A carregar...
Publicação
Developing a Credit Risk model for corporate lending
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.99 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Credit risk models are an important tool for credit risk management in both banking and FinTech
institutions, they may influence both loan approval and pricing decisions. Recently, more attention
was drawn to the development of credit risk models for small and medium enterprises (SME) lending,
and the use of machine learning models has been researched and applied. Also, the sample bias that
arises from only having information for the accepted applicants and the class imbalance that comes
from the low number of defaults pose a challenge to developing models that are adequate for their
use in practical applications. This thesis aims to develop an SME lending model, bringing the use of
LightGBM as a machine learning algorithm, and comparing it to a well-known statistical learning
technique that is used in the industry, Logistic Regression, with two different modifications. Also,
another aim is to introduce techniques to deal with the class imbalance (in this case, Synthetic Minority
Oversampling Technique, SMOTE) and for the reject inference (Label Spreading). The results show that
the reject inference technique seems to be well-applied to the dataset since the expected bad rate for
the rejects is higher than the accepts. Also, the LightGBM algorithm presented better results than both
Logistic Regression algorithms that were used (using Lasso and Ridge penalty), showing an effective
model can be trained in the presence of a low number of defaults and, at the same time, consider
information that comes from the rejects.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Statistics and Information Management, specialization in Risk Analysis and Management
Palavras-chave
Credit scoring Reject inference class imbalance SME credit risk modeling