Sharpe ratio parameters (and other measures)

Discussion in 'Strategy Development' started by macdice, May 27, 2013.

  1. macdice



    My first question is what the 'standard' period and benchmark rates are when people report Sharpe ratios without qualifying how they were computed. The two parameters I am thinking of are (1) the period (is it the mean *daily* return over the standard deviation, or some other period?), and (2) the benchmark rate of return (the Wikipedia article mentions a choice of the risk free rate and indexes like the S&P500, but I can easily imagine that some people use zero too, which might be close enough to the risk free rate currently).

    I have seen some arguments that the period doesn't matter, since the returns should have a standard distributions, so one day vs one month should give the same results. I have trouble seeing exactly why I should believe that. I probably need to do some stats homework... but it seems to me that the monthly figures are smoothed by a type of band pass filter over the returns data, ie removing higher frequency variance, and the question of the distribution of returns in financial markets seems to be contentious anyway. (Of course, when using the standard deviation in the computation we already assumed the standard distribution...) But then daily figures seem no less arbitrary, they filter out hour-by-hour variance, and ultimately I found myself wondering why we use fixed time periods at all, and not returns per position opened and closed, for a strategy that gets periodically into and out of the market.

    But of course the reason to use the Sharpe ratio at all is to be able to compare one strategy with another, so as long as the same parameters are used, it should be useful... hence my question, what parameters are others using when they post numbers? For example, how are the Sharpe ratios on computed? I found some ancient posts via Google that show that that site's recipe has changed at least once (and people complained about their numbers changing), and the latest info I could find suggested that they use monthly returns and refrain from computing numbers until there are three months' figures available (which seems like a less convincing figure to me than a day-based version).

    My second question is what other stats-based measures people recommend for optimising intraday strategies. I have been playing around with Sortino, Sterling, Modigliani 'M2' and others, many of which run into the above problem, along with simpler measures like profit factor. But I keep focussing on Sharpe ratio (whether or not I think it's actually very good for any purpose), because I have been comparing my strategies' performance with Collective2 systems and Sharpe features prominently there.

    Thanks for any feedback!
  2. dom993


    On the 1st part of your question, there is clearly no "standard" way of computing Sharpe ratio, otherwise no-one would be confused :)

    There is a big difference between using daily returns & using monthly returns ... sure, daily-based has a finer grain than monthly-based, but the "large" difference is sqrt(12).

    After thinking about it for a while, I have come to the conclusion that the only useful measure is using daily returns on the net account value, that is including unrealized PnL from open trades (which is the only way to expose martingale position sizing). But this is not the way it is done anywhere that I know of ... and as a result, you'll come across strategies with very high Sharpe ratios hiding a blow-up in the making.