We propose using a Bayesian time-varying coefficient model estimated with Markov chain-Monte Carlo methods to measure contagion empirically. The proposed measure works in the joint presence of heteroskedasticity and omitted variables and does not require knowledge of the timing of the crisis. It...