Article on offline reinforcement learning in dynamic pricing published

19 Juni 2025

The article “Improving network dynamic pricing policies through offline reinforcement learning”, which was developed in collaboration between the Chair of Data Analytics & Statistics (Prof. Brieden) and the Chair of Business Analytics & Management Science (Prof. Steinhardt), has been published in OR Spectrum.

In this research work, offline reinforcement learning algorithms are used to learn a pricing policy from historical sales data that achieves higher revenue than the policy underlying the observed data. The authors thus demonstrate for the first time, on typical benchmark networks, how reinforcement learning can be successfully applied in dynamic pricing. Compared to existing online methods, offline approaches offer the major advantage that no interaction with the real world is required, thereby eliminating the costly learning process of online methods, which arises from the need to explore suboptimal prices.

The open access publication can be accessed here.