I'm trying to develop an indicator based on patterns of movement. One of the questions I'm trying to determine an answer for is the number of bars I'll need to look at. For example if I wanted it to recognize things like double tops, or head and shoulders type patterns trends, etc. I assume I would need at least 20 bars. But would I need 30? 40? more? I don't really know any TA and that's why I'd like the computer to do the recognition work for me. Any suggestions for the number of bares I should include? I was thinking of looking at 2 different time frames. Daily charts is the first and 1 minute bars would be the second. I would love it if the it would work for day trading but I'm assuming patterns may be more prevalent in the daily bars. Thanks in advance, MPO Note: I meant to have the poll question be: How many bars are needed for most TA patterns?
I think just like all TA, the timeframe, number of bars, "severity" of the move, etc, etc are all subjective. :eek: I would say, make your number of bars an input and don't hard code it. Run it against various timeframes with various bar counts and see how it works.
Thats a tough one, because there are so many patterns. With a double top for example, its range based so your bars # could be anything. Same with an abc pattern, it could be as few as 4 or a macro abc, with any number of bars. But for congestion areas, as few as four might do the trick-less than that, well theres your three bar reversal for a start, so its going to depend on which patterns you want to look for, practically everything is a TA pattern of some sort.
Using # of bars is one approach... Another approach is to identify the previous swing high/low which would not be constrained by the # of bars. Of course you would still need to define what is a swing high/low. For example, a swing high may be the highest bar of an upswing with the high of the X previous bars and the X bars that come after it being lower. This would be my preference. Hope this helps.
Thank you for the replies so far. I guess I can rephrase/clarify what my need is... Ideally I would like to separate various chart patterns into different groups based on their similarity. My minimal understanding of TA is that much of it is subjective and you can get several people looking at the same chart who all come up with different interpretations. The goal is to take out the subjectivity. This means that ideally the number of bars I would be using would be enough that there can be meaningful separations between them while not so much that most chart patterns are unique. For example, if you only use 2 bars with the end of bar price or even change in price, there are only 3 possible patterns - bar 1 is higher, bar 2 is higher or they are equal. This obviously isn't enough to be meaningful from a predictive standpoint. At the other extreme, if you use 1000 bars, there are so many permutations that each pattern could theoretically be assigned into it's own group which is also meaningless.
Look at Thomas Bulkowskis' work in this area. He has a couple of excellent books, and has written various articles in magazines such as, Technical Analysis of Stocks and Commodities, and recently Stocks Futures and Options. Very in-depth analysis and objective work in my opinion. Good Luck and Good Trading!! Kelly http://www.amazon.com/s/ref=nb_ss_b...rch-alias=stripbooks&field-keywords=bulkowski Look at the first and third books, and check your local library.
I've got the "Trading Classic Chart Patterns" one. It's the best statistical work on the subject I've seen.