I agree that there's a been a lot of great advice given in this thread. An earlier comment inspired me to add a bit to this: uncorrelated strategies/investments are a great thing to have, and by all means should be utilized as much as they're feasible/available, but I'd be inclined to add in a substantial "fudge factor" to one's perception of correlation amongst one's strategies/positions. The reason for this is that correlations often go up during extreme market movements, and many a trader/hedge fund has been wiped out completely -- and then some -- by overestimating correlation, which seems to be a much more serious (and common) mistake than underestimating it.

Max, have you read any of Ralph Vince's work on portfolio allocation? It's not easy reading, but it's very interesting and establishes a strong mathematical foundation for risk assessment, correlation, profitability and of course allocation.

Sorry but you cannot redefine things your own way. Traders have already defined profit factor PF to be the amount of winners divided by amount of losers. The average win/avg loss ratio is: (amount of winners/number of winners) / (amount of losers/number of losers) This is a good article on the subject: http://www.tradingpatterns.com/profitability.pdf

bigdavediode, I looked up Ralph Vince's books on Amazon and the other reviewers agreed with you. I ordered his newest, it should be interesting. intradaybill, Thanks for pointing out that my reply to you didn't make sense. Here's what it should have been "I was defining profit factor as historic total win/loss ratio. So with a PF of 2 and winning percentage of 10% you would have a high probability of losing money in a week if you did not trade frequently even though the PF is positive. Sorry for replying hastily. (acrary did a better job of explaining his methodology on page 6 of his System Development thread which motivated my analysis) Corey and Euler, At this point it has been made pretty clear that I shouldn't put 100% of my faith in this simple model. I've heard that saying too. I looked for the saying's source but instead found this little analysis of how correlation actually does increase during extremes. He measured correlation between sectors, not strategies. Regards, Max __________________ maxdama.com - The log of my research on and implementation of automated trading strategies

I applaud the effort of trying to apply cs material to the trading world. One approach I did find interesting was using the hill climbing alg to find the best half-life for your moving average. Once you wander into more complex, or rather more aesthetically more complex models, you are in danger of forcing cs concepts onto markets rather than finding appropriate opportunities to to apply ai techniques. btw I didn't know they used scheme anywhere besides at mit. Great work nonetheless

Most top 10 programs seem to use some sort of functional language in first or second year introductory courses. Scheme is particularly popular, perhaps primarily due to the rather famous <a href="http://mitpress.mit.edu/sicp/">SICP</a>. At Cornell, we used <a href="http://www.smlnj.org/">SML/NJ</a>. Most fundamental principles of CS are much easier demonstrated in a functional language, so it serves as a good method of introduction. Scheme seems particularly comprehensible, if you can get past all the parentheses...

Maxdama, Acrary only used 2 variables, PF and trade frequency to calculate confidence intervals. Why you created your own PF and introduced "winning percentage" as variables ?

fluttrader, Here's a quote from page 5 of acrary's thread, "One series of studies I did was to find out what were the important factors in consistency. I did tests on size of expectation, % wins, profit factor, number of trades within a timeframe, and the effect of dispersion of trades (std. deviation) has on the results." I think we used the same variables, but also, our models were a little bit different (monte carlo vs. pure math) so it could be that there are differences. Regards, Max