Backtesting does not work which a great amount of time units

Forums ProRealTime English forum ProOrder support Backtesting does not work which a great amount of time units

  • This topic has 35 replies, 8 voices, and was last updated 2 months ago by avatarWim.
Viewing 15 posts - 16 through 30 (of 36 total)
  • #228351

    Thanks JS.

    Maybe the weirdness is something related to IG Accounts only.

    Do you exprience ‘no info showing under Long and Short’ as per my screenshots on the link below?

    https://www.prorealcode.com/topic/no-or-incorrect-display-of-drawdown-in-v12/#post-228221

     

    #228353
    JS

    Detailed report seems to be good in all cases…

    1 user thanked author for this post.
    #228355

    @JS
    I don’t see a separately listed drawdown for long and short. So you have a problem too.

    1 user thanked author for this post.
    #228356
    JS

    That’s right, there’s definitely a problem with the “DrawDown”… (recently)

    #228357
    Wim

    I had these missing long/short-drawdown-figures already in v11. But then it was rare, in general a restart of the backtest solved the problem. Currently, in v12, the frequency of these missing long/short-details has exploded, and restarting backtests only helps incidentally. I am running out of work-arounds, and getting better in Python.

    1 user thanked author for this post.
    #228359

    Hello

    I’ve been creating back testing strategies for a while now and test up to 200K, I find PRT like an errant misbehaving teenager and not that much better at V12. There are a few things I have found when back testing.  It seems to get into batches of failures where a strategy I am testing reports different results the more I test it, also it will give different results FOR THE SAME PIECE OF CODE AND DATES when run later! I get Around this by Either, (1) Turning off the detailed report and equity profile and running again; or (2) Closing down PRT entirely (I run it through IG) and starting it again, which seems to reset and overcome the issues. I though at first it was because my computer was old and so it could not handle the data, but changing for a new one only made back testing faster, not better. I now think either (a) it is somehow caching the data incorrectly at the PRT server when streaming back and forth over the internet, particularly if changing timeframes for the same code. Or (b) there is a drop out on the packet of data that is sent if the internet is slow and the PRT server buffers are not setup well enough to handle these dropouts, or are continuously storing / caching the incorrect data until I reset the Report / Equity Profile or PRT totally.

    Cheers – Mark

    1 user thanked author for this post.
    #228360

    It seems to get into batches of failures where a strategy I am testing reports different results the more I test it, also it will give different results FOR THE SAME PIECE OF CODE AND DATES when run later! I get Around this by Either, (1) Turning off the detailed report and equity profile and running again;

    I have experienced the same / similar to what you describe above.

    I find it’s worse the more versions (slight difference in code) of the same code that I am running and where I have 3 or 4 equity curves and positions showing on a Chart.

    #228363

    @Nikolas
    Do you now have any information as to why this is and when it will be fixed?
    Do we need to worry?

    #228364

    Yes – I think we need to worry. I would love to be able to TRUST PRT code and its systems; how do we know what is the right results?

    #228365

    I feel the same way… I’ve been using my systems for a long time and I know that they work well. However, the fine-tuning still takes place via a back test. And if that’s not right… how do we know what else is wrong or not working?
    Loss of trust is not good.

    #228366

    Have any of you noticed that we have to very careful after optimisation as to what variable values get submitted to ProOrder??

    Even with all values set as fixed values in the optimiser when we go to ‘Prepare for Automatic Trading’ the vales that show (in the Prepare for Auto Trading screen) are not (always) the values that we have set as fixed in the optimiser??

    Above depends on what ‘routine’ we adopt before for submission ProOrder!

    It was as above in v11 also!

    Isn’t it such a pity that PRT do not read all these nuggets of information we post on this Forum!!??   I say this as PRT have never once (to my knowedge) come back on this Forum and asked for more information etc?

