|
主讲人 |
黄乃静 |
简介 |
<p>How can we learn real-time information about aggregate economic activity from firm-level accounting data? This paper proposes a micro-to-macro nowcasting framework that uses machine learning algorithms to directly exploit accounting information from 21,061 publicly listed U.S. firms to nowcast U.S. aggregate output, thereby preserving the rich information embedded in firm-level heterogeneity and cross-firm interactions that is often lost in aggregated data. The empirical results show that the proposed approach significantly improves nowcasting accuracy, reducing the root mean squared error by more than 70% relative to a random walk benchmark and substantially outperforming models based on aggregated accounting data. Firm-level corporate accounting data also improve nowcasting performance by about 17.35% relative to models using mixed-frequency aggregate macroeconomic and financial predictors. Overall, the results underscore the value of directly exploiting firm-level accounting data for nowcasting aggregate economic activity.</p> |
