Modeling bank default intensity in the USA using autoregressive duration models

This paper employs a duration based approach in order to model the inter-arrival times of bank failures in the US banking system for the period 1934 - 2014. Conditional duration models that allow duration between bank failures to depend linearly or nonlinearly on its past history are estimated and evaluated. We find evidence of strong persistence along with non-monotonic hazard rates which imply a financial contagion pattern according to which, a high frequency of bank failures generates turbulence which shortly after leads to additional fails, whereas prolonged periods without abnormal events signify the absence of contagious dependence which increases the relative periods between bank failure appearance. In addition, we find that mean duration levels of tranquility spells or equivalently the bank fail events intensity is subject to long run shifts. Further, we obtain statistical significant results when we allow duration to depend linearly on past information variables that capture systemic bank crisis factors along with stock and bond market effects.