Citation
Hassan T. Anis and Roy H. Kwon, “End-to-end, decision-based, cardinality-constrained portfolio optimization,” European Journal of Operational Research, 320(3), 739-753, 2025.
Abstract
This paper applies an integrated end-to-end predict-and-optimize framework to cardinality-constrained portfolio optimization. Instead of estimating factor-risk parameters in a separate prediction step and then constructing the portfolio, the model learns the parameters through a downstream sparse portfolio decision. Because the original nonlinear mixed-integer feasible region is discontinuous, the authors compare continuous relaxations of different tightness as implicit neural-network layers. On real financial data, the decision-based model outperforms decoupled baselines across cardinality levels, while also showing that relaxation tightness, problem structure, and experimental design interact in nontrivial ways.