Dumb Quant - EBITDA/EV

[west]

Before I post this, I want to mention (yet again) that I'm not a quant value investor!  I like to research areas where the market has been "rich" in the past so I can hunt there in the future.  Or at least understand areas that might be good in the future based on prior "good industries to hunt in" attributes in the past.

 

(Kraven describes it as making sure you're fishing in the right pond.)

 

With that out of the way...

 

Attached is a dumb EBITDA-to-EV sort returns for the US market between 2002-2007.  The annual selections were picked at the closing date for the prior year, and then sold at the end of the current year.  So, for 2002 for example, the selections were picked on 12/31/2001 and then sold on 12/31/2002.

 

I did this research using Compustat's database so (in theory at least) they are point-in-time and include companies that have subsequently delisted.  Suffice it to say, the returns using this dumb model are "satisfactory"

 

I did the sort for all market cap companies, and for companies for market caps above $500m.

 

Attached is also a screenshot for those who don't have Excel.

 

Further disclaimers (such as many of the super nano/pico cap companies might not have had the liquidity to actually buy them) and specifications of my model will be included in a future post.  (Which is hopefully coming soon...)

 

UPDATE - RhubarbXIV found a bug in my spreadsheet.  The spreadsheet and screenshot attached to this post have been changed from the original ones.

EBITDA-to-EV-Sort.2002-2007.for-CoBF_FIXED.xlsx

[wknecht]

Thanks, interesting. Certainly makes small caps look good. Why not roll this through 2008 and 2009? You're only looking for when markets are rich?

[RhubarbXIV]

Check your 2007 > 500m column.

[west]

Check your 2007 > 500m column.

 

What about it?  On initial glance it looks fine to me...

 

(That doesn't mean it's necessarily bug free though...)

 

[RhubarbXIV]

Check your 2007 > 500m column.

 

What about it?  On initial glance it looks fine to me...

 

(That doesn't mean it's necessarily bug free though...)

 

It's the same as your all cap column. Same for the spreadsheet.

[west]

Check your 2007 > 500m column.

 

What about it?  On initial glance it looks fine to me...

 

(That doesn't mean it's necessarily bug free though...)

 

Whoops.  You're right.  I forgot to do the > $500m sort.

 

I'll repost an updated spreadsheet ASAP.

 

Great catch!

[west]

Thanks, interesting. Certainly makes small caps look good. Why not roll this through 2008 and 2009? You're only looking for when markets are rich?

 

Because my code is awful (or at least R isn't designed to handle data the way I'm asking it to), it takes about two-thirds of a day to run per year.  I actually spun up some VMs on Digital Ocean to do as many years as I could at once to get the data you see.  However, they cost money to run them...

 

I'll probably be processing the rest of the years up to last year on my machine (which will take a few days).  Then I'll update the spreadsheet.

 

[west]

Check your 2007 > 500m column.

 

What about it?  On initial glance it looks fine to me...

 

(That doesn't mean it's necessarily bug free though...)

 

Fixed version (and screenshot) attached.

 

The fix makes quite a bit of a difference!

Whoops.  You're right.  I forgot to do the > $500m sort.

 

I'll repost an updated spreadsheet ASAP.

 

Great catch!

 

New (fixed) versions attached to this post (and the first one).

EBITDA-to-EV-Sort.2002-2007.for-CoBF_FIXED.xlsx

[rukawa]

How did you get access to compustat?

[west]

How did you get access to compustat?

 

A more awesome member of the board helped me get access.  That's all I can say :D

[opihiman2]

That's cool.  I would rather see CAGR #'s instead.  It looks like the better larger cap strategies had a CAGR of around 17%.  That's not bad.  But, I've seen better screens from AAII. 

[west]

That's cool.  I would rather see CAGR #'s instead.  It looks like the better larger cap strategies had a CAGR of around 17%.  That's not bad.  But, I've seen better screens from AAII.

 

Cool.  They may have better ones.  I mean, honestly, this screen is dumb as a rock.

 

My idea with running this back-test is I just wanted to see myself that the strategies actually work.  I always hear "strategy X outperforms the market by N basis points over a time period of three years" or "The Magic Formula has outperformed the market by a million basis points on average since 1982" or statements of the like.  However, for better or for worse (probably for better), I don't really trust people when it comes to money.  If somebody says their strategy outperforms, I want to see the raw data.  Not summary statistics.  Now that I've got access to compustat I can finally do that :D.  And hopefully with me posting this spreadsheet here, with actual picks instead of just summary data, other people can see this as well (at least for the dumb EBITDA/EV screen).

