What type of study is this?

This is a Quantitative Study study.

October 2, 2025Open Access

Predicting Internal Control Deficiencies with Deep Learning: Evidence from IPO Review Records

Key Points

The study found that incorporating proxy information increases PR-AUC and ROC-AUC by 22.78% and 4.52% respectively.
Interpretability analyses demonstrated that metrics like ROA and OPM are crucial in identifying firms at high risk of internal control deficiencies.
Using IPO rejection outcomes as proxy labels significantly advances the construction of predictive models for internal control deficiencies.
The approach optimizes audit procedures and enhances firms’ self-assessments of internal controls, signifying both regulatory and theoretical value.

Abstract

Internal control is a core governance mechanism for improving information quality and reducing information risk. Identification of internal control deficiencies (ICDs) based on annual report disclosures or audit opinions suffers from reporting lags and sparse positives. Leveraging China’s issuance regime, this study uses IPO rejection outcomes as proxy labels and builds a financial-ratio–based prediction model. Incorporating proxy information lifts PR-AUC and ROC-AUC over the baseline by 22.78% and 4.52%; a full-parameter search yields gains of 24.05% and 5.32%. Under fixed thresholds, F1, precision, and recall also rise. Interpretability analyses show that ROA, OPM, and PPE retain persistent importance in identifying high-risk firms. The approach advances label construction and methodology, and offers an operational tool to target regulatory inquiries, optimize audit procedures, and support firms’ internal control self-assessments, with clear policy and theoretical value.

Ask AI

Helpful

Bookmark

View Full Paper