invest(igations)

Did What Works on Wall Street Stop Working?

This article is the sixth in a series about screens designed by famous investors. The first, on Benjamin Graham, can be found here; the second, on William O’Neil, can be found here; the third, on Joel Greenblatt, can be found here; the fourth, on Ken Fisher, can be found here; the fifth, on Joseph Piotroski, can be found here; and for an overview of the subject, see my article “Can Screening for Stocks Still Generate Alpha?”

James P. O’Shaughnessy’s What Works on Wall Street has long been one of my favorite books on investing.

Not because it’s so readable—it includes pages and pages of backtests that are a real slog to get through. And not because it’s impeccable—there are some odd omissions, which I’ll get to later. And not because it strikes an estimable balance—there’s simply not enough about accounting measures, and what’s there is somewhat weak.

It’s because O’Shaughnessy’s approach has always struck me as the most sensible approach to stock selection, and prior to O’Shaughnessy, as far as I’m aware, nobody had proposed it. It is, in a nutshell: study assiduously what stock-picking factors have worked best in the past and combine those factors to pick stocks for the future. This is what I do every day, and it has made me a wealthy man.

What Works on Wall Street (WWOWS) was originally published in 1997, then went through three more editions, in 1998, 2005, and 2012. Each time, O’Shaughnessy refined his factors and found new ones. Each time, he was able to boast that his approach still worked.

But if O’Shaughnessy were to revise his book for a fifth edition, I’m afraid he’d have to temper that claim.

The O’Shaughnessy Screens

At the end of the fourth edition O’Shaughnessy concludes that two strategies out of all the ones he’s outlined are the best—“one for investors willing to take market risk and one for very conservative investors.”

Both start with what O’Shaughnessy calls the “All Stocks” universe, which excludes stocks with an inflation-adjusted market cap less than $200 million. Because O’Shaughnessy sets that $200 million limit in December 2008, the limit today would be $249 million. Using Portfolio123, I can write a rule for that limit using the Consumer Price Index. This leaves us with just a little more than 4,000 stocks. (If you’re a Portfolio123 subscriber, you can access the universe.)

Both use a ranking system O’Shaughnessy calls “Value Composite 2.” It has six equally weighted components: price-to-book, price-to-earnings, price-to-sales, EBITDA/EV, price-to-cash-flow, and shareholder yield (buyback yield, as measured by share reduction, plus dividend yield). (O’Shaughnessy and I actually use the reverse of the first five ratios—earnings yield, etc.—which enables us to rank stocks with negative income, EBITDA, etc.). (Here’s the ranking system.)

For Very Conservative Investors

O’Shaughnessy’s conservative screen invests in the top 25 stocks in the staples sector according to shareholder yield and the top 25 stocks in the utilities sector according to Value Composite 2. Portfolios are formed every month and held for one year. I’m starting my backtesting in January 2010, since the last portfolio that O’Shaughnessy tested was in December 2009. This 50-stock portfolio has an average annual return of 11.57%. An equally weighted portfolio of stocks from the “All Stocks” universe gets only 10.35%, so this strategy is a modest improvement on random stockpicking (though not on the S&P 500). (Here’s the screen, but you first have to copy the universe linked above.)

Trending Value

The strategy for investors willing to take market risks is called “Trending Value.” It takes the stocks that rank in the top 10% of the All Stocks universe by Value Composite 2 and winnows those down by buying those with the greatest six-month price appreciation. (Here’s the screen, but you first have to copy the universe and ranking system linked above.)

Once again we form portfolios every month and hold for one year. The results are not good. The average 25-stock portfolio earned only 8.41%, and a 50-stock portfolio earned only 8.53%. You’d have been better off just choosing stocks at random.

What Went Wrong?

O’Shaughnessy is nothing if not a careful backtester. But backtested results often don’t survive out of sample.

Let’s start with O’Shaughnessy’s Value Composite 2 ranking system. O’Shaughnessy is fond of decile ranking, and gives dozens of examples in his book. If we take Value Composite 2, though, and look at how stocks in the All Stocks universe have performed since January 2010, with annual rebalancing, we get a complete reversal of how they performed in his book:

The red bar at the left is the performance of the S&P 500. The stocks in the All Stocks universe that performed best were those ranked lowest—the most expensive stocks (the blue bar on the left). Those that performed the worst were the cheapest (the green bar on the right).

