Thanks for the bug reports. We are handling them all and will notify everyone as soon as these issues are fixed.
Please give us some concrete examples so that we can fix this. Thanks!
It seems that in the case of NFLX besides Ebitda, NetPlant and Payables are very different too.
Here are the details on âvery different results:â
Original Site Sim Results: 22% Annual Return, 1.38 Sharpe:
https://www.portfolio123.com/port_summary.jsp?portid=1606266
With Compustat âUse Prelimâ on Beta Site 10% Annual Return 0.74 Sharpe:
https://beta.portfolio123.com/port_summary.jsp?portid=1606405
With Compustat âExclude Prelimâ on Beta Site 11% Annual Return, 0.79 Sharpe:
Same Portid as above
With Factset âUse Prelimâ on Beta Site 7% Annual Return, 0.44 Sharpe:
Same Portid as above
With Factset âExclude Prelimâ on Beta Site 4% Annual Return, 0.19 Sharpe:
Same Portid as above
There are going to be differences looking at specific companies. If the EBITDA is substantially different then there pretty much have to be other differences with that company as well.
You can replace EBITDA with nearly any of the accounting lines. To paraphrase Mr. Rogers, everything fits together because itâs all one piece. If you change, say, Payables then something else has to be changed in order to get to the big Total Assets/Total Liabilities+Total Shareholders Equity line being equal.
And thatâs assuming no changes to some lines that I thought were sacrosanct. I looked and there are some companies with different Sales lines, for example. By the time that you get to net income everythingâs in the air, but I expected Sales to be more or less equal. (The differences are probably because of differing treatments of unusual variable costs, by the way.)
Happy to start some tests. My first 2 questions:
1/ SI%float is still available when choosing Factset provider. Where does it come from? Have you found a solution or is it still Compustat data that will disappear in the transition?
2/ My CEF sims donât change at all when switching to Factset. Do CEF fundamental data really come from Factset? (Yield, NAVDisc, NAVDiscAvg(), NAV%Chg12M etcâŚ)
Paul, I donât know why but probably these differences are making roughly a 10% underperformance in my rolling backtest.
SI%Float is not a fundamentals item so weâre still getting Compustat numbers for that. Ditto with the CEF data. The only things on the website that are different between Compustat and FactSet are the fundamental line items and things derived from those.
Any sign of these line items in the factset datastream?
SpcItems
ExpNonOp
Could you publish a complete list of line items from compustat and factset and how you are mapping p123 functions to each?
The corollary being, which p123 functions are we currently using that will map to nothing (no data for an line item) in the FactSet data?
If factset is missing a line item is the beta site mapping to the compustat line item? Donât do that (re: SI%Float @Yuval above), so we can see what we will have when compustat goes away.
Are subscribers now seeing similar results when running simulations on the existing p123 site vs. running simulations in the Compustat with preliminary data option on the Beta site? (Results from simulation run on the existing p123 site should be substantially the same as results from simulations in the Compustat with preliminary data option on the Beta site, correct?)
Additionally, have subscribers noticed unexpected results today when running simulations in the existing p123 site vs. simulation results run on identical models in the existing p123 site earlier in the week or last week?
Thank you.
Hugh
Noticed that FactSet BUD has DbtTotQ of $0. Thatâd be nice. Compustat has $100B. Though it bounces around oddly in Compustat, too.
Best,
Scrooge
I donât believe Howmet Aerospace HWM is in the SP500 and it doesnât have 42mm employees. FactSet beta is showing these.
Thanks for your answer Yuval.
Maybe you donât see a stockâs SI%float and a CEFâs NAV growth or discount as fundamentals, but they are not technical either. Is there a plan to get data about short interest and CEFs before Q3 or must we already throw away models in which they are used? The sooner we know it the better if we must go out at best of such trading strategies. Your migration and your communication should be driven by use cases and userâs risk, not database constraints.
I am also receiving the âinvalid uid in PriceHistoryâ error for the following port https://beta.portfolio123.com/port_summary.jsp?portid=1606334
Also getting the âInvalid uid in PriceHistoryâ on some existing Simulated Strategies on all four datasets, i.e. Compustat and Factset both with or without Prelim.
Did some tests:
- Created a new default simulation on the beta site called "Factest - New " â OK
- Changed an existing simulation to default simulation, same as 1., "Factest - Based on old " â ERROR
- Created a new simulation with import from 2. "Factest - Based on old - Copy " â ERROR
Any idea?
Just one data point from me, but sims for the main ranking system I use across Compustat, Factset with prelim and Factset excluding prelim are all fairly similar. In fact the Factset options actually have better backtests than Compustat. This is for a multi factor ranking system that is not particularly innovative, mostly a âgreatest hitsâ album of various academic papers.
Not sure what learning I should take from that other than a little bit of relief that my backtest results are not a function of me somehow having over optimized to some quirk of the Compustat data set.
I ran rollback backtesting in each data vendor and it seems that the greater differences come from before 2009.
To be very brief, there are two explanations for why results are different between the regular site and the beta site using Compustat and Use Prelims.
-
Preliminaries are treated differently in the beta site than on the regular site. Thatâs not going to change. That will affect results. Basically, we found certain things wrong with how P123 had been processing preliminary results, and those are being corrected on the beta site.
-
There are bugs, which we are working to fix.
Any sign of these line items in the factset datastream?
SpcItems
ExpNonOpCould you publish a complete list of line items from compustat and factset and how you are mapping p123 functions to each?
The corollary being, which p123 functions are we currently using that will map to nothing (no data for an line item) in the FactSet data?
If factset is missing a line item is the beta site mapping to the compustat line item? Donât do that (re: SI%Float @Yuval above), so we can see what we will have when compustat goes away.
FactSet gives a lot more NAs for SpcItems than Compustat does. I suggest adding IsNA to all your SpcItems rules. Thatâs what Iâm doing. Instead of SpcItemsTTM, use IsNA(SpcItemsTTM,0).
It looks like ExpNonOp is missing. Weâll fix that. Thatâs a bug.
Weâll be making a spreadsheet available so that users can see the Compustat to FactSet mapping.
The goal is to have NO current P123 functions that will map to nothing in the FactSet data. Whether that goal is achievable or not remains to be seen.
All fundamental line items are FactSet if you use FactSet on the beta site. SI%Float is not a fundamental line item, so the beta site defaults to Compustat, like it does with all non-fundamental line items. As additional FactSet data gets incorporated into the beta site, that will change.
Noticed that FactSet BUD has DbtTotQ of $0. Thatâd be nice. Compustat has $100B. Though it bounces around oddly in Compustat, too.
Best,
Scrooge
Check BUDâs FactSet DbtTotQ using the âExclude Prelimâ setting. That has the right number. Iâm not sure why it has zero in the âUse Prelimâ setting.
I donât believe Howmet Aerospace HWM is in the SP500 and it doesnât have 42mm employees. FactSet beta is showing these.
Somehow we accidentally added three zeros to all the empl numbers. Weâll fix that.