A Markov Model of Bank Failure Estimated Using an Information-Theoretic Approach (WP 2003-1)

Abstract

In this paper, we develop an early-warning bank failure model designed specifically to capture the dynamic process underlying the transition from financially sound to closure. We model the transition process as a stationary Markov model and estimate the transition probabilities using a Generalized Maximum Entropy (GME) estimation technique. The GME estimation method is a member of the class of information-theoretic methods, is semi-parametric, and is better suited for estimating models in which the data are limited (e.g., few events, and data availability problems), highly collinear, and measured with error — conditions that often exist with micro-level banking data. In addition, this method allows us to incorporate prior information and impose fewer distributional assumptions relative to conventional maximum likelihood (or full information maximum likelihood) methods. We report estimates of the transition probabilities for nine transition states for the population of nationally chartered banks incorporating the effect of bank-specific and macroeconomic variables from 1984 through 1999.

Authors

Dennis Glennon and Amos Golan