THE USE OF HIGH FREQUENCY DATA TO IMPROVE MACROECONOMETRIC FORECAST

Abstract

Monthly measurements generally provide valuable information for future economic movements. This study demonstrates how the high frequency data, through a subset of variables in the monthly model, can be pooled in a systematic way via the quarterly econometric model as well as improve the forecasting accuracy. Three monthly models, VAR, BVAR, and ARIMA are used to capture the monthly information. Single-and tow-quarter-ahead forecasts are combined with the monthly data. Results obtained by using the modified Taiwan government quarterly model indicate the potential for significant reductions in root mean squared errors over the one-quarter-ahead forecasts. However, the gain appears to be of less relevance for the longer-term forecast horizon.