Abstract: Using quarterly Taiwan economic data, we demonstrate that deeper understanding of relations between variables and substantial gains in forecasting can be obtained by applying econometric and statistical tools to the traditional macro-econometric models. The improvement in forecasting accuracy is illustrated by out-of-sample forecasts, and the models employed in the comparison include univariate time series models, macro-econometric models and combined models in which time series techniques are used to describe the dynamic structure of the residual series of econometric models. The paper also considers various issues related to forecasting such as aggregation and model misspecification.
Key words and phrases: ARIMA, out-of-sample forecast, outlier, residual dynamic, structural model, Taiwan macro-econometric time series model.