Beyond Alpha: A Verifiable Framework For Quantifying The Risk Management Value of Interpretable Ai In Emerging Financial Markets

The deployment of artificial intelligence in financial markets, particularly in volatile emerging economies, presents both an opportunity for growth and a significant systemic risk. This research confronts this duality by introducing and validating a novel, end-to-end, and fully reproducible framework for creating and, more importantly, rigorously evaluating interpretable machine learning (IML) models. We train and optimize a suite of IML models—including regularized ElasticNet, constrained Decision Trees, and Explainable Boosting Machines (EBM)—on an extensive, raw historical dataset of the S&P/BVL Peru General Index (SPBLPGPT) from its inception (2011-2025). Our key contribution is the deployment of a sophisticated validation battery that moves beyond standard metrics. While a formal Model Confidence Set (MCS) test reveals that a passive Buy & Hold strategy was statistically superior in terms of average daily returns (p > 0.9), our analysis uncovers a critical paradox: a Second-Order Stochastic Dominance (SSD) test demonstrates that the EBM strategy is formally preferable for any rational, risk- averse investor. We resolve this paradox through a suite of Quantitative Explainable AI (CXAI) techniques, proving that the EBM’s value lies not in traditional alpha generation, but in its superior tail risk management, successfully hedging against the most severe market downturns. By providing statistical proof (p = 0.0054) of the model’s dynamic, context-aware logic and ensuring complete scientific reproducibility via a FAIR-compliant pipeline, this research offers a verifiable paradigm for building trustworthy financial AI that contributes to the stability and sustainability of emerging financial ecosystems.