EViews

29. Juni 2010

Inhaltsverzeichnis
1 Introduction

2 Assignment 1

3 Assignment 2a

4 Assignment 2b

5 Assignment 3

10

6 Assignment 3b

12

7 Assignment 4_neu (using programs)

13

8 Assignment 5b

17

9 Assingment 6

18

Eviews 1 Introduction

1 Introduction
The EViews Window
Component
Drop Down
Menus
Command Line

Status Line

Work Area
Help Menu

Purpose
easy-to-use tools for
implementation of EViews
procedures
alternative to drop down
menus for usage of command
processing language at
advanced level
message area indicating
default directory (Path),
default database (DB),
default workfile (WF) Work
Area
display of various object
windows
provision of helpful advice

Object-orientied language and event-oriented language.

The Help System
Help/Users Guides or Help/Command&Programming Reference: provide a complete documentation of EViews on over 2000 pages
Help/Quick Help Reference/Function Reference: overview of functions available for easy use
Help/Quick Help Reference/Object Reference: overview of objects available that help to organize
your workfile, e.g. equation, series, etc.
Using a Workfile
Open the workfile by clicking File/Open/EViews Workfile and indicating the directory
series: gdp (gross domestic product), m1 (money supply), pr (price level), rs (short term interest
rate), resid (explained later)
coef: c (coefficient vector used to represent parameters of equations)
buttons: view, proc, object, etc.
data set information: range, sample

Eviews 1 Introduction
Empty workfile (File/New/Workfilespecify data) still contains the objects: constant, residual
You always have to save the results of a regression with a proper name thus creating an object.
Working with Objects
Objects are collections of related information and operations that are bundled together into easy-to-use
units that you hold in a workfile.
creating objects: click Object/New Object, select object you desire and name it meaningfully.
For taking notes: Type Text
useful objects: text, series, equation, graph, group, etc.
Examining a Single Series
spreadsheet view: double-click on series, e.g. gdp
different sample period: click Sample and change sample range. Can de done in the main window
or when the series in opened as a graph
descriptive statistics: click View/Descriptive Statistics/... (Histogram & Stats)
graphs: click View/Graph/Line
saving graphs: rightclick on graph and save graph to disk: *.emf for inclusion in word and *.eps
for inclusion in latex documents
copying results into word: copy (STRG+C) and paste (STRG+V)
freezing results: click Freeze and name object via Name as to create a new object in your workfile
Examining Several Series
spreadsheet view: click on series1+STRG+series2, double-click in blue area and select open
group
summary statistics: click View/Descriptive Stats/Common Sample
plotting two series: click View/Graph/Line or View/Multiple Graphs/ Line. Graph shows both
series in one graph and multiple graphs shows one graph for each series
scatter diagram: click View/Multiple Graphs/Scatter/...SCATMAT: matrix containing scatterplots for all series
Group objects: Strg (single marks), Shift (mark tow, inbetween everything is makred). After marking
click right/Open/As group/Name. If you dont name it, you get a window where you can either name
or store the object
Using the Quick Menu

Eviews 1 Introduction
changing sample: click Quick/Sample and proceed as before
graphs: click Quick/Graph/Line etc.
generating new series: click Quick/Generate Series and enter equation according to seriesname=function(series) etc. eg. log()
Using EViews Functions
Basic arithmetic operations and basic math functions can be used when working with series, such as
in operations to generate new series (NOTE: some functions require the @ symbol to be identified
by EViews!):
click Quick/Generate Series and enter equation or
use the Command Line via series seriesname=function(series)
Note that in the command line the first entry is always the object type (series, scalar, etc.) you want
to create!
Descriptive statistics functions can solely be used within the Command Line.
Mean: scalar seriesname=@mean(series)
Std: scalar seriesname=@stdev(series)
can be seen in status line
recall: Define new object: Matrix-Vector-Coef. (insert how many values needed), then:
Vectorname(1)=@mean(series)
Vectorname(2)=@stdev(series)
Getting Data into EViews
open *.xls and memorize number of obs, labels and ordering of variables
close *.xls
create empty workfile via FileNewWorkfile and indicate parameters (structure type; data
range)
click ProcImportRead Text Lotus Excel, select data type *.xls and browse
indicate appropriate parameters (indicate all series separated by a Leerzeichen), label data and
save workfile
Other ways:
Go to the folder where you saved the excel file, then drag it and drop it in the grey part of
EViewsFertigstellen

