Chapter 4

Classical linear regression model assumptions

and diagnostics

1
 Violation of the Assumptions of the CLRM

• Recall that we assumed of the CLRM disturbance terms:

1.
2.
3.
4. The X matrix is non-stochastic or fixed in repeated samples
5.  N(0,2)

2
 Investigating Violations of the
Assumptions of the CLRM

• We will now study these assumptions further, and in particular look at:
- How we test for violations
- Causes
- Consequences
In general we could encounter any combination of 3 problems:
- the coefficient estimates are wrong
- the associated standard errors are wrong
- the distribution that we assumed for the
test statistics will be inappropriate
- Solutions
- the assumptions are no longer violated
- we work around the problem so that we
use alternative techniques which are still valid
3
 Assumption 1: E( ) = 0

• Assumption that the mean of the disturbances is zero.

• For all diagnostic tests, we cannot observe the disturbances and so

perform the tests of the residuals.

• The mean of the residuals will always be zero provided that there is a
constant term in the regression.

4
 Assumption 2: Var( ) = 2 < 

• We have so far assumed that the variance of the errors is constant, 2 - this
is known as homoscedasticity. ˆ +
t

• If the errors do not have a

constant variance, we say
that they are heteroscedastic
e.g. say we estimate a regression
. calculate the residuals, ( )
and
x 2t

-
5
 Detection of Heteroscedasticity: The GQ Test

• Graphical methods
• Formal tests: There are many of them: we will discuss Goldfeld-Quandt
test and White’s test

test is carried out as follows.

1. Split the total sample of length T into two sub-samples of length T1 and T2.
The regression model is estimated on each sub-sample and the two
residual variances are calculated.
2. The null hypothesis is that the variances of the disturbances are equal,
H0:  12   22

6
 The GQ Test (Cont’d)

3. The test statistic, denoted GQ, is simply the ratio of the two residual
variances where the larger of the two variances must be placed in
the numerator.
s12
GQ  2
s2
4. The test statistic is distributed as an F(T1-k, T2-k) under the null of
homoscedasticity.

5. A problem with the test is that the choice of where to split the
sample is that usually arbitrary and may crucially affect the
outcome of the test.

7
 Detection of Heteroscedasticity using White’s Test

• White’s general test for heteroscedasticity is one of the best

approaches because it makes few assumptions about the form of the
heteroscedasticity.
• The test is carried out as follows:
1. Assume that the regression we carried out is as follows
yt = 1 + 2x2t + 3x3t +
And we want to test Var( ) = 2.
We estimate the model, obtaining the residuals,

2. Then run the auxiliary regression

ˆt2  1   2 x2t   3 x3t   4 x22t   5 x32t   6 x2t x3t  vt
8
 Performing White’s Test for Heteroscedasticity

3. Obtain R2 from the auxiliary regression and multiply it by the

number of observations, T. It can be shown that
T R2  2 (m)
where m is the number of regressors in the auxiliary regression
excluding the constant term.

4. Decision rule: If the 2 test statistic from step 3 is greater than the
corresponding value from the statistical table then reject the null
hypothesis that the disturbances are homoscedastic.

9
 Consequences of Using OLS in the Presence of
Heteroscedasticity

• OLS estimation still gives unbiased coefficient estimates, but they are
no longer BLUE.

The Ordinary Least Squares (OLS) estimator is the Best Linear Unbiased
Estimator. This means:

- Best: It has the smallest variance among all linear estimators.

- Linear: It is a linear function of the data.
- Unbiased: It gives the correct estimate on average (the expected value
equals the true parameter).
- Estimator: it is used to estimate the coefficients (β).
This property holds when certain assumptions (such as no perfect
multicollinearity, homoscedasticity, and no autocorrelation of errors) are
met. 10
 Consequences of Using OLS in the Presence of
Heteroscedasticity

• This implies that if we still use OLS in the presence of

heteroscedasticity, our standard errors could be inappropriate and
hence any inferences we make could be misleading.

• Whether the standard errors calculated using the usual formulae are
too big or too small will depend upon the form of the
heteroscedasticity.

