DISCONTINUANCE OF BRAZILIAN COMPANIES SECTOR CONSUMPTION CYCLIC: A METHODOLOGY FOR BALANCING OF DATABASE USING DATA MINING TECHNIQUES

Abstract

Discontinuity of companies is an issue that increasingly is being studied in the field of accounting and finance due to the considerable number of parts of the social fabric affected by the failure of a corporate entity. Banks, investors, governments, auditors, managers, suppliers, employees and many others have great interests in the accuracy of prediction of insolvency of a company. In Brazil, studies on the subject are still suffering the effects of being available only in databases with reduced dimensions, mostly due to data quality. But there is little studied issues in predictive modeling of insolvency. The balance or imbalance of data on insolvency is one of those issues in economic environments typical number of companies classified as solvent is much higher than those classified as insolvent. The aim of this study is to propose a new procedure for balancing of database problems in insolvency prediction with (step) feature selection. Was then constructed a strategy of data mining with the double virtue of selecting attributes and solve the problem of the imbalance. The database was derived from financial statements of Brazilian companies in the consumer cyclical economic sector, listed on the BOVESPA between the years 1996 and 2011. The results and validations performed demonstrate the success of the proposed strategy, improving the ability of the prediction model for the classification of companies belonging to the class of insolvent and thus consolidating it as quite competitive with other strategies presented in the specific literature. Given the nature of the empirical exercise, it is clear that the advantage of the new procedure does not depend on the studied sector.

Keywords: Discontinuity of companies; Brazilian companies from the consumer discretionary sector, accounting variables, Data mining; Balancing database.