Answers are grounded in this paper and cite their sources.
As asset prices evolve, portfolio weights tend to diverge from their target weights at the expense of market impact costs, and we need to periodically rebalance our portfolios to keep them on target. Excessively frequent rebalancing can shrink net-of-trading-cost returns while too infrequent rebalancing can risk unintended risk exposure. Prior work uses rising artificial intelligence and machine learning techniques to optimize this sequential decision problem. This systematic review of the literature, conducted according to PRISMA 2020 guidelines with Science Direct, Springer Link, and IEEE Xplore as databases, begins with 939 identified records and filters away studies with the criteria that the dynamic rebalancing decision itself must be modeled explicitly, leaving twenty for full-text review. Performance reporting assessment shows the Sharpe ratio appears in thirteen of twenty studies, six studies display no risk-adjusted measures, and no two studies display identical measured metrics, yet the published Sharpe ratio varies from 0.08 to 8.83, and above twenty in hourly panels, with no implementation disclosure of annualization method. Regarding technical challenge quantification, fifteen of twenty present point estimates with no seed dispersion, only one models lot-size constraints, and none models market impact costs endogenously. Assumed transaction costs vary from 1 to 125 basis points per trade, and up to 150 basis points round trip, with no empirical justification, and no studies jointly alter rebalancing frequency and transaction rate within a bivariate sensitivity design. Finally, this literature confuses two distinct mechanics-restoration of portfolio weights toward target versus global reselection of asset pools-yielding conflicting effects on optimal rebalancing frequency. This review presents a unique taxonomic nomenclature of six families of techniques, proves that these two operations are operationally distinct, yet yield opposite rebalancing frequency effects, and proposes a minimum set of experimental results to be consistently reported going forward. So, when it comes to advancements in AI-driven portfolio rebalancing, compromises on experimental design and on uniformity of reporting metrics prove more of a constraint than the complexity of the algorithms.
No takes yet. Share an insight, caveat, or question.
Mehran Zahoor (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: