Running multiple variations of the same strategy

Forums › ProRealTime English forum › General trading discussions › Running multiple variations of the same strategy

  • This topic has 31 replies, 10 voices, and was last updated 6 years ago by avatarDerek.
Viewing 15 posts - 16 through 30 (of 32 total)
1 2 3
  • 12/14/2017 at 10:19 PM #55708
    Derek

    Assume the following thought play:

    The optimal time to go long is 6’o clock AM and you always close out at the next days open. Let’s futher assume that entering at 5 or 7 is also profitable but less so. And 4 and 8 are break even before another choice of entry time leads to negative results. These profits have a bell curved shaped around the optimal value. We enter at 4,5,6,7,8 and reap good profits but the best solution would have been: all in at 6am.

    Now, the optimal value starts to change in one direction. It becomes 7am. We enter at 4,5,6,7,8 and our returns are less than in the first case and going all in at 6 would still be better. 

    Furthermore transaction costs have risen due to more transactions and slippage may as well. I haven’t tested it and many different shapes of “profit curves” may exist but it doesn’t appear promising to me.

    12/14/2017 at 10:38 PM #55709
    Vonasi

    But Derek does the optimal position just slip from 6 to 7? What if it jumps from 6 to 8 and then to 4? Then your perfect bell curve becomes a square topped curve(?!) Markets can be wildly moving beasts and maybe they jump about faster than our lovely little optimization backtest can keep up with? Maybe we need to be trying to hit a wildly moving target and a machine gun firing five bullets is better than a rifle firing one?

    12/15/2017 at 12:05 PM #55731
    GraHal

    I agree Vonasi (I use ‘machine guns’! :)) and wouldn’t it be so useful if the ‘Trading Systems Screen’ and the ‘Auto-Trading Screen’ had sort options etc so that you could see all the ‘Strategy Variations’ listed under one another etc??

    The ‘Trading Systems Screen’ is desperate for a revamp and needs a ‘File / folder  management  function’ and the ‘Auto-Trading Screen’ is useless for any analysis / keeping track of Variations as it only auto-sorts by ‘code version date / time’.

    I’m moaning again … I’d best get back in my burrow! 🙂

    GraHal
    PS Yes … I have suggested above improvements to PRT via their ‘on-site Suggestions Form!

    12/15/2017 at 4:53 PM #55759
    Derek

    @Vonasi & Grahal:

    This is an interesting topic. Maybe you can post some screenshots of your varying results? And how the machine gun’s results are superior to the all-in systems?

    12/15/2017 at 5:19 PM #55765
    GraHal

    Mmm I’m not sure they are superior Derek partly because I find PRT does not make analysis easy and I’m often not in the mood for off-Platform copy &  paste, drag and drop into excel etc.

    Also though I think it’s a ‘trading personality thing’ as in I hate drawdowns (I guess we all do) but I frequently stop Strategies if I judge they have triggered in the ‘wrong place’. For example … if I can see that price is at 61.8% fib retrace from a high and my Strategy has just opened a short then I stop / exit and think … “Oh I’ll have to tweak you so that doesn’t happen again!” 🙂

    Then I do a tweak and maybe run the two variants (or more) together for a while to see if my ‘distrust’ is justified etc.

    GraHal         

    12/15/2017 at 6:08 PM #55771
    Vonasi
    Derek wrote:

    @vonasi & Grahal: This is an interesting topic. Maybe you can post some screenshots of your varying results? And how the machine gun’s results are superior to the all-in systems?

    I don’t have any evidence at the moment as I posted the opening question just to gauge opinion from others on whether this was a method worth following up on. Opinion seems divided at the moment with good arguments for both sides – all in or machine gun. I am more interested in understanding whether running strategies with multiple variables is beneficial at the moment rather than multiple exit levels as the latter is a pretty forgone conclusion. The more profit you hold out for the less likely you are to win – it is a fairly straight line. The benefit of different variables in multiple strategies should be something that is also easy to assess. Five different backtests with stake divided between five strategies with one variable adjusted (within a range either side of the best performing variable) and see if the combined profit/loss is greater than if we went all in on the worst performing strategy.   

    12/15/2017 at 8:43 PM #55776
    GraHal

    Unless I’m missing something (?) you can do above from one set of backtest results?

    I chose 7 results ( best and 3 either side of best) at £1 per point …

    Gain on all 7 = 1384 (each rounded as I’m watching tv also! :))

    7 x Gain on worst performer = 1337

    But  Vonasi what is the logic for using the worst performer out of the 7??   

    GraHal   

     

    12/15/2017 at 9:02 PM #55779
    GraHal

    Haha the film is no good anyway so now I’ve thought about it!

