Economics Letters 101 (2008) 172–174

Contents lists available at ScienceDirect

Economics Letters
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / e c o n b a s e

Winner’s curse in toll road concessions

Laure Athias a,b,⁎, Antonio Nuñez c
a
ATOM, University of Paris Sorbonne, 112 bd. de l'Hôpital, 75013 Paris, France
b
IDHEAP and SPAN, Rte de la Maladière 21, 1022 Chavannes-près-Renens, Switzerland
c
LET, University of Lyon, 14, Avenue Berthelot, F-69363 Lyon, France

a r t i c l e i n f o a b s t r a c t

Article history: We empirically assess the effect of the winner’s curse in auctions for toll road concessions, taking into
Received 25 January 2007 account, to our knowledge for the first time, problems of commitment and enforcement, using a unique
Received in revised form 7 July 2008 dataset of 49 worldwide road concessions.
Accepted 11 July 2008
© 2008 Elsevier B.V. All rights reserved.
Available online 25 July 2008

Keywords:
Auctions
Common value
Winner’s curse
Incomplete contracts
Concessions

JEL classification:
D44
H54
L51

1. Introduction
Auctions for toll road concession contracts are common-value
auctions in which an increase in the number of bidders has two
counteracting effects on equilibrium bidding behaviour. First, increasing
competition leads to more aggressive bidding, as each potential bidder
tries to maximize her chances of winning against more rivals: this is the
competitive effect. Second, the winner’s curse1 becomes more severe as
the number of potential bidders increases, and rational bidders will
bid less aggressively in response: this is the winner’s curse effect.2 If the
winner’s curse effect is large enough, i.e. more than compensates for
the increase in competition, prices could actually rise – in the context
of procurement auctions – as the competition increases. As a result,
governments should restrict entry, or favour negotiations over
auctions (Bulow and Klemperer, 1996; Hong and Shum, 2002) when
the winner’s curse is particularly strong.
Nevertheless, these effects stand under the classical assumption that
bidders are able to commit to bidding promises. One obstacle to these
theoretical conclusions may be the realization by the intelligent bidder
that the contract may later be subject to profitable renegotiation. This
fact, highly highlighted in concession contracts (Guasch et al., 2003),
affects bidding behaviour in subtle ways, and may strongly question the
two theoretical effects highlighted above (Milgrom and Weber, 1982).
In this paper, we consider the empirical importance of these
considerations, using an original database of 49 worldwide toll road
concession contracts, self collected. We show that bidders bid less
aggressively in toll road concession auctions when they expect more
competition (i.e. the winner’s curse effect is very strong), and that
bidders will bid more strategically in weaker institutional frame-
works, in which renegotiations are easier.
2. Auctions for road concessions

⁎ Corresponding author. ATOM, University of Paris Sorbonne, 112 bd. de l'Hôpital,
75013 Paris, France. Tel.: +33 1 44 07 83 21; fax: +33 1 44 07 83 20.
E-mail addresses: laure.athias@malix.univ-paris1.fr (L. Athias), abnunez@gmail.com
(A. Nuñez).
1
The winner’s curse is an adverse-selection problem which arises because the
winner tends to be the bidder with the most overly optimistic information concerning
the value of the auction.
2
Thus, what is called winner’s curse effect in the rest of the paper is actually the
internalization of the winner’s curse.
In this paper we study bidding behaviour in first-price, sealed bid
auctions, using data on road concessions. Concession contracts are
awarded in two stages; in the first stage, private consortiums submit
their technical qualifications, following the rules defined by the public
authority. In the second stage, qualified consortiums are allowed to
bid. Concessionaires are usually specific purpose companies, which
shareholders are infrastructure construction companies and financial
institutions, both either local or international.

0165-1765/$ – see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.econlet.2008.07.017
 L. Athias, A. Nuñez / Economics Letters 101 (2008) 172–174 173

Fig. 1. Distribution of TFD.

