Metropolis Sampling in Bilinear Time Series Models

Abstract

Modelling nonlinear relationships plays an important role in recent econometric development. A promising approach to model some of the nonlinearities of a time series is to choose simple extensions of linear time series models such as bilinear models. In this contribution we investigate the bilinear time series model in a Bayesian framework. Inference is made using Markov chain methods like the Gibbs sampler and the Metropolis algorithm. Two versions of the Metropolis algorithm are compared in a numerical study.