Similarity-Based Recession Prediction in Different Interest Rate Environments

We develop a flexible nonparametric similarity-based approach to predict the state of the business cycle in different interest rate environments. Our approach provides methodological advantages over parametric logit and probit models and new empirical perspectives on the usefulness of the term spread as the main leading indicator and its connection to the prevailing interest rate level. Empirical results on the U.S., euro area and Japan show that the predictability of business cycle regimes depends on not just the term spread but also monetary policy conditions measured by the level of the short-term interest rate