I have a dataset with 10 columns and 34 rows. Basically, the dataset contains unique visitors for the publisher websites. I'm trying to forecast numbers for a Network under which several brands are included. The numbers for each brand is correlated with the Network level numbers. How can I forecast the numbers using Multivariate ARIMA in R? Is there a specific package?

  • 1
    $\begingroup$ "dse" package in R allows fitting VARMA models (VARMA stands for vector ARIMA, i.e. multivariate ARIMA) . You would have to choose a severely restricted version of VARMA, though, as an unrestricted VARMA model will have a very high number of parameters (on the order of 100 to 1000) as compared to your sample size. If you drop the MA part and think about VAR, even there the curse of dimensionality will be a big concern; a VAR(1) will have 100 parameters. You should look for something simpler, IMHO, such as univariate exponential smoothing, low-order univariate ARIMA or the like. $\endgroup$ – Richard Hardy May 9 '17 at 14:49
  • $\begingroup$ Or perhaps forecast the overall trend with a univariate model (exponential smoothing, low-order ARIMA, etc.) and then forecast the deviations of individual series from the common trend with univariate models again. $\endgroup$ – Richard Hardy May 9 '17 at 14:51

Your Answer

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

Browse other questions tagged or ask your own question.