Your suggestion seems to be very solid. I have always suspected you should only deploy systems with currently rising equity curves. However, since you have far more experience with systems design than I, I have a question. I have observed that when your run momentum pullback systems on the stocks, there are times when the pullback systems have 3-6 mos of rising equity curves and then they chop for while and lose money. Then they work again. I suspect the same thing happens to mean reversion systems. So would it be better to have a few "signal" systems running but not deployed. You would then monitor the equity curves and see whether we were in a momentum market or a reversion market. You then deploy the best systems you have for that environment. ... Or is it better to develop a portfolio of systems that are designed to smooth out your equity curve as best you can and not try to guess what the market is doing.
I had a typo it should be -Drawdown, this way it becomes a positive number for good systems, otherwise things look strange.
In the commodity world where you only have limited markets to trade , the basket approach to smooth out the curve works best because the system switching can create curve fitting issues, (just switching to the best one). I often develop systems where the parameters are adaptive or the logic allows the system to adapt, for example using an average swing low or high level in RSI over the past 4,5 cycles to define overbought and oversold. In stocks the switching concept might work because you have 1000's of stocks and possibly millions of trades, this reduces the risk of curve fitting.
Interesting, would you be up for developing an sp model? If I recall correctly, it is not included in the basket.
You are correct, it not in the basket. I can do a opening range breakout index model , after we finish this series.
Great look forward to it. I'm finding since '02 most models holding overnight looking for follow through are performing adversely. They tend to do much better exiting on close.
I have rerun this optimization using my custom optimization factor. I also expanded the parameters I am running over. Please take a look at the workbook I created. The optimize factor column is my weighted score. Please analyze these results and tell me which ones you think are best. You need to balance using parameters which had similar results, scores from neighboring sets of parameters with overall high scores.
PS, I will publish my analysis, tomorrow night or during the day on Thursday. Over this weekend we will discuss layering the entry methodology on top of this trend detector.