China Carbon Emission Trading Market Multivariate Time Series Dataset (2014–2022)
Description
This dataset contains raw multivariate time series data from three pilot carbon markets (Beijing, Hubei and Shenzhen), as well as a unified result dataset with completed missing value imputation. All data are stored in Excel format and can be directly loaded for subsequent analysis and modeling without additional missing value imputation procedures. The accompanying code retains scripts for cross-market correlation and nonlinear causality analysis, as well as the full implementation of the proposed MIMO-BiLSTM-Attention-GARCH integrated forecasting framework. This dataset supports relevant research on multivariate time series joint representation learning, cross-market linkage modeling and carbon price forecasting algorithm validation.
Files
Steps to reproduce
1. Data Preparation Place all Excel files (Beijing Raw Data.xlsx, Hubei Raw Data.xlsx, Shenzhen Raw Data.xlsx, Carbon Market Imputed Missing Data Result.xlsx) and Python scripts (correlation_analysis.py, Main_Code.py) in the same project directory. The missing-value-imputed dataset has been fully preprocessed and can be loaded directly without additional imputation procedures. 2. Environment Configuration JetBrains PyCharm is recommended as the IDE with Python 3.12 as the interpreter version. Before execution, install the following third-party dependencies via pip in the PyCharm virtual environment or system environment: Basic data processing: pandas, numpy, openpyxl Signal processing and scientific computing: scipy, PyWavelets (imported as pywt in code) Machine learning utilities: scikit-learn (imported as sklearn in code), xgboost, optuna Time series econometric analysis: statsmodels, arch Deep learning framework: tensorflow Visualization: matplotlib 3. Reproduce Cross-market Correlation and Nonlinear Causality Analysis Open the project in PyCharm, set the correct working directory, and run the correlation_analysis.py script. The script reads the preprocessed multi-market carbon price dataset, performs stationarity tests on time series, calculates pairwise Pearson correlation coefficients, and conducts Granger causality tests. Outputs include correlation matrices, significance test results, and corresponding visualization plots. 4. Reproduce the MIMO-BiLSTM-Attention-GARCH Forecasting Framework Run the Main_Code.py script in PyCharm. The script loads the multivariate time series dataset, and sequentially performs wavelet denoising, data normalization, and recursive feature elimination. Training and test sets are split via a sliding window strategy. The proposed MIMO-BiLSTM-Attention-GARCH integrated forecasting model is built and trained, with the XGBoost benchmark model and GARCH volatility modeling integrated. Optuna-based hyperparameter optimization is supported. Finally, prediction results and evaluation metrics (MAE, RMSE, R²) are generated for model performance validation. 5. Result Verification All outputs from the above scripts are consistent with the empirical analysis sections of the corresponding research paper, covering cross-market linkage characteristics analysis and multi-model forecasting performance comparison.
Institutions
- Chengdu University of TechnologySichuan, Chengdu