EViews.com

Hello,

I am looking at how, if at all, the time-series predictability of stock returns varies depending on whether the market is in a good state or a bad state, as indicated by a dummy variable. Using the dummy variable to split my sample, I want to conduct a variance ratio test. For the bad time sample, I was wondering, does the test with a truncated sample calculate the n-period before variance ratio (from bad states only) or the n-observation before variance ratio? My data are monthly, so my question is does the VR test treat adjacent observations in the dataset as being one month apart (even if they aren't) or does it know how far apart they really are? I'm using EViews 8 and EViews does know that the data are monthly and dated.

Also, were I to run an LM test on an equation estimated over a truncated sample, would EViews treat adjacent observations in the dataset as being one month apart or would it know how far apart they really are?

Many thanks in advance. :D

Actually, I've answered my own question by reconstructing the test in Excel and varying the methodology. EViews calculates quantities in equation 34.54 in the manual for all t in the sample, irrespective of whether t-q satisfies the if condition. In other words, it calculates the q-period behind VR, not the q-observation behind VR.