When optimizing strategy parameters, you look for the parameter set that generates the best objective. I found that the quality of the parameters in walk-forward testing largely depends on how you choose this objective: Profit to drawdown ratio - worst results Sharpe ratio - better results Gross profit - best results I find this a little odd, as we normally go for a good profit to drawdown ratio and not for the highest gross profit. So I would expect that an objective based on the P/D ratio gives the best results. But it's just the other way around. This seems independent on the strategy, as I use a compound strategy with many trade rules and different assets. Maybe the reason is that the gross profit is calculated in the least complicated way, and thus the strategy parameters have the most direct affect on it? Has someone else made experiments with different objectives for optimizing?

Gross profit is a linear function of trade PnL. Profit to drawdown ratio is a non-linear function of trade PnL. It may be that the optimum in this case is a local optimum or all values are local optimum values and global optimum does not exist. This is a complicated subject. Good question though. It also depends on what kind of optimization you perform. Check this blog for details http://bit.ly/oNFi5S If you are using multiple indicators then optimum values do not have any significance IMO.

Profit is certainly a linear function of trade profit, but it's a highly nonlinear function of the parameter vector. But I think you have a point - P/D ratio and Sharpe ratio are possibly "even more nonlinear" and thus give worse results. On a side note, I looked into your blog and found your price action lab software quite interesting. Are you generating trade rules with a decision tree algorithm? I'm asking because I tried that also some months ago, but abandoned it temporarily because it didn't give the quick success that I hoped for.