No. of Recommendations: 16
But in 6 years the optimal parameter values could have changed.
...
If the optimal parameter values have changed, they were just tuned for the data. Which means they are useless going forward.
It's not useful to look at short periods for a signal that triggers only once per year, but you can look at very long periods.
For relatively recent history, since the 1970s say, then a somewhat longer "timeout" on the signal like 115 or 120 days would have worked a bit better, as the key major tops have been pretty rounded except for the crash of '87. It would have avoided a couple of whipsaws. This was analyzed in some detail about 15 years ago.
But the original signal was built using data back to 1916 and 1930. The goal was extreme robustness for all kinds of market conditions. In older decades, markets were not just more volatile, but also more jagged and trendless. Being a bit more nimble worked a lot better. So the 99 day lookback was chosen as what would have worked best on the WHOLE period, sort of a compromise between the older jagged times and the recent trending times--after all, who knows what the market will look like in the next 20 years? Again, the idea being that for a signal this slow, there just aren't enough signals since the 1970s to believe that tuning to the smoothly trending markets since then would be a choice well supported by the data.
In fact, lots of different lookbacks still seem to add a lot of value. For example, "no new trailing-six-month high in the last 3 months" works pretty well for identifying stretches that are particularly clearly ongoing bulls, which is a "timeout" parameter of only 63 trading days.
Jim
