater
in wage goods industries (as represented by the Electrical Appliances and Household
goods sector) than in luxury production (as represented by the Motor Vehicles, Parts
and Accessories sector). Furthermore the data indicate that labour is a good substitute
in both sectors - albeit that the degree of substitutablity is greater in the Motor
Vehicles sector. The fact that these conclusions emerge from a comparison of only
two sectors and using less than the best estimates of the model, at very few points in
the sample, points to the need for two areas of future research before the Translog will
yield convincing contributions to the industrial strategy debate.
Firstly the present study could be extended to incorporate a broader cross- section of
manufacturing sub-sectors. Using both the same data set and methods as employed
here but extending the analysis to all wage and luxury goods manufacturing sub-
sectors would serve two purposes. Not only would it provide more evidence to draw
conclusions about the validity of demand restructuring but it may highlight causes of
the concavity problems experienced here. Should an extended analysis yield estimates
which are as badly conditioned as the etimates which emerge here a second area of
research would present itself which may contribute to resolving the concavity issue.
Concavity problems could stem from the mc data set or the restrictive assumptions
made here concerning the price of capital. Attempts should be made to develop more
appropriate price of capital series and alternaative, longer series of the other variables.
The second important conclusion of the dissertation concerns the appropriate

maintained hypotheses. The data clearly indicate that not only are the hypotheses of
homogeneity, homotheticity and seperability invalid but that homotheticity biases
estimates of economies of scale downwards. While the invalidity of the commonly
imposed maintained hypotheses is sufficient to statistically discredit the results of
applications using either the Cobb-Douglas or CES forms, the finding that imposing
homotheticity inappropriately biases estimates of scale downwards presents an
interesting avenue for research which could partially resurrect the usefulness of Cobb-
126
interesting avenue for research which could partially resurrect the usefulness of Cobb-
Douglas and CES based analyses. In this regard whether the downward bias emerging
from homtheticity is always, in relative terms, of the same magnitude is an important
consideration. If it is, estimates of scale which emerge from homothetic functions such
as Cobb-Douglas could be used to give a crude ranking of sectors according to the
degree of scale experienced. Pursuing the avenues for research and homogeneity on
scale estimates and hence better use of traditional production function analyses for
industrial strategy debates.
The final conclusion to emerge here concerns concavity. The data provide empirical
evidence for the theoretical argument that non-concave estimates ought to be treated
with care. Indeed, the data clearly indicate that concave estimates of scale and
substitution differ from non-concave estimates - albeit in a non-systematic manner.
Again the research programmes suggested above could prove useful for assessing
more completely the effect of non-concavity on estimates of scale and substitution.
127
BmLIOGRAPHY
African National Congress (1994) The Reconstruction and Development Programme:

A Policy Framework, Umanyane Publications, Johannesburg.
AlIen, R. G.D (1938) Mathematical Analysisfor Economists, MacMillan, London.
Arrow, K.J. (1964) Optimal Capital Policy, the Cost of Capital, and Myopic Decision
Rules, in the Annals ofthe Institute ofStatistical Mathematics, 16,21-30.
Arrow, K.J., H.B. Chenery, B. Minhas, and R.B. Solow (1961) Capital-Labour
Substitution and Economic Efficiency, Review ofEconomics and Statistics, 43:5,
August, 225-254
Barten, A.P. (1969) Maximum Likelihood Estimation of a Complete system of

Demand Equations, European Economic Review, NO.I pp7-73.
Baumol, W.J. (1977) Economic Theory-and Operations Analysis, 4th Edition

Prentice-Hall Intemational, London.
Bi0m, E (1989) Taxation, Technology and the User Cost ofCapital, Elsevier Science
Publishers, Amsterdam.
Bemdt, E.R. (1991) The practice ofeconometrics: classic and contemporary,

Addison-Wesley, Reading, Massachusetts.
Bemdt E.R. and L.R.Christensen (l973a) The Intemal Structure of Functional

Relationships: Seperability, Substitution and Aggregation, The Review ofEconomic
Studies VoI. XL (3) No. 123.
Bemdt E.R. and L.R.Christensen (1973b) The Translog Function and the substitution
of Equipment, Structures and Labour in U.S. Manufacturing 1929-1968, Journal of
Econometrics, I, 81-114.
128
Bemdt, E.R, M.N. Darrough, and W.E. Diewert (1977) Flexible Functional Forms and
Expenditure Distributions: An Application to Canadian Consumer Demand Functions,
International Economic Review, 18:3.
Bemdt, E.R. and D.D. Wood (1975) Technology, Prices and the Derived Demand for
Energy, The Review ofEconomics and Statistics, 57:3,259-268
Bemdt, E.R. and D.D. Wood (1982) The Specification and Measurement of Technical
Change in U.S. Manufacturing, in Advances in the Economics ofEnergy and
Resources, Vol. 4, Edited by John R. Moroney, JA! Press, Greenwich, Connecticut.,
Brown, M.B (1968) On the Theory andMeasurement of Technological Change,

Cambridge University Press, Cambridge.
Browne, G.W.C (1943) The Production Function of South African Manufacturing

Industry, in South African Journal ofEconomics, 11,258-268.
Chambers, R.G. (1988) Applied Production Analysis, Cambridge University Press,

New York.
Christensen, L.R. (1977) Simultaneous Estimation of Technical change and Scale

Economies, Unpublished Memorandum, Department of Economics, University of
Wisconsin.
Christensen L.R. and W.H. Greene (1976) Economies of Scale in U.S. Electric Power
Generation, Journal ofPolitical Economy, 84: 4, 655-676.
Christensen L.R. and D.W Jorgensen (1969) The Measurement ofU.S. Real Capital
Input, 1929-1967, in Review ofIncome and Wealth, Dec., 292-320.
Christensen, L.R., D.W. Jorgensen and L.J Lau (1973) Transcendental Logarithmic
Production Frontiers, in Review ofEconomics and Statistics, 55: 1,28-45.
Chiang, A. C. (1984) Fundamental Methods ofMathematical Economics 3rd Edition,

McGraw-Hill, Singapore.
Clemhout, S. (1968) The Class of Homothetic Isoquant Production Functions, in

Review ofEconomic Studies, 35, 91-104.
129
Cluver, F.H. and E. Contogiannis (1984) Energy substitution reponses of South

African manufacturing: An application of the Translog Production Function, Journal
ofStudies in Economics and Econometrics, 18, 5-14.
Cluver, F.H (1981) An application ofa transcendental logarithmic production

function incorporating an energy input to selected sectors ofSouth African
manufacturing, Unpublished Masters of Economic Science Thesis, University of the
Witwatersrand.
Cobb, C and P.H. Douglas (1928) A Theory of Production, in American Economic

Review, Supplement to Volume 18, 139-165.
COSATU (1992) Economic Policy in COSATU: Report ofCOSA TU Economic Policy

Conference, 27-29 March, 1992.
Cramer, l S. (1986) Econometric Applications ofMaximum Likelihood methods,

Cambridge University Press, New York.
Denny,M. and lD.May (1978) Homotheticity and Real Value-Added in Canadian

Manufacturing, in Production Economics: A Dual Approach to theory and
Applications, Vo1.2. Elsevier-North-Holland, Amsterdam.
Denison, E.F. (1962) The Sources ofEconomic Growth in the United States and
Alternatives Before Us, Supplementary Paper No. 13, New York, Commitee for
Economic Development.
Diewert, W.E (1971) An application ofthe Shepherd Duality Theorem: A Generalized

LeontiefProduction Function, Journal ofPolitical Economy, 79:3, May/June 481-
507.
Diewert, W.E. (1982) Duality Approaches to Microeconomic Theory, in Handbook of

Mathematical Economics, Vol II, edited by K.J. Arrow and M.D. Intriligator, North-
Holland, Amsterdam.
Dixon, P.B, S. Bowles and D. Kendrick (1980) Notes and Problems in Micro
economic Theory, Elsevier Science, Amsterdam.
130
Durbin, 1 (1957) Testing for Serial Correlation in Systems of Simultaneous

Regression Equations, Biometrika, Dec., 370-377.
Fuss, M.A. (1977) The Demand for Energy in Canadian Manufacturing: An example
of the Estimation of Production Structures with Many Inputs, Journal of
Econometrics, 5, 89-116.
Fuss, M., D. McFadden and Y. Mundlak (1978) A survey of Functional Forms in the
Economic Analysis of Production, in Production Economics Vol. I, edited by M. Fuss
and D. McFadden, North-Holland, Amsterdam.
Gould, lP. and c.B. Ferguson (1980) Microeconomic Theory, 5th edition, Richard D
Irwin Inc., Homewood, Illinois.
Greene, W.H (1992) LIMDEP Version 6.0: User's Manual and Refemce Guide,
Econometric Software Inc., New York.
Greene, W.H. (1993) Econometric Analysis, 2nd Edition, MacMillan, New York.
Greene, W.H. (1983) Simultaneous Estimation of Factor Substitution, Economies of

Scale, Productivity, and non-neutral Technical Change, in Developments in
Econometric Analyses ofProductivity, ed A. Dogramaci pp 121-144, Kluwer-Nijhoff,
Boston.
Grilliches, Z and V. Ringstad (1971) Economies ofScale and the Form ofthe
Production Function, North-Holland, Amsterdam.
Guilkey, D., c.A. Knox-Lovell and R.C. Sickles (1983) A Comparison of the
Performance of Three Flexible Functional Forms, International Economic Review,
24:3, October, 591-616.
Gujarati, D.N. (1988) Basic Econometrics, 2nd Edition, McGraw-Hill, Singapore.
Hall, R.E. and D.W. Jorgensen (1967) Tax Policy and Invetsment Behaviour, The
American Economic Review, 57:3.
Hall, R.E. (1968) Technical Change and Capital from the point of the Dual. Review of
Economic Studies, 35, 35-46.
131
Hanoch, G. (1971) CRESH Production Functions, Econometrica, 39:5
Hausman, lA. (1978) Specification Tests in Econometrics, Econometrica, 46:6,

1251-1271.
Henderson, lM. and R.E. Quandt (1980) Microeconomic Theory: A Mathematical

Approach, 3rd Edition, McGraw-Hill, New York.
Industrial Development Corporation of South Africa Limited, (1992) Sectoral Data

Series: Manufacturing, Industrial Development Corporation of South Africa Limited:
Department of Economic Research and Development, Johannesburg.
Intriligator, M.D. (1978) Econometric Models, Techniques and Applications,

Prentice-Hall Inc., Englewood Cliffs, New Jersey.
Jehle, G.A. (1991) AdvancedMicroeconomic Theory, Prentice-Hall International,

London.
Jorgensen, D.W. (1963) Capital Theory and Investment behaviour, in American

Economic Review, May 1963.
Jorgensen, D.W. (1974) The Economic Theory of Replacement and Depreciation, in

Econometrics and Economic Theory. Essays in Honour ofJan Tinbergen edited by
W. Sellekaerts, MacMillan, London.
Jorgensen, D.W (1986) Econometric Methods for Modelling Producer Behaviour, in

Handbook ofEconometrics, Vol. 3. edited by Zvi Grilliches and M.D. Intriligator,
North-Holland, Amsterdam.
Judge, G. G., W.E. Griffiths, R. Carter Hill, H. Lutkepohl, Tsoung-Chao Lee, (1985)
The Theory and Practice ofEconometrics, 2nd Edition, John Wiley and Sons,
Singapore.
Kennedy, P (1985) A Guide to Econometrics, 2nd Edition, Basil Blackwell, Oxford.
Kmenta, J (1986) Elements ofEconometrics, 2nd Edition, MacMillan, New York.

132
Koutsoyiannis, A (1979) Modern Microeconomics 2nd Edition, MacMillan, Hong-

Kong.
Lancaster, K (1968) Mathematical Economics, MacMillan, New York.
Lau, L.l (1978) Testing and Imposing Monotonicity, Convexity and Quais-Convexity
Constarints, in Production Economics Vol. I, edited by M. Fuss and D. McFadden,
North-Holland, Amsterdam.
Lipsey, RG. (1971) An Introduction to Positive Economics, 3rd Edition, Harper and
Row, London.
Lombard, lA. and J.P. van den Heever (1990) The Optimal Allocation ofSavings in
the South African economy and the role ofmonetary policy, South African Reserve
Bank, Occasional paper NO.l.
Macroeconomic Research Group (1993) Making Democracy Work: A Frameworkfor

Macroeconomic Policy in South Africa, Centre for Development Studies, Bellville.
Maddala, G.S. (1987) Econometrics, McGraw-Hill, Singapore.
Madden, P. (1986) Concavity and Optimization in Microeconomics, Basil Blackwell,

London.
Malinvaud, E. (1970) Statistical Methods ofEconometrics, Rand McNally, Chicago.
Mark, J. (1986) Problems encountered in measuring single- and multifactor

productivity, in Monthly Labour Review, Dec.
Mark, lA. and W.H. Waldorf(1983) Multifcator Productivity: a new BLS measure, in
Monthly Labour Review, Dec., 3-11.
McElroy, M.B. (1977) Goodness of Fit for Seemingly Unrelated Regressions, Journal
ofEconometrics, 6, 381-387.
McFadden, D (1978) Cost, Revenue and Profit Functions, Chapter 1.1 in Production
economics: A Dual Approach to Theory and Applications, Vol.l, edited by M. Fuss
and D. McFadden, North-Holland, Amsterdam.
133
McMenamin, V (1993) Shifts in ANC Economic Policy, in Transforming the

Economy: Poilicy Options for South Africa, Indicator South Africa, Pietermaritzburg,
245-254.
Mohr, PJ., C. van der Merwe, Z.C. Botha, and J. Inggs (1988) The Practical Guide to
South African Economic Indicators, Lexicon, Johannesburg.
Moll, T.c. (1990) Output and Productivity Trends in South Africa: Apartheid and
Economic Growth Unpublished Ph.D thesis: Faculty of Economics and Politics,
Cambridge University.
Moll, T.c. (1991) Growth Through Redistribution: A Dangerous Fantasy?, in The

South African Journal ofEconomics, 59:3, 313-330.
Nadiri, Ishaq M. (1982) Producers Theory, in Handbook ofMathematical Economics

Vol II, edited by K.J. Arrow and M.D. Intriligator, North-Holland, Amsterdam.
Nerlove, M. (1967) Recent Empirical Studies of the CES and Relalted Production
Functions in Murray Brown ed The Theory and Empirical Analysis ofProduction,
Studies in Income and Wealth, Vo1.32, Columbia University Press for the National
Bureau of Economic Research, New York, 55-122.
Spandau, A. (1973) Cross-section production functions and income shares in South

Mrican Industry, South African Journal ofEconomics, 41:3, 208-233.
Spencer, D.E. and K.N. Berk (1981) A Limited Information and Specification Test,
Econometrica, 49:4, 1079-1085.
Standard Industrial Classification of all Economic Activities, 4th Edition, Central

Statistical Services, Pretoria.
Standish, B. and D. Galloway (1991) Exports Efficiency and Capital in South African
Manufacturing, in Journal ofStudies in Economics and Econometrics, 15: 1.
Takayama, A (1985) Mathematical Economics, 2nd Edition, Cambridge University

Press, New York.
134
Theil, H and A. Nagar (1961) Testing the Independence of Regression Disturbances in

Journal ofthe American Statistical Association, 56, 793-806.
Uzawa, H. (1962) Production Functions with Constant Elasticities of Substitution,

Review ofEconomics and Statistics, 44 no. 4.
van der Wait, I.N.A and 1.1. Swanepoel (1987) Indicative Labour and Capital
Programming: An Exercise on the Frontiers of Empirical Economics, Journal of
Studies in Economics and Econometrics, 11 :2, 19-67.
Zamagni, S (1987) Microeconomic Theory: An Introduction, Basil Blackwell, London.
Zellner, A. and H. Theil (1962) Three-Stage Least Squares: Simultaneous estimation

of Simultaneous Equations, Econometrica, 30: 1, 54-78.
Zellner, A. and N.S. Revankar (1969) Generalized Production Functions, Review of

Economic Studies, 36.
APPENDIX 3
CONCAVITY TESTING
Quasi - concavity (concavity) is fulfilled if the matrix of AESs is negative semi-

definite (negative definite). A matrix [B] is negative semi-definite if
IBll ~ O,IBJ;~ 0, ... ,IBnl ~ 0 if n is odd and IBnl ~ 0 if n is even. The matrix will be
negative definite if the weak inequalities are replaced by strong inequalities (Chiang,
1984, p394).
Where three inputs are employed: capital (k), labour (l) and materials (m), the matrix
of AESs 10'1 will be negative semi-definite, and the Translog quasi-concave, if
10'11 ~ 0,la 2 1~ 0 and la 3 1~ o.
In order for the mechanics of the concavity test to be established the matrix of AESs
needs to be obtained. The matrix of AESs, which is symmetrical, can take one of six
forms:
H a~]
a lk
[aak/ a" ami
a km a'm a mm
H amk lk
[a
akm amm aim ]
a
ak/ ami a"
ak/
[all ami]
a ,k a kk a mk
a'm a km a mm
ami
[all a mm
aM ]
a km
a'm
a ,k a mk a kk
a km
[a_ aim]
a mk a kk a ,k
amI a kl a"
mm a~]
r
a'm
amI a" a kl
a mk a lk a kk
A3-2
Despite six alternative matrices of AESs existing only three elements appear as the
fIrst element on the principle diagonal: 0' Ick ,0' /I and 0' mm' Thus in order for the fIrst
condition of negative semi-defIniteness (10'I1 ::; 0) to be fulfIlled three conditions need
to be satisfIed: 0' Ick ::; 0,0'/1 ::; 0 and 0' mm ::; 0
The second condition for negative semi-defIniteness is that 10'21 ~ O. Six different [0'2]
matrixes exist:
[Gu G,] A3.1

O'k/ 0'/1
[GU G
mk
] A3.2
0' km 0' mm
II
[G
GM ] A3.3
O'/k O'lck
[ G
II G.,] A3.4
O'/m 0' mm
[G.. G..] A3.5

O'mk O'lck
[G.. GI.] A3.6

0' m/ 0' /I
The determinants of matrixes A3.1 and A3.3; A3.2 and A 3.5; and A3.4 and A3.6 are
identical. Testing whether 10'21 ~ 0 will, therefore require fInding the determinant of
three matrixes (either A3.1 or A3.3; and either A3.2 or A3.5; and either A3.4 or
A3.6).
The third, and fInal, condition for negative semi-defIniteness is 10'31::; O. Although six
different [0'3] exist, because they are symmetrical the value of the determinant will be
identical in all cases. Hence the third condition merely requires a test of whether the
determinant of any of the versions of the matrix of AESs is less than or equal to one.
APPENDIX 4.1
DATA SERIES
A4.1.1 ELECTRICAL APPLIANCES AND HOUSEHOLD GOODS (lSIC 3833)
Table A4.1.1.1 Cost of Capital: Electrical Appliances and Household Goods (lSIC 3833)
Year Depreciation (8) Interest Paid (ip) Interest Received Cost of Capital
(ir) (CK )
1972 1000000 0 0 1000000
1973 1000000 1000000 0 2000000
1974 1000000 1000000 0 2000000
1975 2000000 1000000 0 3000000
1976 2000000 1000000 0 3000000
1977 2000000 1000000 0 3000000
1978 2000000 1000000 0 3000000
1979 2000000 2000000 0 4000000
1980 2000000 4000000 0 6000000
1981 3000000 2000000 0 5000000
1982 3000000 7000000 1000000 9000000
1983 4000000 6000000 1000000 9000000
1984 4000000 4000000 2000000 6000000
1985 3000000 10000000 2000000 11000000
1986 5000000 10000000 2000000 13000000
1987 6000000 6000000 2000000 10000000
1988 6000000 5000000 2000000 9000000
1989 6000000 8000000 3000000 11000000
1990 9000000 11000000 4000000 16000000
Note: 1. Cost of Capital (CK) is calculated as CK = 8 + ip - ir
2. Source of 8; ip and ir: mc (1992)
A4.1. - 2
Table A4.1.1.2 Cost of Labour and Price of Labour: Electrical Appliances and Household Goods
(ISIC 3833)
Year Cost of Labour Number of Average Price of Price Index of
(CL) Labourers (L) Labour (AvP L) Labour (PL ) .
1972 7000000 3600 1944.4399 1.0000
1973 9000000 3980 2261.3066 1.1630
1974 12000000 4410 2721.0884 1.3994
1975 15000000 4760 3151.2605 1.6207
1976 17000000 5140 3307.3931 1.7009
1977 17000000 4710 3609.3418 1.8562
1978 19000000 4420 4298.6426 2.2107
1979 19000000 3910 4859.3350 2.4991
1980 20000000 3820 5235.6021 2.6926
1981 26000000 3870 6718.3462 3.4552
1982 29000000 3870 7493.5400 3.8538
1983 27000000 3470 7780.9800 4.0017
1984 25000000 3160 7911.3926 4.0687
1985 26000000 2710 9594.0957 4.9341
1986 29000000 2630 11026.6162 5.6708
1987 43000000 3040 14144.7373 7.2745
1988 67000000 4220 15876.7773 8.1652
1989 68000000 4120 16504.8535 8.4882
1990 72000000 4290 16783.2168 8.6314
Note: 1. AvPL = CL / L
2. PL = AvPL /1944.4399
3. Source of CL and L: mc (1992)
A4.1. - 3
Table A4.1.1.3. Output, Cost of Materials and Price of Materials: Electrical Appliances and
Household Goods (lSIC 3833)
Year Output (y) Value- Real Cost of Material Material Cost of Price
Added (va) Materials Price Price Materials Index of
(rCM ) Index Index (CM) Materials
('90=100) (1990=1) (PM)
(PM1 ) (PM2 )
1972 115000000 41000000 74000000 8.5000 0.0850 6290000 1.0000
1973 141000000 48000000 93000000 9.5000 0.0950 8835000 1.1176
1974 175000000 59000000 116000000 11.2000 0.1120 12992000 1.3176
1975 192000000 62000000 130000000 12.7000 0.1270 16510000 1.4941
1976 195000000 66000000 129000000 15.1000 0.1510 19479000 1.7765
1977 138000000 45000000 93000000 18.1000 0.1810 16833000 2.1294
1978 136000000 46000000 90000000 19.8000 0.1980 17820000 2.3294
1979 172000000 59000000 113000000 22.2000 0.2220 25086000 2.6118
1980 225000000 76000000 149000000 25.4000 0.2540 37846000 2.9882
1981 230000000 77000000 153000000 28.5000 0.2850 43605000 3.3529
1982 218000000 73000000 145000000 33.2000 0.3320 48140000 3.9059
1983 216000000 70000000 146000000 37.1000 0.3710 54165996 4.3647
1984 195000000 66000000 129000000 40.3000 0.4030 51987000 4.7412
1985 194000000 63000000 131000000 46.2000 0.4620 60522000 5.4353
1986 205000000 67000000 138000000 57.2000 0.5720 , 78936000 6.7294
1987 230000000 72000000 158000000 65.9000 0.6590 104122000 7.7529
1988 262000000 85000000 177000000 73.6000 0.7360 130272000 8.6588
1989 251000000 83000000 168000000 89.1000 0.8910 149688000 10.4824
1990 276000000 91000000 185000000 100.0000 1.0000 185000000 11.7647
Note 1: rCM = y-va
2. PM2= PM1 /lOO
3. CM = rCM x PM2
4. PM = PM1 /8.5
5. Source ofy, va, and PM1 : mc (1992)
A4.1. - 4
Table A4.1.1.4 Costs and Cost Shares of Capital, Labour and Materials: Electrical Appliances
and Household Goods (ISIC 3833)
Year Cost of Cost of Cost of Total Costs Cost Share Cost Share Cost Share
Capital Labour Materials (TC) of Capital of Labour of
(CK) (CL) (CM) (SK) (SL) Materials
(SM)
1972 1000000 7000000 6290000 14290000 0.0700 0.4899 0.4402
1973 2000000 9000000 8835000 19835000 0.1008 0.4537 0.4454
1974 2000000 12000000 12992000 26992000 0.0741 0.4446 0.4813
1975 3000000 15000000 16510000 34510000 0.0869 0.4347 0.4784
1976 3000000 17000000 19479000 39479000 0.0760 0.4306 0.4934
1977 3000000 17000000 16833000 36833000 0.0814 0.4615 0.4570
1978 3000000 19000000 17820000 39820000 0.0753 0.4771 0.4475
1979 4000000 19000000 25086000 48086000 0.0832 0.3951 0.5217
1980 6000000 20000000 37846000 63846000 0.0940 0.3133 0.5928
1981 5000000 26000000 43605000 74605000 0.0670 0.3485 0.5845
1982 9000000 29000000 48140000 86140000 0.1045 0.3367 0.5589
1983 9000000 27000000 54165996 90166000 0.0998 0.2994 0.6007
1984 6000000 25000000 51987000 82987000 0.0723 0.3013 0.6264
1985 11000000 26000000 60522000 97522000 0.1128 0.2666 0.6206
1986 13000000 29000000 78936000 120936000 0.1075 0.2398 0.6527
1987 10000000 43000000 104122000 157122000 0.0636 0.2737 0.6627
1988 9000000 67000000 130272000 206272000 0.0436 0.3248 0.6316
1989 11000000 68000000 149688000 228688000 0.0481 0.2973 0.6546
1990 16000000 72000000 185000000 273000000 0.0586 0.2637 0.6777
Note: 1. TC = CK +CL +CM
2.SK =CK ITC
3. SL = CL ITC
4.SM =CM ITC
A4.1. - 5
A4.1.2 MOTOR VEHICLES, PARTS AND ACCESSORIES (lSIC 3840)
Table A4.2.1.1 Cost of Capital: Motor Vehicles, Parts and Accessories (ISIC 3840)
(ir) (CK )
1972 35000000 15000000 5000000 45000000
1973 40000000 18000000 7000000 51000000
1974 47000000 30000000 10000000 67000000
1975 58000000 41000000 16000000 83000000
1976 60000000 50000000 21000000 89000000
1977 52000000 59000000 17000000 94000000
1978 87000000 46000000 25000000 108000000
1979 78000000 56000000 15000000 119000000
1980 104000000 53000000 18000000 139000000
1981 136000000 66000000 28000000 174000000
1982 154000000 111000000 37000000 228000000
1983 208000000 118000000 36000000 290000000
1984 336000000 230000000 72000000 494000000
1985 602000000 324000000 112000000 814000000
1986 441000000 264000000 90000000 615000000
1987 500000000 197000000 81000000 616000000
1988 539000000 217000000 142000000 614000000
1989 668000000 351000000 214000000 805000000
1990 626000000 329000000 288000000 667000000
Note: 1. Cost of Capital (CK ) is calculated as CK = 8 + ip - ir
A4.1.-6
Table A4.1.2.2 Cost of Labour and Price of Labour: Motor Vehicles, Parts and Accessories (ISIC
3840)
(CL) Labourers (L) Labour (AvP L) Labour (PL )
1972 138000000 57510 2399.5828 1.0000
1973 165000000 61230 2694.7576 1.1230
1974 200000000 66590 3003.4539 1.2517
1975 238000000 70840 3359.6838 1.4001
1976 249000000 74450 3344.5266 1.3938
1977 263000000 70230 3744.8384 1.5606
1978 297000000 70700 4200.8486 1.7507
1979 342000000 73300 4665.7573 1.9444
1980 446000000 81310 5485.1802 2.2859
1981 617000000 90250 6836.5649 2.8491
1982 764000000 96260 7936.8379 3.3076
1983 849000000 89540 9481.7959 3.9514
1984 1009000000 90090 11199.9111 4.6674
1985 1055000000 84840 12435.1719 5.1822
1986 1162000000 81200 14310.3447 5.9637
1987 1399000000 80110 17463.4883 7.2777
1988 1675000000 79980 20942.7363 8.7277
1989 2061000000 80100 25730.3359 10.7228
1990 2280000000 82100 27771.0117 11.5733
2. PL = AvP L /2399.5828
3. Source of CL and L: IDC. (1992)
A4.1. -7
Table A4.1.2.3 Output, Cost of Materials and Price of Materials: Motor Vehicles, Parts and
Accessories (ISIC 3840)
Year Output (y) Value- Real Cost of Material Material Cost of Price Index
Added (va) Materials Price Index Price Index Materials of
(rCM ) (1990=100) (1990=1) (CM) Materials
(PM1 ) (PM2 ) (PM)
1972 11025000000 2807000000 8218000000 7.5000 0.0750 616350000 1.0000
1973 12421000000 3287000000 9134000000 8.2000 0.0820 748988000 1.0933
1974 14093000000 3629000000 10464000000 9.5000 0.0950 994080000 1.2667
1975 15094000000 3803000000 11291000000 11.2000 0.1120 1264592000 1.4933
1976 13959000000 3478000000 10481000000 13.0000 0.1300 1362530000 1.7333
1977 11732000000 2918000000 8814000000 14.7000 0.1470 1295658000 1.9600
1978 13507000000 3485000000 10022000000 16.2000 0.1620 1623564000 2.1600
1979 13615000000 3341000000 10274000000 18.4000 0.1840 1890416000 2.4533
1980 17487000000 4340000000 13147000000 20.8000 0.2080 2734576000 2.7733
1981 21252000000 5341000000 15911000000 23.5000 0.2350 3739085000 3.1333
1982 19417000000 5008000000 14409000000 27.4000 0.2740 3948066000 3.6533
1983 18456000000 4633000000 13823000000 29.9000 0.2990 4133077000 3.9867
1984 15963000000 4298000000 11665000000 33.7000 0.3370 3931105000 4.4933
1985 11613000000 2968000000 8645000000 39.2000 0.3920 3388840000 5.2267
1986 10707000000 2771000000 7936000000 49.2000 0.4920 3904512000 6.5600
1987 13244000000 3206000000 10038000000 58.5000 0.5850 5872230000 7.8000
1988 15357000000 3801000000 11556000000 66.4000 0.6640 7673184000 8.8533
1989 15054000000 3728000000 11326000000 79.1000 0.7910 8958866000 10.5467
1990 13277000000 3278000000 9999000000 100.0000 1.0000 9999000000 13.3333
Note 1: rCM = y-va
2. PM2 =
PM1 /lOO
3. CM =rCM x PM2
4. PM = PM1 /7.5
A4.1.-8
Table A4.1.2.4 Costs and Cost Shares of Capital, Labour and Materials: Motor Vehicles, Parts
and Accessories (ISIC 3840)
Capital Labour Materials (rC) of Capital of Labour of
(SM)
1972 45000000 138000000 616350000 799350000 0.0563 0.1726 0.7711
1973 51000000 165000000 748988000 964988000 0.0529 0.1710 0.7762
1974 67000000 200000000 994080000 1261080000 0.0531 0.1586 0.7883
1975 83000000 238000000 1264592000 1585592000 0.0523 0.1501 0.7976
1976 89000000 249000000 1362530000 1700530000 0.0523 0.1464 0.8012
1977 94000000 263000000 1295658000 1652658000 0.0569 0.1591 0.7840
1978 108000000 297000000 1623564000 2028564000 0.0532 0.1464 0.8004
1979 119000000 342000000 1890416000 2351416000 0.0506 0.1454 0.8039
1980 139000000 446000000 27345.76000 3319576000 0.0419 0.1344 0.8238
1981 174000000 617000000 3739085000 4530085000 0.0384 0.1362 0.8254
1982 228000000 764000000 3948066000 4940066000 0.0462 0.1547 0.7992
1983 290000000 849000000 4133077000 5272077000 0.0550 0.1610 0.7840
1984 494000000 1009000000 3931105000 5434105000 0.0909 0.1857 0.7234
1985 814000000 1055000000 3388840000 5257840000 0.1548 0.2007 0.6445
1986 615000000 1162000000 3904512000 5681512000 0.1082 0.2045 0.6872
1987 616000000 1399000000 5872230000 7887230000 0.0781 0.1774 0.7445
1988 614000000 1675000000 7673184000 9962184000 0.0616 0.1681 0.7702
1989 805000000 2061000000 8958866000 11824866000 0.0681 0.1743 0.7576
1990 667000000 2280000000 9999000000 12946000000 0.0515 0.1761 0.7724
Note: 1. TC=CK+CL +CM
2. SK =CKITC
3. SL =CLITC
4.SM =CM ITC
A4.1. - 9
A4.1.3 FURNITURE (ISle 3320)
Table A4.1.3.1 Cost of Capital: Furniture (ISIC 3320)

(fr) (CK )
1972 3000000 3000000 1000090 5000000
1973 3000000 3000000 1000000 5000000
1974 4000000 5000000 1000000 8000000
1975 5000000 6000000 2000000 9000000
1976 5000000 5000000 2000000 8000000
1977 4000000 3000000 1000000 6000000
1978 5000000 4000000 1000000 8000000
1979 5000000 4000000 1000000 8000000
1980 8000000 5000000 1000000 12000000
1981 16000000 9000000 2000000 23000000
1982 15000000 9000000 2000000 22000000
1983 16000000 13000000 4000000 25000000
1984 24000000 29000000 4000000 49000000
1985 15000000 22000000 5000000 32000000
1986 23000000 20000000 5000000 38000000
1987 30000000 20000000 4000000 46000000
1988 34000000 33000000 7000000 60000000
1989 45000000 46000000 9000000 82000000
1990 51000000 .57000000 12000000 96000000
Note: 1. Cost of Capital (CK ) is calculated as CK = cS + ip - ir
A4.1. - 10
Table A4.1.3.2 Cost of Labour and Price of Labour: Furniture (ISIC 332)
(CL) Labourers (L) Labour (AvP L) Labour (PL )
1972 47000000 25950 1811.1801 1.0000
1973 55000000 26820 2050.7085 1.1323
1974 68000000 29680 2291.1052 1.2650
1975 76000000 28220 2693.1255 1.4870
1976 79000000 28570 2765.1382 1.5268
1977 80000000 26700 2996.2546 1.6544
1978 83000000 25200 3293.6509 1.8186
1979 100000000 26320 3799.3921 2.0979
1980 135000000 31000 4354.8389 2.4046
1981 175000000 34900 5014.3267 2.7687
1982 212000000 36770 5765.5698 3.1835
1983 244000000 36400 6703.2969 3.7013
1984 259000000 36100 7174.5151 3.9615
1985 278000000 34180 8133.4111 4.4909
1986 324000000 35000 9257.1426 5.1114
1987 356000000 35000 10171.4287 5.6162
1988 414000000 37300 11099.1953 6.1285
1989 409000000 36300 11267.2178 6.2213
1990 495000000 36300 13636.3633 7.5294
2. PL = AvPL 11811.1801
3. Source of CL and L: mc (1992)
A4.1. - 11
Table A4.1.3.3 Output, Cost of Materials and Price of Materials: Furniture (ISIC 332)
Year Output (y) Value- Real Cost of Material Material Cost of Price Index
Added (va) Materials Price Index Price Index Materials of
(rCM ) (1990=100) (1990=1) (CM) Materials
(PM1 ) (PM2 ) (PM)
1972 1459000000 492000000 967000000 8.8000 0.0880 85096000 1.0000
1973 1613000000 551000000 1062000000 10.0000 0.1000 106200000 1.1364
1974 1708000000 591000000 1117000000 12.4000 0.1240 138508000 1.4091
1975 1762000000 600000000 1162000000 14.1000 0.1410 163842000 1.6023
1976 1700000000 582000000 1118000000 15.8000 0.1580 176644000 1.7955
1977 1717000000 607000000 1110000000 17.7000 0.1770 196470000 2.0114
1978 1782000000 641000000 1141000000 19.2000 0.1920 219072000 2.1818
1979 2059000000 746000000 1313000000 21.6000 0.2160 283608000 2.4545
1980 2346000000 839000000 1507000000 25.4000 0.2540 382778000 2.8864
1981 2411000000 900000000 1511000000 29.3000 0.2930 442723000 3.3295
1982 2231000000 890000000 1341000000 33.1000 0.3310 443871000 3.7614
1983 2081000000 761000000 1320000000 37.0000 0.3700 488400000 4.2045
1984 1888000000 694000000 1194000000 39.9000 0.3990 476406000 4.5341
1985 1759000000 647000000 1112000000 45.2000 0.4520 502624000 5.1364
1986 1788000000 680000000 1108000000 53.1000 0.5310 588348000 6.0341
1987 1734000000 631000000 1103000000 62.0000 0.6200 683860000 7.0455
1988 1894000000 707000000 1187000000 71.5000 0.7150 848705000 8.1250
1989 1894000000 648000000 1246000000 84.5000 0.8450 1052870000 9.6023
1990 1797000000 603000000 1194000000 100.0000 1.0000 1194000000 11.3636
Note 1: rCM = y-va
2. P M2 =
PM1 /lOO
3. CM =rCM x PM2
4. PM = PM1 /8.8
A4.1.-12
Table A4.1.3.4 Costs and Cost Shares of Capital, Labour and Materials: Furniture (ISIC 332)
Capital Labour Materials (TC) of Capital of Labour of
(SM)
1972 5000000 47000000 85096000 137096000 0.0365 0.3428 0.6207
1973 5000000 55000000 106200000 166200000 0.0301 0.3309 0.6390
1974 8000000 68000000 138508000 214508000 0.0373 0.3170 0.6457
1975 9000000 76000000 163842000 248842000 0.0362 0.3054 0.6584
1976 8000000 79000000 176644000 263644000 0.0303 0.2996 0.6700
1977 6000000 80000000 196470000 282470000 0.0212 0.2832 0.6955
1978 8000000 83000000 219072000 310072000 0.0258 0.2677 0.7065
1979 8000000 100000000 283608000 391608000 0.0204 . 0.2554 0.7242
1980 12000000 135000000 382778000 529778000 0.0227 0.2548 0.7225
1981 23000000 175000000 442723000 640723000 0.0359 0.2731 0.6910
1982 22000000 212000000 443871000 677871000 0.0325 0.3127 0.6548
1983 25000000 244000000 488400000 757400000 0.0330 0.3222 0.6448
1984 49000000 259000000 476406000 784406000 0.0625 0.3302 0.6073
1985 32000000 278000000 502624000 812624000 0.0394 0.3421 0.6185
1986 38000000 324000000 588348000 950348000 0.0400 0.3409 0.6191
1987 46000000 356000000 683860000 1085860000 0.0424 0.3279 0.6298
1988 60000000 414000000 848705000 1322705000 0.0454 0.3130 0.6416
1989 82000000 409000000 1052870000 1543870000 0.0531 0.2649 0.6820
1990 96000000 495000000 1194000000 1785000000 0.0538 0.2773 0.6689
Note: 1. TC = CK +CL +CM
2. SK =CKITC
3. SL =CLITC
4.SM =CM ITC
APPENDIX 4.2
INSTRUMENTAL VARIABLES
Table A4.2.1 Instrumental Variables: Prices and Employment data.

