This repo is an attempt to use the code of Lee, Kim, and Gupta (2020) to learn Bayesian synthetic control. I attempt to use their model to duplicate the results from Scott Cunningham's Mixtape. You can find the code for Lee, Kim, and Gupta's (2020) in their supplementary materials.

A second Stan script uses the code of Piironen and Vehtari (2017), which is also used by brms. This version runs without transitions, unlike the code of Lee, Kim, and Gupta (2020).

At this point I have not changed the code of either Cunningham or Lee, Kim, and Gupta, though I have modified the code of Piironen and Vehtari. You can find a modified version of their code in bscm_horseshoe_modified.stan.

A third approach is included in sc_spike_slab.Rmd, which is a spike and slab formulation for synthetic controls.

This is a work in progress, and contributions are welcome. In particular, I am looking for convenient ways of using auxillary variables to help select the weights.