Explainable AI models for portfolio risk assessment: Bridging the gap between black-box predictions and fiduciary transparency

The financial industry's rapid adoption of artificial intelligence (AI) for portfolio risk assessment creates a tension between predictive power and explainability—a critical concern for fiduciary obligations. This research explores the implementation of explainable AI (XAI) methodologies in portfolio risk management, comparing traditional black-box models with transparent alternatives. We evaluate LIME, SHAP, and rule-based explainers against complex neural networks and ensemble methods across diverse market conditions using a comprehensive dataset of financial instruments. Results demonstrate that XAI approaches can achieve 92.4% of black-box performance while providing interpretable insights, satisfying regulatory requirements without significant accuracy sacrifices. Our framework integrates explainability metrics with performance indicators, creating a balanced scorecard for model selection in financial contexts. This work addresses the growing demand for transparent AI in financial decision-making, offering practical guidance for portfolio managers navigating the intersection of advanced analytics and fiduciary responsibility.