Hello, I'm looking for a mathematical formula to compare strategies pnl curves (they have diff # months/years, diff pnl, diff drawdown), I guess my goal is to classify them so the longest consistent result with the smoothest and highest pnl are first.
What do you use?

I cc some ways I found on the web
https://www.tradingblox.com/Manuals/UsersGuideHTML/tradingperformancestatistics.htm

Modified Sharpe Ratio and Annualized Sortino, seems interesting
What do you think?

thanks for your advises, have nice weekend
w

I think Sharpe ratio is an excellent method of distilling system performance down to one number, factoring risk ibto the equation. Personally, the next step I would take would be to find a way to identify outliers in each candidate.
I would have a problem with a system that outperforms others, yet has rare outliers of high loss. The averaging effect of any rating method might tend to disguise a few very bad trades.

Heph333 

I think Sharpe ratio is an excellent method of distilling system performance down to one number, factoring risk ibto the equation. Personally, the next step I would take would be to find a way to identify outliers in each candidate. I would have a problem with a system that outperforms others, yet has rare outliers of high loss. The averaging effect of any rating method might tend to disguise a few very bad trades.

Thanks a lot Heph333
To remove outliers on each candidates, do you remove the biggest abnormal positive and negative (both) trades even if they seems legitimate (like I understand if they are cause by some input data abnormally) but do you still remove them when everything looks fair?
thanks
w

Van Tharp has a system quality number, SQN, designed to make it possible to compare systems. His SQN rewards consistent performance because a system that is consistent and profitable can be ramped. Profitable systems with large swings are much riskier to ramp up your position size due to the increased risk of ruin.

Sharpe ratio is roundly criticized for confusing variability with risk. It is also very popular with the mutual fund industry, therefore it is automatically suspect as possible snake oil.

Let me add because it does not measure market risk just variability. So they imply to customers that it is a risk measure. See how that worked out in the last crash. Your concerns about outliers says you are interested in the deviation in results. Trading systems are hardly ever examples of the normal distribution. Re-read Taleb on fat tails.

Cool to see someone still referencing the Tradingblox website - lots of useful info to be found there.

I personally like MAR since it gives me a decent reflection of maximum pain and the expected gain. However, it can punish one severely if there is only one large drawdown and all other drawdowns are much smaller. Of course, the largest drawdown may still be coming, which will then again influence the MAR.

Once of the big downsides of this type of performance measuring is that performance changes with various start/end dates. A system that ends in a big drawdown when testing stops could be the one that outperforms all other systems over the next year, but when the decision needs to be made which system to use, its performance numbers will be rather poor.

Regarding your two favourites, Sharpe ratio also penalises upside volatility. If you have stops in place, then perhaps Sortino would be a better measure of performance.

Regarding the removal of outliers, I would say you do so at your own risk. Unless you understand why the outliers occurred and you can accurately assess that they won't happen again you could be understating the risk of your system.

Just to reference some similar threads

System Optimization Fitness Functions My hope in starting this thread is that others will share their preferred fitness functions, that together we may come up with a better function. I know Big Mike has one I've never seen posted (hint), hopefully others also have ideas we can share …

Big Mike's Trading Blog: Van Tharp's Max Expectancy and System Quality Number (SQN)