Walk Forward Analysis

Put simply it is just breaking the data down into in sample and out of sample periods and then comparing the results from one in sample period with its out of sample period to give an idea of strategy robustness and likely hood of it working the same going forward.

In PRT you can also use it to find optimized variables. You are looking to see that after the optimization run each IS and OS pairing has similar variable values. Some people use this to fine tune their finished strategy regularly by re-optimizing and using the last set of values as these worked best on the latest data.

Personally I just use it for straight forward IS and OS testing for robustness. I fix all variables as to how they would be in the final strategy and then set a ‘dummy variable’ from 1 to 1 and turn on WF. By running various tests with different length IS and OS periods I have a good idea of how fitted the strategy is.

Hope this helps.

Correct it is – and it is the best sort of IS and OOS.

What walk forward does is break the historical data that you have down into say 5 IS periods that it optimizes the variables on and then it applies the best performing variables at fixed values to the attached connected OOS period. It will do this for all 5 IS and OS pairs. Then it adjusts the results so that even if the OS data period is shorter than the IS the results are comparable.