Forecasting Insolvency of Brazilian Publicly Traded Companies

摘要

Qualified information is a fundamental asset in the decision making process, particularly when high financial resources are involved. For example, anticipated detection of company bankruptcy, for sure, will help both the investor interested in the company as their managers. The challenge of exploring the huge amount of data to consider in such analysis can benefit from adequate data mining tools. A comparative analysis of classification algorithms for this kind of application is presented in order to find a better predictive performance. The experiments were performed on data from publicly traded companies registered in the Sao Paulo Securities, Commodities and Futures Exchange (Bovespa) that appeared at least twice from 1996 to 2010. Moreover, some experiments were carried out with the set of candidate attributes, in this case financial ratios and ledger accounts that could better explain the target variable. The better results were found with a committee of classifiers using both financial ratios and ledger accounts, although, the ledger accounts alone presented a higher prediction power than financial ratios.