ESSAYS ON FINANCIAL STABILITY: MARKET STRUCTURE AND EARLY WARNING SYSTEMS

This thesis bundles together two distinct yet overlapping strands of the banking literature, which have received renewed interest following the global financial crisis: market structure and financial stability, and early warning systems for banking crisis forecasting. The thesis consists of three chapters. The first chapter critically reviews the extant theoretical and empirical literature on the ambiguous concentration- stability nexus and the forecasting of banking crises, highlighting the specific gaps addressed in the following empirical chapters. The second chapter investigates the channels through which bank concentration affects financial stability. Recent evidence points to the presence of non-monotonicities in the relationship between market concentration and stability, implying that the channels identified in the theoretical literature may be at play simultaneously with varying magnitude, which crucially depend upon initial levels of concentration. Using panel data, the chapter tests for the simultaneous presence of the channels and finds evidence that the prevalence of one or the other is a function of the initial degree of market concentration. The third chapter provides a systematic analysis of the role played by the duration of a systemic banking crisis in affecting the relative ability of multinomial and binomial logit models in correctly predicting the arrival of a crisis. The specific hypothesis tested is that the longer the duration of the crisis the better is the multinomial logit model in forecasting systemic banking crises relative to the binomial logit model. Results confirm that the multinomial logit model outperforms alternative binomial models in correctly predicting the arrival of a systemic banking crisis. In particular, the performance of the multinomial model improves over the binomial logit when the average duration of the crisis increases: the longer the average duration of crises in the sample, the better the relative performance of the multinomial over alternative binomial specifications.