A Gradient Boosting Approach to Estimating Tail Risk Contagion

Abstract

We propose a flexible CoVaR-based measure to estimate the tail risk contagion across financial institutes in a high dimensional framework. Considering potential nonlinearity and interaction among the financial institutes, a single-index indicator representing directed spillover is derived from Gradient Boosting Machine based generalized quantile regression. Our approach can be utilized to monitor risk spillover channels for risk supervision. Empirically, we investigate 16 publicly-listed banks in China with our proposed method. We show the outperformance of our method over the linear model. The empirical study suggests that the tail risk is inclined to spill over into the same type of banks. Besides, we find that not only state-owned banks are systemically important, but small and medium banks can play key roles in tail risk contagion, too. The total connectedness peaks when the financial system is under distress.