Abstract: This study predicts the term spread using various machine learning models. Given that numerous macroeconomic variables can be used for term spread prediction, 116 variables are considered, and key variables are selected and extracted using LASSO. The core of the research lies in comparing two methodologies for predicting the term spread. The first method involves directly forecasting the spread itself, while the second method predicts long-term and short-term yields separately and then generates the spread from those predictions. The results indicate that the approach of directly predicting the term spread is statistically significantly superior. Our analysis of various forecasting models reveals that Long Short-Term Memory (LSTM), which can effectively capture nonlinear characteristics, demonstrates particularly strong performance in financial time series forecasting. These findings provide an effective approach to predicting the term spread and may serve as an important foundation for future research.

Keywords: Forecasting; LASSO; Long Short-Term Memory; Machine learning; Term spread

JEL codes: C45, E43, G17

DOI: ...