Photonic Quantum Computing vs. Classical Solvers in Constrained Factor Portfolio Optimization

Battle Count: 77

Highly relevant to quantitative trading and institutional portfolio management. The paper directly addresses constrained factor portfolio optimization with realistic transaction costs, turnover penalties, and higher-moment risk shaping. It provides actionable mandate-specific guidelines for deploying advanced optimization engines. The findings on quantum vs. classical solver trade-offs, RL failure modes, and concentration dynamics are directly applicable to live production systems. The 164-month out-of-sample evaluation on a well-known factor library adds practical credibility.

High implementation complexity due to: (1) proprietary photonic quantum hardware access (Dirac-3 API with queue-based scheduling), (2) commercial MIP solver licensing (Gurobi), (3) deep reinforcement learning training infrastructure for SAC, (4) QUBO formulation and data-to-portfolio pipeline with 6 stages, (5) 48-configuration hyperparameter sweep requiring significant compute, (6) Ledoit-Wolf shrinkage covariance estimation, (7) exponential smoothing and weight reconstruction logic. The SAC pipeline is most accessible; Dirac-3 requires specialized hardware access.

The study uses the publicly available Jensen-Kelly-Pedersen 13-factor equity library and well-known solvers (Gurobi, SAC). However, the Dirac-3 photonic quantum annealer is proprietary hardware from Quantum Computing Inc. with queue-based access, limiting full reproducibility. Cross-seed variance is reported (3 seeds for Dirac-3, 5 for Gurobi and SAC). The 48-configuration sweep and detailed hyperparameter grids are specified. No code repository is mentioned.