Continuous-time mean-variance portfolio selection integrating greenwashing effect and stochastic horizon: A reinforcement learning framework

This paper develops a reinforcement learning (RL) framework to investigate asset pricing distortions caused by corporate greenwashing and investor investment horizon uncertainty. First, corporate greenwashing behavior and stochastic investment horizons are incorporated into a continuous-time meanvariance portfolio model, which balances exploration and exploitation through entropy regularization. The Hamilton-Jacobi-Bellman (HJB) equation for the value function of the self-financing portfolio problem is derived, and we demonstrate that the optimal feedback policy follows a Gaussian distribution with time-decaying variance. Subsequently, the policy improvement theorem and the policy convergence theorem are proven, which leads to the design of a practical RL algorithm. Finally, numerical simulations are conducted to examine the behavioral characteristics of investor portfolio decisions and to validate the algorithm’s convergence and effectiveness. The results indicate that when greenwashing remains undetected, a higher degree of greenwashing results in a more pronounced overvaluation of risky assets, and early investment termination by investors lowers expected returns. Following the exposure of greenwashing, a higher degree of greenwashing induces a greater decline in returns due to the collapse of the “greenwashing bubble”, whereas early termination can help mitigate the accumulation of risk.