Predicting Auditor Switches by Applying Data Mining

Abstract

Auditor dismissals are considered to be a threat to audit quality. Several studies have examined auditor switches by applying typical statistical analysis. In the present study we deal with the auditor switching problem by applying data mining methodologies. Publicly available financial statement and auditing data are used as predictors. The optimum vector of significant input variables is defined by employing feature selection. A number of data mining classification methods are used to develop models capable of predicting the auditor change cases. The methods are compared against the widely used Logistic Regression. According to the results, all the data mining methods outperform Logistic Regression. Significant factors associated with auditor changes are revealed. The results can be useful to auditing firms, managers, investors, creditors and corporate regulators.