SE of regression with GARCH models

[strypste]

Hi there

I estimated a GARCH model and wanted to replicate the S.E. of regression. The formula given in the help for this statistic is given by

square root of (Residual sum of squares)/(Nr of observations - number of regressors).

I noticed, however, that Eviews uses only the coefficients in the mean equation as number of regressors. So for example, if I estimate a AR(2)-ARCH(1) model with a constant in the mean and variance equation. The number of regressors used in the formula above is 3 and not 5. Is this ok?

Thank you!

Best
s

[startz]

A couple of things...

EViews is correct, although if the difference between n-3, n-5, or n is noticeable, then you have to few observations to be relying on GARCH in the first place.

If you are doing GARCH it isn't obvious what the standard error of the regression means, since the variance is time-varying.

[strypste]

Ok great. Thanks!

I am using about 630 observations. Do you think that is not enough?

Best
S

[startz]

That's probably enough.