Year C.P.I C.P.I. App. App. Mot. Mot. Furn. Furn. Man. Emp.
'72=1 Price Price Price Price Price Price Emp. Index
Index Index Index Index Index Index '72=1
'72=1 '72=1 '72=1
1972 10.40 1.000 13.10 1.000 7.70 1.000 14.6 1.000 1135160 1.000
1973 11.40 1.096 14.30 1.092 8.20 1.065 15.9 1.089 1196160 1.054
1974 12.70 1.221 15.70 1.198 8.80 1.143 18.2 1.247 1266170 1.115
1975 14.60 1.404 17.10 1.305 10.40 1.351 19.7 1.349 1313140 1.157
1976 16.00 1.538 19.20 1.466 12.20 1.584 22.0 1.507 1355210 1.194
1977 17.90 1.721 20.70 1.580 13.70 1.779 23.8 1.630 1317120 1.160
1978 19.80 1.904 21.60 1.649 15.10 1.961 24.6 1.685 1314650 1.158
1979 22.40 2.154 24.20 1.847 16.90 2.195 25.9 1.774 1336070 1.177 f\
1980 25.50 2.452 27.30 2.084 18.60 2.416 28.2 1.932 1423180 1.254
1981 29.30 2.817 28.50 2.176 21.30 2.766 36.3 2.486 1510480 1.331
1982 33.70 3.240 35.30 2.695 25.20 3.273. 40.8 2.795 1544280 1.360
1983 37.80 3.635 43.00 3.282 23.10 3.000 44.9 3.075 1466890 1.292
1984 42.20 4.058 44.30 3.382 31.50 4.091 49.3 3.377 1478520 1.302
1985 49.10 4.721 54.30 4.145 37.60 4.883 53.7 3.678 1428990 1.259
1986 58.20 5.596 65.90 5.031 51.00 6.623 60.7 4.158 1415560 1.247
1987 67.60 6.500 69.80 5.328 62.90 8.169 70.1 4.801 1429270 1.259
1988 76.30 7.337 75.60 5.771 73.20 9.506 77.0 5.274 1442580 1.271
1989 87.50 8.413 80.90 6.176 80.70 10.48 90.2 6.178 1449760 1.277
1990 100.0 9.615 100.0 7.634 100.0 12.99 100.0 6.849 1451610 1.279
Notes: 1. Source ofC.P.I, Appliance Prices, Motor Vehicle Prices, Furniture Prices and
Manufacturing employment mc (1992) Sectoral Data Series 1972 to 19903.
2. Emp Index ('72=1) are manufacturing employment numbers indexed with 1972 being
the base year.
A4.2. - 2
Table A4.2.2 Instrumental variables: Government and Private Consumption data

Year Real Pvt. Nom Pvt Pvt Con Real Govt Nom Govt GovtCon
Con: '85 Con '72=1 Con: '85 Con '72=1
prices prices
1972 43452 9503 1.000 10991 1937 1.000
1973 46746 11134 1.089 11475 2219 1.097
1974 49198 13106 1.218 12363 2802 1.286
1975 50984 15167 1.360 13859 3687 1.510
1976 51788 17116 1.511 14621 4465 1.733
1977 53219 18914 1.625 15181 5034 1.882
1978 52180 21086 1.848 15283 5526 2.052
1979 53472 24427 2.089 16017 6329 2.242
1980 58065 30797 2.425 17477 8158 2.649
1981 62306 38086 2.795 17808 9877 3.147
1982 63613 44564 3.203 18934 12361 3.704
1983 65525 51596 3.600 19277 14115 4.155
1984 68536 59705 3.983 20589· 17927 4.941
1985 66167 66167 4.572 21297 21297 5.674
1986 66272 77965 5.379 21785 25672 6.687
1987 68827 93353 6.202 22600 30599 7.683
1988 72453 111324 7.026 22975 35276 8.712
1989 74191 131309 8.093 23786 44308 10.57
1990 75319 152475 9.256 24025 50476 11.92
Note 1. Source of Private Consumption Expenditure and Government Consumption Expenditure is:
SARB (1991) South Africa'S National Accounts 1946 to 1990.
2.Pvt. Con ('72=1) emerges from applying the formula
(Nom. Pvt. Con/Real Pvt. Con: '85 prices) and indexing the series by setting the 1972 value
to 1.
3.Nom. Pvt. Con is Nominal Private Consumption Expenditure and Real Pvt. Con: '85 prices
is Real Private Consumption Expenditure with 1985 prices as the base.
4 Govt Con Nom. is Nominal Government Consumption Expenditure and Real Govt. Con: . 85
prices is Real Government Consumption Expenditure with 1985 prices as the base.
A4.2. - 3
Table A4.2.3 Instrumental Variables: Gross Domestic Product and Disposable Income data
Year NomGDP Real GDP: '85 GDPDEF. NomPDY PDY Index
prices '72=1 '72=1
1972 15535 87599 1.000 10836 1.000
1973 19218 91604 1.183 12062 0.941
1974 23690 97202 1.374 14217 0.955
1975 26646 98850 1.520 16810 1.021
1976 30020 101074 1.675 18131 0.999
1977 33263 100979 1.857 21342 1.060
1978 38247 104023 2.073 22884 1.019
1979 45772 107966 2.391 27595 1.065
1980 60328 115114 2.955 34710 1.084
1981 71080 121285 3.305 39067 1.091
1982 80531 120820 3.758 45393 1.115
1983 91457 118589 4.349 52696 1.118
1984 107221 124636 4.851 62556 1.190
1985 123126 123126 5.639 70760 1.158
1986 142135 123148 6.508 80106 1.136
1987 164524 125735 7.378 98091 1.227
1988 198110 130888 8.535 115248 1.246
1989 232532 133636 9.812 133674 1.257
1990 262650 132405 11.19 154888 1.278
Note: 1. Source of Gross Domestic Product and Personal Disposable Income is: SARB (1991) South
Africa's National Accounts 1946 to 1990.
\
2 GDP DEF ('72=1) emerges from applying the formula (Nom. GDP/Rea1 GDP: '85 price~
and indexing the series by setting the 1972 value to 1. Nom GDP is Nominal Gross Domestic
Product and Real GDP: '85 prices is Real GDP with 1985 as the base year. GDP DEF
('72=1) is therefore the GDP detlator with 1972 as the base year.
3 PDY Index ('72=1) emerges from applying the formula (Nom PDY/GDP DEF ('72 = 1»)
and indexing the series by setting the 1972 value to 1. Nom. PDY is Nominal Personal \
Disposable Income.
APPENDIX 5.1
FORMS OF ALTERNATIVE SPECIFICATIONS OF THE TRANSLOG

SYSTEMl
5.1.1 MODEL 1: TECHNICAL CHANGE AUGMENTED NONHOMOTHETIC

FORM
Cost Function:
Cost Shares:
ISpecifications provided below are theroretical specifications. As discussed in Chapters 3 and 4, the
econometric estimation of the Translog requires the dropping of one of the share equations and
nonnalisation of the prices and total costs in the remaining system by the price of the variable whose
share equation has been dropped. The econometric estimation here involved dropping the cost share
of materials.
A5.1 - 2
5.1.2 MODEL 2: NONHOMOTHETIC FORM (CONSTANT TECHNOLOGY)
Cost Function:
Cost Shares:
aInC
a In PM
=SM =aM +"( My In y+"( MK In PK +"( ML In PL +"( MM In PM
5.1.3 MODEL 3: HOMOTHETIC FORM (CONSTANT TECHNOLOGY)
Cost Function:
In C = In a o + a y In Y + +"( yy
2
(In y ) + a K In P K + a L In P L + a M In PM + +"( KK (In P K )
2
Cost Shares:
aInC
= = +Y MK In PK +Y ML In PL +Y MM In PM
aIn PM
SM aM
A5.l - 3
5.1.4 MODEL 4: HOMOGENOUS FORM (CONSTANT TECHNOLOGY)
Cost Function:
Cost Shares:
aloe
aIn PM
=SM =aM +Y MK In PK +Y ML In PL +Y MM In PM
5.1.5 MODEL 5: VARIABLE RETURNS TO SCALE COBB-DOUGLAS

FORM
Cost Function:
Cost Shares:
APPENDIX 5.2 A PRIORI DATA ANALYSIS
A 5.2.1 ELECTRIC APPLIANCES AND HOUSEHOLD GOODS (ISIC 3883)
Table AS.2.1.1 Input-output coefticients. relative input prices and real avera~e costs (ISIC 3833)
Year Input-output Input-output Input-output Relative Relative Relative Real
of Capital of Labour of Materials price of price of price of Average
Capital Labour Materials Costs
1972 0.4174 0.000031304 0.6435 1.0000 1.0000 1.0000 0.1243
1973 0.4184 0.000028227 0.6596 0.9509 1.0654 1.0239 0.1289
1974 0.4000 0.0000252 0.6629 1.0630 1.1677 1.0994 0.1287
1975 0.4167 0.000024792 0.6771 1.1729 1.2416 1.1446 0.1377
1976 0.3795 0.000026359 0.6615 1.4526 1.1605 1.2121 0.1381
1977 0.4783 0.00003413 0.6739 1.6986 1.1747 1.3476 0.1689
1978 0.4191 0.0000325 0.6618 1.6357 1.3408 1.4127 0.1776
1979 0.3081 0.000022733 0.6570 1.4632 1.3528 1.4138 0.1513
1980 0.2311 0.000016978 0.6622 1.5067 1.2921 1.4339 0.1362
1981 0.2174 0.000016826 0.6652 1.9140 1.5882 1.5412 0.1491
1982 0.2385 0.000017752 0.6651 1.9171 1.4302 1.4495 0.1466
1983 0.2176 0.000016065 0.6759 1.8285 1.2191 1.3297 0.1272
1984 0.2205 0.000016205 0.6615 2.2226 1.2032 1.4020 0.1258
1985 0.2268 0.000013969 0.6753 2.2096 1.1904 1.3113 0.1213
1986 0.2098 0.000012829 0.6732 2.1477 1.1273 1.3377 0.1173
1987 0.1783 0.000013217 0.6870 2.3034 1.3653 1.4551 0.1282
1988 0.1565 0.000016107 0.6756 2.4835 1.4149 1.5004 0.1364
1989 0.1633 0.000016414 0.6693 2.8010 1.3745 1.6974 0.1475
1990 0.1486 0.000015543 0.6703 2.5359 1.1307 1.5412 0.1296
Note: 1 Input-output coefficients were calculated by defining the quantities of factors of production as
the constant Rand expenditure on the different factors and the quantity of output as the constant Rand
value of output.
2. Relative prices are defined as the ratio of the price index of each input to the price index of
output as calculated by the mc (1992)
3. Real average costs is detined as the ratio of nominal average costs to the price index of output,
where nominal average costs are calculated as the ratio of current value of total costs to the constant
rand value of output.
Table AS.2.1.2 Descriptive statistics of relative factor prices and input-output coefficients (ISIC
3833)
Variable Mean Standard Deviation.
Relative price of Capital 1.8056 0.54839
Relative price of Labour 1.2547 0.14454
Relative price of Materials 1.3502 0.18409
Input-output coefficient of Capital 0.28662 0.11039
Input-output coefficient of Labour 0.000020903 0.0000069507
Input-output coefficient of Materials 0.66725 0.96169E-02
A5.2 - 2
Figure A.S.2.I.! Input-output coefficient of capital (ISIC 3833)

0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0+--+-+-__t--+----1---+--+--+-+--+--+--+--+-......-+_---i1---+----4
n ~ N n ~ n n ~ 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
Figure A.S.2.1.2 Input-output coefficient of labour (ISIC 3833)

0.000035
0.00003
0.000025
0.00002
0.000015
0.00001
0.000005
O+--+--+--+-+--+-+--+-+--+--I---+----i~+___t~+___t-+-__t
n ~ N n ~ n n ~ 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
A5.2 - 3
Figure A.S.2.1.3 Input-output coefficient of materials (ISIC 3833)

0.69
0.68
0.67
0.66
0.65
0.64
0.63
0.62 ~-+-+---lI---+--t---+--+--+--t--+-+---l-"""-""'-+--+--+---4
n ~ ~ n n n n n 00 ~ ~ ~ M ~ ~ ~ ~ ~ 00
Figure A.S.2.1.4 Relative price of capital, labour and materials (ISIC 3833)
3 --G- Price of Capital
- Price of Labour
2.5 ~ Price of Materials
1.5
0.5
O~-+-+--t--+---f--+-"""'--+-"""'",--+-"""'I---+-"""'I--"""~I---+--I---1
n ~ ~ n n n n n 00 ~ ~ ~ M ~ ~ ~ ~ ~ 00
A5.2 - 4
A5.2.2 FURNITURE (ISle 3220)
TableAS.2.2.1 InQut-output coefficients relative input prices and real average costs (ISIC 3840)
1972 0.1480 1.78E-05 0.6628 1.0000 1.0001 1.0000 0.0940
1973 0.1333 1.66E-05 0.6584 0.9526 1.0391 1.0429 0.0946
1974 0.1247 1.74E-05 0.6540 1.0206 1.0134 1.1288 0.1006
1975 0.1209 1.60E-05 0.6595 1.1326 1.1001 1.1854 0.1045
1976 0.1229 1.68E-05 0.6576 1.4096 1.0109 1.1888 0.1027
1977 0.1182 1.56E-05 0.6465 1.6421 1.0122 1.2306 0.1007
1978 0.1066 l.4lE-05 0.6403 1.5962 1.0763 1.2913 0.1030
1979 0.0928 1.28E-05 0.6377 1.5191 1.1790 1.3795 0.1069
1980 0.0801 1.32E-05 0.6424 1.6088 1.2320 1.4789 0.1157
1981 0.0946 1.45E-05 0.6267 1.6725 1.1121 1.3374 0.1067
1982 0.1098 1.65E-05 0.6011 1.8450 1.1370 1.3433 0.1085
1983 0.1206 1.75E-05 0.6343 1.9470 1.2006 1.3639 0.1181
1984 0.1319 1.9lE-05 0.6324 2.2196 1.1699 1.3390 0.1227
1985 0.1302 1.94E-05 0.6322 2.4823 1.2172 1.3921 0.1252
1986 0.1292 1.96E-05 0.6197 2.5851 1.2230 1.4438 0.1272
1987 0.1251 2.02E-05 0.6361 2.5423 1.1634 1.4594 0.1297
1988 0.1098 1.97E-05 0.6267 2.6884 1.1496 1.5241 0.1310
1989 0.1103 1.92E-05 0.6579 2.7776 0.9990 1.5419 0.1309
1990 0.1174 2.02E-05 0.6644 2.8069 1.0918 1.6477 0.1440
Note: 1 Input-output coefficients were calculated by derming the quantities of factors of production as
value of output.
output as calculated by the IDC (1992)
3. Real average costs is detined as the ratio of nominal average costs to the price index of output,
Table AS.2.2.2 Descriptive statistics of relative factor prices and input-output coefficients (ISle
3840)
Relative price of Capital 1.8657 0.63557
Relative price of Labour 1.1119 0.81819E-Ol
Relative Drice of Materials 1.3326 0.17112
Input-output coefficient of Capital 0.11719 0.16105E-Ol
Input-output coefficient of Labour O.17 I64E-04 0.23575E-05
Input-output coefficient of Materials 0.64161 0.1685lE-Ol
A5.2 - 5
Figure A.5.2.2.1 Input-output coefficient of capital (ISIC 3320)

0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0+--+--+------lf---+--......-+--+--+-+--+-+--+--+------lf---+--......-+---4
n ~ N ~ ~ n n ~ 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
Figure A.5.2.2.2 Input-output coefficient of labour (ISIC 3320)

2.S0E-DS
2.00E-DS
1.S0E-DS
1.00E-DS
5.00E-D6
O.OOE+OO +--+-+--+-+--+--......-+--......-+------If---+--+-+--+-+--+-+---f
n ~ N ~ ~ n n ~ 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ ~
A5.2 - 6
Figure A.S.2.2.3 Input-output coefficient of materials (ISIC 3320)

0.67
0.66
0.65
0.64
0.63
0.62
0.61
0.6
0.59
0.58
0.57
0.56 +-.......--+--+_-t--_+_--4I--......-+_-t--_+_--4I--......-+_-+--+- I---4---4
n n ~ ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ ~ ~ 00
Figure A.S.2.2.4 Relative price of capital, labour and materials (ISIC 3320)
3
- 0 - Price of Capital
- Price of Labour
2.5 ~ Price of Materials
1.5
0.5
0+-~--+--1---+-+_-+-_+_~-_+_ 1---+--+_-+-+_-+--+---41--...
n n ~ ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ ~ ~ 00
A5.2 - 7
AS.2.3 MOTOR VEHICLES PARTS AND ACCESSORIES (ISIC 3840)
Table A5.2.3.1 Input-output coefficients. relative input prices and real average costs (lSIC 3840)
1972 0.1638 5.22E-06 0.7454 1.0000 1.0000 1.0000 0.0725
1973 0.1533 4.93E-06 0.7354 0.9747 1.0545 1.0267 0.0730
1974 0.1416 4.73E-06 0.7425 1.1148 1.0952 1.1083 0.0783
1975 0.1322 4.69E-06 0.7480 1.1335 1.0366 1.1056 0.0778
1976 0.1365 5.33E-06 0.7508 1.3437 0.8797 1.0940 0.0769
1977 0.1532 5.99E-06 0.7513 1.5085 0.8771 1.1016 0.0792
1978 0.1276 5.23E-06 0.7420 1.3753 0.8927 1.1015 0.0766
1979 0.1296 5.38E-06 0.7546 1.2315 0.8859 1.1178 0.0787
1980 0.1148 4.65E-06 0.7518 1.2999 0.9463 1.1481 0.0786
1981 0.1167 4.25E-06 0.7487 1.5053 1.0299 1.1327 0.0771
1982 0.1536 4.96E-06 0.7421 1.5785 1.0107 1.1163 0.0777
1983 0.1777 4.85E-06 0.7490 2.0007 1.3171 1.3289 0.0952
1984 0.2254 5.64E-06 0.7308 1.8372 1.1409 1.0984 0.0832
1985 0.3281 7.31E-06 0.7444 1.8756 1.0613 1.0704 0.0927
1986 0.3385 7.58E-06 0.7412 1.6312 0.9004 0.9904 0.0801
1987 0.2591 6.05E-06 0.7579 1.5024 0.8909 0.9548 0.0729
1988 0.2140 5.21E-06 0.7525 1.5076 0.9181- 0.9313 0.0682
1989 0.2128 5.32E-06 0.7524 1.6505 1.0231 1.0063 0.0749
1990 0.2439 6.18E-06 0.7531 1.4906 0.8911 1.0267 0.0751
Note: 1 Input-output coefficients were calculated by defining the quantities of factors of production as
value of output.
output as calculated by the mc (1992)
3. Real average costs is defined as the ratio of nominal average costs to the price index of output.
Table A5.2.3.2 Descriptive statistics of relative factor prices and input-output coefficients (lSIC
3840)
Relative orice of Capital 1.4506 0.28540
Relative orice of Labour 0.99220 0.11434
Relative orice of Materials 1.0768 0.87638E-Ol
Input-output coefficient of Capital 0.18538 0.67783E-0l
Input-output coefficient of Labour 0.54475E-05 0.86508E-06
Input-outPut coefficient of Materials 0.74704 0.68690E-02
A5.2 - 8
Figure A.5.2.3.1 Input-output coefficient of capital (ISIC 3840)

0.35
0.3
0.25
0.2
0.15
0.1
0.05
O~-+-+--+-+--+-+--+-+--+-+--+--+--+--+--+--+--+----t
n n ~ ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ ~ ~ ~
Figure A.5.2.3.2 Input-output coefficient of labour (ISIC 3840)

8.00E-06
7.00E-Q6
6.00E-06
5.00E-06
4.00E-06
3.00E-06
2.00E-06
1.00E-06
O.OOE+OO +--+-+--+--~-+---+-+--+--t---+---+--+--+-+-"""'--"I---+-~
n n ~ ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ ~ ~ ~
A5.2 - 9
Figure A.5.2.3.3 Input-output coefficient of materials (ISIC 3840)

0.76
0.755
0.75
0.745
0.74
0.735
0.73
0.725
0.72
0.715 +--+--+---4--+---11---+--1---+--1--.......-+--+-+--+-+--+--+---4
n ~ ~ ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ ~ ~ 00
Figure A.5.2.3.4 Relative price of capital, labour and materials (ISIC 3840)
2.5 ~ Price of Capital
- Price of Labour
2 --<>- Price of Materials
1.5
0.5
0+--+--+---11---+--+--+-+---+--+---4--+---11---+--+--+-+--+--1
n ~ ~ ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ ~ ~ 00
APPENDIX 5.3
MOTOR VEHICLES, PARTS AND ACCESSORIES (ISIC 3840)
5.3.1 Hypothesis Testing:
Table AS.3.l.1 LR test statistics and appropriate chi-square statistics for ZEF and 3SLS
estimates of ISIC (3840)
Null Hypothesis Critical Computed Computed Computed Critical Computed
')
L.R. L.R. L.R. 2 L.R.
X~
X
statistic statistic statistic statistic . statistic statistic
and dffor IZEF IZEF I3SLS and df for I3SLS
lZEF, estimates (AR) estimates I3SLS (AR)
lZEF estimates (AR) estimates
(AR) and estimates
13SLS
estimates
Non-homothetic 11.07 (5) 20.539 19.451 20.359
Homotheticitv 14.07 (7) 56.969 48.584 56.627 5.99 (2) 28.564
Homo~eneity in Output 15.51 (8) 57.651 51.428 57.418 7.81 (3) 31.177
V.R.T.S. Cobb-Douglas 19.7 (11) 75.031 86.309 75.361 12.59 (6) 67.549
Note: 1. For IZEF, IZEF (AR) and I3SLS the alternate hypothesis is that the technology augmented
form is valid. For I3SLS (AR) estimates the alternate hypothesis is that the non-homothetic
form is valid.
2. Numbers in parenthesis appearing after X2 indicate Degrees of Freedom (dO (at the 5% level
obtain the specification of the different null hypothesis from the alternative soecification.
..
T a ble AS.3 1 2 LR t es t starIS rICS
< f or comparison 0 f IZEF and IZEF(AR) estimators ISIC 3840
Model 1 2 3 4 5
Likelihood Ratio Statistic 9.608 9.615 14.995 12.797 -5.618
Table AS.3.l.3 LR test statistics for comparison of I3SLS and I3SLS(AR) estimators ISle 3840
Model 2 3 4 5
Likelihood Ratio Statistic 8.833 14.627 12.765 -6.610
A5.3. - 2
Table AS.3.1.4 LR test of Economies of Scale· IZEF estimates (ISIC 3840)

Critical Chi-Squared Computed Wald- Computed Wald-
Statistic IZEF estimates IZEF (AR) estimates
Model 1 (5) 11.071 205.345 441.078
Model 2 (4) 9.488 121.158 151.682
Model 3 (2) 5.991 9.899 79.499
Model 4 (1) 3.841 16.629 23.580
ModelS (1) 3.841 51.028 22.210
Note: 1. Numbers in parenthesis after the Model Number are the number of degrees of freedom
2. Source of critical cbi-squared values: Guiarati (1988, D685)
Table AS.3.l.SLR t~st ofEconomies of Scale - 13SLS estimates (lSIC 3840)

Critical Cbi-Squared Computed Wald- Computed Wald-
Statistic I3SLS estimates I3SLS (AR) estimates
Model 1 (5) 11.071 208.839
Mode12 (4) 9.488 115.586 154.732
Mode13 (2) 5.991 6.954 55.053
Model 4 (1) 3.841 16.648 16.714
Mode15 (1) 3.841 30.906 11.317
2. Source of critical cbi-SQuared values: Guiarati (1988, 0685)
A5.3. - 3
5.3.2 Durbin-Watson Test Results
., ID ur b'ID- W a ts ons tal'IS l'ICS f or IZEF esl'Ima t es 0 f ISIC 3840

3 2 1 C,omputed and cratIca
T a bleAS•.• .
Model Equation Number of DL DU Computed Ho: p=O'
Explanatory's D-W vs.
Statistic Hl: 0>0
Model I Cost function 14 0.070 3.642 1.080 N.D.
Cost share of Capital 4 0.859 1.848 0.989 N.D.
Cost share of Labour 4 0.859 1.848 1.620 N.D.
Model 2 Cost function 9 0.369 2.783 0.618 N.D.
Cost share of Capital 3 0.967 1.685 0.626 Reiect
Mode13 Cost function 7 0.549 2.396 0.838 N.D.
Cost share of Labour 2 1.074 1.536 1.038 Reiect
Model 5 Cost function 3 0.967 1.685 0.804 Reiect
Cost share of Capital -- -- -- -- --
Cost share of Labour -- -- -- -- --
Note: Significant Lower and Upper D-W statistics are given for the 0.05 level of significance (Source:
Gujarati, 1988,0687).
...
T a ble AS 322 C omputed and cratIca
.. ID ur b in- W atson statIstIcs or S S estimates of ISIC3408
Model Equation Number of DL DU Computed Ho:p=O
Exogenous D-W vs.
Variables Statistic HI: D>O
Model 1 Cost function 15 0.063 3.676 1.097 N.D
Model 2 Cost function II 0.220 3.159 0.611 N.D.
Cost share of Capital II 0.220 3.159 0.619 N.D.
Cost share of Labour II 0.220 3.159 1.517 N.D.
Cost share of Caoital II 0.220 3.159 0.759 N.D.
Cost share of Caoital II 0.220 3.159 0.747 N.D.
Gujarati, 1988, p687).
A5.3. - 4
5.3.3 Parameter Estimates
..•1 IZEF narameter estlma t es f or M ot or Ve h'le Ies (ISle 3840)

T a bl eA 533 .
Parameler Model I Model 2 Model 3 Model 4 Model 5
ao 20.5118 20.5454 20.5084 20.5103 20.5255
(1526.62) (2095.70) (2244.63) (2417.36) (2810.09)
aK 0.0700 0.0896 0.0656 0.0664 0.0659
(7.9168) (11.9729) (11.9376) (12.2346) (12.0302)
aL 0.1744 0.1778 0.1647 0.1638 0.1648
(55.7134) (60.9623) (35.7946) (35.5762) (40.9885)
aM 0.7556 0.7326 0.7697 0.7698 0.7693
ay 0.7755 0.8218 0.9610 0.9593 0.9045

(17.1890) (22.1241) (32.1764) (96.0428) (67.6657)
a l
0.0047
(2.4907)
YKK 0.0467 0.0054 0.0109 0.0089
(3.3846) (0.7636) (1.3662) (1.1478)
YKL 0.0430 0.0441 0.0320 0.0336
(5.1074) (8.2463) (3.6116) (3.7887)
YKM -0.0897 -0.0495 -0.0429 -0.0425
Yu 0.0957 0.0978 0.1079 0.1023

(7.7777) (8.1845) (5.3868) (4.7437)
YLM -0.1387 -0.1419 -0.1399 -0.1359
YMM 0.2284 0.1914 0.1828 0.1784
YKY -0.0761 -0.0798

(-3.7649) (-3.2842)
YLY -0.0666 -0.0664
(-10.297) (-9.9745)
YMY 0.1457 0.1462
Yyy 0.1557 0.0215 0.0277

(2.2549) (0.2592) (0.2943)
Y II
-0.0001
(-0.8793)
YYI 5.29E-05
(0.02116)
YKr 0.0005
(0.6687)
Yu 0.0004
(1.0394)
YMt -0.0009
Dtrm 1.91E-13 5.63E-13 3.83E-12 3.97E-12 9.91E-12

NOle: 1. Numbers in parenthesis are I-statistics referring lo parameter estimates above. I-statistics do
nOl exist for indirectly calculated parameters.
2. Dtrm is the determinant of the residual covariance matrix.
A5.3. - 5
Table 5.3.3.2 IZEF (AR) parameter estimates for Motor Vehicles (ISle 3840).
Parameter Model I Model 2 Model 3 Model 4 Model 5
a o 20.6046 20.5633 20.5690 20.5208 20.5213
(587.168) (1236.01) (458.71) (1245.86) (2187.61)
a K 0.0994 0.0850 0.0522 0.0630 0.0614
(8.4328) (8.1679) (2.0149) (6.5489) (11.8744)
a L 0.1786 0.1761 0.1477 0.1438 0.1625
(11.6508) (42.4787) (17.3615) (15.5765) (41.0398)
aM 0.722 0.7389 0.8001 0.7932 0.7761
ay 0.5803 0.7693 0.8189 0.9215 0.9037

(12.1830) (20.9784) (28.8541) (56.9829) (44.1960)
at -0.0025
(-0.6131)
YKK 0.1130 0.0446 0.0714 0.0253
(9.0302) (2.6401) (2.9026) (1.7897)
YKL . 0.0399 0.0523 0.0562 0.0609
(4.5619) (7.2021) (3.1976) (3.9282)
YKM -0.1529 -0.0969 -0.1276 -0.0862
Yu 0.0510 0.0990 0.0089 0.0044

(2.4153) (6.2152) (0.2914) (0.1396)
YLM -0.0909 -0.1513 -0.0651 -0.0649
Y MM 0.2438 0.2482 0.1927 0.1511
YKY -0.1418 -0.1027

(-10.531) (-4.5118)
YLY -0.0786 -0.0680
(-7.4313) (-8.6887)
YMY 0.2204 0.7070
Yyy 0.1028 0.0745 0.1870

(1.4889) (0.7149) (2.0476)
YII 7.64E-05
(0.2898)
YYt 0.0112
(4.2270)
YKJ -0.0024
(-2.3324)
YL/ 6.05E-05
(0.0575)
Y Mt 0.0023
Dtnn 1.12E-13 3.30E-13 1.665E-12 1.95E-12 l.354E-ll

Note: 1. Numbers in parenthesis are t-statistics referring to parameter estimates above. t-statistics do
not exist for indirectly calculated parameters.
2. Dtnn is the detenninant of the residual covariance matrix.
A5.3. - 6
Table A5.3.3.3I3SLS parameter estimates for Motor Vehicles (ISle 3840)

Parameter Model 1 Model 2 Model 3 Model 4 Model 5
0. 0 20.5125 20.5454 20.5064 20.5104 20.5197
(1523.17) (2057.33) (2169.61) (2418.04) (2658.82)
UK 0.0699 0.0904 0.0651 0.0664 0.0643
(7.9301) (11.8879) (11.7853) (12.2421) (11.7320)
a L 0.1743 0.1777 0.1651 0.1638 0.1638
(55.5600) (60.6326) (36.0186) (35.5759) (40.6735)
aM 0.7558 0.7319 0.7698 0.7698 0.7719
ay 0.7682 0.8247 0.9740 0.9590 0.9201

(16.5304) (20.2754) (27.7927) (95.3638) (64.0343)
at 0.0045
(2.3873)
YKK 0.0479 0.0051 0.0120 0.0089
(3.4770) (0.7193) (1.4781) (1.14327)
YKL 0.0437 0.0438 0.0312 0.0337
(5.1641) (8.1860) (3.5276) (3.7935)
YKM -0.0916 -0.0489 -0.0432 -0.0426
YLL 0.0955 0.0977 0.1097 0.1022

(7.7270) (8.2086) (5.6831) (4.7271)
YLM -0.1392 -0.1415 -0.1409 -0.1359
YMM 0.2308 0.1904 0.1841 0.1785
YKY -0.0755 -0.0826

(-3.7405) (-3.3236)
YLY -0.0664 -0.0654
(-10.2371) (-9.7151)
YMY 0.1419 0.1480
Yyy 0.1941 0.0058 -6.482£-05

(2.3689) (0.0596) (-0.0006)
Y II
-0.0001
(-0.7344)
YYt -0.0002
(-0.0755)
YKl 0.0005
(0.6333)
YLt 0.0003
(0.9805)
YMt -0.0008
Dtnn 1.934E-13 5.647E-13 3.809E-12 3.971E-12 1.021E-ll

Note: 1. Numbers in parenthesis are I-statistics referring to parameter estimates above. I-statistics do
2. Dtnn is the determinant of the residual covariance matrix.
AS.3. - 7
Table A5.3.3.4 I3SLS(AR) oarameter estimates for Motor Vehicles (ISle 3840)
Parameter Model 1 Model 2 Model 3 Model 4 ModelS
a o 20.5644 20.5570 20.5166 20.5186
(1216.58) (500.192) (1313.91) (1932.92)
a K 0.0864 0.0545 0.0641 0.0636
(8.1624) (2.1641) (7.2228) (10.5731)
a L 0.1754 0.1443 0.1432 0.1629
(43.7719) (16.1818) (15.5670) (37.4923)
aM 0.7382 0.8012 0.7927 0.7735
ay 0.7611 0.8323 0.9336 0.9234

(17.2166) (24.1256) (57.4715) (40.5366)
at
YKK 0.0455 0.0660 0.0223
(2.6437) (2.8967) (1.7380)
YKL 0.0526 0.0620 0.0633
(7.4699) (3.8317) (4.2116)
YKM -0.0981 -0.1280 -0.0856
Yu 0.0999 0.0012 0.0080

(6.4407) (0.0377) (0.2513)
YUt -0.1525 '-0.0632 -0.0173
YMM 0.2502 0.1912 0.1569
YKY -0.1092
(-4.7814)
YLY -0.0658
(-8.5811)
YMY 0.1750
Yyy 0.0938 0.2343

