12/18/2024

Heteroskedasticity
(Incomplete)

Dr Arshad Ali Bhatti

School of Economics, IIIE
Fall, 2024

Homoskedasticity
 In the classical regression model, it is
assumed that var(ui) = σ2 for all i and for
all possible values of the independent
variables X1, …, Xk. [Homoskedasticity]

 What happens when this assumption is

violated? [Heteroskedasticity]

Dr Arshad Ali Bhatti/

Spring 2019 2

1
 12/18/2024

Heteroskedasticity: Example

Dr Arshad Ali Bhatti/

Spring 2019 3

Heteroskedasticity: Example

Dr Arshad Ali Bhatti/

Spring 2019 4

2
 12/18/2024

Heteroskedasticity: Example
 Simple form:
 Yi = βo + β1Xi + ui
 var(ui) = Xiσ2
 This may occur when:
 Households have substantially different income levels,
provinces have substantially different population, income
and level of economic development
 Countries differ substantially in population, income, and
level of economic development
 More general form:
 var(ui) = f(Z1i, Z2i, …, Zni) σ2

Dr Arshad Ali Bhatti/

Spring 2019 5

Consequences
 The OLS estimators are still unbiased and consistent. This
is because none of the explanatory variables is correlated
with the error term. So a correctly specified equation will
give us values of estimated coefficient which are very close
to the real parameters.
 Affects the distribution of the estimated coefficients
increasing the variances of the distributions and therefore
making the OLS estimators inefficient.
 Underestimates the variances of the estimators, leading to
higher values of t and F statistics.
Dr Arshad Ali Bhatti/
Spring 2019 6

3
 12/18/2024

Detection Methods
 Informal Tests:
 Graphical Method
 Formal Tests:

Goldfeld-Quandt test The Harvey-Godfrey LM test

Breusch-Pagan LM test The Engle’s LM test
Glesjer test The White’s test
The Park LM test

Dr Arshad Ali Bhatti/

Spring 2019 7

Graphical Method

Dr Arshad Ali Bhatti/

Spring 2019 8

4
 12/18/2024

Graphical Method

Dr Arshad Ali Bhatti/

Spring 2019 9

Graphical Method

Dr Arshad Ali Bhatti/

Spring 2019 10

10

5
 12/18/2024

Graphical Method

Dr Arshad Ali Bhatti/

Spring 2019 11

11

Graphical Method

Dr Arshad Ali Bhatti/

Spring 2019 12

12

6
 12/18/2024

Formal Tests

Dr Arshad Ali Bhatti/

Spring 2019 13

13

Goldfeld-Quandt test
 First test, rarely used today (not available in EViews)
 Step1: Identify one variable that is closely related to
the variance of the disturbances, and order (rank)
the observations of this variable in descending order
(starting with the highest and going to the lowest).
 Step2: Split the ordered sample into two equally
sized sub-samples by omitting c central
observations, so that the two samples will contain
½(n-c) observations.

Dr Arshad Ali Bhatti/

Spring 2019 14

14

7
 12/18/2024

Goldfeld-Quandt test
 Step3: Run and OLS regression of Y on the X
variable that you have used in step 1 for each
sub-sample and obtain the RSS for each
equation.
 Step4: Caclulate the F-stat=RSS1/RSS2, where
RSS1 is the RSS with the largest value.
 If F-stat>F-tab(α, n1-k, n2-k) ; reject the null of
homoskedasticity.

Dr Arshad Ali Bhatti/

Spring 2019 15

15

Breusch-Pagan LM Test
Step1: Estimate the model by OLS and obtain the
residuals
Step2: Run the following auxiliary regression:
= + + +…+ +

Step3: Compute LM=nR2, where n and R2 are from the

auxiliary regression.
Step4: If LM-stat>χ2p-1 critical reject the null and
conclude that there is significant evidence of
heteroskedasticity
Dr Arshad Ali Bhatti/
Spring 2019 16

16

8
 12/18/2024

Breusch-Pagan LM Test
 Implementation in Eviews
 Estimate original regression using OLS
 In estimation window click: View→
Residual tests → Heteroskedasticity tests
→ Breusch-Pagan-Godfrey
 Use F or Chi-square statistic to decide
about the rejection of:
H0: No Heteroskedasticity
Dr Arshad Ali Bhatti/
Spring 2019 17

