Forecasting macroeconomic variables using data of different periodicities

Abstract

A formal statistical method is used in this study to combine forecasts from a quarterly macroeconometric model for Taiwan with monthly time series forecasts. Three monthly models, i.e. vector autoregressive (VAR), Bayesian vector autoregressive (BVAR) and Autoregressive integrated moving average (ARIMA) were alternately applied to examine whether a superior monthly model can achieve better quarterly forecasts. For variables that are observed both quarterly and monthly, combined forecasts are generally found to be superior to the macro forecasts but inferior to the monthly ones. With respect to variables that are available only quarterly, results in this study indicate that the gain in forecasting accuracy due to the inclusion of the monthly data is substantial even when no monthly information is available for the quarter.