Keystone 2: Does it Really Work?
By Moe Chernick (firstname.lastname@example.org)
El Segundo, CA (June 22, 1999) -- In my last column, we introduced the Keystone Screen, the large-cap growth screen developed by Robert Sheard and followed here in the Workshop rankings and returns every week. We discussed how the screen takes the 30 largest stocks ranked 1 or 2 for Timeliness by Value Line and sorts that list by each stock's recent price performance (relative strength). We also saw that this screen has compiled an impressive average annual return of 31% per year from 1987-1998. In tonight's column, we continue our look at Keystone by examining the legitimacy of this screen.
To determine a screen's legitimacy, there are three main items that you should look at before investing. The first is the logic behind the screen: Does it make sense? Suppose you decided to test a strategy of investing only in computer companies that start with D. There is no logic to this, but almost any portfolio that included Dell Computer (Nasdaq: DELL) during last five years would have looked pretty impressive.
The second is data mining or curve fitting: Have the screen's criteria been manipulated to select winners or leave out losers? An example of curve fitting would be a criterion like: Relative Strength greater than 17% but less then 28%. A screen like that might produce very high historical returns, but only because it just happens to capture a few exceptional stocks while eliminating some losers.
The final factor is the crystal ball effect: Do the screen's criteria include items that you could not have known were important without knowledge of the future? An example of this would be a screen that selected Internet companies. The historical returns since 1994 might be very impressive, but who knew that then just how important that would be? And, more importantly, does it tell us anything about the future? The classic crystal ball screen is Buy Microsoft.
The logic that large and fast growing companies (Keystone's first two criteria) are reasonably safe bets is not hard to swallow, nor is the idea that stocks that have been moving up in price recently might continue to do well. Keystone's simple formula helps eliminate data mining as a concern. There just aren't enough factors there to manipulate.
As for the crystal ball effect -- there is room for debate on this topic. In recent years, large-cap stocks have outperformed small-cap stocks, therefore one can argue that by choosing only large caps, this screen is predictive. This is a concern that should not be dismissed. I believe a strong argument can be made that the small stock myth is just a myth and that large stocks have almost always outperformed small stocks. However, let's look at the data on this screen and see if we can confirm the legitimacy of this screen without taking a side on that debate.
The first set of numbers I am presenting comes from a column Robert Sheard wrote on June 19, 1998. In that column, he showed that Keystone's average annual returns from 1/1/86-6/17/98 were as follows:
Top 5 Stocks: 29.4%
Top 10 Stocks: 26.3%
Top 15 Stocks: 24.4%
Top 20 Stocks: 23.5%
Top 25 Stocks: 21.7%
Top 30 Stocks: 21.1%
S&P 500: 17.6%
The results show the legitimacy of this screen in two important ways. First, it validates Value Line's selection criteria for timeliness. Remember that the Standard & Poor's 500 Index is a large stock index, yet the Keystone 30 (large stocks rated 1 or 2 for timeliness by Value Line) beats it by 3.5% a year. This means that even without the relative strength factor, these stocks beat the S&P 500.
Second, look at the symmetry of the returns. The Keystone screen sorts these large, timely stocks by recent price performance -- relative strength. The first 5 stocks, those with the highest relative strength, do the best, the second five stocks do the second best, and so on all the way down the list. This is exactly what you would expect if RS was a legitimate factor. In fact, the symmetry wouldn't have to be nearly this strong to show that relative strength is legitimate.
This data from Robert Sheard's original study is strong evidence that this screen works. However, there are still two potential concerns about the data. First, the data only covers 12 years, and second, all the data is based on a January 1 start date. Unfortunately, at this moment, we are limited to only 12 years of data. However, thanks to Brian Malcolm, we can now look at how Keystone does during the rest of the year. If you would like to download Brian's data you can get it here.
Brian's data shows the following average annual returns from 1987 through 1998 for the Keystone 5 for various start months:
From this data you can see that the returns are strongly affected by the start month,. The screen works best from November-February. While, ideally, the seasonality would not exist, the fact that it does is not all that unusual. We see the same thing with the Foolish Four data. What is important is that no matter which month you start, this screen beats the S&P 500.
Therefore, from both an analytical and an intuitive viewpoint, we see a lot of evidence that indicates that Keystone is a legitimate strategy that is likely to continue to work in the future.
Does this mean that it will? Only time will tell for sure, but like the RP4 screen, this is one screen where the data is solidly in your favor.
Until next time, Fool on!