Fully automated futures trading

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

  1. Kernfusion

    Kernfusion

    The last couple of months were good for me too, not at HWM yet, but if things go the same way for a couple more months I might be..
     
    #4021     Mar 23, 2024
  2. Rob, I'm curious about your beta portfolio: why the large risk overweight to stocks vs bonds? Is it tactical or strategic?
     
    #4022     Mar 24, 2024
  3. could you share which instruments you are trading??
     
    #4023     Mar 25, 2024
  4. See here tab asset classes, you can see my strategic risk weight to bonds is around 7%, and tactically is currently lower.

    Rob
     
    #4024     Mar 25, 2024
  5. Kernfusion

    Kernfusion

    Here it is, some weights are 0, or set to 'don't trade' for various reasons, e.g. repeating instruments with different multipliers..
    There's probably more liquid instruments now, I need to re-check and add them, e.g. the "new" lumber, but the process is laborious and I just can't bring myself to do it :)

    3KTB, AUD, BAX, BRE, BTP, BTS, BZ, CAC40, CAD, CC, CGB, CGF, CHF, CL, CT, DA, DAX, DJSD, DJUSRE, ECO, EMD, EOE, ESTX50, EUR, F1U, FBON, FLKTB, GBL, GBM, GBP, GBX, GC, GE, GF, HE, HG, HH, HO, IR, IXB, IXE, IXI, IXM, IXU, IXV, IXY, JGB, JPY, K200, K200M, KC, KE, LE, M1MS, M2K, MBT, MES, MGC, MHI, MNQ, MXP, N1U, N225M, NG, NIFTY, NQ, NZD, OAT, OJ, PA, PL, QG, QM, R, RB, RP, RS, RUR, RY, SB, SCI, SEK, SI, SMI, SOFR3, SPI, SSG, SX3P, SX7E, SX8P, SXAP, SXDP, SXEP, SXPP, SXRP, SXTP, TSE60, TSEMOTHR, TSEREIT, UB, V2TX, VIX, WIIDN, XIN0I, XT, YT, ZB, ZC, ZF, ZL, ZM, ZN, ZO, ZR, ZS, ZT, ZW
     
    Last edited: Mar 25, 2024
    #4025     Mar 25, 2024
    No.Emotions.Trader likes this.
  6. wopr

    wopr

    I was checking out your strategy report Rob (https://github.com/robcarver17/reports/blob/master/Strategy_report) and I noticed that you don't trade any trend rules for Bitcoin, any reason for that? Micro contract is quite cheap too.
    I noticed as I saw you have a short on, and Bitcoin has been on a good run for the past 6-8 months.

    Thanks!
     
    #4026     Mar 26, 2024
  7. No idea. One of the upsides of a fully automated fitting process is you don't need to know or care where the weights for each one of over 200 instruments is coming from.

    Code:
      BITCOIN:
        assettrend16: 0.015
        assettrend32: 0.02
        assettrend64: 0.02
        carry10: 0.088
        carry125: 0.034
        carry30: 0.089
        carry60: 0.043
        mrinasset1000: 0.02
        relcarry: 0.1
        relmomentum80: 0.041
        skewabs180: 0.144
        skewabs365: 0.144
        skewrv180: 0.129
        skewrv365: 0.114
    
    Rob
     
    #4027     Mar 26, 2024
    newbunch and wopr like this.
  8. wopr

    wopr

    I'm not there yet with confidence level.
    Since I'm typing, small update from me. I've just completed a big effort to add more markets, sort of like what Rob did a year or so ago. I'm at 104 markets right now.
    Since that will cause some reshuffling in the portfolio, I also used this to add more rules from Rob's latest book. I'm just about to deploy that any day now.

    In terms of performance, good year so far, about 3.5% from the HWM. Unfortunately, my vol targeting kicked me out of Cocoa a few weeks ago :)
     
    #4028     Mar 26, 2024
  9. That's very fair; in fact it's only this year that I've trusted the fully automated fitting enough to use it, before I was manually handcrafting. And I wrote the code myself!

    Rob
     
    #4029     Mar 26, 2024
  10. Kernfusion

    Kernfusion

    So what logic is currently included into the automatic rule-weight calculation, apart from dropping expensive rules?

    In the Strategy report "Forecast weights" I can see that e.g. SP500_micro (a cheap to trade instrument) has breakout20=1.2 but breakout160=NaN, so a more expensive variation is included but a cheaper one is dropped. Are you also incorporating the historical SR as described here (or a simpler version as in the spreadsheet here) or even doing a full mean-variance optimization with some robust bootstrapping to fit the rules daily?
    And also, as I understood from the blog post, these forecast weight estimations are done by pooling all instruments, so the same forecast should mostly have the same weight for all instruments (unless some expensive rules were excluded for a particular instrument, which causes re-normalization of the remaining rule-weights for this instrument), but in this report the weights are all over the place from instrument to instrument..
     
    Last edited: Mar 26, 2024
    #4030     Mar 26, 2024
    wopr likes this.