17

The Glesjer LM Test

Step 1: Estimate the model by OLS and obtain the
residuals
Step 2: Run the following auxiliary regression:
| uˆt |= a1 + a2 Z 2t + a3 Z 3t + ... + a p Z pt + vt

Step 3: Compute LM=nR2, where n and R2 are from the

auxiliary regression.
Step 4: If LM-stat>χ2p-1 critical reject the null and conclude
that there is significant evidence of heteroskedasticity

Dr Arshad Ali Bhatti/

Spring 2019 18

18

9
 12/18/2024

The Glesjer LM Test

 Implementation in Eviews
 Estimate original regression using OLS
 In estimation window click: View→
Residual tests → Heteroskedasticity tests
→ Glesjer
 Use F or Chi-square statistic to decide
about the rejection of:
H0: No Heteroskedasticity
Dr Arshad Ali Bhatti/
Spring 2019 19

19

The Park LM Test

Step1: Estimate the model by OLS and obtain the residuals
Step2: Run the following auxiliary regression:

ln uˆ t2 = a1 + a2 ln Z 2t + a3 ln Z 3t + ... + a p ln Z pt + vt

Step3: Compute LM=nR2, where n and R2 are from the

auxiliary regression.
Step4: If LM-stat>χ2p-1 critical reject the null and conclude
that there is significant evidence of
heteroskedasticity
Dr Arshad Ali Bhatti/
Spring 2019 20

20

10
 12/18/2024

The Harvey-Godfrey LM Test

Step1: Estimate the model by OLS and obtain the residuals
Step2: Run the following auxiliary regression:

ln uˆt2 = a1 + a2 Z 2t + a3 Z 3t + ... + a p Z pt + vt

Step3: Compute LM=nR2, where n and R2 are from the

auxiliary regression.
Step4: If LM-stat>χ2p-1 critical reject the null and conclude
that there is significant evidence of
heteroskedasticity
Dr Arshad Ali Bhatti/
Spring 2019 21

21

The Harvey-Godfrey LM Test

 Implementation in Eviews
 Estimate original regression using OLS
 In estimation window click: View→
Residual tests → Heteroskedasticity tests
→ Harvey
 Use F or Chi-square statistic to decide
about the rejection of:
H0: No Heteroskedasticity
Dr Arshad Ali Bhatti/
Spring 2019 22

22

11
 12/18/2024

The Engle’s ARCH Test

 Engle introduced a new concept allowing for hetero-
skedasticity to occur in the variance of the error terms,
rather than in the error terms themselves.
 The key idea is that the variance of ut depends on the
size of the squarred error term lagged one period u2t-1
for the first order model or:
 Var(ut)=α1+ α2u 2
t-1

 The model can be easily extended for higher orders:

 Var(ut)= α1+ α2u 2
t-1+…+ αku 2
t-k

Dr Arshad Ali Bhatti/

Spring 2019 23

23

The Engle’s ARCH Test

 Step1: Estimate the model by OLS and obtain the
residuals
 Step2: Regress the squared residuals to a constant and
lagged terms of squaredαresiduals, the number of lags
will be determined by the hypothesized order of ARCH
effects.
 Step3: Compute the LM statistic = (n-k)R2 from the LM
model and compare it with the chi-square critical value.
 Step4: Conclude

Dr Arshad Ali Bhatti/

Spring 2019 24

24

12
 12/18/2024

The Engle’s ARCH Test

 Implementation in Eviews
 Estimate original regression using OLS
 In estimation window click: View→
Residual tests → Heteroskedasticity tests
→ ARCH
 Use F or Chi-square statistic to decide
about the rejection of:
H0: No Heteroskedasticity
Dr Arshad Ali Bhatti/
Spring 2019 25

25

The White’s Test

Step1: Estimate the model by OLS and obtain the residuals
Step2: Run the following auxiliary regression:
uˆt2 = a1 + a2 X 2t + a3 X 3t + a4 X 22t + a5 X 32t + a6 X 2 t X 3t + vt

Step3: Compute LM=nR2, where n and R2 are from the

auxiliary regression.
Step4: If LM-stat>χ2p-1 critical reject the null and conclude
that there is significant evidence of
heteroskedasticity

