Time-Frequency Analysis of Dynamic Financial Connectedness: A New Approach With Bayesian Time-Varying Vector Autoregressions

Abstract

The global financial crisis has triggered growing interest in quantitatively analyzing connectedness between financial markets. This paper proposes a general framework for measuring financial connectedness that allows for heterogeneity in both the frequency and time domains and incorporates prevalent measures of financial connectedness. We illustrate the proposed method by establishing the connectedness among four major U.S. financial markets. Empirical results suggest that the conventional time-frequency connectedness lack robustness and can have unjustified changes, while our new connectedness measure is more robust and evolves in a more practical manner. Moreover, our new approach can provide certain “early warning” for the emergence of underlying financial crises in the financial system, which cannot be achieved by the conventional time-frequency connectedness. At last, both our new approach and conventional approach identify medium-term, rather than short-term and long-term spillovers as the main spillover channel for systemic risks.

Keywords: Financial connectedness, frequency domain, Bayesian vector autoregression with time-varying hierarchical structure

JEL Classification: c32