Detecting Bid Leakage in Procurement Auctions: A Machine Learning Approach with New Electronic Data: новая интересная тема научного семинара департамента экономики

Докладчик: Ломакин Артемий Сергеевич, преподаватель, аспирант департамента экономики, младший научный сотрудник Международная лаборатория теории игр и принятия решений

Абстракт: We study bid leakage in sealed-bid procurement auctions. Bid leakage is the corrupt practice when an auctioneer reveals a participant's bids to a favored bidder in order to maximize their chances of winning. This paper pursues two main goals: first, to compare bid leakage before and after the 2019 reform, which restricted offline auctions; second, to measure the extent of bid leakage by estimating the share of suspicious auctions. The analysis is based on procurement data between 2014 and 2022. We apply statistical and machine learning methods, including decision trees and gradient boosting, to predict the probability of winning an auction. We show that the 2019 reform effectively eliminated bid leakage in requests for quotations. In the pre-reform period, we estimate that approximately 9% of auctions are affected by bid leakage.