I'm currently at: Strategy actions: wait/skip/enter/long/short/hold/exit/adjust stop/adjust limit Using the feedback loop of consecutive trades, one more action to add: abandon or review strategy (after a larger than expected losing or winning streak) The reasoning as, there is a low probability the current system is still valid given the larger than expected streak. This increases the variety of the strategy, to increase its ability to respond to what the market does, thus increasing the Requisite Variety, I suppose by one bit. I hope there is another concept or wikipedia article to include, to get to the next step ...
imo the requisite variety idea is off the mark and far over-complicates something that is actually very simple. can u post an excel sheet with the trades? date and P&L as % (or otherwise standardized) are all that you need.
I didn't test it out, more trying to reason by the theory alone. I agree, it's probably not a practical approach and there are probably simpler ways to improve trade performance. Also I'm more in the trade independence camp. But, I'm interested in theory that challenges my thinking, assumptions, beliefs, or in new concepts I haven't heard of. For a practical approach, I use a system stop -- trade until it hits a pre-set number of consecutive losses. The next trade may be a win or loss with same probability, or not. I don't know, best to cut losses short, the market may have changed. I've found my strategies don't perform the same forever, they either expire eventually or revert to some longer term probability that my limited backtesting missed (I've seen that discussed in threads too).
OK, looking at my 199 trades done this year and ordered by entry date, I have Prob(A) = 0.6281 Prob(A|B) = 0.6400 Also, just in case it helps, my rolling average of last 20 trades' win rate: AVERAGE 62.81% MEDIAN 65.00% MIN 30.00% MAX 90.00% Is the extra win rate of +0.0119 on a sample size of 199 trades statistically significant to conclude that my system exhibits autocorrelation?
Try a Pearson Correlation Coefficient with your returns staggered by one to see if they're autocorrelated. Zero would be uncorrelated, -1 would be negatively correlated (a loss followed by a win or vice-versa) and 1 is strongly positively correlated (wins followed by wins.) It's not ideal because it assumes a std dev, which itself assumes a normal distribution -- but it's a start. As well that kind of leads you to look at the distribution of your returns to see their normality (or lack thereof), kurtosis, etc. Studying this can only help you. Good luck.
Thank you zed and others for your contribution to this thread. I learned a lot. FWIW. here is the trade series from today, Tues 5/31/11 P1-P2-L3-P4-P5-P6-P7-P8 My goal is to turn the trade series into a indicator/system I came up with a simple rule to wait for a losing trade. take the next two trades. then stop and reset. wait for the next sequence. So I took P4 and P5. netted 1 ES pt. terriable slippage as usual. p=profit. L=loss. 6 trade number. total system generated trades today were 8 feedback, comments, good, bad, ugly all welcome. I will try to keep this post up for the rest of this week. Thanks.
FWIW. trade series from today, Wed 6/1/11 P1-P2-L3-P4-P5-P6-P7-P8-L9-P10-P11-P12 following the rules, waiting for a loss, I took P4, P5 and then P10, P11 net after comm, slippage, 2 ES pts system was 83% profitable. but lost $60 feedback, comments, good, bad welcome. Thanks
All systems fail when market conditions change, there is random distribution of profits and losses.You could get period of 30 losses to 3 wins ,if market conditions become less favourable to your system. Past performance of a system is no guarantee of future.