Dynamic Bivariate Peak Over Threshold Model for Joint Tail Risk Dynamics of Financial Markets

Abstract

We propose a novel dynamic bivariate peak over threshold (PoT) model to study the time-varying behavior of joint tail risk in financial markets. The proposed framework provides simultaneous modeling for dynamics of marginal and joint tail risk, and generalizes the existing tail risk literature from univariate dimension to multivariate dimension. We introduce a natural and interpretable tail connectedness measure and examine the dynamics of joint tail behavior of global stock markets: empirical evidence suggests markets from the same continent have time-varying and high-level joint tail risk, and tail connectedness increases during periods of crisis. We further enrich the tail risk literature by developing a novel portfolio optimization procedure based on bivariate joint tail risk minimization, which gives promising risk-rewarding performance in backtesting.