FactSet beta site v1.0, NOW LIVE

Dear All,

The beta site on which you can compare Compustat to FactSet fundamentals is now live. You can access it here: [size=3]https://beta.portfolio123.com[/size]

Once you sign in, click on your picture in the upper right-hand corner of your screen. You’ll see options for Data Vendor—Compustat or FactSet—and for PIT Method , which I’ll explain below.

If you choose Compustat as your Data Vendor, you should see pretty much what the regular P123 website offers. If you choose FactSet as your data vendor, all fundamental data will be what FactSet has provided us rather than what Compustat has provided us.

The stocks themselves will remain the same. The beta site does not represent universes created with Factset data (we do not expect the universes to differ much). Estimates are also still from CapitalIQ, as well as all remaining datasets. Little by little we’ll be adding other Factset datasets to the beta site.

RBICS is on the beta site, but for informational purposes only. In other words, you can look up a company’s RBICS code or the companies in an RBICS category, as explained here—https://www.portfolio123.com/mvnforum/viewthread_thread,12240—but we haven’t yet incorporated in ranking yet.

As for the PIT Method, here’s how it works.

For Compustat, “Use Prelim” reflects what we have now. Preliminary announcements are exposed when Compustat processes them. “Exclude Prelim” enables you to completely exclude preliminary announcements so that only data from complete statements from filing is exposed. NOTE: this is a brand new feature that did not exist before.

For FactSet, it’s a similar behavior from March 2020 onwards. The main difference for Factset lies when doing historical backtests since we had to make some compromises. The main problem lies in that Factset creates only one record for the as-reported period. This record is first populated with pre-announcement data (if the company pre-announces). When the SEC filing is processed, that record is overwritten, therefore Factset no longer has the original preliminary data.

The handling of preliminary data is the main problem with Factset in the context of a point in time database. NOTE: going forward, we at Portfolio123, are keeping the preliminary data so this will not be an issue for historical data starting around March 2020.

Luckily Factset does give us two key dates for past filings: when the preliminary data was first announced (“News Date”) and when it was filed with the SEC (“File Date”). When you choose “Use Prelim” we expose the final data on the News Date + one day. This means that some data points may suffer from look-ahead biases. Depending your strategy, this may not matter much since many important items are usually always reported in press releases.

If you want a more conservative backtest approach you can choose to “Exclude Prelims” which will expose the data on the File Date + one day. NOTE: this may still suffer from a minor look-ahead bias since Factset may have taken a few days , or even weeks , to process the filing. We hope to revisit this later when we learn more about typical delays.

Lastly, if Factset has neither File Date nor News Date we lag the data 30 days for interims and 45 days for annuals. This is very rare and happens mostly with older periods. We have a new command for these instances, LaggedStmt, which will read true if the statement has been lagged.

As you play with the site, I’m sure a variety of questions will come up. Please don’t hesitate to post them below and we’ll do our best to answer them.

Thank You

Ok, some Compustat/Factset sims have pretty similar results. One that doesn’t seems to be b/c of the Close(0)/EVPS factor. To get rid of any share count issues. I decided to look at MktCap/EV. In many cases, the results are similar but there are exceptions like HUM. Why is the EV so different between the two DBs?

HUM:
Compustat:
EV=37,792.4013
Mktcap=46,704.4013

FactSet:
EV=48,764.4822
MktCap=46,704.4822

Good catch! Compustat has HUM with over $15 billion in cash and equivalents, while FactSet has it with only $4 billion. It looks like we might have accidentally switched Cash and CashEquiv. We’ll fix it ASAP.

I noticed some differences calculating important ratios that depend on some important factors. For example AAPL’s EBITDA TTM and Cash & Equiv TTM differ in each database.

I am unable to run simulations on either Compustat or Factset.

For example, on https://beta.portfolio123.com/port_summary.jsp?portid=1596353 I am getting invalid benchmark error, no matter which benchmark I chose. For https://beta.portfolio123.com/port_summary.jsp?portid=1589247 I am getting invalid uid in PriceHistory.

I only tested these 2 so far.

I am also receiving the “invalid uid in PriceHistory” error for the following port https://beta.portfolio123.com/port_summary.jsp?portid=1606334

So basically run your backtest with prelim, then without prelim, and the actual performance answer is somewhere in the middle of those two.

Beta site does not recognize the “Portfolio” function for both Compustat and Factset.

NFLX’s EbitdaTTM is totally different between dbs.

Georg , Portfolio seems ok for me in the screener. Where do you see the problem ?

Yes, it is. We’re showing the correct numbers, they just calculate it differently. I wish that I could say more, but the S&P site doesn’t go into much detail about their calculation.

When I run with the Compustat Vendor on the beta site, I get very different results than running on the regular site (with either “use prelim” or “exclude prelim”). Reason?

This is now fixed. But it doesn’t help with HUM. FactSet has CashEquivQ as NA for HUM, as well as for hundreds of other stocks, most of them in the financial sector. So HUM’s EV, the way we calculate it from FactSet financials, is now even higher than it was last night.

It’s not that FactSet is wrong and Compustat is right. It’s that Humana’s balance sheet claims a huge number of securities as “investment securities” and therefore current assets, but those securities look to Paul and me (and to FactSet) like long-term securities, and therefore should not all be classified as cash and equivalents. NA is probably more correct than $10.4 billion (which is what Humana reports as short-term investments).

At any rate, it looks like we need to change our fallback formula for EV. If CashEquivQ is NA we’ll need to look at CashQ instead.

Marco, this simple sim does not work on the new platform, with Compustat or Factset data.

Buy: Eval( Ticker(“SPY”) & Portfolio(1451475),ticker(“SPY”),ticker(“ief”))
Sell: nodays>12

Attached are sims on current and new platform.



Thanks Yuval, I appreciate the update and will try a resim today.

Walter

There’s something wrong with WeeksToQ. Under the beta site, it’s returning NA for almost all stocks - and that’s with either Compustat or Factset selected. The non-beta site still returns proper (non-NA) values.

This will impact EV/EBITDA and other important ratios. Is there any solution ?

I use the TSX as a benchmark in the port sim but when I run the sim I get the following error:

Invalid benchmark 51425 for portfolio

Funny I’ve tried SPY as well and get a similar error.

We’re going to recalculate EV so that if CashEquivQ = NA, it falls back to CashQ. If both are N/A, EV will simply not reflect the reported cash that Compustat deducts.

This is going to mostly affect banks. And you should NEVER use EV for banks.

But it affects some other companies too. Take JNJ, for instance. Their most recent 10K lists $17.3 billion in “cash and cash equivalents.” FactSet gives this as N/A because JNJ does not tell us which of these are true cash equivalents and which are “current marketable securities,” i.e. “highly liquid investments with stated maturities of greater than three months from the date of purchase.” For all we know, that $17.3 billion could be all in high-yield junk bonds. Do you really want to subtract all these investments when calculating EV? It might be more prudent to consider the EV of JNJ as simply its market cap plus its debt.

The differences in the case of AAPL are negligible, less than 1%. There is necessarily going to be some differences in data between vendors.