Forecasting nonlinear green bond yields in China: Deep learning for improved accuracy and policy awareness

Abstract

This study develops a convolutional neural network bidirectional long short-term memory model with an attention mechanism to forecast yields in China’s green bond market. The model incorporates macroeconomic indicators, financial variables, policy factors, and issuer heterogeneity to enhance predictive accuracy. Empirical results show the model outperforms traditional approaches in point forecasting. It also offers superior robustness under identical confidence levels, increasing its utility for risk management and policy assessment in green finance. It is a practical tool for regulators, investors, and issuers.