 

[opihiman2]

Wow, I just got this email from AAII:

 

A member recently asked me if a screening strategy with fewer criteria performs better than one with many criteria. As irony would have it, a few days later after I was asked this question, Wesley Gray and his colleagues at Alpha Architect published a paper on SSRN comparing several of the value-oriented AAII Stock Screens to a simple valuation model. The study’s results are not an apples-to-apples comparison to the way we track the performance of the screens (I’ll discuss the differences momentarily), but it did find that only our Piotroski High-F Score screen fared as well as a screen that simply seeks non-financial stocks with low ratios of EBITDA (earnings before interest, taxes, depreciation and amortization) to TEV (total enterprise value).

 

Valuation is among the biggest drivers of stock returns. A strategy solely focused on low valuations will have good returns if it identifies enough stocks.

 

The challenge with any strategy is making it investable. It is quite common for an analysis of indicators to divide the results into deciles, or 10 evenly split groups ranked from lowest to highest. Even if the universe of stocks studied for the analysis is narrowed in some fashion, each decile may still contain far more stocks than the average individual investor is willing to hold or can cost-effectively hold. (In Gray’s study, the EBITDA/TEV screen identified an average of 96 stocks.) There is also a behavioral aspect to consider: How willing are you to hold stocks that are otherwise unattractive?

 

Very coincidental and relevant to this post.  The link to the whole article is here:

 

http://www.aaii.com/investor-update?a=update11614

 

Super interesting.  So, the study used the CRSP and CompuStat database.  They also footnoted this:

 

"Because our results are focused on a different universe of securities, the evidence we present can sometimes differ drastically from the results presented on the AAII website."  Which I somewhat concluded as well given the security exchanges they were looking at for equities.  Anyways, their EBITDA/(T)EV screen came up with 90+ equities on average, and its performance from 1963 till 2013 is around 16.52% CAGR.  The only screen that outperformed it was the Pitroski F Score with slightly better percentages.  From 1997 to 2013, a really good period to look at since, as we know, those were some challenging times with two major booms and two major busts, the CAGR of EBITDA/EV was around 15.4%.  It's not quite fair to compare to the S&P, since both screens, I believe, exclude certain sectors such as the financials.  So, maybe financials just suck and if you exclude it from the S&P, it would drastically outperform.  Without a back test, we won't know. 

 

For Magic Formula, they claim that it actually underperformed the S&P500 over the really long period.  I don't know.  I've seen SO much different benchmark #'s for the MF.  But, consistently, I read that people are benchmarking this that with a lot of leverage.  Like, 175% long, 80% short.  That's insane.  I read another article that showed in a certain time period, MF barely outperformed S&P, maybe by 2%, with huge leverage.  I think 130% long, 80% short.  That's retarded. 

 

Anyways, enough rambling.  Hope that was informative. 

 

[west]

Thanks for the link!

 

So... The EV/EBITDA selection (or EBITDA/EV selection in my case... inverting it makes it easier to sort) may not have been too coincidental ;)  I heard about Gray's statement that a simple EV/EBITDA sort out performs the Magic Formula.  So, due to that (and partially due to my laziness...  an EV/EBITDA sort is ridiculously easy to code), I figured that's what I should test when I got access to compustat.

 

Testing the Magic Formula side of things, for comparison, might take a little longer to do though...

[opihiman2]

By the way, when you did your back tests, how did you handle the sell and buy transactions?  Did you take transaction costs into account?  What about short term/long term taxes?  And, finally, did you rebalance the portfolio?  I think those are all important to consider when doing back testing vs indexing benchmarks. 

[west]

By the way, when you did your back tests, how did you handle the sell and buy transactions?  Did you take transaction costs into account?  What about short term/long term taxes?  And, finally, did you rebalance the portfolio?  I think those are all important to consider when doing back testing vs indexing benchmarks.

 

Transaction costs were not considered.  Luckily, at least with IB and non-penny stock stocks, this is close to being true for me :)

 

I didn't consider taxes as well.  However, I did the study over one year periods exactly so that in theory people could take the Greenblatt approach to taxes, i.e., sell the losers one day before one year to get short-term capital losses on them, and sell the winners one day after one year to get long-term capital gains on them.

 

On rebalancing: I just assume that at the end of each year the positions were all liquidated to cash and put into whatever the top picks were at the current time.

[west]

I should also note that, again, this "study" was very "dumb" (perhaps "unsophisticated" is a better word), and was not a rigorous as (in theory at least) official academic studies.  Any conclusions, beyond what the raw data reports, are left to the reader to make ;)

 

[topofeaturellc]

one year returns for a strategy like this are really shorting the actual results.  A longer-holding period is required.

 

That said, absolutely naive quant value strategies can underperform for years at a time.  Levering them seems really really silly.

[cobafdek]

west, this work of yours (and your posting of the results), is much appreciated.

