Yiyuan Wang

University at Buffalo · Economics · 2026–27 job market

Fields

Finance, Macroeconomics

Job market paper

News Narratives and Omitted Credit-Cycle States in Corporate Bond Markets

Standard credit-risk variables leave a broad common residual component in corporate bond markets. I ask whether topic-level narratives from more than four decades of Wall Street Journal business news help measure these omitted credit-cycle states. Using latent Dirichlet allocation (LDA) for historical diagnostics and a strict train-only online LDA (oLDA) design for real-time tests, I construct monthly narrative-attention measures and relate them to credit spreads and returns across rating segments. After removing the component of monthly spread changes explained by standard rates, slope, equity-market, and option-implied controls, the first principal component of the standardized A, BBB, and high-yield (HY) residuals explains about 76 percent of their variation and assigns similar weights across ratings. On the 408-month text-residual sample, full-sample text mappings are strongest in risky credit. A sparse diagnostic yields a nonempty active set only for HY, while a complementary full-topic Ridge mapping produces a monotonic diffuse-fit gradient from A to BBB to HY. Real-time forecasting evidence is smaller: a spread-based risk filter accounts for most of the HY timing improvement, while the narrative overlay adds a modest increment. The evidence therefore separates a broad omitted credit state from a rating-specific narrative mapping that is strongest in risky credit.