Using Neural Networks to Identify Chart Patterns

Does anyone have a good article on what 'inputs' are best used when trying to train Neural Networks to identify specific chart patterns? I figure that each pattern would require its own well trained network ... but I can't figure out whether or not each network would also need its own unique set of inputs.

Anyway, I was just wondering if anyone had some whitepapers on the subject.

Thanks,
Corey

 

Not even a tipsicle?

 

Isn't this funny I was in the San Diego State library finishing up my EE homework and started looking down the finance isle and so like five books on neural networks and how to incorporate them. Read a little bit of it and thought it was very interesting.

 

There's a book called Predictors that talks about practical examples. Although, I don't recall if they went in to any specifics about the structure of the neural networks they used.

It was more of a non-fiction biography of a team from santa fe that implemented them.

You can also play with a software tool called patternz, it was partially created by thomas bullkowski, the writer of the pattern encyclopedias.

I didn't find it very useful, and it doesn't give the code detail. But, it gives you an idea about how one might be implemented.

 

A collection of dissertations in the field of technical analysis and neural networks.

Some of them discuss which inputs should be used.


Good luck!

/Jossan

 

I greatly appreciate the info guys -- but what I am really looking for is not Neural Networks as applied to price prediction, but rather binary classification of patterns. But those white papers on optimal trader were definitely interesting nonetheless.

Thanks!
-C

 

Just to add extra clarification, I want to say that I don't mean Neural Networks which use technical analysis measures like trend and MACD as predictors (there are plenty of papers on those). I want articles on actually identifying chart patterns.

THanks!

 

Hi Corey,

I am also interested in this subject. However, I am not sure if it is worth, as we will have 2 steps to identify a pattern first, then apply the patterns to price prediction. Errors after 2 steps may make it not worthwhile anymore.

 

Check out a paper by Andrew Lo at MIT (?)

 

I believe that is what the program, patternz does. I'm not positive it is AI based, but using the program, I would assume it is.

It isolates various patterns, flag, etc.. and outputs the captured patterns and comments on their meaning and resolution.