Clustering Dynamics and Persistence for Financial Multivariate Panel Data

We introduce a new method for dynamic clustering of panel data with dynamics cluster location and shape, composition, the number clusters. Whereas current techniques typically result in (economically) too many switches, our results economically more meaningful patterns. It does so by extending standard cross-sectional using shrinkage towards previous means. In this way, different cross-sections are tied together, substantially reducing short-lived switches units between clusters (flickering) birth death incidental, less Monte Carlo simulation, we study how to set penalty parameter data-driven way. A systemic risk surveillance example business model classification global insurance industry illustrates works empirically.