Contagion across asset classes during crises is typically detected with statistical techniques that separate true spillovers from coincident co-movement driven by common shocks, changing volatility, or market microstructure. Early empirical caution comes from Kristin Forbes at MIT Sloan School of Management and Roberto Rigobon at Massachusetts Institute of Technology who showed that naïve correlation increases can overstate contagion when heteroskedasticity and common factors are not controlled for. Their work motivates more robust, time-varying methods.
Time-varying correlation and volatility models
Practitioners frequently use DCC-GARCH to capture dynamic correlations while modeling conditional volatility. Robert Engle at New York University Stern School of Business developed this framework to estimate how correlations evolve during stress, distinguishing persistent links from transitory spikes. Vector autoregressions with volatility adjustments and structural VAR identification extend this by estimating impulse responses and attributing shocks to specific markets, which helps determine directionality of spillovers rather than simple co-movement.
Tail dependence, networks, and causality
Detection of extreme co-movements relies on copula-based methods and extreme value theory to quantify tail dependence; Andrew Patton at Duke University emphasized copulas for uncovering asymmetric dependence during downturns. Forecast-error variance decomposition and connectedness measures derived from VARs provide a network view of spillovers; Francis X. Diebold at the University of Pennsylvania pioneered measures of connectedness that summarize how shocks to one asset class explain variance in others over horizons. Granger causality tests and rolling-window causality checks add temporal direction, while change-point tests identify regime shifts that signal structural contagion rather than ordinary market co-movement.
Understanding causes and consequences requires contextual nuance. Contagion often amplifies through leveraged positions, margin calls, and cross-border funding lines, hitting emerging markets and real-economy stakeholders disproportionately and creating policy trade-offs between capital controls and liquidity provision. Method choice matters: failing to model time-varying volatility or tail behavior can misclassify interdependence as contagion, leading to inappropriate regulatory responses.
For robust inference combine approaches: control for global factors, model conditional volatility, test tail dependence, and use connectedness metrics to map transmission channels. This integrated strategy aligns statistical detection with economic interpretation, improving risk management and policy during crises while recognizing geographic and institutional heterogeneity in vulnerability.