Public procurement cartels: A large-sample testing of screens using machine learning

Fazekas, M., Tóth, B. and Wachs, J. (2023). Public procurement cartels: A large-sample testing of screens using machine learning. GTI-WP/2023:02, Budapest: Government Transparency Institute.

Cartels in public procurement impose high costs on public budgets. Precisely measuring them has a prominent policy and academic importance. The literature so far used data which is not widely available, aimed to identify specific behaviours in isolation, and considered few cases to generalise from. By implication, it has not produced comprehensive and generalisable knowledge able to support policy. We address these gaps in the literature by simultaneously measuring multiple cartel behaviours, drawing on data for 68 cartels from 7 countries during 2004-2021. We apply state-of-the-art machine learning methods to combine diverse cartel screens in a predictive model. As expected, no single indicator or group of indicators can predict a wide set of cartel behaviours. Combining many indicators in a random forest algorithm achieves 77-91% prediction accuracy across countries. Most individual cartel screens contribute to prediction in line with theory. Policy implications are profound, offering to improve cartel investigations and policy making.

Read the full working paper HERE