(0.6466) (1.8995)
YII
YYt
YKt
Yu
YMt
Dtnn 3.457E-13 1.690E-12 1.954E-12
Note: 1. Numbers in parenthesis are I-statistics referring to parameter estimates above. I-statistics do
2. Dtnn is the determinant of the residual covariance matrix.
A5.3. - 8
5.3.4: Monotonicity and Concavity Test Results
... I e d cos t s h ares and 0 wn AES s (:or IZEF estimates Model 1 (ISIC 3840)
T a bl e AS ~ 4 1 F'tt
Year Fitted cost Fitted cost Fitted cost OwnAES OwnAES OwnAES
share of share of share of Capital Labour Materials
Capital Labour Materials
1972 0.0705 0.1748 0.7547 -3.7885 -1.5888 0.0760
1973 0.0607 0.1676 0.7721 -2.7929 -1.5596 0.0880
1974 0.0526 0.1584 0.7898 -1.1250 -1.4986 0.1000
1975 0.0465 0.1500 0.8045 1.0989 -1.4131 0.1099
1976 0.0548 0.1487 0.7973 -1.6904 -1.3964 0.1050
1977 0.0732 0.1644 0.7627 -3.9432 -1.5417 0.0815
1978 0.0594 0.1531 0.7881 -2.5965 -1.4490 0.0989
1979 0.0525 0.1455 0.8027 -1.0962 -1.3518 0.1087
1980 0.0369 0.1341 0.8304 8.2166 -1.1358 0.1270
1981 0.0342 0.1378 0.8299 11.6168 -1.2175 0.1267
1982 0.0443 0.1465 0.8109 2.2042 -1.3668 0.1141
1983 0.0555 0.1616 0.7844 -1.8606 -1.5237 0.0964
1984 0.0740 0.1807 0.7464 -3.9851 -1.6032 0.0702
1985 0.0989 0.1998 0.7014 -4.3369 -1.6077 0.0386
1986 0.0989 0.1947 0.7063 -4.3366 -1.6116 0.0420
1987 0.0822 0.1814 0.7369 -4.2546 -1.6045 0.0636
1988 0.0752 0.1785 0.7474 -4.0381 -1.5986 0.0709
1989 0.0791 0.1837 0.7381 -4.1788 -1.6078 0.0644
1990 0.0767 0.1721 0.7518 -4.0988 -1.5795 0.0739
Table AS.3.4.2. Proper AESs and determinants of matrixes of AESs for IZEF estimates, Modell
(ISIC 3840)
Year Proper Proper Proper Dtrrn of Dtrrn of Dtrrn of Dtrrn of
AES AES AES Matrix 1 Matrix 2 Matrix 3 Matrix 4
Capital : Capital : Labour:
Labour Materials Materials
1972 4.4893 -0.6859 -0.0514 -14.1347 -0.7583 -0.1233 -1.30E-15
1973 5.2302 -0.9154 -0.0719 -22.9997 -1.0836 -0.1424 -0.0138
1974 6.1649 -1.1603 -0.1090 -36.3206 -1.4587 -0.1618 -0.0429
1975 7.1675 -1.3986 -0.1495 -52.9255 -1.8353 -0.1776 -0.0771
1976 6.2810 -1.0542 -0.1702 -37.0910 -1.2888 -0.1757 -0.0413
1977 4.5761 -0.6078 -0.1064 -14.8614 -0.6907 -0.1369 -0.0048
1978 ·5.7292 -0.9167 -0.1494 -29.0614 -1.0970 -0.1656 -0.0285
1979 6.6336 -1.1297 -0.1879 -42.5234 -1.3954 -0.1822 -0.0418
1980 9.6927 -1.9288 -0.2454 -103.2802 -2.6770 -0.2044 -0.2075
1981 10.1103 -2.1560 -0.2127 -116.3609 -3.1769 -0.1994 -0.3332
1982 7.6212 -1.4952 -0.1677 -61.0947 -1.9841 -0.1841 -0.1580
1983 5.7921 -1.0598 -0.0940 -30.7134 -1.3025 -0.1556 -0.0772
1984 4.2158 -0.6241 -0.0283 -11.3842 -0.6693 -0.1134 -0.0227
1985 3.1761 -0.2933 0.0105 -3.1156 -0.2533 -0.0621 -0.0010
1986 3.2326 -0.2837 -0.0087 -3.4611 -0.2627 -0.0678 0.0006
1987 3.8819 -0.4803 -0.0374 -8.2427 -0.5011 -0.1034 -0.0083
1988 4.2056 -0.5969 -0.0399 -11.2314 -0.6425 -0.1149 -0.0200
1989 3.9584 -0.5361 -0.0228 -8.9506 -0.5566 -0.1041 -0.0153
1990 4.2583 -0.5561 -0.0720 -11.6593 -0.6123 -0.1220 -0.Ql15
A5.3. - 9
Table AS.3.4.3 Fitted cost shares and Own AESs for IZEF estimates. Model 2 (lSIC 3840)
Year Fitted cost Fitted cost Fitted cost Own AES OwnAES OwnAES
1972 0.0896 0.1778 0.7326 -9.4881 -1.5306 -0.0084
1973 0.0810 0.1702 0.7488 -10.5244 -1.4994 0.0059
1974 0.0695 0.1606 0.7699 -12.2682 -1.4347 0.0240
1975 0.0618 0.1517 0.7864 -13.7621 -1.3426 0.0379
1976 0.0623 0.1499 0.7879 -13.6673 -1.3183 0.0391
1977 0.0763 0.1653 0.7585 -11.1802 -1.4700 0.0143
1978 0.0653 0.1536 0.7811 -13.0417 -1.3647 0.0335
1979 . 0.0630 0.1453 0.7916 -13.5058 -1.2503 0.0422
1980 0.0449 0.1337 0.8213 -18.5799 -1.0094 0.0662
1981 0.0346 0.1375 0.8280 -23.4084 -1.0990 0.0714
1982 0.0419 0.1458 0.8123 -19.7816 -1.2576 0.0590
1983 0.0503 0.1608 0.7889 -16.7455 -1.4362 0.0400
1984 0.0645 0.1796 0.7559 -13.2019 -1.5360 0.0120
1985 0.0881 0.1983 0.7136 -9.6543 -1.5558 -0.0254
1986 0.0904 0.1924 0.7171 -9.3983 -1.5555 -0.0223
1987 0.0744 0.1788 0.7468 -11.4719 -1.5338 0.0041
1988 0.0651 0.1756 0.7592 -13.0820 -1.5232 0.0149
1989 0.0681 0.1806 0.7513 -12.5115 -1.5385 0.0081
1990 0.0705 0.1681 0.7614 -12.0915 -1.4876 0.0168
Table AS.3.4.4. Proper AESs and determinants of matrixes of AESs for IZEF estimates, Model 2
(ISIC 3840'
Year Proper Proper Proper Dtnn of Dtnn of Dtnn of Dtnn of
Capital: Capital: Labour:
1972 3.7682 0.2459 -0.0894 0.3232 0.0190 0.0048 1.75E-17
1973 4.1991 0.1837 -0.1133 -1.8528 -0.0957 -0.0217 -2.23E-17
1974 4.9506 0.0751 -0.1477 -6.9067 -0.3005 -0.0563 2.01E-16
1975 5.7010 -0.0181 -0.1891 -14.0242 -0.5221 -0.0867 -4. 14E-16
1976 5.7255 -0.0090 -0.2017 -14.7633 -0.5343 -0.0922 -4.14E-16
1977 4.4981 0.1445 -0.1322 -3.7976 -0.1803 -0.0384 -2.94E-16
1978 5.3959 0.0300 -0.1830 -11.3183 -0.4374 -0.0792 -4.00E-16
1979 5.8143 0.0080 -0.2334 -16.9199 -0.5702 -0.1073 2.82E-16
1980 8.3381 -0.3412 -0.2918 -50.7694 -1.3462 -0.1520 -1.99E-15
1981 10.2802 -0.7294 -0.2468 -79.9565 -2.2039 -0.1394 1.92E-15
1982 8.2165 -0.4536 -0.1983 -42.6332 -1.3730 -0.1135 -3.29E-15
1983 6.4532 -0.2473 -0.1188 -17.5944 -0.7306 -0.0715 1.13E-16
1984 4.8051 -0.0150 -0.0451 -2.8105 -0.1587 -0.0205 -5.88E-17
1985 3.5247 0.2127 -0.0030 2.5964 0.2004 0.0396 1.95E-17
1986 3.5344 0.2367 -0.0283 2.1273 0.1532 0.0338 -1.37E-16
1987 4.3163 0.1086 -0.0625 -1.0356 -0.0594 -0.0103 3.59E-17
1988 4.8554 -0.0011 -0.0641 -3.6483 -0.1952 -0.0268 -1.56E-16
1989 4.5841 0.0332 -0.0461 -1.7650 -0.1019 -0.0145 6.15E-17
1990 4.7202 0.0784 -0.1090 -4.2932 -0.2091 -0.0368 3.55E-16
AS.3. - 10
Table AS.3.4.S Fitted cost shares and Own AESs for IZEF estimates. Model 3 (ISIC 3840)
Year Fitted cost Fitted cost Fitted cost Own AES OwnAES Own AES
Caoital Labour Materials
1972 0.0656 0.1647 0.7697 -11.7110 -1.0939 0.0093
1973 0.0659 0.1659 0.7682 -11.6660 -1.1076 0.0080
1974 0.0653 0.1636 0.7711 -11.7606· -1.0811 0.0106
1975 0.0638 0.1585 0.7776 -11.9947 -1.0148 0.0164
1976 0.0609 0.1478 0.7914 -12.4875 -0.8256 0.0283
1977 0.0617 0.1502 0.7881 -12.3383 -0.8745 0.0254
1978 0.0613 0.1491 0.7896 -12.4131 -0.8540 0.0267
1979 0.0592 0.1427 0.7981 -12.7787 -0.7093 0.0340
1980 0.0608 0.1478 0.7914 -12.5043 -0.8269 0.0283
1981 0.0657 0.1635 0.7708 -11.7023 -1.0803 0.0103
1982 0.0662 0.1651 0.7687 -11.6193 -1.0980 0.0085
1983 0.0698 0.1768 0.7534 -11.0928 -1.2045 -0.0053
1984 0.0724 0.1853 0.7423 -10.7295 -1.2540 -0.0154
1985 0.0714 0.1817 0.7468 -10.8619 -1.2355 -0.0113
1986 0.0680 0.1704 0.7616 -11.3503 -1.1523 0.0022
1987 0.0683 0.1717 0.7600 -11.3013 -1.1643 0.0006
1988 0.0704 0.1786 0.7510 -11.0062 -1.2163 -0.0074
1989 0.0715 0.1823 0.7462 -10.8507 -1.2388 -0.0119
1990 0.0651 0.1614 0.7735 -11.7837 -1.0532 0.0127
(lSIC 3840
Year Proper Proper Proper Dtrrn of Dtnn of Dtrrn of Dtrrn of
1972 3.9618 0.1504 -0.1036 -2.8847 -0.1321 -0.0210 -2.73E-16
1973 3.9269 0.1524 -0.0976 -2.4985 -0.1166 -0.0184 -1.50E-16
1974 3.9963 0.1478 -0.1090 -3.2562 -0.1466 -0.0233 -2.47E-16
1975 4.1632 0.1354 -0.1347 -5.1604 -0.2145 -0.0347 -4.96E-16
1976 4.5583 0.1093 -0.1964 -10.4685 -0.3649 -0.0619 -3.93E-16
1977 4.4516 0.1182 -0.1821 -9.0274 -0.3278 -0.0554 -1.02E-15
1978 4.5006 0.1136 -0.1881 -9.6549 -0.3444 -0.0582 -6.81E-16
1979 4.7865 0.0922 -0.2283 -13.8458 -0.4428 -0.0762 -3.45E-16
1980 4.5625 0.1080 -0.1959 -10.4765 -0.3655 -0.0618 -5.65E-16
1981 3.9803 0.1523 -0.1098 -3.2000 -0.1440 -0.0232 -3.17E-16
1982 3.9288 0.1570 -0.1025 -2.6771 -0.1234 -0.0198 -3.78E-16
1983 3.5935 0.1839 -0.0501 0.4480 0.0247 0.0038 7.33E-18
1984 3.3849 0.2020 -0.0173 1.9972 0.1244 0.0190 1.32E-16
1985 3.4648 0.1959 -0.0308 1.4155 0.0838 0.0130 1.26E-16
1986 3.7624 0.1715 -0.0781 -1.0765 -0.0539 -0.0086 -1.55E-16
1987 3.7274 0.1738 -0.0720 -0.7346 -0.0375 -0.0059 -9.40E-17
1988 3.5457 0.1885 -0.0431 0.8144 0.0460 0.0072 1.58E-17
1989 3.4540 0.1961 -0.0284 1.5124 0.0903 0.0139 -3.28E-17
1990 4.0448 0.1485 -0.1209 -3.9502 -0.1719 -0.0280 -5.32E-16
A5.3. - 11
Table A5.3.4.7 Fitted cost shares and Own AESs for IZEF estimates, Model 4 (lSIC 3840)
Year Fiuedcost Fiued cost Fiued cost OwnAES OwnAES Own AES
1972 0.0664 0.1638 0.7698 -12.0416 -1.2922 0.0020
1973 0.0668 0.1648 0.7684 -11.9693 -1.3012 0.0007
1974 0.0661 0.1628 0.7712 -12.0998 -1.2824 0.0033
1975 0.0645 0.1580 0.7775 -12.3723 -1.2317 0.0089
1976 0.0609 0.1484 0.7907 -13.0198 -1.0934 0.0206
1977 0.0615 0.1511 0.7874 -12.8992 -1.1367 0.0177
1978 0.0613 0.1498 0.7889 -12.9417 -1.1162 0.0191
1979 0.0595 0.1433 0.7973 -13.3025 -0.9960 0.0264
1980 0.0610 0.1482 0.7908 -12.9996 -1.0898 0.0207
1981 0.0657 0.1636 0.7706 -12.1529 -1.2905 0.0028
1982 0.0661 0.1653 0.7686 -12.0844 -1.3054 0.0009
1983 0.0697 0.1766 0.7536 -11.5086 -1.3826 -0.0128
1984 0.0723 0.1850 0.7428 -11.1350 -1.4163 -0.0230
1985 0.0711 0.1818 0.7471 -11.3033 -1.4053 -0.0189
1986 0.0676 0.1708 0.7615 -11.8391 -1.3482 -0.0055
1987 0.0681 0.1719 0.7600 -11.7643 -1.3556 -0.0070
1988 0.0702 0.1785 0.7513 -11.4380 -1.3916 -0.0150
1989 0.0714 0.1821 0.7465 -11.2657 -1.4066 -0.0194
1990 0.0650 0.1618 0.7732 -12.2847 -1.2732 0.0051
Table A3.5.4.8. Proper AESs and determinants of matrixes of AESs for IZEF estimates, Model 4
(lSIC 3840)
1972 4.0893 0.1685 -0.0778 -1.1624 -0.0526 -0.0086 -1.12E-16
1973 4.0505 0.1724 -0.0733 -0.8320 -0.0383 -0.0063 -6.36E-17
1974 4.1252 0.1656 -0.0826 -1.4997 -0.0668 -0.Ql1O -2.69E-16
1975 4.2984 0.1519 -0.1060 -3.2369 -0.1337 -0.0222 -2.67E-16
1976 4.7174 0.1175 -0.1581 -8.0180 -0.2825 -0.0476 -7.02E-16
1977 4.6144 0.1230 -0.1426 -6.6307 -0.2440 -0.0405 -6.68E-16
1978 4.6589 0.1214 -0.1502 -7.2601 -0.2616 -0.0439 -5.05E-16
1979 4.9447 0.1034 -0.1897 -11.2008 -0.3617 -0.0623 -4.68E-16
1980 4.7161 0.1191 -0.1596 -8.0746 -0.2836 -0.0481 -9.93E-16
1981 4.1238 0.1610 -0.0777 -1.3221 -0.0596 -0.0096 -2.19E-16
1982 4.0737· 0.1640 -0.0699 -0.8199 -0.0379 -0.0061 -8.48E-17
1983 3.7274 0.1914 -0.0209 2.0177 0.1108 0.0173 2.32E-16
1984 3.5139 0.2081 0.0109 3.4226 0.2123 0.0324 1.26E-16
1985 3.5996 0.2000 -0.0007 2.9272 0.1733 0.0265 1.50E-16
1986 3.9082 0.1749 -0.0447 0.6875 0.0346 0.0054 6.70E-17
1987 3.8693 0.1789 -0.0401 0.9763 0.0500 0.0078 9.05E-17
1988 3.6808 0.1942 -0.0133 2.3693 0.1338 0.0207 1.48E-16
1989 3.5854 0.2022 0.0004 2.9907 0.1780 0.0273 1.45E-16
1990 4.1958 0.1539 -0.0860 -1.9635 -0.0860 -0.0139 -2. 17E-16
AS.3. - 12
Table A5.3.4.9 Fitted cost shares and Own AESs for IZEF AR) estimates, Modell (ISIC 3840)
Year Fitted COSl Fitted COSl Fitted COSl Own AES Own AES Own AES
share of share of share of Capital Labour Malerials
Capital Labour Malerials
1972 0,0970 0.1787 0.7243 2.7005 -2.9994 0.0841
1973 0.0729 0.1686 0.7584 8.5478 -3.1365 0.1053
1974 0.0576 0.1591 0.7832 17.7306 -3.2705 0.1206
1975 0.0455 0.1519 0.8025 33.6065 -3.3736 0.1325
1976 0.0685 0.1574 0.7739 10.4956 -3.2941 0.1149
1977 0.1026 0.1750 0.7221 1.9859 -3.0490 0.0827
1978 0.0705 0.1612 0.7680 9.5444 -3.2407 0.1113
1979 0.0520 0.1545 0.7932 23.6159 -3.3358 0.1268
1980 0.0187 0.1380 0.8430 270.4179 -3.5686 0.1568
1981 0.0107 0.1341 0.8548 898.2686 -3.6208 0.1638
1982 0.0279 0.1435 0.8281 110.0835 -3.4915 0.1479
1983 0.0434 0.1547 0.8014 37.9075 -3.3331 0.1318
1984 0.0754 0.1728 0.7514 7.6252 -3.0796 0.1009
1985 0.1215 0.1973 0.6807 0.4254 -2.7582 0.0571
1986 0.1201 0.1969 0.6824 0.5062 -2.7638 0.0581
1987 0.0835 0.1797 0.7362 5.2428· -2.9854 0.0915
1988 0.0655 0.1721 0.7618 12.0897 -3.0891 0.1074
1989 0.0686 0.1758 0.7549 10.4330 -3.0382 0.1031
1990 0.0639 0.1728 0.7625 13.0086 -3.0791 0.1079
Table A5.3.4.l0. Proper AESs and determinants of matrixes of AESs for IZEF (AR) estimates,
Model 1 (ISIC 3840)
.Capital : Capital: Labour:
Labour Malerials Malerials
1972 3.3024 -1.1763 0.2975 -19.0056 -1.1566 -0.3407 0.0014
1973 4.2457 -1.7658 0.2893 -44.8360 -2.2181 -0.4139 0.0062
1974 5.3565 -2.3914 0.2705 -86.6799 -3.5798 -0.4678 0.0167
1975 6.7747 -3.1876 0.2541 -159.2706 -5.7097 -0.5114 0.0391
1976 4.7010 -1.8853 0.2539 -56.6731 -2.3485 -0.4430 0.0187
1977 3.2218 -1.0633 0.2807 -16.4352 -0.9663 -0.3311 0.0075
1978 4.5100 -1.8234 0.2658 -51.2711 -2.2627 -0.4312 0.0240
1979 5.9705 -2.7102 0.2583 -114.4251 -4.3506 -0.4897 0.0574
1980 16.4569 -8.6957 0.2185 -1235.8405 -33.2140 -0.6073 0.6180
1981 28.8591 -15.7501 0.2071 -4085.3413 -100.9273 -0.6360 2.2074
1982 10.9542 -5.6116 0.2352 -504.3532 -15.2035 -0.5719 0.3194
1983 6.9401 -3.3941 0.2668 -174.5136 -6.5239 -0.5105 0.1287
1984 4.0644 -1.7001 0.2997 -40.0020 -2.1207 -0.4007 0.0364
1985 2.6646 -0.8492 0.3232 -8.2735 -0.6968 -0.2618 0.0099
1986 2.6873 -0.8651 0.3234 -8.6205 -0.7191 -0.2653 0.0110
1987 3.6606 -1.4887 0.3129 -29.0518 -1.7364 -0.3711 0.0339
1988 4.5419 -2.0658 0.3065 -57.9745 -2.9689 -0.4258 0.0670
1989 4.3084 -1.9524 0.3150 -50.2601 -2.7359 -0.4126 0.0627
1990 4.6121 -2.1367 0.3101 -61.3264 -3.1623 -0.4283 0.0791
A5.3. - 13
Table AS.3.4.11 Fitted cost shares and Own AESs for IZEF (AR) estimates. Model 2 ISIC 3840)
1972 0.0850 0.1761 0.7389 -4.5917 -1.4862 0.1012
1973 0.0718 0.1679 0.8242 -4.2781 -1.4443 0.1521
1974 0.0594 0.1585 0.9137 -3.1974 . -1.3687 0.2029
1975 0.0505 0.1497 0.9683 -1.3070 -1.2618 0.2320
1976 0.0585 0.1492 0.9188 -3.0670 -1.2555 0.2056
1977 0.0807 0.1658 0.7869 -4.5435 -1.4297 0.1300
1978 0.0631 0.1531 0.8927 -3.6423 -1.3081 0.1913
1979 0.0555 0.1438 0.9139 -2.5372 -1.1666 0.2029
1980 0.0331 0.1321 1.0822 11.5676 -0.8965 0.2879
1981 0.0253 0.1369 1.1897 31.1176 -1.0230 0.3348
1982 0.0371 0.1459 1.1205 6.4225 -1.2031 0.3052
1983 0.0499 0.1616 1.0649 -1.1189 -1.3970 0.2798
1984 0.0719 0.1816 0.9450 -4.2819 -1.5047 0.2197
1985 0.1042 0.2011 0.7226 -4.4887 -1.5247 0.0914
1986 0.1053 0.1947 0.6843 -4.4747 -1.5245 0.0687
1987 0.0828 0.1805 0.8351 -4.5714 -1.5014 0.1584
1988 0.0717 0.1773 0.9287 -4.2715 -1.4910 0.2110
1989 0.0759 0.1824 0.9087 -4.4346 -1.5069 0.2001
1990 0.0751 0.1689 0.8556 -4.4090 -1.4506 0.1703
Table AS.3.4.12. Proper AESs and determinants of matrixes of AESs for IZEF (AR) estimates,
Model 2 (lSIC 3840)
Year Proper Proper Proper Dtrm of Dtrm of Dtrm of Dtrm of
1972 4.4940 -0.5428 -0.1628 -13.3719 -0.7595 -0.1770 -1.42E-15
1973 5.3352 -0.6366 -0.0932 -22.2858 -1.0558 -0.2283 -2.1333
1974 6.5522 -0.7848 -0.0445 -38.5553 -1.2645 -0.2796 -6.5142
1975 7.9229 -0.9824 -0.0440 -61.1230 -1.2684 -0.2947 -12.2746
1976 6.9875 -0.8018 -0.1035 -44.9749 -1.2735 -0.2689 -7.2480
1977 4.9088 -0.5256 -0.1600 -17.6009 -0.8669 -0.2115 -0.9508
1978 6.4169 -0.7212 -0.1070 -36.4125 -1.2169 -0.2616 -5.2523
1979 7.5538 -0.9108 -0.1513 -54.1007 -1.3444 -0.2596 -7.8717
1980 12.9786 -1.7088 -0.0584 -178.8139 0.4105 -0.2615 -46.3137
1981 16.0915 -2.2181 0.0713 -290.7681 5.4987 -0.3476 -97.5676
1982 10.6560 -1.3293 0.0745 -121.2772 0.1933 -0.3728 -37.0377
1983 7.4903 -0.8247 0.1207 -54.5416 -0.9932 -0.4054 -15.7850
1984 5.0042 -0.4258 0.1183 -18.5995 -1.1221 -0.3446 -4.2581
1985 3.4955 -0.2865 -0.0414 -5.3747 -0.4925 -0.1411 -0.2755
1986 3.5510 -0.3451 -0.1353 -5.7875 -0.4265 -0.1230 0.1979
1987 4.5016 -0.4021 -0.0039 -13.4010 -0.8860 -0.2379 -1.8665
1988 5.1131 -0.4552 0.0813 -19.7747 -1.1085 -0.3212 -4.2138
1989 4.7748 -0.4042 0.0873 -16.1160 -1.0507 -0.3091 -3.2816
1990 5.1201 -0.5073 -0.0467 -19.8199 -1.0081 -0.2492 -2.7493
A5.3. - 14
Table AS.3.4.13 Fitted cost shares and Own AESs for IZEF (AR) estimates Model 3 (ISIC 3840)
Year Fitted cost Fitted cost Fitted cost Own AES Own AES Own AES
CaDital Labour Materials
1972 0.0522 0.1477 0.8001 8.0463 -5.3625 0.0512
1973 0.0500 0.1450 0.8050 9.5622 -5.4724 0.0551
1974 0.0519 0.1479 0.8001 8.2119 -5.3537 0.0512
1975 0.0504 0.1485 0.8011 9.2989 -5.3294 0.0520
1976 0.0546 0.1573 0.7881 6.6203 -4.9971 0.0413
1977 0.0618 0.1633 0.7748 3.4999 -4.7886 0.0304
1978 0.0562 0.1583 0.7854 5.7908 -4.9617 0.0392
1979 0.0461 0.1511 0.8029 12.9505 -5.2292 0.0534
1980 0.0502 0.1530 0.7968 9.4105 -5.1573 0.0485
1981 0.0672 0.1628 0.7700 1.9401 -4.8055 0.0263
1982 0.0713 0.1663 0.7624 1.0099 -4.6919 0.0198
1983 0.0809 0.1706 0.7485 -0.4531 -4.5554 0.0079
1984 0.0911 0.1769 0.7320 -1.3715 -4.3671 -0.0065
1985 0.0918 0.1792 0.7291 -1.4189 -4.3046 -0.0091
1986 0.0825 0.1749 0.7426 -0.6273 -4.4269 0.0029
1987 0.0807 0.1726 0.7468 -0.4244 -4.4963 0.0065
1988 0.0858 0.1746 0.7396 -0.9552 -4.4341 0.0002
1989 0.0885 0.1757 0.7359 -1.1800 -4.4046 -0.0031
1990 0.0709 0.1674 0.7617 1.1066 -4.6563 0.0193
Table AS.3.4.14. Proper AESs and determinants of matrixes of AESs for IZEF estimates, Model
3 (lSIC 3840)
Year Proper Proper Proper Dtrm of Dtnu of Dtrm of Dtrm of
1972 8.2893 -2.0552 0.4491 -111.8606 -3.8120 -0.4761 -1.l2E-14
1973 8.7511 -2.1704 0.4423 -128.9107 -4.1838 -0.4973 -2.08E-15
1974 8.3146 -2.0702 0.4500 -113.0974 -3.8651 -0.4766 -1.34E-14
1975 8.5142 -2.1630 0.4529 -122.0486 -4.1951 -0.4822 -1.67E-14
1976 7.5396 -1.9640 0.4749 -89.9281 -3.5836 -0.4321 -1.86E-14
1977 6.5639 -1.6631 0.4856 -59.8443 -2.6595 -0.3812 -4.41E-15
1978 7.3118 -1.8884 0.4764 -82.1942 -3.3390 -0.4215 -1.58E-14
1979 9.0774 -2.4510 0.4633 -150.1199 -5.3156 -0.4939 -2.54E-14
1980 8.3187 -2.1897 0.4659 -117.7341 -4.3382 -0.4673 -2.29E-14
1981 6.1390 -1.4675 0.4808 -47.0105 -2.1024 -0.3576 -6.50E-15
1982 5.7368 -1.3458 0.4865 -37.6493 -1.7912 -0.3298 -4. 13E-15
1983 5.0710 -1.1069 0.4902 -23.6507 -1.2289 -0.2764 -1.90E-15
1984 4.4876 -0.9142 0.4974 -14.1488 -0.8268 -0.2190 -1.36E-15
1985 4.4183 -0.9071 0.5016 -13.4130 -0.8099 -0.2125 -1.34E-15
1986 4.8966 -1.0835 0.4988 -21.1995 -1.1757 -0.2614 -4.69E-15
1987 5.0373 -1.1181 0.4948 -23.4667 -1.2530 -0.2738 -6.33E-15
1988 4.7510 -1.0111 0.4960 -18.3364 -1.0225 -0.2467 -2.46E-15
1989 4.6169 -0.9602 0.4964 -16.1184 -0.9184 -0.2329 -2.78E-15
1990 5.7377 -1.3638 0.4895 -38.0741 -1.8386 -0.3295 -1.06E-14
A5.3. - 15
Table A5.3.4.15 Fitted cost shares and Own AESs for IZEF (AR) estimates, Model 4 ISIC 3840)
1972 0.0630 0.1438 0.7932 -8.4986 -5.7413 -0.0206
1973 0.0633 0.1408 0.7959 -8.4828 -5.8808 -0.0179
1974 0.0624 0.1442 0.7935 -8.5271 -5.7235 -0.0203
1975 0.0597 0.1452 0.7952 -8.6516 -5.6786 -0.0186
1976 0.0549 0.1556 0.7896 -8.8202 -5.2457 -0.0241
1977 0.0571 0.1622 0.7809 -8.7542 -4.9976 -0.0328
1978 0.0558 0.1567 0.7877 -8.7950 -5.2022 -0.0260
1979 0.0513 0.1490 0.7999 -8.8796 -5.5116 -0.0140
1980 0.0544 0.1509 0.7950 -8.8339 -5.4329 -0.0187
1981 0.0644 0.1612 0.7749 -8.4275 -5.0357 -0.0389
1982 0.0657 0.1650 0.7698 -8.3589 -4.8997 -0.0441
1983 0.0728 0.1692 0.7585 -7.9619 -4.7555 -0.0557
1984 0.0783 0.1759 0.7464 -7.6430 -4.5429 -0.0686
1985 0.0767 0.1786 0.7454 -7.7388 -4.4616 -0.0696
1986 0.0698 0.1745 0.7564 -8.1327 -4.5856 -0.0580
1987 0.0702 0.1719 0.7586 -8.1084 -4.6677 -0.0556
1988 0.0743 0.1739 0.7526 -7.8751 -4.6035 -0.0620
1989 0.0765 0.1749 0.7495 -7.7472 -4.5722 -0.0653
1990 0.0638 0.1669 0.7702 -8.4579 -4.8330 -0.0436
Table A5.3.4.16. Proper AESs and determinants of matrixes of AESs for IZEF (AR) estimates,
Model 4 (ISIC 3840)
1972 7.7223 -0.7250 0.4310 -10.8406 -0.3509 -0.0677 -0.0068
1973 7.8316 -0.7104 0.4208 -11.4484 -0.3531 -0.0720 -0.0080
1974 7.7662 -0.7404 0.4328 -11.5095 -0.3752 -0.0712 -0.0088
1975 8.0252 -0.8155 0.4379 -15.2748 -0.5045 -0.0864 -0.0128
1976 8.1267 -0.9875 0.4717 -19.7757 -0.7628 -0.0962 -0.0168
1977 7.5777 -0.9340 0.4876 -13.6725 -0.5852 -0.0739 -0.0128
1978 7.9620 -0.9604 0.4742 -17.6398 -0.6937 -0.0896 -0.0176
1979 8.9664 -1.1008 0.4556 -31.4547 -1.0878 -0.1306 -0.0326
1980 8.4219 -0.9942 0.4591 -22.9341 -0.8227 -0.1089 -0.0264
1981 6.8676 -0.7273 0.4803 -4.7257 -0.2012 -0.0348 -0.0063
1982 6.6176 -0.7041 0.4890 -2.8364 -0.1274 -0.0232 -0.0041
1983 5.9425 -0.5608 0.4944 2.5497 0.1293 0.0207 0.0044
1984 5.4200 -0.4744 0.5057 5.3446 0.2989 0.0558 0.0099
1985 5.4476 -0.5083 0.5125 4.8504 0.2804 0.0480 0.0094
1986 5.9981 -0.6323 0.5083 1.3162 0.0717 0.0075 0.0027
1987 6.0426 -0.6175 0.5024 1.3347 0.0697 0.0072 0.0030
1988 5.7111 -0.5412 0.5042 3.6366 0.1950 0.0309 0.0083
1989 5.5488 -0.5029 0.5050 4.6331 0.2527 0.0434 0.0113
1990 6.7176 -0.7538 0.4952 -4.2492 -0.1995 -0.0345 -0.0096
A5.3. - 16
Table AS.3.4.l7 Fitted cost shares and Own AESs for I3SLS estimates. Modell (lSIC 3840)
1972 0.0704 0.1746 0.7550 -3.5398 -1.5947 0.0804
1973 0.0606 0.1673 0.7721 -2.4553 -1.5651 0.0920
1974 0.0526 0.1580 0.7894 -0.7045 -1.5037 0.1036
1975 0.0466 0.1496 0.8039 1.6184 -1.4170 0.1132
1976 0.0549 0.1483 0.7968 -1.3241 -1.4007 0.1085
1977 0.0733 0.1640 0.7627 -3.7279 -1.5465 0.0857
1978 0.0595 0.1526 0.7879 -2.2807 -1.4515 0.1026
1979 0.0525 0.1447 0.8028 -0.6543 -1.3501 0.1125
1980 0.0371 0.1333 0.8296 8.8771 -1.1282 0.1299
1981 0.0348 0.1371 0.8281 11.7939 -1.2125 0.1290
1982 0.0449 0.1457 0.8094 2.4774 -1.3642 0.1168
1983 0.0562 0.1607 0.7831 -1.6302 -1.5249 0.0994
1984 0.0748 0.1797 0.7455 -3.8058 -1.6075 0.0739
1985 0.0995 0.1987 0.7018 -4.2121 -1.6138 0.0437
1986 0.0993 0.1934 0.7072 -4.2126 -1.6174 0.0475
1987 0.0827 0.1801 0.7371 -4.0891 -1.6082 0.0682
1988 0.0758 0.1771 0.7471 -3.8571 -1.6016 0.0750
1989 0.0798 0.1822 0.7380 -4.0096 -1.6117 0.0687
1990 0.0770 0.1704 0.7525 -3.9097 -1.5797 0.0787
Table AS.3.4.l8. Proper AESs and determinants of matrixes of AESs for I3SLS estimates, Model
1 (ISIC 3840)
Year Proper Proper Proper Dtnn of Dtnn of Dtrm of Dtnn of
AES AES . AES Matrix 1 Matrix 2 Matrix 3 Matrix 4
1972 4.5552 -0.7234 -0.0560 -15.1049 -0.8078 -0.1313 -8.67E-16
1973 5.3124 -0.9582 -0.0777 -24.3787 -1.1441 -0.1501 -3.74E-15
1974 6.2559 -1.2053 -0.1160 -38.0767 -1.5257 -0.1692 -8.16E-15
1975 7.2750 -1.4474 -0.1577 -55.2194 -1.9118 -0.1852 -2.79E-15
1976 6.3669 -1.0938 -0.1780 -38.6833 -1.3400 -0.1837 -5.13E-15
1977 4.6360 -0.6382 -0.1131 -15.7268 -0.7266 -0.1453 -1.08E-15
1978 5.8125 -0.9531 -0.1581 -30.4742 -1.1424 -0.1739 -3.7IE-15
1979 6.7569 -1.1753 -0.1981 -44.7729 -1.4549 -0.1911 -3.58E-15
1980 9.8403 -1.9783 -0.2584 -106.8474 -2.7602 -0.2134 -1.09E-14
1981 10.1573 -2.1771 -0.2263 -117.4711 -3.2182 -0.2076 -1.52E-14
1982 7.6788 -1.5194 -0.1806 -62.3435 -2.0191 -0.1920 -8.29E-15
1983 5.8372 -1.0810 -0.1060 -31.5868 -1.3306 -0.1628 -5.74E-15
1984 4.2522 -0.6435 -0.0389 -11.9633 -0.6954 -0.1203 -2.63E-15
1985 3.2104 -0.3122 0.0020 -3.5090 -0.2814 -0.0705 -1.07E-15 .
1986 3.2740 -0.3038 -0.0175 -3.9055 -0.2922 -0.0771 -9.27E-16
1987 3.9324 -0.5019 -0.0484 -8.8878 -0.5306 -0.1120 -1.13E-15
1988 4.2545 -0.6169 -0.0522 -11.9232 -0.6699 -0.1228 -2.23£-15
1989 4.0050 -0.5553 -0.0351 -9.5772 -0.5839 -0.1120 -2.52£-15
1990 4.3283 -0.5800 -0.0853 -12.5578 -0.6441 -0.1316 -2.37£-15
A5.3.-17
Table AS.3.4.19 Fitted cost shares and Own AESs for I3SLS estimates, Model 2 (lSlC 3840)
Year Fitted cost Filled cost Fitted cost Own AES Own AES Own AES
1972 0.0904 0.1777 0.7319 -9.4379 -1.5335 -0.0109
1973 0.0815 0.1702 0.7483 -10.5073 -1.5030 0.0037
1974 0.0696 0.1607 0.7696 -12.3101 -1.4398 0.0221
1975 0.0618 0.1519 0.7863 -13.8554 -1.3496 0.0362
1976 0.0624 0.1500 0.7876 -13.7139 -1.3241 0.0373
1977 0.0769 0.1651 0.7580 -11.1431 -1.4730 0.0121
1978 0.0656 0.1536 0.7808 -13.0670 -1.3696 0.0316
1979 0.0633 0.1454 0.7913 -13.5288 -1.2568 0.0403
1980 0.0445 0.1341 0.8215 -18.9104 -1.0238 0.0648
1981 0.0335 0.1379 0.8286 -24.3215 -1.1149 0.0704
1982 0.0411 0.1461 0.8128 -20.3286 -1.2682 0.0579
1983 0.0495 0.1611 0.7894 -17.1075 -1.4425 0.0388
1984 0.0641 0.1797 0.7561 -13.3557 -1.5394 0.0105
1985 0.0886 0.1980 0.7134 -9.6376 -1.5584 -0.0277
1986 0.0912 0.1922 0.7167 -9.3530 -1.5581 -0.0246
1987 0.0745 0.1788 0.7467 -11.4994 -1.5368 0.0022
1988 0.0649 0.1757 0.7594 -13.2062 -1.5268 0.0133
1989 0.0679 0.1806 0.7515 -12.6175 -1.5417 0.0064
1990 0.0707 0.1680 0.7612 -12.1157 -1.4910 0.0149
Table AS.3.4.20. Proper AESs and determinants of matrixes of AESs for 13SLS estimates, Model
2 (lSlC 3840)
1972 3.7266 0.2609 -0.0880 0.5853 0.0345 0.0089 2.13E-17
1973 4.1583 0.1978 -0.1107 -1.4991 -0.0776 -0.0178 -1.57E-16
1974 4.9137 0.0875 -0.1438 -6.4204 -0.2800 -0.0525 9.09E-17
1975 5.6676 -0.0070 -0.1843 -13.4221 -0.5012 -0.0828 -1.49E-15
1976 5.6796 0.0052 -0.1979 -14.0995 -0.5112 -0.0885 1.67E-16
1977 4.4495 0.1609 -0.1304 -3.3845 -0.1607 -0.0348 -2.02E-16
1978 5.3492 0.0447 -0.1797 -10.7177 -0.4148 -0.0756 -1.47E-15
1979 5.7592 0.0235 -0.2296 -16.1665 -0.5461 -0.1034 -1.93E-15
1980 8.3465 -0.3388 -0.2847 -50.3041 -1.3403 -0.1474 8.2lE-16
1981 10.4852 -0.7630 -0.2380 -82.8238 -2.2955 -0.1352 -3.04E-15
1982 8.2987 -0.4652 -0.1912 -43.0870 -1.3930 -0.1100 -4.49E-15
1983 6.4895 -0.2504 -0.1130 -17.4361 -0.7258 -0.0687 -1.28E-15
1984 4.8005 -0.0086 -0.0411 -2.4850 -0.1404 -0.0179 -2.61E-16
1985 3.4964 0.2263 -0.0016 2.7943 0.2153 0.0431 I.77E-16
1986 3.4997 0.2517 -0.0275 2.3255 0.1672 0.0377 5.72E-16
1987 4.2871 0.1213 -0.0599 -0.7069 -0.0405 -0.0070 -7.67E-17
1988 4.8431 0.0072 -0.0603 -3.2925 -0.1763 -0.0240 -4.94E-16
1989 4.5703 0.0419 -0.0425 -1.4351 -0.0829 -0.0117 -3.20E-16
1990 4.6842 0.0920 -0.1062 -3.8763 -0.1889 -0.0335 -4.51E-16
AS.3. - 18
Table A5.3.4.21 Fitted cost shares and Own AESs for I3SLS estimates. Model 3 (lSIC 3840)
1972 0.0651 0.1651 0.7698 -11.5295 -1.0324 0.0116
1973 0.0653 0.1664 0.7683 -11.4979 -1.0480 0.0103
1974 0.0648 0.1640 0.7712 -11.5747 -1.0185 0.0129
1975 0.0634 0.1588 0.7778 -11.7893 -0.9472 0.0186
1976 0.0608 0.1476 0.7916 -12.2069 -0.7396 0.0306
1977 0.0618 0.1499 0.7883 -12.0451 -0.7893 0.0277
1978 0.0612 0.1490 0.7898 -12.1345 -0.7697 0.0290
1979 0.0590 0.1426 0.7984 -12.5003 -0.6184 0.0363
1980 0.0606 0.1478 0.7917 -12.2407 -0.7435 0.0306
1981 0.0655 0.1635 0.7709 -11.4635 -1.0131 0.0126
1982 0.0662 0.1650 0.7688 -11.3740 -1.0313 0.0108
1983 0.0697 0.1769 0.7534 -10.8727 -1.1473 -0.0030
1984 0.0725 0.1853 0.7422 -10.5153 -1.2019 -0.0131
1985 0.0716 0.1817 0.7468 -10.6303 -1.1806 -0.0090
1986 0.0681 0.1702 0.7617 -11.0948 -1.0886 0.0044
1987 0.0684 0.1716 0.7600 -11.0582 -1.1025 0.0029
1988 0.0704 0.1786 0.7510 -10.7787 -1.1597 -0.0051
1989 0.0716 0.1824 0.7461 -10.6317 -1.1849 -0.0096
1990 0.0652 0.1612 0.7736 -11.5210 -0.9819 0.0150
Table A5.3.4.22 Proper AESs and determinants of matrixes of AESs for I3SLS estimates, Model
3 (ISIC 3840)
Capital: Capital : Labour:
1972 3.9029 0.1380 -0.1086 -3.3290 -0.1531 -0.0238 -2.93E-16
1973 3.8705 0.1391 -0.1020 -2.9316 -0.1376 -0.0212 -3.85E-16
1974 3.9365 0.1355 -0.1142 -3.7065 -0.1675 -0.0262 -6.37E-16
1975 4.0994 0.1238 -0.1407 -5.6389 -0.2351 -0.0375 -8.50E-16
1976 4.4787 0.1019 -0.2059 -11.0301 -0.3834 -0.0650 -4.85E-16
1977 4.3697 0.1127 -0.1923 -9.5869 -0.3467 -0.0588 -1.07E-15
1978 4.4215 0.1064 -0.1975 -10.2097 -0.3633 -0.0613 -4.51E-16
1979 4.7073 0.0830 -0.2374 -14.4279 -0.4604 -0.0788 -3.51E-16
1980 4.4865 0.0989 -0.2044 -11.0282 -0.3842 -0.0645 -7.39E-16
1981 3.9107 0.1451 -0.1176 -3.6801 -0.1656 -0,0266 -8.26E-16
1982 3.8582 0.1507 -0.1107 -3.1563 -0.1454 -0.0234 -5.IOE-16
1983 3.5294 0.1777 -0.0573 0.0182 0.0010 0.0002 1.41E-18
1984 3.3235 0.1967 -0.0244 1.5927 0.0993 0.0152 2.11E-16
1985 3.3998 0.1917 -0.0386 0.9914 0.0587 0.0091 1.36E-16
1986 3.6911 0.1673 -0.0868 -1.5461 -0.0772 -0.0124 -2.38E-16
1987 3.6584 0.1687 -0.0802 -1.1927 -0.0608 -0.0097 -1.06E-16
1988 3.4808 0.1833 -0.0507 0.3848 0.0218 0.0034 9.94E-17
1989 3.3911 0.1908 -0.0356 1.0976 0.0656 0.0101 2.53E-16
1990 3.9704 0.1430 -0.1298 -4.4518 -0.1933 -0.0316 -7.60E-16
AS.3. - 19
Table A5.3.4.23 Fitted cost shares and Own AESs for I3SLS estimates. Model 4 (lSIC 3840)
Year Fitted cost Fitted cost Fitted cost OwnAES Own AES Own AES
1972 0.0664 0.1638 0.7698 -12.0416 -1.2959 0.0022
1973 0.0668 0.1648 0.7684 -11.9689 -1.3048 0.0009
1974 0.0660 0.1628 0.7712 -12.1000 -1.2862 0.0034
1975 0.0644 0.1581 0.7775 -12.3734 -1.2358 0.0091
1976 0.0609 0.1484 0.7907 -13.0240 -1.0987 0.0208
1977 0.0615 0.1511 0.7874 -12.9035 -1.1419 0.0179
1978 0.0613 0.1498 0.7889 -12.9456 -1.1213 0.0192
1979 0.0594 0.1433 0.7973 -13.3071 -1.0016 0.0265
1980 0.0610 0.1482 0.7908 -13.0033 -1.0949 0.0209
1981 0.0657 0.1637 0.7706 -12.1545 -1.2946 0.0029
1982 0.0661 0.1653 0.7686 -12.0861 -1.3094 0.0010
1983 0.0697 0.1767 0.7536 -11.5087 -1.3860 -0.0127
1984 0.0723 0.1850 0.7427 -11.1345 -1.4194 -0.0228
1985 0.0711 0.1818 0.7471 -11.3034 -1.4085 -0.0187
1986 0.0676 0.1709 0.7615 -11.8407 -1.3520 -0.0054
1987 0.0681 0.1720 0.7599 -11.7654 -1.3593 -0.0068
1988 0.0702 0.1786 0.7512 -11.4382 -1.3950 -0.0149
1989 0.0714 0.1822 0.7465 -11.2654 -1.4097 -0.0193
1990 0.0649 0.1619 0.7732 -12.2872 -1.2776 0.0052
Table A5.3.4.24 Proper AESs and determinants of matrixes of AESs for I3SLS estimates, Model
4 (ISIC 3840)
Year Proper Proper Proper Dtrm of Dtrm of Dtnn of Dtnn of
1972 4.0985 0.1666 -0.0778 -1.1928 -0.0540 -0.0089 -1.14E-16
1973 4.0596 0.1705 -0.0733 -0.8632 -0.0397 -0.0065 2.02E-17
1974 4.1345 0.1637 -0.0826 -1.5306 -0.0682 -0.0112 -2.27E-16
1975 43083 0.1499 -0.1059 -3.2705 -0.1351 -0.0225 -1.79E-16
1976 4.7287 0.1150 -0.1578 -8.0519 -0.2838 -0.0477 -9.50E-16
1977 4.6252 0.1205 -0.1422 -6.6582 -0.2452 -0.0406 -1.27E-15
1978 4.6700 0.1190 -0.1499 -7.2923 -0.2630 -0.0440 -4.35E-16
1979 4.9571 0.1009 -0.1895 -11.2448 -0.3633 -0.0625 -1.60E-15
1980 4.7276 0.1167 -0.1594 -8.1124 -0.2850 -0.0483 -8.83E-16
1981 4.1329 0.1589 -0.0775 -1.3449 -0.0607 -0.0098 -1.37E-16
1982 4.0825 0.1619 -0.0696 -0.8405 -0.0389 -0.0062 -9.62E-17
1983 3.7349 0.1894 -0.0207 2.0012 0.1100 0.0171 4.55E-17
1984 3.5206 0.2062 0.0111 3.4088 0.2115 0.0323 3.82E-16
1985 3.6065 0.1980 -0.0004 2.9139 0.1726 0.0264 5.99E-16
1986 3.9162 0.1728 -0.0444 0.6719 0.0338 0.0053 1.16E-16
1987 3.8773 0.1768 -0.0398 0.9598 0.0492 0.0077 335E-17
1988 3.6881 0.1923 -0.0131 2.3544 0.1330 0.0206 2.19E-16
1989 3.5923 0.2003 0.0006 2.9764 0.1773 0.0272 236E-16
1990 4.2050 0.1516 -0.0857 -1.9844 -0.0870 -0.0140 -1.33E-16
A5.3. - 20
Table AS.3.4.2S Fitted cost shares and Own AESs for I3SLS (AR) estimates, Model 2 (ISIC
3840)
1972 0.0864 0.1754 0.7382 -4.4789 -1.4541 0.1045
1973 0.0724 0.1675 0.7600 -4.1332 -1.4094 0.1174
1974 0.0592 0.1584 0.7823 -2.9130 -1.3311 0.1306
1975 0.0498 0.1496 0.8004 -0.7471 -1.2207 0.1412
1976 0.0585 0.1489 0.7924 -2.8020 -1.2102 0.1365
1977 0.0819 0.1651 0.7527 -4.4272 -1.3918 0.1131
1978 0.0633 0.1527 0.7837 -3.4400 -1.2648 0.1314
1979 0.0555 0.1434 0.8007 -2.2545 -1.1152 0.1414
1980 0.0315 0.1323 0.8358 15.0913 -0.8502 0.1617
1981 0.0227 0.1377 0.8392 45.4208 -0.9929 0.1637
1982 0.0351 0.1465 0.8179 9.4040 -1.1704 0.1514
1983 0.0483 0.1621 0.7890 -0.1951 -1.3674 0.1345
1984 0.0714 0.1819 0.7461 -4.0800 -1.4783 0.1092
1985 0.1058 0.2006 0.6929 -4.3870 -1.5025 0.0779
1986 0.1073 0.1940 0.6979 -4.3679 -1.5003 0.0808
1987 0.0834 0.1803 0.7356 -4.4483 -1.4731 0.1029
1988 0.0714 0.1775 0.7502 -4.0792 -1.4630 0.1116
1989 0.0758 0.1826 0.7407 -4.2725 -1.4803 0.1060
1990 0.0756 0.1686 0.7547 -4.2673 -1.4171 0.1142
Table AS.3.4.26 Proper AESs and determinants of matrixes of AESs for I3SLS (AR) estimates,
Model 2 (ISIC 3840)
Year Proper Proper Proper Dtnn of Dum of Dtnn of Dum of
Capital : Capital: Labour:
1972 4.4709 -0.5381 -0.1778 -13.4763 -0.7575 -0.1835 0.0099
1973 5.3359 -0.7821 -0.1979 -22.6460 -1.0970 -0.2046 0.0171
1974 6.6083 -1.1173 -0.2310 -39.7927 -1.6286 -0.2271 0.0322
1975 8.0546 -1.4592 -0.2736 -63.9653 -2.2346 -0.2472 0.0560
1976 7.0361 -1.1156 -0.2925 -46.1153 -1.6270 -0.2507 0.0455
1977 4.8890 -0.5907 -0.2273 -17.7404 -0.8495 -0.2090 0.0210
1978 6.4427 -0.9782 -0.2741 -37.1574 -1.4088 -0.2413 0.0428
1979 7.6051 -1.2060 -0.3283 -55.3233 -1.7730 -0.2654 0.0655
1980 13.6208 -2.7249 -0.3794 -198.3571 -4.9848 -0.2815 0.2294
1981 17.8530 -4.1564 -0.3199 -363.8292 -9.8423 -0.2649 0.4425
1982 11.2242 -2.4143 -0.2731 -136.9895 -4.4053 -0.2518 0.1880
1983 7.7188 -1.5748 -0.1921 -59.3130 -2:5064 -0.2208 0.0908
1984 5.0504 -0.8417 -0.1236 -19.4754 -1.1539 -0.1767 0.0350
1985 3.4779 -0.3382 -0.0970 -5.5048 -0.4562 -0.1265 0.0122
1986 3.5266 -0.3102 -0.1260 -5.8838 -0.4493 -0.1372 0.0139
1987 4.5009 -0.6000 -0.1502 -13.7055 -0.8179 -0.1742 0.0309
1988 5.1517 -0.8319 -0.1451 -20.5724 -1.1474 -0.1844 0.0465
1989 4.8021 -0.7479 -0.1276 -16.7354 -1.0122 -0.1731 0.0409
1990 5.1246 -0.7189 -0.1982 -20.2151 -1.0043 -0.2012 0.0510
A5.3. - 21
Table AS.3.4.27 Fitted cost shares and Own AESs for 13SLS (AR) estimates, Model 3 (ISIC
3840)
1972 0.0545 0.1443 0.8012 4.8717 -5.8724 0.0497
1973 0.0527 0.1411 0.8062 5.7711 -6.0263 0.0537
1974 0.0541 0.1446 0.8012 5.0452 -5.8562 0.0497
1975 0.0521 0.1458 0.8021 6.0939 -5.8038 0.0504
1976 0.0546 0.1568 0.7887 4.8464 -5.3293 0.0394
1977 0.0611 0.1635 0.7754 2.3061 -5.0707 0.0283
1978 0.0561 0.1578 0.7861 4.1341 -5.2884 0.0373
1979 0.0465 0.1500 0.8035 10.0338 -5.6120 0.0516
1980 0.0507 0.1518 0.7975 6.9452 -5.5370 0.0467
1981 0.0674 0.1618 0.7708 0.6976 -5.1340 0.0245
1982 0.0712 0.1657 0.7631 -0.0261 -4.9927 0.0179
1983 0.0810 0.1697 0.7494 -1.2817 -4.8526 0.0060
1984 0.0908 0.1762 0.7329 -2.0086 -4.6354 -0.0084
1985 0.0910 0.1791 0.7299 -2.0186 -4.5470 -0.0111
1986 0.0815 0.1751 0.7434 -1.3354 -4.6713 0.0008
1987 0.0801 0.1723 0.7476 -1.1996 -4.7627 0.0044
1988 0.0854 0.1741 0.7404 -1.6605 -4.7027 -0.0018
1989 0.0882 0.1750 0.7368 -1.8526 -4.6752 -0.0050
1990 0.0703 0.1672 0.7624 0.1246 -4.9363 0.0173
Table AS.3.4.28 Proper AESs and determinants of matrixes of AESs for 13SLS (AR) estimates,
Model 3 (ISIC 3840)
1972 8.8837 -1.9314 0.4533 -107.5284 -3.4880 -0.4975 -7.40E-15
1973 9.3319 -2.0110 0.4444 -121.8632 -3.7339 -0.5214 -5.lOE-15
1974 8.9170 -1.9507 0.4547 -109.0579 -3.5543 -0.4980 -1.00E-14
1975 9.1564 -2.0602 0.4594 -119.2068 -3.9371 -0.5039 -6.54E-15
1976 8.2488 -1.9751 0.4889 -93.8702 -3.7099 -0.4491 -2.82E-16
1977 7.2035 -1.7010 0.5015 -63.5847 -2.8279 -0.3951 1.13E-15
1978 7.9996 -1.9012 0.4905 -85.8567 -3.4606 -0.4377 -6.36E-15
1979 9.8906 -2.4273 0.4757 -154.1336 -5.3739 -0.5158 3.93E-15
1980 9.0558 -2.1649 0.4778 -120.4635 -4.3624 -0.4871 8.56E-15
1981 6.6870 -1.4648 0.4933 -48.2978 -2.1285 -0.3690 -1.91E-15
1982 6.2563 -1.3557 0.5001 -39.0103 -1.8383 -0.3396 -4.49E-15
1983 5.5143 -1.1099 0.5029 -24.1880 -1.2397 -0.2823 -1.90E-15
1984 4.8739 -0.9231 0.5107 -14.4446 -0.8352 -0.2217 -1.98E-15
1985 4.8050 -0.9271 0.5165 -13.9094 -0.8371 -0.2162 -4.59E-16
1986 5.3430 -1.1122 0.5145 -22.3096 -1.2381 -0.2683 -1.04E-15
1987 5.4907 -1.1371 0.5094 -24.4349 -1.2983 -0.2807 -4.96E-16
1988 5.1685 -1.0241 0.5099 -18.9052 -1.0458 -0.2515 -1.04E-15
1989 5.0177 -0.9700 0.5099 -16.5159 -0.9316 -0.2366 -3.62E-16
1990 6.2710 -1.3871 0.5044 -39.9406 -1.9219 -0.3399 1.32E-15
A5.3. - 22
Table AS.3.4.29 Fitted cost shares and Own AESs for I3SLS (AR) estimates, Model 4 (ISIC
3840)
1972 0.0641 0.1432 0.7927 -9.1733 -5.5931 -0.0118
1973 0.0646 0.1401 0.7952 -9.1334 -5.7290 -0.0094
1974 0.0635 0.1435 0.7931 -9.2195 -5.5814 -0.0115
1975 0.0606 0.1443 0.7952 -9.4310 -5.5475 -0.0095
1976 0.0549 0.1545 0.7906 -9.8171 -5.1384 -0.0138
1977 0.0567 0.1613 0.7820 -9.7011 -4.8930 -0.0222
1978 0.0558 0.1556 0.7887 -9.7623 -5.0973 -0.0157
1979 0.0515 0.1475 0.8010 -10.0073 -5.4130 -0.0039
1980 0.0546 0.1495 0.7959 -9.8326 -5.3305 -0.0088
1981 0.0644 0.1604 0.7751 -9.1494 -4.9221 -0.0290
1982 0.0655 0.1643 0.7701 -9.0669 -4.7889 -0.0339
1983 0.0727 0.1690 0.7583 -8.5387 ·-4.6362 -0.0459
1984 0.0780 0.1761 0.7460 -8.1567 -4.4216 -0.0586
1985 0.0761 0.1786 0.7453 -8.2922 -4.3471 -0.0593
1986 0.0692 0.1740 0.7568 -8.7945 -4.4823 -0.0474
1987 0.0698 0.1713 0.7588 -8.7479 -4.5639 -0.0453
1988 0.0739 0.1736 0.7525 -8.4458 -4.4956 -0.0518
1989 0.0762 0.1747 0.7492 -8.2841 -4.4634 -0.0553
1990 0.0635 0.1657 0.7709 -9.2211 -4.7447 -0.0332
Table AS.3.4.30 Proper AESs and determinants of matrixes of AESs for 13SLS (AR) estimates,
Model 4 (lSIC 3840)
Year Proper Proper Proper Dtnnof Dtnnof Dtnn of Dtnn of
1972 7.8961 -0.6846 0.3719 -11.0411 -0.3603 -0.0722 7.97E-16
1973 7.9888 -0.6653 0.3602 -11.4950 -0.3569 -0.0759 3.89E-17
1974 7.9510 -0.7005 0.3734 -11.7596 -0.3849 -0.0753 5.80E-16
1975 8.2431' -0.7771 0.3784 -15.6397 -0.5148 -0.0908 -8.27E-16
1976 8.4663 -0.9726 0.4162 -21.2338 -0.8105 -0.1023 -9.74E-16
1977 7.9239 -0.9309 0.4347 -15.3212 -0.6516 -0.0805 1.54E-15
1978 8.2982 -0.9468 0.4189 -19.0988 -0.7432 -0.0954 1.83E-16
1979 9.3273 -1.0733 0.3964 -32.8291 -1.1129 -0.1360 1.75E-15
1980 8.7493 -0.9687 0.4008 -24.1377 -0.8519 -0.1137 2.30E-15
1981 7.1243 -0.7142 0.4267 -5.7222 -0.2451 -0.0395 2.37E-16
1982 6.8778 -0.6961 0.4366 -3.8835 -0.1768 -0.0281 -7.13E-17
1983 6.1539 -0.5535 0.4437 1.7159 0.0853 0.0158 -8.63E-17
1984 5.6105 -0.4716 0.4571 4.5875 0.2556 0.0501 -7.02E-16
1985 5.6582 -0.5099 0.4645 4.0323 0.2317 0.0420 -4.81E-16
1986 6.2570 -0.6347 0.4586 0.2692 0.0142 0.0023 1.83E-17
1987 6.2913 -0.6156 0.4516 0.3442 0.0175 0.0029 3.36E-18
1988 5.9323 -0.5386 0.4541 2.7767 0.1477 0.0268 9.40E-17
1989 5.7575 -0.4998 0.4551 3.8267 0.2080 0.0396 -1.33E-17
1990 7.0216 -0.7500 0.4417 -5.5518 -0.2564 -0.0376 -4.43E-16
A5.3. - 23
A5.3.5 Economies of Scale
...
Table AS 3 S 1 IZEF estimates of Economies of Scale - ISIC 3840
Year Economies Economies Economies Economies Economies Economies
of Scale- of Scale- of Scale- of Scale of Scale- of Scale-
Model I - Model 2 - Model 3 - Model 1 - Mode12 - Mode13 -
IZEF IZEF lZEF IZEF (AR) IZEF(AR) lZEF(AR)
estimates estimates estimates estimates estimates estimates
1972 1.2894 1.2168 1.0406 1.6906 1.2999 1.2212
1973 1.2553 1.2096 1.0370 1.6123 1.2055 1.1888
1974 1.2271 1.2086 1.0333 1.5643 1.0934 1.1563
1975 1.2074 1.2036 1.0312 1.5176 0.9905 1.1394
1976 1.2300 1.2122 1.0336 1.5419 0.9297 1.1587
1977 1.2841 1.2297 1.0387 1.5943 0.8944 1.2041
1978 1.2393 1.2160 1.0345 1.5047 0.8411 1.1670
1979 1.2201 1.1990 1.0343 1.4365 0.7861 1.1650
1980 1.1706 1.1981 1.0269 1.3772 0.7401 1.1048
1981 1.1543 1.2202 1.0213 1.3757 0.7102 1.0620
1982 1.1788 1.2301 1.0239 1.3882 0.6768 1.0814
1983 1.2049 1.2487 1.0254 1.4078 0.6632 1.0926
1984 1.2553 1.2720 1.0296 1.4535 0.6463 1.1260
1985 1.3384 1.2842 1.0390 1.5070 0.6238 1.2068
1986 1.3409 1.2695 1.0415 1.4662 0.5777 1.2294
1987 1.2801 1.2592 1.0351 1.3898 0.5427 1.1721
1988 1.2518 1.2635 1.0307 1.3558 0.5222 1.1352
1989 1.2606 1.2692 1.0313 1.3469 0.4990 1.1401
1990 1.2590 1.2416 1.0350 1.2979 0.4665 1.1714
MEAN 1.2446 1.2343 1.0328 1.4646 0.7742 1.1538
A5.3. - 24
...
Table AS 3 S 2 13SLS estimates of Economies of Scale - ISle 3840
of Scale- of Scale- of Scale - of Scale- of Scale- of Scale -
Model 1 - Mode12 - Mode13 - Model1 - Mode12 - Model 3-
13SLS 13SLS 13SLS 13SLS (AR) 13SLS (AR) 13SLS (AR)
1972 1.3021 1.2126 1.0267 1.3139 1.2015
1973 1.2609 1.2078 1.0267 1.2883 1.1625
1974 1.2261 1.2100 1.0267 1.2750 1.1238
1975 1.2037 1.2067 1.0267 1.2625 1.1039
1976 1.2316 1.2146 1.0267 1.2901 1.1267
1977 1.2979 1.2285 1.0267 1.3376 1.1808
1978 1.2447 1.2176 1.0267 1.2992 1.1365
1979 1.2260 1.2003 1.0267 1.2729 1.1341
1980 1.1636 1.2052 1.0267 1.2445 1.0634
1981 1.1384 1.2327 1.0267 1.2533 1.0141
1982 1.1678 1.2409 1.0267 1.2773 1.0364
1983 1.1969 1.2587 1.0267 1.3066 1.0493
1984 1.2562 1.2790 1.0267 1.3566 1.0881
1985 1.3630 1.2833 1.0267 1.4179 1.1842
1986 1.3733 1.2665 1.0267 1.4082 1.2115
1987 1.2971 1.2613 1.0267 1.3636 1.1425
1988 1.2599 1.2694 1.0267 1.3504 1.0990
1989 1.2713 1.2746 1.0267 1.3602 1.1046
1990 1.2792 1.2434 1.0267 1.3384 1.1417
MEAN 1.2505 1.2375 1.0267 1.3167 1.1213
A5.3. - 25
A5.3.6 Graphs of Economies of Scale
Figure A5.3.6.1 Economies of Scale based on IZEF estimates - ISIC 3840
1'4~ ~~
~ ~~-x:-~-- -~ __
-~-~=t-~
A
1.2 - - :t(:t
1X-X-X-X-X-X-X-X-X-X-X-X-X-X-X-X-X-X-X
0.8
~EOS-1
0.6 -X-EOS· 2
-X-EOS·3
0.4
0.2
O+--+-+--+--......-+--I---+---t--+--+-+--+-+--+--......-+--I---I
n ~ ~ ~ n n n n 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ 00
Figure A5.3.6.2 Economies of Scale based on IZEF (AR) estimates - ISIC 3840
1.8 --<>-EOS-l
-X-EOS-2
1.6' -X-EOS-3
1.4
0.8
0.6
0.4
0.2
0+--+-+--+--......-+---t~-+--+--+--+-+--+-- ......-+---I1---+---t----4
n ~ ~ ~ n n n ~ 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ 00
A5.3. - 26
Figure A5.3.6.3 Economies of Scale based on 13SLS estimates· ISle 3840
1'4~ ~~
_.~ '"'_~_::t:-~"')K-)K-)t-::-~-)K~
-)K-r-~-~~ ~~
A _
1.2 - - v
0.8
0.6
--<>-EOS.1
0.4 -)K-EOS·2
-X-EOS.3
0.2
0+--+-+--+-+-.......-+---+---11--......--+--+---+-+--+-+--+-+--4
n ~ N ~ n n n ~ 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ 00
Figure A5.3.6.4 Economies of Scale based on 13SLS (AR) estimates· ISle 3840
0.8
0.6 1-)K-EOS.21
-X-EOS-31
0.4
0.2
O+--+-+--+-+-.......-+---+---1I--......--+--+---+-+--+-+--+-+-~
n ~ N ~ n n n ~ 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ 00
A5.3. - 27
Figure AS.3.6.S Economies of Scale· Model 1 • IZEF, IZEF (AR) and 13SLS
(AR)
0.8
0.6 ~13SLS. EOS·1
0.4 -X-IZEF. EOS· 1

