Not logged in
Viewing 15 posts - 1 through 15 (of 19 total)

Forums>StrategyQuant>Pre-Sales Questions>Periodically re-generated EA style

  • #113702 | Reply
    Customer
    61 Posts

    Hi,
    (Posted in regular SQ forum, but am posting this in pre-sales also because the demonstration of process thru the different SQ tools may be of interest to prospective buyers :)
    ———
    Has anyone tried 2yrs worth of 1hr bars from your mt4 broker server for SQ EA live trading success yet?
    I remembered to use “page up” with chart unchecked for scrolling to “pull down” from broker server all the backdata they have for us, consequently at FXCM I have 2yrs worth of 1hr bars for eurusd & audusd, the ones that I tried this on so far.
    Am running thru my usual many days random followed by overnight on genetic followed by improver (maybe this last cycle is more than once) and finally optimize, amazing when you think about it all these tools to carve out the best.
    Got only a single thread on my $30/mo CNS server, but 2yrs 1hr bar-open-only is very fast on it, but I do use memory limitation startup option.

    Guess I’m just chattin’, don’t know if final result will be usable as trying this idea a couple times long ago of just X months of backdata to use for only Y weeks or months of live trading is viable or not, however, this 2yrs is larger than my previous attempts in this mode/hope wherein I think I had only 9mo, soooo, its a big *maybe*.

    Anyone try this style before? Only once on some other strategy-generator-forum did I see someone claim they had success in this mode, too bad I didn’t keep that url around. :(

    My mt4 w/ fxcm is for live real acct, thus so is the data, with volume data in the backdata too. What I do to test out such mode is run the test forward each day while checking to see if I should be running another EA generation.

    If such modus-operandi could be good it sure is behooving not having to buy an i7-4770 eh!? Lol.
    Best,
    Jerry

    #130407 Reply
    Customer
    557 Posts

    What use is 1hr data?  You won’t see realistic strategies on 1hr data, you need 1 minute data at a minimum or tick data ideally unless you are creating strategies that run on a monthly timeframe with trades extending into years.

    #130408 Reply
    Customer
    61 Posts

    Sorry, don’t understand your post. 1hrTF does fine.

    Did arrive at a 1min resolution 1hrTF EA that was good on all 7yrs fhdb when I shared an i7-4770, but on my $30/mo server 1hrTF bar-open-only is fine too.
    (BTW, even on such a fast computer SQ is too sloooooow to develop anything on tick data of even a years worth).

    Back to query of the flock, anyone do auto-trading successfully on 2yr 1hrTF re-generating EAs mode?

    #130412 Reply
    Customer
    557 Posts

    I can’t imagine a 1h TF strategy using bar open would be anything like the simulation in reality.  In other words SQ will find strategies, but they would behave completely differently in real trading, so they would be useless to use.

    #130414 Reply
    Customer
    61 Posts

    Its been my mainstay for me, it always replicates how trading would’ve or could’ve occured in live trading vs tick sensitive EAs wherein anti-testing folks get their gripe of tests being no good (due to tick sensitivity from the EA), so I don’t know why you can’t imagine, its the best in “stability” (non tick sensitivity, and replicatability in tests).

    Anyway, back to my query of the flock.

    #130416 Reply
    Customer
    557 Posts

    I can’t imagine how it could work because the price moves a lot in unknown directions with the 1hr bar, and SQ would not know:

     

    1. Whether a stoploss or the take profit level is hit first if they are within the bar.
    2. Whether price gapped in the 1hr, leading to stop losses below your expected level

    For your successful strategies, how many trades have they made and what has the drawdown been like compared to the backtesting results?

     

    And how many strategies have you tested live in this way? 

    #130418 Reply
    Customer
    61 Posts

    Its worked out in the wash by averages.

    #130419 Reply
    Customer
    557 Posts

    Well if it is true it means many of us are wasting time generating strategies using M1 or tick data, since if H1 bar open is good enough we could generate strategies at least 100x faster.

     

    It also means a lot of time and money saved loading and sourcing/buying tick data.  Also backtesting in MetaTrader would be much simpler and very much faster.

     

    All seems too good to be true?

    #130421 Reply
    Customer
    61 Posts

    Lol, true! Give it a shot!
    That’s half what my subject here is about,i.e., no need deluxe computer (and, so, call that opinion if one n all likes).
    Whether a an sl or tp is tripped by bars-end or by tick …..it all still has to “on average work out in the wash”.
    The first half of purpose of subject was to see if anyone had success doing EAs on 2yr period of 1hr bars periodically regenerated for changing market.
    Thanks for listening.
    Jerry

    #130553 Reply
    Customer
    432 Posts

    Can only confirm @ Batch. I generate all my strategies with “Selected Timeframe Only” mode on M30 and H1. Once I have generated a few thousand that passed my pre-selection criteria via the ranking filters, I retest them in tick simulation mode and kick out the ones that now show a net loss. Usually that´s about 30% of the strategies that fail that first “switch-test” when going into tick simulation mode with the “Selected Timeframe Only” generated strategies in the Retester.

