Answers are grounded in this paper and cite their sources.
Compares credit rating effectiveness in predicting yields and defaults, indicating market-based measures are superior for pricing credit risk.
Using a sample of 4,230 new corporate bonds issued by US industrial firms from 2001 to 2020, we compare the performance of issuer-paid credit ratings from Standard & Poor’s with two subscriber-paid alternatives: Egan-Jones Ratings Company (EJR) and Bloomberg Default Grades. This study provides a unified empirical setting to evaluate these measures across four distinct financial outcomes: offering yields, yield spreads, actual market credit default swap (CDS) spreads, and realized default events. To formally compare their relative explanatory power and address model selection, we employ the non-nested likelihood ratio test of Vuong (1989). Our results reveal a significant performance flip between pricing and default prediction. Bloomberg’s market-based structural model consistently dominates traditional credit ratings in explaining yield and CDS spreads, reflecting its superior ability to incorporate real-time equity and volatility data. Although S&P ratings remain influential in determining offering yields—likely due to access to nonpublic information—they exhibit lower explanatory power than Bloomberg, with EJR ratings performing the weakest across spread specifications. In contrast, when predicting realized default events, the relative rankings shift: S&P outperforms EJR in identifying terminal default risk, although the Bloomberg model remains the most robust overall predictor. For investment-grade bonds, all measures lose significant predictive power for defaults due to the extreme rarity of such events in high-quality segments. Overall, the findings suggest that market-based structural credit risk measures provide a statistically superior fit for pricing credit risk; traditional ratings continue to provide valuable information regarding realized default outcomes.
No takes yet. Share an insight, caveat, or question.
Fairchild et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: