Inflation Forecast Combination: Evidence from Taiwan

Abstract We study the benefit of considering the forecast combination over the individual forecasts generated from the single models in predicting the inflation. In particular, we contribute to the current studies since we rarely find, to the best of our knowledge, the published papers that study the performance of the forecast combination in predicting Taiwan’s inflation, at least from both the Bayesian and the forecast stability perspectives. The models that we adopt to generate the individual forecasts include the random-walk (RW) model, Phillips curve-based (PC) model, the vector autoregression and the unobserved component model with stochastic volatility, and the predictive abilities of those (individual) combination forecasts are evaluated relatively in terms of either the full-sample or the rolling-sample perspectives. Several results are found in this paper. First, a clear winner is not found in the inflation forecast competition among the individual models. However, the forecasting performance of both RW and PC models are significantly worse than the other models in the one-step-ahead and four-step-ahead forecast competition respectively. Second, simply averaging the individual forecasts of inflation provides the best way to obtain the inflation forecast combination. Third, we find the benefit of the inflation forecast combination over the individual forecasts. Lastly, the forecast stability of the forecast combination in predicting Taiwan’s inflation is found in the paper.