Hi, The idea behind walk forward is that you select the best combination of parameters in sample (IS), and then apply them to the next out of sample (OOS) period.
At the end of an OOS period, that OOS period becomes part of the next IS period and you repeat the process. Thus, the parameters to be used for each OOS period is awlays the best ones found from optimization of the preceding IS period. So, in answer to your question; You do not have enough data. You must end with an IS period such that the optimal values can be used to enter the next OOS period. That being said, the data you present seems incorrect, or at least incomplete. You are showing a series of (OOS, IS, OOS) which is not possible. How did you get the first OOS results without an IS period from which to determine the parameter values? Did you perhaps mean to write: IS: 20, 4, 4 <-- gives param values for next OOS OOS: 20, 4, 4 <-- based on preceding IS IS: 25, 3, 6 <-- gives param values for next OOS That being the case, then the correct values to use going forward OOS would be (25, 3, 6) since they were the best of the immediately preceding IS optimization. Mike --- In [email protected], "droskill" <[EMAIL PROTECTED]> wrote: > > Ok - so I have a system that has 3 parameters for optimization. When > I optimized it using backtesting, I get 3 parameters: > > 20,3,4 > > Now, I walk-forward optimize it, and I get: > > OOS: 20,4,4 > IS: 20,4,4 > OOS: 25,3,6 > > So my question is this - they all give decent numbers for returns, etc > - so how would you approach figuring out which set to use? > > Thanks! >
