A hybrid model for carbon prices: Integrating investor attention, mixed-frequency data, quantile regression and deep learning

Expert Systems with Applications - WoS (SCIE), Q1

Author: Trần Kim Phúc, Di Sha, Arne Johannssen, Xianyi Zeng - Viện IAD Trường Đại học Đông Á
Abstract:

Reliable carbon price prediction is crucial for policymakers and market participants to effectively manage risks and make informed investment decisions. However, most existing methods rely on point predictions based on same-frequency data, which often results in information loss and fails to capture the inherent uncertainty of carbon prices. To address these limitations, this study proposes an architecture that integrates investor attention, mixed-frequency data, quantile regression and deep learning techniques. Specifically, the model employs energy factors and the Baidu Search Index as input features, and the mixed-frequency data is processed using a mixed data sampling approach. Then, a combined quantile regression, convolutional neural network and bidirectional long short-term memory model is implemented to generate both point and interval predictions. Empirical findings demonstrate that the proposed integrated architecture outperforms other benchmark models.

DOI: A hybrid model for carbon prices: Integrating investor attention, mixed-frequency data, quantile regression and deep learning - ScienceDirect