3
$\begingroup$

There are 100 commodities that I am tracking. As I attempt to improve my time series models, VAR and VARMA look like they might provide some improved predictions. All of the literature I find on creating VAR models assume that the modeler has some assumption of which items might correlate. Are there methods/practices in finding the related items systematically without prior knowledge of the commodities?

Toolset (Python): Working with pandas, statsmodels and matplotlib mostly.

Improve this question
$\endgroup$
4
  • $\begingroup$ Since you have a large number of time series, you may consider using regularization techniques to lessen the sensitivity and decrease the estimation variance of your model. Bayesian VAR modelling could be relevant. $\endgroup$
    – Richard Hardy
    Feb 15, 2016 at 15:59
  • $\begingroup$ Have you considered a dynamic factor model? $\endgroup$
    – hejseb
    Feb 15, 2016 at 18:53
  • $\begingroup$ Hadn't considered either of those approaches; I'll read up on both (always appreciate links if you know of any papers/articles/examples that explain this clearer than others) $\endgroup$
    – Crowson
    Feb 15, 2016 at 19:22
  • $\begingroup$ You could also try GVAR modelling, a summary of this method (essentially hierarchical VAR-modelling) is given in the survey paper of Chudik & Pesaran named 'Theory and Practice of GVAR modelling' $\endgroup$
    – Jeremias K
    Feb 15, 2016 at 21:33

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.