Some weeks ago I decided to see if I could devise a rule that would serve as a tactical market timer. I met with partial success. By partial success it works with some screens, some benchmarks, and one has to monitor and change the boundary parameters to fit the market. The good thing is that it tends to be quite obvious (322.28% vs. 171.57%) when a combination works. It gets you out of the market early and back into the market early with a net positive impact. The rule is “Between((SMA(8,0,#BENCH)/SMA(15,0,#BENCH)),0.96,1.0125)=1”.
Results from real money screen showing statistics with the rule, without the rule, and the Russell 2000 Value index as benchmark over the period 01/02/2021 to 06/09/2023 are;
| Statistics | With Rule | Without | IWN |
|---|---|---|---|
| Total Return % | 322.28 | 171.57 | 12.82 |
| Ann Return % | 81.09 | 50.96 | 5.10 |
| Max Drawdown % | -22.08 | -28.98 | -25.68 |
| Sharpe | 2.03 | 1.24 | 0.08 |
| Sortino | 3.01 | 1.79 | 0.11 |
| StdDev % | 28.40 | 32.47 | 21.62 |
| CorrelBench | 0.72 | 0.82 | |
| R-Squared | 0.53 | 0.68 | |
| Beta | 0.95 | 1.23 | |
| Alpha % | 74.33 | 46.18 | |
| Avg Holding # | 8.11 | 10.00 | |
| Turnover % | 21.00 | 14.00 |
The screen is quality focused with an industry timing overlay and “Core: Value” a the Ranking System. The details are:
ROI%TTM > ROI%PTM*0.98
ROI%TTM > 8.5
OpMgn%Gr%TTM > 9
SMA(5,0)/SMA(9,0) < (SMA(5,0,#INDUSTRY)/SMA(9,0,#INDUSTRY))*1.025
SMA(5,0)/SMA(9,0) > (SMA(5,0,#INDUSTRY)/SMA(9,0,#INDUSTRY))*0.96
Rank System is “Core: Value”
Universe is liquity constrained NOOTC based excluding ENERGY Sector with about 3200 stocks
10 stocks selected
0.25% slippage
Rebalanced weekly
Rank Tolerance is 0.6%
Observations/Implications:
- Control check by randomly assigning weeks to do a forced recalculation produced just noise.
- Control check by stepping out of the market during those weeks was also just noise.
- Using the timing rule against the benchmark increased the period return to 52.3% from 12.8% or 4.1x.
- Using the Russell 2000 and the Russell 2000 Value as benchmarks works while Russell 2000 Growth doesn’t against the screens I have tested. I haven’t worked with other benchmarks.
- The upper bound parameter does most of the work and needs to tuned to the market environment by regular review. So running a long term backtest isn’t practical.
- Working spreadsheet with details is loaded at the end.
Please enjoy the research. I was a bit amazed at the impact.
Cheers,
Rich
Random Experiment 1.xlsx (27.7 KB)