ARIMA models for bank failures: Prediction and comparison

The number of bank failures has increased dramatically over the last twenty-two years. A common notion in economics is that some banks can become "too big to fail." Is this still a true statement? What is the relationship, if any, between bank sizes and bank failures? In this thesis, the proposed modeling techniques are applied to real bank failure data from the FDIC. In particular, quarterly data from 1989:Q1 to 2010:Q4 are used in the data analysis, which includes three major parts: 1) pairwise bank failure rate comparisons using the conditional test (Przyborowski and Wilenski, 1940); 2) development of the empirical recurrence rate (Ho, 2008) and the empirical recurrence rates ratio time series; and 3) the Autoregressive Integrated Moving Average (ARIMA) model selection, validation, and forecasting for the bank failures classified by the total assets.