Index | Recent Threads | Who's Online | Search

Posts: 373    Pages: 38    Prev 26 27 28 29 30 31 32 33 34 35 Next
Last Post
This topic has been viewed 17992 times and has 372 replies
yoda34
Re: FactSet beta site v1.0, NOW LIVE

Curious if anyone has solved the issue with the Greenblatt style. Even excluding financials (which I have always done) the Greenblatt style is under performing significantly from where it was with the old data provider. Has the EV issue been fixed? If so are there any other thoughts on why the Greenblatt has changed so much downwards?

Jun 27, 2020 2:47:15 PM       
mm123
Re: FactSet beta site v1.0, NOW LIVE

Yes. A source of frustration for many of us. Hoping P123 staff can revisit this. My personal Greenblatt model was returning >35% and now it looks like theses results will be cut in half.

Jun 27, 2020 2:58:11 PM       
andy_p123
Re: FactSet beta site v1.0, NOW LIVE

Mine as well - it looks like the CAGAR is a just a little more than 50% reduced.

I'm not technical person, it's hard for me to try and identify why - any help or comments from the more advanced uses / staff would be very much appreciated.

Jun 27, 2020 3:01:23 PM       
andy_p123
Re: FactSet beta site v1.0, NOW LIVE

So I spent the afternoon looking at my greenblatt ports and trying to find differences in the variables etc. I have been running a greenblatt style for many years and I have captured a lot of my old data for specific trades. Based on what I've seen here's what I think, the below is applicable to my simulations and doesn't prove it's the same for other folks but:

- there is about a 40% - 60% overlap on the results between the compustat and factset on greenblatt style rankings
- The differences in EBIT, EBITDA, etc from quarterly to TTM seem to be fairly limited
- The majority of the difference seems to be driven from the EV value. In some cases the EV is much higher than the old data set, in others much lower. If you're using EV as a value indicator this will make some names seem more or less expensive and move their positioning in the ranking, some anecdotal examples:
- Jan 1, 2017 Compustat shows the EV of Humana as 18.8 and Factset has it at 27.7
- Same date, Staples (which wasn't even in my top 50 in the old dataset) is now ranked #2 as the Enterprise Value has dropped significantly

Again, anecdotally, but I have another formula that is also showing quite a difference in value and based on my understanding of the calculation for EV, the overlaps between my Greenblatt calculations and my other formula are in Cash&Cash Equiv and Preferred Equity - these are for Non Financial Sector and Non Utility Sector names.

Can anyone that has more insight into the data speak to any changes to the EV calculation or underlying data that could be driving the above?

Thanks!

Jun 27, 2020 9:10:19 PM       
RTNL
Re: FactSet beta site v1.0, NOW LIVE

Are folks seeing an increase in turnover in their models too?

Jun 27, 2020 9:29:14 PM       
yuvaltaylor
Re: FactSet beta site v1.0, NOW LIVE

So I spent the afternoon looking at my greenblatt ports and trying to find differences in the variables etc. I have been running a greenblatt style for many years and I have captured a lot of my old data for specific trades. Based on what I've seen here's what I think, the below is applicable to my simulations and doesn't prove it's the same for other folks but:

- there is about a 40% - 60% overlap on the results between the compustat and factset on greenblatt style rankings
- The differences in EBIT, EBITDA, etc from quarterly to TTM seem to be fairly limited
- The majority of the difference seems to be driven from the EV value. In some cases the EV is much higher than the old data set, in others much lower. If you're using EV as a value indicator this will make some names seem more or less expensive and move their positioning in the ranking, some anecdotal examples:
- Jan 1, 2017 Compustat shows the EV of Humana as 18.8 and Factset has it at 27.7
- Same date, Staples (which wasn't even in my top 50 in the old dataset) is now ranked #2 as the Enterprise Value has dropped significantly

Again, anecdotally, but I have another formula that is also showing quite a difference in value and based on my understanding of the calculation for EV, the overlaps between my Greenblatt calculations and my other formula are in Cash&Cash Equiv and Preferred Equity - these are for Non Financial Sector and Non Utility Sector names.

Can anyone that has more insight into the data speak to any changes to the EV calculation or underlying data that could be driving the above?

Thanks!


This is quite interesting. I don't know if I can shed much light here, but I'll offer a few remarks.
1. Compustat and FactSet take different approaches to CashEquivQ. Take ADM, for example. Compustat reports its CashEquivQ as $9.832 billion while FactSet reports its CashEquivQ as $4.734 billion. If you look at the 10-Q, ADM reports $4.734 billion in cash and equivalents and another $5.098 billion in segregated cash and investments. FactSet does not include the segregated cash and investments as cash equivalents and Compustat does. You can decide for yourself who is doing the right thing: ADM reports that these funds "represent deposits received from customers of the Company’s registered futures commission merchant and commodity brokerage services, cash margins and securities pledged to commodity exchange clearinghouses, and cash pledged as security under certain insurance arrangements. Segregated cash and investments also include restricted cash collateral for the various insurance programs of the Company’s captive insurance business. To the degree these segregated balances are comprised of cash and cash equivalents, they are considered restricted cash and cash equivalents on the statement of cash flows." You can see there's some room for discussion.
2. I created a Greenblatt ranking system and did some rank performance tests on both the Compustat and FactSet universes. Here's the ranking system: https://www.portfolio123.com/app/ranking-system/370334 and here's the universe that goes with it: https://www.portfolio123.com/app/universe/summary/243659 . In a fifteen-year backtest, rebalancing monthly, the top ventile (the top 5%) earned 7.8% annually with Compustat and 8.8% annually with FactSet. In a ten-year backtest, the top ventile earned 7.5% with Compustat and 7.4% with FactSet. In a five-year backtest, the top ventile earned -2.1% with Compustat and -0.7% with FactSet. Now I did not spend a lot of time finessing this system. The formula for return on capital is the same one I've been using since 2016, which is when I read his book, and his formula for EBIT/EV is standard. The universe rules seem to make sense considering that Greenblatt wasn't investing in microcaps and he excluded stocks in the financial sector. Maybe your parameters are different--I don't know. But I'm not seeing a huge difference in performance here.

