Lo-Mackinlay variance ratio test

[Martijn Kreft]

Hi everyone,

I am using the Lomackinlay command to test for predictability in stock index returns (log returns), however I don't really trust the results. I do this for multiple indices, but one example is this below:

. lomackinlay Hongarije, robust
 
Lo-MacKinlay modified overlapping Variance Ratio statistic for Hongarije
[         1 -        522 ]

q         N         VR          R_s       p>|z|
--------------------------------------------------
2        506       0.484      -5.7851    0.0000
4        506       0.232      -5.1783    0.0000
8        506       0.130      -4.2371    0.0000
16       506       0.074      -3.4151    0.0006

Test statistics robust to heteroskedasticity

The results are similar for the other indices. However, I find the p-values not really believable, especially because another research (which does the same but for a different time period) does only get very few statistically significant results. Am I missing something? What could cause these results?

Thanks for your time and help!

[Martijn Kreft]

BTW, the VR itself seems to be the cause, because it is pretty extreme in my opinion. But I don't know why this happened