Live market has entries (triggers) that back-testing does not- on the same code!

• 07/25/2018 at 2:26 AM #76657
  Stam

  Hi Forum,

  I have a system that has been running live for couple of weeks now. Every day I check it performance, and I also run an over all back-test to compare to past versions.

  I have noticed a very worrying thing, on the same code, the live account was triggered in where as the back-testing did not have a position open at all (not even few candles before/after). It is not a matter of live market slippage at entry and the like, it is actually getting triggered vs. not at all. That obviously change dramatically the ability to asses the the system performance (and yes I am not expecting past performance to be indicative of future performance :), as triggers into the market is clearly not always taken place at the approximate similar point.

  Have anyone encounter something similar? Am I missing something? I have even copied-paste the live running code into new trading system and run it again and the same issue happens.

  How can the back-test and the live market that run the exact same code give completely different entries where one does trigger and the other does not?

  Looking forward to your wisdom 🙂

  

  Stam

  Participant

  Topics: 2

  Replies: 5

  Been thanked: 0 times
  07/25/2018 at 7:31 AM #76660
  Vonasi

  Without the code it is difficult to say whether it is code based or data based. Sometimes a code can be simply effected by the start time – depending on the code of course. Start two identical codes on different candles and they may run out of synch, a weird time frame such as 3 seconds or 7 minutes can also cause this issue. The other issue might be that your back test runs on a fixed spread where as your real life test will have a spread whose width varies micro second to micros second and dependent on the time of day and also dependent on the market and its current volatility.

  

  Vonasi

  Moderator

  Topics: 192

  Replies: 5088

  Been thanked: 1811 times
  07/26/2018 at 2:25 AM #76758
  Stam

  Hi Vonasi,

  Thank you for your answer. The code is running on a one minute on the Dow, limited to the trading hours of the underline market, which mean that when running the back test outside cash market hours there is no affect to the exact minute starting time being off by couple of minuets. The spread is also a constant during these hours and so I run the back-test with that spread (I’d agree that if it was running on FX there could be spread changes around major news times).

  Any other ideas/thoughts/reasons? I am in a lost to why this happens.

  Thanks again for the help!

  

  Stam

  Participant

  Topics: 2

  Replies: 5

  Been thanked: 0 times
  07/26/2018 at 7:29 AM #76762
  Vonasi

  Any tiny change in spread will have your live version opening and closing at different times compared to your fixed spread back test version. This is why we must always forward test on demo! I think you will be surprised how much spread opens and closes up. Put the ‘bid ask’ indicator on your chart and put it on a very fast time frame and zoom in and you will see that spread is far from the fixed amount that you think it is.

  At the moment I am trying some tests on spreads to find out when they actually change:

  https://www.prorealcode.com/topic/when-does-spread-actually-change/

  IG say that spread should be 1.6 after the DJI opens but yesterday I observed it to be 2.4 several hours after the market had opened. Later nearer market close I eye-balled it and it was 1.6. That is a major difference and would mean very different entries and exits compared to a fixed spread back test.

  

  Vonasi

  Moderator

  Topics: 192

  Replies: 5088

  Been thanked: 1811 times
  07/26/2018 at 9:23 AM #76770
  Stam

  Thank you for your answer, I have went back and checked order by order, and it gets weirder then that, it looks like there is a black hole on the back test on the 17th and 18th of July trading days (18, 19 AEST), I have tested it on other markets – FX other indices, and same black hole… Nothing executed on these days on the back test no matter which market or program it was.

  Does this happened to anyone else???

  

  Stam

  Participant

  Topics: 2

  Replies: 5

  Been thanked: 0 times
  07/26/2018 at 10:04 AM #76775
  robertogozzi

  Those two days were the days the new MTF support was installed on demo accounts and maybe PRT were updating their servers and there might be some “hole”, which wasn’t on real accounts.

  That’s my guess, but I am not sure about that.

   

  

  robertogozzi

  Moderator

  Topics: 69

  Replies: 16517

  Been thanked: 3663 times
  07/27/2018 at 2:31 AM #76848
  Stam

  That is very interesting!! I am actually running the back testing on the live account but, I can see why that somehow get affected…

  Thanks so much for the info, it means bar these days, I can go back to relay on the back-test for the system performance.

  Much appreciated!!

   

  

  Stam

  Participant

  Topics: 2

  Replies: 5

  Been thanked: 0 times
• Author
  Posts