Dynamic Mixed Frequency Synthesis for Economic Nowcasting

TL;DR

This paper introduces a Bayesian framework for dynamic mixed frequency data modeling, enhancing nowcasting of U.S. GDP by effectively synthesizing diverse economic indicators and capturing their evolving interdependencies.

Contribution

It presents a novel Bayesian approach that models mixed frequency data as latent factors, improving interpretability and nowcast accuracy over existing methods.

Findings

Improved nowcast performance for U.S. GDP

Effective modeling of time-varying interdependencies

Incorporating intra-quarter data enhances accuracy

Abstract

We develop a novel Bayesian framework for dynamic modeling of mixed frequency data to nowcast quarterly U.S. GDP growth. The introduced framework utilizes foundational Bayesian theory and treats data sampled at different frequencies as latent factors that are later synthesized, allowing flexible methodological specifications based on interests and utility. Time-varying inter-dependencies between the mixed frequency data are learnt and effectively mapped onto easily interpretable parameters. A macroeconomic study of nowcasting quarterly U.S. GDP growth using a number of monthly economic variables demonstrates improvements in terms of nowcast performance and interpretability compared to the standard in the literature. The study further shows that incorporating information during a quarter markedly improves the performance in terms of both point and density nowcasts.