Leveraging Explainable AI, GRU and VAR Based Framework to Uncover Relationships Between Endogenous and Exogenous Volatility Indices for Effective Trading Strategies

Volatility indices (VIX) represent crucial forward-looking measures of market uncertainty and investor sentiment. The global VIX indices exhibit varied degrees of interdependence, necessitating an understanding of their relationships to assist investors in balancing risk and return effectively. However, the existing literature exhibits two notable limitations. First, the majority of predictive models function as opaque black boxes, offering limited transparency into how specific exogenous volatility indices contribute to forecasted outcomes. Second, no existing work compares the lagged effects captured by statistical models with the feature importance derived from sequence-based neural networks, leaving a gap in understanding the consistency and interpretability of volatility forecasting models. This study addresses these gaps by proposing a novel hybrid framework that integrates Vector Autoregressive (VAR), Gated Recurrent Unit (GRU) neural networks, and Explainable AI (XAI) techniques. The framework captures dynamic spillover effects among seven major VIX indices—including India VIX—using VAR, enhances forecast accuracy with GRU models, and resolves interpretability issues through SHAP values and the SHAP-Lorenz curve, which quantify and visualize each variable's influence. Empirical results reveal the Russell 2000 VIX as the dominant predictor, singularly accounting for the top-ranked influence. The SHAP-Lorenz curve highlights significant inequality in feature contributions, with indices like the CBOE VIX and Gold VIX offering only marginal additional impact. The framework offers practical insights for traders prioritizing the Russell 2000 VIX in hedging strategies and equips market regulators or risk managers with a tool to monitor systemic volatility contagion.