I also use dummies for the number of bidders in the regression. There are 6 two bidder auctions, 19 three bidder auctions, 11 four bidder auctions, 11 five bidder auctions, 2 six bidder auctions and 1 eight bidder auction in the data set.
In Table 4, I present descriptive statistics for the homogenized bids. These are presented separately for each bidder along with the number of bids submitted and contracts won. The municipal bus company HKL participated in all auctions. The five most active bidders submitted 93 percent of bids and won all but one contract. Bidders that submitted a large share of winning bids generally have a lower mean of homogenized bids.
This table shows the number of bids and winning bids and standard descriptive statistics for each bidder. HKL = Helsingin Kaupungin Bussiliikenne. CX = Connex Oy. STA = Suomen Turistiauto. CR = Concordia Oy. PKL = Pohjolan Kaupunkiliikenne. OLA = Oy Liikenne Ab. LLR = Linjaliikenne Randell. AAS = Askaisten Auto. LSL = LS-Liikennelinjat Oy. ESL = Etelä-Suomen Linjaliikenne. AAD = Auto Andersson Oy.
Observed characteristics explain most of the variation in bids. Therefore it plausible to assume that unobserved heterogeneity does not bias the results of the tests. Krasnokutskaya (2004) shows that if the data generating process follows independent
private costs with unobserved heterogeneity, then using the estimation procedures pre- sented below leads to erroneous cost distributions. Wrongly estimated distributions have too low means and too high variances. Higher variances may lead to a situation where the null is not rejected even when it should be. My application turns out to be robust to this possible problem because the null is rejected.
2.4 Conducting the test
This section runs through the estimation and the testing procedures. Those familiar with the estimation methods proposed by Li et al. (2002) (denoted LPV) and Campo et al. (2003) (denoted CPV) and the HHS testing method can skip this section. The only novel feature here is result 2. It is the main methodological insight of this paper and makes testing in the asymmetric case meaningful.
2.4.1 Intuition behind the test
A central issue in auctions for assessing the effects of competition on procurement costs is whether the bidders operate in a private or a common cost environment. Common costs refer to a situation where information on the costs of fulfilling the contract is dispersed among the bidders. The bidders then update their beliefs about costs if they learn their competitors’ signals. Private costs refer to a situation where the bidders care only about their own signals. This distinction is called the information paradigm. Milgrom and Weber (1982) show that due to the winner’s curse the effects of competition may change with the information paradigm. The winner’s curse arises in a situation where bidders bid in a common costs environment according only to their own cost estimates. With unbiased estimates and symmetric bidders, the bidder who underestimates his costs the most wins the auctions and may receive a negative payoff. The expected amount of underestimation increases with the number of bidders. Rational bidders take this into account and thus may even bid less aggressively as competition increases. Hong and Shum (2002) find empirical evidence for this effect. Strategic behavior implies that bidders bid more aggressively as the number of bidders increases. With private costs, only this strategic component is in play, whereas in a common cost setting both of these factors matter and so the effect of competition is uncertain.