1. Compute a regression prediction of December based on all the data available up to that time. For example, in 1982 you would have two observations, 1980 and 1981. In 1983, you would have 1980, 1981 and 1982 to fit.
2. Compute the average move in the Decembers up to that time and predict that the next month will be the same.
3. Compute the average move in the last three Decembers, and predict that the next month will be the same as this.
4. Take the prediction generated by the first three methods each year and find which one has the best forecasting record in the past, and use that method for the next prediction.

Now you have four methods of prediction. Does any beat just predicting the average change from month to month based on simulation, by a reasonable amount. If you want to make money, or test seasonality properly you have to use your head.

Dr. Kim Zussman adds:

Looking further at the same monthly data, December moves seem large compared to the prior 11 months. To check this (and eliminate effects of sign), for each year I looked at the ratio of absolute values:

|Dec ret|/|J-N ret|

One would expect each month to contribute something like 1/11 of the return of the prior 11 months. But Decembers are larger, as shown by the data:

The attached plot depicts |Dec|/|J-N| vs. date, and though variability in this fraction has damped out over time, it still seems high. Even discarding two out-lying years of ‘83 and ‘87, the mean ratio is 0.26; almost 3 times 1/11.

Rick Foust comments:

I suppose that there are two major factors (amongst other smaller ones) that cause the December effect.

The first is money flowing into IRAs prior to the end of the year. Someone on the retail side of the business could confirm or refute this.

The second is large fund rebalancing. Some funds operate on the basis of maintaining a fixed ratio in various asset classes (percent stocks to percent bonds…). Periodic rebalancing of the ratios forces them to buy the asset class that has done poorly and sell the asset class that has done well. It seems that rebalancing predominantly takes place towards the end of the year. Surely there is someone here that could confirm or refute this.

Scott Brooks offers:

IRA fund flow is bigger towards the end of March thru about April 20th or so than it is in December (I say April 20th because the envelope the IRA deposit check is mailed in need only be post marked April 15th).

One can also look at index reconstitution as issues are dropped and others added to the indexes. However, this has the greatest effect on the smaller issues (smaller in terms of cap weighting). Most larger capitalized stocks are going to stay in the index and could be bought in an effort to rebalance a portfolio fund back into the index weighting.

As a result, the index funds have to go thru a flurry of rebalancing, selling the issues dropped from the index and buying those that are added….and proportionalizing those stocks that stick (again, mainly the largest capitalized issues).

Something else to consider (for both money managers and individuals) …

Stocks that have a loss are often sold to realize capital losses to offset the fact that …

Stocks that managers or individuals feel have run their course and have a gain are sold.

Another phenomenon that occurs are RMD’s (Required Minimum Distributions) for those with qualified money that are over 70. This is a forced sale for no other reason than realizing taxable income. What’s interesting is that this now becomes money in motion and as a result, opportunity to invest in other areas. As the baby boomers age, this will become more and more of a factor … especially since such a large number of boomers bought into the myth that they will retire in a lower tax bracket than when they were working.

I’m sure there are are many other reasons that our resident bond mavens and options experts (as well as anyone smarter about the market than I) could add to this discussion

An Anonymous Contributor says:

In his post Kim Zussman wrote that:

Looking further at the same monthly data, December moves seem large compared to the prior 11 months. To check this (and eliminate effects of sign), for each year I looked at the ratio of absolute values:

|Dec ret|/|J-N ret|

One would expect each month to contribute something like 1/11 of the return of the prior 11 months.

No, one wouldn’t. Since you already have all the data, go ahead and look at every month relative to the other eleven months of the same year (or to the preceding eleven months, it won’t make much of a difference). My own back-of-the-envelope calculations with unadjusted data show a mean of about .22 over all months, with December being somewhat below average.