Previsão Da Insolvência Empresarial Utilizando Redes Neurais Artificiais

Abstract

Nas negociações de crédito, o risco é um custo que está sempre presente e, portanto, precisa ser quantificado. Neste cenário, existem diversas ferramentas que se propõem à análise do crédito, algumas delas de ordem quantitativa. Neste sentido, esse artigo tem por objetivo propor um modelo capaz de prever a insolvência de empresas por meio da aplicação do modelo de redes neurais artificiais. O estudo é uma pesquisa exploratória de caráter quantitativo, aplicado à área financeira, utilizando-se o modelo tradicional e o modelo dinâmico de análise financeira. Com os resultados, obtiveram-se dois modelos: um contendo apenas as variáveis do modelo tradicional e outro com as variáveis do modelo tradicional e do modelo dinâmico de análise financeira. A comparação entre estes dois modelos de análise de crédito possibilitou verificar a contribuição das variáveis do modelo dinâmico para o modelo final. Os índices que mais contribuíram para acurácia do modelo proposto foram: Índice de Rentabilidade (X5) com 100% de exatidão; Índice de Estrutura de Capitais (X2) com 98,9% de acerto; e Índice do Modelo Dinâmico (X8) com 91% de precisão.Palavras-chave: Análise de balanços. Risco de crédito. Insolvência. Modelos de previsão.ABSTRACTIn credit negotiations the risk is a cost that is always present and therefore needs to be quantified. In this scenario, there are several tools that propose to credit analysis, some of them of quantitative order. In this sense, this article aims to propose a model capable of predicting the insolvency of companies by applying the artificial neural networks model. The study is exploratory research of quantitative character, applied to the financial area, using the traditional model and the dynamic model of financial analysis. The results obtained two models: one containing only the variables of the traditional model and another with the variables of the traditional model and the dynamic model of financial analysis. The comparison between these two models of credit analysis made it possible to verify the contribution of the variables of the dynamic model to the final model. The indexes that contributed most to the accuracy of the proposed model were: Profitability Index (X5) with 100% accuracy; Capital Structure Index (X2) with 98.9% accuracy and Dynamic Model Index (X8) with 91% accuracy.Keywords: Analysis of balance sheets. Credit risk. Insolvency. Forecasting models..

Economic-Financial Evaluation of Brazilian Companies With Open Capital in the Period 2011 to 2018

“…Comparing the results achieved with the ANN estimate of other studies, it was possible to note that the degree of assertiveness estimated in the present study is close to those obtained by Odom e Sharda (1990), Chen and Du ( 2009), Azayite and Achchab (2018), Teixeira et al (2013) and Prado et al (2020). The work of Odom and Sharda (1990) showed approximately 80% of accuracy in the three samples used.…”

“…Chung et al (2008) also used ANN to estimate a model that could predict the financial situation of industrial companies in New Zealand, but presented an assertiveness percentage of 62%, lower than the percentages estimated by the other authors mentioned above. Prado et al (2020) estimated models for 100 companies in the commercial sector and reached 100% and 96.9% accuracy in both models which higher than those found in the literature. According to the authors, the tool has great relevance to evaluate important characteristics of financial statements.…”

“…In summary, mainly in relation to Chen and Du (2009), Azayite and Achchab (2018), Yildirim et al (2011) and Prado et al (2020), it is possible to verify that the ANN method has a high predictive power of companies on their economic and financial situation, thus constituting an important tool for credit risk analysis by financial institutions, managers companies and private investors.…”

“…The authors indicated that there is approximately 85% accuracy in predicting the solvency capacity of companies with ANN use. Prado et al (2020) used ANN to estimate two solvency prediction models for 100 companies in the commercial sector in 2003, with 68 companies for training and 32 for validation. The first model defined 6 variables with 96.9% of accuracy level while the second model has added 2 variables with 100% of accurancy discussed in the literature.…”