Bayesian inference in a multiple contaminated autoregressive model with trend

Abstract

ABSTRACT Two types of outliers that may occur in data are considered in this paper: additive outliers (AO) and innovational outliers (IO). We have generalized the two types of contaminations AO and IO to the multiple case for an autoregressive model of order p with a regression trend. We adopt the Bayesian approach combined with Gibbs sampling to jointly estimate the model parameters and the outliers on the first hand, and on the other hand we use a test based on p-values and other discrimination Bayesian criteria to detect the location and the magnitude of the two types of outliers. An intensive simulation study is presented for illustrating the performance of the method relative to maximum likelihood estimation, mainly for small sample sizes. Our method is applied to a real data set.