Residual Tests in Panel

[Cathy Roger]

Dear all!

Does anybody know how I can test the residuals of a DOLS panel equation (with country fixed effects) for heteroscedasticity (Breusch-Pagan-Godfrey test), functional form misspecification (Ramsey test), and parameter instability (Hansen)?

I tried with: eq01.reset(2, promt) for the Ramsey test and eq01.hettest(bpg) for the Breusch-Pagan-Godfrey test. Eviews informs me that these tests are not available for panel equations. Why? How then can I test these things?

Thanks, again!
Catherine

[EViews Glenn]

Whenever one is dealing with potentially non-stationary data one has to be very careful about blindly applying tests which are developed in non-stationary settings. While it is true that the residuals from a cointegrating regression are by definition stationary, the same is not true for the regressors and gradients of the cointegrating equation. Accordingly, the theory underlying the LM tests that you discuss does not follow directly from the stationary results so we have not implemented those tests in this setting. There are also issues with the specification of the alternative hypothesis that are not trivial to address, even in a stationary panel setting.

You could, of course, replicate versions of these tests yourself using residuals and regressors extracted from the estimated equation.

Note that as to heteroskedasticity, if I recall correctly (I'm away from my computer) there are settings in the coefficient covariance calculation that allow for results that are robust to some forms of heteroskedasticity. When considering the Ramsey test, you might want to take a look at the nonlinear cointegration literature for guidance. There is a breakpoint cointegration literature that describes issues with parameter instability.