Let’s look at how these same stocks performed according to their six-month price appreciation.

That’s a little bit better, but not a whole lot. Stocks with poor six-month price appreciation did not do well, but those in the other nine deciles all performed about equally well. It seems that neither of the factors that O’Shaughnessy was so bullish about have worked since the last date of O’Shaughnessy’s published backtests.

Some Modest Improvements

Without doing any further backtesting, I want to suggest some modest improvements to O’Shaughnessy’s “Trending Value” screen.

1. O’Shaughnessy’s model favors stocks in certain industries, because some industries, by their very nature, have much higher earnings yields, shareholder yields, and so on, than others. For example, running O’Shaughnessy’s “Trending Value” screen today means that 26 out of the 50 stocks will be in the financial sector, and only one will be in the health care sector. It makes common sense to have a more industry-neutral value system. If we rank each stock on its six value ratios against other companies in the same industry rather than against all other companies, our results should improve.
2. It makes little sense to rank stocks on their six-month momentum and then hold them for one year (O’Shaughnessy practically admits as much). And you certainly aren’t buying them at their cheapest price. I suggest ranking them on their twelve-month momentum, measured one month ago. This is hardly a newfangled approach—it’s a classic momentum strategy that has been in use for well over twenty years.
3. Stocks with consistent growth tend to outperform stocks with inconsistent or negative growth. And a great deal of a stock’s intrinsic value depends on its growth prospects. This is a measure that O’Shaughnessy does not examine in his book. Instead he simply looks at one-year EPS growth, which tells you relatively little about a company, and he ranks on that factor, so that stocks with one-year growth of 400% rank much higher than those with one-year growth of 20%. This measure, he finds, largely fails. Most investors I know would rather own a company with consistent 20% yearly EPS growth over the last five years than a one-year EPS growth rate of 400%, which is clearly unsustainable. One way to measure consistent growth is simply to count the number of quarters over the last five years in which a company’s EPS has been greater than its EPS in the same quarter a year ago. You get a number between zero and twenty. I suggest we add this to our ranking system and give it one-quarter of the weight of the system. (You can see my improved ranking system and my improved screen.)

Results of the Improved Trending Value System

With a portfolio of 50 stocks, we now get a one-year return of 12.99% using the same parameters for our backtest; with a portfolio of 25 stocks, we get a one-year return of 13.23%. Both of those are a good deal better than the one-year return of an average stock over the same time period.

Supercharging the Trending Value Screen

I suggest we go farther. Let’s cut the holding period to three months and lower the liquidity threshold to $100 million market cap as of 12/2008 ($125 million as of today).

We now get a 91-day return of 4.10% for the 25-stock portfolio, which is a one-year return of 17.49%. Yes, if you have a cash account, your taxes will go up, and yes, if you’re investing $100 million in this strategy you might suffer from some market impact buying the smaller stocks. And you’ll have to pay more in transaction costs as well. But the difference in returns is so substantial that it may be worth the cost.

Conclusions

O’Shaughnessy, I believe, wanted to keep things as simple as possible in his book. But the way he deals with industries and time frames is just too simple.

Many ratios, whether value, quality, or growth, should be measured against comparable companies in their sectors, subsectors, or industries rather than against stocks as a whole, and that’s not something O’Shaughnessy has ever done in the various editions of WWOWS.

Moreover, O’Shaughnessy doesn’t really indicate the time frame for his measures. He simply says “Price to Earnings” or “Price to Sales” without telling us if he’s using annual figures, trailing-twelve-month figures, or three-year averages (I used trailing twelve-month figures for value ratios). And often you’ll want to look at three-year or five-year measures (particularly for measuring growth and quality). O’Shaughnessy doesn’t do so in his book.

On the whole, I still think WWOWS is one of the best books ever written on investing. Not only is O’Shaughnessy’s approach irreproachable, his writing is excellent and his approach to backtesting is exemplary.

