I didn't just make up the 12 week rule. What I did was compute the compound return from summer of 1987 to Dec. 31 1986, assuming that you remain fully invested in the S&P 500 except during some limited periods after a sell trigger. There were two free parameters: the % bullishness used as a sell trigger, and the number of weeks to remain out of the market after the sell trigger threshold is no longer exceeded. I varied those two parameters in a search of two-dimensional parameters space to find the maximum compound return. The results were a threshold of 55% bullishness averaged over two weeks, and a waiting period of 12 weeks. The reason I used a two-week average was to improve the counting statistics (since typically only a hundred or so responses go into a week's poll). I didn't want to average over too many weeks, since I found by computing the correlation function of the investor sentiment survey and the change in the S&P 500 that the negative correlation between bullish % and future change in the S&P 500 had a time scale of approximately 10 weeks.
The primary limitation of this is not that it is contrived (I don't think it is, because I used real historical data and nothing else) but rather that the historical data itself is limited. It includes only a few significant corrections. Before this year, there were only three occasions that the 55% two-week average was exceeded, and only two of those were actually precursors to corrections. Thus, what I did in effect was identify a method to maximize your return through the 1987 and 1996 corrections, which just happened both to have approximately a 12 week time scale.
There is nothing to say that the present correction can't do something different -- like a six-month downward slide. It's just that the two corrections that happened during the 1987-1996 period when I have investor sentiment data just happened to have approximately a 12-week downward slide.
So when does the present 12-week period end? Well, since we never actually got to the 55% two-week average threshold (the highest two week average was 53.5% for the 2 weeks ending 2/21 and 2/28) it's impossible to say. But I think it would be reasonable to start counting either at 3/1 or 3/14, when the two-week averages were lower than they were after they reached their peak on 2/28, as the first of the 12 weeks. That means the 12-week period would end either Memorial Day weekend, or the following weekend.
Again, the 12-week rule isn't something carved in stone. It's just what would have maximized your returns during the 1987 and 1996 corrections. This correction could turn out to be significantly different.
If there is anything fundamental behind the 12 week period, perhaps it has something to do with the psychology of the masses of investors -- how long it takes to go from euphoria to despair.
Percent/ Percent/ Percent/ Weekly
Date / Bullish/ Neutral/ Bearish/ Close
================================================
04-04-97 29% 40% 31%
03-28-97 27% 40% 33% 773.88
03-21-97 36% 42% 22% 784.10
03-14-97 34% 36% 30% 793.17
03-07-97 43% 37% 20% 804.97
02-28-97 63% 22% 15% 790.82
02-21-97 44% 36% 20% 801.77 |
