Forecasting the Japanese macroeconomy using high-dimensional data

This paper compares several forecasting methods using high-dimensional macroeconomic data from Japan. The diffusion index (DI) model has been widely used to incorporate the information contained in high-dimensional data for forecasting. We propose two selection methods of the number of latent factors in the DI model and compare the DI model with the vector autoregression (VAR) model whose parameters are estimated by lasso-type methods. We find that the DI model tends to be better for short-horizon forecasting, whereas the VAR model tends to be better for long-horizon forecasting. Moreover, we find that the information exploited for forecasting is similar between the DI and VAR models.