Abstract Taxation remains a critical yet underrepresented dimension in classical portfolio optimization frameworks. While Modern Portfolio Theory and its successors focus on balancing risk and return, they often fail to internalize the after-tax realities faced by investors. This research introduces a dual-layer AI-infused framework that integrates machine learning with decision-theoretic optimization to construct tax-aware mutual fund strategies. The first layer conceptually models a predictive machine learning system that forecasts tax-adjusted returns based on financial attributes such as fund turnover, capital gains realization patterns, and dividend timing. This predictive output informs a second layer prescriptive optimizer that allocates fund weights to maximize investor utility by considering risk aversion, tax brackets, and holding period preferences. The integrated system enables customized and tax-efficient portfolio construction across diverse investor profiles. The framework adapts to evolving fiscal policies and simulates behavior under different regulatory regimes, offering a dynamic and forward-compatible solution. Theoretical analysis and conceptual simulations highlight how the model significantly departs from traditional heuristics by treating taxes not as constraints, but as central variables in portfolio design. Furthermore, the approach reflects contemporary innovations in financial technology, aligning with recent literature on machine learning in asset management, tax-efficient investing, and utility-based decision-making. This study lays foundational groundwork for future empirical applications, particularly in designing intelligent, compliant, and tax-sensitive investment platforms.
Panda et al. (Sat,) studied this question.