Best Approach for Walk forward testing

• 01/22/2019 at 11:20 AM #89318
  O-jay8

  Hello All,
  I have a question regarding what is the best approach to use the walk forward analysis?
  Or better, who favours what approach?

  If one has an overall strategy (not optimised yet) with set parameters.
  Is it better to:

  1: start IS/OOS optimisation, e.g. 70/30 ratio, the old way. Afterwards see if OOS is similar to IS. If yes, test WFA for several repetitions to check the robustness.

  2. start WFA analysis directly with one repetition e.g. 70/30 ratio. After that test WFA for several repetitions to check the robustness.
  There I am always coutious because of the look ahead bias. You can see already whether the OOS sample gets better or worse if you only test one or two parameters at a time.

  3. Test directly WFA with severeal repetitions. There it takes ages to run the optimisation due to the staggering number of combinations. (Depends of course on the number of parameters on uses)

  I read all the topics regarding WFA/WFT but couldnt figure out which approach would be the most efficient.

  Personally I prefer approach 2 but am afraid of the look ahead bias.

  I would like to know how you guys tackle/ master the WFA/WFT.

  Kind regards

  

  O-jay8

  Participant

  Topics: 5

  Replies: 207

  Been thanked: 285 times
  01/22/2019 at 1:07 PM #89328
  GraHal

  O-jay8 wrote:

  the old way

  Please can you elaborate a bit on what you mean by the old way? I think I know what you mean so it need not be a long explanation.

  Good question and bullets overall …  I’ve often pondered over the same and never come to a conclusion.

  

  GraHal

  Participant

  Topics: 204

  Replies: 8440

  Been thanked: 2049 times
  01/22/2019 at 2:51 PM #89335
  O-jay8

  GraHal wrote:

  Please can you elaborate a bit on what you mean by the old way? I think I know what you mean so it need not be a long explanation.

  Good question and bullets overall … I’ve often pondered over the same and never come to a conclusion.

  For example I have a period of 10 years available for my optimisation. So I optimise only 7 years in sample for all my parameters. Ratio70/30.
  If the equity curve, Risk reward, Drawdown etc looks good then I check whether the following 3 years are similiar to the previous optimised 7 years.
  Jebsus89 gave a good explaination here: https://www.prorealcode.com/topic/how-i-create-strategies-and-optimize/

  

  O-jay8

  Participant

  Topics: 5

  Replies: 207

  Been thanked: 285 times
  01/22/2019 at 3:40 PM #89340
  GraHal

  O-jay8 wrote:

  then I check whether the following 3 years are similiar to the previous optimised 7 years.

  Thanks, yeah that’s what I thought you meant.

  I do a variation … optimise on the most recent 7 years then check whether the previous 3 years are similiar to the later 7 years.

  Reason is so that my optimised values have a more recent bias (rather than a way in the past bias).

  

  GraHal

  Participant

  Topics: 204

  Replies: 8440

  Been thanked: 2049 times
  01/22/2019 at 3:49 PM #89341
  O-jay8

  @Grahal

  Interesting approach. How is that working out so far?

  

  O-jay8

  Participant

  Topics: 5

  Replies: 207

  Been thanked: 285 times
  01/22/2019 at 5:10 PM #89355
  GraHal

  Getting better, but I try so many different approaches that I haven’t yet come to a final conclusion which modus operandi to go forward with! 🙂

  I am trying lots (on Demo with Live disabled for the last 9 months!) while I wait for PRT Version 11, but at times I feel like I’m suffering from analysis paralysis! 🙂

  

  GraHal

  Participant

  Topics: 204

  Replies: 8440

  Been thanked: 2049 times
• Author
  Posts