Collusion detection in public procurement auctions with machine learning algorithms

Abstract

Collusion is an illegal practice by which some competing companies secretly agree on the prices (bids) they will submit to a future auction. Worldwide, collusion is a pervasive phenomenon in public sector procurement. It undermines the benefits of a competitive marketplace and wastes taxpayers' money. More often than not, contracting authorities cannot identify non-competitive bids and frequently award contracts at higher prices than they would have in collusion's absence. This paper tests the accuracy of eleven Machine Learning (ML) algorithms for detecting collusion using collusive datasets obtained from Brazil, Italy, Japan, Switzerland and the United States. While the use of ML in public procurement remains largely unexplored, its potential use to identify collusion are promising. ML algorithms are quite information-intensive (they need a substantial number of historical auctions to be calibrated), but they are also highly flexible tools, producing reasonable detection rates even with a minimal amount of information.