-;:K-IZEF (AR) - EOS -1
0.2
o+--+-+---+-+---+--+---.--+---.--+---.--+---.--+---t--+-~I---4
72
Figure AS.3.6.6 Economies of Scale· Model 2 • IZEF, IZEF (AR), I3SLS and
13SLS (AR)
1.6
1.4
1.2
0.8
0.6
0.4
0.2
O+---+-+---+--+---'--+---'--+---'--+---'--+---'--+---t--+-~I---I
72 ~ ~ ~ ~ n n ~ 00 ~ ~ ~ ~ ~ ~ ~ 00 ~ 00
A5.3. - 28
Figure A5.3.6.7 Economies of Scale· Model 3 • IZEF, IZEF (AR), 13SLS and
I3SLS (AR)
0.8
0.6
-o-13SLS - EOS - 3
0.4 -o-I3SLS (AR) - EOS - 3
-X-IZEF-EOS-3
0.2 -X-IZEF (AR) - EOS - 3
o+--+-+--+-+--+-t---+---I -+--+--+--+-+--+-+--+-+--I
n n ~ ~ n n n ~ 00 M ~ ~ M ~ ~ ~ ~ ~ 00
APPENDIX 5.4
FURNITURE (ISle 3320)
5.4.1 Hypothesis Test Results
Table A5.4.1.1 LR test statistics and appropriate chi-square statistics for ZEF and 3SLS
estimates of ISle (3320)
Null Hypothesis Critical Compute Compute Compute Critical Compute
2 dL. R. dL.R. dL.R. 2 dL.R.
X X
statistic statistic statistic statistic statistic statistic
and df for IZEF IZEF I3SLS and df for I3SLS
IZEF, estimates (AR) estimates I3SLS (AR)
IZEF estimates (AR) estimates
(AR) and estimates
I3SLS
estimates
Non-homothetic 11.07 (5) 22.683 -- 21.685 -- --
Homotheticity 14.07 (7) 27.637 17.159 26.692 5.99 (2) -Llll
Homogeneity in Output 15.51 (8) 27.990 18.161 26.692 7.81 (3) -23.142
V.R.T.S. Cobb-Douglas 19.7 (ll) 62.748 58.878 61.406 12.59 (6) 17.320
Note: 1. For IZEF, IZEF (AR) and I3SLS the alternate hypothesis is that the technology augmented
fornl is valid. For 13SLS (AR) estimates the alternate hypothesis is that the non-homothetic
form is valid.
2. Numbers in parenthesis appearing after X2 indicate Degrees of Freedom (df) (at the 5% level
obtain the specification of the different null hypothesis from the alternative specification.
Table A5.4.1.2 LR test statistics for comparison of IZEF and IZEF(AR) estimators ISIC 3320
Model 1 2 3 4 5
Likelihood Ratio Statistic N/A -0.233 8.792 8.124 0.335
Table AS.4.1.3 LR test statistics for comparison of I3SLS and I3SLS(AR) estimators ISIC 3320
Model 2 3 4 5
Likelihood Ratio Statistic -20.042 -14.188 7.843 0.268
A5.4. - 2
Table AS.4.1.4 LR test of Economies of Scale - IZEF estimates (ISIC 3320)

Critical Chi-Squared Computed Wald - Computed Wald -
Statistic IZEF estimates IZEF (AR) estimates
Model1 (5) 11.071 34.600 14.640
Mode12 (4) 9.488 54.171 --
Mode13 (2) 5.991 49.848 4.932
Mode14 (1) 3.841 47.417 4.385
Model 5 (1) 3.841 44.803 2.096
2. Source of critical chi-squared values: Gujarati (1988, p685)
Table AS.4.1.5 LR test of Economies of Scale - 13SLS estimates (ISIC 3320)

Critical Chi-Squared Computed Wald- Computed Wald -
Statistic I3SLS estimates I3SLS (AR) estimates
Model 1 (5) 11.071 32.675 --
Mode12 (4) 9.488 54.150 19.599
Mode13 (2) 5.991 51.838 12.058
Mode14 (1) 3.841 50.678 7.334
Mode15 (1) 3.841 42.824 1.276
2. Source of critical chi-squared values: Guiarati (1988. p685)
A5.4. - 3
.
bl AS.421
T a.e . Computed and cnTleaID urb'ID- W a tsonstal'ISl'ICS ~or IZEF esl'Imat es 0fISIC3320 .
Model Equation Number of DL DU Computed Ho: p=O·
Explanatories D-W vs.
Statistic HI: 0>0
Cost share of Labour 4 0.859 1.848 1.l03 N.D.
Cost share of Caoital 3 0.967 1.685 1.041 N.D.
Mode15 Cost function 3 0.967 1.685 0.391 Reiect
Cost share of Caoital -- -- -- -- --
Guiarati, 1988, p687).
4 22 Computed and cntIca

T a bleA S.•. . . ~or I3SLS estimates 0flSIC 332 0
.. IDurb in- W atson statistics
Exogenous D-W vs.
Variables Statistic HI: 0>0
Cost share of Labour 15 0.063 3.676 1.l03 N.D.
A5.4. - 4
T ab IeA 54 ..3.1 IZEF parameter estimates tor F urm't ure (ISle 3320)
0. 0 18.7265 18.7551 18.7293 18.7231 18.7357
(1044.19) (878.141) (1281.77) (1730.12) (1584.90)
aK 0.0298 0.0404 0.0360 0.0358 0.0365
(6.4772) (7.4758) (12.2287) (12.1968) (14.6303)
aL 0.3343 0.3504 0.3348 0.3339 0.3020
(46.6709) (37.9877) (49.1518) (50.1495) (44.8775)
aM 0.6359 0.6092 0.6292 0.6303 0.6615
ay 0.8584 0.6983 0.7952 0.8492 0.7916

(4.5440) (6.3029) (8.8000) (38.7837) (25.4297)
0. 1 -0.0047
(-0.7747)
YKK -0.0044 0.0245 0.0251 0.0265
(-0.2417) (4.7877) (5.2046) (5.7032)
YKL 9.40E-05 0.0457 0.0364 0.0374
(0.0059) (5.4972) (4.2228) (4.3089)
YKM 0.0043 -0.0702 -0.0615 -0.0639
YLL 0.3028 0.2293 6.2382 0.2349

(8.6365) (11.0997) (10.6702) (10.9819)
YLM -0.3029 -0.2750 -0.2746 -0.2723
YMM 0.2986 0.3452 0.3361 0.3362
YKY -0.0354 -0.0101

(-2.1396) (-0.5188)
YLY -0.1279 -0.0815
(-5.2163) (-2.5057)
YMY 0.1633 0.0916
Yyy 0.0946 0.1054 0.1869

(0.2065) (0.3322) (0.6008)
YII 0.0013
0.9738)
yy, -0.0157
(-1.2642)
YKI 0.0017
(1.8831)
YLr 0.0053
(4.7521)
YMI -0.0070
Dtrm 9.97E-13 3.29E-12 4.27E-12 435E-12 2.71E-ll

A5.4. - 5
Table AS.4.3.2 IZEF AR(l) parameter estimates for Furniture (ISle 3320),
Parameter Model 1 Model 2 Mode13 Model 4 ModelS
0. 0 18.7344 18.7637 18.7287 18.7160
(446.019) (471.858) (782.140) (771.828)
aK 0.0290 0.0358 0.0363 0.0366
(5.4532) (6.8730) (8.4681) (15.5720)
aL 0.3294 0.3284 0.3336 0.3026
(24.3411) (19.7286) (19.8748) (45.0868)
aM 0.6416 0.6358 0.6301 0.6608
ay 0.7350 0.6034 0.8381 0.8799

(2.0129) (2.7361) (10.8367) (10.6029)
at -0.0046
(-0.6969)
YKK 5. lE-OS 0.0317 0.0260
(0.0026) (3.0595) (3.1229)
YKL 0.0024 0.0430 0.0374
(0.1529) (2.5357) (2.3841)
YKM -0.0025 -0.0747 -0.0634
Yu 0.2902 0.2169 0.2312

(6.5941) (4.5046) (4.7835)
YLM -0.2926 -0.2599 -0.2686
YMM 0.2951 0.3346 0.3320
YKY -0.0335
(-1.9550)
YLY -0.1112
(-2.7786)
YMY 0.1447
Yyy 0.2462 0.7602

(0.3478) (1.2015)
YI/ 0.0011
(1.5342)
YYt -0.0066
(-0.3615)
YKt 0.0016
(1.7208)
Yu 0.0050
(4.1016)
YMt -0.0066
Dtnn 1.01E-12 2.62E-12 2.77E-12 2.66E-II

Note: I. Numbers in parenthesis are t-statistics referring to parameter estimates above. t-statistics do
A5.4. - 6
Table AS.4.3.3 I3SLS parameter estimates for Furniture (ISle 3320).

a o 18.7291 18.7492 18.7255 18.7250 18.7357
(1052.33) (809.314) (1098.64) (1734.44) (1571.29)
a K 0.0298 0.0398 0.0358 0.0359 0.0365
(6.4595) (7.4180) (12.1265) (12.2893) (14.6304)
a L 0.3342 0.3493 0.3345 0.3343 0.3020
(46.4143) (36.8824) (47.9128) (49.7975) (0.0067)
aM 0.6360 0.6109 0.6297 0.6298 0.6615
ay 0.9120 0.7514 0.8384 0.8416 0.7919

(4.7551) (5.4979) (6.9707) (37.8196) (24.9053)
a l
-0.0076
(-1.2056)
'YKK -0.0065 0.0256 0.0255 0.0259
(-0.3489) (4.8512) (5.0606) (5.5829)
'YKL -0.0009 0.0447 0.0357 0.0369
(-0.0576) (5.3090) (4.0950) (4.2284)
"I KM 0.0074 -0.0703 -0.0612 -0.0628
"ILL 0.3023 0.2258 0.2353 0.2360

(8.4905) (10.3918) (9.9341) . (10.8070)
'Y LM -0.3014 -0.2705 -0.2710 -0.2729
"I MM 0.2940 0.3408 0.3322 0.3357
'Y KY -0.0371 -0.0099

(-2.2135) (-0.5147)
'YLY -0.1285 -0.0784
(-5.1708) (-2.3425)
"I MY 0.1656 0.8830
'Yyy -0.3824 -0.0835 0.0079

(-0.6911) (-0.1959) (0.0188)
"Ill 0.0013
(2.0379)
"I YI -0.0051
(-0.3492)
'Y KI 0.0018
(1.9458)
"I Lt 0.0054
(4.7979)
"I MI -0.0072
Dtnn 1.07£-12 3.35£-12 4.36£-12 4.36£-12 2.71£-11

2. Dtnn is the detenninant of the residual covariance matrix.
A5.4. - 7
Table AS.4.3.4 I3SLS AR(l) oarameterestimates for Furniture (ISle 3320).

