Andreas,
If you want the truth you will read what FactSet has to say about its own data.
No need to believe an authority figure: Yuval, Marc or I guess you are reading what I have to say too. No need to take it on faith.
I can recommend what Francois said: it was short.
Get the facts on your own.
For those that have no need for a wall-of-words I will spoil the ending. FactSet says the data it provides to P123 now is not PIT.
Best,
Jim
well my point is, you can make PIT Data from non pit data.
Easy to implement, you only have to freeze the rows of the database t-1.
So if factset wants to post something in the past, you simply do not let, save it with t-0 and you have pit.
And from what I understand / assume that is a common way how P123 operates on a day to day update of the database
Obviously not possible or harder if changing to a new data provider (e.g. before August 2020).
My best agressive Models that I have in a live ports (not sims) using earnings estimates are up 40% since then (without earnings estimates around 10%), that might be luck, but from the fact that small caps went nowhere since may I give it a little weight.
So my question remains:
Also what I understood is, that p123 takes data from F. and makes it pit and it takes around 24 hours to update the DB → so its pit but lets call it “PIT +24h” (which I could live with, but a look ahead should be really avoided!)
Then maybe P123 should do that.
If you go up in the thread Philip has a suggestion for one way of doing that.
Quantopian (I do not mean Quantonomics) took another approach.
Yep. You just made all my points.
Jim
As I stated from what I know they do this. and always have done this. This is the whole point about p123.
If p123 does it with earnings estimate I am a bit confused but I am sure Yuval will bring the light 
Yuval has said loud and clear that NOT EVERYTHING is PIT 
“Exclude Preliminary Data” is what it says: and exclusion of certain types of data. Specifically, an exclusion of preliminary data as the name would suggest. It is not a lag which would probably have a different name.
Sincerely Andreas, how could it hurt to do a little more reading on your own about this?
Jim
Jim if your example using TALO is the only kind of look ahead bias (ie. one day maximum). Then it is extremely easy to fix. Just make a buy and sell rule that says you can only transact on Tuesday. I think its something like weekday = 3 I forget which number is Tuesday.
Philip,
That is an excellent idea. I have done this with previous models and will see if my sim becomes more similar to my port by doing that.
Like you said that is easy and I can send you the result in an email if you are still interested in some of the things about my sim/port.
Thank you.
Jim
Sure please try that trick and I can take a look at the results.
This is an excellent point, Jim, and well made. We’ll see what, if anything, we can do about this, and I will update you in a subsequent post on this thread.
Yuval,
Thank you very much. I hope you also saw Philip’s post above. That is a good idea that I can try (that anyone can try).
I am trying to think if his method has any limitations. Anyway, even if it does have some limitations (that I have not thought of) that is something that members can do on their own.
Best,
Jim
Well, one limitation is Monday holidays. On weeks with Monday holidays you’d want to rebalance on Wednesdays. Or maybe to be extra safe, you could always rebalance on Wednesdays. Anyway, I’ll get back to you all about the lookahead bias in estimates.
But estimates, except estimate revision counts, are updated daily. Right? What keeps the Record-date from being on Tuesday (or Monday, or …)
Walter
Walter,
I think Philip is talking about backtesting (with a sim). Backtesting is just once per week.
Live ports is updated daily (a great thing) as you say. I don’t think Philip would want us to do that with the port.
Jim
Thanks Jim,
That’s clears up my confusion. I easily forget the intricacies of the p123 environment.
Walter
All right then, all sorted out, thanks for the persistence Jim!
Andreas, Thank you. You proposed a good way to address PIT issues that I am sure Yuval is considering. I think everyone is on the same page. -Jim
In case anyone would like to see an anecdotal example of this.
I used weekday = 4 as Yuval suggested.
I think this may be pretty good. For sure, I would have to prove to myself that there is a problem with the sim now. I stopped at 8/24/21 because I withdrew cash from the port after that date which affected the port’s buying very significantly. But I have been adding and removing cash throughout. That and Force Positions into Universe should guarantee that there would be some difference between the sim and port.
First blush with (hopefully) conservative conclusions: probably no look-ahead bias theoretically as Philip pointed out and a believable sim.
Sim first then the port.
Thank you everyone! I honestly had no idea as to what was going on with my sim when I first posted in this thread.
Jim
One problem with my method (rebalancing on Tuesdays) is that any other non-estimate related data is needlessly lagged one day.
Jim those lines look pretty darn similar although still some subtle differences. At least now the sim isn’t blatantly better than the port.
We’re hoping to roll out a solution so that you won’t have to rebalance your sims on Tuesdays. Stay tuned.