But in a world where screening for stocks is rampant, sophisticated algorithms are governing multi-billion-dollar portfolios, and every stock of a certain size has been scrutinized to death by thousands of professional investors, the simple approaches that O’Shaughnessy suggested may no longer work. One needs to increase the level of sophistication, of complexity, in order to beat the market. O’Shaughnessy was correct in telling us how to find out what works on Wall Street, and was correct in his conclusions. The fact that his strategy design ended up failing should not invalidate his approach. It only took a few minor tweaks to convert it into a successful strategy again, even in a market which has been remarkably unfriendly to value investing over the last ten years.

In other words, What Works on Wall Street still works, even if one of O’Shaughnessy’s particularly favored strategies failed.

A Few “Trending Value” Investments

Here are ten stocks in the All Stocks universe that, according to the revised “Trending Value” screen, should do well over the next year. I list them in order of twelve-month momentum measured a month ago: Apollo Healthcare (AHCCF), Jaguar Mining (JAGGF), Big Five Sporting Goods (BGFV), Quidel (QDEL), Canaccord Genuity (CCORF), Halozyme Therapeutics (HALO), Superior Group (SGC), Big Lots (BIG), B. Riley Financial (RILY), and Tronox (TROX). These stocks rank in the top decile according to the improved Value Composite 2 ranking system using both Compustat and FactSet data, so they seem genuinely low-priced.

My CAGR since 1/1/2016: 48%.

My top ten holdings right now: STRT, PMTS, CTG, RMNI, LMB, EVC, HBP, BKTI, CCORF, WFG.

Waiting for Fair Value

The idea behind value investing is that you buy a stock well below its fair value, wait for it to appreciate to something close to its fair value, and then sell it.

The natural questions, then, are:

• How long does this process take?
• How often does it actually happen?
• Why does it happen?

Is it possible that this whole idea doesn’t actually work? Is it possible that if you buy a stock well under its fair value you have no better chance of it reaching fair value than if you assigned a more or less random price target to a random stock and held it until it reached that target?

I decided to study stocks listed on the major US exchanges over the last 22 years to try to get an answer to these questions. To be honest, at the outset of my study I had no idea what the answer would be.

Assigning a Fair Value to a Stock

As I’ve written, calculating an intrinsic value for a stock is a process fraught with difficulties, and I believe it cannot be done using an algorithm. It is a necessarily discretionary process. The most recent attempt I made at an algorithmic method relied on backtests I’ve done on the last twenty years of data, so that algorithm cannot be properly tested on that same data.

Instead, I’ve come up with a formula that assigns a fair value to a company based on comparing seven measures to those of the average comparable company: its sales, income, shareholder payout, tangible book value, research expenditures, EBITDA, and unlevered free cash flow.

Methodology

Here is my methodology. It gets a bit technical, so you can skip it if you’d like.

I start by using Portfolio123 to come up with aggregate ratios for each item. I’ll give an example with net income. I look at the current fiscal year’s mean EPS estimate for each company in a GICS subsector and multiply it by the number of fully diluted shares; if a stock has no EPS estimate, I’ll use its trailing-twelve-month net income instead. I then divide it by its market cap. I then exclude the top and bottom 16.5% of the results and all zero or negative results. I take a cap-weighted average of the rest. I’ll call this the aggregate subsector earnings yield.

Then I take the company’s estimated income (or actual income if there’s no estimate) and divide it by the aggregate subsector earnings yield for its subsector to arrive at what I’ll call an income-based fair value.

For example, let's take a large cap stock which I've selected randomly, Newmont (NEM). It has an estimated net income of $3.05 billion this fiscal year, according to analysts. The earnings yield of its subsector (which in this case is the same as its sector, materials) is 5.6%. Divide $3.05 billion by 5.6% and you get an income-based fair value of $54.5 billion, which isn’t that far from its market cap ($45.4 billion).

I do the same thing for the company’s sales. For the shareholder yield, I follow a similar course, using total equity purchased minus total equity issued plus total dividends paid, multiplying the latter by the indicated annual dividend divided by the dividend per share of the most recent fiscal year (all numbers here are for the most recent fiscal year). For the tangible book value, I simply use that item based on the most recent quarter’s balance sheet, without relying on estimates. For the research-based fair value, I simply make that zero if the aggregate for the subsector is zero.

