An efficient reinforcement learning approach for goal-based wealth management.

Goals-based wealth management (GBWM), an investment philosophy, is aiming to attain the desired goal or goals specified by an investor, in a long-term investment. A good algorithm for GBWM is able to provide decision-making and enhance the confidence of consumers and investors, under a complex economic situation. However, the existing methods suffer poor adaptability, oversimplified settings, and limited solution space. In this paper, we extend existing models to more realistic scenarios and propose a new hybrid RL-based algorithm, MHPPO to optimize complex discrete and continuous decisions simultaneously. Our algorithm outperforms existing methods in both simple and complicated settings. The prototype of the algorithm will be implemented on the platform of RoyalFlush company. • A new hybrid reinforcement learning for GBWH with extensive practical features. • New separate neural networks embedding in hybrid policy achieve higher utility. • Student's- T distribution in policy sampling beats other distributions. [ABSTRACT FROM AUTHOR]