Risk Metrics/Risk Measurements

Discussion in 'Strategy Development' started by CPTrader, Jun 24, 2005.

  1. An issue I have debated with for many years is what is/are good risk metrics. Sharpe Ratio, Sortino Ratio, etc all have their disadvantages.

    What do you feel are good risk metrics for a trader/fund? How do you assess a trader/fund for "riskiness"? How do you determine that a trader generates good risk-adjusted returns?

    Any thought on this issue and related issues is most welcome.

    Let's brainstorm!
     
  2. kut2k2

    kut2k2

    What's the disadvantage of the Sortino ratio?

    The topic of risk is a hot button of mine, and proof positive that economystics is no science.

    An economystic foolishly chooses the value-free measure of uncertainty called standard deviation to measure the value quality called risk, the other economystics solidify this foolishness in concrete with one of their fake Nobel prizes, and now, rather than act in a mature and truly scientific manner by admitting their mistake and promising to make the required correction, a move has long been afoot to further corrupt the human language by redefining risk to be nothing more than a synonym for uncertainty.

    Those bozos have now "gifted" us with the twin idiocies of "upside risk" (an oxymoron) and "downside risk" (a redundancy). Shameless, and deserving of nothing but utter contempt.
     
  3. I trade futures with a longer term perspective, average trade duration is about 100 days. For me the goal, the only goal, is a very smooth, very linear equity curve (on semilog paper). Thanks to the huge leverage available with futures, I can dial in whatever slope I want. On semilog paper, the slope of the equity curve is equal to the Compound Annual Growth Rate (CAGR%).

    So what I want is a thing that measures smooth linearity of equity curve. I don't give a crap about whether you call it "risk" or "uncertainty" or "wiggleness" or "variability". I just want to run a large number of different backtest simulations, and sort them according to some criterion, and look at the "best" handful. The criterion I want, would bubble the smoothest most linear equity curves to the top of the list.

    I've tried MAR Ratio and Kestner K-Ratio and Seykota Lake Ratio and Ulcer Index and Sharpe Ratio and Sortino Ratio and Return Retracement Ratio and several more. The one that consistently points out the smoothest, straightest equity curves best, in my opinion, is the Sharpe Ratio. So that's what I use. I don't really care whether it's philosophically impure, or whether some people hate its idea of "risk", or whether its inventor is a fervent priest of the church of the Efficient Market Hypothesis. All I care about is, when I do an optimization run and sort the results by Sharpe Ratio, the outputs that Sharpe likes the best are also the ones that *I* like the best.

    But of course this is a matter of individual human preference. Different people have different likes and dislikes. After all, that is WHY there are so many different ways of measuring trading system goodness (Sortino, MAR, Sharpe, Ulcer, Lake, RRR, K, ...), because there is vast disagreement about which is most preferable. There is no universal "best". Instead, there is "what YOU like best".
     
  4. Thanks for the comments.

    Any more comments, ideas?
     
  5. kut2k2

    kut2k2

    You didn't answer my question:

    What's the disadvantage of the Sortino ratio?
     
  6. Phil2K

    Phil2K

    horribilicus:
    Have you thought about just calculating the regression line of the log of your equity curve? Then you can sort on the error term, which when minimized should result in the most linear equity curve. Also, the slope could be interpreted as a performance measure as well. If I'm missing something here, feel free to correct me :)
     
  7. Phil, that idea appears in Bob Pardo's 1992 book (amazon link) at the bottom of page 74. I have tried it, along with lots of other ideas. For example: what's magic about minimizing the sum of the SQUARES of the error distance between the best-fit regression line and the raw data? Why not minimize the sum of the FOURTH POWER (or any other number) of the error? I've tried those too.

    But I keep performing the same experiment and I keep getting the same result.

    Step 1: Take 30 equity curves and rank them from best to worst according to a bunch of different numerical evaluation functions. One of them is "R squared goodness of fit of a linear regression line to the (semilog) equity curve". Another is Sortino Ratio. Another is MAR Ratio. Another is Sharpe Ratio. Another is Return Retracement Ratio. Another is Lake Ratio. Scour the literature and unearth 20 of them. Jack Schwager's book "Managed Trading, Myths and Truths" is a good place to find many possibilites.

    Step 2: Using the horribilicus eyeball and the horribilicus gut, employ subjective human judgement to rank the same 30 equity curves from best to worst (most desirable, to least desirable).

    Step 3: See which of the evaluations in Step 1, gave a ranking that is closest to the human ranking in Step 2.

    For my eyeball and my gut, the Sharpe Ratio always wins the contest. However I don't pretend this is a universal truth, applicable to all people and all eyeballs and all guts. It works for me, that's all.

    As for the slope: in futures trading, slope is infinitely adjustable by adjusting your trading leverage. It magnifies the ups and it magnifies the downs. So I search for the smoothest possible curve, with the fewest mounds and craters, and then I lever it up to a "trading heat" that I prefer.