Uso de aprendizado de máquina supervisionado para mensurar provisão no mercado de crédito estruturado: uma comparação entre modelos

Abstract

The credit expansion has been fundamental to global economic growth, allowing people and businesses to access more financial resources for investment and consumption within an increasingly complex and competitive economic system. Among the various types of credit, this work highlights structured credit, in which debts from various sources are unified into a single financial product. Given the risk associated with each of these credits, accurately measuring a fair provision for these structures is a major challenge for market agents. In this context, the objective of this study is to apply and compare different supervised learning techniques to predict default events in structured credit portfolios, in order to assist financial institutions in making more accurate and secure provisions. After the analyses, it was possible to conclude that the XGBoost algorithm outperformed the others in terms of accuracy and sensitivity, although all of the others obtained satisfactory results.