Comments on: Bliss recipe with a robustness spice http://www.automated-trading-system.com/bliss-recipe-robustness-spice/ Systematic Trading research and development, with a flavour of Trend Following Tue, 07 Feb 2012 09:06:21 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: Jez http://www.automated-trading-system.com/bliss-recipe-robustness-spice/comment-page-1/#comment-234 Jez Sun, 07 Feb 2010 21:44:50 +0000 http://www.automated-trading-system.com/?p=1489#comment-234 Hi RiskCog, This comment is nearly worth a post of its own! Thanks very much for the feedback. I have never dabbled in fuzzy logic but this conceept sounds interesting because - let's be honest - the idea of putting your bliss function in a perfect equation seems a bit difficult. Thanks again - comments like this are one of the reasons why I am glad I started this blog! All the best with your RiscCog website - it looks like a useful tool! Hi RiskCog,
This comment is nearly worth a post of its own! Thanks very much for the feedback.
I have never dabbled in fuzzy logic but this conceept sounds interesting because – let’s be honest – the idea of putting your bliss function in a perfect equation seems a bit difficult.
Thanks again – comments like this are one of the reasons why I am glad I started this blog!
All the best with your RiscCog website – it looks like a useful tool!

]]> By: RiskCog http://www.automated-trading-system.com/bliss-recipe-robustness-spice/comment-page-1/#comment-233 RiskCog Sat, 06 Feb 2010 06:16:11 +0000 http://www.automated-trading-system.com/?p=1489#comment-233 Interesting discussion. If I were going to fully automate my bliss function, I think I would use fuzzy logic instead of composing an equation. The first step is to decide what you want such as: high return, small losses, no flat periods, small market exposure, low sensitivity to input parameters, etc. The second step is to decide how to measure each feature so respectively this might work: cagr, max DD, min rolling 18mo return, % of time in market, worst case bliss function for 5% parameter skew. It makes sense to roll the robustness measure in after the main bliss function is computed as you point out. The third step is to decide for what measurements the objective is met. For example say that I don't consider any annual cagr less than 50% to be "high return", and anything over 100% is definitely high return. Then to fuzzify this you assign 0.0 to returns of 50% and below then linearize so 51% is 0.02, 75% is 0.5, then 100% is 1.0 and 102% is 1.0. After completing step 3 for all the factors I want to measure, I can combine them with a Fuzzy AND operation. Various definitions of AND are possible but using the minimum function works quite well. So my bliss function would be the minimum of all of the measurements; "high return", "small losses" etc. Interesting discussion. If I were going to fully automate my bliss function, I think I would use fuzzy logic instead of composing an equation. The first step is to decide what you want such as: high return, small losses, no flat periods, small market exposure, low sensitivity to input parameters, etc.

The second step is to decide how to measure each feature so respectively this might work: cagr, max DD, min rolling 18mo return, % of time in market, worst case bliss function for 5% parameter skew. It makes sense to roll the robustness measure in after the main bliss function is computed as you point out.

The third step is to decide for what measurements the objective is met. For example say that I don’t consider any annual cagr less than 50% to be “high return”, and anything over 100% is definitely high return. Then to fuzzify this you assign 0.0 to returns of 50% and below then linearize so 51% is 0.02, 75% is 0.5, then 100% is 1.0 and 102% is 1.0.

After completing step 3 for all the factors I want to measure, I can combine them with a Fuzzy AND operation. Various definitions of AND are possible but using the minimum function works quite well. So my bliss function would be the minimum of all of the measurements; “high return”, “small losses” etc.

]]>