0
$\begingroup$

How are neural networks usually used to predict market evolution? My data consists of a set of pairs (time, value), taken at an interval of 15 minutes.

My ideas so far are:

I.Take 40 values (or another arbitrary number) as inputs and try to estimate the following 50 values(again, arbitrary). This has the problem that once the neural network is generated, I can only produce 50 estimations.

II.Take time as an input and the value as an output. This is more slower but can create more estimations.

I would love any answers or references to articles about the subject.

$\endgroup$
  • $\begingroup$ What do you mean by once the neural network is generated, I can only produce 50 estimations? Also, I am skeptical about the idea that time could be driving stock prices (unless time is a proxy for the cumulative effect of inflation or the like, but that should be totally negligible when you have data recorded as frequently as at 15 minutes intervals). Also, I suppose you are talking about stock prices (which are observable) rather than stock values (which are unobservable). $\endgroup$ – Richard Hardy May 13 '15 at 18:03
1
$\begingroup$

I would suggest some memory based recurrent neural network. Google created a so-called Neural Turing Machine, that is able to predict long sequences of data: http://arxiv.org/pdf/1410.5401v2.pdf

But I have a feeling that it's a bit hard to implement ;)

$\endgroup$
0
$\begingroup$

Probably you need read about Neural Network Autoregression. Check it at Google search, you will find a lot of publications and libraries.

For example: http://forecasters.org/wp/wp-content/uploads/Thielbar_Dickey_IIF_Submission-2.pdf

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.