国际清算银行-人工智能与关系借贷（英）-2025.2_36页_867kb

Summary

Key Findings

• AI and Credit Scoring Interaction: The study examines how banks' use of AI in credit scoring interacts with relationship lending. AI adoption helps mitigate the countercyclical effects of relationship lending on firms' credit supply. Specifically, AI reduces asymmetric information issues during normal times and shifts lending focus to firm-specific conditions rather than macroeconomic shocks. During the COVID-19 crisis, AI-integrated banks provide more stable credit conditions compared to traditional banks.

• Crisis Period Effects: Using data from Italian banks matched with credit register information during the COVID-19 pandemic, the research finds that AI investments lead to increased credit supply and lower interest rates for firms with longer relationships. Non-AI banks exhibit countercyclical lending behavior, increasing loan volumes and reducing rates during the crisis, but AI adoption smoothens this effect, making AI lending less responsive to broad economic shocks.

• Real Effects on Firms: The study shows that AI adoption, combined with relationship lending, positively affects firms' investment and employment decisions during crises. However, it eases the smoothing effect of relationship lending, resulting in limited but noticeable dampening of borrowing behavior and credit access for AI-served firms.

• Methodology and Data: A unique dataset of Italian banks' AI investments in credit scoring is used, along with loan-level and macroeconomic data from 2019Q1-2020Q4. Robustness checks confirm findings across various control variables and sector-specific analyses.

• Policy Implications: AI's role in credit scoring decouples aggregate credit supply from general macroeconomic conditions, emphasizing borrower-specific factors. This may influence monetary policy and financial stability, as AI lending responds more to firm performance (e.g., profitability, transaction volumes) rather than broad economic indicators.

Despite AI's benefits, the research suggests it does not fully replicate non-AI banking behavior in crisis relief, underscoring the complementary nature of technology and traditional relationship lending. Policies should encourage AI adoption while maintaining access for all firms during economic downturns.