New Strategy Cruncher user Mugu M. wrote in with an excellent question (shared with permission):
Mugu M:
Does it make sense to send the Longs and Shorts separately into the Cruncher? For example, when I’m using a momentum oscillator like RSI, extremes like 70, 30 will be used for overbought/oversold to decide buy/or sell..but the Cruncher returns one suggested optimal value that does not make 100% sense for both buy/sell.
Dave:
My default is to separate longs and shorts for optimization, assuming there are enough trades in the backtest to warrant this approach.
If you feel there aren’t enough trades in the backtest to get confidence in optimizing, then it is certainly a viable approach to keep them together.
But combining them does cause weird situations like this one for RSI.
The issue is that the RSI for the long side is looking for a value of 70, but the equivalent value for the short side is 30.
This is a perfect example of a column that isn’t normalized, so you’re not giving it the best chance for the cruncher to detect it for importance and therefore improve your strategy using it.
Here’s a simple way to create a normalized version of RSI across direction just for this very purpose.
Add a new column called directional_rsi.
For longs, directional_rsi is the same as rsi.
But for shorts, directional_rsi should be 100 - rsi.
For shorts, a value of 30 for rsi has a value of 70 for directional_rsi.
Now, the Cruncher can measure extremeness across both directions and weed out the weaker trades.
Great question, Mugu – thanks for sharing!
-Dave
P.S. Are you interested in creating your own strategies? If you’ve got a backtest, you owe it to yourself to try the Strategy Cruncher.
Here’s what one user said recently about it:
Matt: “By using the Strategy Cruncher, I can very easily find the very best filters for any type of strategy backtest without being concerned about over-optimization. Last week, in a mean reversion strategy, in 4 iterations of the cruncher, I found filters that reduced drawdown and increased overall profit by 10x.”
