Very concerned about IG’s data and/or PRT execution of data

• 03/14/2024 at 7:13 AM #229743
  Absolute

  Hi guys,

  I have been a user of both companies services since 2017, making use of PRT to provide me with order information on my proprietary developed algorithms executed by the broker, IG.com

  Since, Monday, the 11th of March 2024, POST the US daylight savings implementation of Sunday the 10th of March 2024 my algo’s running on “2 days” candles backtest and order book differs dramatically from last weeks, based on tick-by-tick data for the last almost 13 years, starting 30 July 2010.

  No changes have been made to the code of any of the algo’s during the last 2 months so the only other variable is the “2 days” historic candle data. My orders only execute on the close of the “2 days” candle data.

  The same applies to the live- and demo data. I have spent a lot of time this week to see whether the fault might be on my side before publishing this topic for support from IG, PRT and the ProRealCode community.

  The CFD markets that I trade are on the US Tech 100 Cash, US 500 Cash and Wall Street Cash indices.

  I have in the meantime applied with InterActive Brokers (IBKR) for an account to have a control mechanism.

  I really hope that this issue has a solutions because “algo traders” success is 100% based on the integrity & consistence of data.

  Regards,

  Niel (very concerned)

   

   

  1 user thanked author for this post.

  Bel

  

  Absolute

  Participant

  Topics: 3

  Replies: 30

  Been thanked: 7 times
  03/14/2024 at 8:05 AM #229745
  Johann

  Hi Neil,

  Unfortunately this is a issue for me as well, I have been trading with PRT and IG since 2016. The back test is not reliable, even demo testing presents it’s own results and both of them differ from live trading. The only way you can develop your algos is via live testing and over a long period of time. Off course this will have costing implications. I do not imagine it will be different from IBKR? Perhaps someone can elaborate on this? Also take in to consideration your algos which worked on IG will not necessary work on IBKR…. according to some previous forum comment….

  

  Johann

  Participant

  Topics: 11

  Replies: 66

  Been thanked: 3 times
  03/14/2024 at 8:46 AM #229746
  PeterSt

  Johann wrote:

  Also take in to consideration your algos which worked on IG will not necessary work on IBKR…. according to some previous forum comment…

  Definitely.

  I have no comment on changed data since USA daylight saving. Of course it is an open door that the last candle of the day may me interpreted differently since … the past several months and which is now noticed ? People with 1 day candles should notice similar ?
  But also consider you having changed Time settings. Also consider a difference between V11 and V12 (hence test that !). But that will be for Backtesting only. Live will not be different between V11 and V12.

   

  Absolute wrote:

  I have in the meantime applied with InterActive Brokers (IBKR) for an account to have a control mechanism.

  That will be totally useless, because both ARE different. Better get used to that in advance – and BTW nothing to complain really.

   

  Johann wrote:

  The only way you can develop your algos is via live testing and over a long period of time. Off course this will have costing implications.

  This is true (both sentences). It is also true that Demo is relatively worthless, which leaves Backtesting vs Live. And *there* now is no difference at all. So all what happens there is misinterpretation hence pilot error. I know, because I have all my trades overlaid exactly in Backtest vs Live. But it takes a lot of effort and code to get there.
  In practice both won’t work out the same after all because of all the thrown-outs. This is obvious. This is the same with IBKR (as often but for different reasons). In the end this can be eliminated to zero times thrown out, but it again takes a lot of time (mainly throughput) and more code again. For example, I think I have now mastered IBKR after a full year of trial and error (again 😉 ).

  On all there is (much) more to say, but this could be off topic for this thread.

  ---

  Niel, your claim looks justified. But you should be able to point out where the differences lay. This starts with your backtesting being 100% equal to live.  Try to work on that. And you can, because you can use your Live trades in retrospect. Work as long as it takes to get it done. Know that it can be done (because I can). After that it will be dead-easy to find the culprit like with your observed DLS.
  And let’s hope that more people have the same experience so we can find common denominators and have more search power.

  Regards,
  Peter

   

   

  

  PeterSt

  Participant

  Topics: 73

  Replies: 2501

  Been thanked: 805 times
  03/14/2024 at 2:47 PM #229765
  Absolute

  @PeterSt wrote: “This starts with your backtesting being 100% equal to live.  Try to work on that. And you can, because you can use your Live trades in retrospect. Work as long as it takes to get it done. Know that it can be done (because I can).”

  Peter, the code is 1500 lines, multi-strategy with 3 neural networks that feeds into a risk manager that creates the orders every second day…I do not think I will be able to dissect the specifics relating to each order.

  

  Absolute

  Participant

  Topics: 3

  Replies: 30

  Been thanked: 7 times
  03/14/2024 at 5:33 PM #229784
  PeterSt

  Niel, do you even recognize the Daylight Savings in your code(s) ? If you don’t do that, nothing can work.
  Maybe that’s one of the reasons your code is only 1500 lines. Haha.

   

  1 user thanked author for this post.

  Absolute

  

  PeterSt

  Participant

  Topics: 73

  Replies: 2501

  Been thanked: 805 times
  04/05/2024 at 2:29 AM #231075
  Absolute

  Hi All,

  I just want to report that the IG / PRT “2 day” candle data reverted back yesterday, 4th of April 2024 to the post daylight savings of the 10th of March 2024.

  I still can not explain this phenomenon but keeps on tracking it.

  Regards,

  Niel

  

  Absolute

  Participant

  Topics: 3

  Replies: 30

  Been thanked: 7 times
• Author
  Posts