|
Abstract:
|
We develop a spatiotemporal autoregressive (STAR) model of the unemployment rates of 148 European territorial units (NUTS) from 1983 to 2006. We build on previous STAR specifications (e.g., Pace et al (2000)) by modeling the effects of geographic boundaries---country, language and currency borders---on the covariance structure. Specifically, we parameterize the spatiotemporal adjacency matrix into between- and within-boundary parts with separate autoregressive terms. Compared to a STAR model without boundary effects, our model significantly improves fit for areal units lying along a border. For the European unemployment data, the boundary effects diminish substantially following two major events: the Schengen agreement (1995 to 2001) and adoption of the Euro (1999 to 2002). We introduce sparse filtering methods to speed computation as part of an R package implementing our model.
|
