As a newbie to API's and the ability to backtest strategies I am pretty excited that I have my first strat compiled and ready to test. It was written by a professional for me and I want to complete the backtest portion on my own for experience. This strategy is based on a simple mean reversion day trade process that I've had a little success with on a very small sample size (roughly 30-40 trades manually). I realize this strategy is likely not profitable in the long run , but I want to go through the process of testing regardless.
The strategy that was written for me contains all of my components from signal to risk/reward ratios to trade management. I believe that running an initial optimization on all 11 (signals, stops, targets etc.) of these variables would produce some seriously curve fitted results. So I have been contemplating segmenting the walkforward process into 3 separate segments but I would like opinions on if that is an even worse idea. Ha!
- Initial segment of backtesting would be simply for the signal: I'd lock all stop and trade management variables and simply test and walkforward the "best" signal parameters. For this segment I would define "best" as the highest percent profitable trades, not net profit as is standard. Initial tests would be run on NQ and then I'd see if similar win rates occured on ES and YM to confirm.
- Second segment of testing would then be on the minimum risk/reward ratios and stop sizes. I would lock in the signal variables from segment 1 and focus on entry criteria. For this segment I would define "best" with more of the standard metrics like efficiency, max draw-down and return on capital.
- The last segment of testing would include the remaining trade management variables like break-even stops and target locations (in ticks). Like the segments, I'd lock in the optimum values from the previous segments and just focus on these. This is where I'd focus on net profit and the equity curve.
Basically, I'd like to confirm first that the signal I use is reliable. With my options trading background all I'm concerned with is probabilities. If I can confirm that a signal produces a certain result a certain percentage of the time then I think I can define a trade system around that. I'm just not sure if stitching together segments like this is exactly what the backtest program does on its own, or if it leads to a curve fitted result on its own.