Detecting the collusive bidding behavior in below average bid auction

Collusion detection is an important problem in any kind of auction mechanism. Average bid auction is widely used in public procurement for it can deal with the problem of winner's curse in the first price auction. However, it has also been accused of incentivizing bidders to collude or manipulate the average price to a high level leading to a high winning price. This kind of collusion could bring huge loss for the buyer. To address this concern, this paper proposes a method containing three steps to detect the potential collusion circle in average bid auction based on network cluster and randomization inference. Moreover, we also suggest an ex-post method and one ex-ante method for the buyers to prohibit the losses from potential collusion.