11
 How Do we Deal with Heteroscedasticity?

• If the form (i.e. the cause) of the heteroscedasticity is known, then we can
use an estimation method which takes this into account (called generalised
least squares, GLS).
• A simple illustration of GLS is as follows: Suppose that the error variance is
related to another variable zt by
var t    2 zt2
• To remove the heteroscedasticity, divide the regression equation by zt
yt 1 x2 t x3t
 1   2  3  vt
zt zt zt zt
t
where vt  is an error term.
zt
  t  var t   2 zt2
• Now varvt   var   2
 2
  2
for known zt.
• So the disturbances zt  from
zt zt
the new regression equation will be
homoscedastic. 12
 Background –
The Concept of a Lagged Value

t yt yt-1 yt
1989M09 0.8 - -
1989M10 1.3 0.8 1.3-0.8=0.5
1989M11 -0.9 1.3 -0.9-1.3=-2.2
1989M12 0.2 -0.9 0.2--0.9=1.1
1990M01 -1.7 0.2 -1.7-0.2=-1.9
1990M02 2.3 -1.7 2.3--1.7=4.0
1990M03 0.1 2.3 0.1-2.3=-2.2
1990M04 0.0 0.1 0.0-0.1=-0.1
. . . .
. . . .
. . . .
13
 Autocorrelation

• We assumed of the CLRM’s errors that Cov ( i , j) = 0 for ij, i.e.

This is essentially the same as saying there is no pattern in the errors.

• Obviously we never have the actual u’s, so we use their sample

counterpart, the residuals (the ’s).

• If there are patterns in the residuals from a model, we say that they are
autocorrelated.

• Some stereotypical patterns we may find in the residuals are given on

the next 3 slides.

14
 Positive Autocorrelation

+
û t û t
+

- +
uˆ t 1 Time

Positive Autocorrelation is indicated by a cyclical residual plot over time.

15
 Negative Autocorrelation

+ ût
û t
+

- +
uˆ t 1 Time

- -

Negative autocorrelation is indicated by an alternating pattern where the residuals

cross the time axis more frequently than if they were distributed randomly
16
 No pattern in residuals –
No autocorrelation
ût
+
û t
+

- +
uˆ t 1

No pattern in residuals at all: this is what we would like to see

17
 Detecting Autocorrelation:
The Durbin-Watson Test

The Durbin-Watson (DW) is a test for first order autocorrelation -

i.e. it assumes that the relationship is between an error and the
previous one
where vt  N(0, v2).
• The DW test statistic actually tests H0 : =0 and H1 : 0
• The test statistic is calculated by
∑𝑇𝑡=2 (𝜀 − 𝜀 )
∑𝑇𝑡=2 𝜀

• Rearranging for DW would give 0DW4.

• If DW = 2. So roughly speaking, do not reject the null hypothesis if
DW is near 2  i.e. there is little evidence of autocorrelation
• Unfortunately, DW has 2 critical values, an upper critical value (du)
and a lower critical value (dL), and there is also an intermediate region
where we can neither reject nor not reject H0. 18
 The Durbin-Watson Test: Interpreting the Results

Conditions which Must be Fulfilled for DW to be a Valid Test

1. Constant term in regression
2. Regressors are non-stochastic
3. No lags of dependent variable 19
 Another Test for Autocorrelation:
The Breusch-Godfrey Test

• It is a more general test for rth order autocorrelation:

• The null and alternative hypotheses are:

H0 : 1 = 0 and 2 = 0 and ... and r = 0
H1 : 1  0 or 2  0 or ... or r  0
• The test is carried out as follows:
1. Estimate the linear regression using OLS and obtain the residuals, .
2. Regress on all of the regressors from stage 1 (the x’s) plus

Obtain R2 from this regression.

3. It can be shown that (T-r)R2  2(r)
• If the test statistic exceeds the critical value from the statistical tables, reject
the null hypothesis of no autocorrelation. 20
 Consequences of Ignoring Autocorrelation
if it is Present

• The coefficient estimates derived using OLS are still unbiased, but
they are inefficient, i.e. they are not BLUE, even in large sample sizes.

