A carregar...
Publicação
Desenvolvimento de um Sistema de Previsão de Insolvência das Sociedades Seguradoras no Brasil
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.68 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
O mercado segurador vem adquirindo participação cada vez mais relevante no cenário econômico do país, com a participação no PIB crescendo significativamente nos últimos anos, e a Superintendência de Seguros Privados (Susep), órgão responsável pelo controle e fiscalização de todo o mercado, tem por objetivo zelar pela solvência das companhias seguradoras e garantir o interesse dos segurados. Neste contexto, prever com antecedência a ocorrência de problemas financeiros é fundamental para evitar a quebra de uma companhia e permitir que os consumidores de seguro tenham preservado seu direito a receber as indenizações ou a poupança acumulada por anos. O presente trabalho propõe a utilização de modelos preditivos, mais especificamente a classe de algoritmos baseados em Machine Learning (ML), para sinalização antecipada de situações de insuficiência de capital em sociedades seguradoras e resseguradoras. O caso foi transformado em um problema de classificação binária, cujas variáveis explicativas foram indicadores financeiros e macroeconômicos e outros indicadores que refletem o porte da empresa, pertencimento a conglomerados financeiros, atuação em determinados ramos de seguro e problemas relacionados a controles internos. Na modelagem, foram utilizados diversos algoritmos de aprendizagem supervisionada, desde mais simples, como Naive Bayes, a mais complexos, como Gradient Boosting. Os classificadores foram treinados e avaliados, sendo conduzida uma comparação das performances em diferentes abordagens, para a Recall, Precision e F1-Measure. O modelo de melhor performance foi capaz de atingir uma Recall de 92%, conseguindo prever 11 dos 12 casos de insuficiência no test set.
The Insurance market in Brazil has been taking an even more relevant in the economic outlook, with its GDP participation growing significantly in the last years, and Susep, the entity responsible for monitoring and overseeing the whole market, has the objective of protecting the solvency of insurance companies e ensure the policyholders’ concerns. In this context, it is crucial to predict in advance the occurrence of financial difficulties in order to avoid the bankruptcy of a company and to permit the consumers to receive their claims or their savings accumulated over the years. This thesis aims to study the use of predictive models, more specifically the class of algorithms based on Machine Learning (ML), to create an early warning system for situations of violation of capital requirements in insurance and reinsurance companies. The case was designed to be a binary classification problem, whose independent variables were financial and macroeconomic indicators as well as further indicators that reflect the size of the company, participation in financial conglomerates, insurance lines of action, and internal control issues. In the modeling, a range of supervised learning algorithms was used, from the simplistic ones, like Naive Bayes, to the more complex ones, like Gradient Boosting. The classifiers were implemented and evaluated, by conducting a performance comparison for different approaches, using Recall, Precision, and F1-Measure metrics. The best model was able to reach a Recall of 92%, managing to predict 11 out of 12 instances of the positive class on the test set.
The Insurance market in Brazil has been taking an even more relevant in the economic outlook, with its GDP participation growing significantly in the last years, and Susep, the entity responsible for monitoring and overseeing the whole market, has the objective of protecting the solvency of insurance companies e ensure the policyholders’ concerns. In this context, it is crucial to predict in advance the occurrence of financial difficulties in order to avoid the bankruptcy of a company and to permit the consumers to receive their claims or their savings accumulated over the years. This thesis aims to study the use of predictive models, more specifically the class of algorithms based on Machine Learning (ML), to create an early warning system for situations of violation of capital requirements in insurance and reinsurance companies. The case was designed to be a binary classification problem, whose independent variables were financial and macroeconomic indicators as well as further indicators that reflect the size of the company, participation in financial conglomerates, insurance lines of action, and internal control issues. In the modeling, a range of supervised learning algorithms was used, from the simplistic ones, like Naive Bayes, to the more complex ones, like Gradient Boosting. The classifiers were implemented and evaluated, by conducting a performance comparison for different approaches, using Recall, Precision, and F1-Measure metrics. The best model was able to reach a Recall of 92%, managing to predict 11 out of 12 instances of the positive class on the test set.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science
Palavras-chave
Ciência de Dados Inteligência Artificial Aprendizagem de Máquina Aprendizagem Supervisionada Classificação Binária Sistema de Sinalização Antecipada Classificação Desequilibrada Seguros Data Science Artificial Intelligence Machine Learning Supervised Learning Binary Classification Early Warning System Imbalanced Classification Insurance Insolvency