EViews Guide [8]: Heteroskedasticity

[1] Example 8.1

[2] Example 8.6
[3] Example 8.9

[1] Example 8.1

(The student version of Eviews 4.1 is used to get the results for
heteroscedasticity. If you use a current version of Eviews such as 6th
version , the results might change due to using different set of explanatory
variables especially in white heteroscedasticity test)

 Get the White SE/The Breusch-Pagan (BP) Test

file: wage1.wf1
The usual SE:

Dependent Variable: LWAGE

Variable Coefficient Std. Error t-Statistic Prob.
C 0.321378 0.100009 3.213492 0.0014
MARR_MALE 0.212676 0.055357 3.841881 0.0001
MARR_FEMALE -0.198268 0.057835 -3.428132 0.0007
SINGLE_FEMALE -0.110350 0.055742 -1.979658 0.0483
EDUC 0.078910 0.006694 11.78733 0.0000
EXPER 0.026801 0.005243 5.111835 0.0000
EXPERSQ -0.000535 0.000110 -4.847105 0.0000
TENURE 0.029088 0.006762 4.301613 0.0000
TENURESQ -0.000533 0.000231 -2.305552 0.0215

 In the Equation Specification box, click Options

61
 Choose Heteroskedasticity Consistent Covariance/White in the
Estimation Options box.

 This will bring you the following results.

Dependent Variable: LWAGE

Method: Least Squares

62
 Date: 09/28/04 Time: 21:32
Sample: 1 526
Included observations: 526
White Heteroskedasticity-Consistent Standard Errors & Covariance
Variable Coefficient Std. Error t-Statistic Prob.
C 0.321378 0.109469 2.935791 0.0035
MARR_MALE 0.212676 0.057142 3.721886 0.0002
MARR_FEMALE -0.198268 0.058770 -3.373619 0.0008
SINGLE_FEMALE -0.110350 0.057116 -1.932028 0.0539
EDUC 0.078910 0.007415 10.64246 0.0000
EXPER 0.026801 0.005139 5.215010 0.0000
EXPERSQ -0.000535 0.000106 -5.033361 0.0000
TENURE 0.029088 0.006941 4.190731 0.0000
TENURESQ -0.000533 0.000244 -2.187835 0.0291

 In this example, the White SEs and the usual ones are almost the same.
 Perform the BP test. In the Eq_8_6 window, choose View/Residual
Tests/White Heteroskedasticity (no cross terms)1. Note: the BP test
is a special case of the White test without cross terms.

You’ll get

1
The student version of Eviews 4.1 is used to get the results for heteroscedasticity. If you use a
current version of Eviews such as 6th version, the results might change due to using different set of
explanatory variables especially in white heteroscedasticity test.

63
 F-statistic 1.457277 Probability 0.144024
Obs*R-squared 15.90816 Probability 0.144575

Test Equation:
Dependent Variable: RESID^2
Method: Least Squares
Date: 09/28/04 Time: 21:53
Sample: 1 526
Included observations: 526
Variable Coefficient Std. Error t-Statistic Prob.
C 0.222807 0.132654 1.679608 0.0936
MARR_MALE -0.045922 0.038520 -1.192169 0.2337
MARR_FEMALE -0.043534 0.040276 -1.080894 0.2803
SINGLE_FEMALE 0.007294 0.038493 0.189497 0.8498
EDUC -0.025062 0.020476 -1.224002 0.2215
EDUC^2 0.001229 0.000845 1.453638 0.1467
EXPER 0.006965 0.006350 1.096818 0.2732
EXPER^2 -7.24E-05 0.000231 -0.313872 0.7537
EXPERSQ^2 -3.81E-08 5.78E-08 -0.659291 0.5100
TENURE -0.001891 0.006711 -0.281820 0.7782
TENURE^2 0.000228 0.000363 0.628406 0.5300
TENURESQ^2 -6.54E-08 1.62E-07 -0.404133 0.6863
R-squared 0.030244 Mean dependent var 0.152030

 The F-statistic is 1.457277 (p-value: 0.144024). So, the null

hypothesis of no heteroskedasticity can’t be rejected at 10% level. No
wonder the White SEs and the usual SEs are so close. (All the
coefficients are insignificant individually!)

