Very cool! Looking forward to checking this out.

Thanks, Katie!

Many thanks to my wonderful coauthors: @jonathanpark.bsky.social, Sy-Miin Chow, Anja Ernst, Vladas Pipiras, and @fishingwithzack.bsky.social.

We conducted a simulation study to evaluate the performance of subgrouping multi-VAR across a range of factors and found that it performs well with respect to model/subgroup recovery and quality of estimates. Performance was also assessed in the context of an empirical example.

What's nice about this?

The competing penalty terms govern both the sparsity and heterogeneity of the solution. This allows for the flexible approximation of commonly encountered scenarios (e.g., minimal heterogeneity, subgroup-specific heterogeneity, etc.).

Subgrouping multi-VAR accommodates both quantitative and qualitative heterogeneity in the estimation of group-, subgroup-, and individual-level effects.

Notably, these dynamics are estimated simultaneously using a structured regularization approach.