Performance Measures | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

FX Dealer Performance 92 replies

Performance statistics 4 replies

Question: evaluating performance 1 reply

Performance, Fortune,& the U.S. President 0 replies

FXWINTRADES managed fund performance 3 replies

Performance Measures

  • #1
  • First Post: Oct 29, 2009 1:44pm Oct 29, 2009 1:44pm
I'm hoping to get some discussion going on performance measures, I currently run a wide portfolio of strategies, but they can be pretty much grouped into 2 general styles.

1. Momentum (lose, lose, lose, big win).
2. Mean Reverting (win, win, win, big loss).

I am currently using t-score (basically Sharpe Ratio) to rate each strategy, the problem which is is becoming apparent is that the Sharpe Ratio penalizes both upside & downside profit volatility. In the case of 1) this is bad as the big wins tend to degrade the Sharpe Ratio. Conversely, if I change to say the Sortino Ratio I then get long periods of draw-down due to the characteristics of 1). In my mind I am almost at the conclusion that these two styles cannot be reconciled in the same account and that the actually need to be split out and traded in seperate accounts. Has anybody else tackled this type of problem?
The breaking of a wave cannot explain the whole sea.
  • #2
  • Oct 29, 2009 7:23pm Oct 29, 2009 7:23pm
MAR is quite good, simple ratio (seasonal profit / drawdown) as is basic profit factor.

I studied a lot of ratios when I designed the fitness algo for the framework I use and basically it all comes back (for me at least) drawdown and profit - everything else is just about how you want to study them (seasonally, periods etc).

In the end I basically defined max DD, used profit% and adjusted risk based on these.

I don't know if this helps at all.. The real question is basically what you want to know. For me in the end the important numbers were what was the max yearly DD and yearly profit.
  • #3
  • Oct 29, 2009 7:45pm Oct 29, 2009 7:45pm
Thanks, I'll check out MAR, but I probably need something a little more responsive. Probably something else you could comment on...I'm also looking at portfolio correlation, I saw you mention this a while ago, did you end up implementing anything in your framework?
The breaking of a wave cannot explain the whole sea.
  • #4
  • Oct 30, 2009 3:01am Oct 30, 2009 3:01am
Quoting Craig
Thanks, I'll check out MAR, but I probably need something a little more responsive. Probably something else you could comment on...I'm also looking at portfolio correlation, I saw you mention this a while ago, did you end up implementing anything in your framework?
I did implement it but in the end it all went back to combined portfolio profit and dd and I didn't even use the correlation calculation I implemented. Correlation itself is not always a bad thing, correlated losses are. If you want to know correlation to your portfolio, you can look at how dd and profit change when you add a new strategy - when dd goes down but profit up the strategy is uncorrelated to your portfolio, if both go up in same ratio, strategy is correlated to portfolio.
  • #5
  • Oct 30, 2009 5:16am Oct 30, 2009 5:16am
I think its difficult to separate performance measure completely from strategy. The types of strategies I normally look at, I usually try to optimize average profit per trade vs total trades pr. session.

Average profit is important because a strategy will almost always perform worse live (unless one is very conservative on the simulation assumptions).

Total number of trades/session is important in order to gain statistical significance of the result. It will generally also smooth out the equity curve due to Law of Large Numbers.

Anyway, these observations might be less relevant for longer term strategies. In the end, I think that this is an area where auto traders might need to apply some "discretion".
  • #6
  • Oct 30, 2009 12:16pm Oct 30, 2009 12:16pm
I have been experimenting with a Sortino style measure, but the results have not been good. I am coming to the conclusion that even though t-score might penalize upside volatility somewhat, it is actually better from a portfolio point of view to take this more conservative approach.
The breaking of a wave cannot explain the whole sea.
  • #7
  • Nov 1, 2009 4:36pm Nov 1, 2009 4:36pm
Quoting mikkom
I did implement it but in the end it all went back to combined portfolio profit and dd and I didn't even use the correlation calculation I implemented. Correlation itself is not always a bad thing, correlated losses are. If you want to know correlation to your portfolio, you can look at how dd and profit change when you add a new strategy - when dd goes down but profit up the strategy is uncorrelated to your portfolio, if both go up in same ratio, strategy is correlated to portfolio.
Overlaying the new strategies results over the current equity curve is quite a tricky business, due to timings, overlaps etc, for me it is probably easier to work with correlation. One thing I did pick up during this exercise is that just taking the last n results and calculating the correlations is not good enough. I had two strategies whose results looked correlated on a graph, however when I did the correlation calculation I got quite a low coefficient. What I realized is that one of the curves was 'out of phase' by one result, so what I end up having to do is calculate a set of correlation coefficients which are generated by 'sliding' the results across one another to make sure that correlation is not hidden by a result or two putting the result sets out of phase.
The breaking of a wave cannot explain the whole sea.
  • #8
  • Nov 1, 2009 5:41pm Nov 1, 2009 5:41pm
Curtis Faith, in his book The Way of the Turtle, mentions that the Sharpe Ratio may not be the best measure for alternative investments (futures, commodities).

Though it is an excellent measure for stock portfolios, perhaps it falls short for Forex because we are much more leveraged, shorter term, and less diversified.

He also mentions that MAR Ratio might be better.

He uses a couple of examples, one being LTCM (Long Term Capital Mgmt).
Before their crisis and blowup they had almost 40 percent returns with smooth distribution, hence a decent Sharpe Ratio.
  • #9
  • Nov 1, 2009 5:57pm Nov 1, 2009 5:57pm
Quoting Intu
Curtis Faith, in his book The Way of the Turtle, mentions that the Sharpe Ratio may not be the best measure for alternative investments (futures, commodities).
Interesting point, that is a good book and he spends a lot of time talking about different measures. I'll have to go and reacquaint myself.
The breaking of a wave cannot explain the whole sea.
  • #10
  • Jun 14, 2012 2:11pm Jun 14, 2012 2:11pm
consider using concepts of multi objective optimisation (pareto optimality)
min variance, max mean.
  • #11
  • Jun 15, 2012 10:49am Jun 15, 2012 10:49am
Can you give more details of how to use concepts of multi objective optimisation (pareto optimality)min variance, max mean?Thanks.
  • #12
  • Last Post: Jun 15, 2012 11:32am Jun 15, 2012 11:32am
Refer fig in following link(somehow insert image is not working)
http://upload.wikimedia.org/wikipedi...ont_pareto.svghttp://upload.wikimedia.org/wikipedi...ont_pareto.svghttp://upload.wikimedia.org/wikipedi...ont_pareto.svg

Here is an example of pareto front where one intends to minimize f1 and f2. The point A,B line in same frontier because f1(a)>f1(b) and f2(a)<f2(b). So their is no clear winner. This approach often yields a set of solution which characterises the possible trade off. Notice that c is worse than A and B there fore it's not part of efficient frontier.


MPT has similar concept

http://en.wikipedia.org/wiki/Modern_portfolio_theory


Also you can read the following if in case you are interested in optimisation in general
http://www.amazon.com/Multi-Objectiv.../dp/047187339X
0 traders viewing now