I liked the smooth look of 32's equity curve, rather nice for a simple mean reversion model. But I wanted it to perform a little better, so I tweaked three things. First, in Settings I changed the Max Open Positions to 0 instead of 10. For unknown reasons, this lets the model properly hold up to 10 positions :-) which helps a lot here. Second, I wanted to add another Condition block to the Buy, such that the stock had to be on a good upswing overall. I used 1-month momentum ROC(20) > 1.5% for this. But this also cuts down on the number of trades available more than I'd like. So third, I went to the Qualifier for N Times in N Bars and loosened it up a little, to 12 bars instead of 8 bars in the original.
What did I get? Overall CAGR is up, in line with SPY now. Exposure is only slightly higher (still under 50%). Sharpe ratio has improved a bit to up around 1.0, and MDD is about the same (a little less than SPY). Nice!
But... hmmm, did I overfit this simple strategy? Well, I did play around with some of those params, and tried and rejected some things (like the 10 month momentum should also be positive), so technically, maybe I did. So what happens when we extend the timescale to the full 20 year backtest? Definitely a degraded performance relative to the original. Model 32 shines from 1999-2009 compared to this overfit misfit. You can run the numbers yourself if you want to see just how humbling this simple exercise was for me. At least it beats SPY. But that's about all you can say for it.
APR: 10.75% • WIn Rate: 65.69% • Sharpe: 1.00