Understanding the algos behind

Home » Topics » QuantTrader Forum » Understanding the algos behind
Understanding the algos behind 2017-03-03T15:29:04+00:00
  • Author
    Posts
  • Alexander Horn
    Keymaster
    Post count: 366

    Understanding the algos behind

  • reuptake
    Participant
    Post count: 62

    So to start the discussion. Am I correct about how this works? “Static Ranking (SR): All strategy components receive directly in one step the allocation which maximizes the the modified sharpe ratio of the overall portfolio for the rebalancing period.”

    So in case of 2 component strategy (AAA and BBB ETFs):

    1. Create a portfolio of 100% AAA / 0% BBB, compute modified Sharpe for lookback period
    2. Create a portfolio of 90% AAA / 10% BBB, compute modified Sharpe for lookback period
    .
    .
    .
    10. Create a portfolio of 0% AAA / 100% BBB, compute modified Sharpe for lookback period
    11. Select portfolio with the highest Sharpe

    (I skip other parameters like mean reversion period and so on).

  • Alexander Horn
    Keymaster
    Post count: 366

    Yes, for Static Ranking (SR) the algo permutes (runs through) all possible combinations of the constituents to find the best combination in terms of modified Sharpe.

    In difference, for Dynamic Ranking (DR), the algo would first permute all combinations of all non-bond ETF to find the best combination in terms of modified Sharpe, and then in a second step add the bond ETF in its best mix.

  • reuptake
    Participant
    Post count: 62

    So it’s not a very good idea to run this on, eg. 500 stocks? 🙂

    • Alexander Horn
      Keymaster
      Post count: 366

      No, beside the technical hurdles, I doubt you would get any suitable result running on such huge number of stocks. In our experience it´s better to stay at the index level for a market, or to use a limited number of index constituents (Dow, Nasdaq or sub indexes of the S&P500).

    • Tom Gnade
      Participant
      Post count: 25

      I actually managed to get it to run on the entire list of S&P 500 stocks several times before it became unstable. It takes a while. I’m not 100% sure the program is multi-threaded and using all 4 CPU cores, if it isn’t then that would help, along with using a 64-bit architecture, so larger chunks of data could be processed in memory.

    • Alexander Horn
      Keymaster
      Post count: 366

      Tom, as you commented over the phone, in the long run probably the way to go is probably a cloud-based QuantTrader, which would allow more computing power and alliviate some of the data issues. Let us discuss this further, very much appreciate your input.

You must be logged in to reply to this topic.