Yuval Taylor
Product Manager, Portfolio123
invest(igations)
Any opinions or recommendations in this message are not opinions or recommendations of Portfolio123 Securities LLC.

Jun 27, 2020 11:34:21 PM       
Chipper6
Re: FactSet beta site v1.0, NOW LIVE

Maybe your parameters are different--I don't know. But I'm not seeing a huge difference in performance here.
Unfortunately, some of my EV derived factors did degrade significantly. I understand that it's difficult for you to pinpoint the problem with my private system without me divulging more details, but I can't divulge more, but rest assured that it really is a big deal. I was getting 5-10% oos alpha from a live system which was almost exclusively based on a single EV derived factor, but that has alpha evaporated and the systems have become unusable with FactSet. I don't know why.

One clue is that when I compare the the line items in FactSet with the 10-Ks and 10-Qs I am finding significant differences. I never found any differences with Reuters and I rarely found differences with Compustat. When I have more time I hope to dig deeper into these differences.

1. Compustat and FactSet take different approaches to CashEquivQ. Take ADM, for example. Compustat reports its CashEquivQ as $9.832 billion while FactSet reports its CashEquivQ as $4.734 billion. If you look at the 10-Q, ADM reports $4.734 billion in cash and equivalents and another $5.098 billion in segregated cash and investments. FactSet does not include the segregated cash and investments as cash equivalents and Compustat does. You can decide for yourself who is doing the right thing: ADM reports that these funds "represent deposits received from customers of the Company’s registered futures commission merchant and commodity brokerage services, cash margins and securities pledged to commodity exchange clearinghouses, and cash pledged as security under certain insurance arrangements. Segregated cash and investments also include restricted cash collateral for the various insurance programs of the Company’s captive insurance business. To the degree these segregated balances are comprised of cash and cash equivalents, they are considered restricted cash and cash equivalents on the statement of cash flows." You can see there's some room for discussion.

I don't know if you should count them in Cash & Equiv, but these deposits and loans should be neutral for EV. That's because either they should count both as cash and as debt, or they should not count for either. You can't exclude it from cash but count it as debt. This may be something to look into.

Jun 28, 2020 12:20:33 AM       
judgetrade
Re: FactSet beta site v1.0, NOW LIVE

I am basically back to the performance I had with compustat.
Speed is back where it was, very good!

What I have learned from the new data:

- I basically took my best ranking systems and back tested them on the new data

- My best ranking system did perform also with the new data (multi factor: earnings estimations growth, stock momentum, industry momentum,
Quality, institutional ownership momentum, value (EV, EV / FCF, EV / projected earnings), value (FCFQ / AstTot))

- the best results I got was through changing the weights of the different factors, so I did not change any factors, I just weighted them
differently. BUT - and that is good news - not dramatically, only a bit.
All in all maybe 10 hours of work.

With that method, I got basically the same backtest performance as with the old data.

My conclusion: multi factor systems, as lined out several times from Yuval is a good way to work with the new data.

I am looking forward to finding new ways to leverage the new data, maybe insider information and shorting data and of cause international
stuff gives some great edges.

THANK YOU VERY MUCH P123 Team, VERY, VERY WELL DONE, I am sure this was a tough time!!!

I will have a look at the greenblat systems and try to figure out the performance differences.

Best Regards

Andreas

Jun 28, 2020 1:05:01 AM       
Edit 1 times, last edit by judgetrade at Jun 28, 2020 1:09:20 AM
selti
Re: FactSet beta site v1.0, NOW LIVE

Hello,
I, for one, would have appreciated to know in advance the exact date (once known) of the transition from Compustat to Factset data instead of just noticing it by luck 2 days afterward ! ;( Imho someone should have sent us an email about it before (the last May's email didn't provide a date. I acknowledge that I had in mind the "end of June 2020") !
Regards

Jun 28, 2020 5:29:08 AM       
andy_p123
Re: FactSet beta site v1.0, NOW LIVE

Thanks Yuval - I appreciate the response, unfortunately i think many of use that us EV in our simulations have seen fairly large differences, almost always in a reduction in alpha.

One thing that would be very helpful for me (and possibly others) would be a way to use the data from Factset to recreate a close approximation to the EV value we had from the compustat data. The EV calculation isn't that hard only only has a couple of elements (mrkt cap, debt, preferred equity, minority int, and cash&equiv). If someone on the P123 team that can see both sets of data could look at each element and help identify the differences between compustat and factset then it seems reasonable that we could use the FactSet data to get back to a similar EV value as what we had before (for example you mentioned some differences in cash accounting - it seems like if Factset has not included some things could we add those back in etc)

It would have to be someone who can see the differences in the data and help us construct the equivalent EV formula. Is this something that P123 could help us with?

Jun 28, 2020 12:59:38 PM       
Edit 1 times, last edit by andy_p123 at Jun 28, 2020 1:00:06 PM
Posts: 373    Pages: 38    Prev 26 27 28 29 30 31 32 33 34 35 Next
 Last Post