i use stuff like NextFYEPSMean > CurFYEPS4WkAgo * 1.2 in some of my models
they all explode recently
do we get better estimation data lately or is this just the way it is?
regards
Andreas
Is that in backtests? Estimates are not PIT and they may have stopped taking snapshots.
no, in portfolios
estimate are not pit?
Continues to be a mystery…I asked on another thread to see if I could expect major differences between backtests and live portfolios using estimates and analyst recommendations with no insight.
https://www.portfolio123.com/mvnforum/viewthread_thread,11255
To quote Marco :[quote]
We are also moving away from snapshotting the data. We now completely rely on S&P for the historical data points which is more manageable.
[/quote]Presumably, estimates are provided by Capital IQ as it says on the home page of this site.
Capital IQ data seems to be (to quote Jim) “PIT for when the data was published, not for when it was made available for you to use for your rebalance”. See here
Thank you!
What is important to know is: is it PIT or is it not PIT.
There can be a time lag before it is in the database but then this time lag Needs to be also reflected in the PIT.
Example:
Release Estimate 1. January 2018
In the (p123) database 5. January 2018
PIT would mean 5. January, (and not 1. January, because on the 1. January I could not have used it realtime)
E.g. PIT = Needs to simulate, backtest the same as if you would use it real time.
Is this the case? By the way, everything else would just not be PIT! 
Regards
Andreas
There are a couple of different things involved here. Jim’s post linked to by Chipper6 is actually about fundamental data, and does not have anything to do with estimates. Compustat only provides the fundamentals, and is PIT. Capital IQ has its own fundamentals database as well (called premium financials), which, until very recently, only provided the filing date and not the date that the numbers were available to use in real-time. So the Capital IQ fundamentals were not point in time in the same sense as Compustat is, but P123 does not use this database. Compustat is now owned by Capital IQ, but historically there were two different databases, and they still are different and both maintained.
As for estimates, the story is different. Capital IQ has a database with estimates, and the other big database is Thomson Reuters I/B/E/S. As far as I know, P123 uses the CapIQ estimates, but I’m not sure. Capital IQ estimates have very precise timestamps, and they are generally correct. However, I do know for certain that Capital IQ sometimes retroactively changes the estimates. I don’t know why. It seems that the impact is generally not that big, but it’s certainly big enough to make backtests a little different when you run them again after a while.
I don’t actually have a answer to judgetrade original question that started this thread. But my answer to “are estimates PIT” would be generally yes, but not 100%. I also think that the occurrences of non-PIT estimate values are not significant enough to be really important.
Peter,
- Good to know. But the question is timestamped for what? Timestamp for when the info was public or when it was added to the database?
I have an open mind on this—if you have new information–but then why does Marco even bother to tell us he is moving away from snapshots? What is the difference? Why would he care or think that we would care?
Captial IQ never calls its estimates “PIT” they use “as is” a lot. What does that even mean? You cannot give me a link where they say it is PIT, I think. Timestamped, maybe. PIT….well let me see the link.
-
None of this matters if they are going to go back and change the data for reasons that we do not know anyway. In which case we can just call it less than optimal data and be done. No need for us to understand, I guess.
-
I’m not against the suggestion that it may not make much of a difference. Perhaps, Andreas can do a lot of work with a spreadsheet and figure it out for himself. Maybe he will figure it out maybe he won’t. Maybe is just the market. But he did ask nicely for the community’s help. And we are talking about someone’s retirement dreams, I think.
-
My impression of this data not being exactly PIT came from Marco: and he made quite clear, in his posts a while ago now, that they were going back and changing the database to make it reflect when the data became public. He can correct me if I did not understand.
This was before he moved to snapshots for some–unknown apparently—reason.
You have no understanding of why it is being changed but you are sure that it isn’t this particular reason? Can you give us any clues–at all–as to what is actually happening? If I there is a subtlety that can be made a little clearer to me that would be appreciated.
