On forecasting the Hong Kong economy with Bayesian vector autoregression model

Abstract

This thesis evaluates various specifications of Bayesian Vector Autoregression (BVAR) and their implication on forecast accuracy with respect to Hong Kong data. In particular, I construct a Hierarchical Bayes model within the conventional BVAR framework using the Minnesota Prior. The design aims at bypassing the restriction of cross-equation independence of the coefficients and arbitrary fine- tuning of the hyperparameters in traditional BVARs. Estimation is facilitated by Markov Chain Monte Carlo methods because of the lack of analytical expressions for the solutions. The reduced form model so estimated is then "identified" into a structural counterpart that can offer meaningful economic interpretations. Empirical results from these exercises are consistent with the currency board system in general with monetary variables respondin g endogenously to shocks that hit the Hong Kong economy.

Date of Award	1998
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology