A VAR Framework for Forecasting Hong Kong'S Output and Inflation

This paper develops a multivariate time series model to forecast output growth and inflation in the Hong Kong economy. We illustrate the steps involved in designing and building a vector autoregression (VAR) forecasting model, and consider three types of VAR models, including unrestricted, Bayesian and conditional VARs. Our findings suggest that the Bayesian VAR framework incorporating external influences provide a useful tool to produce more accurate forecasts relative to the unrestricted VARs and univariate time series models, and conditional forecasts have the potential to further improve upon the Bayesian models. In particular, a six-variable Bayesian VAR including domestic output, domestic inflation, domestic investment, world GDP, the best lending rate, and import prices appears to generate good out-of-sample forecasts results.