from John Hussman, 9/20/09 Weekly Market Commentary:

The fact is that yes, on average, the combined September-October period has historically produced slight declines for the S&P 500 whether you look back since 1870, 1900, 1940 or 1970. But the variance around that slightly negative return is large enough that it’s really misguided, in my view, to base predictions on it. All you can say is that maybe in a repeated game of 20 or 30 years, you might find that avoiding stocks in September and October slightly reduces risk without surrendering long-term returns.

Saying “September and October are the worst months for stocks” simply isn’t all that scary, because again, the average change in the S&P 500 is just slightly below zero, and the average total return including dividends is still slightly positive (though generally less than T-bills).

While a two-month period with an average historical return of zero is interesting, what is more interesting is the variation in those returns. Investors rarely operate in a vacuum, so a given piece of data really matters only in the context of other pieces. As investors, we are very much interested in “conditional return,” which requires that we base our analysis on a wider context than a single indicator.

With regard to September-October performance, Jim Stack of Investech points out that if you restrict the set of Sep-Oct periods to only those that followed the end of a bear market, the returns for the S&P 500 during that subset have been positive by a couple of percent, on average. The difficulty, from my perspective, is that the close of a bear market isn’t an observable variable, except with the benefit of hindsight. Better to use “conditioning variables” that can be measured directly.

For example, at any point in time, we can observe directly where the S&P 500 Index is in relation to its 6-month moving average. Using this variable to measure the recency of a market weakness or strength, we find that Jim is still right, but he’s right in an interesting way. Specifically, if you look at the set of periods where, at the end of August, the S&P 500 was more than 10% below its 6-month average, it turns out that the S&P 500 has indeed advanced by an average of about +2.5% during the subsequent September-October period.

On the other hand, if you look at the set of periods where the S&P 500 was more than 10% above its 6-month average (as it was at the end of August this year), we find that the September-October returns have been clearly negative, averaging a -5.6% decline for the S&P 500.

The lesson here really is much broader than September-October seasonality, which is interesting in the same way horoscopes are interesting, but not something I consider useful for investment management. The important point is that what investors believe are “average” tendencies for the stock market are actually very dependent on the prevailing context, be it valuation, market action, overbought/oversold conditions, sentiment, economic considerations, or other factors. The implications of September-October seasonality are largely dependent on the position of the market at the end of August. So knowing only that September-October has arrived is not particularly helpful information taken by itself. Similarly, it is not helpful to say, look, we think this is a recovery, and in a recovery stocks do “X” – without coupling that analysis with a whole set of other conditioning variables that shape the outcome (or call the recovery itself into question).