Collusion Detection in Infrastructure Procurement: A Modified Order Statistic Method for Uncapped Auctions

Collusion occurs when companies conspire to remove competition from the process of bidding and collectively agree who will win an auction by engaging in price fixing and bid-rigging. In this article, we modify an existing method based on order statistics to improve its accuracy and reliability to detect the presence of collusion. The new approach can be robustly implemented in practice and is difficult to defraud. We use data from the Brazilian Federal Police's ongoing "Operation Car Wash" investigation to identify collusive behavior during an uncapped auction. Such a rich dataset enabled the new approach to be tested in a real-life setting. We demonstrate that our method's increased predictive power makes it the best option to determine full collusion in uncapped auctions. Our new approach can be used in conjunction with other collusion detection methods to improve their accuracy and reliability. The contributions of our article are twofold: 1) Public and private sectors procurement authorities are better positioned to detect collusion even when limited auction data is made available; and 2) law enforcement agencies are provided with a robust method that can be used together with prevailing evidence to pursue a conviction for engaging in collusive behavior.