Predicting Hundreds of Time Series at Once (preferably with R)

[IlyaKEightSix]

Alright, something I've been looking for a way to do is this: how the heck do you predict hundreds of time series at once? All the stuff I look for regarding time series is about looking at *one* time series at a time. But say you want to predict the next year of temperature (all 365 days) given 30 years of data, or the returns of 100 stocks for the next 10 days given 100 days worth of data, or something like that.

Right now, there's a problem I'm working on that involves getting two weeks of ten minute data usage for smartphone users and predicting the next day, in ten minute increments, I believe. So it isn't like I have the luxury of fitting an initial ARIMA, looking at residuals, fitting another model, rinse, repeat. Is there some way of taking into account the fact that I have multiple time series (even if independent), and making some sort of batch prediction?

 

[diegosanaz]

For batch prediction you can use VAR models (vector autorregresive). Is about the same, but instead of using one time series (one vector) you use more (a matrix).
http://faculty.washington.edu/ezivot/econ584/notes/varModels.pdf

 

[IlyaKEightSix]

Diego, I'm trying to replicate the code in that book and cannot find the simulate.VAR function that Dr. Zivot uses. Do you know which R package it's in?

 

[diegosanaz]

I have not used it much, but the package vars seems up to the task