S.E. Regression in MLE of ARMA output in EViews 9

[bela]

Hello,

How is the S.E. of regression calculated in EViews 9 for MLE estimation of ARIMA models? What is the theory behind that calculation? It is important, because the forecast intervals are calculated based on that.

The estimated parameters with MLE estimation includes an estimation of error variance (sigma square). So shouldn't the corresponding standard error of regression (sigma) be just the sqaure root of that? But it seems (from p115 of EViews Users Guide vol II) that it is constructed by taking the square root of

Sum Squared Resid divided by (n - no. of parameters in the model)

where Sum Squared Resid is obtained from the MLE of the error variance - multiplying the latter by sample size n, as expected.

So my question is regarding the division of SSR by (n-no.of parameters). Why do we still need that now that we are using MLE?

Thanks in advance for your help.