SIGNAL-TO-NOISE ANALYSIS OF CRISIS INDICATORS IN GLOBAL FINANCE USING ARTIFICIAL INTELLIGENCE

Abstract

Financial stress indicators are designed to flag systemic risk before crises unfold, yet their informational clarity is often obscured by high market volatility and overlapping noise from global macroeconomic shocks. This study addresses that challenge by introducing an artificial intelligence–driven signal-to-noise analysis framework that disentangles meaningful crisis signals from stochastic fluctuations in major financial indices. Using a unified weekly panel spanning 1990–2025 that integrates the VIX (market volatility), the FRED STLFSI4 (U.S. financial stress), and the ECB CISS (European systemic stress), we evaluate multiple definitions of signal strength, deterministic, event-aligned, spectral, predictive, and information-theoretic. The framework fuses econometric decomposition, spectral analysis, and machine learning–based predictive modeling, including Random Forests and XGBoost, with SHAP-driven explainability to quantify both the magnitude and interpretability of crisis signals. Results reveal distinct signal dynamics across markets: CISS exhibits strong low-frequency structural coherence, STLFSI4 responds sharply to systemic shocks, and VIX encodes transient volatility bursts. Predictive signal-to-noise ratios peak consistently in pre-crisis windows, validating the framework’s capacity for early-warning detection. SHAP-based interpretation further exposes cross-market lead–lag dependencies, illustrating how European stress patterns anticipate U.S. volatility surges during contagion phases. Collectively, the findings demonstrate that integrating deep signal-processing and explainable AI can transform noisy financial indicators into transparent, data-driven tools for real-time systemic risk monitoring and policy decision support.