• Thus, if the standard error estimates are inappropriate, there exists the
possibility that we could make the wrong inferences.

• R2 is likely to be inflated relative to its “correct” value for positively

correlated residuals.

21
 “Remedies” for Autocorrelation

• If the form of the autocorrelation is known, we could use a GLS

procedure – i.e. an approach that allows for autocorrelated residuals
e.g., Cochrane-Orcutt.

• But such procedures that “correct” for autocorrelation require

assumptions about the form of the autocorrelation.

• If these assumptions are invalid, the cure would be more dangerous

than the disease! - see Hendry and Mizon (1978).

• However, it i s unli k el y to b e the ca s e t h a t t h e f o rm of t h e

autocorrelation is known, and a more “modern” view is that residual
autocorrelation presents an opportunity to modify the regression.

22
 Multicollinearity

• This problem occurs when the explanatory variables are very highly correlated
with each other.

• Perfect multicollinearity
Cannot estimate all the coefficients
- e.g. suppose x3 = 2x2
and the model is yt = 1 + 2x2t + 3x3t + 4x4t +

• Problems if Near Multicollinearity is Present but Ignored

- R2 will be high but the individual coefficients will have high standard errors.
- The regression becomes very sensitive to small changes in the specification.
- Thus confidence intervals for the parameters will be very wide, and
significance tests might therefore give inappropriate conclusions.

23
 Measuring Multicollinearity

• Method 1: The easiest way to measure the extent of multicollinearity is

simply to look at the matrix of correlations between the individual
variables. e.g.
Corr x2 x3 x4
x2 - 0.2 0.8
x3 0.2 - 0.3
x4 0.8 0.3 -

• But another problem: if 3 or more variables are linear

- e.g. x2t + x3t = x4t

• Note that high correlation between y and one of the x’s is not
muticollinearity.
• Method 2: Variance inflationary factor
24
Measuring Multicollinearity
 Solutions to the Problem of Multicollinearity

• “Traditional” approaches, such as ridge regression or principal

components. But these usually bring more problems than they solve.

• Some econometricians argue that if the model is otherwise OK, just

ignore it

• The easiest ways to “cure” the problems are

- drop one of the collinear variables
- transform the highly correlated variables into a ratio
- go out and collect more data e.g.
- a longer run of data
- switch to a higher frequency

26
 Adopting the Wrong Functional Form

• We have previously assumed that the appropriate functional form is linear.

• This may not always be true.
• We can formally test this using Ramsey’s RESET test, which is a general test
for mis-specification of functional form.

• Essentially the method works by adding higher order terms of the fitted values
(e.g. yt2 , yt3 etc.) into an auxiliary regression:
Regress on powers of the fitted values:
•
Obtain R2 from this regression. The test statistic is given by TR2 and is
distributed as a  2 ( p  1) .

• So if the value of the test statistic is greater than a  2 ( p  1) then reject the null
hypothesis that the functional form was correct.
27
 But what do we do if this is the case?

• The RESET test gives us no guide as to what a better specification

might be.

• One possible cause of rejection of the test is if the true model is

yt  1   2 x2t   3 x22t   4 x23t   t
In this case the remedy is obvious.

• Another possibility is to transform the data into logarithms. This will

linearise many previously multiplicative models into additive ones:

yt  Axt e t  ln yt     ln xt   t

28
 Testing the Normality Assumption

• Why did we need to assume normality for hypothesis testing?

Testing for Departures from Normality

• The Bera Jarque normality test

• A normal distribution is not skewed and is defined to have a
coefficient of kurtosis of 3.
• The kurtosis of the normal distribution is 3 so its excess kurtosis (b2-3)
is zero.
• Skewness and kurtosis are the (standardised) third and fourth moments
of a distribution.