Eviews 2 Assignment 1
Click right in the grey part of EViews Open Foreign data as workfile choose excel file
Fertigstellen OR Open EViews Workfile Dateityp: .xlsChoose excel file Fertigstellen

2 Assignment 1
1.3
Rename an object: click right on objectrename (for workfile: save)
To change the order of the variables in a group, use Spec
1.4
ViewDescriptive statsCommon table (for a graph)Freeze (to save)
1.5
ViewGraphScatterSimple scatterFreeze (to save)
To change graph:
Add titel: AddText
Scaling: OptionAxes/Scaleschoose Axis under Edit Axis and Scaling User specified
Symbols: OptionLine/SamplesDifferent options
Save to disk: ProcSave graph to disc
2.1
QuickEstimate equationenter variables: food_exp c income OR food_exp=c(1)+c(2)*income
and choose Method: LS save the output through name
2.2
c and resid are now filled, but are overwritten after each equation is computed
2.3
(i) c(1) insignificant on 5% significance level, c(2) significant on all conventional significant levels
(ii) If income increases by 1 unit (100$), food_exp increases by 10 units (1$)
(iii) R2 =0,385
3.1
To create predicted values after estimating:
series predictedvalues=@coefs(row of object c)+@coefs(row of object c)*x-variables
Quick way: after estimating click on Forecast and enter a name to save it
3.2
Series resid1=resid (this way the residuals are saved and dont change after the next regression)
Series resid1=food_exp - predictedvalues

Eviews 3 Assignment 2a
3.3
Open predictedvalues and resid1 as a group and use scatterplot.
Quickgraphscatterenter series names (with Leerzeichen).
You see heteroscedasticity
Line/shade: indicate orientation and position (0)

3 Assignment 2a
2.1
FileNewProgram
Mark whether a sentence is supposed to run or not: text prevents running
2.2
To open a workfile: wfopenPath Run
2.3
To form a group: group name objects (eg. group grp1 advert price sales)
Open the descriptive statistics: name.stats (eg. grp1.stats)
2.4
Estimate a regression: equation name.typ y-var c x1-var x2-var (eg. equation reg1.ls sales c price
advert)
Procs:
Compute residuals: name.makeresid name (eg. reg1.makeresid residuals)
Compute predicted values: reg1.forecast name (eg. reg1.forecast predicted_data)
Create scatterplot: graph name.scat name1 name2 (eg. graph regscat.scat predicted_data residuals)
Compute R-squared: scalar name=name.@r2 (eg. scalar firstscalar=reg1.@r2)
To save: save
Programs are superior for repetitive tasks
3.1
t-stats:
scalar name=@coefs(No)/@stderrs(No)
(Besser: scalar name=reg1.@coefs(No)/reg1.@stderrs(No))
3.2
p-value (2-seitig):
scalar name=2*(1-@ctdist(name t-stat,n-k-1)) (bei positivem t-stat)
scalar name=2*(@ctdist(name t-stat,n-k-1)) (bei negativem t-stat)
3.3
Critical value:
scalar name=@qtdist(1-/2,n-k-1)

Eviews 4 Assignment 2b
3.5
t-stats:
scalar name=@coefs(No)-Wert/@stderrs(No)
p-value (1-seitig):
scalar name=1-@ctdist(name t-stat,n-k-1)) (bei positivem t-stat)
scalar name=@ctdist(name t-stat,n-k-1)) (bei negativem t-stat)
Critical value:
scalar name=@qtdist(1-,n-k-1)
Appendix
Data members of equation (are computed together with eq., so we can access them using the following
comments):
@aic
Akaike information criterion
@coefcov(i,j)
Covariance of coefficient estimates i and j
@coefs(i)
i-th coefficient value
@f
F-statistic
@logl
Value of the log likelihood function
@meandep
Mean of the dependent variable
@ncoef
Number of estimated coefficients
@r2
R-squared statistic
@regobs
Number of observations in regression
@schwarz
Schwarz information criterion
@sddep
Standard deviation of the dependent Variable
@se
Standard error of the regression
@ssr
Sum of squared residuals
@stderrs(i)
Standard error of coefficient i
@tstats(i)
t-statistic value for coefficient i
c(i)
i-th element of default coefficient vector for equation
@coefcov
Covariance matrix for coefficient estimates
@coefs
Coefficient vector
@stderrs
Vector of standard errors for coefficients
@tstats
Vector of t-statistic values for coefficients

