Backtesting reliability

Forums › ProRealTime English forum › General trading discussions › Backtesting reliability

Viewing 14 posts - 1 through 14 (of 14 total)
  • 04/08/2018 at 10:32 PM #67567
    bezieh

    Hello guys!

    Basically always when i backtest my strategies, it sometimes ends up making alot of money if i backtest a couple of months or weeks, but if i backtest longer periods, sometimes i see that it completely empties the account, and when i look where those trades that were bad occured, i see that most often the bot didn’t even follow my rules that i set up. (I use the wizard ofcourse, but shouldn’t be any bugs..).

    Now i wonder, what should i do? Should i kind of trust a shorter timeframe where i see that the trades followed my system, and is it also quite reliable because over 200 or 300 trades happened, so i can get an estimated value of what percentage win i have?

    Looking forward to get some nice answers guys!

    Take care and good luck 🙂

    04/09/2018 at 8:17 AM #67578
    GraHal

    Might be easier for u to help if you post your code on here?

    (use the blue button, top row, far right in the tool bar to insert nice looking formatted code please)

    04/09/2018 at 1:47 PM #67746
    Nicolas
    bezieh wrote:

    i see that most often the bot didn’t even follow my rules that i set up.

    Strange behavior that must be analyzed by reading the code. Do you use any indicator? with CALL instruction?

    04/09/2018 at 4:10 PM #67763
    bezieh

    im using a moving average crossover, combined with MACD, basically.

    Im using the wizard to create the code fast and easily basically, but just because its simplified, doesnt mean it should error or do something weird i suppose?

    People told me it’s really hard, or impossible to make money over time with a bot created with the wizard.

    04/09/2018 at 9:33 PM #67774
    bezieh

    Hey, ive got a strategy that would have gotten me 800% within the past 5 years, with 40% accuracy in the trades.

    And when i backtest the past months, same result, 40% accuracy, and making some percent.

    But when i backtest like 1 or 2 years, suddenly it says -100%, and lower accuracy of trades.

    How come? What should i trust, really? Haha, because somehow i see that when it says -100%, and i look at the chart, i can see that the trades at some places doesn’t execute like a coded, but at the other backtests its correct..

    Thanks for respondents!

     

    Moderator’s edit: this post has been moved from being its own topic, to become merged with current topic, as it doesn’t appear it needed to be a separate topic

     

    04/10/2018 at 4:23 PM #67927
    GraHal
    GraHal wrote:

    Might be easier for us to help if you post your code on here?

     

    04/13/2018 at 8:54 AM #68247
    jebus89

    https://www.prorealcode.com/topic/thoughts-on-profitable-trading-systems/

    Its so easy making a system that works for 1-12 months.

    if you backtest for 3 years and the system looks like shit, its because its just that, its just shit.

    If you backtest something for 10 years and it looks profitable and every year is green? Well maybe then you got something that could work..

     

    I would personally NEVER run something thats profitbale for just 1-2 years and shit the rest…. i would not feel confident or comfortable running that system. I want ALL OF MY DATA to be backtested so that i can be confident that if something changes in the market/turns down in the market, i will not only stay alive, but maybe even profit a little bit 😉

     

    Edit: a big experience for me was when i first got my hands on 200K bars backtest (aka 2-3x as much data as a normal PRT user, i opened a premium account via ig.com) and i saw that 9/10 systems that i had made with only 2-3 years backtest data was SHIT when i got my hands on 10 years data…. i only had 1 system that “survived” getting 2-3x amounts of data… i obviously threw away the other 9 systems… they where curve-fitted as f**k 🙂 be careful!

    Everyone can MAKE money in the market. The clue is staying alive 🙂 become a RISK MANAGER more then a trader if you want to stay alive!

    04/13/2018 at 1:01 PM #68266
    bezieh

    Nice man!

    would you say a macd + moving average combination with low values like 15/20, on a 1h timeframe, with backtest of 4 years is reliable?

    It would make 40% winning trades, with 1% stop loss, and makes from 1k euro to like 3 – 4k euros within those 4 years, with a total of about 800 trades or so.

    That in my opinion should be quite reliable, and not curve fitted, since its on a low frame with low MA values.. Do you agree? Couldn’t do a bigger backtest, since the setting was 99000 units :p

    04/13/2018 at 2:39 PM #68268
    jebus89

    without actually being able to test the code myself its hard to say. 40% winrate isnt exactly “having the odds with you” as much as against you..

     

    If the backtest is looking good i would guess the variables really isnt that much to curve-fit. Curve-fitting isnt easier to do, or not to do, based on timeframe (at least from what i know). Curve-fitting is when you take a data sample, be it 1 min or 1 week candles, and then you model something that fits that data and gives a good eq curve and profitable trades.

     

    The problem is then of course: is that model (system/strategy/algo) fitted for *this* spesific data only, or is it robust enough to handle new data?
    One way to do that is to let someone here backtest using 200K bars, i could do it for you if you wanted. Dont worry about sharing the code, for startes youve already said enough for anyone here to re-create it, secondly MA and macd all alone is probably not gonna give you much i would guess. Im gonna take a wild guess and say that these 2 indicators are 2 of the most used indicators for many traders.

    You could also check how the same system works in dax/nikkei/hang seng/italy/spain/us500/wall st. and see if you get similar results. I get real sceptical if my systems only works on 1 market. Not that i start using the systems on all those markets, but i would like to see similar, profitable in the long run kind of results.

    04/13/2018 at 2:48 PM #68270
    bezieh

    Thanks for your reply jebus, is there any other way for us to have contact, some private message thing on this forum that’s available perhaps?

    I would love to communicate with you further and share ideas with you! 🙂

    04/13/2018 at 2:58 PM #68272
    jebus89

    @Nichlas can give you my mail i think.

    04/13/2018 at 3:07 PM #68274
    bezieh

    What’s your mail, ill mail you

    04/13/2018 at 7:09 PM #68306
    jebus89

    @nicolas i ment* 🙂

     

    I dont think its allowed to post emails here, but i give my permission to let nicolas send u a mail with my email. If thats possible?

    04/13/2018 at 11:03 PM #68326
    bezieh

    bro just email me:

    darvin.98@hotmail.se

    🙂

  • Author
    Posts
Viewing 14 posts - 1 through 14 (of 14 total)

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
backtesting and definition of differens emas.
backtesting does not work which a great amount of time units
backtesting kaput ? (ib)
help test new backtesting software
how to change colors of buy/sell icons in backtesting?
issue with inconsistent backtesting results
no backtesting results

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