Fully automated futures trading

Discussion in 'Journals' started by globalarbtrader, Feb 11, 2015.

  1. In 2022, there still exists someone trading discretionarily? :)

    You should not spare constructive criticism, just because he is a gentleman :)
     
    #3351     Jun 9, 2022
  2. Gambit

    Gambit

    I have a question Rob. Would adding a mean reversion system on a short term time frame (ie daily) be complementary to a trend following system that trades on the monthly and longer time horizon?

    And is there any benefit to exploring trend trading in spreads such as WTI/Brent? Thank you.
     
    #3352     Jun 9, 2022
    newbunch likes this.
  3. I have a question for Rob too:
    what's your take on the trend (pun intended) in the CTA industry towards alternative assets (including less liquid futures and OTC contracts)? It looks like every major player has launched a programme which, at least in some part, has exposure to such assets. In theory, the more assets the better, but I wonder how can multi-billion funds trade something like cheese futures... My suspicion is that such assets actually get a very small risk allocation because of liquidity issues so that mostly becomes a marketing trick to justify higher fees.

    I was also curious what "residual" stand for in your daily automatic pnl report. It is often a larger factor than the actual performance of your dynamic_TF_carry strategy.
    Thanks.
     
    #3353     Jun 9, 2022
    KevinBB, Gambit and newbunch like this.
  4. Now that was a pleasant day. My futures trading system achieved its best daily profit ever. And I had to transfer out excess cash twice within the last 24 hours, to avoid that my portfolio is getting too large to my liking.
     
    #3354     Jun 14, 2022
    Kernfusion, newbunch, Gambit and 2 others like this.
  5. Well done. My MTM was down -0.5% yesterday but up 1% today (because I close before the US does, I sometimes get p&l running over to the next day). So small up, rubbing the HWM a little higher. And excess cash is only about 2.5% right now, not worth the transfer :)

    Rob
     
    #3355     Jun 14, 2022
    Kernfusion and newbunch like this.
  6. Kernfusion

    Kernfusion

    Yeah, the futures system is killing it, my prod system is almost at HWM, paper system (more capital) is making new HWMs for several days.
    Stocks and bonds are the opposite of course, so net net I'm somewhat down..
     
    #3356     Jun 14, 2022
  7. There is a bug in the psystemtrade cost calculation for dynamic optimisation which only came up when I was writing up the methodology for my book.

    See discussion here.

    It's trivial to fix, but I'm currently checking the backtest to see what effect it has.

    Rob
     
    #3357     Jun 15, 2022
    wopr and Kernfusion like this.
  8. Problem is fixed, though I need to tidy up the production code, but I had to recalibrate the shadow_cost parameter to 50, otherwise I was trading far too slowly. Latest commit should work, but be careful if you have shadow_cost in your private_config.yaml as that will need overwriting.

    If you're using your own implementation of this method then hopefully it's clear what the difference is but ping me if you want more clarification.

    Rob
     
    #3358     Jun 15, 2022
    wopr likes this.
  9. Kernfusion

    Kernfusion

    I think I'm calculating costs slightly differently, because I express both costs and GapSigma in USD.

    I.e. when I first calculate risk of the gap portfolio I use daily percentage returns: (TodayPrice-YesterdayPrice)/YesterdayPrice and standard deviations of these returns, (the weights are calculated as FullNotionalUSDPrice*Units/BaseSystemCapital) so my portfolio sigma is initially in daily percentage units, I then multiply it by BaseSystemCapitalUsd which gives daily risk of the gap portfolio in USD and then I add trading costs to it which is also in USD.

    GapSigmaUSD = GapSigmaPercentDaily*BaseSystemCapitalUsd+CostsUSD

    Trading Costs are calculated for each product in the gap portfolio as
    CostsUSD=(BidAskSpreadUSD+CommPerCtrUSD)*NumUnins * shadowCostsDaily
    and then summed up for all instruments in the gap portfolio.

    My ShadowCosts parameter equals to 30, but I treat it as annual value, therefore I divide it by 16 to get daily (shadowCostsDaily=1.875) because everything else is also "in daily units".

    My buffer is currently 0.017, and I still think the system is trading a bit too fast (up to 2x from before, didn't fully analyzed it yet). Not sure if my ShadowCostAnnual is really in the same units as in pysystemtrade or not, 30 and 50 look pretty close..
     
    Last edited: Jun 15, 2022
    #3359     Jun 15, 2022
  10. Yes, this problem only occurs because I translated everything into portfolio weight terms.

    I'm currently playing around with the shadow cost parameter to see if 50 is optimal or not.

    Rob
     
    #3360     Jun 16, 2022
    Kernfusion likes this.