 

I have a general question about these databases upon which quant research is performed.  I'm assuming, in the case of studies examining EV/EBITDA, that EV is defined as MV + Preferred + Debt - Cash.  Have any of the well-known reputable researchers actually gone to the 10Ks to check the actual numbers?

 

I'm sure you've had the experience of finding a stock on, say, Yahoo Finance or Morningstar, with an EV/EBITDA < 2, but when you simply look at the actual 10K, you find that you need to make obvious adjustments so that the adjusted multiple is higher, such as > 6.  Such a large adjustment is possible, for instance, in the case of accounting for minority interest.

 

In other words, running an EV/EBITDA screen on any database will have some percentage of "errors," or false positives, and it could be a very large percentage (possibly > 25%).

 

This concern of mine does not detract from these research results, but, on the contrary, I think it might actually strengthen them.  That is, an EV/EBITDA screen, that identified a group of outperformers, but contains a sizable number of such false positives, could be called a "Really Dumb" quant screen THAT STILL OUTPERFORMS, whereas the screen that could be culled into containing only the correctly adjusted numbers (the true positives) is merely a "Dumb" screen.

 

I haven't looked at the famous quant papers yet, with the purpose of answering my question, and I haven't yet read Carlisle and Gray's Quantitative Value.  Do you, or does anyone, know whether any researchers have performed the laborious task of checking the database numbers with the simple 10k numbers?  Or are these databases already adjusted?  Does anyone have a handle on the effect of false positives?

[west]

cobafdek,

 

I'm on my phone so I'm going to have to keep this shorter than I'd like.

 

First, I'm very happy someone is asking questions.  I never posted how I did my calculations or how I got rid of bad companies, and no one has said anything.  It kind of makes me wonder... :)

 

Anyway, so have I checked the Compustat data that I've worked with?  Not thoroughly, but I'm pretty surprised that when I do check it, it *almost* always seems to be correct.  Even with small, foreign companies where the financials are poorly disclosed, but theoretically available.  So I have a lot of faith in Compustat so far.

 

On the "bad data" stocks that show up in screens in other places: These honestly don't seem to be showing up in Compustat.  For real.

 

(Have I mentioned that I love Compustat!  I just wish it didn't cost whatever it does.  Something like $50k or so a year.  And as an individual, I theoretically can't even buy a subscription if I wanted to.)

 

On the EV calculation: Great question.  For minority interest (in a perfect world) I would at least try to find out the industry P/B ratio and apply it to the book value of whatever minority interest was.  Since I was doing quick, get-it-out-the-door coding, I just added book value of minority interest as it was in the EV calculation.  For preferred shares, I dropped all companies with a complex capital structure (more than just one type of stock outstanding) from my list to simplify things.  I figured this wouldn't grossly skew the data since most companies in the US have just one class of stock.

[oddballstocks]

I have found professional data services to be night and day different from the free ones.  I hate to say it, but the old maxim rings true "you get what you pay for"

[frommi]

Are the 10/30/50 stocks always the cheapest 10/30/50 or is it a random pick out of decile #1?

Are netnets in it or are they left out because of a negative EBITDA/EV rating?

Is there a free/cheap screener that can do exactly that type of screening?

 

If not, west why don`t you set up a website and charge a fee for it? :)

 

( Can`t wait to see data from 2008/2011. )

[west]

Are the 10/30/50 stocks always the cheapest 10/30/50 or is it a random pick out of decile #1?

Are netnets in it or are they left out because of a negative EBITDA/EV rating?

Is there a free/cheap screener that can do exactly that type of screening?

 

If not, west why don`t you set up a website and charge a fee for it? :)

 

( Can`t wait to see data from 2008/2011. )

 

It is always the cheapest 10/30/50.

 

Net-nets should be in the cheap EV/EBITDA list as long as they qualify by that metric.  However, negative EV companies *were* dropped since at the time I was sure how to code them in the sort.  This should be fixed.

 

Why don't I set up a website?  Setting up a website means dealing with customers.  I subscribe to oddballstocks notion that unless they're paying me big bucks, I'd just prefer not to deal with customers.  Plus, I'd have to figure out the whole licensing the data thing.

 

On whether there's a website that does this: I hear screener.co licenses compustat, but I haven't verified this.  I tied the free trial a while ago, and I was very unwhelmed by its lack of flexibility.  However, for a dumb, current EV/EBITDA sort it might work fine?  The last time I check it gave a free 20 day or so trial if you want to check it out.

[frommi]

Thanks for your answer. I used my free time on screener.co already and was pleased with the data, but i am just to cheap to pay for data. Perhaps i should just do it.

I read on alpha architect that a shiller PE screen had very pleasing results, too. Do you think when you take the average EBITDA of the past 5-8 years you get better results? Or did you do that already?

[Scorps]

Portfolio123 supposedly uses compustat data for their screens.