[2] Example 8.6  WLS (duplicate Table 8.1)

Use 401ksubs_a.wf1 (includes 750 observations only). Results will be

different than the book as we use only 750 of all observations.
eq01
Dependent Variable: SAV
Method: Least Squares
Sample: 1 100
Included observations: 100
Variable Coefficient Std. Error t-Statistic Prob.
C 124.8424 655.3931 0.190485 0.8493
INC 0.146628 0.057549 2.547897 0.0124
R-squared 0.062127 Mean dependent var 1582.510
Adjusted R-squared 0.052557 S.D. dependent var 3284.902
S.E. of regression 3197.415 Akaike info criterion 18.99787
Sum squared resid 1.00E+09 Schwarz criterion 19.04997
Log likelihood -947.8935 F-statistic 6.491778
Durbin-Watson stat 1.536387 Prob(F-statistic) 0.012391

64
 eq02

Dependent Variable: NETTFA

Method: Least Squares
Date: 03/31/09 Time: 15:57
Sample: 1 750
Included observations: 750
Weighting series: INC^(-1/2)
Variable Coefficient Std. Error t-Statistic Prob.
C -10.53002 2.459160 -4.281958 0.0000
INC 0.716128 0.072553 9.870356 0.0000
Weighted Statistics
R-squared 0.072976 Mean dependent var 13.42219
Adjusted R-squared 0.071737 S.D. dependent var 37.26687
S.E. of regression 35.90530 Akaike info criterion 10.00231
Sum squared resid 964314.3 Schwarz criterion 10.01463
Log likelihood -3748.866 F-statistic 97.42393
Durbin-Watson stat 2.068947 Prob(F-statistic) 0.000000
Unweighted Statistics
R-squared 0.157110 Mean dependent var 17.76764
Adjusted R-squared 0.155983 S.D. dependent var 56.03648
S.E. of regression 51.48093 Sum squared resid 1982414.
Durbin-Watson stat 2.047916

eq03

Dependent Variable: NETTFA

Method: Least Squares
Date: 03/31/09 Time: 15:59
Sample: 1 750
Included observations: 750
Variable Coefficient Std. Error t-Statistic Prob.
C -30.97435 3.991226 -7.760610 0.0000
INC 0.831059 0.079463 10.45851 0.0000
(AGE-25)^2 0.026924 0.004646 5.795071 0.0000
MALE 2.975827 4.494008 0.662177 0.5081
E401K 13.87599 3.912373 3.546694 0.0004
R-squared 0.216530 Mean dependent var 17.76764
Adjusted R-squared 0.212323 S.D. dependent var 56.03648
S.E. of regression 49.73302 Akaike info criterion 10.65786
Sum squared resid 1842663. Schwarz criterion 10.68866
Log likelihood -3991.697 F-statistic 51.47439
Durbin-Watson stat 2.014239 Prob(F-statistic) 0.000000

65
eq04

Dependent Variable: NETTFA

Method: Least Squares
Date: 03/31/09 Time: 16:02
Sample: 1 750
Included observations: 750
Weighting series: INC^(-1/2)
Variable Coefficient Std. Error t-Statistic Prob.
C -18.44427 2.683550 -6.873086 0.0000
INC 0.610794 0.073316 8.331028 0.0000
(AGE-25)^2 0.022355 0.003134 7.133370 0.0000
MALE 2.643732 2.942505 0.898463 0.3692
E401K 8.532961 2.749420 3.103550 0.0020
Weighted Statistics
R-squared 0.143210 Mean dependent var 13.42219
Adjusted R-squared 0.138610 S.D. dependent var 37.26687
S.E. of regression 34.58780 Akaike info criterion 9.931523
Sum squared resid 891255.4 Schwarz criterion 9.962324
Log likelihood -3719.321 F-statistic 41.51430
Durbin-Watson stat 2.040807 Prob(F-statistic) 0.000000
Unweighted Statistics
R-squared 0.201081 Mean dependent var 17.76764
Adjusted R-squared 0.196791 S.D. dependent var 56.03648
S.E. of regression 50.22096 Sum squared resid 1878998.
Durbin-Watson stat 2.022764
 How do we get eq02 and eq04? Taking eq02 for example.

 Click Options in the Equation Specification window

66
You’ll see the following:

 Choose Weighted LS/TSLS and type the weight inc^(-1/2) in the box.
Click OK. And, OK again.

