Firms increasingly develop dynamic pricing policies to maximize revenue for perishable products with limited inventory over a finite selling horizon. This trend is enabled by the growing availability of sales data and is observed across industries such as airlines, hotels, cruise lines, fashion, and seasonal retail. Given customer heterogeneity, firms may further adopt discriminatory pricing across customer groups. However, excessive price disparities can trigger legal risks and consumer backlash, motivating price fairness constraints that bound inter-group price differences in each selling period. We formulate this problem as an action-constrained Markov decision process (ACMDP) with unknown demand functions and adopt a model-free deep reinforcement learning (DRL) framework. However, standard DRL algorithms for unconstrained MDPs cannot directly handle these fairness constraints. Therefore, we introduce an optimization-based shielding mechanism. From the DRL pricing agent’s perspective, this mechanism converts the ACMDP into a shield-induced unconstrained MDP. Meanwhile, it guarantees constraint satisfaction for all executed prices. Building on this framework, we propose the Shield Soft Actor-Critic (Shield-SAC) algorithm. This is the first Shield-SAC method for fairness-aware pricing under instantaneous and hard price fairness constraints. We test it in two simulated markets of different scales and validate that Shield-SAC achieves strong revenue performance while consistently enforcing the price fairness constraints during both training and deployment.
Qiao et al. (Mon,) studied this question.