Re-thinking the Foolish Four
The great returns that we thought the Foolish Four should provide for investors turned out to be mostly a result of random variations in our original sample. When we expanded our study, the high January returns were just a statistical fluke, as were the higher returns we got from dropping the highest ranked stocks.
By Ann Coleman (TMF AnnC)
December 18, 2000
More numbers! I can hear the clamor for more numbers. Better numbers, deeper numbers, more decimal places.
You asked for it, you're gonna get it. OK, no one is actually begging for more numbers, but I have to pretend that you are to get through all this. There are a lot of numbers to report on.
I will try to get through the rest of the interesting stuff we found in the Center for Research in Security Prices (CRSP) database today. Then we will get back to the future (the new Workshop Portfolio) and stop worrying about the past, except to remember the lessons we've learned.
One of the things we looked at was the "January effect," as we define it, which is the seasonality we saw in the original study where the highest returns were clustered around portfolios that started/renewed in December/January. We didn't have any real proof of the seasonal effect but concluded that it was a good idea to start in December or early January, just in case. �
We had lots of theories about what could cause such an effect, but what we are seeing now in the new database is absolutely no seasonality at all. Nothing, nada, zip. It looks like the whole December/January thing was simply a random statistical variation.
That random variation is one of the reasons why we so dramatically over-estimated the returns, by the way. The first studies all used January portfolios only. Since January happened to be the best month, we were essentially using the top number in a series and assuming that it represented the average. Wrong. Then when we saw that January was the highest, we came up with a theory that suggested it was a repeatable phenomenon, and that by investing each January you could get the same high returns that January had shown in the past. Wrong again. If there really was a repeatable January effect, it should have showed up when we expanded the database in time and included some large cap, non-Dow stocks. Instead it disappeared. �
(Note to fans of Beating the Dow: As near as we can tell, the January problem affects the Beating the Dow returns as well. In fact, we based our original assumption that the January returns were representative of all months on that book, mainly because it was plain that the author had looked at returns from all months, even though he quoted only January returns.)
The second thing we checked was whether or not one should drop the first stock. As you may recall, the Foolish Four strategy buys the second through fifth stock because our original, smaller database showed that the first stock tended to underperform. In the expanded study, we don't see any evidence for that theory either. Overall, the strategy performed almost the same whether the top stock was included or not. There is a slight difference in favor of dropping the top stock, but I doubt that it is significant. In fact, we see that including the top stocks worked better in the '80s. Conclusion: There was no good reason to drop that top stock, but including it didn't help much either.
So although it looks to me like the high yield/low price formula we've been using might have been responsible for a small part of the outperformance we thought we saw, most of the extraordinarily high returns in the original sample (which we were loudly proclaiming to everyone who would listen) were a product of "datamining" (finding random correlations and considering them valid and repeatable). �
If someone wants to look at the numbers and conclude that the strategy never worked at all, that's fine with me. When we change the definition of our Top 30 stocks, it's easy to conclude that. Once we take the Dow stocks out, most of the rest of the outperformance goes away. That could be because, when you take the Dow stocks out, a different type of stock replaces them. No one has ever suggested that the strategy should work on just any old company. At the least you need very large companies paying significant dividends. We controlled for size but not for yield.
I have posted a full summary of the results on our discussion boards.
When we took the Dow stocks out, we expected to find that the Foolish Four formula still picked stocks that outperformed. After all, it should do that regardless of whether the stock is listed on the Dow or not as long as it is a big U.S. company. But we don't see that. We see a difference of 0.31% over 50 years. That's not going to convince anyone (including me) that the Foolish Four is a valid stock-picking method. But that may not be the best test.
Essentially, we had two tests for validity. The first was the database with Dow stocks included but looking only at returns prior to the discovery period. Virgin data. The Foolish Four did pretty well on that test. The second was the subset of returns with Dow stocks removed. There we don't find any support, and I don't know what to make of that.
Probably part of the explanation is the Microsoft <% if gsSubBrand = "aolsnapshot" then Response.Write("(Nasdaq: MSFT)") else Response.Write("(Nasdaq: MSFT)") end if %> effect. With the Dow stocks removed, Microsoft and other high growth companies move into the Top 30 (some of them were added to the Dow last year) but, since they pay insignificant dividends, none would be selected as Foolish Four stocks. What that might mean for the future is anyone's guess.
If anyone wants to contend that the lack of support when Dow stocks are included invalidates the whole thing, they are welcome to. I think the pre-discovery test with the Dow stocks included is a better test for past validity because there is no reason to arbitrarily throw out the Dow stocks as long as they weren't part of the original sample. However, excluding Dow stocks may be a better test for future performance since the Dow itself is changing in the direction of lower-yielding, higher-growth stocks.
We were also able to test straight high-yield strategies. As long as the Dow stocks were included, the High Yield 5 and High Yield 10 did a bit better than the Top 30 and a bit worse than the Foolish Four. I'll leave it to the statisticians to figure out if the differences are statistically significant. The High Yield 5 averaged 14.58% and the High Yield 10 averaged 14.19% over 50 years, compared to 13.46% for the Top 30 stocks and 15.20% for the Foolish Four.
We did not test the Beating the Dow method of ranking stocks first by yield and then by price. That would have been nice, but the programming would be formidable. Since the Foolish Four consistently outperformed that method using Dow stocks in the original time periods, I don't see any reason to assume that the BTD method would somehow do better if we did test it.
Well, that's what we found. When you boil it all down it spells WARNING for all of the yield-based Dow strategies. It spells STOP if you are expecting market-doubling returns and probably if you are holding any of these strategies in a taxable account. It spells CAUTION if you are using these strategies in a retirement account and expect to outperform the market.
So why am I keeping the Foolish Four around for another year? I want to see how it does in a recovery year. In the past it has been very strong during recovery years and I just hate the idea of selling it before it has a chance to shine. (Obviously, I'm anticipating a good year next year. We're doomed.)
I also like the idea of balancing the high-growth stocks that the Workshop screens usually pick with some old economy stocks that actually generate steady and substantial profits. Despite its bad performance this year, the Foolish Four is beating most of the Workshop portfolios right now. If we have a good year next year and the Foolish Four still doesn't beat the market, I will conclude that no matter what happened in the past, it is unlikely to be useful in the future even as a value strategy. �
Fool on and prosper!
