Can Credit Growth and The Yield Curve Predict Financial Crises?

Key Takeaways

• Decision tree ensembles outperform logistic regression in crisis prediction
• Credit growth relative to GDP is a primary early warning signal
• Flat or inverted yield curves signal upcoming financial turmoil
• Global variables amplify domestic credit and yield curve risks
• Shapley analysis reveals individual predictor contributions to model forecasts

Pulse Analysis

Machine learning has moved from a niche academic exercise to a cornerstone of macro‑financial analysis. Traditional econometric models, such as logistic regression, often struggle with non‑linear interactions and high‑dimensional data, limiting their predictive power for rare events like financial crises. Tree‑based ensembles—extremely randomised trees and random forests—address these gaps by capturing complex patterns without overfitting, delivering robust out‑of‑sample forecasts across a century‑spanning dataset of 17 economies. This methodological shift aligns with a broader industry trend toward data‑driven risk management.

The ECB paper highlights two variables that consistently rise to the top of the predictive hierarchy: sustained credit growth relative to GDP and a flattening or inverted yield curve. Credit expansion signals leverage buildup, while an inverted curve reflects market expectations of tightening monetary conditions and potential recession. By applying a Shapley value framework, the authors quantify each variable’s marginal impact, revealing that even after controlling for recession indicators, the yield curve retains independent predictive strength. The model correctly flagged the 2007‑08 global crisis well in advance, underscoring the practical value of these early‑warning metrics.

For policymakers and institutional investors, the implications are clear. Incorporating machine‑learning‑derived signals into macro‑prudential dashboards can enhance surveillance, allowing pre‑emptive actions such as counter‑cyclical capital buffers or targeted liquidity provisions. Moreover, the transparency afforded by Shapley explanations mitigates the “black‑box” concern, fostering trust in algorithmic outputs. As financial markets grow more interconnected, the combination of credit‑growth monitoring, yield‑curve analysis, and advanced analytics promises a more resilient financial system.