BEYOND ISOLATED RATIOS: ACCRUAL–REVENUE–CASH FLOW INDICATORS AND INTERACTION-AWARE LEARNING FOR DETECTING FINANCIAL REPORTING IRREGULARITIES IN U.S. PUBLIC COMPANIES
Keywords:
Financial reporting irregularities, Accruals, SEC enforcement releases, Fraud detection, Gradient boosting, Anomaly detection, Explainable artificial intelligenceAbstract
Financial reporting irregularities are rare, evolve over time, and surface only after costly enforcement, which makes their detection from public filings a persistent challenge for regulators, auditors, and investors. This paper examines whether accounting-based indicators—accruals, revenue growth, operating cash flow, financial ratios, and the relations among them—identify misreporting risk out of sample. Using 146,045 firm-years of 10-K fundamentals of U.S. public companies (fiscal years 1990–2014) labeled with SEC Accounting and Auditing Enforcement Releases (964 misstated firm-years from 412 enforcement cases), we construct an Accrual–Revenue–Cash (ARC) indicator system of 53 indicators in six families, including cross-statement divergence and multi-year accrual accumulation, complemented by eight anomaly-lens scores based on ECOD. Unsupervised detectors, the F-score logit, RUSBoost, gradient boosting, a multilayer perceptron with piecewise linear encoding (PLE-MLP), and their rank ensemble (ARC-Lens) are evaluated under a rolling protocol with a 24-month gap and serial-fraud correction. ARC indicators raise the AUC of a logit model from 0.666 to 0.702 and of gradient boosting from 0.697 to 0.714. ARC-Lens reaches a mean AUC of 0.717 in 2003–2008, significantly above the F-score (+0.051) and the 42-input boosting benchmark (+0.020), on par with RUSBoost, and captures 14.9% of misstated firm-years in the top 5% of scores (10.7% for RUSBoost). Unsupervised outlier detectors perform at chance level (AUC 0.48–0.50): misreporting firms are statistically typical rather than extreme. Ablation and TreeSHAP analyses identify accrual accumulation as the dominant signal and attribute 58.9% of model attribution to pairwise interactions, chiefly between accrual measures and firm scale and financing.References
[1] Dechow P M, Sloan R G, Sweeney A P. Causes and consequences of earnings manipulation: An analysis of firms subject to enforcement actions by the SEC. Contemporary Accounting Research, 1996, 13(1): 1-36.
[2] Karpoff J M, Koester A, Lee D S, et al. Proxies and databases in financial misconduct research. The Accounting Review, 2017, 92(6): 129-163.
[3] Beneish M D. The detection of earnings manipulation. Financial Analysts Journal, 1999, 55(5): 24-36.
[4] Dechow P M, Ge W, Larson C R, et al. Predicting material accounting misstatements. Contemporary Accounting Research, 2011, 28(1): 17-82.
[5] Sloan R G. Do stock prices fully reflect information in accruals and cash flows about future earnings? The Accounting Review, 1996, 71(3): 289-315.
[6] Richardson S A, Sloan R G, Soliman M T, et al. Accrual reliability, earnings persistence and stock prices. Journal of Accounting and Economics, 2005, 39(3): 437-485.
[7] Jones J J. Earnings management during import relief investigations. Journal of Accounting Research, 1991, 29(2): 193-228.
[8] Dechow P M, Sloan R G, Sweeney A P. Detecting earnings management. The Accounting Review, 1995, 70(2): 193-225.
[9] Kothari S P, Leone A J, Wasley C E. Performance matched discretionary accrual measures. Journal of Accounting and Economics, 2005, 39(1): 163-197.
[10] Perols J. Financial statement fraud detection: An analysis of statistical and machine learning algorithms. Auditing: A Journal of Practice & Theory, 2011, 30(2): 19-50.
[11] Bao Y, Ke B, Li B, et al. Detecting accounting fraud in publicly traded U.S. firms using a machine learning approach. Journal of Accounting Research, 2020, 58(1): 199-235.
[12] Bertomeu J, Cheynel E, Floyd E, et al. Using machine learning to detect misstatements. Review of Accounting Studies, 2021, 26(2): 468-519.
[13] Achakzai M A K, Peng J. Detecting financial statement fraud using dynamic ensemble machine learning. International Review of Financial Analysis, 2023, 89: 102827.
[14] Liu F T, Ting K M, Zhou Z H. Isolation forest. In: Proceedings of the 8th IEEE International Conference on Data Mining (ICDM), 2008: 413-422.
[15] Li Z, Zhao Y, Hu X, et al. ECOD: Unsupervised outlier detection using empirical cumulative distribution functions. IEEE Transactions on Knowledge and Data Engineering, 2023, 35(12): 12181-12193.
[16] Han S, Hu X, Huang H, et al. ADBench: Anomaly detection benchmark. In: Advances in Neural Information Processing Systems (Datasets and Benchmarks Track), 2022, 35.
[17] Grinsztajn L, Oyallon E, Varoquaux G. Why do tree-based models still outperform deep learning on typical tabular data? In: Advances in Neural Information Processing Systems (Datasets and Benchmarks Track), 2022, 35.
[18] Gorishniy Y, Rubachev I, Babenko A. On embeddings for numerical features in tabular deep learning. In: Advances in Neural Information Processing Systems, 2022, 35.
[19] Shwartz-Ziv R, Armon A. Tabular data: Deep learning is not all you need. Information Fusion, 2022, 81: 84-90.
[20] Hribar P, Collins D W. Errors in estimating accruals: Implications for empirical research. Journal of Accounting Research, 2002, 40(1): 105-134.
[21] Amiram D, Bozanic Z, Rouen E. Financial statement errors: Evidence from the distributional properties of financial statement numbers. Review of Accounting Studies, 2015, 20(4): 1540-1593.
[22] Walker S. Compared to what? Does Benford’s law really detect corporate fraud? Econ Journal Watch, 2022, 19(1): 30-46.
[23] Seiffert C, Khoshgoftaar T M, Van Hulse J, et al. RUSBoost: A hybrid approach to alleviating class imbalance. IEEE Transactions on Systems, Man, and Cybernetics—Part A: Systems and Humans, 2010, 40(1): 185-197.
[24] Brown N C, Crowley R M, Elliott W B. What are you saying? Using topic to detect financial misreporting. Journal of Accounting Research, 2020, 58(1): 237-291.
[25] Chen T, Guestrin C. XGBoost: A scalable tree boosting system. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016: 785-794.
[26] Ke G, Meng Q, Finley T, et al. LightGBM: A highly efficient gradient boosting decision tree. In: Advances in Neural Information Processing Systems, 2017, 30: 3146-3154.
[27] Lundberg S M, Erion G, Chen H, et al. From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2020, 2(1): 56-67.