Frosty's Strategy Analysis

Discussion in 'Strategy Building' started by frostengine, Dec 5, 2007.

  1. Well found something slightly interesting for strategy #2. The long trades seem to be the only ones doing very well. The short trades seem to be where most of the losses are comming from, however it does make about 4k over time from the short trades, but the curve is horrible.

    Here are the charts for strategy 2, long and then short only.
     
    #31     Dec 8, 2007
  2. After seeing these results the next steps to improve the strategy i will do will be to focus on strategy #2. Right now its a stop and reverse strategy. I will lave the current long entry signal alone, but convert the reverse to short signal into just a go flat signal if currently long. Then I will work on finding a good short only strategy that complements this long only version.

    Here is the chart showing the short only trades from strategy #2.


    Anyone have any comments or suggestions about the way i'm handling analyzing and improving the strategy?
     
    #32     Dec 8, 2007
  3. RedRat

    RedRat


    Hi Frosty,

    During 2003 year S&P market was at about 800 level. At 2007 it was trading at ~1400 level. Strategy performance depends upon the level of the market. It is possible that during 2003 all your profits were taken as commissions. Later it started to gain extra money, exceeding commissions.

    Then your strategy is optimized for bullish market. Your longs brings more then short deals. What about current performance?

    If you have 3 different strategies the best choise is not to select one of them, but instead to trade all of them together with proper % of capital.

    How many strategy parameters do you have? What a pity you started discretionary trading instead of improving your old strategies.

    RR
     
    #33     Jan 20, 2008
  4. RedRat,

    Hey thanks for reviving this thread. Although I have a discretionary thread running, that is not where all my focus is. I am still actively developing automated strategies as well. The same basic strategy I started this thread with, I have still been tweaking. I have made a number of improvements to the original strategy. All strategy development are made still using the SAME testing period as I started this thread with. Attached is a PL screen shot using the same testing period as the first screen shots of this thread. This is currently my best strategy based on the same principles from the start of this thread.

    In my next message I will post a screen shot of PL from the time that test period ended through this past Fridays price action showing how the strategies are performing going forward.
     
    #34     Jan 20, 2008
  5. Here is the PL chart for how it performed AFTER that last one ended. As you can see its not doing very well in this STRONGLY bearish market... however, for a strategy that performs so well in an upward market, it is holding up fairly well in such negative conditions.... mostly just treading water.

    At least that is how I view it.. thoughts?
     
    #35     Jan 20, 2008
  6. RedRat

    RedRat


    Frosty,

    You did not answer, how many parameters do you have in your strategy? It can be overfitted.

    Is this chart "in sample"? Did you test your strategy using "out of sample data"? If you adjust parameters for the 4 years period, you should leave at least 1 year of data for out-of-sample testing.

    RR
     
    #36     Jan 21, 2008
  7. All conditions, long entry, long exit, short entry, and short exit, each have 4 unique parameters involved in making the decision.

    Two of the parameters for each decision are for length. How many bar periods to use in calculating my indicator. And 2 parameters are the values used in comparison to make the decision.
     
    #37     Jan 21, 2008
  8. RedRat

    RedRat


    I suppose that there are too many parameters. In fact it depends upon the number of deals. If you have many many historical deals, then it does not matter.

    I suggest the following:
    You take data from 2003 till the April of 2007 as in-sample data. You optimize your strategy parameters using that data.

    Then you test performance (without changing parameter values) using data from April 2007 till 2008.

    It will be "proof of concept". If your strategy is profitable OOS, then you can continue your development. Otherwise adding new parameters/filters will just lead to overfitting.

    PS. Most of TA indicators/oscillators does not work OOS.
     
    #38     Jan 21, 2008
  9. slowBear

    slowBear

    You might want to look at Stridsman:"Trading Systems That Work". He describes several techniques to analyze the trades in a system and improve system performance. It all sounded very logical to me when I read it. Part of the analysis is used to improve stop placement. Also he doesn't do any optimization of parameters, so curve fitting isn't of any concern.
     
    #39     Jan 21, 2008