AI adoption noticeably lifts profitability at China’s listed commercial banks by cutting operating costs and improving capital allocation; gains are largest at banks with higher digital investment and tighter deposit–liability coupling, and vary by ownership, size and policy cycle.

Main Finding

AI adoption by Chinese commercial banks causally increases bank profitability. The effect operates mainly through two channels—improved operational efficiency and better resource allocation—and is robust to propensity-score matching, placebo tests, and an event-study test of parallel trends.

Key Points

• Sample: 17 A‑share listed Chinese commercial banks, 2009–2022.
• Identification: multi‑period difference‑in‑differences (staggered adoption) combined with propensity score matching (PSM); parallel trends supported by event‑study; placebo simulations conducted.
• Aggregate result: AI adoption → significant positive effect on bank profitability.
• Mechanisms:
  • Operational efficiency: AI raises the effectiveness of cost use (reported as higher cost elasticity of income) and improves liquidity/funding‑cycle management that sharpens deposit–loan turnover adaptability.
  • Resource allocation: AI enhances cross‑business capital allocation by improving risk–return matching across diversified operations.
• Heterogeneity: effects are stronger for banks with higher digital investment intensity and tighter customer‑stickiness–liability‑cost coupling; effects vary by ownership type (state vs. non‑state), bank size, and policy cycle timing.
• Contribution: introduces a disclosure‑based measure of AI adoption constructed from bank annual reports and exploits staggered adoption timing for causal inference.

Data & Methods

• Data: panel of 17 publicly listed Chinese commercial banks (A‑share), annual data 2009–2022.
• Treatment measure: binary (or timing) AI adoption indicator derived from firms’ annual report disclosures.
• Empirical strategy:
  • Multi‑period DID leveraging staggered adoption dates to estimate average treatment effects while controlling for time and bank fixed effects.
  • Propensity score matching used to balance observables before DID estimation.
  • Event‑study specification used to test and confirm the parallel trends assumption.
  • Placebo simulations run to assess the possibility of spurious inference.
• Mechanism tests: mediation/heterogeneity analyses linking AI adoption to (i) cost/income relationships, (ii) deposit–loan turnover and liquidity/funding‑cycle metrics, and (iii) measures of intra‑bank capital allocation efficiency and risk–return alignment.

Implications for AI Economics

• Measurement: disclosure‑based indicators from annual reports are a practical, scalable way to identify firm‑level AI adoption in financial institutions; useful for other empirical work on digital technologies.
• Causal inference: staggered DID plus PSM and event‑study checks provide a credible template for studying adoption of technologies with staggered rollouts.
• Theory/practice: AI raises bank performance not only by cutting costs but by reallocating capital across activities more efficiently and improving liquidity management—models of firm productivity should incorporate both operational process gains and improved internal capital markets.
• Policy: regulators and policymakers should view AI investment as a lever for enhancing profitability and risk management, but outcomes depend on complementary investments (digital intensity) and business model features (customer stickiness, liability structures).
• Research gaps: validate generalizability beyond listed banks and China, examine long‑run risk implications (e.g., tail risks, systemic effects), and use microdata (product‑level, client‑level) to further unpack the mechanisms.