0

I have a set of $n$ objects where an object, $n_x$, consists of two time series lists of numbers. ${n_1, n_2, n_3,...,n_n}$ where $n_x={2,3,4,1,3,5,34,...},{4323,23,42,34,1,1,1,...}$.

For each $n_x$ object, time series one has been shown to be granger causal at some lag for time series two.

The lag that is most predictive for each object is not necessarily the same.

I am a complete novice to mathematical modeling, however in my lab I sadly am the person with the most mathematical background. For this problem, what type of model is appropriate for using time series one values to predict future values of time series two? In my mind, I would like to have a training subset of the objects, which fits a model, and then the model is tested for its ability to predict the values for the other objects.

This is not a homework question, I work in a biology lab as a graduate student.

2012ssohn
  • 3,827
Justin Gardin
  • 1
  • It sounds like you have in mind fitting model parameters separately for individual objects. If that's so, the Question of how to use "time series one values to predict future values of time series two" begs the issue of how to use the time series to determine the "lag that is most predictive". – hardmath Jan 25 '14 at 00:51
  • I want to use a single model fitting to best predict untrained data. – Justin Gardin Jan 25 '14 at 20:20
  • As an update, I am attempting to use the Matlab Neural Network modelling application (NARX version), something simpler would be desirable as an alternative. – Justin Gardin Jan 26 '14 at 00:23
  • Does the phrase "machine learning" fit with your project? I know a little about parameter fitting and time series (really little), so I thought I'd mention the sister site CrossValidated as possibly better at this sort of question. – hardmath Jan 26 '14 at 02:20

0 Answers0