Upgrading Credit Pricing and Risk Assessment through Embeddings

Abstract

Credit ratings are central in fixed income markets, defining mutual fund benchmarks and risk-based capital regulation of insurance companies. Credit ratings explain a large share of the variation in corporate credit spreads, but a surprising amount of variation remains across firms. Asset pricing theory implies that equilibrium bond prices depend not only on credit ratings but all information that investors use to form their portfolios. We extract a high dimensional representation of this information, called firm embeddings, from US corporate bond holdings of mutual funds and insurance companies. Within broad credit rating categories, firm embeddings explain credit spreads and the volatility of credit spreads better than credit ratings and the distance to default. Therefore, firm embeddings can augment (and eventually replace) credit ratings to provide more timely and accurate information for fixed income markets. We illustrate the potential impact of an improved rating system on the risk-based capital regulation of insurance companies.

Keywords: Artificial intelligence, Asset pricing, Credit rating, Credit spread, Machine learning, Risk-based capital regulation

JEL Classification: G12, G22, G23