Public procurement auctions in Brazil

This thesis provides an empirical analysis of data generated by ComprasNet, the online procurement bidding platform developed and used by the Brazilian federal government. ComprasNet is a large bidding platform used since 2001 by more than 2200 public purchasing units who list around one million lots each year. Over 70,000 unique bidders have participated in these auctions. In 2010, 46 percent of all procurement for the federal government was conducted through ComprasNet, totaling R$ 27 billion, or 0.7 percent of Brazil’s GDP. In short, these auctions represent a large share of federal tenders and a substantial amount is contracted through them each year. Chapter 1 provides an overview of ComprasNet. After reviewing the literature on various topics which this dissertation contributes to, I describe the institutional background surrounding ComprasNet. I then present the baseline data used throughout the remainder of this dissertation. Chapter 2 addresses one important aspect of designing an online ascending auction, namely how to end the auction. ComprasNet varied its ending rules over time, providing an unique opportunity to test theories of bidder behaviour, as well as assessing the impact of ending rules on auction outcomes. Chapter 3 analyses a two-stage auction format which ComprasNet uses. Two-stage designs have long been proposed by the theoretical literature, but there are virtually no empirical works apart from experimental studies. Finally, chapter 4 analyses a bid preference programme targeted at small and micro enterprises (SMEs). The programme consists of setting aside eligible lots for SMEs. We first use eligibility rules as a source of exogenous variation in the treatment assignment to estimate the effects of the programme on auction outcomes. We then set up an open auction model with endogenous entry and asymmetric bidders and estimate the model’s primitives. In particular, we estimate entry costs, which we interpret as red tape costs