Accounting Information Inconsistencies and Their Effects on Insolvency Prediction Models

Many studies have shown that avoiding political costs is an incentive for firms to manipulate accounting information, e.g., McNichols and Wilson, 1988; Jones, 1991; Kato et al., 2001. The majority of them use discretionary accruals models as proxies to manipulation. This paper introduces a new variable (DIF) that measures data inconsistencies present in financial reports, replacing discretionary accruals in the detection of manipulation. Accounting information was collected from 2,033 Brazilian health maintenance organizations (HMOs). This information was then processed and financial ratios derived from insolvency prediction models and thresholds established by the regulatory agency were taken as attributes to differentiate solvent HMOs from insolvents. During this data processing, inconsistencies were identified and instead of being removed, were used to determine the value of the attribute DIF. Processed data was then analysed using data mining techniques and a series of classifiers were created. The classifiers found have high accuracy in terms of discriminating distressed companies, especially when the DIF variable is used. In addition, the attributes selected and the structure of the classifiers can be supported by traditional models of analysis based on financial ratios. Results are relevant for those who are interested in assessing firms' insolvency risk, because data inconsistencies may signal firms' performance, therefore shall not be removed from analysis