The Cost of Fraud Prediction Errors

Abstract

We compare seven fraud prediction models with a cost-based measure that nets the benefits of correctly anticipating instances of fraud (true positives), against the costs borne by incorrectly flagging non-fraud firms (false positives). Our cost-based measure supplements traditional comparison metrics and is estimated separately for auditors, investors, and regulators. We find that the higher true positive rates in recent models come at the cost of higher false positive rates, and that even the best models trade off false to true positives at rates exceeding 100:1. Indeed, the high number of false positives makes all seven models considered too costly for auditors to implement, even when we consider extreme subsamples in which a priori the likelihood of misreporting is greater. For investors, M-Score and, when used at higher cut-offs the F-Score, are the only models providing a net benefit; the other models are only economically viable when applied to extreme subsamples of misreporting risk. For regulators, several models are economically viable as false positive costs are limited by the number of investigations they can initiate, and by the relatively low market value loss a ‘falsely accused’ firm would bear in denials of requests under the Freedom of information Act (FOIA). Our results are similar irrespective of whether we consider fraud or two alternative restatement samples.