Detecting Suspicious Auctions in Procurement Market

We propose a novel machine-learning-based approach to detect bid leakage in first-price sealed-bid auctions. Bid leakage is when procurer illegally provides favored participant with information about other bids, ensuring her win. We extract and analyze the data on more than 1.4 million Russian procurement auctions. The direct classification is impossible since the fact of bid leakage in each auction is unknown. To overcome this problem, we first train a classifier, namely, gradient boosting of decision trees, to distinguish the winners from the runner-ups. Crucially, we use features that are irrelevant for fair auctions but reflect the bid leakage well. We then suggest methods to estimate the probability of bid leakage presence for each auction in the dataset based on classifier’s predictions and overall performance. We validate our results using synthetic placebo dataset. We find that at least 10% of auctions are exposed to bid leakage, which is consistent with the previous estimates in the literature.