BAYESIAN ANALYSIS OF UNRESTRICTED VECTOR AUTOREGRESSIVE MODEL WITH NON-NORMALITY INNOVATIONS

บทคัดย่อ

Assumption of normality in innovations is often used in Vector Autoregressive (VAR) model, but this assumption is very sensitive to outliers. Outliers pose a major challenge to econometricians and practitioners. However, the unconditional distribution using t-distribution has fatter tails than the usual normality assumption. In this work, new efficient Bayesian approach when the innovations of VAR model follow a t-distribution was developed and discussed. The empirical example used a simulated data and data on USA economy by extending the usual normality to assume a t-distribution. The results show that the developed Bayesian approach is good for forecasting while small degree of freedom was capable to diminish the effects of outliers.