Abstract
This paper investigates the predictive modelling of the credit spread term structure by incorporating macroeconomic uncertainty indicators and statistical learning methodologies. We propose a new penalty loss function designed to enhance the economic interpretability and temporal stability of the estimated Nelson-Siegel credit spread parameters. We predict these parameters by extending the set of features generally used in the literature to include macroeconomic and financial uncertainty measures. We compare linear regression models with ensemble-based machine learning techniques. Empirical analysis using SHapley Additive exPlanation (SHAP) values show the influence of firm-level variables and reinforce the link between economic uncertainty and credit spreads. Boosted trees outperform other approaches to predict the term structure of credit spread.

This publication has 36 references indexed in Scilit: