Reinforcement Learning for E-Commerce Dynamic Pricing

With the quick development of artificial intelligence technology, it has been applied in many fields. Motivated by applications in financial services, we consider a seller who offers prices sequentially for online products, to maximize the long-term revenue, as well as increase costumers’ satisfaction. This paper investigates how reinforcement learning methods can help optimize profits for e-commerce. We model the dynamic pricing problem as a Markov decision process and apply two reinforcement learning methods: Q-learning and Sarsa for pricing. Then, we give three predetermined demand models: linear-, quadratic- and exponential models with a variety of learning rates for numerical experiments. Results suggest that Q-learning has better performance than Sarsa as it achieves higher profits and lower volatility except the learning rate is 1.