Comments on: Debate over value of Sharpe Ratio in HF analysis continues in new academic study http://allaboutalpha.com/blog/2009/07/13/debate-over-value-of-sharpe-ratio-in-hf-analysis-continues-in-new-academic-study/ Hedge funds, portable alpha, 130/30 and alpha-centric investing Fri, 19 Mar 2010 07:52:26 +0000 http://wordpress.org/?v=abc hourly 1 By: Ranjan Bhaduri http://allaboutalpha.com/blog/2009/07/13/debate-over-value-of-sharpe-ratio-in-hf-analysis-continues-in-new-academic-study/comment-page-1/#comment-186097 Ranjan Bhaduri Sat, 15 Aug 2009 16:16:30 +0000 http://allaboutalpha.com/blog/?p=4958#comment-186097 If one ranks a set of hedge funds via Sharpe, and via Omega, then one should (in most cases) expect to see similar rankings for some of the funds (though in comparing a set of rankings, one should take care in selecting the appropriate threshold for the Omega function). Funds that have a high Omega ranking compared to a low Sharpe ranking, will most typically be funds that are being penalized for upside volatility by the Sharpe. Funds that have a high Sharpe and a low Omega score, are those that most typically have some unfavorable characteristics in their higher statistical moments that are being ignored by the Sharpe. The Sharpe Ratio is only good for funds that have a Gaussian (i.e. Normal) Distribution since it only takes the first two statistical moments into consideration (look at its formula). The Omega function encodes all of the higher statistical moments (again, look at its formula - as some mathematicians say "A formula is worth a 1000 pictures"). The fact that some of the rankings via Sharpe will be similar to the Omega, does not in any way justify the use of the Sharpe. One would simply be using inferior statistical analysis via the Sharpe, and making themselves more vulnerable to tail events. The Sharpe ratio is mathematically uncivilized, and it's rather embarrassing for the industry that some folks still use the Sharpe, and the Sharpe's cousin - the Information Ratio - which is just as bad, if not worse. I do encourage readers of this blog, to read Dr. Shadwick's posts. If one ranks a set of hedge funds via Sharpe, and via Omega, then one should (in most cases) expect to see similar rankings for some of the funds (though in comparing a set of rankings, one should take care in selecting the appropriate threshold for the Omega function). Funds that have a high Omega ranking compared to a low Sharpe ranking, will most typically be funds that are being penalized for upside volatility by the Sharpe. Funds that have a high Sharpe and a low Omega score, are those that most typically have some unfavorable characteristics in their higher statistical moments that are being ignored by the Sharpe. The Sharpe Ratio is only good for funds that have a Gaussian (i.e. Normal) Distribution since it only takes the first two statistical moments into consideration (look at its formula). The Omega function encodes all of the higher statistical moments (again, look at its formula – as some mathematicians say “A formula is worth a 1000 pictures”). The fact that some of the rankings via Sharpe will be similar to the Omega, does not in any way justify the use of the Sharpe. One would simply be using inferior statistical analysis via the Sharpe, and making themselves more vulnerable to tail events. The Sharpe ratio is mathematically uncivilized, and it’s rather embarrassing for the industry that some folks still use the Sharpe, and the Sharpe’s cousin – the Information Ratio – which is just as bad, if not worse. I do encourage readers of this blog, to read Dr. Shadwick’s posts.

]]> By: S Jacobs http://allaboutalpha.com/blog/2009/07/13/debate-over-value-of-sharpe-ratio-in-hf-analysis-continues-in-new-academic-study/comment-page-1/#comment-184034 S Jacobs Thu, 13 Aug 2009 02:32:01 +0000 http://allaboutalpha.com/blog/?p=4958#comment-184034 garbage in ... garbage out! Regardless of the risk measure statistic, using only month end data, when all is marked to market daily -- and sometimes minute to minute if highly leveraged & volatile -- ignores the interday risk we all live with, as traders & investors. How many positive, smooth, month end to month end returns really consist of x number of std deviations within the month? Hence, trading system back/acid walk forward test design, on a trade by trade time period basis, yields hypothetical data events far superior for valid statistical analysis than the month end to month end statistical snap shots of legacy track records over a systemically skewed time preference marketplace. In other words, all else being equal, does an investor give their funding to a trader with a "superior" risk/reward 5 year month end to month end real track record, or a new trader with 10+ years of hypothetical back/acid test walk forward results, trade-by-trade, marked to market, that can be R&D duplicated to measure data & algorithm integrity? I'd go with the new trader every time. garbage in … garbage out! Regardless of the risk measure statistic, using only month end data, when all is marked to market daily — and sometimes minute to minute if highly leveraged & volatile — ignores the interday risk we all live with, as traders & investors. How many positive, smooth, month end to month end returns really consist of x number of std deviations within the month? Hence, trading system back/acid walk forward test design, on a trade by trade time period basis, yields hypothetical data events far superior for valid statistical analysis than the month end to month end statistical snap shots of legacy track records over a systemically skewed time preference marketplace.

