Title: Nowcasting Japanese GDP using text data and machine learning
Authors: Mototsugu Shintani - University of Tokyo (Japan) [presenting]
Abstract: A nowcasting analysis of Japanese GDP using text data and machine learning is conducted. We employ a machine learning approach because the estimation of mixed-data sampling (MIDAS) models without parameter restriction, as well as models with text information, involves high dimensional data. Based on the unrestricted MIDAS model with 15 monthly hard macroeconomic indicators and newspaper articles, we find that the model with news data outperforms the model using only hard data, especially during the period of COVID-19 crisis.