Trade analyzer/neural network: does this exist?

Hi everyone,

Im looking for a solution to something that has been bothering me for a while now.

I use Forex Tester to backtest my manual trading strategy , but I am looking for something that can analyze the data chart itself, and find similarities among multiple trade signals/periods of time that I can "highlight" on the chart (across multiple time frames as well), and compute a result leading to it telling me a "signal alert", based on the indicators I have specified and their values..


Ex. Im looking at EURUSD on the 1DAY TF. I have 4 indicators to manually view the chart. A trade movement occurs. After the fact (days later), I can see my indicators lined up for it, but I want to "highlight" the region and have a program look at it (all timeframes and indicators in use) and tell me the next time a setup like this occurs.



Its more difficult than a simple indicator email alert thing in MT4. This may require some Neuroshell program, or ???? It seems Neuroshell can do similar things but I am not totally sure.

Thanks for any advice.

Neural networks exist, trends exist so you should be able to use a neural network to solve your problem.....

But its not necessarily easy. You must have a signal in mind to help you determine your trend. What is that signal, well that would be the configuration of indicators your using indicating some condition exists.

So your neural network would look back over historic data and analyze your information and generate coefficients which would help you make an informed decision.

Because of the nature of neural networks you want volumes of signals not just a few so you might want to consider using a neural network on short time frames 1m, 5m, 15m. You need hundreds or more occurrences to train your network. Do you get many signals on the daily? If not then daily won't work too well.

How I would do it is export data from lets say MT4 and put it into excel. Then get a program like MatLab which has the capability to analyze a neural network relatively easily. In that case you won't have to get into programming.

I'm sorry in advance if this reply is light on specific information. But you asked a very open ended question, which is fine but you need to understand you'll have to do a fair bit of homework to get this up and running as well as educate yourself on what a neural network really is and how it works. If you want to do that you can watch on iTunesU a class from Caltech which digs into machine learning.

Finally if you don't want to break your back mathematically you could elect to just use some statistical analysis and just preform some linear regression on your data, which might not be the best but it's a start and can be done in excel.

vladrath,

Thank you for your post, it was very thoughtful and led me to some more options to consider.

The signals I am looking for are on 4HR primarily, with "sub-signals" or confirmations many times on the 1hr and 1DAY. The signals are pretty dependable looking back at backtested data from 2000 (farthest I can go with Forex Tester). Signals show depending on a few factors like "price touching of a whole price number eg 1.5000, 1.5100, 1.5200, etc) and "how long a specific indicator has been positive vs negative". My hope is that the neural network, or even Matlab, can discover patterns from them in various timeframes or even across different currencies and then develop some sort of "probability of good trade" so I can adjust lot size accordingly.

Is there a way to make a neural network like Neuroshell's advertised functions without putting up $4G for it? Ive seen FANN and the other "googleable" info on creating your own NN but it doesnt seem straightforward. I do have heavy experience in networking, IT security, and some in database administration, but am still learning programming.

Last time I looked up NN stuff was a year or so ago and now I'm ready to apply some strategies to it...just looking for the best method on doing this.