Evaluating Skewed Strategy Returns at Realistic Capital Sizes
Summary
The document considers how to compare backtested strategies when returns are not normally distributed, especially when negative skew and leverage make a Sharpe ratio an incomplete description of risk. It raises maximum drawdown as a possible measure of severe losses, but the response recommends using several complementary metrics, including downside deviation and the Sortino ratio, alongside Sharpe and drawdown.
It also emphasizes evaluating candidate strategies at realistic capital sizes. Trading costs, market impact, and broker fees can change with position size, so ratios that abstract from scale may not describe implementable results. Comparing a small set of capital allocations allows these effects to be incorporated directly. The response treats choosing one strategy as simpler than combining several, since portfolio allocation assumptions add complexity. It offers practical guidance rather than an industry standard for measuring skew, and gives no empirical comparison or rule for setting leverage; metric choice and sizing still depend on the intended use and implementation.
Key ideas
- Sharpe alone may not adequately characterize strategies with nonnormal or negatively skewed returns.
- Downside deviation, Sortino ratio, and maximum drawdown can complement Sharpe in evaluating downside risk.
- Strategy performance should be assessed at candidate capital sizes because costs and market impact can scale with size.
- Choosing among combined strategies requires capital allocation assumptions beyond those needed to select one strategy.
- The discussion offers no universal skew metric or leverage-setting rule.
Tags
Full text
# Evaluating trading strategies by the skewness of returns # Evaluating trading strategies by the skewness of returns How to deal with skewness of returns when evaluating different trading strategies? More specifically, I'm back testing different strategies to be implemented as an automated black box strategy. While e.g. Sharpe ratio is a convenient and intuitively simple way to compare the strategies, but if the returns are not normally distributed it is vastly inadequate measure if one considers e.g. the use of leverage. This point is all the more important, when one considers e.g. many of the dynamic strategies like momentum that typically has a negative skew. This leads me to my question: what is the most convenient way/is there an industry standard to comparing the skew of a trading strategy? Or does everybody simply use max drawdown to evaluate the possibility of the strategy blowing up and as a guideline when considering how much leverage to use? ## Answer by Brian B (score 4) https://quant.stackexchange.com/a/28076 You do not state whether your evaluations will result in potentially implementing multiple strategies or just one of them. This matters because if you are going to be combining multiple ones then you need some reasonable capital allocation assumptions, which increases complexity immensely. Let's take the simpler case where you just want to choose one. Multiple metrics depend on the strategy size, of which unscalable mean/variance (Sharpe) ratios are just one. For example, size affects trading costs, market impact, and broker fees. Thus, I suggest you choose a few candidate strategy sizes (in terms of capital) and do your analysis just on them. You can then properly add costs and market impact, and you do not need to worry about scaling. I'll further add that if negative skew is important you may wish to look at downside deviation and Sortino ratios in addition to Sharpe and max drawdown.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.