Choosing from optimized results

Discussion in 'Strategy Development' started by BrooksRimes, Jun 29, 2005.

  1. Attached is an Excel spreadsheets from an optimization run in Amibroker. The two optimized variables are the points of fixed profit and the points of fixed loss. 110 combinations altogether.

    There are many positive results here. One could simply pick the parameters with the highest Net Profit but that would ignore risk and the bumpiness of the equity curve.

    Amibroker provides a number of other statistics that combine reward and risk such as: Recovery Factor, CAR/MDD, RAR/MDD, RRR, Ulcer Perf Index, Sharpe Ratio and K-Ratio. (I will provide definitions in a Part II). These all result in some "give back" in total net profit.

    Amibroker lets you choose ONE of these statistics, sort on it, then display really nice 2D and 3D charts. Then, you can finalize selection by finding parameters which work well in a large area.

    Some of you seasoned system developers: How do you choose from optimized results? What is the MOST important statistics when all statistics are good?

  2. Part II

    Attached is a definition of the Amibroker statistics.
  3. I wouldn't call myself seasoned with 2 1/2 years of realtime results, but I am a profitable systems trader.

    I optimize using sharpe ratio and then check the profit factor for acceptability.
  4. Thanks for the post. Ok, this is a good example.

    The max profit from any test is $48681 with a Sharpe ratio of 45.

    The best Sharpe ratio is 277 with a corresponding profit of $18827, which is 61% lower.

    The 6th best Sharpe ratio is 215 with a profit of $28676 or 41% below the max profit.

    All of the above have profit factors of 19:1 or better.

    I believe any Sharpe above 2.0 is consider good.

    My thought is that is too much profit to give up for a better Sharpe ratio. Is it wrong to think that way?

  5. Your numbers sound incredible! I thought I had a good system.
    19:1 R/R is fantastic!

    For what it's worth (probably nothing) I separated my backtest data into 2 pieces. One I optimized on and the other I tested blind. I found the sharpe ratio and profit factor to be the best predictors of future profitability.

    I didn't worry about dollar profits until money management was applied. The absolute dollars aren't important because when you use mmgt you switch the trades to a percent of available equity. The dollars that come out of that are more important.

    Here's a shot of my first system. I went live near the end of 2002. Notice how the results before and since are about the same. I made a change to adjust trade size based on volatility but no changes to the trade selection or mmgt pieces.
  6. Yes, too incredible. I just discovered a flaw where I was looking at the end of a bar before it was finished. The system is still profitable but not as good. Profit factor is now under 2.0.

    Thanks for the thoughts on MM.

    I divided my data into thirds, optimized on the middle third and tested out of sample on the other 2/3s.

    Your results are great - congratulations!

    What are you trading and is it intraday or longer term?