Data-Driven Credit Risk Assessment in U.S. Public Finance: A Review of Rating-Based, Market-Based, and Machine Learning Approaches Under Sparse Default and Delayed Disclosure

Credit risk assessment in U.S. public finance faces two persistent constraints. Conventional predictive validation is limited by the rarity of municipal defaults and financial disclosure may not be timely with underlying fiscal conditions. This review compares three approaches to municipal credit risk assessment: rating-based, market-based, and machinelearning approaches. They are assessed using a common framework that includes information basis, reliability, timeliness, predictive usefulness, assessment purpose, constraint adaptation and institutional maturity. Rather than applying each paradigm’s conventional evaluation standard, the review holds the two constraints constant and examines how each approach manages the resulting trade-offs. Ratings rely on institutional judgment to include information beyond structured data but have raised reliability issues with agency incentives. Market signals update continuously but also reflect liquidity, tax treatment, and dealer behavior. Machine learning models can use broader and earlier information sources but are severely constrained in their ability to be validated due to the sparsity of default labels. Across all three paradigms, a central distinction emerges between assessing ongoing creditworthiness and predicting the rare event of default. This distinction helps explain why apparent weaknesses may reflect a mismatch between method and assessment purpose rather than methodological failure. The review contributes a structured comparison of these paradigms around the joint constraints of sparse defaults and delayed disclosure, which existing research has largely examined in isolation.