The RATS Software Forum

as the title,
parameters estimated by MLE is assumed the distribution follows normal distribution.
however, if we set the distribution follow non-normal one, then, we have to apply quasi-MLE.
But, I couldn't find the function of QMLE in WinRats. does it have this function?
or we don't have to account this issue?

Can someone answer me this ? thanks a lot.

Frank

With QMLE, you maximize a likelihood using standard methods that you either know or suspect is misspecified. However, there are quite a few situations where the misspecified likelihood is adequate for obtaining consistent estimates of the parameters - an obvious example would be least squares, which is the MLE for i.i.d. Gaussian disturbances, but consistent under much broader circumstances.

While the parameter estimates might be consistent, a covariance matrix that comes from the inverse of the information matrix won't be correct. In general, the ROBUSTERRORS option corrects for the failure of the information equality when you're not using the correct likelihood. That's available on DDV, GARCH, LDV, MAXIMIZE, NLLS, NLSYSTEM and SUR, as well as LINREG.

I've attached some fairly generic lecture notes on QMLE. (These are from the State Space/DSGE course materials). As described above, the general treatment with RATS is to use your standard instruction for estimation, but include the ROBUSTERRORS option.