As April draws to a close, the old adage “Sell in May and go away” may be on some investors’ minds. The saying refers to the tendency of markets to underperform during the period from May to October (as compared with better performance from November through April), advising us to sell and wait for brighter days ahead.
This sounds like an old wives’ tale, but history suggests there may be something to it. According to Investopedia.com, since 1950, the Dow has returned only 0.3 percent on average from May to October, compared with 7.5 percent for November through April. It’s hard to argue with a gap like that. Or is it?
Super Bowl wins, dairy products, and other spurious correlations
The so-called Super Bowl indicator, which predicts the market will be up if the NFC team wins and down if the AFC team wins, also has a strong record. Despite making no sense at all, this indicator has been right 80 percent of the time. Again, hard to argue with that . . . except, sometimes, it's spectacularly wrong (most recently in 2008).
If this seems strange, wait until you hear which stats actually have the highest correlation with stock market returns. They are—and I am not making this up—butter production in Bangladesh, cheese production in the U.S., and the combined sheep population of Bangladesh and the U.S.
Personally, I don’t trade on dairy products, either here or in Bangladesh. Correlation, as we can see, is not causation, an observation more honored in the breach than the observance, especially in investing. “Sell in May and go away” is an example of spurious correlation, no doubt enhanced by the fact that many investors go on vacation in the summer months. Also contributing, I suspect, is the fact that some spectacular crashes have occurred during that period. One big crash can make it look like an entire month is dangerous, and several can taint a whole half year.
Indeed, I could come up with my own adage—let’s call it “the rule of 7s and 8s”—which warns against investing in years ending in those digits. I haven’t run the numbers, but I suspect the crashes of 1987 and 2008 would mean that the returns for those years, averaged over a couple of decades, would be lower than most. Again, completely spurious, but I could make a case for it.
How to tell the difference between correlation and causation?
Empirical data, as with the sell in May rule, is a good start, but it’s not enough. The rule also has to make sense on a fundamental, theoretical basis. Something that not only shows up, but that you also understand, can be the basis of a good rule. Something that you just observe doesn’t pass that test.
In many ways, this replicates the development of the scientific method. In years past, science was based on observation—and was often wrong. Chemistry, for example, was for centuries a collection of ad hoc recipes, mere alchemy. Medicine was the same. Even today, scientists face a similar challenge in developing new treatments. The answer is double-blind testing and rigorous statistical analysis. And even then, the results are often wrong.