The answer to the OP is it depends on the length of the optimization period. For example getting a Profit Factor of 99 is obviously an overly curve fitted skewed result. The problem with NinjaTraders optimization is that you can't tell it to optimize based on multiple criteria. For example, ideally you want to optimize on both the Total Net Profit and Profit Factor. The reason you want to do this is because sometimes optimizing on only one or the other will overly curve fit for that optimization period. So this week I'm going to test/compare the optimization results for both Total Net Profit and Profit Factor and ideally use the nearest or matching settings from both optimization criteria. I hope they improve this in NT v8.
If you just use net worth the performance is biased towards the latter end of the time range being optimized (assuming fairly exponential growth). To avoid this I decided to use the mean normalized increase. This allows an early increase of 4000 to 4400 to have the same impact on the algorithm performance as an increase from 40000 to 44000 when the algorithm has more capital to throw around. However, if you're optimizing an algorithm, biasing the performance towards the latter end of the time range being optimized might be beneficial. This would yield algorithm parameters that are biased for the most recent market behavior. Anyway those are my two cents.
Thanks Byron, I already took that into account, I use a steady # of contracts to trade to avoid distorting profits in certain timeframes.
EliteTraderNYC Would you explain what you believe PF is exactly and how you exactly calculated it? Thus far neither of you explanations are making sense.