Machine Learning Predictions of Credit and Equity Risk Premia

Abstract

The emergence of algorithmic high-frequency trading in the market for credit risk affords accurate inference of new risk measures. When combined with machine learning predictive methods, these measures forecast substantial future changes in firms' credit and equity risk premiums in out-of-sample. Parallel measures estimated from firms' stocks fail to predict risk premiums, indicating that credit-market-based risk measures contain valuable information for forecasting firms' risk premia in both markets. The innovative high-volume high-frequency trading has not alleviated short-horizon pricing deviations across firms' equity and credit markets, an epitome of latent arbitrage in the market for credit risk.