    Results above are bound to be and 7 x best performer is bound to be the most gain!

    If different backtest periods are used then one could not make any firm assumption re which is better … machine gun or all-in??

    GraHal    

    12/15/2017 at 10:18 PM #55781
    Vonasi

    The logic behind using the worst result is that if you had one bullet in your rifle then there is a chance that that one bullet would hit the lowest performer. Yes it might shoot higher but you have to assume the worst possible result is the one you will get when backtesting in a working range of results. If you fire your machine gun you will hit all the targets but not with such big bullets. Yes one bullet might perform better if it hits anything in the top half of the range but it has a pretty good chance of performing worse when the target moves about so much. At least five bullets from a machine gun means that you have a good chance of averaging a good result and is better than hitting the worst target repeatedly with one bullet. That is my logic behind the argument anyway.

    12/15/2017 at 11:36 PM #55786
    GraHal

    Yeah in a way the PRT Optimiser has a ‘lot to answer for’?

    It takes a lot of experience and deep thought (like yours above) to appreciate that using the best optimised result (or even near best etc) is a recipe for failure? The nearer to the best result chosen the more curve fitted could be the Strategy?

    All obvious stuff I know, but I thought it’d be a change from guns! 🙂 🙂 🙂

    GraHal  

    12/17/2017 at 5:49 PM #55893
    Derek

    Alright, before I do some manual testing (I completely stopped using proorder a year ago) here is some input from Perry Kaufmann on the topic. Reference is to his book “Trading Systems and Methods” from 2013, chapter 23, Entering a position.

    My ratonale for comparing our discussion with his is, that using the same entry technique with several indicator values is like averaging into a position and/or waiting for better prices after the first trade is entered.

    • Averaging into a position using a trend following strategy can lead to improved results.
    • Noisy markets allow for better entries when waiting for a better price with the potential of missing a really big move. This should not concern us right now, since we discuss stacked openings and at least one trade will be entered.

    I guess the big similarity is that we receive one initial signal (go long for example) and then increase the position size. We learn that not all trades up to the maximum position size must be taken since some strategy-indicator combinations will just not get to the point of entering the market.

    Of course, this really brief summary does not summarize the points made in the book. I hope they help to advance our discussion.

    12/17/2017 at 6:34 PM #55894
    GraHal

    We’ll be interested on your findings Derek, thank you.

    I used to do  a lot of stacked trades when manual trading. I’d wait until the first trade was in profit by spread x 3 and so as soon as took the trade 2 I was still in profit overall then same again up to 5 trades total.

    Logic was that if I judged Trade 1 entry wrong then my loss was small and I exited.

    Of course it can bee seen that if the Trade 1 entry was good then more profit could have been made by entering total lot size for 5 trades as Trade 1. But it was fun seeing all the trades stacked one on top each other all going green and profit growing! 🙂

    I generally manually trade on a 5 second timeframe and sometimes  I exited individual trades or if I felt there had been a ‘good run up / down’ then exit all trades together often in 1 minute or so … then go and make coffee &toast feeling very pleased with myself! 🙂

    When I have auto-trades running, my screen gets so cluttered it gets on my nerves so I’ve not done stacked trades for a while.

    I did put a suggestion in to PRT (using their on-site Suggestions Form) that a optional tick-box be provided to not show auto-trades … sadly this has not been provided (not yet anyway!). 

    GraHal  

    12/17/2017 at 7:22 PM #55897
    Despair

    Robert Carver advocates in his book “Systematic Trading” running the same strategy with several different sets of parameters to avoid curve fitting. If you are interested in this practice I recommend reading his book.

    12/17/2017 at 8:37 PM #55901
    Henrik

    Grahal, Just have one acount for auto and one for manual trades and problems solved 🙂

    12/17/2017 at 9:50 PM #55906
    GraHal

    Thanks Henrik, I do have a 2nd account but never funded it due to several on here getting emails from IG a few months back saying they were not trading enough to justify having their account ‘free’.

    I thought I’d let the dust settle, but we never got to the bottom of the ‘criteria for free’ and anyway I’ve not seen any further reports of similar ’emails from IG’.

    So yeah maybe I’ll revisit the 2nd account idea again.

    Cheers
    GraHal 

  • Author
    Posts
Viewing 15 posts - 16 through 30 (of 32 total)
1 2 3

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
running two instruments at the same time
adding input from custom indicator into strategy
andrew cardwell strategy on rsi
array/bubble sort in strategy doesn’t work, but works as an indicator
arrays & same candle programming
auto restart automated strategy
bollinger+fractal strategy idea
breakout box strategy (tick by tick not working for entries)
can you add compounding to a strategy?
central bank dealers range – reversal strategy help please!

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