A1175
Title: Achieving shrinkage in the time-varying parameter models framework
Authors: Angela Bitto - WU Wien (Austria) [presenting]
Sylvia Fruehwirth-Schnatter - WU Vienna University of Economics and Business (Austria)
Abstract: We investigate shrinkage for time-varying parameter models based on the normal-gamma prior which has already been introduced for standard regression models. Our approach extends previous work in which the Bayesian Lasso prior has been considered. The Bayesian Lasso is a special case of the normal-gamma prior. We show how the normal-gamma prior can easily be extended to the time-varying parameter models and focus on inducing shrinkage on the square root of the variance of the prior of the error term in the non-centered state equation. We present both a univariate and a multivariate application. First we choose EU area inflation modelling based on the generalized Phillips curve, then we draw our attention to a multivariate time series with a time-varying covariance matrix and analyse DAX-30 data. Our findings suggest, that the normal-gamma prior bears advantages over the Bayesian Lasso prior in terms of statistical efficiency and performs significantly better when drawing attention to the predictive performance.
