EViews.com

Hi

i am running a monthly time series regression with stock return of firm as a dependent variable and independent variables are exchange rate changes and market portfolio return. all variables are stationary.
It is well known that the one of the assumptions of regression is that the errors should be normally distributed.

My query is whether the individual variables or series should also be normally distributed in time series regression. What if the errors are not normally distributed in time series regression and all other residual test like LM serial correlation and ARCH are OK.

Please help.

Hi

...
It is well known that the one of the assumptions of regression is that the errors should be normally distributed.

No it isn't. The Gauss-Markov theorem does not require normal errors. The only place that normal errors matters is for small sample distribution of the coefficients being normal.

My query is whether the individual variables or series should also be normally distributed in time series regression. What if the errors are not normally distributed in time series regression and all other residual test like LM serial correlation and ARCH are OK.

The series being normal is irrelevant for everything. Asymptotic test statistics do not require normality.
If deviations from normality are very large, then one does need a larger sample than otherwise for asymptotic theory to kick in.

Thanks a lot for reply.