Clustering macroeconomic variables

Numerous studies have highlighted the structural instability in certain macroeconomic time series. This issue has been typically addressed through three econometric methodologies: structural breaks, Regime-Switching, and time-varying parameter models, all requiring some ex ante structure to define the changes. Drawing on the recurrent Chinese restaurant process, a model for an autoregressive process is introduced and estimated via a particle filter. This methodology is employed to study the instability in post World War II US inflation. The application displays a good fit to the data, producing a clusterization of the time series that can be interpreted in terms of economic history, given a relative small number of estimated clusters. In addition, it is able to recover key data features without making restrictive assumptions, as in the case of one-break or time-varying parameter models.