Sensitivity of Priors in the Presence of Collinearity in Vector Autoregressive Model: A Monte Carlo Study

Abstract

The use of prior distribution plays a great role in Bayesian approach. This work examines the sensitivity of priors in Bayesian Vector Autoregressive (BVAR) model when the time series data are correlated. The three scenarios of collinearity namely; Low Collinearity (LC), Moderate Collinearity (MC) and High Collinearity (HC) were considered while the forecast evaluation criteria were used to judge the performance of these BVAR priors. Results from Monte Carlo experiment showed that all the forecast criteria evaluation for all the scenarios of collinearity have the same pattern of performance. However, Normal-Wishart prior performed best in the three scenarios of collinearity especially in small sample sizes.