29
 Testing for Normality

• Bera and Jarque formalise this by testing the residuals for normality by
testing whether the coefficient of skewness and the coefficient of excess
kurtosis are jointly zero.
• It can be proved that the coefficients of skewness and kurtosis can be
expressed respectively as:
𝐸[ ] 𝐸[ ]
( ) / ( )

The Bera Jarque test statistic is given by

 b12 b2  32 
W T   ~  2 
2

6 24 
• We estimate b1 and b2 using the residuals from the OLS regression, .

30
 What do we do if we find evidence of Non-Normality?

• It is not obvious what we should do!

• Could use a method which does not assume normality, but difficult and
what are its properties?

• Often the case that one or two very extreme residuals causes us to reject
the normality assumption.

• An alternative is to use dummy variables.

e.g. say we estimate a monthly model of asset returns from 1980-1990, and
we plot the residuals, and find a particularly large outlier for October
1987:

‘Introductory Econometrics for Finance’ © Chris Brooks 2013 31

 What do we do if we find evidence
of Non-Normality? (cont’d)

ût
+

Oct Time
1987

Create a new variable:

D87M10t = 1 during October 1987 and zero otherwise.
This effectively knocks out that observation. But we need a theoretical
reason for adding dummy variables. 32
 Omission of an Important Variable or
Inclusion of an Irrelevant Variable

Omission of an Important Variable

• Consequence: The estimated coefficients on all the other variables will be
biased and inconsistent unless the excluded variable is uncorrelated with
all the included variables.
• Even if this condition is satisfied, the estimate of the coefficient on the
constant term will be biased.
• The standard errors will also be biased.

Inclusion of an Irrelevant Variable

• Coefficient estimates will still be consistent and unbiased, but the
estimators will be inefficient.

33
 Parameter Stability Tests

• So far, we have estimated regressions such as yt = 1 + 2x2t + 3x3t + ut

• We have implicitly assumed that the parameters (  1 ,  2 and  3 ) are

constant for the entire sample period.

• We can test this implicit assumption using parameter stability tests. The
idea is essentially to split the data into sub-periods and then to estimate up
to three models, for each of the sub-parts and for all the data and then to
“compare” the RSS of the models.

• There are two types of test we can look at:

- Chow test (analysis of variance test)
- Predictive failure tests

34
 The Chow Test

• The steps involved are:

1. Split the data into two sub-periods. Estimate the regression over the
whole period and then for the two sub-periods separately (3 regressions).
Obtain the RSS for each regression.
2. The restricted regression is now the regression for the whole period
while the “unrestricted regression” comes in two parts: for each of the sub-
samples.
We can thus form an F-test which is the difference between the RSS’s.

The statistic is RSS   RSS1  RSS2  T  2k


RSS1  RSS2 k

35
 The Chow Test (cont’d)

where:
RSS = RSS for whole sample
RSS1 = RSS for sub-sample 1
RSS2 = RSS for sub-sample 2
T = number of observations
2k = number of regressors in the “unrestricted” regression (since it comes
in two parts)
k = number of regressors in (each part of the) “unrestricted” regression

3. Perform the test. If the value of the test statistic is greater than the critical
value from the F-distribution, which is an F(k, T-2k), then reject the null
hypothesis that the parameters are stable over time.
36
 A Chow Test Example

• Consider the following regression for the CAPM  (again) for the
returns on Glaxo.

• Say that we are interested in estimating Beta for monthly data from
1981-1992. The model for each sub-period is

• 1981M1 - 1987M10
0.24 + 1.2RMt T = 82 RSS1 = 0.03555
• 1987M11 - 1992M12
0.68 + 1.53RMt T = 62 RSS2 = 0.00336
• 1981M1 - 1992M12
0.39 + 1.37RMt T = 144 RSS = 0.0434

37
 A Chow Test Example - Results

• The null hypothesis is

H0 : 1   2 and 1  2
• The unrestricted model is the model where this restriction is not imposed

00434
.   00355
.  000336
.  144  4
Test statistic  
00355
.  000336
. 2
= 7.698

Compare with 5% F(2,140) = 3.06

• We reject H0 at the 5% level and say that we reject the restriction that the
coefficients are the same in the two periods.
38