Anyone paying attention to market news must know that September is the weakest month for stocks. Mark Hulbert, one of our favorite writers, calls it The Cruelest Month. He provides a table of returns, summarized as follows:
Notice from the table that in all but one of the last 11 decades, September was a below-average performer. In more than half the decades, in fact, the month's rank was dead last.
Why, given such an overwhelming record, would anyone question September's bad record? Because there is no good theory for why the month should be such an awful month for the stock market. And, without such an explanation, there's the distinct possibility that the statistical pattern is just a fluke.
As one can see, Hulbert is well aware of data mining. He mentions the popular "butter production in Bangladesh" example.
Hulbert next considers a number of hypotheses and invites readers to share their own ideas. His thought is that a good explanation would make the September story more convincing, although he shares the information that day traders do not wait for any hypothesis testing! The sophisticated audience of "A Dash" may not find that very convincing.
Hulbert's Market Watch colleague, Irwin Kellner, has a number of reasons for September weakness, including the possibility of a self-fulfilling prophecy.
Why Both Articles are Wrong
Here at "A Dash" we are veteran debunkers of mythical market lore. Most people (including our employers when we started in the business) just want to see the data --- all of it! Traders all believe in patterns. The more data the better.
The idea of random results is lost on most, like the "day traders" Hulbert mentions. There are twelve months. There will be a distribution. Some will be good and others will be bad. Always.
What if there is a reason? A hypothesis does not really help. When you already know the outcome, any smart person can invent a compelling reason. New readers can revisit our discussion of this topic, where a group of very smart grad students were given a list of findings and asked to provide reasons. They did very well. Only after the class were they told that all of the relationships were reversed!
The scientific method works only when one begins with the hypothesis.
Let us try an experiment. Instead of taking the currently constituted months of September, instead put all of the individual trading days in a basket. (We know that this basket has a negative bias, but bear with us). From this basket we create trading months. From the days in the other months, we create comparison months.
This is an interesting approach to getting beyond the data mining issue. It is (unfortunately) not our idea but that of Andrew Moe. He writes as follows:
When comparing the months composing September to a random basket of days the results are random. Attempts to find seasons of non-randomness are frequently subject to data mining bias, as the same permutation test debunking the September drift is easily used to identify (falsely) statistically significant periods.
The study. Running a bootstrap permutation study on Dow data from 1960 to 2008 we estimate the empirical distribution of differences in monthly return between September and other months. We test the hypothesis that a random September is no more bearish than a composition of random days sampled with replacement. We find that the mean difference between populations is 0.0695%, yielding a p-value of 0.3612 – random.
Here are four good reasons to ignore the September weakness articles.
- This is a widely advertised theory. Even if you do not believe in completely efficient markets, one would expect some anticipation.
- We have already had a decline of nearly 2%, exceeding the expected monthly decline on the first day. Should we now expect normal trading for the rest of the month?
- The evidence shows that the September pattern is not a statistically significant deviation.
- Other seasonal methods (Sell in May, Presidential Cycle) have not worked well in this time of turmoil