Chinese firms' digital transformations correlate with better audited accounts but costlier audits: digitized clients show lower earnings management yet trigger greater auditor effort and higher fees, especially for non–Big Four firms.

Main Finding

Corporate digital transformation (DT) among China’s A‑share firms (2010–2023) improves audit quality (lower discretionary accruals) primarily by strengthening internal controls and increasing information transparency, but it also raises auditor inputs and costs (longer audit lags and higher audit fees). Effects are heterogeneous by auditor type: non‑Big Four auditors show larger audit‑quality gains but face substantially higher increases in audit effort and fees, while Big Four firms experience limited changes—evidencing a “digital divide” in audit capabilities.

Key Points

• Sample: 37,173 firm‑year observations of Shanghai & Shenzhen A‑share listed firms, 2010–2023; financial firms and incomplete records excluded; winsorized continuous variables (1st/99th pct).
• Main dependent variables:
  • Audit quality: AbsDA (absolute discretionary accruals) from a modified Jones model — higher AbsDA = lower audit quality.
  • Audit inputs: AuditLag (ln days from fiscal year‑end to audit report) and audit fees.
• Key independent variable: firm DT intensity (DCG_word) constructed by textual analysis of annual reports (frequency of DT‑related keywords, log(1+count) and scaled).
• Mediators: internal control quality (IC index from DIB) and information disclosure transparency (exchange disclosure grade transformed into numeric scale).
• Main empirical strategy: multivariate regressions of AbsDA (and audit inputs) on DT with standard controls (size, leverage, BM, Tobin’s Q, cashflow, receivables, fixed assets, executive pay, ownership variables, years listed, auditor opinion, audit fees, etc.), two‑step mediation tests for IC and disclosure.
• Core empirical findings:
  • DT is significantly associated with lower AbsDA (improved audit quality).
  • DT positively affects IC and disclosure transparency, and these channels mediate the DT → audit quality effect.
  • DT is also associated with increased audit inputs: longer audit lags and higher fees.
  • Heterogeneity by auditor type: non‑Big Four auditors show larger reductions in AbsDA but larger increases in audit inputs/fees; Big Four show marginal changes—consistent with differing IT capabilities and resource endowments.
• Interpretation: DT reduces information asymmetry (helpful for auditors) but raises complexity and new risks (requiring more and/or specialized audit effort). Big Four firms’ advanced technology and human capital allow them to adapt with smaller marginal increases in input and to capture fee premiums; smaller auditors face steeper cost/effort increases.

Data & Methods

• Data sources: CSMAR for firm financials; DIB for internal control index; Shanghai & Shenzhen exchange disclosure evaluations; auditor/type and audit fees from filings.
• DT measure (DCG_word): keyword dictionary for DT topics (AI, cloud, blockchain, IoT, big data, digital platform, etc.); count occurrences in annual reports with Python Jieba segmentation; ignore negated mentions; log(1+count) and scale by 100.
• Audit quality metric: AbsDA from modified Jones / Dechow approach (firm‑scaled total accrual less non‑discretionary accruals).
• Audit input proxies: AuditLag = ln(days between fiscal year‑end and audit report); audit fees as reported.
• Mediators:
  • IC: internal control index (DIB).
  • Disclosure: exchange evaluation coded so higher numeric value = worse transparency.
• Estimation: OLS regressions with firm‑year controls and robustness checks; two‑step mediation tests to evaluate internal control and disclosure channels.
• Robustness and limitations noted by authors: large panel, winsorization, but potential concerns remain (textual DT proxy, possible endogeneity/reverse causality, China‑specific institutional context).

Implications for AI Economics

• Demand for skilled audit labor and AI tools will increase: DT raises the complexity and volume of digitally native evidence (big data, on‑chain/off‑chain reconciliation, system logs), creating higher demand for auditors with data‑science, IT audit, and AI competencies.
• Heterogeneous adoption creates market power and barrier effects: Big Four firms’ advanced analytics/technology investments let them handle DT more efficiently and extract fee premiums, potentially increasing market concentration and raising entry barriers for smaller audit firms—an important redistribution effect in labor and capital markets in the digital economy.
• Investment incentives for audit‑AI: The results imply strong returns to investing in audit automation, AI analytics, and IT audit capabilities (efficiency and competitive positioning). This shapes firm and industry investment models and can accelerate adoption of AI audit tools.
• Fee dynamics and welfare tradeoffs: While DT can improve audit quality, it also increases audit costs for many auditors (especially smaller ones). Policymakers and market participants should weigh improved transparency/quality against higher monitoring costs and possible consolidation.
• Policy and regulation considerations:
  • Support for smaller audit firms (subsidies, shared tooling, training) could mitigate the digital divide and preserve competition.
  • Standardization and disclosure of firms’ digital processes (data schemas, system architectures, key IT controls) would reduce audit search/learning costs and improve comparability.
  • Oversight of AI tools in auditing: ensure transparency, validation standards, and auditability of algorithmic procedures (model documentation, explainability) to preserve audit reliability when AI is used.
• Research directions in AI economics:
  • Quantify returns to audit‑technology investments and their effects on auditor pricing and market structure.
  • Causal identification of DT → audit outcomes (instrumental variables, natural experiments) and cross‑country comparisons.
  • Microdata studies on the adoption and performance of specific AI tools in audit tasks, and labor market impacts on auditors (re‑skilling, wage premiums).
  • Interaction effects between enterprise DT and use of AI in auditing—does auditor use of AI moderate the DT → audit input tradeoff?

Limitations to keep in mind: results are based on Chinese listed firms (institutional and managerial capabilities may differ elsewhere); DT is proxied by keyword frequency (textual measures capture emphasis but not necessarily implementation depth or effectiveness); potential endogeneity between DT and audit outcomes is acknowledged though controls are extensive—causal mechanisms would benefit from stronger identification in future work.