Banca de DEFESA: Cayan Atreio Portela Bárcena Saavedra

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : Cayan Atreio Portela Bárcena Saavedra
DATE: 31/05/2023
TIME: 14:30
LOCAL: PPGA
TITLE:

Statistical Learning and Credit Risk: Modeling Probability of Default

KEY WORDS:

Machine Learning; Survival Analysis; Probability of Default; Credit Risk; Statistical Learning

PAGES: 80
BIG AREA: Ciências Sociais Aplicadas
AREA: Administração
SUMMARY:

The use of machine learning models for decision making is consolidated in the industry. Machine learning automated solutions are part of the expected credit loss estimation process. The present work investigates credit risk applications to estimate the default probability parameter. Different challenges, that arise at different stages of the automation pipeline, are addressed. The first chapter, investigates the use of sensitive variables when estimating probability of default. Different strategies, from pre-processing step do modeling classifier algorithms, are combined to identify the drop in performance caused by the absence of features that might bring sensitive information. The second chapter uses Survival Analysis to estimate probability of default. Since the implementation of International Financial Reporting Standard 9 (IFRS 9), the method required for estimating the PD parameter has experienced significant changes, demanding frequent improvement on techniques to comply with the new regulation. The third chapter propose a statistical learning method that incorporate information on prepayment, as a secondary risk, when estimating the probability of default. In this way, this work provides an overview of applications in PD estimation, using different datasets, and different methods of machine learning algorithms.

BANKING MEMBERS:
Externo à Instituição - ELI HADAD JUNIOR - UPM
Externo à Instituição - FABIANO GUASTI LIMA - USP
Presidente - 1997721 - HERBERT KIMURA
Externa ao Programa - 1804045 - JULIANA BETINI FACHINI GOMES
Externo à Instituição - LEONARDO FERNANDO CRUZ BASSO - UPM