Dr Arshad Ali Bhatti/

Spring 2019 26

26

13
 12/18/2024

The White’s Test

 Implementation in Eviews
 Estimate original regression using OLS
 In estimation window click: View→
Residual tests → Heteroskedasticity tests
→ White
 Use F or Chi-square statistic to decide
about the rejection of:
H0: No Heteroskedasticity
Dr Arshad Ali Bhatti/
Spring 2019 27

27

Solutions

1. Generalized Least Squares (GLS)

2. Weighted Least Squares (WLS)
3. Heteroskedasticity-Consistent Estimation
Methods

Dr Arshad Ali Bhatti/

Spring 2019 28

28

14
 12/18/2024

GLS: If variances are known

Suppose: Yi = β o + β1 X 1i + β 2 X 2i + L + β k X ki + ui
where: var(ui ) = X jiσ 2
Multiplying both sides by 1/ X ji
Yi 1 X X X ki u
= βo + β1 1i + β 2 2i + L + β k + i
X ji X ji X ji X ji X ji X ji
ui 1
var( )= var(u i )
X ji X ji
1
= ( X jiσ 2 ) = σ 2
X ji
Dr Arshad Ali Bhatti/
Spring 2019 29

29

GLS: If variances are known

Define: Y
Y%i = i
X ji
1
X% oi =
X ji
X
X% 1i = 1i
X ji
M
X ki
X% ki =
X ji
Dr Arshad Ali Bhatti/
Spring 2019 30

30

15
 12/18/2024

GLS: If variances are known

Estimate:
Y%i = β o X% oi + β1 X% 1i + β 2 X% 2i + L + β k X% ki + ε i
ui
where: ε i =
X ji

Dr Arshad Ali Bhatti/

Spring 2019 31

31

GLS: If variances are known

 This is another example of a generalized least-squares
(GLS) estimator
 This estimation procedure is equivalent to minimizing a
weighted sum of squares where the weight on the
squared error term for observation i is 1/Xji

 This procedure is also called a weighted least-

squares (WLS) estimator.

 Intuition: places more weight on more precise

information and less weight on less precise information
 The resulting estimates are BLUE.
Dr Arshad Ali Bhatti/
Spring 2019 32

32

16
 12/18/2024

GLS: If variances are known

 More general known functional form
 If var(ui) = f(Z1i, Z2i,, …, Zmi, )σ2,
repeat the steps above, but divide the
dependent and all independent
variables by:

f ( Z1i , Z 2i ,..., Z mi )

Dr Arshad Ali Bhatti/

Spring 2019 33

33

GLS: If variances are known

 Implementation in Eviews
 Generate a new variable for weights
genr w=1/(Xji)^0.5
 Estimate original regression using OLS
 In estimation window click: Estimate→
Options → weighted LS/TSLS, put w in weight
box → OK
 This method is known as weighted least
square (WLS).
Dr Arshad Ali Bhatti/
Spring 2019 34

34

17
 12/18/2024

If Functional Form is Unknown

 Use results from Breusch-Pagan or White auxiliary
regression to specify a functional form and apply
the procedure above using the estimated function.
 Use White’s Heteroskedasticity Consistent
Covariance Matrics (HCCM) procedure
 weights observations by squared sample
residuals
 this procedure produces corrected standard
errors and appropriate t-statistics when the
functional form of the variance is not known
 available in most econometric packages
Dr Arshad Ali Bhatti/
Spring 2019 35

35

If Functional Form is Unknown

 Implementation in Eviews
 In estimation window click: Estimate→
Options → HAC; ( for hetero-auto
consistent standard errors)
 Tests based on a HAC are consistent
asymptotically

Dr Arshad Ali Bhatti/

Spring 2019 36

36

18
 12/18/2024

Application
• Estimation of Consumption Function
(Quarterly data)

Dr Arshad Ali Bhatti/

Spring 2019 37

37

References
• Johnston and Dinardo, Econometric Methods,
McGraw-Hill, 1997.
• Dimitrios, Applied Econometrics using Eviews,
Palgrave, 2006.
• Green, W. H., Econometric Analysis, Pearson,
2002.
• Gujrati 4e (undergraduate text)
• Dougherty (undergraduate text)

Dr Arshad Ali Bhatti/

Spring 2019 38

38

19