2
$\begingroup$

I would like to predict stocks of a company for 6 months. I would like to use neural networks for this prediction.

  1. Can anyone suggest how many hidden layers and hidden nodes to be used?

I have modeled a neural network with 3 input nodes and 2 hidden nodes and 1 output node in R but I am getting forecast accuracy of around 35% only.

  1. Can anyone suggest how I can improve the forecast accuracy of my model?
  2. Are there any other models which give better accuracy for this kind of data?
$\endgroup$
6
  • 2
    $\begingroup$ Are you predicting the stock price or the returns on the stock price (this is equivalent but it matters when you measure forecast accuracy)? Or are you predicting stocks of goods in the storage facilities? Or something else? Also, what is your forecast accuracy measure? $\endgroup$ Commented Sep 25, 2015 at 9:42
  • $\begingroup$ i actualy want to predict the stock price of product company but not goods. i am using normal rmse for forecast acuracy,i donnow if it is right.thanks $\endgroup$
    – alily
    Commented Oct 8, 2015 at 10:31
  • $\begingroup$ If your accuracy measure is RMSE and you are predicting a stock price in dollars (or another currency), how can you get a value in percentage terms (35%) rather than dollars? Also, stock price (price of a share of a company traded on a stock exchange) is normally unpredictable. That is, the best prediction is the last observed value (unless it is some exotic exchange). If you get low forecast accuracy, that is natural. It would be strange if you got high forecast accuracy. $\endgroup$ Commented Oct 8, 2015 at 10:40
  • $\begingroup$ Regarding use of neural networks, your network looks quite small (just two hidden nodes). If you want to capture the nonlinearities better you may try including more nodes. You could determine the number of nodes using cross validation. $\endgroup$ Commented Oct 8, 2015 at 10:41
  • $\begingroup$ Thanks richard, i am converting dollars in percentage correctly predicted based on the amount i gain or loose.i will add more nodes and try.Are there any other models which are commonly used for stock price prediction? $\endgroup$
    – alily
    Commented Oct 14, 2015 at 8:04

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.