Adam, I., Fazekas, M., Kazmina, Y., Teremy, Zs., Tóth, B., Villamil, I., R. and Wachs, J. (2022). Public procurement cartels: A systematic testing of old and new screens. GTI-WP/2022:01, Budapest: Government Transparency Institute.
Though cartels are thought to be common, they tend to be hard to find. Successful prosecutions are even more rare, and usually begin with an exceptional event: when a cartel member makes a sloppy mistake or decides to blow the whistle. Researchers have long studied these cases to learn how cartels function, and how collusive behavior sends signals in data. These signals are used to build cartel screens, methods to scan data on prices and activity for evidence of collusion. Even though these screens are theoretically sound and tend to work very well on the cases they are designed to highlight, less is known about their external validity. As competition authorities are collecting more and more data, there is a growing need to evaluate the general applicability of cartel screens.
In this report we collect data on 156 proven cartel cases in public procurement from around the world. We link the firms involved in these cartels to procurement market activity in 77 cases from six countries, covering a rich variety of sectors. We test the efficacy of a suite of cartel screens across these contexts. We could apply at least one screen to 47 cases. We come to several conclusions:
- Data quality and availability are severe hinderance for testing and applying cartel screens. This finding is especially disconcerting given that public procurement data ought to be transparent and accessible to citizens.
- Most cartels set off one or more screens in our toolkit, some set off as many as five screens. Specifically, 33 of the 47 cartels set off one screen, 13 set off at least 3 screens, and 5 set off 5 or more screens. This suggests that some cartels will be highly visible under a multi-screen microscope.
- No single screen works in even a majority of cases, reflecting the rich variety of cartel types operating in the economy and proving that only a multi-screen approach could lead to effective indicator-based investigations.
Reflecting on our findings, we suggest that by combining multiple screens, machine learning methods, such as random forests, can be applied to predict collusion risks of groups of firms hence guiding investigative priorities. A rigorous cartel screening program, applying multiple tested screens to clean data has the potential to be a gamechanger for competition authorities working in procurement. By monitoring signals of collusion, cartels will be forced to collude in increasingly sophisticated ways, increasing costs and adding uncertainty to the long list of challenges of coordinating a cartel.
Read the full paper here