Detecting financial restatements using data mining techniques

Financial restatements have been a major concern for the regulators, investors and market participants. Most of the previous studies focus only on fraudulent (or intentional) restatements and the literature has largely ignored unintentional restatements. Earlier studies have shown that large scale unintentional restatements can be equally detrimental and may erode investors' confidence. Therefore it is important for us to pay a close to the significant unintentional restatements as well. A lack of focus on unintentional restatements could lead to a more relaxed internal control environment and lessen the efforts for curbing managerial oversights and instances of misreporting. In order to address this research gap, we focus on developing predictive models based on both intentional (fraudulent) and unintentional (erroneous) financial restatements using a comprehensive real dataset that includes 3,513 restatement cases over a period of 2001 to 2014. To the best of our knowledge it is the most comprehensive dataset used in the financial restatement predictive models. Our study also makes contributions to the datamining literature by (i) focussing on various datamining techniques and presenting a comparative analysis, (ii) ensuring the robustness of various predictive models over different time periods. We have employed all widely used data mining techniques in this area, namely, Decision Tree (DT), Artificial Neural Network (ANN), Naïve Bayes (NB), Support Vector Machine (SVM), and Bayesian Belief Network (BBN) Classifier while developing the predictive models. We find that ANN outperforms other data mining algorithms in our empirical setup in terms of accuracy and area under the ROC curve. It is worth noting that our models remain consistent over the full sample period (2001-2014), pre-financial-crisis period (2001-2008), and post-financial-crisis period (2009-2014). We believe this study will benefit academics, regulators, policymakers and investors. In particular, regulators and policymakers can pay a close attention to the suspected firms and investors can take actions in advance to reduce their investment risks. The results can also help improving expert and intelligent systems by providing more insights on both intentional and unintentional financial restatements.