AC-DQN

Abstract

This paper proposes a framework called AC-DQN that combines Mixed Integer linear programming models with Deep Q-Network (DQN) learning to provide an effective approach for personal asset management. ALM models, developed to manage investment strategies considering future liabilities, are well-suited for addressing the specific financial goals and constraints of individuals. The AC-DQN framework extends value-based reinforcement learning to handle continuous action spaces by leveraging MIP representation of networks. This enables the inclusion of action-related equations as constraints, allowing for optimal investment decisions considering goals and constraints. Experimental results demonstrate the effectiveness of the proposed approach in personal asset management, considering goals and constraints.