Does the asymmetric exponential power distribution improve systemic risk measurement?

The measurement of systemic risk using parametric modeling suffers from fat-tailedness, asymmetric kurtosis and asymmetric tails. Prior research shows that the asymmetric exponential power distribution (AEPD) can potentially avoid overfitting and underfitting problems because it can be reduced to a Gaussian distribution and a generalized error distribution. This paper implements a parametric estimation for the systemic risk measure CoVaR (ie, conditional value-at-risk) of Huang and Uryasev and compares the goodness-of-fit and backtesting performance of the AEPD with other commonly used distributions (ie, the normal, Student t and skewed t distributions). Based on data from the Chinese banking sector from 2008 to 2019, the empirical results show that AEPD has the best goodness-of-fit. Moreover, it is the only distribution that provides a validated estimation for CoVaR.