Hi! In a VAR/VEC framework, there are several methods to get normally-distributed residuals. Is there a way to get realiable t-values also with non-normal residuals? I know there is an algorithm to get reliable t-values in the face of ARCH-effects, but if there are no ARCH-effects?
Good for you! Too many users don't bother to wonder if normality is even an assumption they are making, much less what to do if it is untrue. In econometrics this sin is very widespread, it seems. R makes it easy to err this way.
Depending on your needs, you might want an FFT as an alternative to ARCH.
Thre are good alternatives to the t-test, if that helps; http://thestatsgeek.com/2014/04/12/is-the-wilcoxon-mann-whitney-test-a-good-non-parametric-alternative-to-the-t-test/
This problems comes up in several fields. Here is a non-finance paper that might help
http://rctdesign.org/techreports/arphnonnormality.pdf
Thanks! I'm using Stata, and just discovered robust regression. It seems to do the trick, right?
Not very familiar with Stata - so many packages these days, and I think it may be more popular in Europe than the US? There are a lot of types of robust regression. A pretty good summary is here; http://www.stata.com/meeting/germany12/abstracts/desug12_jann.pdf
- Badges
- Science topic