Fully automated futures trading

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

  1. Hi GAT, I just saw the below on BBG. What are your thoughts on machine learning?

    Thanks, B

    Man Group is more interested
    in hiring engineers and data scientists than MBAs. The reason: artificial intelligence is proving a winner. The firm's AHL Dimension fund started using machine learning in 2014 and has returned double the industry average over the past three years. But there's still room for humans. They get to examine unusual AI trades before execution.
    #1001     Sep 27, 2017
  2. truetype


    Seeing some mid-month Sep numbers... Evo up, other AHL programs down.
    #1002     Sep 29, 2017
  3. I also saw the news from Bloomberg on Man seemly success story of applying machine learning to trading and was wondering about it.
    #1003     Sep 29, 2017
  4. AHL has never hired MBAs. When I was working there we only had one, and he wasn't a researcher or PM.

    And I've said before that 'machine learning' is a meaningless term. There's supervised learning, which includes things like doing traditional regression, hardly a new concept. Then there's unsupervised learning, which terrifies the hell out of me.

    And 'Data scientist' is an equally stupid meaningless term for a job that's existed for decades.

    Having said that this weird stuff is a relatively small part of AHL's AUM, and if I'm being cynical, a bit of a marketing tool. Not a pop at AHL specifically, this goes for most funds (ignoring the newer, smaller funds which are using only AI and for which frankly I'm pessimistic of their future).

    #1004     Oct 1, 2017
    Elder and lovethetrade like this.
  5. sle


    Just FYI if you are using VIX or VSTOXX as "implied" you will be getting consistently incorrect results. I will "leave it to the reader" to work out what exactly the issue is.
    #1005     Oct 1, 2017
  6. sle


    Besides the obvious ML stuff (e.g. NLP for news or filings analysis), there are pockets where SML techniques have proven to be very useful, for example where you need to be combining signals from multiple factors. I, however, have never seen it as a source of alpha.
    #1006     Oct 1, 2017
  7. Elder


    Yes I read that report too, and imho it did read as having been written by some wide-eyed reporter stuck right up someone's 'PR machine' to put it politely.

    On a separate topic, I have been looking at how to set up rolling in my system. Following your guidance, I understand you continue to use the signal from the legacy contract until you have fully rolled out to the next one on the delivery calendar. Do you find that this sometimes causes you to adjust the position size immediately upon the roll due to the different underlying volatility and price of the new contract? The problem I see could be acute if say I am only trading a single calendar month (Dec) due to seasonality. Also, the carry on the far Dec might be far lower than the carry on the near Dec also potentially increasing the required adjustment. Any guidance on your experience here would be much appreciated.
    #1007     Oct 2, 2017
  8. The position size won't be adjusted immediately as the vol estimate is done on a rolling basis (ewma). But yes if the vol jumps on the new contract then over the subsequent weeks the position size will adjust.

    I also use a smoothed estimate for carry so again over a few weeks the signal will adjust here.

    These effects are small and don't increase trading unduly as long as you avoid trading eg the very front of Eurodollar where the vol jumps are significant.

    #1008     Oct 2, 2017
    Elder likes this.
  9. Elder


    Yup. Got it, thanks.
    #1009     Oct 2, 2017
  10. Out of curiosity I just eyeballed GE vol and the ratio between the 2nd and 1st contracts is around 1.67:1

    Don't trade this.

    #1010     Oct 2, 2017