Demand forecasts for infrastructure projects include many competition increases. A good measure for this correction is the
uncertainties. Thus, each bidder’s traffic appraisal represents an relative discrepancy between the forecast and the actual traffic.
estimate, subject to error. No bidder knows what future traffic We have data on the traffic forecasts included in the bids
will be and each one realizes that the other bidders may possess submitted by the winning bidders, and on actual traffic coming from
information or analyses he would find useful for his own forecast. traffic counts. The average ratio between them is called Traffic
As a result, in toll road concession auctions, the winning bidder Forecast Deviation (TFD), defined as:
may be the one who most overestimates future traffic. Bidders who
would fail to take this selection bias into account at the bidding stage 1 t0 þn−1 forecastt
TFD ¼ ∑ ð1Þ
would be subject to the winner’s curse. Reasonably sophisticated n t¼t0 actualt
bidders should then bid cautiously. These considerations lead us to the
following proposition: where n is the number of years for which we could compute this
deviation. One aspect of the contractual record draws immediate
Proposition 1. The greater the number of bidders, the more likely attention: the prevalence of traffic overestimation, as highlighted by
bidders will be conservative to correct for traffic overestimation, i.e. the the existing literature (Flyvbjerg et al., 2003), since the average
greater the effects of the winner’s curse. overestimation is 25%. Fig. 1 gives the distribution of this variable in
the sample.
However, imperfect enforcement leading to renegotiations is a
major characteristic of concession contracts (Guasch et al., 2003).
Thus, when bidders expect a high likelihood of renegotiation that
Table 1
renders it possible to avoid any losses, they have strong incentives to
Data definitions and descriptive statistics
submit bids containing promises difficult to satisfy, with the sole
purpose of being awarded the tender. Uncertainty in forecasts is then Variable Obs Mean Std. dev. Min Max Definition
used in a strategic way by bidders, which is exacerbated by TFD 49 1.253 0.453 0.8 3.399 Ratio forecast
traffic/actual traffic
information asymmetries in concession projects. This feature leads
NUMBER OF BIDDERS 49 3.918 1.891 1 9 Number of bidders
to the following proposition, which has, to our knowledge, never been (NB) for the contract,
tested: after the
prequalification
Proposition 2. The lower the likelihood of contract renegotiation, the stage
CIVIL LAW 49 0.735 0.446 0 1 1 if the country in
more likely bidders will be conservative as the number of bidders
question is under
increases, i.e. the greater the effects of the winner’s curse. civil law regime; 0
otherwise
To test this double prediction, we now turn to the empirical part of HIGH INCOME COUNTRY 49 0.531 0.504 0 1 1 if the country in
(HIC) question is a high
the paper.
income country; 0
otherwise (source:
3. Data on road concession contract auctions World Bank)
GOVERNMENT 49 2.531 3.056 0 10 Number of
We have constructed a unique dataset of 49 worldwide toll road LEARNING concessions the
public authority
concession contract auctions (highways, bridges and tunnels). Most of
has awarded
the data included in the database was provided by concessionaires before the present
and by regulators. Some others come from scientific and professional project
press. LENGTH 49 107.089 112.997 0.5 510 Length of the
As explained above, bidders need to correct for traffic over- facility in
kilometres
estimation and increase their correction on their estimate when
174 L. Athias, A. Nuñez / Economics Letters 101 (2008) 172–174

