Walk forward but more versatile

• 05/08/2019 at 11:51 PM #98004
  Paul

  Just a thought…

   

  First a walk forward test with say 2 variables (x&y) so each period has an optimum parameter.

  Next you could lock that graph and values for each period and add another global “z” criteria and see the results & chart update (and maybe optimise that new “z” criteria) or use the locked values with a different “z” value in a new chart to compare.

   

  Does that add anything interesting?

   

  Or would you say just add that “z” fixed parameter before and then do a walk forward test.

  And it’s possible to run multiple WF (x&y) test at the same time with each a different “z” parameter. Even of course a WF with x&y&z is possible.

  The way it is now.  But would the results be the same?

   

  I ask this sometimes I do a forward test and then questions what would happen if I changed that “z” parameter. Then I have to forward the test again which is time consuming.

   

  Daft question or does it make sense somewhere?

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  05/09/2019 at 7:20 AM #98013
  jebus89

  So if i understand u correct, u have 3 variables but ur only walk forward testing 2 of them and then u think “what would happen if i change variable #3?”

  I would say just walk forward all 3 variables at the same time, unless u are sure u will not change #3, if it is constant then i wouldnt bother testing it.

  1 user thanked author for this post.

  Paul

  

  jebus89

  Participant

  Topics: 118

  Replies: 586

  Been thanked: 238 times
  05/09/2019 at 2:25 PM #98083
  Paul

  If values are locked for all the periods in the chart and then changing “z” it may create a better looking chart.

  But real trading results will probably be completely different.

  Indeed, or running multiple WF test with different “z” parameter at the same time or just all 3 at once!

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  05/19/2019 at 5:03 PM #98849
  Jan

  Hallo Paul,

  are you looking for automated optimalisation for 2 variables x and y, which are obviously overfitted, by a variable z  which would then result in an acceptable robust system ?

  Presumably the million dollar question.

  It is very time consuming, but if you optimise for variable x and y for each consistent set of data/period/month, write down the optimised setting for x and y with testing IS 66% and 1 OOS 36%, doing this at least for 12 periods/ months, maybe you can see a pattern of the actual prices per period/month.

  Maybe one set of data/period/month is trending, which give optimised settings A for x and Y,  and it is comparable with another trending period/month with quite simular settings for X and Y. And other periods appear more or less non-trending, and here the setting for X and Y are more or less the same in those periods.

  Maybe the trending indicator like the ADX will show trending and non trending periods in those monthly periods, and can be used as a variable Z, to compensate settings X and Y, with multiplying or dividing the settings X and Y with variable Z

  Also you can do a bigger set of data test at once, where you look which periods does not perform well with the overall optimized setting, and have a closer look at the price-developments(trending/non-trending) in those periods

  It is an quit abstract story above, I am aware of it, it is difficult finding the Z-factor.  Making it work is another thing !

  KR Jan

   

  

  Jan

  Participant

  Topics: 10

  Replies: 90

  Been thanked: 29 times
  05/20/2019 at 12:54 PM #98887
  Paul

  The idea was to use it for the exit.

  i.e. scenario one trade a day which closes each day.

  What would happen to the WF chart with saved values x&y for each period, if I changed the trailing stop / stoploss or add other exit criteria afterwards (z)?

  Anyway, I put that idea in the bin.

   

  Gr Paul

  

  Paul

  Participant

  Topics: 40

  Replies: 812

  Been thanked: 660 times
  05/27/2019 at 2:50 PM #99531
  Jan

  Hi all,

   

  we all know the problem of overfitting,  having an optimised setting, but then the market movements structure changes overtime and the setting is much less profitable.

  Does anyone of you a daily or weekly optimalisation during time ?  So adjusting the settings of your life strategy after some time as a result of new optimalisation runs ?

  Looking forward for your experiences. Thanks in advance.

  

  Jan

  Participant

  Topics: 10

  Replies: 90

  Been thanked: 29 times
• Author
  Posts