Thank you MAESTRO for the encouraging comment, and engaging this discussion. Also, I've wanted to thank you for your comments on this thread: can linear regression predict the future? http://www.elitetrader.com/vb/showthread.php?s=&threadid=181558&perpage=6&pagenumber=1 I was able to follow the thread but unfortunately never figured out the conclusion. Regardless, the things I had to study (random walk, expected number of steps, non-Gaussian distributions, and more) to follow the thread have since proved to be very useful and helpful in conducting my strategy research. Definitely a treasure on ET. Sorry, I couldn't figure out how the cubic spline was connected, but it was interesting to learn about. And thanks to everyone else contributing to this thread. 2-in-1 systems are something I've had some ideas about, perhaps putting them into words will help myself and others. It's good to hear different perspectives. thanks !!!!!
How about this: since I control my trades, they can be independent or not independent, from how I make my system, it's my choice. More important : from the point of trade entry, is the price movement after entering next trade going to be independent or not independent from what happened in my previous trade. Perhaps this view was intended or assumed. Best to state it to be sure. New info can be read as the trade progresses, and open several lots instead of just one when placing trade, to give me another couple of control techniques -- add and reduce lots as trade goes on -- to be able to absorb more of the variety from the price.
Thank you all for your contribution to this thread. I think I am slightly more confused then when I first started. So there appears to be two schools of thought: 1) All the trades in a trade series are 'independent', regardless of the system being traded. 2) The trades in a trade series of a 'properly' designed system are 'depentent' or somewhat correlated. The implication being the 'current trade' is effecting the win/loss probability of the next trade? Thanks.
I don't know, but I think that sounds right. For 2), you can get some new info from past trade outcomes (or some other stream of data knowable in the present as you go) for the decision making for the next trade. Everything I've read, and my general experience, is that trades in my systems are independent, wins and losses randomly distributed over the backtest data, and going forward, the last trade or streaks do not offer any info for the next trade. That is why I am very interested to follow the converse argument, look for what I've missed or never considered.
You can do a simple analysis that will help. You don't have to guess or assume! if you have a large enough sample of trades that are arranged in chronological order: 1. calculate the simple probability that a trade is a winner. this is simply the win% for the set. (This is Prob(A)) 2. now calculate the conditional probability that a trade is a winner, given that the previous trade was a winner. (This is Prob(A|B)) Go through your trades and select only the trades that follow winning trades, and copy them to a separate sheet. Now, calculate the probability of this set. if Prob(A) <> Prob(A|B) you may have some autocorrelation in your system returns. there are issues of sample size, etc to consider, but this is a very simple and rough way to get some intuition about the concept. imo, everyone is overcomplicating this. some systems are uncorrelated, most aren't. you can do analysis and figure it out. in my experience though, a strategy like the OP suggested usually does not work very well. there certainly could be exceptions.
Well, here are some stats from the system. So, I don't plan to trade the actual system, but trade or at least pay attention to trades that happen after 1 loss, 2 losses. So the system is deliberately designed to take advantage of the 'trade series.' Total Number of Trades: 830 Winners: 629 Losers: 201 Percent Profitable: 75% long trades: 414 short trades: 416 I think what interests me is the Losing series analysis Losing Series -- Number of Series 1 ---- -- 117 2 ------ 24 3 ---- --- 6 4 ---- --- 2 5 ------- 2 So there were 6 occurances where there were 3 consecutive losers in a row. and 2 occurances where there were 4 consecutive losses in a row. you get the picture. So this is what is of interest to me. how the Number of Series drops off sharply after 2 consecutive losses in a row. So the idea here is to wait to take a trade only after 1 or 2 losers... any thoughts, comments, good or bad? Thanks
I think, per AdamG_SMB's comment, it should be looked at as "of those that made it here, how many continued?" Of 117, 24 lost again = 24 / 117 = 20.5% Of those 24, 6 lost again = 6 / 24 = 25% Of those 6, 2 lost again = 2 / 6 = 33.3 % Assuming there's some error, those are close to 25%, which was the initial loss rate. It's like there is some sort of mostly constant "decay" rate, this seems to draw out an example of independence, I think this is similar to AdamG_SMB's post. I suppose what we'd look for would be stats like "117 - 116 - 115 -2 - 2", or "117 - 116 - 20 - 20 - 20 - 20" to suggest non-independence.
Plugging your trade stats into my expected max losing streak formula: s = ln( 1/ number of trades ) / ln(loss rate) s = ln (1/830) / ln(.25) s = -6.72... / -1.386... s = 4.84 if that's right, then a losing streak of 5 would be expected, as the maximum losing streak.