Issue with All-Stars: Graham Ranking System

I have been working with a list of high growth stocks and curiously found that All-Stars: Graham was the best ranking system for these stocks, which is totally unexpected. I looked into it further and I believe that I have found a problem with the RS. The node:

Oper PE

is implemented as MktCap/ OpInc(0,TTM)

The problem is that OpInc can be negative. If it is negative then the node will rank higher than any stock with positive OpInc.

This node should be changed from Oper PE to Oper Yield, implemented as OpInc(0,TTM) / MktCap. Of course, the node has to be changed to “higher is better”.

A second problem has to do with the node EPS5YCGr%, or at least I believe that it may be a problem. How does one calculate earnings growth if earnings start out with a negative value? Is the result NA or is there some sort of calculation performed?

Operating income cannot be coupled with market cap because it’s a measure of the entire firm’s performance, not just the equity portion. Operating income has to be coupled with enterprise value, which is the EBIT/EV ratio that Greenblatt writes about. I noticed this problem too a week or two ago. So the node has two problems, not just one.

All growth rates are calculated by taking the difference between now and before and dividing by the absolute value of before. In this case, this would then be annualized by adding one, taking the fifth root (^0.2), and subtracting one.

For example, salesforce had an EPS of -0.40 five years ago and now has an EPS of 1.44. That’s a difference of 1.84. Divide that by 0.4 and you get 4.6. The fifth root of 5.6 is 1.41. Subtract one and you get 41% , which is exactly what EPS5YCGr% gives.

Thanks Yuvol - it has been a while since I did any serious math, but isn’t the growth rate bogus when starting with a negative value? If a company had zero earnings and 5 years later the growth rate would be infinite. I think that negative earnings should return NA.

If you were to exclude from growth measurements all companies who posted negative income numbers, you’d be missing a huge part of the picture. One-half of the Russell 3000 have had negative TTM EPS at least once in the last five years.

Unfortunately, the way we measure growth isn’t ideal. But nobody, as far as I know, has come up with a better way to deal with negative past performance. And the way we do it is standard among financial data providers.

If you want to write your own custom formula for growth, put this in the denominator: Max(0.05,Abs(EPSExclXor(20,TTM))). That way you’ll never be dividing by zero, and you’ll be normalizing growth rates that can seem kind of insane because a company was once earning between -0.04 and 0.04 per share. That’s not a perfect solution either, but it’ll have to do until someone comes up with a better one.

So you come up with a number for all Russell 3000 stocks, but the numbers aren’t meaningful for half the stocks. It seems to me this does more damage than good. Is it not better to have half the stocks give a value of NA but the rest meaningful, as opposed to the negative earnings giving nonsensical values?

Back to my original observation, Mr. Graham was the father of value investing, yet the RS does a wonderful job, not on value stocks, but growth stocks.

Steve

Steve,

I am sure you already thought about this but here are my 2-cents…

  • The second node favors stocks that have grown for the last 5 years and have done so steadily. So, there is a decent chance that the historical growth will continue in the future (i.e. more likely that stocks with the same 5y growth but a more unstable = unpredictable growth or no growth at all)

  • The first node finds those that a cheaper relative to the others. So, say, we take a bucket of similar ranking stocks on the 2nd node, the first node will rank higher those that are relatively cheaper / better value for whatever reason. They should therefore have a potential for price appreciation relative to the others.

  • One can probably make the same reasoning starting with the first node: we create a number of buckets for what appears to be good value stocks. Within a given bucket of good-value stocks, I will probably prefer to pick those that have a 5y history of stable growth as opposed to those with no growth or a wildly fluctuating growth.

I would be curious if you could let me know if your results improve if you remove (e.g. in a buy rule) the “best” outliers for each node i.e. exclude the best (cheapest) 2% or 5% on the value node and the best (highest growth) 2% or 5% on the growth node. The idea is to avoid “cheapness” that is too good to be true (there is a good reason for it to be cheap but it is not captured by any of the 4 factors) and those that had a past growth that is actually unsustainable.

