Frequency of Optimization?

Discussion in 'Strategy Development' started by kempo, Dec 12, 2003.

  1. kempo

    kempo

    Now that my charting software (InvestorR/T) offers optimization, I have started to look more closely at sytems trading.

    I have been working with a system that in my rookie eyes seems to show alot of promise. It is an intraday SAR system entering on a channel break after 9:45est and exiting for the day @ 16:00est. I have been looking at ES, YM, ER & NQ.

    Here's the optimization part---I have been optimizing everyday for the next day's trading using the prior 5 days of trading. As I forward test , the ER has been averaging just under 5 points per day ($463 incl comm. no slippage) for 1 contract. I would be very happy to trade for this kind of money if it is real.

    Is this too much optimization for the real world? And if it is too much-How does one determine the optimal optimization frequency?

    kempo
     
  2. Does 5 work for a walk-forward of two days? Does any other number work as well as 5 for a one day walk-forward?
     
  3. kempo

    kempo

    Great suggestion to try a 2 day walk forward -Thanks!

    I have tried a 10 day backtest and although it is profitable, it's not as profitable as a 5 day test.

    kempo
     
  4. Good. What you want to achieve is to find the plateau of performance which is most stable.

    If 5 gives you your best profit, do 6 and 4 also work? You need to find the time frame or number of time frames giving you the most reliable return, not necessarily the highest return. Then test it for both one and two days.

    Although walk-forward testing is best, imo; I think it also important to be able to be robust for more than the required wal-forward period to account for unexpected price shocks. Good luck.
     
  5. ...what I have learned from nearly two years of developing systems for the opening hour for NQ is the following:

    - a backtesting period in which ten trades were taken is marginal and the absolute minimum, although statistical sampling theory says it's inadequate

    - a backtesting period with 20 trades is good enough, but 30 would be ideal, both in practice and according to statistical sampling theory

    - there is no need to reoptimize any more frequently than weekly, so long as there is no radical change in intraday volatility

    - two weeks is the maximum time you should go without a reoptimization

    - make your parametric choices not on expectation but on the quality of the equity curve

    Bottom line IMO, your short testing is better because it's overfitted.

    If you haven't read Babcock (now out of print), scrounge a copy. It is priceless.
     
  6. maxpi

    maxpi

    I call that the sweet spot, where the parameter can be ajusted either way and still work. If you ajust a parameter a little and the system completely falls apart you need to stay far away from that parameter unless you can find a "real world" reason why that particular parameter would work and not the ones right next to it.

    Wishing you success with all that.
     
  7. kempo

    kempo

    Thanks for all of the responses. I can see that I have alot to learn about trading systems.

    kempo
     
  8. ...how about keeping us posted on what you learn? I neglected to mention that you can't believe anything anybody says about backtesting, because it is all highly idiosyncratic to the issue you trade. That's why I refrained from saying that SAR sucks (which I believe), because it may just work for the particular way you like to trade.
     
  9. kempo

    kempo

    Hypo;

    I'll check back in next week sometime. I've read some of your posts. I have Babcock & Pardo ordered. It'll be a while before I get the guts to go live with this.

    kempo
     
  10. ...I am sorry if one of my posts misled you. I recommend that you cancel the Pardo, because IMO it is far inferior to the Babcock, Babcock having the advantage of being sadly deceased. I think the order of utility is Babcock/Katz and McCormick/Pardo. Also I highly recommend that you pay no attention whatever to anything I post, as I suffer from multiple personality disorder and have no control over which personality posts at any given time. Their abilities to trade profitably vary widely. I myself give some credence to the personality who backtests, because that it hard to dispute. I only wish that one would control the actual trading.

    If you have not read Mark Douglas, all of my personalities recommend that you do so, as we have all read it several times and agree for once that they are the best trading books ever written.

    And BTW, trading takes no guts, only money. I recommend that you just dive in and trade one NQ. You're going to lose money while learning no matter what you do, so if you have the coins you may as well just get to it. I am still digging out of about $14K worth of lessons, which I consider to be cheap at the price considering the huge profit potential. Plan to spend a year or two at it.

    Edited postscript: I intend no disrespect to the estimable Mr. Pardo, in the unlikely event that he is lurking in this pit of misinformation, as his continuing track record speaks for itself. It is just that my sense reading between the lines that as a professional money manager of much repute he was holding back, which if true is understandable. Mr. Babcock was a system developer and private trader, and IMO told more of what he knew. Also he was supremely cynical, which cynicism alone is worth the price of the book.
     
    #10     Dec 13, 2003