For EBITDA and unlevered cash flow, I also privilege estimates, but divide by enterprise value rather than market cap and use the entire universe of non-OTC stocks rather than the subsector to calculate the aggregate; I then subtract from the result the company’s total debt, noncontrolling interest, and preferred equity, and add its cash and equivalents, to get back to its market-cap based fair value.

I end up with seven factor-based fair values. I exclude the highest and lowest of these and average the middle five to come up with my final fair value.

But for companies in the financial sector, I don’t use the EBITDA, unlevered free cash flow, or research values, and I take the median of the other four.

Fair Value Examples

Here are some examples of fair values for a dozen S&P 500 stocks, chosen completely at random:

As you can see, most of these companies are relatively fairly valued at the moment. Western Union and Capital One seem underpriced and might be good value-based investments; Arthur J. Gallagher seems comparatively overpriced.

Please note that I did no backtesting at all to come up with this complicated formula. The seven measures I chose are all old, dating from the 1980s or earlier; it has long been standard to compare market-cap-based value measures to similar companies, and it has long been standard to compare EV-based measures to the market as a whole.

Now let’s see how investing in underpriced companies and holding until they are fairly priced actually works in practice.

Testing a Fair Value-Based Strategy

My test is to check every week among all stocks listed on US exchanges and to buy them if a) their market cap is $30 million or more; b) their price is $1.00 or more; and c) their fair value is three times their market cap or more. I will then hold them through thick and thin until their market cap becomes greater than their fair value—or until they’re delisted or are bought by another company.

I tested this going back to 1999 using Compustat’s database, building a portfolio that at its peak numbered over 400 stocks. The CAGR of the equity portion of the portfolio (ignoring the cash portion) is 15.77%, which is respectable, considering the CAGR of the S&P 500 is 7.28% over the same period, and considering that you’d often be holding almost as many stocks at a time. A total of 3,118 stocks were bought. 53% of the stocks were winners. The average return of a stock was 36.98%; the average holding period was 464 days (about 15 months). Here is a year-by-year summary:

As one would expect of a strictly value-based strategy, it outperformed over the 21st century as a whole, but underperformed the S&P 500 in six out of the last ten calendar years.

Now some of these stocks rose to their fair value; others were sold because they were delisted or acquired. To answer the first two questions I posed at the beginning of this article:

• Of the 2,356 stocks that were bought more than five years ago, 78.7% of them “rose” to their fair value within five years; 20.5% were delisted or acquired; and 0.8% remained in the portfolio for over five years. (I put “rose” in quotation marks because in many cases the fair value dropped so much that the stock was sold at a loss.)
• Of all the stocks that “rose” to their fair value within five years (and this includes some stocks bought in the last five years), it took an average of 370 days for them to do so, which is just over a year; they had an average return of 34.13%.

We can conclude that the average stock, if bought at a discount of 67% to its fair value, will take about a year to “rise” to its fair value.

It’s important to note that this system buys each stock at more or less the precise moment when its price dips below 1/3 of its fair value. Its price could continue to fall for a while afterwards. It sells it at more or less the precise moment when its price reaches its fair value. Its price could continue to rise afterwards.

The Mean Regression Puzzle

Now it’s entirely possible that what we saw with fair value is simple mean regression. By this I mean that as long as one can come up with a “fair value” for a stock that is based on an average of some kind, then, by the strict and immutable laws of probability, stocks that are very cheap or very expensive compared to the average will be more likely to have average prices later. This has nothing to do with how much a stock is actually worth.

If, on the other hand, our measure of “fair value” is a good indication of how much a stock is worth, then a measure of “fair value” that is simply based on mean regression would produce a distinctly worse outcome.

The Fundamental Difference

So here's how I came up with a mean-regression-based model.

Most fundamentals of a company are positively correlated with its market value. The reason the seven components of my fair value have been chosen by value investors over the years is not just that they’re correlated with market value but that they also make a positive contribution to it. If, for example, one had chosen total liabilities instead of book value, or cost of goods sold instead of R&D, one would have favored companies that are likely to do poorly rather than companies that are likely to do well.

