Calculating the probability of collusion based on observed price patterns

We present a method for estimating the probability of collusion using observed price patterns. Having these probabilities, we can also estimate the impact of the number of firms and other relevant variables on the probability of collusion, as well as the price increases and total expected overcharge caused by collusion. These estimates are essential to inform policies about how to best prevent collusion. We apply the method to 28,863 auctions in the Swedish generic pharmaceuticals markets and predict 64% of the auctions to be part of price patterns at least partly caused by collusion. We also find that nearly all collusions take the form of bid-rotation (i.e., firms taking turns in offering the lowest price) and that moving from competition to collusion increases average prices by 65%. Moreover, the results demonstrate that multimarket contact significantly increases the risk of collusion, and that increasing the number of firms from two to four reduces the risk of collusion by approximately one half. Still, we find collusion to be an important problem even with four or more firms.