3 months and your equity curve changes completely…

• 04/07/2020 at 8:32 PM #125140
  Roberto S.

  Hi, I’m using a system since October 2019.
  I was happy until Coronavirus hit.
  I run the walkforward to get the more recents params to use in the following few months.
  Now I run a test again because I was testing an idea to prevent entering on volatility peaks we had in march.
  I had a very bad equity curve running from 2018-04 (1st days) to 2020-03 (the end) (see from_2018-04.png screenshot).

  I replaced the starting date with the one used in the previous test I made in January (so starting from 2018-01) and the equity is totally different (see from_2018-01.png screenshot).

  I’m using optimization through “Gain” (if the translation from Italian is correct).

  Any suggestion to make the equity not so dependent on the period used for testing? 3 months are not so much…

  Regards
  R

  EDIT: I fixed the end date of test to the right one 2020-03. In the 1st version of this topic I put the wrong one 2019-03

  

  Roberto S.

  Participant

  Topics: 10

  Replies: 17

  Been thanked: 4 times
  04/09/2020 at 2:23 PM #125443
  Roberto S.

  Probably my question is not clear.

  Let m know and I’ll try to explain myself better.

  Regards
  R

  

  Roberto S.

  Participant

  Topics: 10

  Replies: 17

  Been thanked: 4 times
  04/09/2020 at 2:54 PM #125449
  GraHal

  Yes I have read it a few times and still not quite sure what you are having trouble with?

  As few words as possible is always the best. 🙂

  Roberto S. wrote:

  Any suggestion to make the equity not so dependent on the period used for testing?

  ‘BackTesting’ or ‘Walk Forward Testing’ or ‘Live Forward Testing’?

   

  1 user thanked author for this post.

  Roberto S.

  

  GraHal

  Participant

  Topics: 204

  Replies: 8440

  Been thanked: 2049 times
  04/09/2020 at 4:33 PM #125462
  Nicolas

  The equity curve is different if you start your analysis from a different date. Keep in mind that walk forward analysis divide the periods into X occurrences divided into IS and OOS periods..! In each occurrence, only the In-Sample period is optimized and the Out-Of-Sample period serve for testing the best settings found.

  So, in brief, if you start the backtest earlier in history, all the IS periods will be different (moved backwards) and hence the very different results at the end of the chosen backtest date. Hope I’m clear 🙂

  2 users thanked author for this post.

  Roberto S., GraHal

  

  Nicolas

  Keymaster

  Topics: 81

  Replies: 20830

  Been thanked: 4744 times
• Author
  Posts