One can classify fundamentals into roughly four groups:

1. Positive fundamentals. These are the seven fundamentals I chose to base fair value on (income, sales, EBITDA, free cash flow, R&D, shareholder payout, and tangible book value): when a company’s positive fundamentals go up, it’s taken as a good sign.
2. Negative fundamentals. These are things like cost of goods, liabilities, debt, and inventory: when a company reduces its negative fundamentals, that’s taken as a good sign, and if you compare two companies with the same market cap, you might favor the company that reports lower figures for these items.
3. Neutral fundamentals. These are things like total assets, number of employees, SG&A expenditures, PP&E, depreciation and amortization, and so on. All of these correlate roughly with a company’s market cap but they are not normally used for valuation purposes; decreases or increases in these fundamentals are not necessarily good or bad, and if two companies with the same market cap had different levels of one or more of these items, it would not be considered a bullish or bearish sign.
4. Uncorrelated fundamentals. In this category would be all ratios, since ratios have no correlation with market cap; also included would be items that only appear sporadically on a company’s statements and items that vary wildly from company to company or from statement to statement.

In order to determine if value really is a significant improvement on mean regression, I have come up with a measure that we can use as a base rate: a fair value measure that is based on neutral fundamentals rather than positive ones. We’ll call it, for the sake of jest, the company’s “unfair value.” Like the fair value measure, it has seven components: total assets, number of employees, SG&A expenditures, gross PP&E, depreciation and amortization, working capital, and intangibles. I’ll compare total assets and working capital to enterprise value and the rest to market cap, using exactly the same sorts of calculations as I made for the seven components of the fair value estimate (for firms in the financial sector, I’ll use only assets to market cap, employees, and SG&A). Please remember that these are ratios that, as far as I know, nobody actually uses because they make little financial sense. Once again, I’ll buy companies at one-third of their unfair value and sell them when they reach their unfair value.

The “unfair value” should be taken as a base rate to which to compare the results of the “fair value” experiment.

Price-Based Value

Just to be on the safe side, I’ve also created a “price-based fair value” that has nothing to do with a company’s fundamentals.

I’ll set the “price-based fair value” of a stock as equal to the average of its price one month ago, three years ago, and five years ago, each adjusted for the growth of the sector it’s in since then. (I'm avoiding its price six months ago and one year ago because of the momentum factor.) Once again, we’ll buy stocks that are priced at one-third of their price-based fair value and sell them when their price is greater than their price-based fair value. This system disregards fundamentals altogether, simply buying stocks whose prices have fallen quite drastically.

Base Rate Results

If the success of the “fair value” measure were due entirely to mean regression, we should find the same success with the “unfair value” measure and the “price-based value” measure. But we do not. Below is a table comparing the results.

On every measure—CAGR, average holding period, return per stock, and percentage of stocks that rise to fair value—the original “fair value” method works far better than the “unfair value” or the “price-based value” method.

Conclusion

Value-based investing works. If you buy a stock at a deep discount to its fair value with a plan to sell it when its market value is more or less equal to its fair value, you will more likely than not end up realizing a decent profit. And your success will not be due solely to the law of mean regression. Stocks actually do have a fair value, no matter how difficult it might be to come up with a reasonable estimate of it. I won't say that value investing is the only method that works in the stock market. But as a portfolio management strategy, it's hard to beat buying at a discount and selling at full value.

A Few Stocks That Are Priced at a High Discount to Fair Value

Currently, quite a few stocks have market caps of less than a third of fair value, according to my seven-factor formula. But most of those are pretty tiny companies. Of companies in the S&P 1500, the following five are clearly underpriced: Brighthouse Financial (BHF), Endo (ENDP), Genworth Financial (GNW), H&R Block (HRB), and Seneca Foods (SENEA). That doesn't mean, however, that they are all good investments. If you're willing to go smaller, I would favor the following underpriced stocks for a sell-when-price-approaches-fair-value approach, though some of them may be difficult to buy in quantity due to limited liquidity: Big 5 Sporting Goods (BGFV), Ingles Markets (IMKTA), Medical Facilities (MFCSF), Mannatech (MTEX), and Summer Infant (SUMR). Not only are they selling for less than a third of their fair value, they all rank pretty highly on the various ranking systems I've designed on Portfolio123.

My CAGR since 1/1/2016: 48%.

My top ten holdings right now: STRT, CTG, PMTS, LMB, RMNI, USAK, CCORF, WFG, BELFB, PTSI.