Democritus University of Thrace (DUTH)
Αποτίμηση κινδύνων των ευρωπαϊκών τραπεζών και συστήματα έγκαιρης προειδοποίησης
Abstract
dc:descriptionFinancial crises valid forecasts ensure the stability of the entire financial edifice in general and the banking sector in particular. In this thesis two approaches are implemented in order to forecast the onset of systemic banking crises within countries of EU-14, several quarters in advance, using mainly standard macroeconomic, banking and market variables. First, following the binary approach, classification models by applying discriminant analysis, logistic, probit and linear regression are exported, in order to anticipate crises -12 to -7 quarters before they materialize. For this scope a new threshold extraction and evaluation criterion (GoF) and a new combined classification approach is implemented. Country – blocked 14 – fold cross validation indicates 82.4% classification accuracy, 78.4% true positive rate and 80.6% positive predictive value.According to the second ternary approach, two distinct periods of systemic banking crisis forecast, the so-called early warning period (-12 to -7 quarters prior to the crisis) and the late warning (-6 to -1 quarters prior to the crisis) are detected. For this multilevel classification, neural networks classification, polynomial logistic regression and multiclass linear discriminant analysis are utilized. Country – blocked 14 – fold cross validation indicates 85.7% correct classification under the optimal classification method, which is the multiclass linear discriminant analysis.Applying the presented framework, policy makers are able to predict a banking crisis up to 12 quarters in advance and therefore exercise a case by case macroprudential policy, using only public available and freely accessible data.
Degree
thesis:*- Grantor dc:publisher
- Democritus University of Thrace (DUTH)
- Year dc:date
- 2018
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Stavroulias, Pantelis
- Σταυρούλιας, Παντελής
Subjects
dc:subject × 12Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/44525