Skip to content
Book Open access

Predicting Decision-Time Reliability of Equity Market Risk under Delayed Inputs and Non-Stationary Conditions

Nov 2026 · 0 citations · 21 references

Abstract

Value-at-Risk (VaR) models are standard in automated risk control, yet operational pipelines consume risk reports with a one-day reporting lag while market regimes shift faster than the estimator adapts. This mismatch creates a decision-time gap in which a nominally current risk estimate is stale. We formalise the Decision-Time Reliability problem and propose a predictive monitoring layer that estimates next-day failure probability using only decision-time information. Leave-one-out ablation and SHAP analysis independently confirm that risk-volatility divergence is the dominant predictive signal—dominant by a 2 × margin in mean absolute SHAP value—and that removing noisy features improves XGBoost ROC-AUC from 0.642 to 0.685. We compare four models across six global equity markets: S&P 500, FTSE 100, Nikkei 225, DAX, Hang Seng, and EM ETF. A reliability-gated three-level exposure policy reduces Expected Shortfall Breach (ESB) by 49–69% consistently across all six assets. Our novel Regime-Conditional Reliability Estimator (RCRE) achieves the best tail-severity reduction (50% ESB reduction on S&P 500) via soft Gaussian regime mixing. The COVID-19 volatility shock, used as a fully held-out stress test, confirms early warning capability during the acute mid-March crash window.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.