4 Assignment 2b
1.1-2.6
Siehe Assignment 2a
3.1
Set sample range: smpl 1 n obs

Eviews 4 Assignment 2b
3.2
Lower bound of confidence interval: scalar name=@coefs-@stderrs*critical value
Upper bound of confidence interval: scalar name=@coefs+@stderrs*critical value
3.3
We can choose any c inside the confidence interval.
4.2
series name=@dnorm(x) (x here a series)
scalar name=@dnorm(x) (x here a scalar)
4.3
series name=@dtdist(x,df) (x here a series)
scalar name=@dtdist(x,df) (x here a scalar)
4.4
Take the series compute in 4.2 and 4.3, open as a group and than viewgraphline
4 using a program
Compare the distributions of a normal distribution and a t-distribution (with 4 different degrees of
freedom) in a graph:
How to create a vector:
vector (No. of elements) vectorname
vectorname(1)=number or path
vectorname(2)=number or path
... (depending on number of element)
Text of program:
Vector(4) df
df(1)=4
df(2)=10
df(3)=100
df(4)=2000
series normal=@dnorm(x)
for!i=1 to 4
series tdist{!i}=@dtdist(x,df(!i))
group joint{!i} normal tdist{!i}
freeze(grafik{!i}) joint{!i}.line
next
freeze(grafikall) grafik1 grafik2 grafik3 grafik4
show grafikall.align(2,1,1)
Elemente:
Vector name: df
Group name: joint

Eviews 4 Assignment 2b
Graph name: grafik
Joint graph name: grafik all
Elements of align: align(n,h,v) with n=number of columns, h=horizontal space, v=vertical space
freeze(name of graph) to save a graph and not only show it
{!i} as a replacement
5.1
c = z0 = z1 = 0

0 0 1 0
0
c

R = 0 0 0 1 = = r = 0
z0
0 1 0 0
0
z1

(1)

5.2
scalar fstat=reg1.f
scalar pval=1-@cfdist(fstat,3,2773) (allg.: scalar name=1-@cfdist(fstat,#r,n-k))
5.3
scalar critval=@qfdist(0.95,3,2773)
5.4
Equation objectViewCoefficient testsWald-Coefficient Restrictionsc(3)=0, c(4)=0
5.5
ViewCoefficient testsWald-Coefficient Restrictionsc(3)=0, c(4)=0, c(2)=0.005

0 0 1 0
0
c

0
R = 0 0 0 1 =
z0 = r =
0 1 0 0
0.005
z1

(2)

5 using a program
Estimate the equation
equation reg1.ls delta_p c delta_q q qv
Estimate the variance-covariance matrix
c: series interc=1 (we define a constant)
X: group reggr interc delta_q q qv (create a group containing all variables)
| {z }
name for X

n-k: scalar degfr=reg1.@regobs-reg1.@ncoef (calculate the degrees of freedom)

XX: matrix(4, 4) secmom
=@
|
{z
} | {z }
define object name for XX
0
(X X)1 : matrix insecmom

{z

inner
| {z }

(reggr) (Alternative: XX=@transpose(X)*X)

for the innter product

} =@inverse(secmom)

namef or(X 0 X)1

Eviews 5 Assignment 3
ee/n-k: scalar ssqrd=reg1.@ssr/degfr (mit ssr: sum of squared residuals)
matrix estvarcov=insecmom*ssqrd
Short Way:
matrix
cov
| {z } evmat=reg1.@coef
| {z }

object type

d
V ar(b|X)

