Transfer Prediction for the Price Volatility of Carbon Trading with Hybrid Gated Recurrent Unit

Carbon trading is a market-based mechanism for reducing greenhouse gas emissions that provides economic incentives for mitigating climate change and promotes the development of a low-carbon economy. However, China's carbon trading market is still only in the early stages of its development. The late establishment of the trading mechanism in this emerging market has led to limited data availability for deep learning modeling. Consequently, accurately predicting the price volatility in China’s carbon trading market is a challenging task. To address this issue, we propose a hybrid model that integrates generalized autoregressive conditional heteroskedasticity (GARCH) and gated recurrent unit (GRU) to predict the volatility of carbon price. A transfer learning (TL) model is developed based on the hybrid (baseline) model to achieve comparable prediction accuracy to ordinary deep learning but with a significant reduction in required training data. The effectiveness of the TL model is verified through an ablation study method. Furthermore, we propose a new factor to measure the transferability of TL, enabling us to verify the effectiveness of TL before actual modeling and provide relevant guidance for time series data selection of source domains. Finally, we present the empirical results based on actual data to demonstrate the superiority of the proposed transfer learning framework in predicting carbon trading price volatility as well as the effectiveness of the proposed transferability measurement factor