The extended Hodrick-Prescott (HP) filter for spatial regression smoothing

The extended Hodrick-Prescott (HP) method was developed by Polasek (2011) for a class of data smoother based on second order smoothness. This paper develops a new extended HP smoothing model that can be applied for spatial smoothing problems. In Bayesian smoothing we need a linear regression model with a strong prior based on differencing matrices for the smoothness parameter and a weak prior for the regression part. We define a Bayesian spatial smoothing model with neighbors for each observation and we define a smoothness prior similar to the HP filter in time series. This opens a new approach to model-based smoothers for time series and spatial models based on MCMC. We apply it to the NUTS-2 regions of the European Union for regional GDP and GDP per capita, where the fixed effects are removed by an extended HP smoothing model.