Indicators: Next price predictor using Neural Network

New comment
 

Next price predictor using Neural Network:

An indicator that uses a Feed Forward Neural Network to predict the next few open prices. The network is trained using a Backpropagation method. The training is set up automatically, resulting in a self-trained network and a self-learning indicator.

Author: Vladimir

 

Hi!

Excelente thread!

Is possible to do the same thing, but with de values of indicators? For example, RSI, Stochastic or Williams' Percent?

I think that, these(RSI, Stochastica, etc) indicators can show the "future" direction of the prices... then.. if we use the same idea of your indicator, but, with the prices of the indicators.. then, we will have more chance for to know the future direction...

What you think about it?

 

seems too good but how to trade

 

it looks sophisticated, but aytime i apply it to chart, it points in different direction (random), so there is no consistent forecast. but it definitely looks sophisticated, so thanx :)

 

Hi Gpwr

Have you tried kohonen map(2D SOM) as a classifier and/ or as a dimension reducer for indicators?Thanks.

 
geordie:

Hi Gpwr

Have you tried kohonen map(2D SOM) as a classifier and/ or as a dimension reducer for indicators?Thanks.

No, not yet. That would be my next step. Any opinions about Kohonen maps for forex before I dive into them?

 

Good to see this excellent price predictor. Please gpwr can you tell us the exact timeframe to use for this indicator and moreso what currency pair do you think is most suitable. Apart from the 12-5-1 network layer value to be chosed, is their any other important value needs to be change by trial and error. I have been working on it and its great but still want to see more testing.

Thanks

yusnas

 
Yusnas:

Good to see this excellent price predictor. Please gpwr can you tell us the exact timeframe to use for this indicator and moreso what currency pair do you think is most suitable. Apart from the 12-5-1 network layer value to be chosed, is their any other important value needs to be change by trial and error. I have been working on it and its great but still want to see more testing.

Thanks

yusnas

You may try to increase nep to 10,000 and, if your computer is fast, to 100,000. It will give you a lower training error and more consistent results. I have not tried other currencies or timeframes.

 

Sorry if this is a very "noob" question, but could the neural net's predicted price be rendered inacurate if there is a news release that deviates from what the market expected? could analysing past, predicted, and actual fundemental news releases increase the accurate of predictions? I think adding EMA smoothing was a great move, but would incorporating something like these:

http://codebase.mql4.com/4543
http://articles.mql4.com/523

or trend/pattern indicators like these:

http://codebase.mql4.com/5161
http://articles.mql4.com/729
http://forum.mql4.com/14855


help increase accuracy in any way? I'm sorry if this is a stupid question, I'm still learning MQL4 and Forex ;p

Thanks,

Ian

 
macleod77:

Sorry if this is a very "noob" question, but could the neural net's predicted price be rendered inacurate if there is a news release that deviates from what the market expected? could analysing past, predicted, and actual fundemental news releases increase the accurate of predictions? I think adding EMA smoothing was a great move, but would incorporating something like these:

http://codebase.mql4.com/4543
http://articles.mql4.com/523

or trend/pattern indicators like these:

http://codebase.mql4.com/5161
http://articles.mql4.com/729
http://forum.mql4.com/14855


help increase accuracy in any way? I'm sorry if this is a stupid question, I'm still learning MQL4 and Forex ;p

Thanks,

Ian


Thanks. I will look at those indicators.

 

Thanks for reply gpwr. I tried using 100,000 for nep and it freezed my computer at times. for the input I used 13-7-3 it somehow work well but I am looking at the right timeframe that will not been giving me flip flop results. you can try this code (13-7-3) in your system and the timeframe you ve been using before and compare the result with default 12-5-1. You will see that there is some better accuracy in the former.


Thanks

123456
New comment
Reason: