trade optimization

[ctd]

What about omega? I've been quite pleased using a modified version of omega (ratio with a single threshold as opposed to the ratio across thresholds) in backtesting my strategies.



If there is an "industry standard" I think it would have to be Sharpe or IR, though I too prefer Sortino given its omission of upside vol. Obviously though, these two only speak to two moments of your return distribution, so you're only getting a partial picture--omega has the benefit of completely describing the distribution of returns.



IMHO, the "best measure" depends quite heavily on the nature of your trading system(s) (ie, the nature of your return distribution) and your objectives. Can you provide any more color regarding your systems/objectives?

[nodoodahs]

Since meteor started the thread, it would be nice to see that additional color from him.



I would just like to clarify in terms of color on systems/objectives, good information might include the amount of money run through it (6 digits? 11 digits? imbetween?), customers for the system (or is it for personal use?), issues to be traded (stocks, futures, options?), timeframes (holding and/or trade frequencies).



Some of those might impact which type of metrics are recommended.  I tend to (but don't most of us?) answer through the lens of what I would be testing for my use ...

[cpptrader]

hadnt heard of the omega measurement before, so I did a quick search and found the Keating and Shadwick 2002 paper, publicly available, and the application by some people from Winton. both are attached below. seems like an interesting measure that avoids the heavy MM nodoodahs is opposed to.



[url=/User%20Files/2463/GammaPub%20-%20Keating.pdf]Attached File: GammaPub - Keating.pdf[/url]



[url=/User%20Files/2463/edge%20fund%20omega.pdf]Attached File: edge fund omega.pdf[/url]

[meteor]

nodood, I have been looking at the different metrics and I havent yet found a clear winner since these different metric provide me different set of parameters.



I have decided now to step back a bit and look at the meaning of the paramters I am optimizing and how this meaning can be translated into values of these paramters (which make sense to me )



Then my goal is to select a couple of paramters region I am confident with (ie make sense to me) and look at different performance metrics they produce.



cptrader, thanks for the papers

[nodoodahs]

I discovered the same thing. 



Not only do different metrics give different parameters/"best" systems, the same metric can give different answers if the benchmarks or hurdles are changed.



Examples of the benchmark change: regress the excess returns of multiple stock trading systems against the U.S. total market, S&P 500, and Russell 2000 indices.  You will come up with a different Alpha and Beta for each benchmark for each system, sometimes making one system better or worse than another.



Example of the hurdle change:  calculate a Sharpe, Sortino, or for that matter, an Omega, at different RF or MAR rates for every system in test.  The ranking of the systems may well change due to the different benchmark parameter (I believe the paper shows that in one or two examples).



I "solved" this problem for myself by looking at the different metrics through the lens of what I wanted for results, and finding the metrics that fit my desires.  In the end, I went with a combination of very simple metrics and a set of rules (hurdle these two metrics, then maximize this other one) rather than one complex metric.

[ctd]

Seems like a reasonable strategy. When I started backtesting, I found it somewhat interesting to calculate ALL of my metrics for a given strategy and attempt to glean some insight into what kind of time-series behavior may be driving high or low value



It's a bit of an ethereal exercise, and the verdict is out as to whether or not it helped me in any real way, but it was an interesting one regardless, particularly given my lack of experience with performance measurement in general.



Regarding the omega measure--I didn't mean to leave you hanging and it appears cptrader has taken up the slack, but one of the papers I've found helpful in the past can be found here:

Omega

[Tradenator]

cpptrader, thanks also for the papers. 



ctd, can you please check your link, thanks.

[zinmaster]

More along the lines of what I'm looking for:

"How to price Hedge Funds: From two-to four-moment CAPM",

Angelo Ranaldo & Laurent Favre



"Portfolio Selection With Higher Moments", Campbell Harvey, et al.



"INTRODUCING HIGHER MOMENTS IN THE CAPM: SOME BASIC IDEAS", Gustavo M. de Athayde and Renato G. Flôres Jr.



and there's a Barone-Adesi paper on 3-moment CAPM that I can't locate just this second.



The Athayde & Flores paper is interesting in that it purports to eliminate the utility parameter & build orthogonal factors of higher moments, and price them.

[ctd]

woops, thx...updated.

[shavinOccam]

i still prefer sharpe to sortino, because i think that sortino biases your choice to rare positive events. my assumption is that whatever goes FOR you can go the same way AGAINST you. so i punish systems for excessive upside and downside vola.



yet i more and more think that below a certain level, let's say a sharpe of 1, there is a lot of chance involved anyways. and above that a look at the chart tells you what the numbers might not.



a system that trades above 100 times per year and has a sharpe above 1.5 will not show bad stats using any other measure ... what is more important to me than the choice of the figure is persistence in certain environment, like bull/bear, low/highVola.