Automated Chart Pattern Recognition

Discussion in 'Trading Software' started by hprop, Jun 26, 2008.

  1. hprop


    I'm looking around for Chart Pattern Recognition Software that can be automated to create and export a report (preferably with charts) every day after going over EOD-data.

    Nirvana's OmniTrader has a Chart Pattern module that looks good - they've backtested chart patterns and ranked profitability and reliability and allow you to select a threshold for what signals to receive. I'm however uncertain if it's possible to export charts with the patterns marked - they do have some scripting/programming support.

    Autochartist is a subscription service which seems to deliver it but not really offer that much customization or control. Service does look good (and cheap) however.

    AIQ Systems has a plugin which I've read somewhere on the net (unreputable, I know) basicly is Autochartist added to the program. Unfortunately AIQ datafeeds doesn't have the market I'm looking at for this.

    PatternExplorer for Amibroker is a plugin that can be bought separately. It's got some nice features, and of course Amibroker is open ended enough to allow for report generation.

    Equis Metastock also has a plugin for this.

    Anyone care to expand or comment on what I've written, or share their own experience with this type of software?
  2. Autochartist is good but minimal user preferences and flexibility.

    I submitted a thread for info on visualtrader under software. Not much content to my questions on the website, looking for more feedback. It''s pattern recognition claims to be first rate.
  3. hprop


    I think VisualTrader and OmniTrader shares the same pattern recognition module, only that it's an addon to OT.
  4. trade ideas
  5. For some reason or other most pattern recognition is reactionary, meaning it is a history report that follows a past opportunity.

    The patterns which are anticipatory for entry to making money (CW approaches and otherwise) are not tabulated on pattern recognition analytical type software. It is also true that the related exit patterns for such entries are not correlated nor free standing

    Finally, there are those cases where a "hold through in profit making" on about 13 types of patterns (in two distinct and non over lapping subsets) that occur on the next faster fractal as compared to the trading fractal. thses patterns are the "causal" patterns of most lesser skilled trading of egdge type traders. they also account (into two cases) for scaling in and scaling out as a consequence of high eprcieved but not actual risk.

    It will probably take a while for the pattern detecting people to move from their historical raction orientation to a more profitiable anticipatory orientation.

    there are four big areas to consider and aside from non CW ATS traders, the patern anticipation ground remains unbrocken and quite frustrating.
  6. monti1a


  7. neoticker
  8. frankl


    Try out VectorBull. It is specialized in pattern scanning.
  9. You might also consider Patternz By Bulkowski (free):

    Tymora also scans for some patterns

    Tradestation 8 apperars not to have been mentioned - maybe because it will not scan the whole market (TS2000i can with the right data feed).

    Andrew Lo at MIT has a paper on pattern recognition algos.

    The impression I get is that off the self scanning programs or plug ins are generally dissapointing. This is my experience with eg Tradestation or Metastock pattern scanner code written by others. I write my own as I also look for non textbook patterns. Off the shelf scanners fail to select by other important factors such as position of pattern in the context of historical price/volume action - with some patterns what is a buy when occurring at at one level could be meaningless or a sell signal at another. The context in which the pattern occurs is important.

    My experience with writing pattern finding algos is that you can be overprecise regarding parameters and you really need to break a pattern down into its essential elements/building blocks* in the context of historical price action and be prepared to look at a lot of duds rather than try and code for a complete text book pattern. I also screen in other ways before I code for the patterns and add qualifying factors that are not part of textbook descriptions. You pretty much have to know WHY a pattern occurs IMHO- hence maybe the failing by third party programmers code.

    Coding for a pattern is a quantum leap from recognising a pattern when you see it, and coding for a pattern you don't understand based on text book paramerters is barking in the wrong forest eg coding a flag based on height of pole degrees slope and length has validity as there is a reason why these parameters can be valid, but is only part of the story, this pattern occurrs for a reason and works/fails for a reason - not fully captured by simple dimensions/volume pattern.

    IMHO watching/screening the tape and then looking at the chart for confirmation is far superior - I use scans primarily for markets that trade in my overnight or trade too many stocks to watch the tape and as a backstop/safety net for markets I can watch.

    *(you would be surprised how effective the seemingly overly simple and non customiseable scans in IB TWS can be if you use the screening properly - and it scans the whole market - a rare feature. With the addition of a bit more ability to screen and also combine 2 or more parameters this would be even more useful)
    #10     Jun 28, 2008