No. of Recommendations: 14
My final conclusions on the 3 momentum screens.
The screens were backtested on GTR1. The output is the daily portfolio values for each of the 21 cycles of the 21 day holding period. It also gives the minimum, maximum, and average CAGR for the 21 cycles. Mathematically, the numerical average is nonsense. No cycle has this return.
The quandary is how to assess the return and risk statistics for the screen. We have 21 sets of returns and they range more than +/- 10% different. So which cycle should we look at to figure what our real-world investment in the screen will likely be?
We could look at the worst one, and figure that our actual results would be at least this. But that is probably overly pessimistic. In fact, the worst cycle in one of the screens is quite a bit worse than all the other 20 cycles.
We could look at the median cycle, but that is probably over optimistic.
I settled on using the one at the 25th percentile. For that one, 75% are better and only 25% are worse. So that is cautiously optimistic.
Because of the high volatility of these screens it is necessary to use some sort of timing. Without timing the drawdowns were in the range of -70%, with recovery periods up to 10 years.
I applied 7 different timing schemes to see which was best. And it turned out that it doesn't much matter which one was used. They all cut the drawdowns by more than half, with little or no drop in the CAGR. Even so, there were recovery periods as long as 2 years.
(I settled on one that is easy to compute; sell at 4% below the 52 week SMA of the Nasdaq Composite Index, ^IXIC, taken at the scheduled monthly trade date.)
The 25th percentile CAGR returns (with timing) in the periods:
1985-2026 2006-2026
RS-52wk 5 HTD 12: 25.3% 25.3%
PHL-52wk 5 HTD 12: 26.7% 24.1%
Overlap 0-5: 22.3% 23.2%
The weighted average number of stocks held in the Overlap screen was 3.2 stocks.
The range of CAGRs (untimed):
1985 to 2026 Average of top 5 cycles
RS-52wk 5 HTD 12: 24.1% - 29.5% 28.4%
PHL-52wk 5 HTD 12: 27.1% - 32.2% 31.4%
Overlap 0-5: 23.8% - 34.1% 33.3%
The overlap screen has a much wider range than the others. Nonetheless, what makes it worthwhile to do is that the returns at the upper end are much larger than that of the others.
