Government bond yields reflect investors' expectations regarding future economic conditions, inflation, and monetary policy. The spread between 10-year and 2-year government bond yields has long been regarded as a leading indicator of economic activity, with yield curve inversions frequently preceding periods of recession.
A substantial body of empirical research has documented the relationship between the yield curve and future recessions, particularly for the United States. However, the predictive performance of the yield spread may differ across countries because of variations in financial market structure, monetary policy frameworks, and macroeconomic conditions.
This study investigates whether the 10-year--2-year government bond yield spread can predict recessions approximately one year ahead across five countries: India, the United States, the United Kingdom, Germany, and France. Recession forecasting is formulated as a binary classification problem using the continuous yield spread as the predictor variable. Two supervised machine learning models, Logistic Regression and Random Forest, are evaluated using a time-series cross-validation framework to preserve chronological ordering and avoid look-ahead bias.
The empirical results indicate that the predictive content of the yield spread varies considerably across countries. While the yield spread provides useful forecasting information in some economies, its performance is less consistent in others, suggesting that it should be viewed as one component of a broader recession forecasting framework rather than a universally reliable standalone indicator.
The study contributes to the literature by providing a comparative cross-country evaluation of yield curve forecasting using a consistent machine learning framework. It also highlights the importance of appropriate time-series validation, careful treatment of class imbalance, and cautious interpretation of predictive performance when recession events are relatively rare.