1
$\begingroup$

I am trying to build sparse coefficient time series models but couldn't find a good resource/book to learn from. Can anyone share with me a good resource/book? It will be great if it comes with R codes.Can anyone use this kind model? Thanks.

$\endgroup$
  • 1
    $\begingroup$ Could you be a little more explicit as to what you mean by "sparse" in this context and how this differs from the Box-Jenkins approach? $\endgroup$ – Stephan Kolassa Nov 21 '12 at 10:06
  • $\begingroup$ For example:if we build an ARMA(p,q) model,generally,we need estimate p+q coefficients.The sparse means some of the p+q coefficients may equal to zero.So the model's coefficients are sparse.Maybe I don't describe clearly or in time-series these models don't call sparse coefficient models.If like this what they named? Thanks. $\endgroup$ – XingHua.Ma Nov 21 '12 at 12:49
1
$\begingroup$

Are you familiar with the Lasso, Elastic Net and similar regularization techniques? The Lasso specifically is used in regression with many regressors to yield a result where "many" cofficients are zero, which looks like what you are aiming for.

It looks like there is some interest out there in applying the Lasso and similar concepts to ARMA models. See, e.g., here or here or here.

I would recommend that you look into regularization and google for terms like "regularization technique time series technique".

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ I'd love to see an answer to the question posted at your first link... (stats.stackexchange.com/questions/10425/…) $\endgroup$ – Zach Nov 21 '12 at 16:14
  • 1
    $\begingroup$ @Zach: you and me both, it would make a lot of sense... But the paper behind my third link notes that simulations were done in R using lars and glmnet, so it might be worthwhile to mail the authors and ask for their code. $\endgroup$ – Stephan Kolassa Nov 21 '12 at 19:37

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.