I have not done the testing, but I strongly suspect that taking every trade will beat out trading after 1 loss or trading after 2 losses. in net profit i mean. for example today, the system generated 12 trades. 3 losers and 9 winners. almost a 70% win rate. after each loser there was a profitable trade. psychologically, I have an issue trading in a 100% mechanical manner. I need to express or discharge my creativity in some way. so trying to find a way to reach a happy medium.
Well, I wouldn't want to argue with RD. I didn't know that, and it's very interesting. I just have to wonder why. It wouldn't seem like a statistically outlying winner would have any impact on whether the next trade is a winner or a loser. It almost seems superstitious in a way. Then again, his net worth has a few more zeros than mine, so as tomahawk suggested, maybe it's worth back testing?
I think it has to do with the volatility cycle. The Turtles systems were looking for the big run ... after big runs markets tend to rest (chop), which of course is deadly for breakout systems. I would imagine they tested it both ways (taking every trade vs waiting for a fail).
This is a good question. If you can use the win losses then it means there is auto correlation. That means that there is some pattern to wins losses. Some people treat these as "independent". Some systems will show auto correlation -- others wont. A good example would be a simple over sold/bought indicator will tend to produce batches of losing trades because the market when going into high oversold/overbought is often the result of a strong trend. You have to think about what you are measuring. Let us say you test after 3 losers then you get a winner 90% of time. You are really measuring the cycle period in the market, i.e you are measuring how long the market tends to stay overbought or sold. So, there may be some deeper logic there that you could exploit -- if it existed.
The probability of the next trade being a winner or a loser is always 50% +/- a statistically significant edge, if it exists. There are methods in statistics to estimate the edge and its significance. They are taught in graduate level. Since we have no way of knowing from a limited set of observations whether you indeed have an edge, the probability of your next trade to be a winner is 50% +/- a random edge that is normally distributed with mean exactly 0. Anything else, I am afraid to say, including statements that the probability is equal to some win rate, are outcomes of a misunderstanding of statistics and reality. http://en.wikipedia.org/wiki/Gambler's_fallacy
was this a joke? here's a better approach for your type of thinking. bet someone on the next toss of a coin. Since its 50/50 then you know that the next toss will be the opposite of the first.
Some good food for thougth, thank you for your comments. A while back, in another thread, I exchanged a post with intradaybill and he mentioned 'waiting' for high probability trades was the best lesson that he learned. I believe one has to know the system, what mkt behaviour is it capturing. So this approach would be very different from a system built on a pure datamining analysis. For example Michael Harris APS, which I believe just combs through data looking for patterns. With this approach, one would start with a clear goal. for exmple the goal would be to capture intraday trends or oversold/overbought mkt conditions. Unlike traditional system trading, the system itself would NOT be of importance, but the trade series or mkt behavior that it tries to capture...
I am sorry, but none of you above are correct what so ever. The solution is in Bayes' theorem and the most vivid illustration of it is presented via the Monty Hall paradox. Please educate yourself. MAESTRO http://en.wikipedia.org/wiki/Monty_Hall_problem http://en.wikipedia.org/wiki/Bayes'_
+ 1 And you are all comparing oranges with apples... If you have a system that is for example 65% right, then every next trade has the same probability. A winner or loser does not change that. On the other side of the story,if you have a distribution of your trades and you can make some assumptions that you will have a new high after X trades, you can use a money management rule that takes advances of this. It has nothing to do with your probability or win or lose trades in a row...
Please re-read my answer early in the thread. This is often correct, but not always. Some systems do show significant periods of autocorrelation in their trades. Maestro's answer is correct in a theoretical sense (though the Monty Hall parallel is a bit tenuous at best), but in actual work, not always. It is fairly easy to evaluate this from a Bayesian perspective. If you have a large enough sample size, consider Prob(A), the probability that any trade is a winning trade, compared to Prob(A|B) where B is the probability the previous trade was a winning trade. if Prob(A) != Prob(A|B), allowing for variability, then you have a system that may have some autocorrelation. This is precisely the element of system development that makes Kelly sizing suicide with a lot of systems--they have more runs of winners and losers than theory would suggest they should. Bottom line is that this is not as simple as many people think. It is important to understand the theory, but equally important to understand where practice diverges from theory.