Matrix of the linear restrictions

R: matrix (2, 4) linrest
(Define empty matrix)
| {z } | {z }
dimension

name

linrest(1,3)=1 (H0 : z0 = 0)
linrest (2, 4) =1 (H0 : z1 = 0)
| {z }
field of matrix

Vector of the restrictions

r=Rb: vector(2) restvec=linrest* reg1.@coef s
|
{z
}

vector with all coefficients

scalar nrest=@rows(linrest) (number of restrictions)

Quadratic form
vector(1) f_ratio=@transpose(restvec)*@inverse(linrest*estvarcov*@transpose(linrest))*restvec/nrest
Testing
scalar fcritval=@qfdist(0.95,nrest,degfr) (compute the critical value for the F-distribution; in general:
@qfdist(1-,#r,n-k))
Use Conditional Statement:
if fcritval>f_ratio(1) then
%dec=The null hypothesis cannot be rejected at a 5% significance level
else
%dec=The null hypothesis can be rejected at a 5% significance level
end if
% used to define a string
alpha decision=%dec
show decision

5 Assignment 3
1.4
ViewDistributionKernel Density Graphs (nonparametric method of smoothing)Kernel (doesnt
matter with large sample size= leave standard)
2.1
ls wage c educ exper (zum Speichern: equation name.ls wage c educ exper)

10

Eviews 5 Assignment 3
2.3
If you divide the dependent variable by a constant, the parameters also have to be divided by the
same constant
2.6
If you multiply the independent variable by a constant, the parameters changes accordingly being
divided by that constant.
2.7
Standardized series:
series wage_z=(wage-@mean(wage))/@stdev(wage)
2.8
bj : @coefs(i);
bj : @stdev(xi);
by : @stdev(y)
scalar name=@stdev(xi)/@stdev(y)*eq1.@coefs(i)
3.1
log(variable) is the function for the natural logarithm of a variable.
Appendix: Theory
Linear transformation of regressors
Let x be the nxk matrix of the regressors and A a kxk nonsingular linear transformation matrix.
Postmultiplying x by A yields:
xA = x[a1 , ..., ak ] = [xa1 , ..., xak ]
|{z}
B

A projection matrix looks like this: B(B 0 B)1 B 0

This leads to: PB = PAx = xA(A0 x0 xA)1 A0 x0 = xAA1 (x0 x)1 (A0 )1 A0 x0 = x(x0 x)1 x0 = Px
With linear transformations we only change the coefficients, but the fitted values stay the same. Since
the vectors of fitted values and residuals e depend on x only through Px and Mx , they are invariant
to any nonsingular linear transformation of the columns of x.
Adding or substracting constants to an independent variable doesnt change the coefficient.
Create deviation from the mean for a series:
series educabw=educ-@mean(educ) (the sum of this series is 0)
For the regression wage c educabw and wage educabw the coefficient is the same (as c and educabw
are orthogonal).
Standardization
Starting point: original OLS equation
yi = b0 + b1 xi1 + ... + bk xik + ei
yi y = b0 b0 + b1 xi1 b1 x1 + ... + bk xik bk xk = b1 (xi1 x1 ) + ... + bk (xik xk ) + ei
Let
by be the sample standard deviation of y,
bx the sample standard deviation of x.

bx
(xi1 x1 )

bx
(xik xk )
ei
(yi y)
= 1 b1
+ ... + k bk
+

by

by

bx1

by

bxk

by

(3)

If xj increase by one standard deviation, then yb changes by bj standard deviations.