     

    It´s actually a good pre-evaluator to generate them in “Selected Timeframe Only” mode first and then retest in tick simulation mode as this will already kick-out the strategies that relay to much on intrabar movements. And the strategies done this way don´t have trades that last years or anything, they last a few hours usually before trailed out or SL is hit and through the “Selected Time Frame Only” mode for generating strategies SQ will never generate strategies that are dependend on intra-bar movement, which makes them even more stable. On 15 years of data they usually have between 1200 to 5000 trades. Of course, as said, you need to re-evaluate them in Tick Simulation mode later on in the Retester and run all the robustness tests and you´ll lose a good amount of the strategies (I usually manage to cut down 2000 initially generated strategies to 40 in the end), but you´ll find many REALLY stable ones this way.

     

    And I tell you, these strategies (6 of them so far on 4 different pairs) are running on my private account and made me 12% this month already with 1,5% drawdown and a super stable equity curve;)

     

    Now should I have kept that “trick” for us Batch? ;) Hmm…..

     

    Mikeyc: don´t assume to much what should theoretically work and what not, just TEST TEST TEST. I never assume anyhting, I just go ahead and test and test and test again, that´s the difference between success and failure in most things in life. People that only assume don´t get far and stick to a world of theory, telling themselves why this and that can not work or why this and that SHOULD work, while both doesn´t apply if they´d just have tested their ideas in and out:)

    #130556 Reply
    Customer
    61 Posts

    Can only confirm @ Batch. I generate all my strategies with “Selected Timeframe Only” mode on M30 and H1. Once I have generated a few thousand that passed my pre-selection criteria via the ranking filters, I retest them in tick simulation mode and kick out the ones that now show a net loss. Usually that´s about 30% of the strategies that fail that first “switch-test” when going into tick simulation mode with the “Selected Timeframe Only” generated strategies in the Retester.
     
    It´s actually a good pre-evaluator to generate them in “Selected Timeframe Only” mode first and then retest in tick simulation mode as this will already kick-out the strategies that relay to much on intrabar movements. And the strategies done this way don´t have trades that last years or anything, they last a few hours usually before trailed out or SL is hit and through the “Selected Time Frame Only” mode for generating strategies SQ will never generate strategies that are dependend on intra-bar movement, which makes them even more stable. On 15 years of data they usually have between 1200 to 5000 trades. Of course, as said, you need to re-evaluate them in Tick Simulation mode later on in the Retester and run all the robustness tests and you´ll lose a good amount of the strategies (I usually manage to cut down 2000 initially generated strategies to 40 in the end), but you´ll find many REALLY stable ones this way.
     
    And I tell you, these strategies (6 of them so far on 4 different pairs) are running on my private account and made me 12% this month already with 1,5% drawdown and a super stable equity curve;)
     
    Now should I have kept that “trick” for us Batch? ;) Hmm…..
     
    Mikeyc: don´t assume to much what should theoretically work and what not, just TEST TEST TEST. I never assume anyhting, I just go ahead and test and test and test again, that´s the difference between success and failure in most things in life. People that only assume don´t get far and stick to a world of theory, telling themselves why this and that can not work or why this and that SHOULD work, while both doesn´t apply if they´d just have tested their ideas in and out:)

    Way to go geektrader, more intense than my modus-operandi, seems your doing better for it, my route is similar but less, due to limited setup on a 2003 32bit server. But well done!

    #130557 Reply
    Customer
    432 Posts

    Only time will tell if I am “right” Batch, but so far so good. The generating of strategies in “selected time frame only” mode and only do the retesting and further steps in “tick simulation” mode seems to be the way to find stable strategies.

     

    Oh and why not get yourself a newer computer? It will pay itself over time as you will be able to generate even more stable strategies…

    #130627 Reply
    Customer
    97 Posts

    Only time will tell if I am “right” Batch, but so far so good. The generating of strategies in “selected time frame only” mode and only do the retesting and further steps in “tick simulation” mode seems to be the way to find stable strategies.

     

    Oh and why not get yourself a newer computer? It will pay itself over time as you will be able to generate even more stable strategies…

    Can I ask what computer you are working with?

     

    Thanks

    #130629 Reply
    Customer
    432 Posts

    2 of those: https://contabo.com/?show=servers Dedicated Server X

    #130630 Reply
    Customer
    97 Posts

    2 of those: https://contabo.com/?show=servers Dedicated Server X

     

     

    Thanks for quick reply…..

     

    One more question, if you don’t mind………

     

    When you say two of those……. are they linked to process the job together……?

     

    Thanks again.

     

    daveM

Viewing 15 posts - 1 through 15 (of 19 total)
Reply To: Periodically re-generated EA style

You can use BBCodes to format your content.
Your account can't use Advanced BBCodes, they will be stripped before saving.

Your information: