The RATS Software Forum

I am running a partial adjustment model (which includes lagged dependent) and Durbin h test suggests that there is still higher order autocorrelation. I included a second lag of the dependent variable and then tested for serial correlation in residuals with Breusch-Godfrey Test but I am still rejecting the null hypothesis of no autocorrelation. Any suggestions what else I can do to correct for higher order autocorrelation?

Your help will be greatly appreciated.

Thanks

Petros

Other that the obvious step of adding more lags, the other thing is to check whether there is some other problem with the regression. For instance, the BG test assumes homoscedasticity as well. ARCH effects, for instance, can cause a model to reject lack of serial correlation even though there is no obvious adjustment to the mean model that could eliminate serial correlation---it just throws the serial correlation test off enough that you can never get it to pass.

Many thank Tom