Optimization of the systems in today’s time?

Forums ProRealTime English forum ProOrder support Optimization of the systems in today’s time?

Viewing 15 posts - 1 through 15 (of 23 total)
  • #178850

    I have one question about optimizations … if I optimize a system that runs well to very well over 200,000 bars, over 200,000 bars, I get a passable performance. I always optimize with WF70 / 30.
    But how would I adapt / optimize it more to the present day? At the moment it is often the case that my insample curve is steeper than my outsample curve. How do I get it the other way around? The insample runs passably, but outsample better? Simply optimize the WF70 / 30 over 50000bars, i.e. shorten the optimization time, works quite well. But is the result still robust?
    A good example is SP500 or Nasdaq, which 8 years ago were significantly lower than today’s.

    How do you optimize systems with WF that outofsample delivers better results? So more into the present day? Can someone answer that?

    #178851

    Results for Out of Sample (OOS) is usually not as good as In Sample (IS).  You can liken it to results in Live not being as good as during a backtest optimisation.

     

    #178853

    Using the example of Nasdaq … trend conditions, entry conditions, but I have SL and TP as exit conditions. Today completely different values are better than 5 years ago … if I leave all conditions as above the 200000bars, except TP / SL.
    So if I ONLY optimize TP / SL these days, my backtest OOS looks much better. And optimize these values from time to time. Can one make it like this? Would that still be robust? From experience?

    #178854

    So if I ONLY optimize TP / SL these days, my backtest OOS looks much better.

    How are you defining / getting backtest OOS.

    By definition OOS is not optimised and yet you say above you optimised TP /SL.

    #178858

    No, sorry the translator.
    My trend conditions and my entry conditions are fixed. I only optimize SL / TP over 200000bars WF70 / 30 and get a good result.
    My question: Does it make sense to optimize the SL / TP over, for example, 50000bars WF70 / 30? Since then you are optimizing a little more into the present day. For me it makes sense to optimize SL / TP from time to time because the market is changing. All other criteria remain fixed.
    Does that make sense or do I have a mistake?

    #178861

    I consider yes, that does make sense.

    Depending on the timeframe, you could even optmise over less e.g. 10k bars?

    Explain further … on DJI we have just gone through 2 weeks of downturn and likely now to go through an upturn.  You would not want to optimise for (just) the 2 weeks of downturn if you think it is going to change to an upturn?

    So you need to optimise over at least a few weeks upturn AND a few weeks downturn.

    You coud try and be clever and optimise over just a few weeks upturn and then you be set for the few weeks upturn that is to come (in my opinion).

    As you can see there is no black and white / definitive answer … else we would all be billionaires! 🙂

    You could try my idea above re optimise over just an upturn period and then set this version going in Demo on virtual money.

    Also set the version optimised over both downturn and upturn periods in Demo (or Live?) so you can compare results.

    Let us know how you get on please.

     

    #178862
    JS

    In my opinion, you never know when markets are changing, so optimization over the longest period is my preference.
    When you optimize over shorter periods you maybe “running after the markets”.

    #178863

    I personally noticed it especially with mean reversion systems in the EurUsd and trend following systems in the Nasdaq. I leave all trend conditions and entry conditions the same … but if I optimize the exit conditions separately over 50000bars in the M15, I get a much better result than over 200000bars. Still WF, of course. You adapt the exit conditions to today’s times, so to speak. But is that already over-optimized?

    #178864

    is that already over-optimized?

    Best way to find out is to run 2 / several different versions … this is what our Demo Accounts are for.

    Let us know how you get on.

    #178865

    I was hoping someone has done that before and could give us an answer. 😉
    Nicolas maybe?

    #178867
    JS

    When your system and your money management are two separate things than it is not a bad idea to optimize your money management on a regular bases.
    In my systems the money management is not separate from the system (rules) so it is already dynamic.
    Maybe you can try to make your money management also dynamic…

    #178869

    What suggestions do you have for making the management dynamic?
    I tried ATR and “% profit”. That didn’t work for the Nasdaq in particular. Are there other options?

    #178871
    JS

    In my systems I use the standard deviation as a dynamic input and that works well.

    #178872

    Mmh, I also use that in the Dax. But why doesn’t it work on the Nasdaq?

    #178873
    JS

    It works for me on the DAX and the Nasdaq.
    You can sort of standardize the standard deviation for multiple markets:

    (Standard Deviation / Close) * 100

Viewing 15 posts - 1 through 15 (of 23 total)

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