基于GATv2-RF-SHAP模型的系统性金融风险预警研究OA

Effective early warning of systemic financial risk is crucial for maintaining financial security.This study integrates Random Forest algorithm,Factor Analysis,and an improved Graph Attention Network ﹙GATv2)to develop a more effective sys-temic financial risk early warning model.Based on seven dimensions—the macro economy,stocks,bonds,exchange rates,com-modities,interest rates,and financial institutions—16 indicators are selected to construct a systemic financial risk warning system for China.By combining GATv2 with factor analysis,an efficient Financial Stress Index is extracted,overcoming the path depen-dence of traditional factor models on linear assumptions.GATv2's multi-head dynamic attention mechanism uncovers nonlinear relationships between indicators,while the Random Forest algorithm determines the importance of each indicator.The SHAP model is used to visualize the feature contribution of different indicators to risk.The results show that the Random Forest algorithm per-forms well in risk prediction.In the risk warning indicator system,short-term and medium-to-long-term loan interest rates,bond market custody balance,and the loan-to-deposit ratio emerge as key risk drivers,with short-term loan interest rates negatively in-fluencing systemic financial risk.Based on these findings,recommendations for risk prevention and control are provided to support financial security decision-making.