From Black Box to Glass Box: Trustworthy AI for Live Quantitative Investment

The mainstream adoption of machine learning in investment management has created a widening gap between predictive performance and practical trustworthiness. Models that achieve impressive backtest results frequently fail in live deployment due to overfitting, regime sensitivity, or a lack of the transparency required by risk officers, compliance teams, and regulators. Unlike earlier phases of AI adoption in finance, which focused on whether these models could outperform traditional methods, the central challenge today is whether they can be trusted, explained, and safely operated in production environments where real capital is at stake.This Research Topic addresses that challenge directly. It brings together researchers, quantitative analysts, and financial engineers working at the intersection of explainable AI, deployment-grade robustness, and the emerging use of large language models (LLMs) and alternative data for investment signal generation. The aim is to build a definitive body of work on how AI-driven investment systems are made interpretable, how their reliability is validated beyond standard backtesting, and how they are governed once deployed. Thematic emphasis sits firmly on methodological rigour, transparency, and operational accountability, not on forecasting accuracy alone.This Research Topic does not duplicate work on broad AI applications in econometrics, general asset pricing methodology, or classical algorithmic trading. It is specifically concerned with the trustworthiness, interpretability, and governance of AI systems at the point of production deployment in investment and portfolio management contexts.We welcome original research, reviews, and methodological contributions addressing, but not limited to, the following themes:· Explainable AI (XAI): interpreting deep learning in portfolio contexts.· SHAP and surrogate modelling: applying interpretability tools to trading signals.· Market robustness: adapting to distribution shifts and regime changes.· Backtesting rigour: preventing overfitting with statistical corrections.· LLMs and alternative data: extracting signals from unstructured text.· Model governance: managing risk and compliance for live deployment.· Hybrid decision architectures: combining algorithms with human oversight. Topic editors: Wei Li, Zhonghua Li, Michael Chak Sham Wong Source: https://www.frontiersin.org/research-topics/81968/from-black-box-to-glass-box-trustworthy-ai-for-live-quantitative-investment