The model that makes up about 80% of my stock picking process has undergone several minor adjustments, one following quickly after the other. Since I have referred to it quite a bit recently I thought it might be useful to thrash out the details as well as its history.
The model detailed here replaced a previous effort which I used through 2008-2010. The old model focused almost exclusively on the balance sheet rather than on earnings, but eventually I had to admit that earnings might be important so I started this new one from scratch, basing it on the strongest research I could find.
Version 0.1 – Fair value and the Piotroski F-score
First I calculate an approximate fair value for the company and hence its shares. I do this by taking the average of ROE10, ROE5 and ROE3 (which in turn are the 10, 5 and 3 year ROE averages) and using that as an estimate of the average future ROE under normal conditions. By blending those historic averages I weight recent results more heavily than results many years ago.
There are various issues with this approach that need to be manually looked at when selecting a company. For example, companies with extreme ROE figures in some years due to their book value being unusually small at some point. This might skew otherwise weak ROE results upwards and makes them look attractive.
I then assume that fair value is ten times my normal return estimate, which produces a target price from which I can calculate the upside from the current share price. This upside is the first factor in the model as it measures the capital gains you might reasonably expect in the long term.
Sorting companies by upside alone gives a top scoring decile (50 companies out of a universe of 500 with positive 5 year ROE) with these attributes: Normal ROE 15.8%; price/book ratio 0.64; Piotroski F-score 5.6; net gearing 45%; upside 234%.
I use the Piotroski F-score in an attempt to measure where the company is in the business cycle and therefore how far it may be from recovering and being re-priced back to normal levels. It’s better to have a company with 50% upside that may return that amount in 6 months rather than one that may take several years. The F-score may be useful here as it measures changes in returns, cash flow, leverage, liquidity, margins, and asset turnover among other things, all of which help define a positive or negative trend.
To factor in the Piotroski F-score I simply multiply it by the upside so that the higher the F-score the better the company scores in my model.
Sorting on this score gives a top decile with these attributes: Normal ROE 16.6%; price/book ratio 0.7; Piotroski F-score 6.0; net gearing 52%; upside 231%. This gives a higher F-score than before but somewhat higher gearing too. High gearing may or may not be an issue for companies that are well into the turnaround phase.
Version 0.2 – Adds net gearing
The previous version gives an average top decile net gearing of 45%, but this is the product of some companies with almost no gearing and many with over 100% or even 200% gearing. In the name of caution I decided to control this risk by adding net gearing as a factor. This is done by multiplying the version 0.1 score by POWER(2, -(net gearing/100)), in excel speak. This means a net gearing of 0% doesn’t change the score, 100% becomes ½ and so halves the score, 200% becomes ¼ and so on. Net cash produces a number greater than one and so increases the score. This helps bring down the average gearing of the top scoring companies.
Sorting on this score gives a top decile with these attributes: Normal ROE 16.1%; price/book ratio 0.7; Piotroski F-score 5.9; net gearing 23%; upside 223%. This group has slightly lower upside than before but half the average net gearing.
Version 0.3 – Adds absolute ROE
With the previous models, two companies with the same upside and gearing would score the same even if one had a normal ROE of 20% while the other could only manage 10%. After reading the Magic Formula book I was swayed by the argument that a higher ROA or even ROE is better regardless of price as earnings can be reinvested more efficiently. To factor in ROE as an absolute number rather than just in relation to the current price/book ratio version 0.3 multiplies the version 0.2 score by the expected normal ROE, so the higher the ROE the better the score.
Sorting on this score gives a top decile with these attributes: Normal ROE 19.2%; price/book ratio 0.9; Piotroski F-score 5.8; net gearing 24%; upside 218%. Once again the upside has dropped slightly but gearing and the F-score remain steady while average ROE has gone up by 3%.
In comparison, the universe of 500 companies has normal ROE 14.9%; price/book 2.3; F-score 5.9; net gearing 43%; upside -3%. So the top decile has better ROE, lower gearing, about the same amount of fundamental momentum and is lower than half the price (relative to book) of the universal average.
The upside of -3% across the whole universe shows that perhaps there is some sense to my estimate of fair value, given that on average the 500 stocks are about fair value.
One other telling metric is the current ROE in comparison to the expected normal ROE. The top decile has a current ROE some 28% below their expected normal level, which suggests that investors are pricing the companies on an excessively short timescale. The universe as a whole is down 16%, which makes sense as we are in the great recession. The bottom decile, containing the most overvalued companies, has current returns almost 9% above their normal average. This suggests that investors are perhaps overpaying for short term good results, something which value investors have been saying for almost a century.
Unfortunately the only way to know if the differences between these models makes any difference is to test them all together in real time going forwards, which isn’t going to happen. I think it’s likely that they would all perform reasonably well, but each adjustment makes me slightly happier with the selection of companies it produces, which is important if the strategy it to be stuck with over the long term.