Thank you, Marco and p123.
I appreciate the rollback (and effort) but think there is at least one additional cause of quirky but real model degradation which does not fall into either of the “We found a longstanding error while we were making other changes and corrected it” or “Your model sucks and should have been abandoned years ago” buckets.
I could of course be wrong, but I can’t find any mention of what follows in the recent threads and I think it is a very good example of unexpected consequences created by hastily made changes even when they sound like no-brainers.
And what is this hidden change?
In theory, it seems that removing companies without fundamental data makes sense. Removing these useless companies is a primary difference between the new universes and the old, correct? It is also one change p123 has not rollbacked even temporarily?
Here is the problem that creates…
When using a rule such as Frank(“p123 function”)>20, that rule is now cutting out a different slice of companies than would have been the case if the now-removed-from-the-universe missing data securities were still sitting in the middle of that ranking.
Frank is a fabulous tool. My models that use it the most have been hit hardest by the changes.
There may be other related consequences to the removal of these companies I’m not imagining. By the way, I think this is one reason p123 is seeing such a difference in subscriber reactions. If you’re not using rule-based ranking functions, you’re probably a lot less unmoored by the recent changes.
Would it be possible to recalibrate? Absolutely, but that takes time and some stability regarding other changes being considered on a parallel basis.
I have two additional considerations:
-
For those who are using CompuStat data, S&P and MSCI are supposed to announce what could be significant changes to the GICS classifications any day now. In fact, the announcement was scheduled last month but delayed. Possible changes to GICS were published last fall for comment. Is p123 ready in this regard? I mention this in part because it seems crazy to begin a major overhaul based on p123’s recent changes and then get hit in the middle of that work with a need to do another overhaul based on GICS revisions.
-
Sims seem to be running much slower since p123’s recent changes. As mentioned in a previous post, it’s a subjective observation but also a really worry. Even if multiyear sims ran as quickly as daily production screens, it would still take a lot of time to recalibrate and especially when relying on multiple models. With sims running as slowly as they seem to be, it feels like retuning anything is going to take forever.
In the meantime, it’s clear that even the remaining non-rollbacked changes are causing an expected (and may I add unnecessary) loss of alpha.
Help!
Thanks.
Hugh