Subspace shrinkage in conjugate Bayesian vector autoregressions.

Area of Science:

• Econometrics
• Macroeconomic modeling
• Bayesian statistics

Background:

• Economists often choose between large vector autoregressions (VAR) and factor models for analyzing extensive datasets.
• Both VAR and factor models have limitations when applied to large-scale macroeconomic data.

Purpose of the Study:

• To develop a unified Bayesian vector autoregression (VAR) framework that integrates the strengths of factor models.
• To introduce a subspace shrinkage prior that allows for simultaneous estimation of the number of factors and shrinkage intensity.

Main Methods:

• Development of a conjugate Bayesian VAR model incorporating a subspace shrinkage prior.
• Theoretical analysis of the prior's properties.
• Simulation studies to evaluate factor number detection and forecasting performance.

Main Results:

• The proposed subspace shrinkage prior successfully identifies the number of factors in simulations.
• The Bayesian VAR with subspace shrinkage prior demonstrates improved forecasting accuracy on US macroeconomic data.
• The model allows for flexible estimation of the factor model subspace and shrinkage strength.

Conclusions:

• The conjugate Bayesian VAR with a subspace shrinkage prior offers a powerful and flexible approach for large-scale macroeconomic analysis.
• This integrated method enhances forecasting performance compared to traditional standalone models.
• The approach provides a principled way to combine information from VAR and factor models.