EXPLAINABLE IDENTIFICATION OF TAX-RELATED FINANCIAL REPORTING ANOMALIES IN U.S. PUBLIC COMPANIES: CROSS-FITTED TAX-CONSISTENCY RESIDUALS AND OWEN-VALUE SHAPLEY PROPAGATION
World Journal of Economics and Business Research · 26.02.2023 · 1 (1) · 33-45
DOI: 10.61784/wjebr4125 ↗Kurz gefasst
Fachartikel in der Zeitschrift World Journal of Economics and Business Research, online erschienen am 26.02.2023. Verfasst von JiaYing Chen und 2 weiteren Autor:innen. Die Arbeit wurde bisher 12-mal zitiert (Crossref). Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Income-tax accounts are reported under two regimes, financial accounting and the tax code, so their consistency with the rest of a firm’s financial statements is a natural check on reporting quality. Existing machine-learning detectors of financial misreporting treat tax items as generic inputs, and their explanations rarely reach the line items that an examiner can verify. We propose TRACE, an explainable framework for identifying tax-related financial reporting anomalies in U.S. public companies. First, for every fiscal year TRACE estimates label-free peer expectations of income-tax expense, taxes payable and their changes with cross-fitted conditional median and dispersion models, so that the resulting tax-consistency residuals cannot absorb the anomaly they measure. Second, it embeds these residuals in an imbalance-aware gradient-boosting ensemble that combines dollar-scale and within-year peer-rank representations of 28 Compustat items and is selected on validation years only. Third, it explains each risk score with Propagated Shapley Explanations, which chain exact single-baseline TreeSHAP, an exact Owen-value allocation over the reported, expected and dispersion components of each residual, and guarded propagation to 33 line items, with a proven completeness guarantee. On 146,045 U.S. firm-years labelled with SEC Accounting and Auditing Enforcement Releases and evaluated strictly out of time (2003–2008), TRACE achieves the highest NDCG at the 1% screening cut-off (0.0330), an improvement of 99% over RUSBoost and 89% over LightGBM (bootstrap p = 0.019 and 0.010) at comparable AUC. Its label-free tax-consistency score alone outperforms four generic outlier detectors. The explanations are exactly complete, agree closely with KernelSHAP (Spearman correlation 0.98) at 2.7 times lower cost, and show that detected misstatements most often combine tax expense growing faster than peers’ with taxes payable below peer expectations.
Zitieren
APA
Chen, J., Zhang, X., & Han, X. (2023). EXPLAINABLE IDENTIFICATION OF TAX-RELATED FINANCIAL REPORTING ANOMALIES IN U.S. PUBLIC COMPANIES: CROSS-FITTED TAX-CONSISTENCY RESIDUALS AND OWEN-VALUE SHAPLEY PROPAGATION. World Journal of Economics and Business Research, 1(1), 33-45. https://doi.org/10.61784/wjebr4125
Vancouver
Chen J, Zhang X, Han X. EXPLAINABLE IDENTIFICATION OF TAX-RELATED FINANCIAL REPORTING ANOMALIES IN U.S. PUBLIC COMPANIES: CROSS-FITTED TAX-CONSISTENCY RESIDUALS AND OWEN-VALUE SHAPLEY PROPAGATION. World Journal of Economics and Business Research. 2023;1(1):33-45. doi:10.61784/wjebr4125
BibTeX
@article{Chen2023,
author = {Chen, JiaYing and Zhang, XuGuang and Han, Xu},
title = {EXPLAINABLE IDENTIFICATION OF TAX-RELATED FINANCIAL REPORTING ANOMALIES IN U.S. PUBLIC COMPANIES: CROSS-FITTED TAX-CONSISTENCY RESIDUALS AND OWEN-VALUE SHAPLEY PROPAGATION},
journal = {World Journal of Economics and Business Research},
year = {2023},
volume = {1},
number = {1},
pages = {33--45},
doi = {10.61784/wjebr4125}
}
EUVIMED