Loading ...
Sorry, an error occurred while loading the content.

Re: Optimizing parameters for different tickers...

Expand Messages
  • imcharliehm
    I was surprised to hear you found wildly different results with one security. Al, I just found more widely divergent results when testing on a single
    Message 1 of 10 , Jul 2, 2002
    View Source
    • 0 Attachment
      "I was surprised to hear you found wildly different results with one
      security."

      Al,

      I just found more widely divergent results when testing on a
      single security with regards to various moving average types. For
      SPY, using a 12/55 parameter set, no stops, the most recent 500
      trading days, trading both L&S, and five different moving average
      types (MA, EMA, WMA, DEMA TEMA) with the same rules:

      Sigavg1 = Optimize("Sigavg1",12,2,12,2);
      Sigavg2 = Optimize("Sigavg2",55,13,70,6);
      Buy = Cross(MA(Close,sigavg1),MA(Close,sigavg2));
      Sell = Cross(MA(Close,sigavg2),MA(Close,sigavg1));
      Short = Cross(MA(Close,sigavg2),MA(Close,sigavg1));
      Cover = Cross(MA(Close,sigavg1),MA(Close,sigavg2));
      [To create the other four rule sets, I simply substituted the name of
      that average for "MA".]

      I got the following results:

      B&H MA EMA WMA DEMA TEMA
      SPY -25 10.5 13.2 27 35.8 27.2

      However, I think another factor is driving results besides
      the uniqueness of the data series (as was the case with EUK.) Using
      just 4 of Worden's 31 sector groups, I got the following:

      B&H MA EMA WMA DEMA TEMA
      Metals&Mining 51.6 4.2 9.0 4.5 0.1 -13.9
      REIT's 28.2 32. 0.6 2.1 -4.4 -4.5
      Comglomerates -11 5.8 3.0 13.3 19 32.7
      Telecoms -50.2 22.3 5.9 30.6 -2.6 25.5

      Fussing with the slow/fast parameters would changed the results, of
      course, but what I found striking was that, when everything else was
      held constant, how the various MA's preformed against sectors that
      were headed up, down, or sideways. That seems to be the driving
      factor, lending confidence to the results obtained for SPY. So,
      optimizing for a single ticker can be done afterall, it seems.

      Live and learn. Charlie
    • Al Venosa
      Very interesting, Charlie. However, you are reporting only one dependent variable, that is, net % profit. Have you looked at differences among the various
      Message 2 of 10 , Jul 3, 2002
      View Source
      • 0 Attachment
        Very interesting, Charlie. However, you are reporting only one dependent
        variable, that is, net % profit. Have you looked at differences among the
        various moving averages with respect to drawdowns, profit factor, W/L ratio,
        % profitable, etc., etc.? It's possible that the moving averages that give
        you the highest return may also be giving you the highest drawdowns. There
        are other factors besides overall profit upon which to guage a good trading
        system. Whatever you finally decide on, be consistent. Thanks.

        Al V.


        ----- Original Message -----
        From: "imcharliehm" <charliehm@...>
        To: <amibroker-ts@yahoogroups.com>
        Sent: Wednesday, July 03, 2002 12:00 AM
        Subject: [amibroker-ts] Re: Optimizing parameters for different tickers...


        > "I was surprised to hear you found wildly different results with one
        > security."
        >
        > Al,
        >
        > I just found more widely divergent results when testing on a
        > single security with regards to various moving average types. For
        > SPY, using a 12/55 parameter set, no stops, the most recent 500
        > trading days, trading both L&S, and five different moving average
        > types (MA, EMA, WMA, DEMA TEMA) with the same rules:
        >
        > Sigavg1 = Optimize("Sigavg1",12,2,12,2);
        > Sigavg2 = Optimize("Sigavg2",55,13,70,6);
        > Buy = Cross(MA(Close,sigavg1),MA(Close,sigavg2));
        > Sell = Cross(MA(Close,sigavg2),MA(Close,sigavg1));
        > Short = Cross(MA(Close,sigavg2),MA(Close,sigavg1));
        > Cover = Cross(MA(Close,sigavg1),MA(Close,sigavg2));
        > [To create the other four rule sets, I simply substituted the name of
        > that average for "MA".]
        >
        > I got the following results:
        >
        > B&H MA EMA WMA DEMA TEMA
        > SPY -25 10.5 13.2 27 35.8 27.2
        >
        > However, I think another factor is driving results besides
        > the uniqueness of the data series (as was the case with EUK.) Using
        > just 4 of Worden's 31 sector groups, I got the following:
        >
        > B&H MA EMA WMA DEMA TEMA
        > Metals&Mining 51.6 4.2 9.0 4.5 0.1 -13.9
        > REIT's 28.2 32. 0.6 2.1 -4.4 -4.5
        > Comglomerates -11 5.8 3.0 13.3 19 32.7
        > Telecoms -50.2 22.3 5.9 30.6 -2.6 25.5
        >
        > Fussing with the slow/fast parameters would changed the results, of
        > course, but what I found striking was that, when everything else was
        > held constant, how the various MA's preformed against sectors that
        > were headed up, down, or sideways. That seems to be the driving
        > factor, lending confidence to the results obtained for SPY. So,
        > optimizing for a single ticker can be done afterall, it seems.
        >
        > Live and learn. Charlie
        >
        >
        >
        > To unsubscribe from this group, send an email to:
        > amibroker-ts-unsubscribe@yahoogroups.com
        >
        >
        >
        > Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
        >
        >
        >
      • imcharliehm
        Al, You re absolutely right. Largest profit might be less important than number of trades, max drawdowns, etc. --and they are to me in the end-- but those
        Message 3 of 10 , Jul 5, 2002
        View Source
        • 0 Attachment
          Al,

          You're absolutely right. Largest profit might be less important
          than number of trades, max drawdowns, etc. --and they are to me in
          the end-- but those considerations are global to any trading system,
          not the local conditions I was testing for at that point. In other
          words, I was still at the stage of looking at individual trees, --
          i.e., deliberating testing for single variables-- not stepping back
          to see the forest, both because I wanted to discover what the
          individual factors were and because my computer can't handle the
          exponential number of calculations I'd otherwise be creating. My
          results surprised me, so I reported them, because a lot of us are
          probably exploring similar ideas and approaches.

          This stuff ain't rocket science, nor is anything an individual
          discovers likely to remain proprietary for very long. So I write
          about anything I find, here or to a sounding board of friends I use,
          because trying to describe something and explain it helps me to
          understand it, so I can move on to the next step of the research
          process. Doing things piecemeal probably glosses over some of the
          systemic interactions that might occur, but I don't see any easy
          workaround. Besides, my assumptions are two: robust parts are
          interchangeable, and interchangeable parts make for systems that are
          both robust and adaptable. So, I log my experiments and results in
          journals so I can refer back to them later without having to repeat
          the testing, and sometimes the journals take the informal format of a
          message board post.

          Thanks for your comments. Charlie

          PS If you're looking for a piece of published research that is really
          sloppy and begs to be redone, take a look at this month's Futures
          Magazine and the article on "Building a Better MACD Indicator" by
          some academics. Tearing into their assumptions and procedures is the
          sort of thing Amibroker lends itself to. The work I've done with the
          article suggests that Kaufman is right in asserting that stops might
          penalize results more than they help and that trading rules
          themselves are a sufficient (and perhaps better) exit strategy
          (depending on one's intention and how the system is built, of course.)
        Your message has been successfully submitted and would be delivered to recipients shortly.