Heteroscedasticity test for Weighted Least Square statistics

[cheeyoong1]

Using unstructured data in eviews, I have produced a set of weighted statistics and wanted to confirm there's no more heteroscedascity in the weighted least square model. I used the Glesjer test for this purpose.

I obtained the weighted residuals and made it absolute, absuhat=abs(resid) from the weighted least square model and run the regression, ls absuhat c w*x1 w*x2 and got the glesjer test results output. (w=1/sigma or standard deviation. * is multiply).

The problem is how come my output is different from the eviews glesjer test button output ? The glesjer test button output is awresid = c + x1*wgt +x2*wgt (wgt = weight= 1/sigma or std dev, awresid is absolute weigthed residuals).

If I used Breusch-Pagan test and Harvey-Godfrey test, it's still different between my output and the button's output.

Hope to receive the reply urgently. You can email to cheeyoong1@gmail.com.

Sincerely,
Eric

[EViews Glenn]

Without seeing your data it is hard to say whether this is the cause of the discrepancy, but I do note from your notation that you didn't weight the constant in your auxiliary regression....

[latisha11]

yes, more JMK Design information is needed

[EViews Glenn]

This was pushed to my RSS feed by the recent response so I went in to see whether I could figure out what might be going on. Most likely, the issue is the scaling of the default weights.

Suppose we estimate the following WLS equation:

Code: Select all

equation eq01.ls(w=f) y c x

The following auxiliary regression matches the Glejser test output run on the full sample:

Code: Select all

eq01.makresid res1 equation testeq.ls abs(res1)*f/@mean(f) c x*f/@mean(f)

Note that if there are automatic sample adjustments made in the original estimation, these adjustments must also be made to the call to @mean.

Lastly, I don't know whether the original poster is going to see this or not. We do not as a general rule respond to forum posts directly to email when the topic is presumably of general interest.