Reinforcement Learning Approach for Dynamic Pricing

Abstract

With the introduction of digital technologies, it becomes easier for customers to compare prices and choose the product that is most profitable for them, this leads to instability of demand, which means that there is a need for market players to review pricing policies in favor of one that could take into account the characteristics of producer’s resources and current demand status. Dynamic pricing seems to be an adequate solution to the problem, as it is adaptive to customer expectations. In addition, with the digitalization of the economy, unique opportunities arise for using this apparatus. The purpose of this study is to evaluate the possibility of applying the concept of dynamic pricing to traditional retail. The goal of solving the dynamic pricing problem in the framework of this study is to maximize profits from the sale of a specific associated product at an automatic gas station. To solve this problem, the authors propose using machine learning approaches that adapt to the external environment, one of which is reinforcement learning (RL). At the same time, an approach is proposed to restore the demand surface for subsequent training of the agent.