A Multilayer Network and Machine-Learning Framework for Measuring Systemic Risk in the New Energy Industry: PageRank, XGBoost, and SHAP
Description
This dataset supports the study of systemic risk transmission in China’s new energy sector through a firm–municipality–industry-chain multilayer network. It contains daily stock-return data for 15 listed new energy firms, market-return inputs, a monthly systemic-risk and macroeconomic panel, firm business classifications, industry-chain association matrices, municipality-level GDP data for 2025, and map-software-based intercity distance measures. Firm-level financial data were derived from CSMAR inputs; monthly macroeconomic indicators were obtained from the CEI Statistical Database; business classifications follow the Industrial Classification for National Economic Activities; and the city gravity-network inputs combine GDP and distance information. The accompanying Python scripts construct the systemic-risk index, estimate firm-level Granger-causality networks, city gravity networks, industry-chain networks, multilayer PageRank measures, XGBoost prediction models, SHAP-based feature attributions, and robustness tests. The research hypothesis is that systemic risk is transmitted across firm, regional, and industry-chain layers, and that both lagged risk conditions and macroeconomic variables help explain changes in sectoral risk. The reproduced results show that the firm layer carries the largest aggregate PageRank weight, while Kunming, downstream industry segments, and selected firms emerge as prominent network nodes. The XGBoost results provide out-of-sample risk predictions and identify lagged risk measures, financial conditions, and macroeconomic indicators as influential predictors. Users should run python run.py --check after installing the listed dependencies, then execute the required tasks through run.py; all results are written to task-specific folders under outputs.