Lists
Lists
Kanellopoulos, Konstantinos (2022) Analysis and forecasting of asset quality, risk management and financial stability for the Greek banking system. Doctoral thesis, London Metropolitan University.
The increase in non-performing loans (NPLs) during the financial crisis of 2008, which has been converted into a fiscal crisis, as well as the risk of a medium-term increase due to the COVID-19 pandemic has put into question the robustness of many banks and the financial stability of the whole sector. As far as the banking sector is concerned, the management of non-performing loans represents the most significant challenge as their stock reached unprecedented levels, with the deterioration in asset quality being widespread. Addressing the problem of non-performing loans with the assistance of credit risk modeling is important from both a micro and a macro-prudential perspective, since it would not only improve the financial soundness and the capital adequacy of the banking sector, but also free-up funds to be directed to other more productive sectors of the economy.
This Thesis extends earlier research by employing a short-term monitoring system with the aim to forecast “failures” i.e. NPL creation. The creation of such a monitoring system allows the risk of a “failure” to change over time, measuring the likelihood of “failure” given the survival time and a set of explanatory variables. The application of Cox proportional hazards models and survival trees to forecast NPLs can be usefully employed in the Greek corporate sectors.
The research aim of this thesis consists of two domains: The first aim is the investigation of the determinants that contribute to the NPLs formation. Two GAMLSS models are being tested, a linear GAMLSS model and a nonlinear semi-parametric GAMLSS model which includes smoothing functions that capture potential nonlinear relationships between the explanatory variables to model the parameters favorably. The explanatory variables of the models consist of credit risk variables, macroeconomic variables, bank-specific variables and supervisory and market variables, while the response variable is the non-performing loans.
The second aim is to provide answers on whether proportional hazards Cox models and survival tree models can forecast NPLs of loans that are provided in specific corporate sectors in Greece by the use of the most granular data set of corporate borrowers. By evaluating a series of Cox models, a short-term monitoring system has been created with the aim to forecast “failures” i.e. NPL creation. The Cox proportional hazards regression models are incorporating time-to-event, involving a timeline, described by the survival function, indicating the probability that a loan becomes an NPL until time t. The time period counts from the origination of the loan until the “death” of the loan, i.e. its termination, incorporating an “in between” observation point. The event is when the loan is initially being “infected”, i.e. has become NPL. Regarding survival trees, the data set was divided into more subsets, which are easier to model separately and hence yield an improved overall performance. Such models are then beneficial to implement with different machine learning techniques. Predictors (or covariates) are defined as the sectors of the Greek economy and the model is fitted both for the whole sample and for the sample of early terminated loans.
The Thesis is organized as follows: Chapter 1 - Introduction addresses the role of banks in financial intermediation, the evolution of credit risk and some issues regarding the Greek banking sector. Chapter 2 constitutes a literature review on research focused on improving the predictive performance of different credit risk assessment methods. Chapter 3 outlines the competitive conditions in the banking sector to demonstrate whether the increase in concentration had affected the competitive conditions in the Greek banking system. In Chapter 4, the funding and the liquidity conditions in the Greek banking sector are being addressed. Chapter 5 contains the selection of aggregate sample, results and analysis of GAMLSS models that have been used for determining NPLs. Chapter 6 provides an introduction to the granular database on Large Exposures, which is used for deriving the panel sample of corporate borrowers whereby models of forecasting and prediction are being employed. Chapter 7 contains the application of Cox models and decision trees, the estimation procedure, parameters, model fit, estimation results and empirical findings. Chapter 8 provides an evaluation and applicability of models as well as the implications for further research. Finally, a conclusion is provided by summarizing my contribution to the research community and my recommendations to the banking industry.
 |
View Item |