Table 2 identify in Models 2 to 4, if the theoretical effects still hold when we

Estimation results account for the possibility for bidders to renegotiate the contract.
Model 1 Model 2 Model 3 Model 4 Results are reported in Table 2.
NUMBER OF BIDDERS −0.220⁎⁎⁎ − 0.169⁎⁎ −0.132⁎ −0.347⁎⁎⁎ The first striking result is that the number of bidders is clearly an
(NB) (−2.87) (− 1.97) (−2.72) (−3.28) important variable, driving the value of bidders’ tenders. Model 1
GOVLEARN ⁎ NB − 0.010 shows a negative impact of a fiercer competition on the traffic forecast
(− 1.13)
deviation variable. It means that, overall, bidders are more conserva-
HIC ⁎ NB −0.136⁎⁎⁎
(−2.44) tive the more bidders there are, i.e. the effect of the winner’s curse in
CIVIL LAW ⁎ NB 0.119⁎ toll road concession contract auctions is strong.
(1.71) Results of models 2 to 4 show that this phenomenon is significantly
LENGTH − 0.056⁎⁎ −0.075⁎⁎⁎ −0.070⁎⁎⁎ stronger when bidders expect a lower likelihood of renegotiation. In
(− 2.44) (−3.30) (−2.97)
Constant 0.452⁎⁎⁎ 0.641⁎⁎⁎ 0.730⁎⁎⁎ 0.768⁎⁎⁎
particular, Model 2 indicates that the effect of the variable GOVLEARN
(4.37) (4.73) (5.74) (5.45) interacted with the number of bidders is negative, though almost not
R2 0.149 0.276 0.342 0.340 significant, on the TFD variable. In addition, as predicted, the effect of
Adjusted R2 0.131 0.228 0.298 0.297 the variable HIC interacted with the number of bidders is significantly
N 49 49 49 49
negative on the traffic forecast deviation, meaning that bidders are
Significance levels: +0.15 ⁎ 0.10 ⁎⁎ 0.05 ⁎⁎⁎ 0.01. more cognizant of the winner’s curse in HIC. Also, we find as predicted
t-stat are in parenthesis. The dummy variables are not taken as logarithms in the model. that the variable CIVIL LAW interacted with the number of bidders is
positive on the traffic forecast deviation, implying that bidders less
internalize the winner’s curse when bidding in civil law countries.
The propositions to be tested formulated above suggest two main Endogeneity of bidders is a concern for empirical studies on
factors that are likely to influence the bidding behaviour and then auctions. However, it is not a major issue here because, first, our
TFD: the number of bidders and the likelihood of contract renegotia- dependent variable is not the bid itself but traffic forecast deviation, so
tion. The actual number of bidders, after the prequalification stage, that the potentiality of unobservable determinants of traffic forecast
accounts for the level of competition. The hypothesis is that bidders deviation is weak and, second, bidder endogeneity tends to mitigate
will be more conservative the larger is the number of bidders, i.e. we the winner’s curse yet we find statistical evidence of the winner’s
expect a negative impact of the NUMBER OF BIDDERS variable on our curse.
TFD variable.
Regarding the likelihood of contractual renegotiation, we used 5. Conclusion and policy implications
three variables. The first one, the variable GOVLEARN, reflects the
experience of the procuring authority in awarding concession Using an original database, we show that the winner’s curse effect
contracts. As a large number of prior concessions should decrease is particularly strong in toll road concession contract auctions. More
the probability of renegotiation (Guasch et al., 2003), we expect a precisely, the winner’s curse effect prevails on the competitive effect so
negative impact of this variable interacted with the number of bidders that bidders bid less aggressively when they expect more competition.
variable on our dependent TFD variable. We also observe that the effect of the winner’s curse is weaker
The second one is the indicator “high income country” (HIC) when the likelihood of renegotiation is higher, so that bidders will bid
developed by the World Bank in 2006. As in Laffont (2005), the more strategically in weaker institutional frameworks. These original
intuition behind is that wealthier countries are more able to invest in results point out the necessity to improve on the current theoretical
institutions than poorer countries, so that the probability of framework by taking into account as a primary concern the impact of
renegotiation is smaller and then the winner’s curse effect will be the perspective of later profitable renegotiation on equilibrium
stronger. We expect therefore a negative impact of the crossed bidding behaviour.
variable HIC ⁎ NUMBER OF BIDDERS on our TFD dependent variable.
The third one, CIVIL LAW reflects the legal system of the country in References
which the project takes place. The intuition is that institutional
features that traditionally characterize a common law regime make it Bulow, J., Klemperer, P., 1996. Auctions versus negotiations. American Economic Review,
86, 180–194.
more difficult to renegotiate under such a legal regime than under a Flyvbjerg, B., Bruzelius, N., Rothengatter, W., 2003. Megaprojects and Risk — An
civil law system (La Porta and Shleifer 2003). We expect therefore a Anatomy of Ambition. Cambridge University Press.
lower winner’s curse effect in civil law countries. Guasch, J.L., Laffont, J.J. and Straub, S., 2003, Renegotiation of Concession Contracts in
Latin America. Mimeo.
In addition, we include the physical LENGTH of the infrastructure
Hong, H., Shum, M., 2002. Increasing competition and the winner’s curse: evidence
as a control variable since it captures the heterogeneity in the sample from procurement. Review of Economic Studies 69 (4), 871–898.
(as the length is highly correlated with the forecast errors3) and re- Laffont, J.J., 2005. Regulation and development. Collection Frederico Caffe Lectures.
normalize the regression errors (See Table 1 for descriptive statistics). Cambridge University Press.
La Porta, R., Shleifer, A., 2003. Courts. Quarterly Journal of Economics 118.
Milgrom, P., Weber, R., 1982. A theory of auctions and competitive bidding.
4. Econometric results Econometrica 50 (5), 1089–1122.

To test our double prediction, we first analyse the overall impact of

the number of bidders on bidding behaviour (Model 1). We then

3
Regressing the squared errors on LENGTH and the other control variables, we find a
very significant (t-test = −3.05) negative correlation.