Ideal Optimization

Discussion in 'Automated Trading' started by mvalley, Sep 18, 2006.

  1. mvalley


    I've recently gotten into automated trading first by using stockfetcher and then translating my stockfetcher queries into AMI Broker.

    Since AMI Broker has an optimization feature, I've been using that to optimize some of my queries.

    Im running into some questions about what is the "ideal" optimization to trade.

    I've been trading the ones that have the highest Sharpe/Profit Factor ratio, but I've notice that the higher the sharpe, many times the lower the ROI.

    The highest ROI results tend to be the parameters that yield lower % profit per trade and lower sharpe ratios.

    There's a bunch of other measures that I haven't ever touched like RRR, CAR, etc

    What measure does everyone here use to determine which parameters are best to trade?
  2. man


    before the desaster happens to this thread i quickly throw in MY personal opinion: sharpe ratio is what you actually "feel". ROI is a function of leverage. without an additional risk figure it does not mean too much.
    any figure that does not relate return to risk can only serve as additional information (IMHHHHO).


    i give this thread three and a half pages of content, and additional four for bashing.
  3. Sharpe Ratio has nothing to do with "feel"...
    And ROI is not a "function of leverage"...
    And your use of the English language is sloppy and undisciplined in the extreme... bordering on reckless...
    And very likely mirrors your money management style.

    Arbitrage funds that grind out a steady return...
    Say 15-20% per year when riskless return is about 4-5%...
    But do it year in and year out with minimal month to month fluctuation...
    Perhaps with one losing month/year...
    Will often have a Sharpe Ratio in the 3.0 to 4.0 range.

    Doing this requires very sophisticated strategies and technology...
    And you will get rich slow... but you WILL get rich.
  4. man



    .sharpe measures best what traders "feel", at least better than profit factor, hit ratio, payoff or things alike. use a little fantasy to find out what i might mean with "feel" ...
    .well, if i want to change the ROI for any of my systems, I can lever it up or down. but that is just ME and MY systems.
    .this is an internet board, not grammar school.

    rest of your post is suprisingly inspired and professional compared to the first paragraph. nevertheless the whole post reduces my odds by 50%. bashing in post three. almost record. given the fact that we never had contact before ... it IS record.

    good trading to you nonetheless.
  5. So...

    if High Sharpe == Good feel...

    I'll make millions trading when I'm high on crack
  6. man


    i am not completely sure i get you ... but i'd say you hit it right on ... :)
  7. taowave


    Sharpe is OK,but i dont like any measure that penalizes you for large gains to the upside...Remember with Sharpe,the denominator is the standard deviation of returns...

    As I trade my own capital,I am far more concerned with my max drawdown as oppose to the volatility of my earnings.As you have experienced,maximising sharpe usually leds to a reduction in profitability.

    To make a long story short,if ypu are using Ami,check out the Recovery ratio and to a lesser degree the Ulcer index.
  8. tireg


    The Sortino Ratio is often used in lieu of Sharpe ratio, where S = E[R-Rf]/stDev(negative excess returns). It does not penalize for upside volatility.

    I like to use MAR ratio in conjunction with Profit Factor in developing trading systems individually. Combining them, I take into account Sharpe as my goal is consistency and smoothness of equity curve.
  9. Sortino is great for measuring past results, but I would never use it for optimization (others may disagree). All optimization includes some curve fitting. As such, large optimized gains could easily be large losses going forward. The fact that the system was correct on all big moves in the past doesn't mean it will be so in the future.

    I'll give an example. Let's say your system is correct 60% of the time and wrong 40% of the time. And let's say it is October 1987. Your system is correct and you make a killing. But there was a 40% chance you would have lost a lot of money in the crash.

    I think you should include all volatility when optimizing systems and not just downside volatility.

    To reply to the original post, never use ROI alone. I can easily set my system to make 1000% a year, but of course I'd lose 100% eventually. By trying to make a moderate amount of money, risk is kept within tolerance.

    If you were to chart leverage against return, you would obviously see them rising together at first. Eventually, returns would start slowing and at the end return actually declines because leverage will cause massive losses which overwhelm the large gains. Finally, return will be -100% from a single 100% loss.

    But sharpe is not the end all be all of return/risk ratios. Somebody else commented about looking at return vs max drawdown. There are many ratios you can use and I suggest using multiples ratios (you can just create a geometric average of the ratios and optimize that value).
  10. man


    my problem with sortino is, that the number of observations that really make sortino different from sharpe is often very low. so it is just a dozen outliers. and the fewer observations the less reliant the stat. same true with MAR and all draw down ratios. very few observations. in fact, by itself, draw down is less meaningful as one might think. and i think drawdown must increase over time to keep the shape of return distribution constant.
    #10     Sep 21, 2006