11

Eviews 6 Assignment 3b

6 Assignment 3b
3.1
Enter for the observation: 1 200
3.2
For educ: eq.@coefs(2)1.96*eq.@stderrs(2) (bzw. statt 1.96: -@qtdist(0.025,n-k))
The one with the lower n is wider (as we have smaller stderrs for higher n and a different t-stat).
Our aim is a narrow interval.
3.3
1 =85 cannot be rejected.
We do not reject for all values inside the confidence interval: [64,88]
4.1
log(variable) for ln
4.3
We can reject at any conventional significance level as p-value is close to 0.
4.4
Centered R2 : 1-SSR/SST
SSR=eq6.@ssr; SST=(eq6.@sddep)2 *(n-1)
Adjusted R2 : 1-(n-1/n-k)SSR/SST
4.5
Adjusted R2 : (5) 0.129; (6) 0.179
Akaike: (5) 0.973; (6) 0.916
Schwarz: (5) 0.989; (6) 0.941
all criteria point toward equation 6
4.6
AIC=log(SSR/n)+2k/n
SBC=log(SSR/n)+log(n)k/n
Maximized log-likelihood under the normality assumption:
e 0 (y X )
e
L = n/2 log(2) n/2 log(e
) 2e1 2 (y X )
e0 e
1
L = n/2 log(2) n/2 log( n ) e0 e e0 e
)
n
SSR
2 SSR
n
log( SSR
n )
2(

L = n/2 log(2) n/2

log( SSR
n )

L = n/2 log(2) n/2 n/2

L = n/2(1 + log(2) + log( SSR
n ))
The formula used by Eviews:
AIC=-2(L/n)+2k/n
SBC=-2(L/n)+k*log(n)/n

12

Eviews 7 Assignment 4_neu (using programs)

For the L use: eq6.@logl
4.7
Siehe 4.5

7 Assignment 4_neu (using programs)

1
u
Create a workfile: workfile simultest
| {z } |{z}
name

11000
| {z }

undated samplesize

1.1
Construction of the mode:
x: series Regress = @ runif (|{z}
6 , |{z}
12 )
| {z }
| {z }
N ame

unif ormdist

-vector:
vector(2) Beta
| {z }
N ame

Beta(1)=5
Beta(2)=0.5
series

yb =Beta(1)+Beta(2)*Regress
|{z}

N ame

About distributions: q=quantil, d=density, r=random

2
2.1
Decision variables: !m=1000
vector(!m) betavec
| {z }
N ame

smpl @first 20 (we cut the sample size down to 20)

The loop:
For !k=1 to !m
series y_stoch=b
y + @rtdist(5)
| {z }
epsilon

equation reg01.ls y_stoch c Regress

betavec(!k)=reg01.@coefs(2)
Next
2.2
Series are dependent on the sample size, vectors arent. But series have lots of view, that cant be
used with a vector.
Change sample size to 1000 as in the vector: smpl @first !m

13

Eviews 7 Assignment 4_neu (using programs)

Transform:

mtos
| {z }

( betavec
| {z } , |betaser
{z } ) (where betaser is a series that is created new here)

matrixtoseries vectorname seriesname

Plot histogram: freeze(Kern

| {z }) betaser.hist (include Jarque-Bera test)
N ame

Plot empirical distribution test: freeze(Tests) betaser.edftest (4 tests to test for a normal distribution)
3
General:
{!i} as part of a name
(!i) as a variable or parameter
3.1
workfile simultest u 1 2000
Definition of the sample sizes:
vector(4) stp
stp(1)=10
stp(2)=20
stp(3)=300
stp(4)=2000
Construction of the model as in 1.1
Decision variables: !n=1000
The loops:
For !i=1 to 4
!stp=stp(!i)
smpl @first !stp (sets the sample sizes)
vector(!n) betavec{!i} (This time we create four vectors; one for every sample size)
For !k=1 to !n
series y_stocj=b
y +@rtdist(5)
equation reg01.ls y_stoch c Regress
betavec{!i}(!k)=reg01.@coefs(2)
Next
Next
3.2
Just change the parameters of @runif(a,b)
4
4.1
Creation of series and graphics:
smpl @first !n

14

Eviews 7 Assignment 4_neu (using programs)

For !i=1 to 4
mtos(betavec{!i},betaser{!i})
freeze(Kern{!i}) betaser{!i}.hist
freeze(Tests{!i}) betaser{!i}.edftest Next
(This loop could be part of the upper loops after the first next)
freeze(endgra) Kern1 Kern2 Kern3 Kern4
show endgra.align(2,3,1) (N 1: number of columns; N 2: horizontal space; N 3: vertical space)
Added task: use of either the uniform or the t-dist
Construction of the model as in 1.1; definition of the sample sizes as in 3.1
Decision variable:
!n=1000
!switch=1
!vert=1
The loop:
For !i=1 to 4
!stp=stp(!i)
smpl @first !stp
vector(!n) betavec{!i}
For !k=1 to !n
If !vert=1 then
Series y_stoch=b
y +@runif(-2,2)
Else
series y_stoch=b
y +@rtdist(5)
End if
equation reg01.ls y_stoch c Regress
betavec{!i}(!k)=reg01.@coefs(2)
Next
Next
5
5.1-5.4
Subroutine name(output, arguments)
...
endsub
subroutine jbt(scalar p,series y)
series spow=(y@mean(y))2
4
series fpow=(y @mean(y))
scalar smom=@mean(spow)
scalar sch=thmom/(smom( 3/2))

scalar thmom=@mean(thpow)

series thpow=(y@mean(y))3
scalar fomom=@mean(fpow)

scalar woelb=fomom/(smom( 2))

15

Eviews 7 Assignment 4_neu (using programs)

scalar jbtest=(@obs(y)/6)*(sch2 +((woelb 3)2 )/4)

scalar p=1-@cchisq(jbtest,2) end-

sub
Formeln:
JB =

n 2 (K 3)2
(S +
)
6
4

(4)

mit
1
n

P
(y y)3
S= 1P
( n (y y)2 )3/2

(5)

und
K=

1 P
(y y)4
n
P
( n1 (y y)2 )2

(6)

5.5
We use the programm text from the added task, then we compute:
smpl @first !n
For !i=1 to 4
scalar p{!i}
mtos(betavec{!i},betaser{!i})
Alternative computation (instead of those two lines)
call jbt(p{!i},betaser{!i})
if !switch=1 then
freeze(Kern{!i}) betaser{!i}.kdensity
else
freeze(Kern{!i}) betaser{!i}.hist
endif
freeze(Tests{!i}) betaser{!i}.edftest
next
Composite graph:
freeze(endgra) Kern1 Kern2 Kern3 Kern4
show endgra.align(2,3,1)
5.6
Local (after subroutine, before the name): prevents the subroutine from producing everything. With
local only the thing specified as output is stored and the other ones are lost after the output is
created.

16

Eviews 8 Assignment 5b

8 Assignment 5b
Theory CAPM
E[rjt rf ] = j E[rmt rf ]

(7)

where:
rjt =risky asset return on asset j (in period t)
rmr =risky return of the market portfolio
rf =riskless return (time invariant)
Cov(rjt ,rmt )
= V ar(r
=proportionality factor
mt )
Regression model:
vjt = rjt E(rjt )
vmt = rmt E(rmt )
7 becomes then:
rjt vjt rf = j (rmt vmt rf )
rjt rf = j (rmt rf ) + vjt j vmt
|
{z
}
jt

It holds:
E(jt ) = E(vjt ) j E(vmt ) = 0
Since j =

Cov(rjt ,rmt )
V ar(rmt )

E[(rjt E(rjt ))(rmt E(rmt ))]

E[(rmt E(rmt ))2 ]

E[vjt vmt ]
2 )
E(vmt

It holds:
E[jt (rmt rf )] = E[(vjt j vmt )(rmt rf )] = E[(vjt j vmt )(vmt + E(rmt ) rf )]
|
{z
}
= E[(vjt j vmt ) vmt ] = E[vjt vmt ]

2 ]
j E[vmt

= E[vjt vmt ]

non-stochastic
E[vjt vmt ]
2 ]=0
E[vmt
2 )
E(vmt

1.1
Open workfile, then ProcStructure/Resize Current Page=> Enter data structure
1.4-2.2
Use a program:
for !i=1 to 10
series r{!i}rf r{!i}-rf
equation eq{!i}.ls r{!i}rf c rmrf
next
2.3
Before loop: vector(10) tstat
As part of loop: tstat(!i)=(eq{!i}.@coefs(2)-1)/eq{!i}.@stderrs(2)

