i was curious to get any opinions on what a minimum # of trades you might expect to accuratelly judge the accuracy of a system? Obviouslly more the better but would you expect a small # such as 100 trades adequate or do you think it requires more than that?
30 trades or more should represent a "statistically significant" set; but those should be trades that occur "out-of-sample" (i.e., occuring on forward data not seen by the system during optimization, etc). Almost any system will look profitable in a simple backtest. Just as important, your evaluation should include more than 1 forward test period. (Walk-forward testing is what I am suggesting here). saxon
If memory serves (since I am far too lazy to go looking for my old statistics texts of long ago), 30 is a "rule of thumb" number. The proper sample size is dependent on the variability of the data set. Therefore, the greater the variability in your trade results, the larger the sample size should be in order to be representative. I have absolutely no recollection of the appropriate formula in this regard. Of course, I could be way off here. If so, please feel free to correct me.