STATISTICS Oex Options Expiration Trading Patterns by David K. Moy
There's a statistically significant bias in the weekly trading of the Standard & Poor's 100 index, in that the week that options expire is noticeably stronger than the week that follows. Statistics that check for significantly different means and ratios can help options traders detect these patterns. Not only that, options expiration patterns can also be useful neural network inputs.
Can the tail wag the dog? It can, if the tail is the expiration of Standard & Poor's 100 index options and the dog is the stock market. These contracts cease trading on the third Friday of each month. During that particular week, the market is often volatile with an upward bias. The following week, however, is often a reversal with a downside tendency.
Figure 1 shows that the average weekly change for the S&P 100 (OEX) was a gain of 0.62 points for the period of January 17, 1986, to May 12, 1995. The biggest difference is the 2.03-point average gain during options expiration versus the average loss of 1.62 points in the week following. I chose the OEX to look at instead of the S&P 500 because the options are more actively traded. Further, I decided to analyze point changes instead of percentage changes because striking prices are governed by absolute rather than relative changes in the OEX.
To begin with, the hypothesis is that differences exist between the expiration, post-expiration and nonexpiration-affected weeks, while the null hypothesis being tested is that there are no differences between the expiration, post-expiration and nonexpiration-affected weeks. If statistically significant results are uncovered, then the null hypothesis would be rejected.
The t -test is a measure of the likelihood that observed differences in the averages (or the means) is due to chance. If the t -statistic is a large positive or large negative number, then one can be fairly confident that differences are meaningful. A 95% confidence level, or alpha, is typical; this means there is still a 5% chance that the apparently significant large (or very negative) t -statistic results are due to simple bad luck. Statistics can't prove a particular viewpoint beyond any doubt, but they can suggest what is probable or improbable.