Detecting Semantic Mismatches in XBRL Tag Mapping for SEC 10-K Filings: A Text Comparison and Historical Consistency Analysis

In one sentence

Summary

The paper addresses the critical issue of semantic mismatches in XBRL tag mapping within SEC 10-K filings, which undermines the reliability of automated financial analysis. Despite the mandate for Inline XBRL, tag mapping accuracy remains a persistent challenge, with custom tag usage rates fluctuating between 16% and 23% across filer categories. The research positions itself at the intersection of financial data quality and NLP, aiming to provide a computationally efficient method for detecting errors that standard validation rules often miss, particularly those involving semantic nuances rather than structural violations. The methodology employs a tiered text comparison approach combining lexical similarity scoring (TF-IDF and BM25) with domain-specific contextual features extracted from the SEC Financial Statement Data Sets (2014–2024). The system analyzes the alignment between reported line-item labels and assigned taxonomy elements, incorporating polarity, hierarchical, and measurement features. Additionally, it utilizes cross-period consistency analysis and SIC-code industry peer benchmarking to identify anomalous tag selection changes. The study leverages the XBRL US Data Quality Committee validation rules and SEC trend reports to establish baselines and detect deviations indicative of data quality degradation. Findings reveal persistent heterogeneity in custom tag rates, with non-accelerated filers and complex industries like insurance and real estate showing higher error propensity. The analysis identifies three primary mismatch categories: qualifier, granularity, and concept boundary errors. The proposed lightweight verification methods are designed for integration into disclosure management workflows, offering pre-tagging, post-tagging, and pre-submission checkpoints. While effective for reducing computational overhead, the approach relies on taxonomy definition quality and requires sufficient peer data for benchmarking, limiting its utility for highly specialized industries with few filers.

Research problem

The paper investigates the accuracy of XBRL tag mapping in SEC 10-K filings, specifically focusing on semantic mismatches between financial statement line-item labels and their corresponding taxonomy elements. It addresses the challenge of detecting these mismatches, which affect the reliability of automated financial analysis, and explores whether text comparison and historical consistency methods can identify data quality issues distinct from substantive business changes.

Core contributions

• Proposes a tiered text comparison framework combining TF-IDF and BM25 with domain-specific contextual features for XBRL tag verification.
• Introduces a cross-period consistency analysis and SIC-code industry peer benchmarking method to detect anomalous tag selection changes.
• Categorizes semantic mismatches into qualifier, granularity, and concept boundary types with specific detection strategies for each.
• Provides empirical analysis of custom tag usage patterns across filer categories and industries from 2014 to 2024.
• Designs lightweight verification checkpoints for integration into existing disclosure management workflows without complex infrastructure.

Method

• Lexical similarity scoring using TF-IDF and BM25 on line-item labels vs. taxonomy definitions.
• Domain-specific contextual feature extraction including polarity, hierarchical, and measurement features.
• Cross-period tag variation detection comparing tag assignments across consecutive filing periods.
• SIC-code industry peer comparison flagging tags used by fewer than 5% of peers for the same line item.
• Integration of Loughran-McDonald Master Dictionary for domain-specific financial vocabulary.
• Use of FinBERT embeddings for semantic distance measurement in temporal consistency analysis.

Results & findings

• Custom tag rates range between 16% and 23% across filer categories, with non-accelerated filers consistently showing higher rates.
• Third-party filing agents account for a significant portion of custom tag variance, particularly among smaller filers.
• Industries with complex financial instruments (insurance, real estate) exhibit higher-than-average custom tag rates.
• Qualifier mismatches are the most frequent error type, often involving net/gross or current/noncurrent distinctions.
• Tag stability temporarily declined during 2019-2020 due to ASC 842 adoption and expanded iXBRL requirements.
• The proposed lightweight methods process typical 10-K filings within seconds on standard hardware without GPU acceleration.

Data & experiments

• Utilizes SEC Financial Statement Data Sets (TAG, PRE, NUM, SUB files) for 10-K filings from 2014 to 2024.
• Leverages XBRL US Data Quality Committee validation rules and SEC Office of Structured Disclosure trend reports.
• Analyzes custom tag rates across filer categories (Large Accelerated, Accelerated, Non-Accelerated, Smaller Reporting Companies).
• Conducts cross-sectional tabulation at benchmark years (2014, 2017, 2019, 2020) and aggregate trend analysis through 2024.
• Performs industry peer benchmarking using SIC codes to establish expected tag usage distributions.

Limitations & caveats

• Relies on the quality of taxonomy element definitions, which vary in specificity.
• Industry peer benchmarking requires a minimum number of active filers per SIC code.
• May not produce reliable benchmarks for highly specialized industry codes with few filers.
• Does not include controlled experiments with disclosure professionals to measure error detection rates.
• Focuses on semantic mismatches rather than structural or mathematical errors covered by DQC rules.

Curator notes

Discovery status: report_analysis_focus_promoted. Source links: XBRL Agent: Leveraging Large Language Models for Financial Report Analysis