Question about Walk Forward Analysis

Hi,

With WFA, why do we start at the oldest data, optimise on a percentage of the data, then forward test on a percentage of the data, move forward and repeat.

Instead, why don’t we take a random slice of the complete history (say 20% of the data), optimise on that, then test on the rest of the data (skipping the slice), then choose another random slice, test on the remainder of the data, and repeat this many hundreds of times.

Seems that would give a better clue to how well a strategy copes with changing markets and how sensitive it is to parameter changes.

Thoughts?

Not a bad idea it is like a combination of some  MC functions and  WFM in a combined test.

yes, I wouldn’t call it Walk-Forward then, it is more like Random Optimization robustness verification method. But it is not difficult to add it there, I’ll add it to our tasks.

+1

+1,

This idea is awesome, seems like a good option to add to SQ4,

when at the WFA/WFO there could be a (RANDOMIZATION) option to check/uncheck,

when this option is checked the data % parts will be randomized,

when unchecked the data % will be in order,

this idea is a great feature for checking strategy robustness for sure!

Yes it is karish. Bravo mikeyc excellent idea

It could indeed help to sort and remove strategies before WFM  and should belong to the Robustness tests for automation purpose. 

We can take the IS-Validation period, optimize on that and test the optimized strategy on IS-Training and OOS 🙂