Pre-Conference Talk by YOUNG Ezekiel Ong | GuidedRec: Behaviourally Constrained Portfolio Recommendation with Certified Slate Optimisation

Financial asset recommendations must balance what an investor is likely to accept against how an asset affects their portfolio. A stock that matches an investor’s preferences can increase risk, while a valuable diversifier may be behaviourally implausible. We introduce GuidedRec, a holdings-aware framework that first identifies eligible, currently unheld stocks through a behavioural gate calibrated using only past data. It then jointly chooses a fixed number of additions to maximise each investor’s risk-adjusted utility for the resulting equally weighted portfolio, accounting for interactions among new and existing holdings. 

We show that when the estimated covariance matrix has a factor-plus-isotropic structure, its isotropic component does not affect which fixed-size slate is optimal. This enables a compact factor-based formulation and rigorous search bounds. Our Ledoit–Wolf covariance estimator satisfies this condition exactly. GuidedRec uses a branch-and-bound solver that either proves global optimality or reports a feasible recommendation with a certified optimality gap. 

We evaluate GuidedRec across 20 prespecified recommendation dates, non-overlapping 30-trading-day evaluation windows, and 16,850 matched investor-date observations. Compared with preference-only and leading portfolio-aware baselines, GuidedRec achieves the largest observed date-balanced mean gains in realised portfolio utility and risk-adjusted returns, accounting for behavioural deployability. Every recommendation respects its behavioural gate, and results remain leading across gates derived from different preference models. The solver certifies all main optimisations and matches exhaustive search where feasible. Ablations show that removing the gate raises predicted utility but harms realised outcomes, while slate-level optimisation and personalised risk aversion materially shape recommendations.

This is a Pre-Conference talk for 7th ACM International Conference on AI in Finance (ICAIF 2026).

Ezekiel YOUNG is a PhD candidate at the School of Computing and Information Systems, Singapore Management University, supervised by Prof. Hady W. Lauw. His research focuses on the intersection of economics and computation, with applications to product bundling, competitive markets, and portfolio recommendation.