?
A Noisy-Labels Approach to Detecting Uncompetitive Auctions
Despite several rounds of institutional reform starting from 2005, the public procurement process in Russia remains marred by low competitiveness and inefficiency. In the years 2014–2018, almost half of the studied auctions failed to attract more than a single-bidder. But are single-bidder auctions necessarily uncompetitive? The auction format we study is a sealed-bid auction, where a bidder is unaware of who else is participating. If they were to submit a bid with the expectation of competition that fails to materialise, for all means and purposes the auction can be said to be competitive. More importantly, the presence of many bidders does not guarantee competition – we could be facing a cartel, or the restriction of competition via a corrupt procurer, such as the case of bid-leakage. We assume that bids in multi-bidder auctions are predominantly competitive while bids in single-bidder auctions are not, and apply generalized confident learning, a method for classification in the presence of noisy labels, to attempt to separate competitive and uncompetitive bids. This allows us to identify behaviour patterns resembling monopolists and cartels.