my trading system backtest

Discussion in 'Strategy Development' started by hanhao, Aug 13, 2007.

  1. hanhao


    here is the result of my trading system i am backtesting
    what i do is run the system onto every security and get the % gain after my system went through them.
    it's done in a bear to bull market from 1 Jan 2000 to 27 Apr 2006
    comments please.

    Average percent gain 36.34%
    Standard Dev 41.38%

    Total count 185

    Average percent gain -11.17%
    Standard Dev 7.19%

    Total count 157

    Aveage percent gain 12.59%
    Average Standard Dev 24.28248053
    Total Count 342

    Win/Lost count ratio 1.178343949
    Percent gain/lost ratio 325.31%

    i think the Standard Dev of the winnings is insanely high! but the good thing is the losings are relatively low.
  2. R0bert


    How many stocks did you take into account? Did you apply filters on price/volume?
    How long does your system hold positions (in average, min, max)?
    How are the entry signals of your system distributed in time? May be they all happen in one day.
    What is the result of your system for 2006-2007?
  3. maxpi


    Not bad, reward to risk is about 3:1 and winners 55% or so. That is pretty much into the comfort zone.
  4. hanhao


    took account into all stocks. those with little liquility lesser than $100k are filtered. Where liquidity is the past 2 month's average of volume x past 2 month's average of price

    my system holds position as long as they are profitable. losing positions of more than 8% are cut. A minor low trailing stop is used to lock profits

    i dont really understand what you mean by my entry signal distribution

    2006-2007 is not tested because its a bull market. i wanted to ensure this system is tested in a bear and bull run equally. From the plunge of 2000, it took till 2006 to recover to the same level as it was in 2000.
  5. hanhao


    would anyone with statical knowledge like to comment?
  6. Basically you should test your system on out-of-sample data
    (any time period other than that used to develop the system, to fine tune parameters etc). For example, test on 2006-2007, or on years before 2000.

    If your in-sample results show 90% winners, and out-of-sample results show 50% winners, the system is probably overfitted and will not work in the future.
  7. The back testing I use through TI is event based back testing. That is I have an event like a V bottom and when I test it, I test it against the whole market at once. Of course I have filters for volume and price etc but I am testing against everything that would have come through. I really think that is the only way to do it.
  8. hanhao


    excellent, perhaps it is curve fitted
  9. hanhao


    the Standard Dev of my winnings ,41.38% is really bothering me

    it's like 36.34% +- 41.38%
  10. I forgot to add the picture
    #10     Aug 14, 2007