Would it make sence?
Would it make sence?
To trade those instruments only from the side that had better performance during the backtest? For example: I backtest simple trend following system and on Eurodollar it made 200k on long side and only 30 k on short side since 1984 until today. On Cotton same system made money on short side 50k and lsot 25k on long side same period and so we go until Wheat. Now would it make sence to start trading this system in a next manner: Eurodollar from long side only, Cotton from short side only and all others in the same order?
If there are 25 markets in the portfolio, this idea adds 25 new parameters to the trading system: one parameter per market. (namely: do I trade this particular market LongOnly? ShortOnly? LongAndShort?)
Perhaps choosing the new parameters on one timeslice of data (such as 1988-1998), and then simulating with those parameter choices on a different timeslice of data (such as 1999-2008) would help you decide. This would be a type of "walk-forward" testing.
Perhaps choosing the new parameters on one timeslice of data (such as 1988-1998), and then simulating with those parameter choices on a different timeslice of data (such as 1999-2008) would help you decide. This would be a type of "walk-forward" testing.
thanks so much sluggo.. yeah its tested on 25 futures markets.. you're right it'd be just as you said the decision wether I trade this commodity long, short or long/short. I'd have to add then to the code for each commodity what side it should trade and then see the results and then change the time scale..I use EOD data, what you'd suggest changing to. Weekly or more frequency like Intrda day?
It seems like excessive curve fitting to micro manage each market in that manner. The future is seldom so very close to the past as we know. Sluggos' suggestion to slice up your data into windows of time and optimize your concept on window A then test it on B C D. Then optimize B and test on A C D etc. Also run the method without micro managing trade direction, take all trades using parallel time windows. When this is complete you may compare wether the micro or blunt approach best fits the modeling period. Some have observed that a strictly long only system had certain plusses which appealed to their personality. You could structure your test and log the results in excel to manage the test more efficiently.
thanks RedRock..I am back to testing and numbers make some sence to me...gotta do some more A,B, C time windows back testing too as you suggested in order to prevent some certain period curve fitting. My system is 2 days hold on average, ran it on 20 years of EOD, but still wanna have it as robust as I can get..