In other words, all else being equal, does an investor give their funding to a trader with a “superior” risk/reward 5 year month end to month end real track record, or a new trader with 10+ years of hypothetical back/acid test walk forward results, trade-by-trade, marked to market, that can be R&D duplicated to measure data & algorithm integrity? I’d go with the new trader every time.

]]> By: Skip Thevenot http://allaboutalpha.com/blog/2009/07/13/debate-over-value-of-sharpe-ratio-in-hf-analysis-continues-in-new-academic-study/comment-page-1/#comment-172513 Skip Thevenot Thu, 23 Jul 2009 18:10:42 +0000 http://allaboutalpha.com/blog/?p=4958#comment-172513 A Sharpe Ratio is a sharpe ratio no matter what it is measuring, i.e. equities, commodities, other investments... correct ? What is, say an industry accepted and desired level of Sharpe Ratio... anything above 0.5 considered attractive ? A super attractive investment is one greater than a Sharpe of 1.0 ? I heard that Goldman targets 0.5 - 1.0, does this sound reasonable ? Thanks A Sharpe Ratio is a sharpe ratio no matter what it is measuring, i.e. equities, commodities, other investments… correct ?
What is, say an industry accepted and desired level of Sharpe Ratio… anything above 0.5 considered attractive ? A super attractive investment is one greater than a Sharpe of 1.0 ? I heard that Goldman targets 0.5 – 1.0, does this sound reasonable ? Thanks

]]> By: Peter Urbani http://allaboutalpha.com/blog/2009/07/13/debate-over-value-of-sharpe-ratio-in-hf-analysis-continues-in-new-academic-study/comment-page-1/#comment-170397 Peter Urbani Sun, 19 Jul 2009 01:26:35 +0000 http://allaboutalpha.com/blog/?p=4958#comment-170397 Further ETL a.k.a Expected Shortfall (ES) and Conditional Value at Risk (CVaR) stands for Expected Tail Loss and not Extreme Tail Loss which invites spurious comparisons with Extreme Value Theory. For more on this see Kevin Dowd's 'Measuring Market Risk' or any of several other references. The ratio of the right tail measure to the left tail measure is equivalent to an Upside Potential Ratio measure of which there are also several forms including the Gaussian and Forsey and Sortino's 3-parameter lognormal version. Empirically given 100 observations of your P&L sorted in descending order your 95th percentile Historical VaR will be the 95th observation and the 95th percentile Historical CVaR will be the average of the 5 losses greater than the 95th loss i.e. the expected / probability weighted VaR conditional on the 95th percentile having been exceeded. This measure meets all of the measures of mathematical coherence as proposed by Artnzer et al most particularly sub-additivity which VaR does not save for the special case of multivariate normality. Further ETL a.k.a Expected Shortfall (ES) and Conditional Value at Risk (CVaR) stands for Expected Tail Loss and not Extreme Tail Loss which invites spurious comparisons with Extreme Value Theory. For more on this see Kevin Dowd’s ‘Measuring Market Risk’ or any of several other references. The ratio of the right tail measure to the left tail measure is equivalent to an Upside Potential Ratio measure of which there are also several forms including the Gaussian and Forsey and Sortino’s 3-parameter lognormal version.

Empirically given 100 observations of your P&L sorted in descending order your 95th percentile Historical VaR will be the 95th observation and the 95th percentile Historical CVaR will be the average of the 5 losses greater than the 95th loss i.e. the expected / probability weighted VaR conditional on the 95th percentile having been exceeded. This measure meets all of the measures of mathematical coherence as proposed by Artnzer et al most particularly sub-additivity which VaR does not save for the special case of multivariate normality.

]]> By: Peter Urbani http://allaboutalpha.com/blog/2009/07/13/debate-over-value-of-sharpe-ratio-in-hf-analysis-continues-in-new-academic-study/comment-page-1/#comment-167720 Peter Urbani Tue, 14 Jul 2009 09:09:11 +0000 http://allaboutalpha.com/blog/?p=4958#comment-167720 270 funds is hardly a representative sample. We have tested more than 8000 hedge funds using 13 different distributions and four different goodness of fit measures and find less than 12% of those funds to have returns that are normally distributed. 270 funds is hardly a representative sample. We have tested more than 8000 hedge funds using 13 different distributions and four different goodness of fit measures and find less than 12% of those funds to have returns that are normally distributed.

]]>