Downloads: 0
Original Research | Computer Science | Volume 15 Issue 7, July 2026 | Pages: 1767 - 1772 | India
A Methodology for Identifying Crisis Mechanisms in Equity Markets: Externally Grounded Labeling, Gradient Boosting, and Synthetic Validation
Abstract: This paper proposes a methodology for classifying financial market stress into four market regimes comprising normal conditions, liquidity shocks, credit events, and systemic contagion using gradient-boosted tree classifiers. The principal contribution is an externally grounded labeling protocol in which crisis labels are defined exclusively from non-equity financial stress indicators before any equity return features are examined, thereby eliminating circular identification between labels and predictors. The study presents the complete mathematical formulation of the gradient boosting objective, SHAP interpretation framework, and walk-forward cross-validation procedure. The methodology is validated using synthetic data generated from label-conditional distributions calibrated to established financial crisis theories (Brunnermeier and Pedersen; Lo). The synthetic experiments demonstrate strong classification performance (macro-averaged F1 of 0.93) and recover theoretically expected feature-importance patterns through SHAP analysis in all three crisis regimes. The proposed framework, including labeling, feature construction, model training, walk-forward validation, and SHAP analysis, is provided as a self-contained, reproducible Python script and is presented as ready for empirical application to real market data.
Keywords: financial crisis typology, labeling protocol, gradient boosting, SHAP, liquidity spirals, synthetic validation
How to Cite?: Arnav Chakole, "A Methodology for Identifying Crisis Mechanisms in Equity Markets: Externally Grounded Labeling, Gradient Boosting, and Synthetic Validation", Volume 15 Issue 7, July 2026, International Journal of Science and Research (IJSR), Pages: 1767-1772, https://www.ijsr.net/getabstract.php?paperid=SR26716040907, DOI: https://dx.doi.org/10.21275/SR26716040907