Previsão de insolvência corporativa no Brasil considerando a regionalidade

• Autor(a) principal: Duarte, Denize Lemos
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFU
• Texto Completo: https://repositorio.ufu.br/handle/123456789/28779; http://doi.org/10.14393/ufu.di.2020.95
• topic: Falência; Bankruptcy; Regionalidade; Regionality; Risco de Crédito; Credit Risk; Regressão Logística; Logistic Regression; Florestas Aleatórias; Random Forest; CNPQ::CIENCIAS SOCIAIS APLICADAS::ADMINISTRACAO

The prediction of financial distress in the context of the credit analysis plays a crucial role for the market because the link with losses and high costs involved in the unfolding of the insolvency and credit recovery process. This research develops a comparison and evaluates two insolvency prediction models, one based on machine learning, called Random Forest (RF), and another on traditional statistics, Logistic Regression (LR), by using data from Brazilian companies between 2005 to 2018. We also verify the performance of the models considering a regionality property (trough the mesoregion of Triângulo Mineiro and Alto Paranaíba and Sul Goiano). In order to deepen the knowledge, we carried out a systematic literature review on financial distress and bankruptcy where we detected that artificial intelligence technology is constantly improving to predict companies in financial distress. After that, we focused on the analyzes of the model performances. The main results according to the accuracy, brier score, forecast errors and area under the ROC curve (AUC) metrics showed that the RF classifier surpasses the LR model, considering the AUC, the predictive capacity in the national scenario was 96, 7%, and 93.4% and in the region's companies was 96.3% and 90.9% respectively. However, compared to other studies, the LR model presented a satisfactory result.