Previsão de insolvência nas operadoras de planos de saúde no Brasil: um estudo com técnicas de análise multivariada

Abstract

This study aims to predict the beginning of insolvency state in health maintenance organizations using to this end financial ratios and statistical classification techniques. To achieve this objective, it was construced a sample of 491 organizations, 83 of which became insolvente, and 408 maintained theirselves solvente during the 2011 through 2013. To predict the beginning of insolvency state, three classification techniques were used: logistic regression, classification and regression trees and random forest. Thirteen financial indicators were calculated using the annual reports of those organizations that composed the sample, based on the year before the event happened or not. To be able to determine which technique had the best results, it was made a ROC curve analysis. The results show that the best technique to classify the organizations on solvents or insovents was the random forest. Eventhough, the orther techniques also proved themselves eficiente for this kind of classification.