What is missing for me in your scenario is a notion of quality. If you get good results I assume that your universe / list of high growth stocks has somehow been filtered be decent quality stocks - at least quality enough that we can trust reported numbers.

Anyway, my modest thoughts. I am still learning to think about all this in some logical way and sometimes it works, sometimes it does not…

Jerome

Jerome - I’m simply picking on the RS. It’s one of the flagship ranking systems. I think the concept is good but the implementation is wrong. I’d like to see the RS deprecated and replaced with one that implements the “cheap” and “growth” properly.
Steve

Continuing with my attempt at reasoning…

Did you state that when you try this ranking on a value universe, it does not work well?

Going back to the dividend growth model:

  • We start with a universe of stocks that appear cheap by some metrics
  • We look for the ones that exhibit stable strong growth (node 2)
  • Yet because their price is low (in relative terms), it means their risk is perceived as very high (rightfully or wrongfully but for reasons that are not picked up by the factors used in this RS) relative to others so that the denominator “r-g” is high enough to depress the price.
  • In other words, one has found stocks that are perceived to exhibit very risky growth.

Ideally, node 1 should then help us identify those that are truly risky growth (-> avoid) and those that are perceived as riskier than they really are (-> can go into the portfolio). Currently, node 1 looks for those that are “cheap” so not helpful to help sort this out and probably likely to fall into many value traps.

I actually wonder if we would not get better results in this case by inverting the valuation ranking (i.e. for node 1 only) and set it as “more expensive” is better.
This way we fish the less risky end of the risky growth stocks and therefore more likely to get stocks where the risk will not materialize.

Thoughts and feedback welcomed,

Jerome

Steve, OK.
FYI - I wrote my 2nd piece w/o having had the chance to read your prior answer

I fixed the node in question. It’s now EBIT/EV.

I don’t want to mess with the EPS5YCGr%. That’s standard across the industry.

“standard across the industry”

It isn’t about what others do and in any case, others aren’t employing ranking systems. The growth formula accentuates negative earnings at the expense of legitimate value stocks actually growing earnings. You really need to reconsider what you are doing with this.

Growth is percentage change. If you go from an EPS of -0.40 to an EPS of 0.60, what percentage change did you make? 1.00/0.4, or 250%. If you go from an EPS of -0.40 to an EPS of -1.40, what percentage change did you make? -1.00/0.4, or -250%. If you make any growth rates with negative numbers N/A, you’re making it impossible to evaluate businesses. Remember that valuation is impossible without a measurement of growth, and that practically every company begins its life with negative earnings (though sometimes by the time they go public they’ve fixed that) because outlays are always going to precede profits. You simply can’t throw up your hands and say “I don’t want to calculate this company’s growth rate because their earnings used to be negative,” or else you won’t be able to evaluate the company’s growth and you won’t be able to figure out what the company is worth. The formula does NOT accentuate negative earnings. A company that has increased its EPS from -0.40 to 0.60 has actually grown five times as much as a company that increased its EPS from 0.40 to 0.60. EPS growth should be used in conjunction with other growth measures–sales, operating income, analyst projections–but throwing it away whenever a company hasn’t broken even in a particular time period in its past is senseless.

Yuvol - if you want to measure improvement in earnings, I’m sure you are smart enough to devise a formula to do so. There are a couple of problems with the existing formula that need to be addressed. One problem is that there is a pole at zero. Anything around zero will have an outsized growth figure when it really shouldn’t. Growth approaches infinity as you approach zero. Second is that an adjustment of negative earnings isn’t really growth at all. I have to stand by my comment that this formula isn’t doing what you want it to do and is actually contaminating the RS.

What is better: growing from -1 to +1 or growing from -0.5 to +1?

It’s easy to fix the problem of the vertical asymptote at $0.00 EPS. You just insert “Max(0.05,” before “Abs” in the denominator. No growth formula is perfect. We’re not going to change the underlying workings of EPS5YCGr% because a lot of people use it as is in their ranking systems and simulations, and it matches what’s used throughout the industry. But if you want to substitute your own formula, fine.