Bioenergy market predictions using AI: Integrating climate change and green finance

This study aims to improve the accuracy of bioenergy return prediction by applying advanced deep learning models. The volatility of the bioenergy market and its susceptibility to multiple influencing factors necessitate the use of advanced methods to improve prediction accuracy and reliability. Therefore, this study employs 1965 daily data points sourced from the Bloomberg database, covering the period from April 1, 2015, to October 19, 2022. We utilized six hybrid deep learning models: LSTM, GRU, CNN-BiLSTM, CNN-BiLSTM-Attention, INFO-CNN-BiLSTM, and NRBO-BiLSTM-Attention. These models were applied to predict global bioenergy returns, using the Global Climate Policy Uncertainty Index, Global Green Bond Index Returns, Arctic Oscillation Index, and the Nasdaq BioClean Fuel Index as inputs. The forecast horizons were set to 5 days, 10 days, and 15 days. Model performance was assessed using Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and MCS testing. Our findings indicate the following: (1) Although the INFO-CNN-BiLSTM model exhibited the smallest prediction error, the NRBO-CNN-BiLSTM-Attention model's predictions were the closest to the actual values, and Multiple Comparison Testing confirmed its superior performance in predicting bioenergy returns; (2) Across the 15-day, 10-day, and 5-day horizons, the NRBO-CNN-BiLSTM-Attention model consistently improved prediction accuracy and overall performance. These findings contribute to the optimization of global energy structures and the promotion of sustainable energy development and application.