Parameter Model I Model 2 Model 3 Model 4 Model 5
ao 18.5046 18.5418 18.7388 18.7108
(111.630) (134.981) (819.578) (690.890)
aK 0.0339 0.0322 0.0366 0.0365
(4.9303) (6.9514) (8.3166) (14.1616)
aL 0.2396 0.3393 0.3329 0.3003
(1.8319) (13.0888) (19.1357) (42.9850)
aM 0.7265 0.6285 0.6305 0.6632
ay 2.5387 2.3240 0.7981 0.8959

( 1.9785) (2.2786) 00.7084) (9.7263)
at
YKK 0.0305 0.0360 0.0267

(2.9084) (3.2999) (3.0043)
YKL 0.0044 0.0329 0.0397
(0.1945) (1.7508) (2.5003)
YKM -0.0349 -0.0689 -0.0644
YLL 0.2291 0.2418 0.2319

(3.9789) (4.4774) (4.7675)
YLM -0.2335 -0.2747 -0.2716
YMM 0.2684 0.3436 0.3380
YKY -0.0214
(-1.0205)
YLY -0.0691
(-1.6471)
YMY 0.0905
Yyy -6.1347 -5.0693

(-1.5235) (-1.5672)
Ytt
YYt
YKI
I
YLt
YMr
DUln 1.02E-II 9.59E-12 2.82E-12 2.67E-II
Note: I. Numbers in parenthesis are I-statistics referring to parameter estimates above. I-statistics do
2. Dtrm is the detemlinant of the residual covariance matrix.
A5.4. - 8
5.4.4 Monotonicity and Concavity Test Results
Table AS.4.4.l Fitted cost shar.es and Own AESs for IZEF estimates. Modell (ISIC 3320)
Year Fitted cost Fined cost Fined cost OwnAES Own AES Own AES
1972 0.0315 0.3396 0.6289 -35.1804 0.6809 0.1649
1973 0.0300 0.3310 0.6390 -37.1553 0.7428 0.1663
1974 0.0298 0.2974 0.6729 -37.5816 1.0613 0.1733
1975 0.0301 0.3088 0.6611 -37.0642 0.9375 0.1706
1976 0.0321 0.2922 0.6757 -34.3995 1.1245 0.1741
1977 0.0329 0.2861 0.6809 -33.4076 1.2035 0.1754
1978 0.0337 0.2907 0.6756 -32.5863 1.1432 0.1740
1979 0.0308 0.2851 0.6841 -36.1596 1.2178 0.1763
1980 0.0279 0.2659 0.7061 -40.5103 1.5210 0.1827
1981 0.0280 0.2672 0.7048 -40.3098 1.4984 0.1823
1982 0.0320 0.2878 0.6801 -34.4899 1.1812 0.1752
1983 0.0360 0.3139 0.6500 -30.1318 0.8873 0.1683
1984 0.0405 0.3294 0.6301 -26.3529 0.7550 0.1650
1985 0.0444 0.3440 0.6116 -23.7467 0.6521 0.1632
1986 0.0455 0.3376 0.6169 -23.0954 0.6947 0.1636
1987 0.0484 0.3284 0.6232 -21.5337 0.7626 0.1642
1988 0.0469 0.3057 0.6474 -22.3073 0.9691 0.1678
1989 0.0485 0.2649 0.6865 -21.4791 1.5392 0.1769
1990 0.0523 0.2837 0.6639 -19.7182 1.2366 0.1712
(ISIC 3320\
1972 1.0088 1.2171 -0.4182 -24.9724 -7.2820 -0.0627 0.0008
1973 1.0095 1.2240 -0.4322 -28.6164 -7.6784 -0.0633 0.0009
1974 1.0106 1.2148 -0.5138 -40.9066 -7.9904 -0.0800 0.0013
1975 1.0101 1.2160 -0.4839 -35.7689 -7.8014 -0.0742 0.0014
1976 1.0100 1.1981 -0.5343 -39.7009 -7.4231 -0.0897 0.0017
1977 1.0100 1.1917 -0.5546 -41.2245 -7.2802 -0.0965 0.0019
1978 1.0096 1.1891 -0.5422 -38.2715 -7.0852 -0.0951 0.0018
1979 1.0107 1.2043 -0.5530 -45.0577 -7.8246 -0.0911 0.0022
1980 1.0127 1.2183 -0.6129 -62.6433 -8.8851 -0.0978 0.0030
1981 1.0126 1.2178 -0.6084 -61.4238 -8.8300 -0.0971 0.0033
1982 1.0102 1.1973 -0.5474 -41.7585 -7.4765 -0.0927 0.0026
1983 1.0083 1.1835 -0.4845 -27.7531 -6.4717 -0.0854 0.0020
1984 1.0070 1.1684 -0.4595 -20.9098 -5.7144 -0.0866 0.0017
1985 1.0062 1.1583 -0.4399 -16.4967 -5.2177 -0.0870 0.0015
1986 1.0061 1.1532 -0.4545 -17.0573 -5.1084 -0.0929 0.0016
1987 1.0059 1.1425 -0.4801 -17.4340 -4.8415 -0.1053 0.0017
1988 1.0066 1.1415 -0.5307 -22.6321 -5.0460 -0.1190 0.0021
1989 1.0073 1.1291 -0.6653 -34.0751 -5.0751 -0.1703 0.0029
1990 1.0063 1.1238 -0.6079 -25.3968 -4.6388 -0.1579 0.0024
A5.4. - 9
Table AS.4.4.3 Fitted cost shares and Own AESs for IZEF estimates, Model 2 (lSIC 3320)
1972 0.0404 0.3504 0.6092 -8.7417 0.0137 0.2886
1973 0.0370 0.3373 0.6257 -8.1321 0.0508 0.2835
1974 0.0314 0.3082 0.6604 -6.0045 0.1692 0.2773
1975 0.0340 0.3158 0.6502 -7.2058 0.1325 0.2785
1976 0.0356 0.3086 0.6558 -7.7651 0.1675 0.2778
1977 0.0369 0.3055 0.6576 -8.1057 0.1835 0.2776
1978 0.0353 0.3020 0.6627 -7.6522 0.2027 0.2770
1979 0.0321 0.2907 0.6772 -6.3792 0.2733 0.2761
1980 0.0293 0.2737 0.6970 -4.6066 0.4077 0.2758
1981 0.0324 0.2774 0.6902 -6.5139 0.3750 0.2758
1982 0.0363 0.2920 0.6717 -7.9449 0.2644 0.2763
1983 0.0397 0.3085 0.6518 -8.6452 0.1679 0.2783
1984 0.0440 0.3215 0.6345 -9.0728 0.1079 0.2814
1985 0.0465 0.3308 0.6227 -9.1757 0.0725 0.2844
1986 0.0450 0.3224 0.6326 -9.1249 0.1043 0.2818
1987 0.0419 0.3097 0.6484 -8.9104 0.1617 0.2788
1988 0.0388 0.2904 0.6708 -8.4958 0.2754 0.2764
1989 0.0324 0.2565 0.7111 -6.5012 0.5864 0.2764
1990 0.0325 0.2634 0.7041 -6.5911 0.5087 0.2761
(ISIC 3320)
AES AES AES Matrix I Matrix 2 Matrix 3 Matrix 4
1972 4.2283 -1.8523 -0.2883 -17.9979 -5.9543 -0.0792 3.81E-15
1973 4.6616 -2.0317 -0.3031 -22.1441 -6.4334 -0.0775 5.59E-15
1974 5.7203 -2.3843 -0.3511 -33.7378 -7.3499 -0.0763 8.0lE-15
1975 5.2599 -2.1784 -0.3392 -28.6214 -6.7528 -0.0781 1.24E-14
1976 5.1577 -2.0049 -0.3590 -27.9025 -6.1766 -0.0823 7.62E-15
1977 5.0542 -1.8934 -0.3688 -27.0324 -5.8351 -0.0851 4.53E-15
1978 5.2921 -2.0049 -0.3739 -29.5575 -6.1395 -0.0836 9.41E-15
1979 5.8960 -2.2288 -0.3969 -36.5055 -6.7286 -0.0821 1.00E-14
1980 6.6957 -2.4350 -0.4417 -46.7104 -7.2001 -0.0826 5.40E-15
1981 6.0887 -2.1413 -0.4363 -39.5151 -6.3817 -0.0869 1.13E-14
1982 5.3154 -1.8821 -0.4019 -30.3538 -5.7379 -0.0885 8.47E-15
1983 4.7308 -1.7124 -0.3676 -23.8317 -5.3384 -0.0884 4.56E-15
1984 4.2297 -1.5142 -0.3481 -18.8689 -4.8459 -0.0908 3.82E-15
1985 3.9681 -1.4222 -0.3351 -16.4107 -4.6319 -0.0917 8.49E-15
1986 4.1477 -1.4643 -0.3484 -18.1549 -4.7159 -0.0920 6.60E-15
1987 4.5224 -1.5844 -0.3694 -21.8932 -4.9946 -0.0914 1.07E-14
1988 5.0576 -1.6984 -0.4116 -27.9187 -5.2327 -0.0933 7.57E-15
1989 6.5071 -2.0514 -0.5075 -46.1554 -6.0053 -0.0955 1.29E-14
1990 6.3330 -2.0644 -0.4829 -43.4603 -6.0814 -0.0928 1.93E-14
A5.4. - 10
Table AS.4.4.S Fitted cost shares and Own AESs for IZEF estimates Model 3 (ISle 3320)
1972 0.0360 0.3348 0.6292 -7.4105 0.1382 0.2596
1973 0.0336 0.3307 0.6357 -6.5279 0.1544 0.2586
1974 0.0295 0.3054 0.6650 -4.0923 0.2792 0.2563
1975 0.0321 0.3154 0.6525 -5.8157 0.2241 0.2569
1976 0.0344 0.3024 0.6632 -6.8499 0.2980 0.2563
1977 0.0361 0.2988 0.6651 -7.4496 0.3215 0.2563
1978 0.0347 0.2991 0.6662 -6.9702 0.3190 0.2562
1979 0.0327 0.3009 0.6664 -6.1098 0.3075 0.2562
1980 0.0315 0.2944 0.6742 -5.4297 0.3518 0.2562
1981 0.0349 0.2990 0.6661 -7.0456 0.3199 0.2562
1982 0.0379 0.3066 0.6555 -7.9095 0.2723 0.2567
1983 0.0403 0.3174 0.6423 -8.3580 0.2139 0.2578
1984 0.0438 0.3210 0.6352 -8.7453 0.1963 0.2587
1985 0.0456 0.3239 0.6305 -8.8602 0.1833 0.2594
1986 0.0446 0.3165 0.6389 -8.8019 0.2185 0.2582
1987 0.0417 0.3010 0.6573 -8.5438 0.3069 0.2566
1988 0.0400 0.2883 0.6717 -8.3094 0.3973 0.2562
1989 0.0350 0.2528 0.7122 -7.0728 0.7710 0.2585
1990 0.0344 0.2561 0.7095 -6.8544 0.7265 0.2582
(ISle 3320)
1972 4.0200 -1.7151 -0.3035 -17.1849 -4.8656 -0.0563 3.14E-15
1973 4.2766 -1.8793 -0.3063 -19.2970 -5.2199 -0.0539 2.82E-15
1974 5.0335 -2.1301 -0.3519 -26.4787 -5.5861 -0.0523 5.59E-16
1975 4.5911 -1.9325 -0.3345 -22.3820 -5.2284 -0.0543 1.96E-15
1976 4.5016 -1.6974 -0.3692 -22.3056 -4.6369 -0.0599 2.97E-15
1977 4.3722 -1.5593 -0.3819 -21.5109 -4.3404 -0.0635 -1.43E-16
1978 4.5074 -1.6611 -0.3780 -22.5397 -4.5451 -0.0611 4.32E-15
1979 4.6988 -1.8219 -0.3694 -23.9573 -4.8848 -0.0577 3.64E-15
1980 4.9298 -1.8991 -0.3837 -26.2137 -4.9975 -0.0571 3.35E-15
1981 4.4883 -1.6456 -0.3788 -22.3987 -4.5133 -0.0615 3.89E-15
1982 4.1329 -1.4760 -0.3663 -19.2341 -4.2086 -0.0643 2.61E-15
1983 3.8463 -1.3763 -0.3470 -16.5818 -4.0486 -0.0653 3.36E-15
1984 3.5903 -1.2120 -0.3466 -14.6069 -3.7313 -0.0694 1.13E-15
1985 3.4631 -1.1377 -0.3448 -13.6172 -3.5929 -0.0713 1.81E-15
1986 3.5800 -1.1591 -0.3580 -14.7398 -3.6160 -0.0718 3.56E-16
1987 3.9015 -1.2448 -0.3879 -17.8439 -3.7416 -0.0717 4.96E-15
1988 4.1581 -1.2900 -0.4180 -20.5913 -3.7927 -0.0729 1.35E-15
1989 5.1162 -1.4690 -0.5250 -31.6286 -3.9864 -0.0763 4.27E-15
1990 5.1324 -1.5208 -0.5111 -31.3213 -4.0828 -0.0736 4.29E-15
A5.4. - 11
Table AS.4.4.7 Fitted cost shares and Own AESs for IZEF estimates. Model 4 (lSIC 3320)
1972 0.0358 0.3339 0.6303 -6.2563 0.1120 0.2597
1973 0.0333 0.3297 0.6371 -5.1147 0.1280 0.2587
1974 0.0291 0.3048 0.6661 -2.0667 0.2476 0.2565
1975 0.0318 0.3147 0.6535 -4.2427 0.1944 0.2570
1976 0.0343 0.3022 0.6635 -5.6082 0.2631 0.2565
1977 0.0361 0.2988 0.6651 -6.3801 0.2843 0.2565
1978 0.0346 0.2991 0.6663 -5.7694 0.2826 0.2565
1979 0.0325 0.3006 0.6669 -4.6700 0.2728 0.2564
1980 0.0312 0.2941 0.6746 -3.8293 0.3152 0.2564
1981 0.0348 0.2989 0.6662 -5.8655 0.2834 0.2565
1982 0.0380 0.3066 0.6554 -6.9559 0.2373 0.2569
1983 0.0405 0.3173 0.6423 -7.5285 0.1817 0.2580
1984 0.0441 0.3211 0.6348 -8.0543 0.1640 0.2590
1985 0.0461 0.3240 0.6299 -8.2230 0.1513 0.2598
1986 0.0450 0.3167 0.6383 -8.1384 0.1844 0.2585
1987 0.0420 0.3014 0.6566 -7.7911 0.2680 0.2568
1988 0.0403 0.2889 0.6708 -7.4958 0.3531 0.2564
1989 0.0352 0.2540 0.7109 -6.0073 0.7045 0.2586
1990 0.0345 0.2571 0.7083 . -5.7313 0.6637 0.2583
(lSIC 3320
1972 4.1288 -1.8319 -0.2938 -17.7475 -4.9805 -0.0573 1.29E-15
1973 4.4101 -2.0151 -0.2965 -20.1039 -5.3838 -0.0548 1.13E-15
1974 5.2171 -2.2970 -0.3412 -27.7303 -5.8064 -0.0529 1.62E-15
1975 4.7373 -2.0745 -0.3241 -23.2672 -5.3939 -0.0551 2.51E-15
1976 4.6131 -1.8114 -0.3579 -22.7560 -4.7199 -0.0606 7.l9E-16
1977 4.4636 -1.6587 -0.3703 -21.7378 -4.3877 -0.0642 4.87E-16
1978 4.6137 -1.7710 -0.3665 -22.9166 -4.6160 -0.0618 -2.96E-16
1979 4.8300 -1.9498 -0.3582 -24.6031 -4.9993 -0.0584 1.70E-15
1980 5.0750 -2.0356 -0.3722 -26.9628 -5.1256 -0.0577 -9.26E-16
1981 4.5923 -1.7539 -0.3672 -22.7515 -4.5804 -0.0622 -3.58E-15
1982 4.2127 -1.5675 -0.3551 -19.3979 -4.2441 -0.0651 -2.19E-15
1983 3.9133 -1.4587 -0.3363 -16.6819 -4.0705 -0.0662 1.82E-17
1984 3.6386 -1.2804 -0.3360 -14.5606 -3.7256 -0.0704 -1.67E-15
1985 3.5038 -1.2003 -0.3343 -13.5205 -3.5768 -0.0724 -2.43E-15
1986 3.6225 -1.2233 -0.3471 -14.6238 -3.6005 -0.0728 -3.70E-15
1987 3.9525 -1.3157 . -0.3760 -17.7096 -3.7319 -0.0726 -7.58E-16
1988 4.2130 -1.3641 -0.4051 -20.3964 -3.7826 -0.0736 -3.77E-15
1989 5.1879 -1.5562 -0.5083 -31.1459 -3.9751 -0.0762 1.67E-16
1990 5.2131 -1.6131 -0.4950 -30.9807 -4.0826 -0.0736 -1.69E-15
A5.4. - 12
Table AS.4.4.9 Fitted cost shares and Own AESs for IZEF AR) estimates Model 1 (] SIC 3320)
1972 0.0306 0.3344 0.6350 -31.6253 0.6047 0.1570
1973 0.0288 0.3270 0.6444 -33.6361 0.6559 0.1588
1974 0.0282 0.2953 0.6776 -34.3472 0.9411 0.1669
1975 0.0289 0.3066 0.6666 -33.5721 0.8251 0.1640
1976 0.0315 0.2908 0.6815 -30.7260 0.9933 0.1680
1977 0.0327 0.2853 0.6869 -29.5728 1.0603 0.1696
1978 0.0330 0.2898 0.6820 -29.2237 1.0044 0.1682
1979 0.0298 0.2858 0.6893 -32.4511 1.0544 0.1703
1980 0.0270 0.2688 0.7099 -35.9650 1.2965 0.1769
1981 0.0277 0.2706 0.7088 -35.0571 1.2684 0.1766
1982 0.0319 0.2895 0.6867 -30.2748 1.0080 0.1696
1983 0.0359 0.3138 0.6592 -26.7801 0.7606 0.1621
1984 0.0408 0.3278 0.6414 -23.4843 0.6503 0.1582
1985 0.0448 0.3410 0.6252 -21.3147 0.5630 0.1555
1986 0.0457 0.3350 0.6310 -20.8418 0.6006 0.1564
1987 0.0482 0.3257 0.6385 -19.7236 0.6651 0.1577
1988 0.0467 0.3049 0.6616 -20.3856 0.8417 0.1627
1989 0.0479 0.2658 0.7003 -19.8384 1.3446 0.1738
1990 0.0513 0.2831 0.6803 -18.4591 1.0891 0.1677
Table A?4.4.10. Proper AESs and determinants of matrixes of AESs for IZEF (AR) estimates,
Model 1 (ISIC 3320)
Year Proper Proper Proper Dum of Dum of Dum of Dum of
1972 1.2345 0.8713 -0.3780 -20.6490 -5.7259 -0.0479 0.0025
1973 1.2547 0.8654 -0.3887 -23.6362 -6.0909 -0.0469 -0.0077
1974 1.2878 0.8694 -0.4621 -33.9817 -6.4893 -0.0564 -0.0851
1975 1.2711 0.8701 -0.4315 -29.3161 -6.2614 -0.0509 -0.1358
1976 1.2623 0.8834 -0.4766 -32.1130 -5.9435 -0.0602 -0.2553
1977 1.2576 0.8886 -0.4930 -32.9377 -5.8060 -0.0632 -0.3379
1978 1.2507 0.8890 -0.4802 -30.9174 -5.7056 -0.0616 -0.3232
1979 1.2814 0.8785 -0.4854 -35.8598 -6.2997 -0.0560 -0.3679
1980 1.3307 0.8696 -0.5335 -48.3982 -7.1187 -0.0553 -0.5401
1981 1.3204 0.8726 -0.5258 -46.2114 -6.9510 -0.0525 -0.6460
1982 1.2597 0.8860 -0.4717 -32.1038 -5.9181 -0.0516 -0.5503
1983 1.2128 0.8945 -0.4147 -21.8397 -5.1412 -0.0487 -0.4424
1984 1.1795 0.9045 -0.3918 -16.6638 -4.5340 -0.0506 -0.4004
1985 1.1572 0.9107 -0.3724 -13.3390 -4.1433 -0.0512 -0.3696
1986 1.1566 0.9134 -0.3840 -13.8553 -4.0934 -0.0536 -0.4053
1987 1.1529 0.9188 -0.4069 -14.4474 -3.9540 -0.0607 -0.4356
1988 1.1685 0.9191 -0.4505 -18.5246 -4.1612 -0.0660 -0.5546
1989 1.1883 0.9255 -0.5716 -28.0866 -4.3041 -0.0931 -0.8083
1990 1.1652 0.9284 -0.5196 -21.4622 -3.9572 -0.0873 -0.6789
A5.4. - 13
Table A5.4.4.11 Fitted cost shlires and Own AESs for IZEF (AR) estimates. Model 3 ISIC 3320)
Year Fitted cost Fitted cost Fitted cost OwnAES Own AES OwnAES
1972 0.0358 0.3284 0.6358 -2.1990 -0.0339 0.2549
1973 0.0328 0.3237 0.6435 -0.0019 -0.0194 0.2540
1974 0.0280 0.3007 0.6714 5.7706 0.0733 0.2528
1975 0.0311 0.3103 0.6586 1.5691 0.0302 0.2530
1976 0.0342 0.3006 0.6652 -1.1604 0.0739 0.2529
1977 0.0365 0.2984 0.6650 -2.6273 0.0846 0.2529
1978 0.0347 0.2980 0.6673 -1.4847 0.0866 0.2528
1979 0.0321 0.2985 0.6694 0.6075 0.0843 0.2528
1980 0.0306 0.2924 0.6770 2.1556 0.1169 0.2529
1981 0.0350 0.2980 0.6670 -1.6657 0.0867 0.2528
1982 0.0387 0.3059 0.6554 -3.6681 0.0490 0.2532
1983 0.0416 0.3161 0.6423 -4.7209 0.0074 0.2541
1984 0.0460 0.3208 0.6331 -5.7609 -0.0098 0.2553
1985 0.0484 0.3241 0.6275 -6.1237 -0.0207 0.2561
1986 0.0471 0.3175 0.6354 -5.9464 0.0022 0.2550
1987 0.0436 0.3031 0.6533 -5.2707 0.0618 0.2533
1988 0.0417 0.2916 0.6667 4.7404 0.1213 0.2528
1989 0.0358 0.2596 0.7046 -2.1962 0.3667 0.2547
1990 0.0350 0.2620 0.7030 -1.6855 0.3427 0.2546
Table A5.4.4.12. Proper AESs and determinants of matrixes of AESs for IZEF (AR) estimates,
Model 3 (ISIC 3320)
Year Proper Proper Proper Dtnn of Dtrrn of Dtrrn of Dtrrn of
1972 4.6575 -2.2818 -0.2448 -21.6177 -5.7673 -0.0685 -7.73E-16
1973 5.0525 -2.5417 -0.2476 -25.5277 -6.4610 -0.0662 -3.29E-15
1974 6.1133 -2.9783 -0.2875 -36.9493 -7.4115 -0.0641 -2.35E-15
1975 5.4496 -2.6414 -0.2719 -29.6511 -6.5801 -0.0663 3.04E-16
1976 5.1792 -2.2805 -0.2999 -26.9102 -5.4942 -0.0713 -7.88E-16
1977 4.9428 -2.0737 -0.3096 -24.6538 -4.9646 -0.0744 -2.67E-15
1978 5.1593 -2.2270 -0.3069 -26.7466 -5.3350 -0.0723 1.85E-16
1979 5.4872 -2.4759 -0.3007 -30.0583 -5.9763 -0.0691 -2.36E-15
1980 5.8031 -2.6040 -0.3129 -33.4238 -6.2358 -0.0684 4.79E-15
1981 5.1276 -2.2035 -0.3075 -26.4369 -5.2768 -0.0726 7.20E-16
1982 4.6339 -1.9459 -0.2964 -21.6533 -4.7153 -0.0754 2.01E-15
1983 4.2705 -1.7954 -0.2802 -18.2716 4.4234 -0.0766 1.62E-15
1984 3.9125 -1.5640 -0.2794 -15.2511 -3.9166 -0.0806 3.22E-15
1985 3.7431 -1.4616 -0.2778 -13.8839 -3.7048 -0.0825 3.59E-15
1986 3.8741 -1.4944 -0.2885 -15.0219 -3.7494 -0.0826 1.83E-15
1987 4.2506 -1.6200 -0.3126 -18.3931 -3.9593 -0.0821 3.98E-15
1988 4.5387 -1.6892 -0.3367 -21.1748 -4.0519 -0.0827 3.24E-15
1989 5.6280 -1.9616 -0.4210 -32.4793 -4.4074 -0.0838 4.17E-15
1990 5.6904 -2.0371 -0.4109 -32.9581 -4.5790 -0.0816 6.41E-15
A5.4. - 14
Table AS.4.4.13 Fitted cost shares and Own AESs for IZEF (AR) estimates, Model 4 ISIC 3320)
1972 0.0363 0.3336 0.6301 -6.8167 0.0799 0.2492
1973 0.0338 0.3294 0.6368 -5.8333 0.0950 0.2484
1974 0.0296 0.3049 0.6655 -3.1490 0.2072 0.2470
1975 0.0323 0.3146 0.6530 -5.0533 0.1572 0.2472
1976 0.0347 0.3025 0.6628 -6.2121 0.2208 0.2470
1977 0.0365 0.2992 0.6643 -6.8795 0.2403 0.2470
1978 0.0350 0.2994 0.6656 -6.3475 0.2390 0.2470
1979 0.0329 0.3009 0.6662 -5.3930 0.2302 0.2470
1980 0.0317 0.2945 0.6738 -4.6463 0.2700 0.2471·
1981 0.0352 0.2993 0.6655 -6.4312 0.2397 0.2470
1982 0.0383 0.3069 0.6548 -7.3880 0.1963 0.2471
1983 0.0408 0.3174 0.6418 -7.8885 0.1442 0.2479
1984 0.0444 0.3213 0.6343 -8.3329 0.1273 0.2486
1985 0.0463 0.3242 0.6295 -8.4706 0.1152 0.2492
1986 0.0452 0.3170 0.6377 -8.4006 0.1461 0.2483
1987 0.0423 0.3019 0.6558 -8.1035 0.2241 0.2471
1988 0.0405 0.2896 0.6699 -7.8418 0.3035 0.2470
1989 0.0354 0.2553 0.7094 -6.4901 0.6307 0.2501
1990 0.0348 0.2584 0.7069 . -6.2485 0.5931 0.2498
Table AS.4.4.14. Proper AESs and determinants of matrixes of AESs for IZEF estimates, Model
4 (lSIC 3320)
Year Proper Proper Proper Dtrm of Dtrm of Dtrm of Dtrmof
1972 4.0884 -1.7719 -0.2778 -17.2598 -4.8380 -0.0573 -5.73E-16
1973 4.3581 -1.9445 -0.2805 -19.5469 -5.2298 -0.0551 2.15E-15
1974 5.1375 -2.2136 -0.3238 -27.0466 -5.6779 -0.0537 1.36E-15
1975 4.6772 -2.0034 -0.3072 -22.6702 -5.2627 -0.0555 -2.43E-15
1976 4.5663 -1.7590 -0.3396 -22.2232 -4.6285 -0.0608 I.72E-15
1977 4.4250 -1.6153 -0.3513 -21.2337 -4.3082 -0.0641 -3.32E-15
1978 4.5685 -1.7215 -0.3478 -22.3885 -4.5314 -0.0619 -2.54E-15
1979 4.7740 -1.8898 -0.3400 -24.0321 -4.9032 -0.0587 -3.90E-16
1980 5.0109 -1.9720 -0.3535 -26.3640 -5.0370 -0.0582 -4.47E-16
1981 4.5482 -1.7054 -0.3485 -22.2271 -4.4967 -0.0622 -7.64E-16
1982 4.1808 -1.5273 -0.3366 -18.9291 -4.1585 -0.0648 -1.46E-15
1983 3.8887 -1.4221 -0.3185 -16.2596 -3.9779 -0.0657 -1.69E-15
1984 3.6223 -1.2516 -0.3180 -14.1816 -3.6383 -0.0695 8.60E-16
1985 3.4909 -1.1745 -0.3162 -13.1623 -3.4908 -0.0713 1.45E-15
1986 3.6079 -1.1975 -0.3285 -14.2440 -3.5197 -0.0717 I.22E-15
1987 3.9317 -1.2881 -0.3565 -17.2738 -3.6616 -0.0717 -4.93E-16
1988 4.1877 -1.3364 -0.3845 -19.9162 -3.7232 -0.0728 -1.75E-15
1989 5.1422 -1.5268 -0.4834 -30.5354 -3.9542 -0.0759 -3.63E-15
1990 5.1650 -1.5805 -0.4707 -30.3834 -4.0587 -0.0735 -2.79E-15
A5.4. - 15
Table AS.4.4.1S Fitted cost shares and Own AESs for 13SLS estimates, Model 1 (ISIC 3320)_
1972 0.0316 0.3396 0.6288 -37.1549 0.6766 0.1532
1973 0.0303 0.3311 0.6386 -39.1291 0.7372 0.1550
1974 0.0301 0.2976 0.6723 -39.3887. 1.0526 0.1630
1975 0.0304 0.3090 0.6606 -38.9867 0.9295 0.1599
1976 0.0322 0.2924 0.6754 -36.3714 1.1158 0.1639
1977 0.0329 0.2864 0.6808 -35.4509 1.1944 0.1655
1978 0.0338 0.2911 0.6752 -34.3166 1.1326 0.1638
1979 0.0309 0.2856 0.6835 -38.1201 1.2051 0.1663
1980 0.0280 0.2665 0.7055 -43.0156 1.5045 0.1733
1981 0.0279 0.2677 0.7044 -43.2360 1.4830 0.1729
1982 0.0319 0.2883 0.6798 -36.6924 1.1682 0.1652
1983 0.0360 0.3145 0.6495 -31.7649 0.8766 0.1573
1984 0.0405 0.3300 0.6295 -27.6768 0.7456 0.1534
1985 0.0444 0.3447 0.6109 -24.8043 0.6434 0.1509
1986 0.0456 0.3384 0.6160 -24.0430 0.6849 0.1514
1987 0.0488 0.3294 0.6218 -22.2270 0.7505 0.1522
1988 0.0473 0.3067 0.6460 -23.0553 0.9531 0.1565
1989 0.0491 0.2661 0.6848 -22.0705 1.5105 0.1666
1990 0.0532 0.2851 0.6617 -20.1040 1.2114 0.1602
1 (lSIC 3320)
Year Proper Proper Proper Dtnnof Dtnn of Dtnn of Dtnn of
1972 0.9161 1.3724 -0.4114 -25.9774 -7.5772 -0.0656 -1.66E-15
1973 0.9102 1.3828 -0.4254 -29.6743 -7.9772 -0.0667 -3~71E-15
1974 0.8996 1.3656 -0.5063 -42.2694 -8.2860 -0.0848 -3.99E-16
1975 0.9041 1.3689 -0.4764 -37.0573 -8.1092 -0.0783 3.68E-15
1976 0.9043 1.3406 -0.5261 -41.4012 -7.7588 -0.0939 -2.98E-15
1977 0.9044 1.3308 -0.5461 -43.1603 -7.6365 -0.1006 1.51E-15
1978 0.9084 1.3246 -0.5337 -39.6928 -7.3768 -0.0993 3.32E-15
1979 0.8981 1.3500 -0.5441 -46.7436 -8.1602 -0.0957 4.03E-15
1980 0.8794 1.3747 -0.6031 -65.4895 -9.3424 -0.1 031 2.20E-15
1981 0.8794 1.3768 -0.5983 -64.8911 -9.3707 -0.1016 -2.57E-15
1982 0.9022 1.3409 -0.5379 -43.6788 -7.8579 -0.0964 3.84E-15
1983 0.9206 1.3163 -0.4756 -28.6916 -6.7284 -0.0883 3.47E-15
1984 0.9326 1.2904 -0.4508 -21.5059 -5.9097 -0.0889 -3.88E-15
1985 0.9412 1.2727 -0.4314 -16.8460 -5.3616 -0.0891 -1.23E-15
1986 0.9417 1.2633 -0.4460 -17.3530 -5.2365 -0.0952 2.05E-15
1987 0.9440 1.2439 -0.4716 -17.5718 -4.9298 -0.1082 -2.03E-15
1988 0.9379 1.2423 -0.5212 -22.8528 -5.1517 -0.1224 -3.45E-15
1989 0.9311 1.2202 -0.6538 -34.2052 -5.1667 -0.1758 -6.22E-16
1990 0.9406 1.2103 -0.5976 -25.2391 -4.6857 -0.1630 -3.31E-15
A5.4. - 16
Table AS.4.4.17 Fitted cost shares and Own AESs for I3SLS estimates. Model 2 (lSIC 3320)
1972 0.0398 0.3493 0.6109 -7.9644 -0.0122 0.2763
1973 0.0363 0.3366 0.7018 -7.1294 0.0221 0.2670
1974 0.0308 0.3081 0.7784 -4.5091 0.1331 0.2778
1975 0.0334 0.3156 0.7915 -6.0065 0.0983 0.2806
1976 0.0354 0.3083 0.7701 -6.8213 0.1319 0.2761
1977 0.0368 0.3053 0.7791 -7.2820 0.1470 0.2779
1978 0.0351 0.3020 0.8119 -6.7138 0.1646 0.2853
1979 0.0318 0.2911 0.9336 -5.1545 0.2291 0.3199
1980 0.0291 0.2746 1.0501 -3.1242 0.3529 0.3568
1981 0.0323 0.2783 1.0635 -5.4260 0.3224 0.3610
1982 0.0363 0.2925 0.9875 -7.1086 0.2202 0.3368
1983 0.0397 0.3086 0.9162 -7.9457 0.1306 0.3145
1984 0.0442 0.3212 0.8266 -8.5167 0.0753 0.2890
1985 0.0468 0.3302 0.7623 -8.6773 0.0426 0.2747
1986 0.0453 0.3219 0.7842 -8.5988 0.0725 0.2790
1987 0.0422 0.3094 0.7771 -8.3171 0.1268 0.2775
1988 0.0391 0.2905 0.8646 -7.8387 0.2331 0.2993
1989 0.0329 0.2572 0.9043 -5.7360 0.5259 0.3109
1990 0.0330 0.2638 0.8584 -5.7826 0.4536 0.2975
Table A5.4.4.18. Proper AESs and determinants of matrixes of AESs for I3SLS estimates, Model .
2 (ISIC 3320)
1972 4.2153 -1.8914 -0.2676 -17.6718 -5.7775 -0.0750 2.48£-15
1973 4.6556 -1.7574 -0.1452 -21.8321 -4.9922 -0.0152 -3.3724
1974 5.7046 -1.9284 -0.1279 -33.1423 -4.9711 0.0206 -6.8137
1975 5.2365 -1.6569 -0.0828 -28.0114 -4.4306 0.0207 -6.6503
1976 5.0950 -1.5784 -0.1392 -26.8586 -4.3749 0.0170 -5.3745
1977 4.9740 -1.4493 -0.1372 -25.8112 -4.1243 0.0220 -5.3670
1978 5.2163 -1.4665 -0.1033 -28.3146 -4.0661 0.0363 -6.7798
1979 5.8221 -1.3651 0.0048 -35.0783 -3.5123 0.0733 -11.7235
1980 6.5961 -1.3014 0.0618 -44.6112 -2.8083 0.1221 -17.5628
1981 5.9722 -1.0461 0.0859 -37.4170 -3.0532 0.1090 -14.8950
1982 5.2139 -0.9632 0.0636 -28.7505 -3.3221 0.0701 -10.4978
1983 4.6490 -0.9329 0.0432 -22.6512 -3.3695 0.0392 -7.5978
1984 4.1521 -0.9262 -0.0188 -17.8813 -3.3193 0.0214 -5.0851
1985 3.8957 -0.9724 -0.0747 -15.5463 -3.3289 0.0061 -3.6959
1986 4.0664 -0.9798 -0.0715 -17.1591 -3.3589 0.0151 -4.2431
1987 4.4267 -1.1456 -0.1252 -20.6505 -3.6205 0.0195 -4.4971
1988 4.9307 -1.0773 -0.0768 -26.1385 -3.5067 0.0639 -7.2314
1989 6.2861 -1.3642 -0.1633 -42.5309 -3.6443 0.1368 -11.2485
1990 6.1380 -1.4836 -0.1944 -40.2980 -3.9217 0.0972 -9.2296
A5.4. - 17
Table AS.4.4.19 Fitted cost shares and Own AESs for 13SLS estimates. Model 3 (ISIC 3320)
Year Fitted cost Fitted cost Fitted cost Own AES Own AES OwnAES
1972 0.0358 0.3345 0.6297 -7.0366 0.1134 0.2497
1973 0.0334 0.3304 0.6362 -6.0645 0.1287 0.2489
1974 0.0294 0.3055 0.6651 -3.4953 0.2477 0.2474
1975 0.0320 0.3153 0.6527 -5.3331 0.1952 0.2477
1976 0.0344 0.3024 0.6632 -6.5039 0.2660 0.2474
1977 0.0362 0.2988 0.6650 -7.1595 0.2886 0.2474
1978 0.0347 0.2992 0.6661 -6.6426 0.2861 0.2474
1979 0.0327 0.3010 0.6664 -5.7087 0.2749 0.2474
1980 0.0314 0.2945 0.6740 -5.0015 0.3172 0.2476
1981 0.0349 0.2991 0.6660 -6.7242 0.2870 0.2474
1982 0.0379 0.3066 0.6555 -7.6415 0.2416 0.2476
1983 0.0403 0.3172 0.6425 -8.1168 0.1860 0.2483
1984 0.0439 0.3208 0.6354 -8.5448 0.1693 0.2490
1985 0.0458 0.3236 0.6307 -8.6751 0.1570 0.2496
1986 0.0447 0.3162 0.6390 -8.6112 0.1906 0.2486
1987 0.0419 0.3010 0.6572 -8.3364 . 0.2751 0.2475
1988 0.0402 0.2884 0.6714 -8.0972 0.3615 0.2475
1989 0.0353 0.2534 - 0.7113 -6.8697 0.7183 0.2507
1990 0.0347 0.2567 0.7086 -6.6365 0.6757 0.2504
3 (ISIC 3320)
1972 3.9812 -1.7148 -0.2866 -16.6479 -4.6977 -0.0538 -6.20E-16
1973 4.2383 -1.8832 -0.2891 -18.7439 -5.0561 -0.0515 -3.02E-15
1974 4.9770 -2.1319 -0.3336 -25.6365 -5.4100 -0.0500 -1.20E-15
1975 4.5409 -1.9324 -0.3168 -21.6613 -5.0550 -0.0520 -4.77E-15
1976 4.4355 -1.6858 -0.3511 -21.4036 -4.4513 -0.0574 -3.29E-15
1977 4.3018 -1.5436 -0.3637 -20.5720 -4.1542 -0.0609 -1.5lE-15
1978 4.4378 -1.6475 -0.3597 -21.5942 -4.3580 -0.0586 -U8E-15
1979 4.6323 -1.8127 -0.3512 -23.0278 -4.6985 -0.0553 -1.95E-15
1980 4.8567 -1.8890 -0.3651 -25.1744 -4.8068 -0.0547 -U8E-15
1981 4.4185 -1.6317 -0.3605 -21.4527 -4.3262 -0.0590 -1.97E-15
1982 4.0695 -1.4611 -0.3485 -18.4070 -4.0267 -0.0617 1.39E-15
1983 3.7910 -1.3623 -0.3298 -15.8808 -3.8713 -0.0626 -6.00E-16
1984 3.5370 -1.1958 -0.3297 -13.9571 -3.5576 -0.0665 -1.32E-15
1985 3.4115 -1.1208 -0.3280 -13.0003 -3.4214 -0.0684 -1.37E-15
1986 3.5238 -1.1411 -0.3410 -14.0587 -3.4432 -0.0689 5.17E-16
1987 3.8338 -1.2248 -0.3702 -16.9906 -3.5635 -0.0689 -1.62E-15
1988 4.0788 -1.2672 -0.3996 -19.5638 -3.6101 -0.0702 -2.98E-16
1989 4.9896 -1.4364 -0.5036 -29.8310 -3.7856 -0.0735 -1.60E-15
1990 5.0096 -1.4896 -0.4899 -29.5802 -3.8805 -0.0709 -5.l2E-17
A5.4. - 18
Table AS.4.4.21 Fitted cost shares and Own AESs for I3SLS estimates. Model 4 (ISle 3320)
Year Fitted cost Fitted cost Fitted cost OwnAES Own AES Own AES
1972 0.0359 0.3343 0.6298 -6.7591 0.1204 0.2585
1973 0.0334 0.3301 0.6365 -5.7345 0.1364 0.2575
1974 0.0293 0.3051 0.6656 -2.9700 0.2575 0.2553
1975 0.0320 0.3150 0.6530 -4.9371 0.2038 0.2559
1976 0.0343 0.3023 0.6633 -6.1537 0.2743 0.2554
1977 0.0362 0.2988 0.6650 -6.8477 0.2964 0.2554
1978 0.0347 0.2991 0.6662 -6.2967 0.2943 0.2553
1979 0.0326 0.3008 0.6666 -5.3061 0.2838 0.2553
1980 0.0313 0.2943 0.6744 -4.5401 0.3268 0.2553
1981 0.0349 0.2990 0.6661 -6.3834 0.2952 0.2553
1982 0.0380 0.3066 0.6554 -7.3703 0.2487 0.2557
1983 0.0404 0.3173 0.6422 -7.8864 0.1923 0.2568
1984 0.0440 0.3211 0.6349 -8.3499 0.1747 0.2577
1985 0.0459 0.3240 0.6301 -8.4946 0.1619 0.2585
1986 0.0449 0.3166 0.6385 -8.4217 0.1958 0.2573
1987 0.0419 0.3013 0.6568 -8.1148 0.2809 0.2557
1988 0.0402 0.2887 0.6711 -7.8477 0.3678 0.2553
1989 0.0351 0.2536 0.7113 -6.4786 0.7265 0.2576
1990 0.0345 0.2568 0.7087 -6.2289 0.6844 0.2573
Table AS.4.4.22 Proper AESs and determinants of matrixes of AESs for I3SLS estimates, Model
4 (ISIC 3320)
1972 4.0746 -1.7776 -0.2962 -17.4166 -4.9072 -0.0566 -2.85E-15
1973 4.3444 -1.9522 -0.2989 -19.6554 -5.2877 -0.0542 -5.27E-15
1974 5.1258 -2.2192 -0.3438 -27.0390 -5.6832 -0.0524 -8.12E-15
1975 4.6644 -2.0083 -0.3266 -22.7630 -5.2967 -0.0545 -4.79E-15
1976 4.5550 -1.7576 -0.3608 -22.4359 -4.6607 -0.0601 -4.34E-15
1977 4.4145 -1.6114 -0.3732 -21.5174 -4.3453 -0.0636 -4.48E-15
1978 4.5577 -1.7189 -0.3694 -22.6258 -4.5625 -0.0613 -3.53E-15
1979 4.7626 -1.8896 -0.3610 -24.1884 -4.9254 -0.0579 -7.07E-15
1980 5.0003 -1.9713 -0.3750 -26.4872 -5.0450 -0.0572 -6.48E-15
1981 4.5374 -1.7026 -0.3701 -22.4719 -4.5286 -0.0616 -3.52E-15
1982 4.1697 -1.5240 -0.3579 -19.2199 -4.2074 -0.0645 -7.86E-15
1983 3.8771 -1.4195 -0.3390 -16.5484 -4.0404 -0.0655 -3040E-15
1984 3.6114 -1.2476 -0.3387 -14.5008 -3.7087 -0.0697 -2.69E-15
1985 3.4803 -1.1701 -0.3369 -13.4877 -3.5649 -0.0716 -5.22E-15
1986 3.5977 -1.1923 -0.3499 -14.5920 -3.5883 -0.0720 -7.09E-15
1987 3.9225 -1.2814 -0.3791 -17.6654 -3.7167 -0.0719 -3.72E-15
1988 4.1800 -1.3281 -0.4086 -20.3589 -3.7673 -0.0730 -5.68E-15
1989 5.1422 -1.5133 -0.5130 -31.1488 -3.9593 -0.0760 -6. 13E-15
1990 5.1637 -1.5680 -0.4994 -30.9262 -4.0615 -0.0733 -1.39E-14
A5.4. - 19
Table AS.4.4.23 Fitted cost shares and Own AESs for 13SLS (AR) estimates, Model 2 (ISle
3320)
1972 0.0339 0.2396 0.7265 -1.9586 0.8171 0.1321
1973 0.0290 0.2315 0.7396 2.8145 0.9561 0.1386
1974 0.0270 0.2036 0.7695 5.8374 1.6163 0.1537
1975 0.0281 0.2093 0.7626 3.9721 1.4531 0.1502
1976 0.0351 0.1926 0.7722 -2.7408 1.9821 0.1551
1977 0.0384 0.1849 0.7768 -4.3397 2.2946 0.1575
1978 0.0353 0.1850 0.7797 -2.8428 2.2887 0.1590
1979 0.0288 0.1802 0.7910 3.0798 2.5038 0.1647
1980 0.0255 0.1653 0.8092 8.6887 3.3345 0.1741
1981 0.0292 0.1636 0.8072 2.5757 3.4462 0.1731
1982 0.0338 0.1734 0.7928 -1.8572 2.8506 0.1657
1983 0.0366 0.1874 0.7760 -3.5516 2.1865 0.1570
1984 0.0432 0.1931 0.7637 -5.8060 1.9663 0.1508
1985 0.0469 0.1985 0.7546 -6.4624 1.7780 0.1461
1986 0.0466 0.1901 0.7633 -6.4119 2.0794 0.1506
1987 0.0461 0.1782 0.7757 -6.3457 2.6046 0.1569
1988 0.0444 0.1595 0.7962 -6.0484 3.7385 0.1674
1989 0.0444 0.1247 0.8309 -6.0432 7.7093 0.1853
1990 0.0439 0.1332 0.8229 -5.9482 6.3989 0.1811
Model 2 (Isle 3320)
Year Proper Proper Proper Dtrm of Dtnn of Dtrm of Dtrm of
1972 1.5417 -0.4171 -0.3414 -3.9772 -0.4326 -0.0087 -3.44E-16
1973 1.6561 -0.6285 -0.3641 -0.0516 -0.0051 -0.000 1 -3.50E-18
1974 1.8012 -0.6812 -0.4908 6.1908 0.4333 0.0076 6.79E-16
1975 1.7471 -0.6260 -0.4632 2.7196 0.2048 0.0037 -9.50E-17
1976 1.6505 -0.2871 -0.5695 -8.1567 -0.5076 -0.0169 -4.48E-16
1977 1.6206 -0.1714 -0.6261 -12.5841 -0.7127 -0.0307 -9.06E-16
1978 1.6741 -0.2685 -0.6188 -9.3088 -0.5240 -0.0191 1.53E-16
1979 1.8483 -0.5332 -0.6378 4.2950 0.2230 0.0057 5.45E-16
1980 2.0438 -0.6913 -0.7456 24.7954 1.0348 0.0246 7.12E-16
1981 1.9222 -0.4827 -0.7679 5.1813 0.2129 0.0068 4.25E-16
1982 1.7516 -0.3041 -0.6982 -8.3622 -0.4002 -0.0152 -6.25E-16
1983 1.6415 -0.2290 -0.6055 -10.4603 -0.6102 -0.0233 -1.28E-15
1984 1.5275 -0.0577 -0.5835 -13.7493 -0.8788 -0.0440 -7. 14E-16
1985 1.4722 0.0149 -0.5592 -13.6578 -0.9447 -0.0529 -1.64E-15
1986 1.4969 0.0185 -0.6092 -15.5734 -0.9659 -0.0580 -2.IOE-15
1987 1.5353 0.0248 -0.6895 -18.8849 -0.9963 -0.0668 -1.28E-15
1988 1.6216 0.0124 -0.8392 -25.2415 -1.0126 -0.0785 -3.98E-15
1989 1.7953 0.0531 -1.2531 -49.8115 -1.1224 -0.1420 -1.68E-15
1990 1.7526 0.0334 -1.1296 -4 l.l 340 -1.0785 -0.1169 -4.59E-15
A5.4. - 20
Table AS.4.4.2S Fitted cost shares and Own AESs for I3SLS (AR) estimates, Model 3 (ISIC
3320)
1972 0.0322 0.3393 0.6285 4.6650 0.1531 0.2788
1973 0.0288 0.3355 0.6357 9.6389 0.1677 0.2772
1974 0.0250 0.3099 0.6651 18.5277 0.2908 0.2732
1975 0.0281 0.3198 0.6521 10.9928 0.2376 0.2745
1976 0.0330 0.3057 0.6613 3.7533 0.3162 0.2735
1977 0.0362 0.3016 0.6623 0.8791 0.3429 0.2734
1978 0.0338 0.3022 0.6639 2.8855 0.3383 0.2733
1979 0.0305 0.3045 0.6650 6.9050 0.3237 0.2732
1980 0.0292 0.2979 0.6729 8.9349 0.3678 0.2727
1981 0.0342 0.3021 0.6638 2.5557 0.3396 0.2733
1982 0.0381 0.3094 0.6525 -0.4685 0.2938 0.2745
1983 0.0408 . 0.3202 0.6390 -1.8922 0.2354 0.2765
1984 0.0460 0.3233 0.6308 -3.7133 0.2204 0.2782
1985 0.0486 0.3259 0.6255 -4.3354 0.2084 0.2795
1986 0.0477 0.3183 0.6340 -4.1444 . 0.2447 0.2775
1987 0.0447 0.3027 0.6525 -3.3607 0.3351 0.2745
1988 0.0434 0.2898 0.6669 -2.9126 0.4286 0.2731
1989 0.0391 0.2537 0.7072 -1.0331 0.8149 0.2730
1990 0.0378 0.2573 0.7049 -0.2819 0.7659 0.2729.
Model 3 (ISIC 3320)
Year Proper Proper Proper Dtrm of Dtrmof Dtrm of Dtrm of
1972 4.0113 -2.4045 -0.2882 -15.3765 -4.4814 -0.0404 2.56E-15
1973 4.4027 -2.7602 -0.2881 -17.7667 -4.9471 -0.0365 1.38E-16
1974 5.2423 -3.1400 -0.3328 -22.0931 -4.7973 -0.0313 2.6lE-15
1975 4.6608 -2.7590 -0.3174 -19.1118 -4.5947 -0.0355 5.65E-16
1976 4.2610 -2.1572 -0.3588 -16.9695 -3.6267 -0.0423 -3.57E-16
1977 4.0174 -1.8772 -0.3755 -15.8379 -3.2834 -0.0472 3.55E-15
1978 4.2166 -2.0667 -0.3689 -16.8034 -3.4825 -0.0437 2.6IE-15
1979 4.5419 -2.3965 -0.3566 -18.3930 -3.8566 -0.0387 3.00E-15
1980 4.7790 -2.5040 -0.3704 -19.5529 -3.8329 -0.0369 6.19E-15
1981 4.1865 -2.0367 -0.3701 -16.6586 -3.4497 -0.0442 3.20E-15
1982 3.7883 -1.7691 -0.3607 -14.4886 -3.2582 -0.0495 2.07E-15
1983 3.5173 -1.6415 -0.3426 -12.8169 -3.2180 -0.0523 4.29E-15
1984 3.2146 -1.3771 -0.3471 -11.1521 -2.9294 -0.0592 1.99E-15
1985 3.0773 -1.2662 -0.3476 -10.3730 -2.8148 -0.0626 3.00E-15
1986 3.1662 -1.2780 -0.3612 -11.0392 -2.7835 -0.0625 3.06E-15
1987 3.4300 -1.3610 -0.3905 -12.8913 -2.7747 -0.0606 4.31E-15
1988 3.6187 -1.3832 -0.4215 -14.3434 -2.7085 -0.0606 1.80E-15
1989 4.3156 -1.4912 -0.5310 -19.4661 -2.5057 -0.0595 5.47E-15
1990 4.3796 -1.5836 -0.5147 -19.3966 -2.5846 -0.0559 2.03E-15
A5.4. - 21
Table AS.4.4.27 Fitted cost shares and Own AESs for I3SLS (AR) estimates, Model 4 (ISIC
3320)
1972 0.0366 0.3329 0.6305 -6.3905 0.0886 0.2642
1973 0.0340 0.3285 0.6375 -5.3349 0.1049 0.2631
1974 0.0296 0.3039 0.6665 -2.3356 0.2204 0.2605
1975 0.0324 0.3138 0.6538 -4.4432 0.1684 0.2612
1976 0.0347 0.3021 0.6632 -5.6505 0.2309 0.2606
1977 0.0365 0.2990 0.6644 -6.3721 0.2492 0.2606
1978 0.0350 0.2991 0.6659 -5.7888 0.2489 0.2605
1979 0.0329 0.3003 0.6667 -4.7513 0.2414 0.2605
1980 0.0316 0.2939 0.6745 -3.9059 0.2823 0.2604
1981 0.0352 0.2990 0.6657 -5.8802 0.2494 0.2605
1982 0.0385 0.3068 0.6547 -6.9479 0.2042 0.2611
1983 0.0410 0.3175 0.6415 -7.5144 0.1510 0.2625
1984 0.0447 0.3217 0.6336 -8.0119 0.1325 0.2637
1985 0.0467 0.3247 0.6286 -8.1713 0.1197 0.2646
1986 0.0456 0.3175 0.6369 -8.0864 0.1507 0.2631
1987 0.0424 0.3024 0.6552 -7.7357 0.2293 0.2611
1988 0.0406 0.2900 0.6694 -7.4246 0.3089 0.2604
1989 0.0351 0.2556 0.7093 -5.8116 0.6370 0.2620
1990 0.0345 0.2586 0.7069 -5.5450 0.6008 0.2618
Model 4 (ISIC 3320)
Year Proper Proper Proper Dtnn of Dtnn of Dtnn of Dtrm of
1972 4.2583 -1.8774 -0.2940 -18.6998 -5.2131 -0.0630 4.22E-15
1973 4.5505 -2.0599 -0.2971 -21.2665 -5.6464 -0.0606 5.30E-15
1974 5.4093 -2.3626 -0.3410 -29.7751 -6.1904 -0.0588 3.4IE-15
1975 4.9023 -2.1325 -0.3239 -24.7809 -5.7081 -0.0609 4.4IE-16
1976 4.7859 -1.8841 -0.3557 -24.2094 -5.0224 -0.0663 3.34E-15
1977 4.6325 -1.7346 -0.3670 -23.0478 -4.6691 -0.0697 5.89E-15
1978 4.7891 -1.8466 -0.3637 -24.3762 -4.9179 -0.0675 2.94E-15
1979 5.0131 -2.0233 -0.3564 -26.2786 -5.3313 -0.0641 5.29E-15
1980 5.2750 -2.1154 -0.3701 -28.9286 -5.4919 -0.0635 2.7IE-15
1981 4.7669 -1.8296 -0.3644 -24.1897 -4.8793 -0.0678 3.32E-15
1982 4.3652 -1.6376 -0.3520 -20.4136 -4.4960 -0.0706 3.79E-15
1983 4.0470 -1.5220 -0.3336 -17.5126 -4.2890 -0.0717 3.60E-15
1984 3.7590 -1.3426 -0.3327 -15.1918 -3.9151 -0.0757 2.69E-15
1985 3.6173 -1.2615 -0.3307 -14.0631 -3.7532 -0.0777 2.62E-15
1986 3.7439 -1.2881 -0.3430 -15.2351 -3.7871 -0.0780 4.10E-15
1987 4.0954 -1.3891 -0.3709 -18.5462 -3.9493 -0.0777 4.85E-15
1988 4.3749 -1.4457 -0.3989 -21.4330 -4.0236 -0.0787 6.89E-15
1989 5.4268 -1.6682 -0.4980 -33.1518 -4.3055 -0.0811 4.88E-15
1990 5.4522 -1.7240 -0.4857 -33.0575 -4.4236 -0.0787 7.08E-15
A5.4. - 22
AS.4.S Economies of Scale
Table A5.4.5.1 IZEF estimates of Economies of Scale (ISle 3320)

of Scale- of Scale - of Scale- of Scale- of Scale- of Scale-
Model I - Model 2 - Mode13 - Model1 - Mode12 - Mode13 -
lZEF lZEF IZEF IZEF (AR) IZEF(AR) IZEF(AR)
1972 1.1867 1.4320 1.2575 1.3729 1.6573
1973 1.1903 1.4083 1.2286 1.3335 1.4713
1974 1.1854 1.3798 1.2126 1.2998 1.3827
1975 1.2127 1.3798 1.2041 1.3074 1.3390
1976 1.2360 1.3776 1.2139 1.3299 1.3896
1977 1.2588 1.3727 1.2112 1.3377 1.3752
1978 1.2768 1.3660 1.2011 1.3312 1.3237
1979 1.2787 1.3397 1.1633 1.2800 1.1557
1980 1.2781 1.3116 1.1313 1.2345 1.0368
1981 1.3079 1.3089 1.1248 1.2412 1.0150
1982 1.3587 1.3272 1.1434 1.2905 1.0796
1983 1.4138 1.3468 1.1607 1.3414 1.1450
1984 1.4748 1.3673 1.1857 1.4070 1.2510
1985 1.5314 1.3829 1.2046 1.4616 1.3413
1986 1.5558 1.3748 1.2002 1.4599 1.3194
1987 1.5801 1.3710 1.2085 1.4738 1.3611
1988 1.5819 1.3458 1.1849 1.4298 1.2473
1989 1.5744 1.3241 1.1849 1.4105 .1.2473
1990 1.6297 1.3361 1.1988 1.4517 1.3126
MEAN 1.3743 1.3607 1.1905 1.3576 1.2869
A5.4. - 23
...
Table AS 4 S 2 I3SLS estimates of Economies of Scale (ISle 3320)
of Scale- of Scale- of Scale- of Scale- of Scale- of Scale-
Model 1 - Model 2 - Model 3 - Model 1 - Model 2 - Model 3 -
I3SLS 13SLS I3SLS 13SLS (AR) I3SLS (AR) 13SLS (AR)
1972 1.1027 1.3308 1.1927 0.3939 0.4303
1973 1.1531 1.1823 1.1916 0.5193 0.5508
1974 1.1709 0.9802 1.1910 0.6323 0.6557
1975 1.2038 0.8956 1.1906 0.7209 0.7313
1976 1.1869 0.8232 1.1910 0.6218 0.6456
1977 1.1997 0.7659 1.1909 0.6463 0.6672
1978 1.2265 0.7316 1.1905 0.7577 0.7633
1979 1.3226 0.6908 1.1889 2.3037 1.7312
1980 1.4188 0.6379 1.1874 -2.7437 -11.9165
1981 1.4613 0.5961 1.1871 -1.8699 -4.4982
1982 1.4222 0.5622 1.1880 -16.1364 5.8456
1983 1.3924 0.5349 1.1888 2.7666 1.9088
1984 1.3410 0.5163 1.1899 1.0464 0.9832
1985 1.3074 0.4900 1.1907 0.7202 0.7267
1986 1.3202 0.4608 1.1905 0.7749 0.7731
1987 1.2938 0.4346 1.1908 0.6743 0.6904
1988 1.3501 0.4148 1.1898 1.0576 0.9986
1989 1.3256 0.3913 1.1898 1.0465 0.9986
1990 1.3012 0.3714 1.1904 0.7829 0.7889
MEAN 1.2895 0.6743 1.1900 -0.2781 0.1829
A5.4. - 24
A5.4.6 Graphs of Economies of Scale
Figure A5.4.6.1 Economies of Scale· IZEF estimates· ISIC 3320
1.8l
1.6
1.4 -%-%-X-~-~_~_~_~_%.-A_
-X-X-X-X-X-X-X
12 X- -X-X-X~X
•
v
-X-X-X- X-X-X-X-X-X-X-X
v
0.8
0.6 -O--EOS-1
0.4
-%-EOS-2
-X-EOS-3
0.2
0+--+-+----+--1---+----11---+---4--+--+-+--+-+----+--+--+--1--...
n ~ ~ ~ n n n ~ 00 M ~ ~ ~ ~ ~ ~ ~ ~ 00
Figure A5.4.6.2 Economies of Scale· IZEF (AR) estimates· ISIC 3320
0.8
0.6 1'r----~
-O--EOS-1
0.4
-X-EOS-3
0.2
0i--+-+--+-1---+--1---+---4--+--+-+--+-+----+--t---+----11---t
n ~ ~ ~ n n n ~ 00 M ~ ~ ~ ~ ~ ~ ~ ~ 00
A5.4. - 25
Figure AS.4.6.3 Economies of Scale - I3SLS estimates· ISle 3320
-x-x-x-x-x-x-x-x-x-x-x-x
0.8
0.6
~EOS-1
0.4 -X-EOS-2
-X-EOS-3
0.2
O+--+-+--+-+--+--+---f--+---f--+---f--+---f-......--l--+---l~-4
n n n ~ n n n ~ 00 ~ ~ ro ~ ~ ~ ~ ~ ~ ~
Figure AS.4.6.4 Economies of Scale· I3SLS (AR) estimates· ISIC 3320.

1.2
~ J-~
\-~-\
::+~
0.4 if"
0.21. I-X-EOS-2
" .-X- EOS - 3
o I I I I I I
n n n ~ n n n ~ 00 ~ ~ ro ~ ~ ~ ~ ~ ~ ~
A5.4. - 26
Figure AS.4.6.S. Economies of Scale - Model 1 - ISIC 3320

1.8
1.6
14
.
1.2 t
-X-X-X-~=~==~~~
_ ~_~~ ~}F..-
V
0.8
0.6 -D-13SLS - EOS - 1

-X-IZEF _- EOS - 1
0.4
-X-IZEF(AR)- EOS-1
0.2
0+--+-+--+-+--+-+--+-+--+-+--+-+--+-+--+-0+-......- ....
n n ~ ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ ~ ~ 00
Figure AS.4.6.6. Economies of Scale - Model 2 - ISIC 3320
1~ T .
1.4 X-X-x-x-x-x-x-x_x_x_x_x-x-x-x-x-x_x_x
1.2
0.8
0.6
0.4 -Q-13S'LS - EOS • 2

~ 13SLS (AR) . EOS • 2
0.2
-X-IZEF - EOS - 2
o+---+--+=::::;:=+=i==+==+=+-+--+-~+---+--+--+--+-+--I
n n ~ ~ n n n n 00 ~ ~ ~ M ~ ~ ~ ~ ~ 00
A5.4. - 27
2h
Figure A5.4.6.7. Economies of Scale - Model 3 - ISle 3320
1.8
1.6
-D-13SLS. EOS - 3
~13SLS(AR)-EOS-3
-X-IZEF- EOS·3
-)I(-IZEF (AR) - EOS - 3
1.4 ......)1(-)1(-)1(-)1(- ........)1(-)1(-)1(, _ _)I(
1.2 - -~-~-~-~-~,-f¥ilJ=C}-{~~~~~-~-~~~
0.8
0.6
0.4
0.2
O+--+-+---+-+---+--+---f--+---f--+---f--+---f--+-~"""-+----1~~
n n N ~ n n n ~ 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ 00
APPENDIX 5.5
ELECTRICAL APPLIANCES AND HOUSEHOLD GOODS (ISIC 3833)
.
T a ble A5.521
. C ompu too ,nd cri"fleaIDur b·IR- Wa tsonstat'ISt'ICS tor IZEF est'Imates 0 f ISIC 3833
Explanatory's D-W vs
Statistic HI: P>O
Cost share of Labour 4 0.859 1.848 0.6343 Reject
Model 3 Cost function 7 0.549 2.396 0.1923 Reject
A5.5. - 2
Tab~e AS.S.2.2 Computed and critical Durbin-Watson statistics for I3SLS estimates of ISIC
383333
Exogenous D-W vs
Variables Statistic Hl: P>O
Model 1 Cost function 15 0.063 3.676 0.7985 N.D
Cost share of Capital 15 0.063 3.676. 1.5004 N.D.
Note: Significant Lower and Upper D-W statis.tics are given for the 0.05 level of significance (Source:
A5.5. - 3

Table AS.S3.! IZEF parameter estimates for Electrical Appliances and Household Goods (ISle
3833).
Parameter Modell Model 2 Mode13 Model 4 ModelS
ao 16.5504 16.5493 16.2943 16.3011 16.3063
(598.017) (437.412) (482.088) (499.395) (544.891)
aK 0.0961 0.0908 0.0827 0.0830 0.0803
(8.9697) (8.9555) (12.8733) (12.9137) (18.4029)
aL 0.4973 0.5012 0.3713 0.3700 0.3465
(25.3261) (21.4250) (19.9111) (19.5238) (19.1818)
aM 0.4066 0.4080 0.5460 0.5470 0.5732
ay 0.7529 0.5244 0.9943 0.9817 0.9233

(6.9942) (6.7923) (20.7089) (59.5296) (36.6742)
at -0.0147
-3.2152
YKK 0.0231 -0.0225 -0.0089 -0.0087
(0.5269) (-1.2139) (-0.5762) (-0.5593)
YKL -0.0085 0.0103 0.0155 0.0208
(-0.1772) (0.5016) (0.8660) 0.1463)
YKM -0.0146 0.0122 -0.0066 -0.0121
Yu 0.1418 0.1500 0.1679 0.1648

(2.3773) (5.2564) (6.2324) (6.2168)
YIM -0.1333 -0.1603 -0.1834 -0.1856
YMM 0.1479 0.1481 0.1900 0.1977
YKY -0.0099 -0.0101

-0.2775 (-0.4564)
YLY -0.1016 -0.2524
-1.6519 (-5.8322)
YMY 0.1115 0.2625
Yyy 0.3083 -0.0955 0.0006

(0.7497) (-0.8449) (0.0058)
YIt 0.0009
(1.1526)
YYt -0.0163
(-1.0072)
YKt -0.0017
(-0.5695)
YLt -0.0072
(-1.8300)
YMc 0.0089
Dtrrn 2.49E-11 6.13E-l1 2.92E-1O 3.07E-1O 1.56E-09

Note: 1. Numbers in parenthesis are I-statistics refering to parameter estimates above. I-statistics do
not exist for indirectly calculated parameters. .
2. Dtrm is the determinant of the residual covaraince matrix.
A5.5. - 4
Table AS.S.3.2 IZEF AR(l) parameter estimates for Electrical AppUances and Household Goods
(ISle 3883),
aa 17.2338 16.4796 16.2533 16.2264 16.4235
(108.687) (229.222) (244.493) (237.248) (326.193)
a K 0.1632 0.1061 0.0704 0.0752 0.0801
(7.2401) (5.5632) (4.1855) (4.5064) (18.243)
a L 0.2889 0.3988 0.3064 0.2954 0.3669
(4.2649) (7.5443) (6.0229) (5.4589) (20.746)
aM 0.5479 0.4951 0.6232 0.6294 0.5530
ay 0.3484 0.5869 0.8890 0.9902 0.7489

(5.8932) (4.9258) (11.978) (33.756) (10.344)
at -0.1018
(-5.3029)
'YKK 0.1807 0.0139 0.0343 0.0242
(6.3684) (0.5281) (1.0530) (0.7416)
'YKL -0.1340 0.0582 0.0417 0.0524
(-3.3832) (2.2454) (1.4040) (1.8168)
'YKM -0.0467 -0.0721 -0.0760 -0.0766
'YLL 0.2935 0.1107 0.1082 0.0993

(4.0092) (2.5425) (2.3244) (2.1941)
'Yud -0.1595 -0.1689 -0.1499 -0.1517
'Y MM 0.2062 0.2410 0.2259 0.2283
'YKY -0.0152 -0.0475

(-0.5487) (-1.6851)
'Y LY -0.0956 -0.1178
(-2.2950) (-2.5098)
'YMY 0.1108 0.1653
'Yyy 0.6159 0.1327 0.1974

(4.4423) (0.8148) (1.2870)
'Y1t 0.0071
(5.7044)
'Y Yt -0.0018
(-0.2995)
'Y Kt -0.0124
(-4.7011)
'YLt 0.0132
(3.1161)
'Y Mt -0.0008
Dtnn 3.24E-12 2.30E-l ] 5.18E-ll 5.5IE-ll 2.99E-09

