Related to Andreas’s questions:
Marco’s initial post in this thread noted that P123 will maintain PIT preliminary data from Factset from March of 2020 on. So if I’m using the Factset engine, is there a difference between “Use Prelim” and “Exclude Prelim” for dates after March 2020? More specifically, does using or excluding prelims affect live recommendations, screens, or ranks after March?
You won’t have to do anything.
By June 26.
I don’t understand this.
Yes, of course.
The fundamental charts are point in time and the screen results reflect Compustat’s restatements, which are applied in a point-in-time manner. In other words, if you run the same screen with an “as of” date a year ago or so you’ll see numbers that are more in line with the ones you see on the fundamental chart–and the numbers you see from FactSet.
yes of course. prelim data will be ignored with live systems if you choose to exclude them.
It’s another tool in the p123 arsenal. This might be desirable depending on your strategy since prelim data can have a lot of NA’s , trigger the fallback behavior, produce temporary erroneous values , etc.
Example most academic paper are done with 10K data and only final ones .
Factset interim restatements is an ongoing project for them (we noticed some back-filling so we asked). They also seem to be missing an effective date for the data. But this might not be so hard to overcome with some spot check tricks. In addition , after 2 years pass since the restated interim period, it’s usefulness greatly diminishes since ratios like TTM growth rate are no longer affected.
It affects a very small set of companies. We’ll get back to it soon. For now we are ignoring this interim restated data until we feel confident on how to use it to make Factset better and truer.
Hi Yuval/Marco,P123,
I’m confused. I’m struggling with Healthcare providers (HUM, ANTM, etc) and pseudo insurance (FAF,FNF). If I look at CashEquivq (Factset) it shows for HUM as 17,158.00 and cashq as 6,054.00. In a previous post you talked about cashq being a fallback for cashequivq for HUM in EV. However, if I look at the snapshot it shows cashequiv as N/A when using Factset.
Can you tell me the formula for EV? I’m close with
max(5,price*sharesfdq + dbttotq - isna(cashequivq,cashq) + isna(noncontrolintq,0) + isna(pfdequityq,0))
Thanks,
David
If you look on the Trello board you’ll see that this bug has not yet been fixed. It’s in the “in progress” column. See Trello . . .
Once it is fixed, CashEquiv will still be N/A for some stocks. Banks, insurance companies, and health-care providers may be among them. In general, don’t use EV for those companies. Debt and cash are treated completely differently by those companies and EV is a very misleading measure of what they’re worth. Instead, use market cap.
Your formula looks correct for non-financial companies.
Hi Yuval,
first of all thank you for your answers and the great work of p123, must be a hercules projdect!
My last question, that I did not state the right way:
Will the performance of the simulation (e.g. how fast they are) be in the range of the old performance engine.
Explainaition: I experience with ‘“current” and “factsheet” and “PIT Prelimary” on’ performance of the simulations, that is about 5 times slower then
when I simulate with legacy.
Thank you
Best Regards
Andreas
Is function EV “Enterpise Value” working correctly with FactSet?
I am using a simple 2-factor ranking system with
(Eval(EBITTTM<0,NA, EBITTTM)) /EV
and
(Eval(EBITTTM<0,NA, EBITTTM)) / ( WorkCapTTM + AstNonCurOtherTTM)
A sim using Current Server and Compustat shows from 2007 to now an annualized return = 42.13%
With FactSet annualized return = 31.79%
That’s a huge difference. Perhaps it’s not EV, but some of the other functions which causes the discrepancy.
https://www.portfolio123.com/port_summary.jsp?portid=1610828
Yes, that will be fixed.
Currently EV gives lots of NAs. That will be changed soon. Also, EBIT is quite different between FactSet and Compustat. I’ve been unable to ascertain precisely why. Lastly, AstNonCurOther is measured differently by FactSet and Compustat.
Thanks for the option to exclude Prelim data. After using this my system appears to be much smoother and better, probably because i use Formula Weight as sizing method. The weights are based on fundamentals and every change in the data triggers buy/sell orders and therefore additional costs. Took me a while to understand this, but with Prelim data active the fluctuations in the data are much bigger than i thought. So please keep this option, it keeps the data more stable.
Thanks Yuval.
Good to know what causes the difference.
I have other sims that perform better with FactSet for some unknown reason.
Yuval and others,
Talking about EV…
While overalll the P123 Greenblatt ranker works a lot less well (2% pa) on the subsets of the SP500 or R1000 that I use as separate universes, I wanted to share that out of the 20-year period since 01 Jan 1999, the bulk of the difference is concentrated on the period Nov 2016 to April 2018 included.
It probably accounts for 90% of the difference. This is true whether I pick 10, 20 or 25 stocks (I have not checked for more yet)
Most (but not all) of my universes are mutually exclusive so it would tend to infer a systemic issue over this period rather than due to a handful of same stocks picked up across all systems.
I can not yet explain why and will look into it further but wanted to share now in case
Thank you
Jerome
Agree with this. My Greenblatt model is significantly worse. I added a few rules that filters for capital efficiency, eliminates some underperforming industries, and allows for some insurance stocks, but otherwise it’s identical to the Magic Formula. Bottom line, results since 1999 are worse by 8% per annum when backtesting the new FactSet Data. A lot of this has to do with factors like EV and GMgn% not working. I’d really hate to lose this excess return.
I have not studied it in depth at all, but in late October 2016 financial stocks went zooming up compared to other sectors, and they stayed up, way up, until April/May 2018.
I suggest you run your systems excluding financial stocks (GICS(40) or Sector=Financial) for both Compustat and FactSet and see if there’s still a marked performance difference.
If there isn’t, then the difference will lie in the way Compustat and FactSet calculate data for financial companies.
If the difference remains large, then we have to hunt elsewhere for an explanation.
I think you are onto something with EV. Alot of my best strategies are using “EV” as part of the calculations and it would explain why my strategies sucks so much on Factset.
Yuval,
Good point.
With financials Compustat = 17.9% AR
W/O financials Compustat = 15.6% AR
W/O financials Factset = 15.0% AR
However, for another one of my sims there is still a big gap even excl financials i.e. 19.32%AR vs 17.44%AR
But the difference really comes from the H2 2017 period so could be due to a few “lucky” buys on one hand
Can you give us an ETA as for when EV will be fixed?
This might well help even out the Greenblatt performances across both datasets.
Thank you,
Jerome
I think I have found the reason for the different results between FactSet and Compustat data.
The problem is that when you run a sim with FactSet it defaults to the All Stocks - USA universe even if your setting is All Fundamental - USA. This also explains the longer run time which Andreas reported.
If you choose All Stocks as your universe then FacSet and Compustat will give same results. I tested this with a sim using Stock Factor and ignoring ranking by selecting number of positions large enough that all stocks were selected as per Stock Factor input.
Yuval you have a bug in the algo - please fix and let us know when done.
Um, you’re wrong. I ran a simulation on the All Stocks universe in FactSet and reran it on the All Fundamentals universe and checked which stocks it picked. On the All Stocks universe it picked plenty of stocks that weren’t in All Fundamentals. On the All Fundamentals universe it didn’t.
The longer run time is because the legacy engine is powered by more servers than the current engine.
The All Stocks universe is the same for FactSet and Compustat. The All Fundamentals universes are totally different. See https://docs.google.com/document/d/17Q0cVnGdGwKGEXFZOuwf297itKdqUC_KGN2UHcMTdYg/edit?usp=sharing part 1.