ľ¹Ï¸£ÀûÓ°ÊÓ

Collusion on the public procurement market has adverse effects on social welfare. It is however difficult to detect, especially in an algorithmic manner. In a new Working Paper of the Utrecht ľ¹Ï¸£ÀûÓ°ÊÓ School of Economics (U.S.E.) Vitezslav Titl (Assistant Professor at Utrecht ľ¹Ï¸£ÀûÓ°ÊÓ School of Economics and a member of the Utrecht ľ¹Ï¸£ÀûÓ°ÊÓ Centre for Public Procurement) and co-authors Bruno Baranek and Leon Musolff (both Princeton ľ¹Ï¸£ÀûÓ°ÊÓ) come up with a new and accurate structural test of collusion for a multistage auction mechanism used in some post-soviet countries such as Georgia, Kyrgyzstan, Moldova, and Ukraine.

The implementation of an accurate structural test of can have a large impact on the ability of the authorities to detect and prosecute collusion on this market, and thereby, can improve the procurement market outcomes and the provision of public goods.

Core idea

The main idea behind the proposed detection algorithm is that firms cooperating in a cartel are less likely to update their bids in the subsequent rounds of the auction. Imagine two firms that submit similarly high bids. The fact that their bids are similar likely means that they have similar cost functions. Then it should be easy for the initially losing firm to undercut the bid of the initial winner. The authors found however that for 45% of all tenders, there are no updates. This finding suggests that, in many contracts, firms do not behave competitively.

Building the algorithm

To find pairs of colluding firms, the authors analyse probabilities that a firm from a pair of two firms undercuts the other firm from the pair. Based on a theoretical model and the equilibrium of the market, the authors compute the predicted probabilities of undercutting a competitor and then compare observed behaviour and model predictions. This gives a good guess about whether the two firms are likely to behave competitively against each other, i.e., whether the firms constitute a collusive pair.

Findings

The soundness of the collusion detection algorithm is validated on a sample of 863 prosecuted collusive firms from Ukraine. The authors document that the algorithm assigns a high probability of being cartel members for the prosecuted firms that participated in 23,515 tenders.