Magic Formula Data

[west]

I started using MFI with real money in 2007.  I've added a quick general market screen on whether to invest in large (1B+ market cap) or small market cap based on P/E multiples of the sp500 vs russell 2000.

 

Its basically been investing in all large caps since 2007...large cap only MFI has done very well...slightly better than Pabrai over that time frame (2007-2013). we'll see if that continues

 

For what it's worth, for investors who don't know how to do *real* financial analysis and qualitative analysis (99.99% of the "investors" out there, to make a number up) but still want to use the Magic Formula should stick with large cap selections.  There's too much potential for bombs to pop up in the small cap selections.  I mean, think about Herbalife.  It's pretty small as far as big companies go.  But as of today it wouldn't even qualify for the $5b+ market cap selection cutoff through the MF website.  And yet still look at how much press coverage it's getting for being a potential fraud.  With large caps, the big players are keeping companies honest.

 

This, plus, in my experience the large cap threshold selections have done as well if not better than the small cap threshold selections from the Magic Formula website.

[jb85]

possibly, though i'd argue the market is probably efficient enough so that, over the long term, even 100M+ company screen should do about as well as any other market cap screen.  this is of course assuming that you pick enough companies in your portfolio say 20+ companies

 

large cap vs small cap outperformance can go in long cycles...longer than 7 years...i think its more just luck/random that large caps happen to have outperformed.  if you look at the late 90s early 2000s small cap investing would been the way to go based on market segment P/E's

[west]

possibly, though i'd argue the market is probably efficient enough so that, over the long term, even 100M+ company screen should do about as well as any other market cap screen.  this is of course assuming that you pick enough companies in your portfolio say 20+ companies

 

While I agree to some degree (I might bump your threshold to $500m), if we're talking about mom and pop investors, who don't want to look at their portfolio and see a potential fraud in it, why risk it?

 

large cap vs small cap outperformance can go in long cycles...longer than 7 years...i think its more just luck/random that large caps happen to have outperformed.  if you look at the late 90s early 2000s small cap investing would been the way to go based on market segment P/E's

 

I agree here as well.  I'm pretty sure large caps outperforming over the last four years (for the MF selections) is not going to be persistent for every year in future going forward.  There's some time period stuff involved.

[west]

possibly, though i'd argue the market is probably efficient enough so that, over the long term, even 100M+ company screen should do about as well as any other market cap screen.  this is of course assuming that you pick enough companies in your portfolio say 20+ companies

 

While I agree to some degree (I might bump your threshold to $500m), if we're talking about mom and pop investors, who don't want to look at their portfolio and see a potential fraud in it, why risk it?

 

I should clarify.  When I say a potential fraud, I mean a basket of potential frauds, as was the case in late 2010 with the Chinese reverse mergers.  And I re-looked at the data, and some of them were above a $100m market cap.

[oddballstocks]

I wonder how much of Greenblatt's results are from his superior data?  There's an interview floating around where he mentions he spent $20m to hire analysts to recalculate and normalize all of the market's financial statements.

 

The website that anyone can access isn't driven by his database, his DB is proprietary to his fund. 

[cubsfan]

Yes, Greenblatt has talked about the adjustments they make to the financial statements and building the database with his team.

 

In addition, he made a point of discussing how poor the international databases were.

 

From what I remember - he didn't want to go into too much specifics and completely give away his edge.

 

I assumed plenty of the performance edge is due to his superior data.

[cobafdek]

Not sure if this paper has been cited here before, but here's Montier's work (from back in 2006, and reprinted in his Behavioral Investing book), which includes international stocks:

 

http://www.poslovni.hr/media/forum-user-upload/files/9a/9a5c2b2f55b461120e8d01095cf09a34.pdf

[topofeaturellc]

I suspect that if you implement a very diverse portfolio the data errors to the good and the bad balance out.

 

Its when you try to reduce the portfolio size that it matters.

 

 

[west]

Hey dudes and (allegedly) dudettes.  I have updated my Magic Formula data website, including making the design not plain painful to look at!

 

Please find it here:

 

www.dusthimer.net/MF-DATA/

 

Enjoy!

[cobafdek]

Hey dudes and (allegedly) dudettes.  I have updated my Magic Formula data website, including making the design not plain painful to look at!

 

Please find it here:

 

www.dusthimer.net/MF-DATA/

 

Enjoy!

 

Impressive work - thanks for posting!

 

One suggestion.  How about analyzing the same data but exclude nothing?  In other words, no dropouts; instead, impute an end-price-per-share of $0.  This assumption would be a worst-case scenario for that particular stock, and the resulting returns would be biased downwards.  It would be interesting to see those numbers compared to what you've done so far by excluding delistings, bankruptcies, etc.  If the return numbers are still good, it's all the more impressive for magicformula.

 

(Such parallel analysis is common in medical research.  For clinical trials, there is the Treatment group vs. the Control group.  During the course of the trial, some patients drop out from each group, or some inadvertently cross-over from one group to the other.  It is useful to see the results doing what is called an "Intention-to-Treat" analysis, and assume all trial participants remained with their initial designation.  Such an assumption is conservative, and may actually understate the true treatment effect.  However, if a treatment effect still remains, it's all the more impressive.)

[west]

That is an exercise I'll leave to the reader :).  It isn't hard to take the average return, multiply by the number of companies in the computation, and then divide by the total number of companies in the original selection set.

 

To do this myself would be to misrepresent the data, which isn't fair to the Magic Formula selections in the slightest.  Especially since there were far more buyouts than bankruptcies...

[peter1234]

Thanks for sharing! Lots of great stuff.

 

:)

[west]

Thanks for sharing! Lots of great stuff.

 

No problem!  I hope someone can get something useful out of it :)

[west]

I finally got access to the actual Compustat Point-in-Time database, not just the normal Compustat database, which I've had access to for about a month and a half now.  Interestingly the "point-in-time" feature only goes back to 1987:

 

https://www.capitaliq.com/home/what-we-offer/information-you-need/financials-valuation/compustat-point-in-time.aspx

 

In his Magic Formula book, Greenblatt says they went back to 1982 (if I remember correctly).  If this is the case, they may have used the normal Compustat database, which has lots of restated items in it, to come up with their results in the book.

 

Hrm...

[Scorps]

west,

 

From what I recall, he posted Magic Formula Results starting from 1988.

[west]

Whoops.  I dug through my giant pile of books and found my copy.  You're right.  He starts in 1988.

 

I should have double-checked that date before I posted.  For some reason I had 1982 in my head.

 

Thanks for catching that!

[jb85]

can you run some backtests for us?

 

have you been able to match greenblatts results?

 

here's a good thread on backing into greenblatts exact formula:

 

http://justadrone.blogspot.com/2011/03/calculation.html

[west]

can you run some backtests for us?

 

have you been able to match greenblatts results?

 

here's a good thread on backing into greenblatts exact formula:

 

http://justadrone.blogspot.com/2011/03/calculation.html

 

That's part of the goal :D

 

Unfortunately/fortunately I've got the database in raw format (versus using one of Compustat's good-to-go applications), so parsing it and understanding it is a little challenging.  Once I "get it", backtesting (including the Magic Formula) should be pretty easy to do.