Headwind or Tailwind? How Banks Interpret Borrowers' AI Exposure (European Central Bank)

Summary

• As public attention to AI rose following the release of ChatGPT, banks lowered assessed default risk (PD) for borrowers in highly AI-exposed industries by about 5%.p.2
• The study analyzes a panel of non-listed, non-financial corporate borrowers across Germany, France, Italy, and Spain, containing roughly 6.8 million firm-bank-quarter observations between 2021Q1 and 2024Q4.p.4
• The favorable industry-level reassessment was weaker for labor-intensive firms, falling by roughly 7% for borrowers with below-median labor intensity but only about 3% for those above it.p.4
• Within narrow product markets, banks reassessed market leaders less favorably, with market leaders' default risk rising by about 8% relative to the smallest competitor per standard deviation of AI attention.p.5
• Credit allocation adjusted alongside risk assessments, with market leaders receiving about 3.8% less drawn credit and 2.4% less committed credit relative to the smallest market competitors.p.5
• In highly AI-exposed markets, the quarterly probability of a lending relationship terminating for a market leader rose by 1.2 percentage points against a base of 7.1%.p.5

Key insights

1. Banks interpret industry-level generative AI exposure as a credit tailwind on average, reducing probability-of-default estimates for exposed borrowers as public attention to AI rises.p.2
2. Lenders with pre-existing sectoral expertise reassess AI-exposed borrowers significantly more favorably than non-specialist co-lenders evaluating the exact same firm.p.4
3. Long-standing bilateral relationship lending does not lead to differential reassessments of AI-exposed borrowers relative to co-lenders.p.5
4. The favorable credit risk reassessment is substantially tempered for labor-intensive firms where technology deployment presents greater operational and disruption risks.p.4
5. Generative AI exposure triggers a systematic reordering of credit risk within product markets, compressing the historical risk advantage enjoyed by market leaders.p.2
6. Credit allocation shifts in tandem with bank internal risk ratings, leading to reduced credit lines and higher relationship exit rates for established market leaders in exposed sectors.p.5
7. The relative risk reordering assigned by banks inside AI-exposed markets accurately predicts subsequent bad-credit performance over the following four quarters.p.5
8. Internal bank default probabilities act as forward-looking indicators of technological displacement before changes appear in firm financial statements.p.6