A carregar...
Publicação
Interpretable models of loss given default
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1011.46 KB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A gestão do risco de crédito é uma área em que os reguladores esperam que os bancos adotem modelos de risco transparentes e auditáveis colocando de parte o uso de modelos de black-box apesar destes serem mais precisos. Neste estudo, mostramos que os bancos não precisam de sacrificar a precisão preditiva ao custo da transparência do modelo para estar em conformidade com os requisitos regulatórios. Ilustramos isso mostrando que as previsões de perdas de crédito fornecidas por um modelo black-box podem ser facilmente explicadas em termos dos seus inputs.
Credit risk management is an area where regulators expect banks to have transparent and auditable risk models, which would preclude the use of more accurate black-box models. Furthermore, the opaqueness of these models may hide unknown biases that may lead to unfair lending decisions. In this study, we show that banks do not have to sacrifice predictive accuracy at the cost of model transparency to be compliant with regulatory requirements. We illustrate this by showing that the predictions of credit losses given by a black-box model can be easily explained in terms of their inputs. Because black-box models fit better the data, banks should consider the determinants of credit losses suggested by these models in lending decisions and pricing of credit exposures.
Credit risk management is an area where regulators expect banks to have transparent and auditable risk models, which would preclude the use of more accurate black-box models. Furthermore, the opaqueness of these models may hide unknown biases that may lead to unfair lending decisions. In this study, we show that banks do not have to sacrifice predictive accuracy at the cost of model transparency to be compliant with regulatory requirements. We illustrate this by showing that the predictions of credit losses given by a black-box model can be easily explained in terms of their inputs. Because black-box models fit better the data, banks should consider the determinants of credit losses suggested by these models in lending decisions and pricing of credit exposures.
Descrição
Mestrado em Econometria Aplicada e Previsão
Palavras-chave
Loss Given Default XGBoost Model interpretability Black-Box model.
Contexto Educativo
Citação
Matos, Sara Madeira (2020). "Interpretable models of loss given default". Dissertação de Mestrado. Universidade de Lisboa. Instituto Superior de Economia e Gestão.
Editora
Instituto Superior de Economia e Gestão