Forecasting GDP based on the Mixed Frequency Data Model

One of the most attractive data releases by the Japanese statistical system is the quarterly real gross domestic product (GDP). Official publication of this statistic, however, comes at a significant delay of up to one and half months. This study aims to demonstrate the nowcasting to predict of the present, the very near future, and the very recent past based on the information provided monthly and quarterly. The dimension reduction by the PCA (principal component analysis) method is conducted to reduce 19 economic variables into 10 PCA factors. We apply the MIDAS (Mixed Data Sampling) regression to deal with mixed frequencies, and the ARDL model as the benchmark. The result indicates the MIDAS framework, using higher frequency economic indicators, performs better than the sample time average method of the ARDL model in predicting the GDP