Dynamic model for a first order autoregression process with Bayesian methodology

Abstract

A ramification of a first order autoregression process is provided. It comprises randomized and variant coefficients in time and assumes a structure of dependency of randomized coefficients that leads towards adapted Kalman's Filter. Although the Kalman Filter model is a generalization of the ordinary Kalman Filter, its analysis produces technical difficulties. It does not seem to be impossible to find a closed form for the filter. Monte Carlo's simulation was applied to Markov's Chain by Gibbs-Sampling and Metropolis-Hasting algorithms to infer parameters of model and work out forecasts of data for a time series of indexes of shares and meat prices.