A multiscale analysis of spillover effects and price forecasting in China's carbon market

As China transitions from regional carbon trading pilots to a unified national Emission Trading Scheme (ETS), understanding spillover effects between China's national and pilot ETSs and accurately predicting national carbon price becomes vital. However, existing studies often overlook multi-scale spillover effects, and rarely embed causal information into prediction models. To address these gaps, this study proposes a causality-guided multi-scale forecasting framework that integrates pilot markets' data, applies Multivariate Empirical Mode Decomposition (MEMD), and incorporates machine learning models to analyze spillover dynamics and predict national carbon price. The results show that: (1) at the raw data level, no Granger causality relationship is observed between national and pilot ETSs; (2) markets appear uncorrelated at short time scales (2–4 days), exhibit strong bidirectional spillover effects at intermediate scales (approximately one week to one year) likely driven by major events and policy changes, and demonstrate a clear linear trend relationship at long-term scales; (3) incorporating pilot market information significantly improves prediction accuracy, reducing Mean Absolute Percentage Error (MAPE) by approximately 0.10%–37.98%. Overall, pilot ETSs offer valuable signals, and the causality-guided multi-scale forecasting framework helps manage the complexity of carbon price prediction. The findings offer insights into the development and improvement of China's national ETS and enhance the effectiveness of carbon price forecasting.