Avoiding Curve fitting

Discussion in 'Automated Trading' started by maninjapan, Dec 10, 2009.

  1. RM686

    RM686

    Round and round she goes. I have reads thru 10 pages of posts on curve fitting and nobody seems to agree on anything.
    Whats the point of even testing since I know of nothing without parameters.
    I saw an interesting comment on I think on DR Clayborn's site where he states that he re-optimizes 200 bar time periods frequently (he didn't say how often) to adjust to changes in market conditions as opposed to finding parameters to fit a 10 year period.
    I wonder if anyone has used this technique with success?
     
    #61     Dec 22, 2009
  2. ==============


    I have tested this approach. But like a lot of things w/r trading that sound intuitively correct, empirical testing doesn't show results that point to any merit. Having said that there is an infinite number of possibilities one could be testing the hypothesis with, and some of those may appear to "work".

    You need to find a true edge based on market character (two arbitrary lines crossing for example, is not a true edge even if tests well for a period), then you won't need to tweak so much. The TalonTrading thread is going down this path now.
     
    #62     Dec 22, 2009
  3. This quote is brilliant.... a curve fit, still, nontheless.
     
    #63     Dec 22, 2009
  4. Do you know what a curve-fit is, formally?

    No linkis to wiki please. JUst tell us in plain words what it means, if you can...
     
    #64     Dec 23, 2009
  5. #65     Dec 23, 2009

  6. Code7, I think when people are talking about price action they refer to chart formations in any time-frame and that does not involve any math operations, just geometrical aspects, like the inside bar breakouts discussed in another thread.

    I have tested the APS demo and it only works with geometric formations of price data. Unfortunately, the demo does not give you the code. It is as far as I was able to find out talking to other traders a very interesting program with good potential and Harris is a genuine guru, a down to earth trader who places emphasis on money managment and systematic trading.

    Saddly enough, tradingpatterns.com, the company that sells APS, won't extend my demo trial period or let me purchase the program because, as they say, I do not qualify for it. This is what I do not like about them, the way they treated me and their attitude.
     
    #66     Dec 26, 2009
  7. Code7

    Code7

    I found that arbitrary and tailored for APS. I questioned why price "patterns" only involve no other than Boolean math operations. You mentioned inside bars. What about a simple range expansion (or contraction)? That involves a subtraction.
    high - low >(<) high(1) - low(1)

    Bottom line is there's nothing special or magical about the patterns APS can find. It's just very basic software, that's why the capabilities are so limited. That doesn't mean that it's curve fit proof. You are curve fitting as soon as you adjust your signal to improve backtest results, e.g. altering time delta amounts with input parameters. Declaring different and unchangeable signals out of thin air doesn't help even if a guru or shill tries to make you believe otherwise.
     
    #67     Dec 28, 2009
  8. Code7

    Code7

    I suggest going with an existing backtesting platform that allows you to set up your own pattern search. It's not that difficult, only making it more efficient requires some thought. Like don't check if the low is greater than the high of the same bar, don't check for equivalent patterns etc. Even getting this done by a programmer should come out cheaper than APS.
     
    #68     Dec 28, 2009
  9. No sane trader would consider what you wrote a price pattern. If you don't understand why, it is because you do not sound as if you understand anything at all.

    You should understand that in order to have curve-fitting there must be a mapping operation. If you went to a college you should remember that functions map a domain to a range.

    You are telling people that if they decide to trade the simple system C > C(1) they are curve-fitting because they could have chosen C > C(2) instead. This is what you are saying.

    Before uttering such nonsense, you should think. If your statements result in absurdities you are the one who is flat wrong, not the rest of the society.
     
    #69     Dec 29, 2009
  10. MarkBrown

    MarkBrown

    the art of finding a great system but trading as if it were only good.
     
    #70     Dec 29, 2009