Anticipating corporate’s distresses

Autores

Flávio Barboza; Denize Lemos Duarte; MICHELE APARECIDA CUNHA

Dados bibliográficos

Ano: 2022

Local de publicação: Exacta

ISSN/ISBN: 1678-5428

DOI: 10.5585/exactaep.2021.17494

Resumo

The objective of this study is to present a model for predicting Financial
Distress (FD) from the perspective of machine learning techniques (TAMs). We
applied and compared the XGBoost, Random Forest and Logistic Regression models
using financial indicators to seek better forecasts of FDs one year before the event in
Latin American companies in the period from 2000 to 2017. Our results showed that
TAMs outperform the logit model, reaching an overall accuracy of 96% (XGBoost). In
addition, five indicators were relevant to its success. The study expands knowledge
and discussions by focusing on the predictive power of comparing models,
highlighting the benefits of using algorithms applied to financial research. It helps in
risk management, loss prevention, allowing greater balance and health for the
financial system, which contributes to the economic, social and sustainable
development of a society.