Note: 1. Numbers in parenthesis are [-statistics refering to parameter estimates above. [-statistics do
2. Dtrm is the determinant of the residual covaraince matrix.
A5.5. - 5
Table AS.S.3.3 13SLS parameter estimates for Electrical Appliances and Household Goods (ISle
3883)
<:x. o 16.5515 16.5573 16.2873 16.2981 16.3122
(598.487) (430.076) (470.832) (494.906) (549.177)
<:X. K 0.0958 0.0876 0.0825 0.0829 0.0803
(8.9262) (8.3544) (12.8434) (1'2.8951) (18.3882)
<:X. L 0.4976 0.5093 0.3721 0.3701 0.3470
(25.2327) (20.9961) (20.1549) (19.5746) (19.2032)
<:x. M 0.4066 0.4031 0.5454 0.5470 0.5727
<:X. Y 0.7193 0.5173 1.0016 0.9873 0.9138

(6.5405) (6.5282) (20.0166) (58.4458) (34.8265)
<:x' t -0.0138
(-2.9929)
"fKK 0.0243 -0.0279 -0.0089 -0.0083
(0.5491) (-1.4845) (-0.5829) (-0.5340)
"fKL -0.0081 0.0165 0.0116 0.0201
(-0.1655) (0.7950) (0.6461) (1.1079)
"fKM -0.0162 0.0114 -0.0027 -0.0118
"fu 0.1386 0.1433 0.1690 0.1651

(2.2510) (4.9568) (6.3001) (6.2785)
"fLM -0.1305 -0.1598 -0.1806 -0.1852
"f MM 0.1467 0.1484 0.1833 0.1970
"fKY -0.0086 -0.0004

(-0.2321) (-0.0179)
-
"fLY -0.1015 -0.2720
(-1.5978) (-6.0244)
"fMY 0.1101 0.2724
"fyy 0.4953 -0.1266 0.0198

(1.1569) (-1.0533) (0.1780)
"fll 0.0012
(1.4198)
"f Yt -0.0224
(-1.3477)
"f Kt -0.0018
(-0.5792)
"fLt -0.0073
(-1.7918)
"f Mt 0.0091
Dtnn 2.52£-11 6.21E-l1 2.87E-1O 3.02E-I0 1.57E-09

Note: 1. Numbers in parenthesis are I-statistics refering to parameter estimates above. I-statistics do
2. Dtnn is the detenninant of the residual covaraince matrix.
A5.5. - 6
Table A5.5.3.4 13SLS(AR) parameter estimates for Electrical Appliances and Household Goods
(ISle 3883)
rarameter Model 1 Model 2 Model 3 Mode14 ModelS
ao 16.6000 16.2592 16.2267 16.4578
(298.552) (244.292) (237.981) (302.244)
aK 0.1021 0.0691 0.0751 0.0804
(7.3573) (3.9853) (4.5303) 07.8105)
aL 0.4808 0.3081 0.2957 0.3526
03.3939) (6.0150) (5.4667) (18.7296)
aM 0.4171 0.6228 0.6292 0.5670
ay 0.3279 0.8780 0.9898 0.6645

(2.6837) 01.4151) (33.3289) (8.6257)
at
"(KK -0.0108 0.0371 0.0243

(-0.5531) 0.1012) (0.7532)
"(KL 0.0380 0.0395 0.0521
(1.5366) (1.2900) 0.8078)
"(KM -0.0272 -0.0766 -0.0764
"(u 0.1408 0.1101 0.0998

(4.1079) (2.2649) (2.1776)
"(LM -0.1788 -0.1496 -0.1519
"(MM 0.2060 0.2262 0.2283
"(KY -0.0308
(-1.2218)
"( LY -0.2253
(-4.4711)
"(MY 0.2561
"(yy 0.2929 0.2079

(1.7089) (1.3825)
"(u
"(Ye
"(Kt
"(Lt
"(Mt
Dtnn 2.44E-l1 5.19E-11 5.5lE-ll

Note: 1. Numbers in parenthesis are t-statistics refering to parameter estimates above. t-statistics do
2. Dtnn is the determinant of the residual covaraince matrix.
A5.5. - 7
5.5.4 Monotonicity and Concavity Test Results
Table AS.5.4.l Fitted cost shares and Own AES's for IZEF estimates Modell (ISle 3833)
1972 0.0944 0.4901 0.4155 -7.0010 -0.4501 -0.5500
1973 0.0886 0.4685 0.4429 -7.3418 -0.4885 -0.5039
1974 0.0856 0.4419 0.4726 -7.5326 -0.5369 -0.4538
1975 0.0841 0.4277 0.4882 -7.6247 -0.5628 -0.4279
1976 0.0869 0.3999 0.5131 -7.4471 -0.6138 -0.3871
1977 0.0906 0.4141 0.4953 -7.2228 -0.5879 -0.4162
1978 0.0864 0.4212 0.4924 -7.4816 -0.5749 -0.4208
1979 0.0797 0.3923 0.5281 -7.9116 -0.6278 -0.3633
1980 0.0762 0.3491 0.5747 -8.1460 -0.7010 -0.2922
1981 0.0770 0.3573 0.5657 -8.0918 -0.6880 -0.3055
1982 0.0776 0.3488 0.5735 -8.0477 -0.7014 -0.2940
1983 0.0776 0.3318 0.5906 -8.0535 -0.7258 -0.2692
1984 0.0807 0.3244 0.5948 -7.8435 -0.7351 -0.2631
1985 0.0800 0.3252 0.5948 -7.8907 -0.7342 -0.2632
1986 0.0773 0.3023 0.6205 -8.0728 -0.7563 -0.2275
1987 0.0732 0.2987 0.6281 -8.3508 -0.7585 -0.2172
1988 0.0712 0.2786 0.6502 -8.4891 -0.7625 -0.1882
1989 0.0711 0.2542 0.6746 -8.4926 -0.7395 -0.1573
1990 0.0693 0.2234 0.7073 -8.6222 -0.6350 -0.1182
Table A5.5.4.2. Proper AESs and Determinants of matrixs of AESs IZEF Estimates, Model 1
(ISle 3833
1972 0.8163 0.6278 0.3454 2.4845 3.4568 0.1282 U8E-15
1973 0.7953 0.6281 0.3575 2.9541 3.3047 0.1183 1.36E-16
1974 0.7752 0.6389 0.3616 3.4431 3.0101 0.1128 7.55E-16
1975 0.7637 0.6444 0.3616 3.7082 2.8471 0.1101 5.77E-16
1976 0.7555 0.6727 0.3505 4.0006 2.4304 0.1148 9.74E-16
1977 0.7735 0.6746 0.3501 3.6478 2.5508 0.1221 1.52E-15
1978 0.7663 0.6567 0.3573 3.7138 2.7171 0.1142 1.01E-15
1979 0.7281 0.6530 0.3565 4.4366 2.4480 0.1010 1.00E-15
1980 0.6804 0.6665 0.3356 5.2470 1.9363 0.0922 1.31E-15
1981 0.6910 0.6648 0.3405 5.0901 2.0303 0.0943 -2.01E-16
1982 0.6862 0.6721 0.3337 5.1737 1.9141 0.0948 1.63E-16
1983 0.6697 0.6813 0.3198 . 5.3966 1.7036 0.0931 1.64E-15
1984 0.6754 0.6959 0.3093 5.3096 1.5796 0.0978 1.01E-15
1985 0.6733 0.6932 0.3109 5.3398 1.5965 0.0966 1.12E-15
1986 0.6361 0.6955 0.2892 5.7011 1.3531 0.0884 U6E-15
1987 0.6112 0.6824 0.2896 5.9606 1.3485 0.0809 1.99E-15
1988 0.5715 0.6846 0.2642 6.1462 1.1288 0.0737 1.87E-15
1989 0.5300 0.6958 0.2228 5.9994 0.8519 0.0667 6.74E-16
1990 0.4507 0.7020 0.1564 5.2723 0.5259 0.0506 1.38E-16
A5.5. - 8
Table A5.5.4.3 Fitted cost shares and Own AESs for IZEF estimates Model 2 (ISle 3833)
1972 0.0908 0.5012 0.4080 -12.7423 -0.3981 -0.5613
1973 0.0908 0.4550 0.4542 -12.7397 -0.4733 -0.4837
1974 0.0879 0.4039 0.5081 -13.2813 -0.5564 -0.3944
1975 0.0859 0.3843 0.5298 -13.6883 -0.5865 -0.3599
1976 0.0809 0.3633 0.5558 -14.7878 -0.6161 -0.3198
1977 0.0823 0.4370 0.4807 -14.4642 -0.5029 -0.4394
1978 0.0853 0.4525 0.4622 -13.8218 -0.4773 -0.4703
1979 0.0855 0.3933 0.5212 -13.7722 -0.5728 -0.3735
1980 0.0818 0.3167 0.6015 -14.5798 -0.6620 -0.2532
1981 0.0792 0.3330 0.5878 -15.2047 -0.6503 -0.2726
1982 0.0779 0.3406 0.5814 -15.5419 -0.6429 -0.2818
1983 0.0764 0.3324 0.5913 -15.9514 -0.6509 -0.2676
1984 0.0735 0.3497 0.5768 -16.7631 -0.6330 -0.2886
1985 0.0728 0.3601 0.5671 -16.9875 -0.6203 -0.3028
1986 0.0725 0.3345 0.5930 -17.0594 -0.6489 -0.2652
1987 0.0728 0.3214 0.6058 -16.9792 -0.6593 -0.2472
1988 0.0705 0.2898 0.6397 -17.6979 -0.6646 -0.2013
1989 0.0695 0.2777 0.6528 -18.0558 -0.6559 -0.1843
1990 0.0676 0.2389 0.6935 -18.7301 -0.5577 -0.1340
(ISle 3833)
Year Proper Proper Proper Dlnn of Dtrm of Dtrm of Dtrm of
1972 1.2263 1.3293 0.2161 3.5686 5.3851 0.1767 -6.24E-16
1973 1.2493 1.2958 0.2243 4.4693 4.4835 0.1786 -8.84E-16
1974 1.2900 1.2730 0.2190 5.7251 3.6172 0.1715 -5.02E-17
1975 1.3120 1.2680 0.2126 6.3070 3.3178 0.1658 1.27E-16
1976 1.3503 1.2712 0.2060 7.2878 3.1133 0.1546 -1.32E-15
1977 1.2863 1.3082 0.2368 5.6196 4.6439 0.1649 -4.05E-16
1978 1.2669 1.3096 0.2336 4.9922 4.7856 0.1699 1.19E-16
1979 1.3062 1.2738 0.2180 6.1828 3.5218 0.1664 1.53E-16
1980 1.3975 1.2479 0.1584 7.6996 2.1343 0.1425 -5.48E-16
1981 1.3904 1.2620 0.1810 7.9546 2.5529 0.1445 -1.15E-16
1982 1.3881 1.2693 0.1907 8.0657 2.7683 0.1448 -1.55E-15
1983 1.4058 1.2702 0.1843 8.4061 2.6560 0.1402 -9.07E-16
1984 1.4006 1.2877 0.2053 8.6491 3.1800 0.1406 -1.67E-15
1985 1.3930 1.2956 0.2150 8.5963 3.4651 0.1416 6.39E-16
"1986 1.4245 1.2836 0.1918 9.0415 2.8773 0.1353 6.05E-16
1987 1.4401 1.2766 0.1767 9.1198 2.5676 0.1317 -5.69E-16
1988 1.5039 1.2704 0.1352 9.4999 1.9491 0.1155 4.04E-17
1989 1.5339 1.2690 0.1158 9.4905 1.7174 0.1075 -9.29E-16
1990 1.6381 1.2604 0.0325 7.7619 0.9211 0.0737 8.20E-17
A5.5. - 9
Table AS.S.4.S Fitted cost shares and Own AESs for IZEF estimates Model 3 (lSIC 3833)
Year Filled cost Filled cost Filled cost OwnAES Own AES OwnAES
1972 0.0827 0.3713 0.5460 -12.3932 -0.4754 -0.1942
1973 0.0840 0.3768 0.5392 -12.1706 -0.4713 -0.2011
1974 0.0839 0.3809 0.5352 -12.1776 -0.4681 -0.2052
1975 0.0837 0.3853 0.5309 -12.2104 -0.4644 -0.2095
1976 0.0804 0.3668 0.5528 -12.8117 -0.4783 -0.1872
1977 0.0785 0.3518 0.5697 -13.1808 -0.4859 -0.1699
1978 0.0806 0.3648 0.5546 -12.7797 -0.4796 -0.1854
1979 0.0817 0.3644 0.5539 -12.5713 -0.4798 -0.1861
1980 0.0806 0.3546 0.5648 -12.7686 -0.4848 -0.1749
1981 0.0812 0.3797 0.5391 -12.6582 -0.4691 -0.2012
1982 0.0800 0.3734 0.5466 -12.8900 -0.4739 -0.1935
1983 0.0785 0.3617 0.5598 -13.1794 -0.4814 -0.1800
1984 0.0762 0.3528 0.5710 -13.6501 -0.4855 -0.1686
1985 0.0766 0.3631 0.5603 -13.5808 -0.4805 -0.1795
1986 0.0758 0.3499 0.5743 -13.7344 -0.4866 -0.1652
1987 0.0776 0.3677 0.5547 -13.3596 -0.4778 -0.1853
1988 0.0773 0.3693 0.5534 -13.4250 -0.4768 -0.1866
1989 0.0750 0.3436 0.5814 -13.9219 -0.4882 -0.1579
1990 0.0735 0.3270 0.5995 -14.2604 -0.4879 -0.1394
Table AS.5.4.6. Proper AESs and Determinants of matrixs of AESs IZEF Estimates, Model 3
(lSIC 3833)
1972 1.5048 0.8538 0.0953 3.6270 1.6773 0.0832 5.31E-16
1973 1.4898 0.8542 0.0974 3.5168 1.7176 0.0853 4.63E-16
1974 1.4848 0.8531 0.1003 3.4958 1.7707 0.0860 8.24E-17
1975 1.4803 0.8516 0.1035 3.4789 1.8324 0.0865 1.75E-16
1976 1.5255 0.8515 0.0955 3.8013 1.6739 0.0804 -1.79E-16
1977 1.5611 0.8524 0.0849 3.9673 1.5134 0.0754 1.08E-15
1978 1.5273 0.8523 0.0935 3.7963 1.6423 0.0801 9.76E-17
1979 1.5205 0.8542 0.0914 3.7196 1.6103 0.0810 1.76E-16
1980 1.5421 0.8551 0.0842 3.8124 1.5026 0.0777 -1.65E-16
1981 1.5025 0.8493 0.1040 3.6802 1.8259 0.0836 9.40E-16
1982 1.5189 0.8491 0.1014 3.8015 1.7738 0.0814 1.06E-16
1983 1.5458 0.8499 0.0942 3.9546 1.6504 0.0778 -3.18E-16
1984 1.5764 0.8484 0.0895 4.1426 1.5810 0.0738 2.08E-16
1985 1.5575 0.8461 0.0986 4.1002 1.7224 0.0765 9.32E-16
1986 1.5841 0.8484 0.0873 4.1731 1.5493 0.0728 5.84E-16
1987 1.5430 0.8467 0.1008 4.0018 1.7590 0.0784 -1.97E-16
1988 1.5430 0.8457 0.1026 4.0196 1.7894 0.0784 -7.93E-16
1989 1.6016 0.8486 0.0820 4.2315 1.4782 0.0704 1.01E-16
1990 1.6452 0.8502 0.0645 4.2512 1.2649 0.0638 4.32E-16
A5.5. - 10
Table A5.5.4.7 Fitted cost shares and Own AES's for IZEF estimates Model 4 (ISIC 3833)
Year Fitted cost Fitted cost Fiued cost Own AES Own AES Own AES
1972 0.0830 0.3700 0.5470 -12.3111 -0.4989 -0.1674
1973 0.0845 0.3750 0.5405 -12.0579 -0.4947 -0.1734
1974 0.0845 0.3792 0.5362 -12.0451 -0.4910 -0.1773
1975 0.0845 0.3839 0.5316 -12.0564 -0.4866 -0.1815
1976 0.0805 0.3666 0.5529 -12.7609 -0.5015 -0.1620
1977 0.0781 0.3522 0.5697 -13.2243 . -0.5108 -0.1462
1978 0.0806 0.3644 0.5549 -12.7391 -0.5031 -0.1600
1979 0.0818 0.3634 0.5548 -12.5280 -0.5038 -0.1602
1980 0.0804 0.3539 0.5657 -12.7833 -0.5099 -0.1499
1981 0.0817 0.3795 0.5388 -12.5361 -0.4908 -0.1750
1982 0.0803 0.3736 0.5461 -12.8047 -0.4959 -0.1682
1983 0.0784 0.3623 0.5593 -13.1661 -0.5046 -0.1560
1984 0.0758 0.3544 0.5698 -13.7047 -0.5096 -0.1461
1985 0.0764 0.3649 0.5586 -13.5694 -0.5028 -0.1566
1986 0.0753 0.3516 0.5730 -13.8100 -0.5110 -0.1430
1987 0.0777 0.3691 0.5533 -13.3154 -0.4997 -0.1616
1988 0.0774 0.3708 0.5518 -13.3731 -0.4983 -0.1630
1989 0.0743 0.3456 0.5801 -14.0454 -0.5137 -0.1363
1990 0.0722 0.3293 0.'5985 -14.5133 -0.5170 -0.1190
(ISIC 3833)
Year Proper Proper Proper Dtnn of Dtnn of Dtnn of Dtnnof
1972 1.6773 0.7335 0.0830 3.3287 1.5230 0.0766 7.39E-16
1973 1.6566 0.7350 0.0844 3.2212 1.5505 0.0787 6.06E-16
1974 1.6488 0.7331 0.0873 3.1959 1.5984 0.0794 1.09E-15
1975 1.6413 0.7306 0.0906 . 3.1731 1.6547 0.0801 7.60E-16
1976 1.7046 0.7282 0.0843 3.4943 1.5364 0.0741 l.15E-15
1977 1.7559 0.7281 0.0749 3.6711 1.4031 0.0691 1.25E-15
1978 1.7078 0.7296 0.0823 3.4929 1.5064 0.0738 9.62E-16
1979 1.6998 0.7333 0.0795 3.4228 1.4691 0.0744 1.05E-15
1980 1.7311 0.7340 0.0729 3.5211 1.3776 0.0711 1.25E-15
1981 1.6706 0.7253 0.0922 3.3618 1.6674 0.0774 l.10E-15
1982 1.6934 0.7240 0.0903 3.4828 1.6300 0.0753 3.90E-l0
1983 1.7320 0.7241 0.0840 3.6439 1.5293 0.0716 9.3IE-16
1984 1.7742 0.7199 0.0809 3.8356 1.4836 0.0679 3.34E-16
1985 1.7456 0.7167 0.0895 3.7754 1.6109 0.0707 6.15E-16
1986 1.7853 0.7197 0.0789 3.8699 1.4573 0.0669 l.12E-15
1987 1.7256 0.7185 0.0910 3.6755 1.6354 0.0725 1.63E-15
1988 1.7248 0.7167 0.0929 3.6884 1.6656 0.0726 1.06E-15
1989 1.8104 0.7191 0.0744 3.9377 1.3980 0.0645 6.23E-16
1990 1.8745 0.7201 0.0583 3.9895 1.2081 0.0581 6.68E-16
A5.5. - 11
Table A5.5.4.9 Fitted cost shares and Own AESs for IZEF AR) estimates Modell (lSIC 3833)
1972 0.1508 0.3021 0.5471 2.3148 0.9058 -0.1389
1973 0.1166 0.3174 0.5660 5.7121 0.7629 -0.1231
1974 0.1055 0.3105 0.5840 7.7644 0.8232 -0.1077
1975 0.0993 0.3133 0.5874 9.2492 0.7984 -0.1048
1976 0.1317 0.2674 0.6009 3.8245 1.3649 -0.0931
1977 0.1463 0.2794 0.5744 2.6105 1.1812 -0.1160
1978 0.1073 0.3303 0.5624 7.3682 0.6628 -0.1262
1979 0.0700 0.3385 0.5915 23.5963 0.6075 -0.1012
1980 0.0643 0.3063 0.6294 29.1427 0.8633 -0.0683
1981 0.0638 0.3344 0.6018 29.7345 0.6341 -0.0923
1982 0.0694 0.3316 0.5990 24.1062 0.6538 -0.0947
1983 0.0740 0.3189 0.6071 20.4736 0.7505 -0.0877
1984 0.0977 0.3034 0.5989 9.6878 0.8926 -0.0949
1985 0.0889 0.3254 0.5857 12.6124 0.6988 -0.1063
1986 0.0769 0.3180 0.6051 18.5553 0.7580 -0.0894
1987 0.0458 0.3536 0.6006 65.3062 0.5194 -0.0934
1988 0.0388 0.3498 0.6114 95.2199 0.5397 -0.0840
1989 0.0469 0.3228 0.6302 61.7558 0.7186 -0.0676
1990 0.0458 0.2984 0.6558 65.3716 0.9452 -0.0454
Table A5.5.4.l0. Proper AESs and Determinants of matrixs of AESs IZEF (AR) Estimates,
Model 1 (lSIC 3833)
1972 -1.9414 0.4340 0.0350 -1.6723 -0.5099 -0.1271 7.35E-17
1973 -2.6204 0.2925 0.1121 -2.5088 -0.7889 -0.1065 -1.62E-16 .
1974 -3.0914 . 0.2418 0.1205 -3.1651 -0.8950 -0.1032 3.70E-16
1975 -3.3064 0.1995 0.1333 -3.5478 -1.0093 -0.1014 6.23E-16
1976 -2.8047 0.4099 0.0074 -2.6466 -0.5242 -0.1272 -1.08E-16
1977 -2.2798 0.4441 0.0060 -2.1138 -0.5000 -0.1370 1.52E-16
1978 -2.7799 0.2263 0.1413 -2.8441 -0.9810 -0.1036 3.12E-16
1979 -4.6561 -0.1279 0.2034 -7.3457 -2.4050 -0.1028 5.81E-16
1980 -5.8022 -0.1538 0.1727 -8.5054 -2.0148 -0.0888 6.16E-16
1981 -5.2818 -0.2166 0.2075 -9.0422 -2.7923 -0.1016 1.89E-15
1982 -4.8234 -0.1233 0.1969 -7.5034 -2.2985 -0.1007 5.51E-16
1983 -4.6777 -0.0392 0.1761 -6.5148 -1.7971 -0.0968 9.08E-16
1984 -3.5196 0.2021 0.1221 -3.7400 -0.9598 -0.0996 -2.61E-17
1985 -3.6319 0.1032 0.1631 -4.3771 -1.3510 -0.1009 4.37E-16
1986 -4.4806 -0.0036 0.1711 -6.0099 -1.6592 -0.0970 1.49E-15
1987 -7.2743 -0.6976 0.2489 -18.9956 -6.5836 -0.1105 4.17E-15
1988 -8.8705 -0.9684 0.2542 -27.2938 -8.9367 -0.1100 3.13E-15
1989 -7.8458 -0.5791 0.2161 -17.1800 -4.5076 -0.0952 3.97E-15
1990 -8.8096 -0.5554 0.1849 -15.8182 -3.2741 -0.0771 2.60E-15
A5.5. - 12
Table AS.S.4.1l Fitted cost shares and Own AESs for IZEF (AR) estimates Model 2 (lSIC 3833)
Year Fitted cost Fitted cost Fitted cost OwnAES Own AES OwnAES
1972 0.1061 0.3988 0.4951 -7.1903 -0.8 liS -0.0366
1973 0.0977 0.3749 0.5274 -7.7789 -0.8798 -0.0297
1974 0.0892 0.3540 0.5568 -8.4644 -0.9414 -0.0186
1975 0.0868 0.3488 0.5643 -8.6737 -0.9570 -0.0153
1976 0.0810 0.3423 0.5767 -9.2266 -0.9766 -0.0094
1977 0.0927 0.3756 0.5317 -8.1727 -0.8777 -0.0283
1978 0.0971 0.3818 0.5211 -7.8224 -0.8598 -0.0315
1979 0.0849 0.3485 0.5666 -8.8512 -0.9580 -0.0142
1980 0.0688 0.3111 0.6201 -10.5927 -1.0706 0.0141
1981 0.0779 0.3331 0.5890 -9.5428 -1.0045 -0.0031
1982 0.0788 0.3382 0.5829 -9.4491 -0.9889 -0.0062
1983 0.0755 0.3335 0.5910 -9.8030 -1.0033 -0.0021
1984 0.0785 0.3465 0.5750 -9.4812 -0.9641 -0.0102
1985 0.0829 0.3569 0.5603 -9.0417 -0.9330 -0.0171
1986 0.0753 0.3393 0.5854 -9.8331 -0.9856 -0.0050
1987 0.0759 0.3368 0.5873 -9.7676 -0.9931 -0.0040
1988 0.0706 0.3246 0.6048 -10.3785 -1.0300 0.0054
1989 0.0637 0.3126 0.6236 -11.2720 -1.0660 0.0162
1990 0.0534 0.2904 0.6562 -12.8499 -1.1309 0.0358
Table AS.S.4.12. Proper AESs and Determinants of matrixs of AESs IZEF (AR) Estimates
Model 2 (lSIC 3833)
Year Proper Proper Proper Dtnn of Dtnn of Dtnn of Dtrm of
1972 2.3755 -0.3725 0.1446 0.1919 0.1245 0.0088 -2.17E-19
1973 2.5890 -0.3992 0.1458 0.1412 0.0713 0.0048 -4.73E-18
1974 2.8430 -0.4519 0.1432 -0.1148 -0.0464 -0.0029 1.64E-17
1975 2.9216 -0.4715 0.1420 -0.2354 -0.0899 -0.0056 1.84E-17
1976 3.0988 -0.5435 0.1444 -0.5924 -0.2088 -0.01l7 -5.76E-18
1977 2.6722 -0.4632 0.1543 0.0329 0.0164 0.0010 3.28E-18
1978 2.5695 -0.4246 0.1510 0.1233 0.0662 0.0043 6.67E-18
1979 2.9673 -0.4990 0.1446 -0.3257 -0.1232 -0.0073 1.00E-17
1980 3.7175 -0.6890 0.1244 -2.4787 -0.6239 -0.0306 -3.16E-17
1981 3.2420 -0.5707 0.1391 -0.9255 -0.2960 -0.0162 -1.02E-16
1982 3.1828 -0.5691 0.1434 -0.7865 -0.2648 -0.0144 3.47E-17
1983 3.3106 -0.6152 0.1430 -1.1252 -0.3582 -0.0184 5.90E-17
1984 3.1393 -0.5969 0.1522 -0.7149 -0.2596 -0.0133 -1.49E-16
1985 2.9677 -0.5527 0.1552 -0.3716 -0.1508 -0.0081 -4.11E-18
1986 3.2791 -0.6364 0.1497 -1.0603 -0.3562 -0.0175 -1.73E-16
1987 3.2779 -0.6184 0.1462 -1.0444 -0.3435 -0.0174 -1.51E-16
1988 3.5402 -0.6892 0.1397 -1.8435 -0.5312 -0.0251 -2.00E-16
1989 3.9220 -0.8147 0.1338 -3.3663 -0.8460 -0.0351 3.39E-19
1990 4.7525 -1.0570 0.1136 -8.0541 -1.5770 -0.0534 3.43E-16
AS.S. - 13
Table AS.S.4.13 Fitted cost shares and Own AESs for IZEF (AR) estimates Model 3 (lSIC 3833)
1972 0.0704 0.3064 0.6232 -6.2839 -1.1112 -0.0230
1973 0.0695 0.3076 0.6229 -6.2873 -1.1074 -0.0232
1974 0.0718 0.3115 0.6167 -6.2745 -1.0951 -0.0275
1975 0.0746 0.3162 0.6092 -6.2411 -1.0803 -0.0328
1976 0.0748 0.3092 0.6160 -6.2386 -1.1023 -0.0281
1977 0.0726 0.3012 0.6262 -6.2664 -1.1274 -0.0209
1978 0.0732 0.3069 0.6199 -6.2593 -1.1098 -0.0253
1979 0.0697 0.3031 0.6272 -6.2867 -1.1216 -0.0201
1980 0.0678 0.2972 0.6351 -6.2875 -1.1398 -0.0145
1981 0.0791 0.3187 0.6022 -6.1606 -1.0725 -0.0376
1982 0.0794 0.3166 0.6040 -6.1531 -1.0791 -0.0364
1983 0.0777 0.3103 0.6120 -6.1886 -1.0990 -0.0308
1984 0.0798 0.3091 0.6111 -6.1445 -1.1028 -0.0315
1985 0.0843 0.3177 0.5980 -6.0368 -1.0757 -0.0405
1986 0.0795 0.3076 0.6129 -6.1516 -1.1074 -0.0302
1987 0.0835 0.3187 0.5978 -6.0566 -1.0726 -0.0406
1988 0.0852 0.3211 0.5937 -6.0110 -1.0650 -0.0435
1989 0.0788 0.3045 0.6168 -6.1669 -1.1173 -0.0275
1990 0.0746 0.2937 0:6318 -6.2420 -1.1506 -0.0169
Table AS.S.4.14. Proper AESs and Determinants of matrixs of AESs IZEF Estimates Model 3
(ISIC 3833)
1972 2.9332 -0.7323 0.2150 -1.6211 -0.3919 -0.0207 1.67E-16
1973 2.9499 -0.7551 0.2177 -1.7393 -0.4242 -0.0217 7.86E-17
1974 2.8656 -0.7174 0.2198 -1.3399 -0.3418 -0.0181 1.02E-16
1975 2.7671 -0.6718 0.2218 -0.9145 -0.2464 -0.0137 6.86E-17
1976 2.8028 -0.6496 0.2131 -0.9790 -0.2468 -0.0144 I.72E-17
1977 2.9066 -0.6714 0.2052 -1.3838 -0.3202 -0.0186 -5.85E-17
1978 2.8554 -0.6738 0.2120 -1.2067 -0.2957 -0.0168 3.74E-16
1979 2.9730 -0.7375 0.2114 -1.7877 -0.4174 -0.0221 3.34E-16
1980 3.0709 -0.7663 0.2058 -2.2639 -0.4958 -0.0258 5.77E-16
1981 2.6546 -0.5958 0.2190 -0.4396 -0.1231 -0.0076 1.14E-16
1982 2.6581 -0.5842 0.2161 -0.4261 -0.1171 -0.0074 5.96E-17
1983 2.7295 -0.5981 0.2106 -0.6491 -0.1668 -0.0105 -6.83E-18
1984 2.6902 -0.5579 0.2063 -0.4610 -0.1179 -0.0079 3.89E-17
1985 2.5577 -0.5081 0.2110 -0.0484 -0.0137 -0.0010 3.23E-18
1986 2.7051 -0.5598 0.2049 -0.5054 -0.1273 -0.0085 7.85E-17
1987 2.5672 -0.5225 0.2132 -0.0943 -0.0268 -0.0018 2.07E-17
1988 2.5238 -0.5018 0.2136 0.0323 0.0094 0.0007 -9.00E-18
1989 2.7386 -0.5641 0.2017 -0.6096 -0.1486 -0.0099 2.34E-17
1990 2.9044 -0.6133 0.1920 -1.2532 -0.2708 -0.0175 3.21E-16
A5.5. - 14
Table A5.5.4.15 Fitted C()st shares and Own AESs for IZEF (AR) estimates Model 4 HSIC 3833)
Year Fitted cost Fitted cost Fitted cost OwnAES OwnAES Own AES
1972 0.0752 0.2954 0.6294 -8.0185 -1.2473 -0.0125
1973 0.0755 0.2955 0.6290 -8.0000 -1.2470 -0.0128
1974 0.0775 0.2996 0.6228 -7.8716 -1.2315 -0.0170
1975 0.0800 0.3048 0.6152 -7.7157 -1.2122 -0.0223
1976 0.0773 0.3006 0.6221 -7.8863 -1.2279 -0.0175
1977 0.0736 0.2939 0.6325 -8.1191 -1.2529 -0.0104
1978 0.0760 0.2979 0.6261 -7.9678 -1.2379 -0.0148
1979 0.0737 0.2928 0.6335 -8.1119 -1.2570 -0.0097
1980 0.0709 0.2877 0.6414 -8.2877 -1.2763 -0.0041
1981 0.0820 0.3097 0.6082 -7.5951 -1.1935 -0.0270
1982 0.0813 0.3087 0.6100 -7.6411 -1.1973 -0.0258
1983 0.0784 0.3035 0.6182 -7.8205 -1.2170 -0.0202
1984 0.0783 0.3044 0.6173 -7.8220 -1.2137 -0.0208
1985 0.0828 0.3131 0.6041 -7.5499 -1.1808 -0.0298
1986 0.0777 0.3032 0.6191 -7.8623 -1.2179 -0.0196
1987 0.0830 0.3131 0.6039 -7.5367 -1.1808 -0.0299
1988 0.0843 0.3160 0.5997 -7.4559 -1.1702 -0.0327
1989 0.0763 0.3007 0.6230 -7.9515 -1.2274 -0.0169
1990 0.0710 0.2907 0.6382 -8.2824 -1.2648 -0.0064
Table A5.5.4.16. Proper AESs and Determinants of matrixs of AESs IZEF (AR) Estimates
Model 4 (ISle 3833)
1972 3.3589 -0.6184 0.1841 -1.2807 -0.2821 -0.0183 l.34E-16
1973 3.3491 -0.6130 0.1838 -1.2405 -0.2737 -0.0179 7.54E-17
1974 3.2555 -0.5861 0.1871 -0.9049 -0.2094 -0.0140 1.66E-16
1975 3.1480 -0.5554 0.1909 -0.5570 -0.1367 -0.0094 6.78E-17
1976 3.2552 -0.5928 0.1887 -0.9129 -0.2131 -0.0141 7.59E-17
1977 3.4222 -0.6453 0.1839 -1.5392 -0.3323 -0.0208 1.09E-16
1978 3.3144 -0.6097 0.1866 -1.1215 -0.2539 -0.0165 1.78E-16
1979 3.4274 -0.6403 0.1822 -1.5508 -0.3314 -0.0210 8.48E-17
1980 3.5679 -0.6835 0.1778 -2.1519 -0.4328 -0.0263 3.68E-16
1981 3.0627 -0.5355 0.1948 -0.3156 -0.0818 -0.0057 1.38E-17
1982 3.0887 -0.5452 0.1945 -0.3913 -0.1002 -0.0069 2.95E-17
1983 3.2036 -0.5814 0.1913 -0.7456 -0.1797 -0.0120 8.32E-17
1984 3.1978 -0.5840 0.1926 -0.7329 -0.1782 -0.0118 -1.03E-17
1985 3.0220 -0.5322 0.1981 -0.2176 -0.0585 -0.0041 6.46E-17
1986 3.2245 -0.5926 0.1919 -0.8214 -0.1970 -0.0129 8.77E-17
1987 3.0166 -0.5287 0.1978 -0.2010 -0.0540 -0.0038 -7.IOE-18
1988 2.9668 -0.5148 0.1994 -0.0766 -0.0213 -0.0015 3.63E-18
1989 3.2850 -0.6121 0.1903 -1.0315 -0.2403 -0.0155 2.28E-16
1990 3.5376 -0.6898 0.1825 -2.0393 -0.4232 -0.0253 -5.18E-17
A5.5. - 15
Table A5.5.4.17 Fitted cost shares and Own AESs for I3SLS estimates Model 1 nSIC 3833)
1972 0.0940 0.4903 0.4157 -6.8882 -0.4630 -0.5567
1973 0.0883 0.4684 0.4433 -7.2067 -0.5032 -0.5094
1974 0.0855 0.4417 0.4728 -7.3728 -0.5536 -0.4588
1975 0.0841 0.4274 0.4884 -7.4534 -0.5809 -0.4324
1976 0.0870 0.4000 0.5130 -7.2832 -0.6337 -0.3919
1977 0.0902 0.4144 0.4954 -7.1015 -0.6061 -0.4208
1978 0.0857 0.4210 0.4932 -7.3576 -0.5932 -0.4245
1979 0.0791 0.3919 0.5289 -7.7566 -0.6491 -0.3663
1980 0.0759 0.3489 0.5752 -7.9586 -0.7275 -0.2951
1981 0.0769 0.3567 0.5665 -7.8972 -0.7142 -0.3081
1982 0.0774 0.3483 0.5743 -7.8635 -0.7287 -0.2964
1983 0.0772 0.3314 0.5914 -7.8746 -0.7555 -0.2715
1984 0.0803 0.3242 0.5955 -7.6851 -0.7659 -0.2655
1985 0.0796 0.3247 0.5957 -7.7297 -0.7652 -0.2652
1986 0.0767 0.3019 0.6214 -7.9060 -0.7917 -0.2293
1987 0.0727 0.2979 0.6294 -8.1564 -0.7950 -0.2185
1988 0.0708 0.2777 0.6515 -8.2741 -0.8037 -0.1894
1989 0.0706 0.2537 0.6758 -8.2910 -0.7882 -0.1586
1990 0.0687 0.2231 0.7082 -8.4086 -0.6976 -0.1195
Table A5.5.4.18. Proper AESs and Determinants of matrixs of AESs I3SLS Estimates Model 1
(ISle 3833)
1972 0.8243 0.5854 0.3597 2.5099 3.4916 0.1283 2.52E-16
1973 . 0.8042 0.5862 0.3715 2.9794 3.3273 0.1183 -1.72E-16
1974 0.7855 0.5992 0.3751 3.4644 3.0234 0.1132 -7.01E-16
1975 0.7747 0.6057 0.3749 3.7294 2.8563 0.1106 1.95E-18
1976 0.7673 0.6370 0.3640 4.0269 2.4486 0.1159 -8.52E-17
1977 0.7832 0.6374 0.3644 3.6905 2.5818 0.1222 -8.52E-16
1978 0.7756 0.6169 0.3716 3.7631 2.7424 0.1137 -4.51E-16
1979 0.7388 0.6129 0.3705 4.4893 2.4653 0.1005 -9.46E-16
1980 0.6941 0.6288 0.3498 5.3083 1.9536 0.0924 -5.27E-16
1981 0.7045 0.6279 0.3541 5.1441 2.0390 0.0947 -6.12E-16
1982 0.6995 0.6356 0.3476 5.2404 1.9268 0.0952 -7.54E-16
1983 0.6835 0.6453 0.3341 5.4821 1.7216 0.0935 -5.25E-16
1984 0.6888 0.6612 0.3240 5.4115 1.6034 0.0984 -4.64E-16
1985 0.6865 0.6582 0.3253 5.4433 1.6169 0.0971 7.46E-17
1986 0.6502 0.6602 0.3043 5.8364 1.3773 0.0890 1.03E-16
1987 0.6261 0.6460 0.3040 6.0928 1.3649 0.0813 -5.59E-16
1988 0.5883 0.6490 0.2787 6.3042 1.1456 0.0745 -1.03E-16
1989 0.5475 0.6603 0.2387 6.2355 0.8787 0.0680 -4.51E-16
1990 0.4713 0.6670 0.1740 5.6437 0.5599 0.0531 1.81E-16
A5.5. - 16
Table AS.S.4.19 Fitted cost shares and Own AESs for I3SLS estimates Model 2 (lSIC 3833)
1972 0.0876 0.5093 0.4031 -14.0513 -0.4110 -0.5675
1973 0.0902 0.4583 0.4514 -13.5083 -0.4997 -0.4870
1974 0.0894 0.4032 0.5075 -13.6836 -0.5987 -0.3943
1975 0.0881 0.3819 0.5300 -13.9546 -0.6359 -0.3585
1976 0.0816 0.3624 0.5560 -15.4396 -0.6682 -0.3186
1977 0.0788 0.4439 0.4773 -16.1824 -0.5256 -0.4436
1978 0.0826 0.4586 0.4588 -15.2002 -0.4992 -0.4746
1979 0.0858 0.3941 0.5202 -14.4554 -0.6149 -0.3739
1980 0.0842 0.3126 0.6031 -14.8047 . -0.7325 -0.2501
1981 0.0818 0.3286 0.5896 -15.4011 -0.7160 -0.2692
1982 0.0793 0.3380 0.5826 -16.0412 -0.7042 -0.2792
1983 0.0770 0.3307 0.5923 -16.6847 . -0.7136 -0.2653
1984 0.0720 0.3513 0.5766 -18.2674 -0.6853 -0.2879
1985 0.0712 0.3618 0.5670 -18.5359 -0.6692 -0.3021
1986 0.0713 0.3353 0.5933 -18.5007 -0.7077 -0.2639
1987 0.0735 0.3192 0.6073 -17.7842 -0.7264 -0.2442
1988 0.0722 0.2852 0.6425 -18.1880 -0.7446 -0.1969
1989 0.0698 0.2750 0.6551 -19.0417 -0.7415 -0.1806
1990 0.0682 0.2350 0.6967 -19.6434 -0.6605 -0.1296
Table AS.S.4.20. Proper AESs and Determinants of matrixs of AESs I3SLS Estimates Model 2
(lSIC 3833)
1972 1.3698 1.3228 0.2216 3.8989 6.2240 0.1841 6.55E-16
1973 1.3989 1.2799 0.2277 4.7925 4.9403 0.1915 2.67E-16
1974 1.4580 1.2514 0.2189 6.0673 3.8297 0.1882 -1.52E-15
1975 1.4906 1.2443 0.2106 6.6518 3.4542 0.1836 -1.40E-15
1976 1.5578 1.2512 0.2069 7.8905 3.3533 0.1701 1.70E-15
1977 1.4717 1.3031 0.2458 6.3396 5.4813 0.1728 1.25E-15
1978 1.4357 1.3009 0.2405 5.5265 5.5214 0.1790 9.29E-16
1979 1.4883 1.2556 0.2204 6.6732 3.8292 0.1813 -3.34E-16
1980 1.6266 1.2244 0.1525 8.1986 2.2028 0.1599 -2.40E-16
1981 1.6140 1.2365 0.1753 8.4229 2.6171 0.1620 1.83E-15
1982 1.6154 1.2467 0.1886 8.6869 2.9239 0.1610 -1.34E-15
1983 1.6478 1.2499 0.1841 9.1912 2.8643 0.1554 -2.26E-16
1984 1.6522 1.2745 0.2113 9.7898 3.6345 0.1527 -5.92E-16
1985 1.6402 1.2823 0.2210 9.7141 3.9562 0.1534 1.88E-15
1986 1.6897 1.2693 0.1969 10.2383 3.2708 0.1480 5.65E-16
1987 1.7037 1.2555 0.1757 10.0156 2.7669 0.1465 -1.22E-15
1988 1.8007 1.2456 0.1281 10.2993 2.0297 0.1302 -8.71E-16
1989 1.8591 1.2492 0.1131 10.6631 1.8791 0.1211 -1.77E-16
1990 2.0288 1.2397 0.0241 8.8588 1.0079 0.0850 1.05E-15
A5.5. - 17
Table AS.S.4.21 Fitted cost shares and Own AESs for I3SLS estimates Model 3 (lSIC 3833)
1972 0.0825 0.3721 0.5454 -12.4288 -0.4669 -0.2173
1973 0.0836 0.3780 0.5384 -12.2319 -0.4628 -0.2250
1974 0.0835 0.3819 0.5346 -12.2528 -0.4598 -0.2292
1975 0.0832 0.3861 0.5307 -12.3004 -0.4563 -0.2335
1976 0.0804 0.3669 0.5528 -12.8175 -0.4701 -0.2092
1977 0.0788 0.3516 0.5696 -13.1143 -0.4771 -0.1907
1978 0.0806 0.3650 0.5544 -12.7789 -0.4712 -0.2073
1979 0.0817 0.3650 0.5533 -12.5764 -0.4712 -0.2086
1980 0.0809 0.3551 0.5641 -12.7300 -0.4759 -0.1967
1981 0.0809 0.3797 0.5394 -12.7170 -0.4615 -0.2239
1982 0.0799 0.3731 0.5471 -12.9182 -0.4662 -0.2155
1983 0.0787 0.3611 0.5602 -13.1518 -0.4732 -0.2010
1984 0.0766 0.3516 0.5718 -13.5664 -0.4771 -0.1883
1985 0.0767 0.3618 0.5615 -13.5436 -0.4729 -0.1996
1986 0.0763 0.3487 0.5750 -13.6347 -0.4779 -0.1847
1987 0.0777 0.3667 0.5557 -13.3497 -0.4703 -0.2060
1988 0.0773 0.3680 0.5546 -13.4190 -0.4694 -0.2071
1989 0.0756 0.3422 0.5822 -13.7860 -0.4791 -0.1769
1990 0.0745 0.3255 0.6000 -14.0319 -0.4771 -0.1575
Table AS.S.4.22 Proper AESs and Determinants of matrixs of AESs I3SLS Estimates Model 3
(ISIC 3833
1972 1.3779 0.9400 0.1101 3.9041 1.8172 0.0893 -6.50E-16
1973 1.3670 0.9400 0.1125 3.7916 1.8684 0.0915 -9.06E-16
1974 1.3638 0.9395 0.1154 3.7734 1.9254 0.0920 -6.53E-17
1975 1.3610 0.9389 0.1186 3.7606 1.9911 0.0925 -6.27E-16
1976 1.3934 0.9392 0.1094 4.0845 1.7992 0.0864 -1.6IE-16
1977 1.4185 0.9399 0.0981 4.2446 1.6173 0.0813 3.25E-16
1978 1.3944 0.9396 0.1075 4.0773 1.7666 0.0861 -5.17E-16
1979 1.3890 0.9403 0.1058 3.9962 1.7397 0.0871 -1.33E-15
1980 1.4041 0.9408 0.0983 4.0864 1.6191 0.0840 -6.52E-16
1981 1.3775 0.9381 0.1182 3.9707 1.9675 0.0894 -5.34E-16
1982 1.3894 0.9382 0.1151 4.0923 1.9032 0.0872 -4.68E-16
1983 1.4084 0.9387 0.1073 4.2403 1.7619 0.0836 -1.29E-15
1984 1.4306 0.9384 0.1016 4.4257 1.6734 0.0795 -2.96E-16
1985 1.4178 0.9373 0.1110 4.3944 1.8247 0.0821 -3.17E-16
1986 1.4360 0.9385 0.0992 4.4539 1.6375 0.0784 -1.23E-15
1987 1.4073 0.9374 0.1136 4.2972 1.8711 0.0840 -1.64E-15
1988 1.4076 0.9371 0.1152 4.3182 1.9012 0.0840 -3.80E-16
1989 1.4484 0.9386 0.0936 4.5066 1.5576 0.0760 -3.04E-16
1990 1.4785 0.9396 0.0754 4.5091 1.3274 0.0695 -1.23E-15
A5.5. - 18
'{able AS.S.4.23 Fitted cost shares and Own AESs for 13SLS estimates Model 4 (lSIC 3833)
1972 0.0829 0.3701 0.5470 -12.2705 -0.4966 -0.1698
1973 0.0843 0.3752 0.5405 -12.0282 -0.4925 -0.1758
1974 0.0844 0.3794 0.5362 -12.0152 -0.4888 -0.1797
1975 0.0843 0.3840 0.5317 -12.0250 -0.4845 -0.1840
1976 0.0805 0.3666 0.5529 -12.6987 -0.4993 -0.1642
1977 0.0782 0.3521 0.5697 -13.1412 -0.5084 -0.1483
1978 0.0806 0.3644 0.5550 -12.6785· -0.5009 -0.1623
1979 0.0817 0.3635 0.5548 -12.4782 -0.5015 -0.1625
1980 0.0804 0.3539 0.5657 -12.7228 -0.5074 -0.1521
1981 0.0817 0.3794 0.5389 -12.4824 -0.4888 -0.1773
1982 0.0803 0.3735 0.5462 -12.7387 -0.4939 -0.1705
1983 0.0785 0.3622 0.5593 -13.0837 -0.5024 -0.1582
1984 0.0760 0.3541 0.5699 -13.5946 -0.5073 -0.1481
1985 0.0766 0.3646 0.5588 -13.4643 -0.5007 -0.1587
1986 0.0755 0.3514 0.5731 -13.6946 -0.5087 -0.1451
1987 0.0778 0.3688 0.5534 -13.2233 -0.4976 -0.1638
1988 0.0775 0.3705 0.5519 -13.2775 -0.4963 -0.1651
1989 0.0745 0.3453 0.5802 -13.9180 -0.5113 -0.1384
1990 0.0725 0.3290 0.5985 -14.3624 -0.5142 -0.1209
Table AS.S.4.24 Proper AESs and Determinants of matrixs of AESs 13SLS Estimates Model 4
(lSIC 3833)
Year Proper Proper Proper Dtrm of Dtrm of Dtnn of Dtnn of
1972 1.6551 0.7398 0.0852 3.3545 1.5356 0.0770 3.41E-16
1973 1.6354 0.7411 0.0867 3.2489 1.5653 0.0791 2.17E-16
1974 1.6278 0.7392 0.0896 3.2232 1.6131 0.0798 -2.16E-16
1975 1.6207 0.7368 0.0929 3.1995 1.6692 0.0805 1.17E-15
1976 1.6810 0.7350 0.0862 3.5152 1.5451 0.0746 1.66E-16
1977 1.7299 0.7352 0.0767 3.6884 1.4088 0.0695 4.87E-17
1978 1.6841 0.7363 0.0842 3.5145 1.5154 0.0742 -3.09E-16
1979 1.6766 0.7397 0.0816 3.4472 1.4801 0.0748 4.72E-16
1980 1.7065 0.7405 0.0749 3.5441 1.3870 0.0716 7.88E-17
1981 1.6484 0.7320 0.0942 3.3837 1.6774 0.0778 3.93E-16
1982 1.6701 0.7310 0.0922 3.5022 1.6378 0.0757 2.55E-16
1983 1.7069 0.7313 0.0857 3.6602 1.5346 0.0721 2.61E-16
1984 1.7469 0.7276 0.0823 3.8454 1.4847 0.0684 -2.51E-16
1985 1.7194 0.7244 0.0910 3.7857 1.6120 0.0712 -5.51E-16
1986 1.7574 0.7274 0.0803 3.8786 1.4578 0.0674 8.82E-16
1987 1.7004 0.7259 0.0926 3.6889 1.6386 0.0729 -1.91E-16
1988 1.6996 0.7243 0.0944 3.7008 1.6680 0.0730 3.33E-16
1989 1.7813 0.7270 0.0756 3.9437 1.3972 0.0650 6.57E-17
1990 1.8422 0.7282 0.0594 3.9915 1.2061 0.0586 -2.49E-16
A5.5. - 19
Table A5.5.4.25 Fitted cost shares and Own AESs for I3SLS (AR) estimates Model 2 ISIC 3833)
1972 0.1021 0.4808 0.4171 -9.8303 -0.4708 -0.2134
1973 0.0981 0.4377 0.4642 -10.3120 -0.5498 -0.1982
1974 0.0918 0.3934 0.5148 -11.1718 -0.6322 -0.1652
1975 0.0891 0.3777 0.5332 -11.5777 ~0.6606 -0.1509
1976 0.0822 0.3626 0.5552 -12.7583 -0.6870 -0.1329
1977 0.0888 0.4292 0.4820 -11.6360 -0.5656 -0.1880
1978 0.0934 0.4412 0.4654 -10.9497 -0.5432 -0.1976
1979 0.0877 0.3852 0.5271 -11.8141 -0.6471 -0.1557
1980 0.0769 0.3168 0.6063 -13.8228 -0.7536 -0.0890
1981 0.0796 0.3371 0.5834 -13.2769 -0.7274 -0.1089
1982 0.0789 0.3454 0.5757 -13.4149 -0.7149 -0.1155
1983 0.0759 0.3387 0.5854 -14.0399 -0.7252 -0.1071
1984 0.0750 0.3578 0.5672 -14.2425 -0.6950 -0.1228
1985 0.0767 0.3692 0.5541 -13.8775 -0.6756 -0~1338
1986 0.0727 0.3445 0.5829 -14.8038 -0.7165 -0.1093
1987 0.0734 0.3331 0.5935 -14.6359 -0.7331 -0.1001
1988 0.0691 0.306~ 0.6248 -15.7430 -0.7641 -0.0728
1989 0.0646 0.2943 0.6411 -17.0577 -0.7723 -0.0586
1990 0.0580 0.2589 0.6831 -19.4565 -0.7619 -0.0224
Table A5.5.4.26 Proper AESs and Determinants of matrixs of AESs I3SLS (AR) Estimates
Model 2 (ISIC 3833)
1972 1.7741 0.3613 0.1084 1.4806 1.9674 0.0887 5.37E-16
1973 1.8848 0.4029 0.1199 2.1171 1.8820 0.0946 5.16E-16
1974 2.0520 0.4245 0.1171 2.8518 1.6655 0.0907 1.04E-15
1975 2.1287 0.4277 0.1121 3.1173 1.5643 0.0871 4.39E-16
1976 2.2745 0.4042 0.1118 3.5914 1.5319 0.0788 1.57E-15
1977 1.9974 0.3643 0.1358 2.5917 2.0545 0.0879 1.37E-15
1978 1.9225 0.3740 0.1293 2.2519 2.0238 0.0906 6.12E-16
1979 2.1254 0.4113 0.1195 3.1278 1.6701 0.0865 3.35E-16
1980 2.5591 0.4168 0.0691 3.8684 1.0563 0.0623 3.00E-16
1981 2.4170 0.4139 0.0907 3.8162 1.2743 0.0710 8.13E-16
1982 2.3947 0.4010 0.1009 3.8560 1.3884 0.0724 8.69E-16
1983 2.4775 0.3882 0.0981 4.0434 1.3532 0.0681 1.39E-15
1984 2.4152 0.3610 0.1189 4.0660 1.6183 0.0712 7.09E-16
1985 2.3423 0.3599 0.1260 3.8898 1.7267 0.0745 1.35E-15
1986 2.5179 0.3579 0.1094 4.2663 1.4899 0.0663 1.50E-15
1987 2.5548 0.3754 0.0956 4.2027 1.3239 0.0642 1.03E-15
1988 2.7970 0.3696 0.0653 4.2063 1.0102 0.0514 1.25E-15
1989 2.9980 0.3436 0.0522 4.1854 0.8818 0.0425 1.26E-15
1990 3.5313 0.3134 -0.0110 2.3533 0.3379 0.0170 7.36E-16
A5.5. - 20
Table A5.5.4.27 Fitted cost shares and Own AESs for 13SLS (AR) estimates Model 3 ISIC 3833)
1972 0.0691 0.3081 0.6228 -5.7018 -1.0858 -0.0225
1973 0.0679 0.3096 0.6225 -5.6811 -1.0815 -0.0227
1974 0.0702 0.3134 0.6164 -5.7170 -1.0699 -0.0270
1975 0.0732 0.3180 0.6088 -5.7373 -1.0559 -0.0323
1976 0.0741 0.3105 0.6154 -5.7385 -1.0787 -0.0277
1977 0.0723 0.3021 0.6256 -5.7337 -1.1037 -0.0205
1978 0.0725 0.3081 0.6194 -5.7347 -1.0858 -0.0249
1979 0.0686 0.3046 0.6268 -5.6942 -1.0963 -0.0197
1980 0.0668 0.2986 0.6346 -5.6564 -1.1142 -0.0141
1981 0.0783 0.3200 0.6017 -5.7199 -1.0499 -0.0372
1982 0.0789 0.3177 0.6034 -5.7142 -1.0569 -0.0360
1983 0.0775 0.3111 0.6114 -5.7264 -1.0767 -0.0305
1984 0.0802 0.3095 0.6104 -5.7014 -1.0818 -0.0312
1985 0.0846 0.3181 0.5973 -5.6360 -1.0557 -0.0402
1986 0.0799 0.3080 0.6121 -5.7042 -1.0863 -0.0300
1987 0.0836 0.3192 0.5971 -5.6530 -1.0522 -0.0403
1988 0.0855 0.3215 0.5930 -5.6213 -1.0451 -0.0431
1989 0.0793 0.3047 0.6160 -5.7102 -1.0962 -0.0273
1990 0.0753 0.2937 0.6310 -5.7370 -1.1285 -0.0167
.Table A5.5.4.28 Proper AESs and Determinants of matrixs of AESs I3SLS (AR) Estimates
Model 3 (ISIC 3833)
Year Proper Proper Proper Dtrrn of Dtrrn of Dtrm of Dtrm of
1972 2.8554 -0.7799 0.2204 -1.9618 -0.4801 -0.0241 8.57E-18
1973 2.8785 -0.8115 0.2237 -2.1420 -0.5297 -0.0255 -4.47E-17
1974 2.7947 -0.7696 0.2256 -1.6939 -0.4379 -0.0220 -4.36E-17
1975 2.6965 -0.7186 0.2273 -1.2132 -0.3311 -0.0175 -1.81E-16
1976 2.7170 -0.6797 0.2170 -1.1917 -0.3033 -0.0173 1.82E-16
1977 2.8092 -0.6943 0.2085 -1.5633 -0.3646 -0.0209 -1.63E-16
1978 2.7689 -0.7065 0.2162 -1.4403 -0.3564 -0.0197 8.88E-17
1979 2.8895 -0.7807 0.2164 -2.1063 -0.4975 -0.0253 1.86E-16
1980 2.9797 -0.8064 0.2105 -2.5764 -0.5704 -0.0286 1.53E-16
1981 2.5762 -0.6254 0.2230 -0.6313 -0.1785 -0.0107 4.76E-17
1982 2.5751 -0.6081 0.2195 -0.5917 -0.1640 -0.0101 1.80E-16
1983 2.6385 -0.6169 0.2135 -0.7956 -0.2060 -0.0128 1.32E-16
1984 2.5925 -0.5657 0.2080 -0.5536 -0.1423 -0.0095 -5.I1E-17
1985 2.4673 -0.5152 0.2125 -0.1376 -0.0390 -0.0028 1.02E-17
1986 2.6052 -0.5661 0.2064 -0.5909 -0.1496 -0.0101 5.95E-17
1987 2.4796 -0.5340 0.2152 -0.2009 -0.0574 -0.0039 4.37E-17
1988 2.4372 -0.5113 0.2154 -0.0650 -0.0191 -0.0014 7.64E-18
1989 2.6340 -0.5672 0.2028 -0.6787 -0.1660 -0.0113 2.16E-17
1990 2.7852 -0.6113 0.1927 -1.2830 -0.2779 -0.0183 1.64E-16
A5.5. - 21
Table A5.5.4.29 Fitted cost shares and Own AESs for I3SLS (AR) estimates Model 4 (ISle 3833)
1972 0.0751 0.2957 0.6292 -8.0071 -1.2404 -0.0126
1973 0.0754 0.2958 0.6288 -7.9899 -1.2400 -0.0129
1974 0.0774 0.3000 0.6226 -7.8625 -1.2246 -0.0172
1975 0.0799 0.3051 0.6150 -7.7075 -1.2055 -0.0224
1976 0.0772 0.3008 0.6220 -7.8739 -1.2215 -0.0176
1977 0.0736 0.2941 0.6324 -8.1028 -1.2465 -0.0104
1978 0.0759 0.2981 0.6259 -7.9548 -1.2315 -0.0149
1979 0.0736 0.2931 0.6333 -8.0987 -1.2501 -0.0098
1980 0.0709 0.2879 0.6412 -8.2718 -1.2694 -0.0043
1981 0.0819 0.3100 0.6081 -7.5855 -1.1874 -0.0271
1982 0.0812 0.3089 0.6099 -7.6301 -1.1913 -0.0259
1983 0.0783 0.3036 0.6181 -7.8069 -1.2110 -0.0203
1984 0.0783 0.3044 0.6172 -7.8062 -1.2079 -0.0209
1985 0.0827 0.3132 0.6040 -7.5365 -1.1753 -0.0298
1986 0.0777 0.3033 0.6190 -7.8458 -1.2122 -0.0197
1987 0.0829 0.3133 0.6038 -7.5243 -1.1752 -0.0300
1988 0.0843 0.3161 0.5996 -7.4438 -1.1648 -0.0328
1989 0.0763 0.3007 0.6230 -7.9335 -1.2217 -0.0169
1990 0.0711 0.2907 0.6382 -8.2599 -1.2589 -0.0064
Table A5.5.4.30 Proper AESs and Determinants of matrixs of AESs I3SLS (AR) Estimates
Model 4 (ISle 3833)
Year Proper Proper Proper Dtnnof Dtnn of Dtnn of Dtnn of
1972 3.3461 -0.6168 0.1836 -1.2641 -0.2792 -0.0180 -1.18E-17
1973 3.3367 -0.6119 0.1834 -1.2257 -0.2713 -0.0176 -2.29E-16
1974 3.2436 -0.5850 0.1867 -0.8922 -0.2071 -0.0138 3.05E-17
1975 3.1366 -0.5543 0.1904 -0.5464 -0.1345 -0.0092 2.20E-18
1976 3.2426 -0.5904 0.1881 -0.8965 -0.2097 -0.0138 -5.30E-17
1977 3.4082 -0.6421 0.1831 -1.5155 -0.3277 -0.0205 l.33E-16
1978 3.3015 -0.6073 0.1860 -1.1037 -0.2503 -0.0162 -3.64E-17
1979 3.4140 -0.6383 0.1816 -1.5312 -0.3280 -0.0207 5.64E-17
1980 3.5534 -0.6810 0.1772 -2.1262 -0.4286 -0.0260 3.12E-16
1981 3.0515 -0.5335 0.1942 -0.3046 -0.0792 -0.0055 -2.97E-17
1982 3.0770 -0.5428 0.1938 -0.3787 -0.0972 -0.0067 -2.50E-17
1983 3.1910 -0.5784 0.1906 -0.7288 -0.1759 -0.0117 3.55E-17
1984 3.1849 -0.5803 0.1916 -0.7139 -0.1737 -0.0115 6.06E-17
1985 3.0103 -0.5287 0.1971 -0.2038 -0.0548 -0.0038 -1.61E-17
1986 3.2112 -0.5887 0.1909 -0.8011 -0.1923 -0.0126 7.84E-17
1987 3.0051 -0.5256 0.1969 -0.1882 -0.0507 -0.0036 1.95E-17
1988 2.9555 -0.5117 0.1986 -0.0648 -0.0180 -0.0013 -1.00E-18
1989 3.2712 -0.6077 0.1892 -1.0082 -0.2350 -0.0151 1.20E-16
1990 3.5216 -0.6845 0.1813 -2.0035 -0.4158 -0.0248 -6.20E-17
A5.5. - 22
A5.5.5 Economies of Scale
-
AS S S 1 IZEF estimates of economies of scale ISIC 3833
Year EOS 1- EOS2- EOS 3- EOS 1- EOS2- EOS3 -
IZEF lZEF IZEF IZEF (AR) IZEF(AR) lZEF(AR)
1972 1.3576 1.9069 1.0057 2.8852 1.7039 1.1249
1973 1.2823 2.0176 1.0056 2.1383 1.6319 1.0762
1974 1.2082 2.1302 1.0055 1.6769 1.5696 1.0289
1975 1.1944 2.1993 1.0054 1.5414 1.5524 1.0099
1976 1.1950 2.0699 1.0054 1.5014 1.5302 1.0068
1977 1.3832 1.8542 1.0056 2.1763 1.6227 1.08ll
1978 1.4402 1.9227 1.0056 2.2631 1.6438 1.0845
1979 1.3320 2.0131 1.0055 1.7076 1.5531 1.0326
1980 1.2147 2.0572 1.0053 1.3249 1.4580 0.9790
1981 1.2525 2.2300 1.0053 1.3313 1.5036 0.9748
1982 1.2998 2.1557 1.0053 1.3898 1.5150 0.9850
1983 1.3205 2.0708 1.0054 1.3918 1.5023 0.9867
1984 1.3990 1.9688 1.0054 1.5201 1.53ll 1.0068
1985 1.4481 2.0259 1.0054 1.5466 1.5554 1.0078
1986 1.4300 1.9700 1.0054 1.4564 1.5122 0.9969
1987 1.4130 2.1293 1.0053 1.3412 1.5047 0.9748
1988 1.3687 2.1954 1.0052 1.2152 1.4724 0.9510
1989 1.3955 2.0078 1.0053 1.2357 1.4460 0.9587
1990 1.3520 1.9458 1.0052 1.1423 1.3968 0.9418
A5.5.5.2 13SLS estimates of Economies of Scale - ISIC 3833

