Computational antitrust has found more and more applications in recent years, including through the use of machine learning. However, the availability of labelled data to train algorithms proves to be an obstacle. In this paper, Valentin Forster, Jürgen Fleiß, Dominik Kowald and Vicky Robertson explore the use of unsupervised machine learning to detect resale price maintenance (RPM) in price data. They develop assumptions that RPM prices exhibit increased similarity, a right-skewed distribution including a cut-off point, and fewer price changes over time compared with non-RPM prices. Based on these assumptions, they extract features based on simple statistical coefficients and perform clustering to detect products with price characteristics consistent with RPM. Subsequently, this can serve as a sufficient basis to conduct more in-depth antitrust investigations. They test our approach on five real-world product datasets scraped from an Austrian price comparison website. They show that their screen successfully clusters products with price patterns indicative of RPM. Their contribution, published in the Journal of Competition Law & Economics, is available in open access here.
