|
主讲人 |
王法 |
简介 |
<p>In mixed-frequency data sets, observed low-frequency series are generated from latent high-frequency processes through temporal aggregation that depend on whether the underlying variables are stocks or flows. This paper develops a high-dimensional mixed-frequency factor model that explicitly incorporates the temporal aggregation relationship within a unified factor structure. We propose a mixed-frequency EM estimator and establish the consistency, convergence rates, and limiting distributions for the estimated factors and loadings. Monte Carlo simulations show that the proposed estimator accurately recovers latent high-frequency factors and common components, and substantially outperforms methods that treat mixed-frequency observations merely as missing data across different error structures and aggregation schemes. An application to U.S. macroeconomic data constructs a monthly measure of U.S. GDP growth and demonstrates the usefulness of the proposed procedure for nowcasting quarterly U.S. GDP growth. </p> |
