City Research Online

Abstract

The recent crisis highlighted the failure of former early warning signals models. This research attributes this partly to dependent variable specification, independent variable specification, model empirical design, and looking at models in isolation (different empirical methodologies and macro and micro applications). This research uses a traffic analysis matrix to synthesize the output of the different models, which are applied on a macro and micro level, while similarly attempting to improve on all the aforementioned in the individual applications. This approach results in significant improvement in out-of-sample results and lead-time compared to earlier work and a number of key insights for regulation and policymaking.

A dependent variable innovation compared to earlier literature in the component models of the traffic lights matrix lies in adopting an ex-ante near-crisis variable compared to an ex-post cost of crisis variable used before. This variable is applied in the macro and micro applications throughout. Near crises is a necessary and sufficient condition for prediction of full-fledged crises. Near crises always precede crises and then either develops into fully-fledged crises or they don’t.

The first paper applies a macro signal extraction framework and looks at the 30 OECD countries over a 30-year period (1979 to 2007). A number of variables were found to be significant in predicting near crises, including banking assets growth, banking assets to GDP, liquidity and a proxy for corporate sector health. The second paper is a macro application comprising a dynamic logit model and a macro Z-score model. The third paper is a Z-score methodology applied on a micro level to 139 banks. The micro application is an important extension in two ways. Systems that have more institutions under stress are scaled on a composite traffic light matrix as worse. The second extension is with regards to credit ratings or rankings within a system, whereby the micro application would allow regulators to do so.

Different models invariably have different output in some aspects and strengths and weaknesses. Signal extraction performed best in terms of Type I errors, the Logit model in terms of NTSR and the Z-score model in terms of Type II errors. The overlay of the micro model improves the traffic lights matrix substantially. These findings reinforce the need by regulators to use a suite of models and a holistic macroprudential approach in judging the build up of systemic vulnerabilities.

Publication Type:	Thesis (Doctoral)
Subjects:	H Social Sciences > HG Finance
Departments:	Bayes Business School > Finance Doctoral Theses Bayes Business School > Bayes Business School Doctoral Theses

Export

Downloads