sparse-portfolio

This paper proposes an end-to-end decision-focused learning framework for sparse tangent portfolio optimization that replaces discrete asset selection with a smooth top-k operator, enabling gradient flow through prediction and optimization to directly maximize Sharpe ratio.