I have been using BSTS package for quite a while and I have found it pretty effective with respect to ARIMAX models.

I was wondering whether it would be possible to share regression coefficients in the regression part (the equation being: $Y_t = \mu_t + \mathbf{x_t} \boldsymbol{\beta} + S_t + \epsilon_t$, with $\mu_t$ being the level at time $t$, $S_t$ being the seasonal component at time $t$) across different time series.

Furthermore I would like to have also mixed effects so that my new equation would become something like:

$Y_{it} = \mu_t + \mathbf{x_t} \boldsymbol{\beta} + \mathbf{Z_t} \boldsymbol{\eta_i}+ S_t + \epsilon_t$.

Question: Is this achievable through BSTS?

Having read Steven Scott presentation I see he implemented Lasso Variable selection and other stuff which makes it difficult for me to implement it directly in STAN..


It looks like version 0.7.0 added support for this model.

As of version 0.7.0, bsts supports having multiple observations at the same time point. In this case the basic model is taken to be

$y_{t, j} = Z^T\alpha_t + \beta^Tx_{t, j} + \epsilon_{t, j}$

This is a regression model where all observations with the same time point share a common time series effect.

  • $\begingroup$ Thank you for the help ! Unfortunately I see no reference to mixed effects $\endgroup$ – Tommaso Guerrini Jul 14 '17 at 7:22

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.