国际清算银行-信贷损失率洞察：全球数据库（英）-2023.5-31页

Summary of BIS Working Paper No 1101: Insights into Credit Loss Rates

Authors: Li Lian Ong, Christian Schmieder, Min Wei
Date: May 2023

1. Purpose and Key Contribution

• Closes a significant data gap in reliable economy-level credit risk information, crucial for financial stability analysis.
• Derives time series of actual, forward-looking market-implied, and macro-implied credit loss rates for global jurisdictions.
• Provides a public dashboard with downloads, enabling scenario analyses based on GDP projections.

2. Typology of Metrics

• Forward-looking Metrics:
  • Individual Firm Data: Combines NUS-CRI (Probability of Default, PD) with World Bank (Loss Given Default, LGD), adjusted.
  • Macro-implied Rates: Projects losses using GDP trajectories, calibrated to historical crisis patterns.
• Contemporaneous Realized Metrics: Uses bank reports (BankFocus) for impairment and charge-off rates.
• *NPL-Based Metrics: Derives accrual ratios from National Bank of Statistics NPL data.

3. Data Sources

• Public: IMF, World Bank, BankFocus, NUS-CRI.
• Private: Vendors (unavailable to public), vendor credit loss rates.
• Adjustments: Handles missing data, outliers, and standardizes series; applies value thresholds (min: 0.05%, max: 30%).

4. Applications

• Financial Stability Analysis: Compares cross-country loss patterns (e.g., during GFC and COVID-19).
• Scenario Analysis: Simulates credit losses under varying GDP trajectories, especially useful for anticipating crisis impacts.
• Regulatory Use: Supports prudential supervision and policy calibration.

5. Contribution

• Addresses gaps in credit loss data for EMEs and LIDCs.
• Offers a suite of metrics tailored to different analytical needs (anticipating v. nowcasting losses, stock v. flow assessments).
• Highlights limitations: data scarcity, policy-induced distortions, and slower market adoption.

Key Insights

• Credit risk data remains fragmented and largely untapped for systematic financial stability analysis.
• GDP elasticity of credit losses varies by economy (AEs have lower sensitivity than EMEs).
• Forward-looking market-implied and macro-implied metrics complement realized data, but predictive accuracy is constrained by crisis unpredictability.