I think the point of their paper is that you get very different results with different intervals. They're pretty clear that they're neither trying to support nor trying to disprove market efficiency - just to show that a lot of results aren't very robust.
I haven't given the paper a thorough look just yet, but a couple things come to mind after my quick once-over:
1) Any real, exploitable predictability found by these guys has probably already disappeared (if not, I sure hope the paper's authors get filthy rich over the next few months).
2) Isn't this analysis just another way of teasing "information" out of noise, sort of like the Bible Code? I mean, if an interval of 4 days doesn't work for some range of dates, try 5. If that doesn't work, try 6... etc. Is there really an effect, here, or not?
Commenting by HaloScan
