Walk Forward Efficiency

I've been reading Robert Pardo's book "Evaluation and Optimization of Trading Strategies", in which he discusses Walk Forward Analysis at length.

He states that a reliable measure of strategy robustness is the Walk Forward Efficiency (WFE), which is simply the ratio of the out of sample profit (or other performance measure) to the in sample profit (ie the profit generated on the optimized training data set).

In Zorro's WFO performance report, we currently get the out of sample results. Is there any easy way to get the in sample results and compute the WFE? Or would I need to set up individual optimization runs that correspond to the training periods used in the WFO?

Is this a feature that is worth including in a future Zorro release?

We have tested this some time ago with our strategies, and found that the WFE was not correlated to strategy robustness. Some quite robust strategies had a low WFE and unstable strategies had a high WFE.

That's really interesting, and thank you for sharing your findings. I will read Pardo with a somewhat more critical mind!

I have found that sometime insanely high WFE values are caused by wrong sign management in the WFE formula.

The WFE formula is:

out of sample annualized rate of return / in sample annualized rate of return

Now... the point is that if we choose (by error or by will) a negative in sample value we're going to mess things up because if the out of sample profit will be positive the result will be negative (bad)... and that's wrong since the the out of sample is good. Vice-versa if the out of sample is negative the global WFE sign will be positive and thats also wrong.

The point is that we shouldn't never use negative in sample return while using only the sign on the out of sample in the WFE formula (note: I'm not saying to use the ABS of the in sample profit, just never ever choose negative in sample profit).

I wrote that because I have found that a couple of tools I've used to manage WFE if forced to choose not the best profit but, let's say, the best dd% can naively pick a negative in sample profit ruining in fact every WFE calculation.