Using the Entirety of the Term Structure to Forecast Recessions

Abstract

Using a logit model and quarterly data from 1962 to 2021, we test the forecasting power of the yield spread, a popular leading indicator, and show that forecasting models that include the entirety of the term structure of interest rates provide more accurate estimates of future economic downturns. We also show that models that only include the yield spread are implicitly imposing restrictions in the coefficients of the model resulting in lower predictive power and omitted variable bias issues.