Is the Hawkes Graph Approach Applicable for Examining the Bankruptcy Risk Dependence Structure? An Empirical Analysis of Firms' Bankruptcies in Japan

Abstract

In this research, we examine the types of bankruptcy risk dependence structures of Japanese firms. For this purpose, we classify bankruptcy events observed in Japan into multiple event types based on industry type and firm size and then use a multidimensional Hawkes process that has been drawing attention recently in the field of finance research so as to model the self-exciting and/or mutually exciting properties of bankruptcy among the event types. As for estimation of the intensity processes associated with the multidimensional Hawkes process, we attempt a new approach using a nonparametric estimation for piece-wise constant kernel functions, introduced by Embrechts and Kirchner (2) to effectively obtain a weighted directed graph representing the estimation result called the Hawkes graph, as well as a conventional maximum likelihood estimation for an exponentially decaying kernel function. We compare and consider the estimation results of the two specifications via visualization with Hawkes graph representations to find that the Embrechts and Kirchner's estimation is applicable for examination of bankruptcy risk contagion.