Large Bayesian VARs for Binary and Censored Variables

2025-06-02

AI Summary

This paper addresses the limitation of traditional vector autoregressive (VAR) models in jointly modeling binary, censored, and continuous variables. We propose the first scalable Bayesian VAR framework capable of systematically integrating all three variable types into a dynamic joint model. Methodologically, we combine data augmentation, Gibbs sampling, and sparse prior–driven shrinkage of high-dimensional covariance matrices to enable efficient Bayesian inference. Our key contribution lies in relaxing the conventional requirement of continuous-valued variables, thereby enabling unified modeling and structural identification of heterogeneous variables—such as binary policy indicators, censored survey data, and continuous macroeconomic series—within high-dimensional macroeconomic systems. Empirical results demonstrate substantial improvements in out-of-sample forecasting accuracy for recession probabilities and short-term interest rates. Moreover, the framework successfully quantifies the time-varying impact of financial shocks on recession risk, offering novel insights for macroeconomic and financial policy analysis.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Bayesian Learning
• Intelligent Robots: State Estimation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Psychology-informed user models and recommender systems

Abstract

We extend the standard VAR to jointly model the dynamics of binary, censored and continuous variables, and develop an efficient estimation approach that scales well to high-dimensional settings. In an out-of-sample forecasting exercise, we show that the proposed VARs forecast recessions and short-term interest rates well. We demonstrate the utility of the proposed framework using a wide rage of empirical applications, including conditional forecasting and a structural analysis that examines the dynamic effects of a financial shock on recession probabilities.

Problem

• Extend VAR to model binary, censored, continuous variables jointly
• Develop efficient estimation for high-dimensional settings
• Forecast recessions and interest rates accurately

Innovation

• Extends VAR for binary, censored, continuous variables
• Efficient estimation for high-dimensional settings
• Forecasts recessions, interest rates accurately