Assessing Systemic Risk Based on the CARE System

Abstract

Expectile is a coherent and elicitable risk measure that responds to the catastrophic losses more properly than a quantile-based approach. In this paper, we extend the univariate conditional autoregressive expectile (CARE) model of Kuan et al. (2009) to a multivariate CARE system to model the systemic risk of financial institutions. We propose an estimation method and derive its asymptotic properties. In the empirical study, we model the 12-variate CARE system on the global systemically important banks' (G-SIBs') returns for the period Jan 2015 to Jun 2020. We show that the CARE system identifies the asymmetric transmissions of the banks' systemic risk, i.e., the square positive and the square negative lagged returns generate different patterns of associations in the expectile connectedness. In addition, the out-of-sample Expectile-based Value-at-Risk prediction captures the 2020 mid-March US trading curb turbulence.