Abstract
Dynamic pricing is considered a possibility to gain an advantage over competitors in modern online markets. The past advancements in Reinforcement Learning (RL) provided more capable algorithms that can be used to solve pricing problems. In this paper, we study the performance of Deep Q-Networks (DQN) and Soft Actor Critic (SAC) in different market models. We consider tractable duopoly settings, where optimal solutions derived by dynamic programming techniques can be used for verification, as well as oligopoly settings, which are usually intractable due to the curse of dimensionality. We find that both algorithms provide reasonable results, while SAC performs better than DQN. Moreover, we show that under certain conditions, RL algorithms can be forced into collusion by their competitors without direct communication.
Highlights
In modern-day online trading on large platforms using the correct price is crucial
Our goal is to evaluate the performance of two examples of reinforcement learning (RL) algorithms on dynamic pricing problems
We studied pricing competition motivated by online markets in order to provide insights for practitioners to assess in how far reinforcement learning can be used to automate frequent repricing in a self-adaptive way
Summary
In modern-day online trading on large platforms using the correct price is crucial. If your goods’ prices are way off the competition, customers might go for cheaper competitors or ones that offer a better service or a similar product. Many traders nowadays can make use of dynamic pricing algorithms, that automatically update their price according to the competitors’ current offers. Those price updates might occur at a high frequency. Markets offer the advantage, that optimal solutions can still be computed via dynamic programming (DP), cf., e.g., Schlosser and Richly (2019), which provides an opportunity to compare and verify the results of reinforcement learning (RL). The second algorithm is Soft Actor Critic (SAC), which is a recent iteration in the group of policy gradient algorithms. It is based on two components, the actor and the critic. The two subsections contain a deeper introduction to both algorithms
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
