FlashFolio: A GPU-Accelerated Solver for Portfolio Optimization
This study addresses the severe computational challenges in large-scale single- and multi-period portfolio optimization when incorporating factor risk models, bid–ask spreads, and nonlinear market impact, particularly the inefficiency and instability of solving multi-period problems. The work presents the first efficient application of GPU parallel computing to multi-period portfolio optimization with nonlinear market impact, integrating customized numerical algorithms to construct a quadratic and nonlinear programming solver capable of handling complex real-world constraints. Compared to the commercial MOSEK solver, the proposed method achieves speedups of up to 12.9× in single-period settings and 48× in multi-period scenarios, while substantially improving solution robustness and accuracy on difficult problem instances.