     

    #228367

    Presumably the big institutions that have the money don’t use Autobots like PRT? Is it just us, the little fish, who want a piece of the cake but aren’t that important?

    #228368

    Hi everyone, I’ve been following the conversation here and totally get where everyone’s coming from with the backtesting hiccups on ProRealTime, especially with the hefty time units we like to throw at it. Trust me, I’ve been there recently, staring at my screen and scratching my head over the same things you’re all talking about. 🤯

    First off, big thanks to everyone for sharing their experiences. I’ve noticed a few of you mentioning issues with backtesting not working as expected, and yeah, it can be super frustrating, especially when you’re trying to fine-tune your strategies.

    I wanted to share a couple of thoughts on this. For starters, reporting bugs is key. I know it feels like throwing a message in a bottle into the ocean sometimes, but it really does make a difference. If you’re dealing with a broker (such as IG), it might seem like your feedback gets lost in translation…

    About the whole v12 situation with IG and autotrading with IB – it’s pretty much early days. It’s like expecting a newborn to start running marathons. I’ve seen the changelogs too, and it’s clear the folks at ProRealTime are on it, fixing things up and rolling out updates. It just takes a bit of time for everything to click into place, especially when they have to make sure it works across all different brokers.

    And yes, the backtest consistency issue is a real pain. It’s like trying to hit a moving target sometimes. I’m planning to flag this up again on my end. It’s important they know we’re serious about needing reliable tools to test our strategies.

    Just wanted to say, let’s keep the feedback coming and stay patient. These tech challenges are part of the game, but with a community like ours, we can help shape things for the better. Plus, it gives us all good battle stories for the forum, especially on rainy Sundays! right? 😆

    3 users thanked author for this post.
    #228383

    Hello Nicolas (and all)

    This issue with back testing isn’t new, it’s been going on for at least 3 years and V11. Three of us in my team used PRT at V11 for about 6 months and got so mad with it, especially when the three of us would run the same code on 3 PC’s and get three sets of different results, that we went back to manual trading. I contacted IG who said it was PRT to resolve, and I contacted PRT, who said it was an IG issue – so, between them, they couldn’t host a kids tea party. Overall, I have been using IG for at least 12 years now and they improve strength upon strength and answer calls withing the same day and often within minutes; but if I contact PRT, I get silence; the last issue I had they wrote to me about 4 months later with an answer!!! So, 3 Questions

    (1) What is the specific Email to send Tech issues to?

    And, (2) is there anybody at PRT, we can actually talk to; for example, I would like to know why I can not get MTF’s (MultiTimeFrames) to function correctly all of the time, I have found almost minimal help files on it (Unless it is in the 80 Euro advance course, but the syllabus does not mention it).

    (3) Have PRT considered a ‘Live Feed’ question desk (even if I had to pay for it, I wouldn’t mind, I’m wasting so many hours to get ‘basic’ code to run correctly

    Cheers – Mark

    1 user thanked author for this post.
    #228413

    When you get random backtests that provide no results (no gain-loss curve), this is a problem that we have noticed and that we’re working on. The problem is server-side, so it won’t be solved with the build. To get around it, make sure that the preloadbar doesn’t start before the 1st instrument data, for example:

    If the instrument starts to quote on January 1, 2020
    and the chart starts on the same day
    and the preloadbar is not zero

    Then it may happen at random that the backtest doesn’t send any orders. The fix requires a lot of testing, so it takes a bit of time.

    run the same code on 3 PC’s and get three sets of different results

    with exact same fees, spread, start and end date, and units displayed, with tick by tick mode checked, with same custom hours and same timezone?

    I would like to know why I can not get MTF’s (MultiTimeFrames) to function correctly all of the time

    What do you mean, do you have an example to post in a separate topic please? I haven’t see any of your topics before? 🙂

    1 user thanked author for this post.
Viewing 15 posts - 16 through 30 (of 36 total)

Create your free account now and post your request to benefit from the help of the community
Register or Login