What Spread to use for BackTesting

• 04/29/2021 at 9:50 AM #168354
  deletedaccount051022

  Hi,

  I was wondering what spread methodology people are using in their backtesting so wanted to create this post to hopefully gather peoples feedback and ideas.

  For testing purposes I’ve been trying different approaches in my never ending attempt to reduce the minor differences between testing and production.  One idea I’d considered was a weighted spread approach based on data from the broker (IG) where I trade CFD GBP Contracts , but fear this might be too low where a strategy is not constrained by start and end times.  So am looking for inspiration please.

   

  Thank you very much.

  

  deletedaccount051022

  Participant

  Topics: 26

  Replies: 139

  Been thanked: 99 times
  04/29/2021 at 10:13 AM #168359
  GraHal

  I use worst case scenario, so for example, on DJI worst case is 4.8 (apart from market closed and flash crashes) so I always use spread = 5 for DJI.

  Which are you concerned about …

  1. Trades not being executed on backtest due to your use of  incorrect spread?
  2. Profits being overstated due to your use of too low a spread on backtest?

  My view on 2. above is … if your strategy is resulting in such a low overall profit that spread value is a big factor in low profit, then you should rework the strategy to get higher profit overall.

  1 user thanked author for this post.

  

  GraHal

  Participant

  Topics: 219

  Replies: 9000

  Been thanked: 2289 times
  04/29/2021 at 11:43 AM #168371
  Vonasi

  I’d suggest testing on best case scenario and worst case scenario too and a few spreads in between. This way you get to see the best possible result (knowing that you will never ever achieve this) as well as what sort of spread might break your strategy.

  If it takes a very big spread to break your strategy then the actual spread becomes rather insignificant and you can then concentrate on robustness testing to ensure that you haven’t just got a curve fit strategy.

  1 user thanked author for this post.

  

  Vonasi

  Moderator

  Topics: 192

  Replies: 5088

  Been thanked: 1811 times
  04/29/2021 at 3:16 PM #168396
  deletedaccount051022

  Thank you GraHal for sharing what you use, I like that idea.

  In response to your question, I am looking to align that actual profits I make as closely as possible to the results in baktesting.  I run various strategies live, and each week I compare the actual results to those shown on the backtest within PRT, and there are occasional differences in profit/loss for each of the strategies.  The backtesting is always slight higher, no more than 5-10%.

  1 user thanked author for this post.

  Kovit

  

  deletedaccount051022

  Participant

  Topics: 26

  Replies: 139

  Been thanked: 99 times
  04/29/2021 at 3:19 PM #168398
  deletedaccount051022

  Thank you Vonasi for sharing your ideas.  I like the worst case approach and will consider adopting this in future testing as part of the suite of tests.

  I am currently running this strategy which I shared previously on a live account and there are minor differences of around 5-7 points on one trade out of every 5 on average.  This tells me I am not allowing for enough of a spread.

  https://www.prorealcode.com/topic/improving-strategys-trailing-stop-loss/

   

  

  deletedaccount051022

  Participant

  Topics: 26

  Replies: 139

  Been thanked: 99 times
  05/01/2021 at 12:54 PM #168508
  vschmitt

  Hi @samsampop,

  I define a spread as two times greater than the habitual spread while the optimization of my trading strategy. After that, I process the spread increase test to see how my trading strategy reacts. If the strategy becomes losing too rapidly after a little spread increase, I consider it is over-fitted. If the strategy is insensitive to the spread increase, I consider it is too under-fitted.

  https://artificall.com/prorealtime/validate-a-trading-strategy-thanks-to-stress-tests/#Simulation-of-spread-increasing-on-Probacktest

   

  Be careful if you set a too different spread than the real spread because the defined spread will impact the stop loss and target exécution. That may produce misleading results of your backtest and distort the behavior while the real execution of your strategy.

   

  Regards

  2 users thanked author for this post.

  GraHal

  

  vschmitt

  Participant

  Topics: 17

  Replies: 61

  Been thanked: 24 times
• Author
  Posts