Detection of Collusive Networks in Multistage Auctions

We develop a method for detecting cartels in multistage auctions. Our approach allows firms to act collusively against cartel members yet competitively otherwise. In these auctions, bidders submit initial bids and can later revise them downward. As initial bids are shaded, close initial bids indicate similar costs and thus create strong incentives to undercut rivals. We classify firm pairs as collusive if they ignore this incentive by failing to update bids against each other, relative to other pairs with similarly close initial bids. Our algorithm predicts Ukraine's Antimonopoly Committee sanctions: Classified pairs are nine times more likely to be sanctioned.