I have no complaints about the way the data is now. Transparency would be good but I can probably figure it out for myself. Perpetuation of what may be a an intentionally false or misleading impression by Capital IQ: not so good.
For now, I am unconvinced that it is even possible for Capital IQ to be more rigorous with the analyst data than they are with fundamental data. Analyst do have some obligations on their reporting to protect themselves–and the people who write their paychecks–from allegations of insider trading. I think those who use this data at P123 may be seeing the benefit of those requirements. But do the analysts report immediately to Capital IQ? I understand this is slightly away from the subject.
I think there is a lag and I may not know exactly why. But so far, I have seen no disagreement on one thing: They go back and change the data.
-Jim
Jim,
I think that the timestamp is when the estimate became “public”, ie. submitted by an analyst. By “precise timestamps” I meant that those timestamps include times with hours, minutes, and even seconds.
That does not mean there is no lag. There is a small lag (up to 8 hours) caused by the fact that the estimates are only made available 3 times per day. But that doesn’t really matter. The estimates are updated by P123 only once a week anyway.
My gut feeling is that most estimates have 0 lag, but that some are lagged. Maybe there are contractual obligations between the analyst firm and CapIQ. (The analyst firm first sells it for a higher price to direct clients perhaps). I don’t know.
Something like 90%+ of the historical estimates never change. Many of the ones that do change, change within 1 or 2 days. In some extreme cases, estimates are changed months or even years back. Though those could be data error corrections or something.
CapIQ is working on adding the exact timestamp that the estimate was available in the database as well. But you need to subscribe to the (very expensive) detailed estimates package for that. I doubt P123 will be able to provide that.
Peter,
Thank you for your comments. I think there may be no disagreement. That was my impression: “when it became public.”
Even if there is some disagreement, I like the data. If anyone has any ideas where to get better data I am open to suggestions. I will put in my $0.02 and say I am pretty sure Zacks would not be an improvement. For the longest time I downloaded tickers meeting various criteria into P123 (from Zacks) and tried massaging this data to get a new port that may supplement what I was doing on P123 (e.g., tickers with Zacks rank 1). I was unable to get anything of value.
We may be able to agree that the data is pretty good no matter what else we may say about it.
Anyway, your information and ideas are much appreciated.
-Jim
Everything pvdb said is right.
Our Estimates come from CapIq and we only subscribe to the aggregates. Aggregate estimates change frequently , sometimes multiple times a day, since it takes just one analysts to revise #'s to cause a change to aggregate. We turn this “continuous” time series of estimates into weekly data points by saving the latest value from CapIQ as of Sunday around 11PM for each week. The only way for past weekly values to change is if we rebuild them which has happened a few times this past year. Current values can change after each update at 10:30PM and 3:00AM.
Marco, thanks for the clarification. I’m glad to hear that you do snapshot the estimates data.
P.S. Why would you need to rebuild them?
So then live simulations and backtests should give nearly identical results since they are both using weekly data (at least when it comes to estimates).
On that note, if I use a sell rule such as “Sell when average recommendation drops”, the results should be the same on a weekly rebalance?
The concept of snapshots was introduced many , many years ago when we could not afford a point in time data. We now are going to rely on S&P for historical correct data. We still do snapshots today to make the loading of our simulation engines faster since we hold everything in the server memory.
Estimates are rebuilt for a variety or reasons. The latest one was when we added support for non-primary issues. Other times was because we found problems in our logic in some special cases. And yes, we would be picking up small differences.
Current data changes daily. So a rebalance on , for example, Wednesday cannot be replicated in a simulation. Only live rebalances on Monday should match simulations.
Thanks
Thank you Marco!!! Thanks for a great platform. And I do appreciate the ability to understand what is going on.
-Jim
thank you Marco!
So its confirmed that simulations using estimate data are not biased upwards due to data problems (ie. estimates available in the database before they were actually available in real life?)