Detecting Financial Statement Fraud: An Alternative Evaluation of Automated Tools Using Portfolio Performance

This article investigates the effect of using financial statement fraud detection models in constructing investment portfolios. Three financial statement fraud detection models are recreated and used to inform portfolio construction. Portfolio performance is compared between two strategies investing in companies on the S&P 500 predicted to have the highest (lowest) likelihood of financial statement fraud according to three models. Investment performance under the two strategies and across the three models are assessed using Fama-French regressions over a trading period from 2003 to 2021 and during market shocks. The portfolio of companies with the highest likelihood of fraud underperforms, characterized by inadequate returns relative to risk exposures. In the case of low-likelihood firms, results are consistent with risk-reward expectations. Financial results were consistent across all three fraud models, indicating that each model effectively discriminates between companies predicted to exhibit financial statement fraud. This research investigates the effect of financial statement fraud risk on investment performance and provides an alternative evaluation of financial statement fraud detection models, complementing the traditional accounting analysis of such models.