From the Performance Report page in the manual:

Prediction error: Estimated error of the results in percent. This is an empirical value calculated from trade parameters that affect the result accuracy, such as the number of trades and the deviation of sample cycles.

Could we get some more information about this? How is it calculated? What practical purpose does it serve?

Is it the same thing as standard error (ie standard deviation of trade results in backtest/sqrt(number of trades in backtest) ??

Thanks!

No, it has nothing to do with the standard error, and has no closed formula.

It is calculated from the performance parameters and test method, which each add a factor with a certain weight to the error estimate.

OK, thanks for the answer. And in terms of practical use, how is the figure interpreted? What does it actually mean? How can I use it to improve the performance of my systems?

I'm not sure what the figure is actually telling me, but if I got a prediction error of 18%, does that mean that my performance statistics actually lie somewhere within +/- 18% of the quoted values? Shouldn't the prediction error then be accompanied by some sort of confidence interval?

Confidence intervals only make sense when you can calculate them with a precise formula, which is not the case here. The prediction error is an estimate. An error of 18% means that your result is rather precise. At 50% it were rather unprecise, at 100% it's worthless.

Thanks jcl, those rules of thumb are handy. Can I suggest adding them to the manual?

Yes, will do.