Global supply chains are sensitive to China's economic sectors, yet direct, model-based evidence that Chinese sectoral stock returns forecast global supply-chain distress is limited. We turn the typical analysis on its head by predicting Global Supply Chain Pressure Index (GSCPI) from market data rather than GSCPI as an explanatory variable. We use 233 months of China's 10 sectoral stock returns to predict GSCPI a month ahead using a cast of deep learning as well as traditional machine learning models: a Feedforward Neural Network (FFNN), a Gated Recurrent Unit (GRU), Ridge regression, ARX (1), Random Forest, and XGBoost, along with an ensemble combining Ridge and FFNN. Models are compared on out-of-sample (OOS) R².

The combined ensemble achieves maximum accuracy (out-of-sample R² ≈ 0.85) and comprehensively dominates individual models; the FFNN as a solo model is the runner-up. Meanwhile, the ARX (1), GRU, and XGBoost models have negative R² and thus demonstrate limited generalization on this dataset. These findings imply that blending linear and nonlinear learners best extracts cross industry signals that predict subsequent GSCPI movements. Though results are particular to China's setting, this new approach provides an actionable early-warning device against worldwide supply chain disruption by casting real-time financial signals into forecast of stress.