"sequentially" trading different systems

Discussion in 'Strategy Development' started by NoWorries, May 16, 2007.

  1. Suppose I have 5 uncorrelated systems that I feel confident to trade (entirely mechanical). I know for each system its historical (backtest) performance, which is robust and acceptable for all five, even though differences exist. However, I also know that future performance will likely be different from past performance as a result of randomness.

    Modern Portfolio Theory says I should trade all systems at once, like a portfolio, and allocate my equity over the systems according to some weighting scheme. This can decrease risk substantially.

    Now suppose I can't trade all systems at once (e.g. because of a lack of capital or technical infrastructure).

    It seems the optimal approach is now to "sequentially" trade the five systems. I.e. if I can trade only one system at a time, I should throw a dice each month and randomly pick a system for that month, instead of trading the single system that performed best in the backtest for the entire year.

    Does this reasoning make sense?
  2. gongish


  3. why not?
  4. nove


    Because you should pick the best combination of systems for the resources you have and scale in more systems as your size permits.
  5. TraDaToR


    Every trade is important within every system and the difference between a profitable system and a loser is thin, so I don't think it's a good idea.

    Perhaps if you have really consistent systems that makes a lot of trades...
  6. The reason why not NoWorries is because each system will have periods of success and period of drawdown, if you randomly jump from system to system you will effectively destroy whatever edge a given system possesses, and your overall system success will then become 50/50 at best (and you could hit multiple systems' drawdowns with this cherrypicking approach, in which case your overall success rate could become 40/60 or worse).

    As has been mentioned it is far better to go with the system that is most affordable to you at the moment, and scale-in to more systems as your account equity grows.

    This will also allow you the time you need to acquire greater and greater knowledge of your systems, and how they perform in different markets, so that you may one day be able to increase whatever edge they already posses, thus making your trading more effective overall.

    Good trading all,

  7. I can certainly pick the best system based on past criteria, but I don't know if this will be the best of the five in the coming year as well.

    E.g. my backtest might show a ranking (on whatever criteria, or combination of criteria) like: 5-2-3-1-4 (i.e. system 5 is best, system 4 is worst). But because of randomness, a year from now I might discover that actual performance was 1-3-4-5-2. Thus, I should have traded system 1 instead of 5.

    Because I can't predict the future ranking with 100% accuracy, why shouldn't I prefer to get the average performance across the five systems, rather than facing the possibility of ending up with the worst of the five.

    If I could trade each system for an infinitesimal amount of time, I would get something very close to the true average in theory (as if I would have held all systems in a portfolio at the same time). In practice a weekly or monthly switch would be more realistic but would also decrease the chance of getting the true average across the systems.
  8. nove


    If historical performance has any relevance to you (eg, you believe its true until the system proves it isn't any more) then ranking shouldn't create a problem.

    You should consider the performance of the combination not just the system (eg 1 and 2 may be best individually but 1 & 5 may have lower correlation across a range of market conditions and thus give a better paired result).

    You can't switch on a time basis unless the time is relatively long compared with the system holding time. Complexity = more room for failures = milking mice for little result.
  9. JimmyJam,
    Thanks for your explanation. I agree that focusing on a single system will increase knowledge about the system and its edge, and might help to improve it.

    I can also see your first point (loss of edge, compounding drawdowns) if I would change between systems in a non-random way. E.g. if I would switch out of systems during a drawdown out of fear. But if the change is random and happens often enough, I don't understand why that would affect the edge.
  10. If paired or multiple positions are allowed the solution is straightforward: use an optimizer and find the optimal portfolio weights that maximize some sort of reward/risk criteria.

    I am specifically interested in a situation where you can choose from multiple systems but can trade only one.

    If a backtest shows that one system outperformed all others every year (or month) over and over again, that could be a strong argument to expect it will dominate the others next year as well. But it seems less obvious if historically there is no single system that dominated in every timeperiod.
    #10     May 16, 2007