Chapter V

Testing Linear Restriction

Dr Hédi ESSID
Topics to be Covered

 Testing restrictions about

- individual parameters
- groups of parameters

 T-tests for individual parameters

- zero value under H0
- some other value under H0

 F-tests for groups of parameters

- the overall significance of a regression
- group exclusion tests
- other linear restrictions

Econometrics 2
Testing Restrictions : Example Cases

 Consider a simple log-linear demand function with 3

seasonal dummy variables

lnQi = β1 + β2 lnPi + β3 lnYi +1D1i + 2D2i+ 3D3i+ ui

 As well as the overall F-test : H0: β2 =β3=1=2=3=0;

 and standard individual T-tests of zero values for individual

parameters : H0: βj =0 or H0:  j =0;

 you might want to test for example;

(i) 1=2=3=0 (the seasonal effects can be excluded) or

(ii) β2 = -1 (the price elasticity of demand is -1) or

(iii) β2 = -1 and β3 = 1

Econometrics 3
 Testing Restrictions : Example Cases

 Now consider a simple log-linear (Cobb-Douglas) production

function

lnQi = β1 + β2 lnLi + β3 lnKi + ui

 As well as the overall F-test : H0: β2 =β3=0;

and standard individual t-tests of zero values for individual

parameters : H0 : βj =0;

 you might want to test for example;

(i) H0: β2 + β3= 1 (constant returns to scale) or

(ii) H0: β2 = 1 and β3 = 1 .

Econometrics 4
 T-Tests of Non-Zero Parameter Values

 We want to test H0: β=β*

against H1: β≠β*

ˆ j
 Instead of calculating the simple t-ratio : tc 
s.e.( ˆ j )

ˆ j   *
 We compute tc  ∼ St (n  k )
s.e.( ˆ j )
where β* is the value we are testing.

Econometrics 5
F- Tests of Zero Restrictions (Group Exclusion Tests)

 Suppose we have a model of the form

Yt=β1+β2Xt2+β3Xt3+…+βkXtk+ut

 Suppose we wish to test the hypothesis that some

subset of variables (for the sake of argument X2, and X3)
can be dropped from the regression (excluded).

 Here the null hypothesis is

H0: β2=0, β3=0

H1: at least one of the parameters β2, β3, is ≠ 0

Econometrics 6
F -Tests of Other Linear Restrictions

Sometimes we might want to conduct a test of the form

β2+β3+…+βk=1

This is a linear restriction.

Another example of a linear restriction is

β2-β3=0

Which is just another way of writing the restriction

β2=β3

Econometrics 7
F-tests of Other Linear Restrictions
The F value and decision rule

 RSSR  RSSU  / m
FCal  ∼F  m, n  k 
RSSU / n  k

where m is the number of restrictions,

RSSR is the Residual Sum of Squares in the
restricted model (under H0)
RSSu is the Residual Sum of Squares in the
unrestricted model (under H1)

Econometrics 8
F- Tests of Other Linear Restrictions
The F value and decision rule

H0: restrictions are valid

Decision rule:

Reject H0 if Fcal > F(m,n-k-1) ( or if P-value < 0.05)

Accept H0 if Fcal < F(m,n-k-1) ( or if P-value > 0.05)

Econometrics 9
Other Issues Concerning Testing Restrictions :
The Chow Test

What is a Chow Test ?

 The Chow test tells us if the regression coefficients are different

for split data sets.

 Basically, it tests whether one regression line or two separate

regression lines best fit a split set of data.

Econometrics 10
Other Issues Concerning Testing Restrictions :
The Chow Test

Split Data Sets and the Chow Test

Sometimes your data will have a break point or structural point (a period
of significant or violent change), splitting a data set into two parts. For
example:

 Donations given to an organization before and after a natural disaster.

 Stock market prices before and after Black Friday.

 House prices before and after a significant interest change.

 Asset prices before and after civil war.

Econometrics 11
The Chow Test

 The dataset on the left has a single regression line .

 The set on the right has a break point in the middle and two
regression lines.

Econometrics 12
The Chow Test

 If the two parts can be represented by one single regression line, we

say that the regression can be “pooled.”

 Let’s say your linear regression analysis of two parts of a data set
(shown on the right) resulted in the following two linear regression
equations :

 First part of the data: yt = X1*1 + u1

 Second part of the data: yt = X2*2 + u2

 The Chow test would tell you if the coefficients 1 = 2 ( H0 ) .

 If they are equal, the data set can be represented with a single
regression line.

Econometrics 13
The Chow Test

Running the Test

The null hypothesis for the test is that there is no break point (i.e. that the
data set can be represented with a single regression line).

1. Run a regression for the entire data set (the “pooled regression”).
Collect the Residual Sum of Square data (RSS).

2. Run separate regressions on each half of the data set. Collect the
RSS data for the two regressions.

3. Calculate the Chow F-statistic using the RSS from each subsample.

Econometrics 14
The Chow Test

A formal test is performed by calculating the F-statistic

(RSSR  RSSU )/ k
Fcal 
RSSU / n  2k

(RSSR  (RSS1  RSS2 ))/ k

 ∼ Fk,n2k
(RSS1  RSS2 )/ n  2k

Econometrics 15
The Chow Test

where:

RSSU = The Unrestricted Residual Sum of Squares (pooled) for the

whole data set, will be

RSSU = RSS1 +RSS2 .

RSS1 = regression line before break.

RSS2 = regression line after break.

RSSR = The Restricted Residual Sum of Squares.

4. Find the F-critical value from the F-table.

5. Reject the null hypothesis if your calculated F-value falls into the
rejection region (i.e. if the calculated F-value is greater than the F-
critical value). Econometrics 16
The Chow Test

Example :

 The following model is regressed using data in quarterly form from

1990 to 2005 (64 observations) for Malaysian stock prices against
output (structural break in 1997).

st=β1+β2yt+ut

 Note : Structural breaks can occur in time series data or cross

sectional data, when there is a sudden change in the relationship
being examined.

- Examples include sudden policy changes such as a change in

government or sudden move in asset prices (1987) or serious
international disaster such as a civil war.

- We then need to decide whether 2 separate regression lines are

more efficient than a single regression.
Econometrics 17
The Chow Test

Example :

The first regression using all the data produced a RSSR = 0.56,
then 2 regressions were run on a sub-sample of the data:

- from 1990 to 1997, giving a RSS1 = 0.23

- from 1998 to 2005, producing a RSS2 = 0.17,

with n = 64 and k = 2.

Econometrics 18
The Chow Test

Example:

(0.56  (0.23  0.17)) / 2

Fcal   12
(0.23  0.17) /(64  4)

• As the critical value for F(2, 60) = 3.15 with significance (5%).

• As 12 > 3.15, we reject the null hypothesis of structural stability.

 We conclude that there is a structural break in this model, we

need to split the data into 2 sub-samples or use another method to
overcome the break.

Econometrics 19