Remarks:

 Do the same steps for eq04.

 The WLS output contains both weighted and unweighted summary
statistics. The weighted summary statistics are based on fitted
residuals, computed using the weighted data while the unweighted
summary results are based on the residuals computed from the original
(unweighted) data.
 You may compare the R2 from the “unweighted” results with the OLS
R2. You may find the unweighted R2 (eq02/eq04) is different from the
WLS R2 reported in Table 8.1. Use the EViews R2!
 It is assumed that var(ui | inci )   2inci so the weight is 1/(inci)1/2 (or
(inci)-1/2). Alternatively, if var(ui | inci )   2inci2 is assumed, the weight
is 1/inci.

67
[3] Example 8.9

 use GPA1.wfl
 generate parcoll: parcoll=(fathcoll>0 or mothcoll>0). That is,
parcoll=1 if either fathcoll>0 or mothcoll>0; =0, otherwise.
 OLS with the standard SE

Dependent Variable: PC
Method: Least Squares
Sample: 1 141
Included observations: 141
Variable Coefficient Std. Error t-Statistic Prob.
C -0.000432 0.490536 -0.000881 0.9993
HSGPA 0.065394 0.137258 0.476435 0.6345
ACT 0.000565 0.015497 0.036427 0.9710
PARCOLL 0.221054 0.092957 2.378024 0.0188
R-squared 0.041526 Mean dependent var 0.397163
Adjusted R-squared 0.020537 S.D. dependent var 0.491055
S.E. of regression 0.485986 Akaike info criterion 1.422685
Sum squared resid 32.35700 Schwarz criterion 1.506338
Log likelihood -96.29930 F-statistic 1.978510
Durbin-Watson stat 1.759636 Prob(F-statistic) 0.120096

 OLS with the White robust SE

Dependent Variable: PC
Method: Least Squares
Sample: 1 141
Included observations: 141
White Heteroskedasticity-Consistent Standard Errors & Covariance
Variable Coefficient Std. Error t-Statistic Prob.
C -0.000432 0.495880 -0.000871 0.9993
HSGPA 0.065394 0.141487 0.462194 0.6447
ACT 0.000565 0.016071 0.035126 0.9720
PARCOLL 0.221054 0.088038 2.510900 0.0132
R-squared 0.041526 Mean dependent var 0.397163
Adjusted R-squared 0.020537 S.D. dependent var 0.491055
S.E. of regression 0.485986 Akaike info criterion 1.422685
Sum squared resid 32.35700 Schwarz criterion 1.506338
Log likelihood -96.29930 F-statistic 1.978510
Durbin-Watson stat 1.759636 Prob(F-statistic) 0.120096

 Note: The White SE is somewhat different from that of the textbook

(equation 8.48). Use the EViews one.
 Now, perform WLS as follows.

68
1. Estimate the model by OLS and obtain the predicted pc (click
“forecast” in the OLS output window): pcf (automatically
generated by EViews).
2. Since pcf is always between (0,1), go ahead to generate weight:
h=pcf*(1-pcf).
3. Re-estimate the model by WLS, using 1/h^0.5 as the weight.

Dependent Variable: PC
Method: Least Squares
Sample: 1 141
Included observations: 141
Weighting series: 1/H^0.5
Variable Coefficient Std. Error t-Statistic Prob.
C 0.026210 0.476650 0.054988 0.9562
HSGPA 0.032703 0.129882 0.251790 0.8016
ACT 0.004272 0.015453 0.276455 0.7826
PARCOLL 0.215186 0.086292 2.493703 0.0138
Weighted Statistics
R-squared 0.026172 Mean dependent var 0.390076
Adjusted R-squared 0.004847 S.D. dependent var 0.484103
S.E. of regression 0.482928 Akaike info criterion 1.410059
Sum squared resid 31.95104 Schwarz criterion 1.493712
Log likelihood -95.40919 F-statistic 2.224037
Durbin-Watson stat 1.731452 Prob(F-statistic) 0.088163
Unweighted Statistics
R-squared 0.040928 Mean dependent var 0.397163
Adjusted R-squared 0.019927 S.D. dependent var 0.491055
S.E. of regression 0.486138 Sum squared resid 32.37717
Durbin-Watson stat 1.757520

69