I've been thinking about the issue of statistical significance. This has been brought up on another site where it was pointed out that one needs to analyze the statistics to determine the likelihood that a given performance record is not due to random chance.
So what I've been thinking is this: Nine years elapsed from when Bob went fully invested until his model turned "unfavorable." Once he had gone fully invested in January of 1991, what is the probability of the next call occurring within two-and-a-half months of the top, as it did?
I figure that with one newsletter a month, he had to consult his model at least once a month to determine whether it was still favorable, or if it had turned unfavorable, so a call based on random chance had as much chance of happening in any month as in any other month. January 1991 through January 2000 was a period of 108 months, so that call had one chance in 108 of occurring in any one month, for a probability of 1/108. The chance of it happening within +/-2.5 months of the top, which is roughly the five month period from January 2000 through May 2000 is five time that, or 5/108 = 4.6%.
The next time his model had a change in position was March of 2003, whereas the bottom was in October of 2002, so he got that one within five months. January 2000 to March 2003 is a period of 39 months, so the call had a 1/39 chance of occurring in any given month. The chance of it occurring within a +/-5 month period of the top, i.e., a ten month period, is ten times that or 10/39 = 26%.
The probability of both those events occurring is the product, or 4.6% x 26% = 1.1% probability of those two calls coming as close to the actual top and bottom as they did by random chance.
Valid analysis or science fiction? You be the judge!