Year EOS 1- EOS 2- EOS 3- EOS 1 - EOS 2- EOS 3-
13SLS. 13SLS 13SLS 13SLS (AR) 13SLS (AR) 13SLS (AR)
1972 1.4349 1.9331 0.9984 3.0497 1.1390
1973 1.2952 2.0802 0.9944 2.6252 1.0865
1974 1.1706 2.2332 0.9902 2.2813 1.0360
1975 1.1427 2.3240 0.9884 2.1788 1.0157
1976 1.1472 2.1636 0.9881 2.0543 1.0124
1977 1.4553 1.8816 0.9948 2.4685 1.0918
1978 1.5394 1.9599 0.9951 2.6022 1.0954
1979 1.3447 2.0907 0.9905 2.1993 1.0398
1980 1.1624 2.1709 0.9853 1.8301 0.9828
1981 1.1995 2.3736 0.9849 1.9324 0.9784
1982 1.2680 2.2734 0.9859 1.9622 0.9892
1983 1.3008 2.1692 0.9861 1.9147 0.9910
1984 1.4261 2.0331 0.9881 1.9887 1.0124
1985 1.4925 2.0955 0.9882 2.0543 1.0135
1986 1.4643 2.0388 0.9871 1.9187 1.0018
1987 1.4150 2.2387 0.9849 1.8828 0.9784
1988 1.3370 2.3320 0.9824 1.7644 0.9531
1989 1.3893 2.1022 0.9832 1.6989 0.9613
1990 1.3254 2.0387 0.9814 1.5655 0.9434
A5.5. - 23
A 5.5.6 Graphs of Economies of Scale
Figure 5.5.6.1 IZEF estimates of scale- ISle 3833

2.5
~EOS-1
0.5 -X-EOS-2
-X-EOS-3
0+--+-+--+-+----+--+--+-...-,1---+----4--+---+-+--+-+--+-+-....
n n N ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ 00 ~ ~
Figure 5.5.6.2 IZEF (AR) estimates of scale- ISle 3833

3
~EOS-1
-X-EOS-2
2.5
-X-EOS-3
1.5 -X-X- x-x-==~
1Xj-X-X-X-X_X-X-X-X-X-X-X-X-X-X-X_X_X_X
0.5
o
A5.5. - 24
Figure 5.5.6.3 I3SLS estimates of scale- ISIC 3833

2.5
1.5
-<)-EOS·1
0.5 -)I(-EOS-2
-X-EOS.3
o+--+-+--+-+---+--t---+---II---+-.......-+--+-+--+-+--+-+----4
n n M ~ ~ n n ~ 00 ~ ~ ~ ~ ~ ~ ~ ~ ~ 00
Figure 5.5.6.4 I3SLS (AR) estimates of scale- ISIC 3833

3.5
-)I(-EOS-2
3
-X-EOS-3
2.5
2
-)K_/--~ .
~-)I(-)I(-)I(-)I(-)I(--)I(-)I(--)I(
-)1(......""
1.5 '"
1X1-X-X-X-X_X-X-X-X-X-X-X-X-X-X-X-X-X-X
0.5
o
A5.5. - 25
Figure 5.5.6.5 IZEF, IZEF (AR) and 13SLS estimates of Model 1- ISIC 3833
3
2.5
1.5
-X-IZEF
-X-IZEF (AR)
0.5
~13SLS
O+--+-+____+_-+__+_-.....-+----I~_+___+-+__+-+____+_-+__+_-+_
.....
n n ~ ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ ~ ~ ~
Figure 5.5.6.6 IZEF, IZEF (AR), I3SLS and 13SLS (AR) estimates of Model 2 -
ISIC 3833
-X-IZEF
3.5
-X-IZEF (AR)
3 ~13SLS
~13SLS(AR)
2.5
-X-X_w _ X - X _w w
1.5 .....- X .....- X - X - X - X - X -.....- X - X - X - X _ X
0.5
O+--+-+____+_-+__+_ I---+----I-_+_--+-+--+-+-_+_-t---+-_I---4
n n ~ ~ n n n ~ 00 ~ ~ ~ M ~ ~ ~ ~ ~ ~
A5.5. - 26
Figure 5.5.6.7 IZEF, IZEF (AR), I3SLS and I3SLS (AR) estimates of Model 3-
ISle 3833
1.2
0.8
0.6 ~IZEF
-:C-IZEF (AR)
0.4
~13SLS
-o-I3SLS (AR)
0.2
O+----+--.--+---I-_+_--+--+__+_~~_+__+-+__+_~~_+___..-_6__...."
72

Cobbledick John 1995

Uploaded by

Copyright:

Available Formats

Cobbledick John 1995

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

Cobbledick John 1995

Uploaded by

Copyright:

Available Formats

AN APPLIED TRANCENDENTAL LOGARITHMIC

COST FUNCTION: ECONOMIES OF SCALE

A dissertation submitted to the Department ofEconomics,

University of Natal, Durban

I am deeply indebted to a number of people who supported me - both emotionally and

Julian Hofineyr provided me with outstanding supervision. He was always willing to

TABLES AND FIGURES Vlll

CHAPTER 1: CHOICE OF AN APPROPRIATE FUNCTIONAL FORM 5

CHAPTER 2: THE TRANSCENDENTAL LOGARITHMIC COST FUNCTION 38

CHAPTER 4: DATA, ESTIMATION AND TESTING PROCEDURES 64

CHAPTER 5: ECONOMETRIC RESULTS 93

CHAPTER 6: CONCLUSION 123

3 CONCAVITY TESTING A3-1

4 DATA SERIES AND INSTRUMENTAL VARIABLES A4-1

4.1 Data series A4-1

5 ECONOMETRIC RESULTS A5-1

5.1 Forms of alternative specifications of the A5-1

5.2. A priori data analysis A5.2-1

5.3 Motor vehicles, parts and accessories (ISIC 3840) A5.3-1

5.4 Furniture (ISIC 3320) A5.4-1

5.5 Electrical appliances and household goods A5.5-1

TABLES AND FIGURES

4.1.1.1 Cost of capital: electrical appliances and A4.1-1

5.2.3.3 Input-output coefficients of materials (ISIC 3840) A5.2-9

5.3.4.12 Proper AESs and determinants of matrices A5.3-13

5.3.5.2 13 SLS estimates of Economies of Scale- A5.3-24

5.4.4.6 Proper AESs and determinants of matrices A5.4-10

5.4.4.26 Proper AESs and determinants of matrices of A5.4-20

Fig5.4.6.1 Economies of Scale - IZEF estimates - A5.4-24

5.5.2.1 Computed and critical Durbin-Watson A5.5-1

5.5.4.4 Proper AESs and Determinants of matrices A5.5-8

5.5.4.24 Proper AESs and Determinants of matrices A5.5-18

Fig 5.5.6.1 IZEF estimates of scale - ISIC 3833 A5.5-23

The demand restructuring approach to growth through redistribution is premised on

Empirical estimation of the characteristics of the technology underlying either the

Constrained Cobb-Douglas functions, where economies of scale are held constant,

Cobb-Douglas estimates do not, however, contribute to an answer to the second

The application of duality theory to economic problems has provided an alternative

comment on this. Furthermore the precise econometric technique employed2 is not

This dissertation presents an application of the Translog system to three Standard

CHOICE OF AN APPROPRIATE FUNCTIONAL FORM

The empirical analysis of technology is undertaken for many reasons in different

• Distribution (i.e. the share of different factors of production in income) which

• scale (i.e. the existence of constant, increasing or decreasing returns)

The constraint of an implausible maintained hypothesis concerning the validity of a

Although the presence of homotheticity and separability are empirical questions in

2.2. ECONOMIES OF SCALE AND HOMOTHETIC FUNCTIONS

2.2.1 ECONOMIES OF SCALE: A GRAPHICAL ANALYSIS

Source: Gould and Ferguson, 1980, p190.

In the long-run the entrepreneur is able to choose between three investment

Figure 2.2 Long -Run Average Cost Curve

Source: Gould and Ferguson, 1980, p191.

Source: Gould and Ferguson, 1980, p176.

2.2.2 ECONOMIES OF SCALE: A MATHEMATICAL EXPOSITION

Mathematically, two different concepts are used to describe the phenomenon of

2.2.2.1 ELASTICITY OF SCALE

e(x) = dIn !(kx)1 2.1

Marginal productivity may be measured by a unit-free measure, the elasticity of output,

An implication of this relationship is that changes in the relative importance of different

J!::..... ----.:~ ::::."__ Xl