17

Eviews 9 Assingment 6
For p-value:
Before loop: vector(10) pvalue
As part of loop: pvalue(!i)=2*(1-@cnorm(@abs(tstat(!i))))
Matrix instead of vector:
matrix (2, 2)
a
|{z}
| {z }
Zeilen,Spalten N ame

a(1,1)=Zahl
a(1,2)=Zahl
a(2,1)=Zahl
a(2,2)=Zahl
2.4
Before loop: vector(10) constant
As part of loop: constant(!i)=eq{!i}.@coefs(1)
3.1
for !i=1 to 10
euqation equ{!i}.ls r{!i}rf c rmrf smb hml umd
next
3.2
2 with @rbar2, Akaike with @aic and Schwarz with @schwarz. Example for R2 :
Use Radj
adj
Before loop: vector(10) radj
As part of the loop: radj(!i)=equ{!i}.@rbar2
3.3
freeze(wald{!i}) Equ .wald Conditions (here: c(3)=0, c(4)=0, c(5)=0)
|{z}
N ame

All p-values are close to zero

9 Assingment 6
Theory
1. Homoscedasticity assumption:
V ar(i |x) = E(i |x) = 2 for i=1,...,n
2. Heteroscedasticity:
V ar(|x) = E(2 |x) = i2
3. Consequences of heteroscedasticity:
Under assumption 1.1-1.3 and 1.5 OLS estimator unbiased but no longer BLUE and t- and F-tests
not valid
V ar(b) = (X 0 X)1 X 0 V ar()I X(X 0 X)1 = 2 (X 0 X)1 (under homoscedasticity)
| {z }

18

Eviews 9 Assingment 6
is always smaller than under heteroscedasticity. To estimate without Assumption 1.4:
P
P
P
P
d
Vd
ar(b) = (X 0 X)1 e2 xi x0 (X 0 X)1 = ( xi x0 )1 e2 xi x0 ( xi x0 )1 = Avar(b)
i

4. Testing for heteroscedasticity:

Idea: start with the linear model: yi = 0 + 1 xi1 + ... + k xik + i
H0 : Assumption of homoscedasticity is true: E(2i |x) = 2 i
If H0 false, then expected value of 2i given the independent variable can be any funktion of the
xij : E(2i )=f(x)
Breusch-Pagan-Test
f(x): linear function of x
2i = 0 + 1 xi1 + ... + k xik + vi
H0 : 1 = 2 = ... = k = 0 F-test OR
n R22 a 2 (k) Langrange Multiplier test
White-Test
f(x): linear function of x, cross product and squares of the independent variables:
2i = 0 + 1 xi1 + ... + k xik + k+1 x2i1 + ... + 2k x2ik + 2k+1 xi1 xi2 + 2k+2 xi1 xi3 + ... + vi
To test more general forms of heteroscedasticity:
H0 : all i = 0 for i 6= 0
Test statistic: R22 a 2 (k)
1.3
series lnname=log(name) (name=Variable)
2.1
series predname=eq1.@coefs(1)+eq2.@coefs(2)*lnwage+eq3.@coefs(3)*lnoutput+eq4.@coefs(4)*lncapital
2.2
series residuals=lnlabour-predlnlabour
2.3
Residual plot: y-axis with residuals, x-axis with predicted values
Open as groupViewGraphXY
click right on graphOptionsAxes/ScalesZero-line
2.4
Seems to be rather heteroscedastic. From that follows:
Estimates/coefficients: dont change
Std.errors: lower than with robust test
Inference: We reject to easy
R2 : dont change
3.1
series residuals2=residuals2
3.2
equation ls residuals2 c lnwage lnoutput lncapital

19

Eviews 9 Assingment 6
3.3
H0 : we have homoscedasticity
BP = n R22 2 (k) (here: k=3)
scalar bp=n*eq2.@r2
3.4
scalar pvalue=qcchisq(bp,3)
4.1
series ln2name=lnname2
series lnname1name2=lnname1*lnname2
4.2
equation ls residuals2 c lnwage lnoutput lncapital ln2wage ln2output ln2capital lnoutputwage lncapitalwage lncapitaloutput
4.3
White=n R22
scalar white=n*eq3.@r2
4.4
scalar whitecrit=@qchisq(0.95,9)
White and Breusch-Pagan both rejecto n the 5% significant level

5.1
ViewEstimate equationEnter the same equation as for OLSOptions Heteroscedasticity-consistentWhiteO

20