Correlation meets causality: A holistic measure of financial contagion

Abstract

This study introduces a new measure of financial contagion. We argue that a rapid increase in correlations between two series is necessary but not sufficient for contagion to occur, and develop a contagion test that combines dynamic conditional correlations with time-varying Granger causality. We empirically illustrate our new approach using systemic risk data covering the period 1996–2023. We show that there are periods when correlations increase rapidly without causality, as well as periods when causality is present but correlations do not increase. The proposed test enables data-driven detection of contagion episodes and provides a clear distinction between interconnectedness and contagion.