2

I am new to Time Series analysis, but I read a lot of questions here which deal with how you can forecast values of a time series based on its history alone.

Now, the data I have is of this kind :

TimeSeries1:

timestamp value
..        ..
..        ..

Now, I have a huge number of other timeseries, which follow the same schema. Also the data is not seasonal. If plotted, the time series will look like a rectangular wave.

What I want to do is forecast values of TimeSeries1, based on the aggregate behavior of all the remaining timeseries data that I have. I had a look at this question, but my problem involves incorporating a large number of other time-series rather than just one like in the above question.

What are the possible approaches to this problem?

Community
  • 1
pnpranavrao
  • 23
  • Have you got a good idea of the underlying behavior of the process that generates your multiple observed time series? Do you know that there is a common process? – GeorgeWilson Jun 01 '13 at 12:12

2 Answers2

1

First you should check if the time series you want to forecast is Granger-Caused by your other variables, if it is, it makes sense to fit a VAR(vector autoregression)-model. In a VAR your dependent variable regress on lagged values of itself and the other variables as well. Might this be what you are looking for?

fredrikhs
  • 1,254
0

I think what you are looking for is an automated process for analysing the available information and selecting the best model out of a list of all possible models. If so, you might want to a software refered to as LDT. I should mension that it (currently) searches a subspace of stationary VAR models.

disclaimer: I am its developer.

rmojab63
  • 216