The Evolution of Portfolio Theory Under Risk & Uncertainty: From Mean–Variance to AI‐Augmented Investing

ABSTRACT

This article summarizes the evolution of portfolio theory from mean–variance optimization to AI‐augmented investment systems. Rather than treating portfolio models as isolated techniques, it organizes the literature as a sequence of responses to different forms of uncertainty: variance, systematic risk, expected‐return estimation error, downside risk, tail loss, parameter ambiguity, risk‐budgeting instability, conditional volatility, regime dependence, high‐dimensional prediction, and AI governance. Combining a systematic literature review, public‐metadata bibliometric analysis, and thematic synthesis, the article covers leading academic journals and practitioner‐oriented outlets, including JF, JFE, RFS, RF, JFQA, Management Science, FAJ, JPM, FMPM, and JAM. The survey highlights how classical portfolio theory, CAPM and multifactor allocation, Bayesian/Black–Litterman models, CVaR, robust optimization, risk parity, volatility management, regime‐switching allocation, parametric portfolio policies, machine learning, and LLM‐based agents form a cumulative intellectual architecture.

The article is designed as a compact roadmap for doctoral students, academic researchers, and quantitative researchers in hedge funds and asset management who seek a rapid but rigorous understanding of how portfolio theory has evolved and how its core problems remain relevant in the AI era.