Please use this identifier to cite or link to this item: http://repositorio.ufla.br/jspui/handle/1/57037
Title: Risco de insolvência: uma análise de diferentes métodos de previsão e do efeito da estrutura de capital
Other Titles: Insolvency risk: an analysis of different forecasting methods and the effect of capital structure
Authors: Prado, José Willer do
Ávila, Ednilson Sebastião de
Andrade, Lélis Pedro de
Guimarães, Paulo Henrique Sales
Keywords: Determinantes do risco de insolvência
Modelos preditivos
Insolvência empresarial
Determinants of the risk of insolvency
Predictive models
Business insolvency
Issue Date: 21-Jun-2023
Publisher: Universidade Federal de Lavras
Citation: COSTA, T. R. da. Risco de insolvência: uma análise de diferentes métodos de previsão e do efeito da estrutura de capital. 2023. 169 p. Dissertação (Mestrado em Administração)–Universidade Federal de Lavras, Lavras, 2023.
Abstract: The overall objective of the study was to analyze the insolvency risk of Brazilian non-financial companies listed on B3, from the perspective of insolvency forecasting and capital structure, analyzing the assertiveness of forecasting models during the pandemic moment and the effects of COVID-19 on the influence of capital structure on the risk of insolvency. The specific objectives were: To carry out a bibliometric review on prediction and risk of insolvency (article 1); identify, among different forecasting techniques, which provides more assertive results for the pandemic period (article 2); analyze the effect of capital structure on the risk of insolvency and the impact of the pandemic as a moderator in this relationship (Article 3). Bibliometric techniques (article 1), discriminant analysis, logistic regression, k-nearest neighbors, decision tree, random forest, artificial neural networks and support vector machines (article 2) and multilevel data regression were used in panel (article 3). As a result, article 1 showed that the field of study developed slowly until the 2000s, with a tendency to grow from 2008 onwards. most relevant was that of Altman (1968). The main country for the field of studies was the United States, but China had the highest volume of publications. The most relevant journal was Expert Systems With Application, and when analyzing the keywords of the field of study, the main ones were classification, financial ratios, risk, neural-networks and models. Trending topics were computational like big data, machine learning, and deep learning. In article 2, the technique with the best percentage of accuracy was decision tree, followed by random forest, logistic regression and discriminant analysis, artificial neural network, k-nearest neighbors and support vector machine. This result showed that the white box, gray box and traditional techniques had better accuracy in relation to the black box techniques (ANN and SVM). Through Article 3, it was possible to identify that some of the capital structure variables showed significant and positive results, confirming the positive influence of debt on the risk of insolvency. The financial indebtedness variable and its quadratic term showed significant results, which confirmed the U-shaped relationship between financial indebtedness and the risk of insolvency. Regarding the moderating effect of the pandemic, only the hypothesis that the COVID-19 pandemic strengthened the positive influence of long-term debt on the risk of insolvency was confirmed. This study contributed to the risk and insolvency prediction literature by performing a comprehensive bibliometric analysis, considering two relevant databases (Web of Science and Scopus); by developing insolvency prediction models considering different forecasting techniques and analyzing the assertiveness for the moment of a pandemic; and analyzing the influence of capital structure on insolvency risk, as well as the moderating effect of the pandemic.
URI: http://repositorio.ufla.br/jspui/handle/1/57037
Appears in Collections:Administração - Mestrado (Dissertação)



This item is licensed under a Creative Commons License Creative Commons