Analysis of mergers in first-price auctions∗
Klaus Gugler†
Michael Weichselbaumer‡
Christine Zulehner§
preliminary and incomplete – please do not quote
March 2015
Abstract
In this paper, we analyse mergers in bidding markets. We utilize data from procurement
auctions in the Austrian construction sector and estimate models of first-price sealed-bid
auctions. Based on estimated cost and markups, we run merger simulations and disentangle
the market power effect from potential cost efficiencies. In addition, we compare static and
dynamic models of first price auctions. Bidders may consider the strategic consequences of
their backlog and price into their bids also the lost option value of winning today’s than
tomorrow’s auction. Ignoring this effect may yield a biased assessment of mergers. Finally,
we compare the outcomes of the merger simulations with the effects of actual mergers. Our
aim is to evaluate the predictions from the auction model with real world outcomes.
JEL Classifications: D44, L10, L13
Keywords: construction procurement, first-price auctions, private values, merger simulation,
evaluation of mergers
∗
We thank seminar participants at the University Würzburg for their comments and suggestions. We are
especially grateful to Josef Aicher (University Vienna) for explaining to us the legal background of Austrian
procurement auctions. This project received funding from the Austrian Science Fund. All errors and opinions are
the authors’ sole responsibility.
†
WU Vienna University of Economics and Business, Address: Welthandelsplatz 1, A-1020 Vienna, Austria,
Email: [email protected].
‡
WU Vienna University of Economics and Business, Address: Welthandelsplatz 1, A-1020 Vienna, Austria,
Email: [email protected].
§
Goethe University Frankfurt, Austrian Institute of Economic Research Vienna and CEPR, Address: Grueneburgplatz 1, D-60323 Frankfurt am Main, Germany, Email: [email protected].
1
Introduction
Merger analysis — the study of the determinants and effects of mergers on the merging firms and
on the markets they operate — has an impact on many areas of industrial organization research.
It answers old questions and raises new questions of firm behavior. The most direct research
agenda are questions of market power and efficiencies. Other fields of industrial organization
and economics in general have derived fruitful environments from mergers and acquisitions
for formulating models, deriving hypotheses and testing. Examples are financial economics,
principal-agents models and the economics of research and development. Understanding of
mergers, eventually, is the understanding of the competitive process, both for economic theory,
empirical modelling and policy applications.
Some ground-breaking research on merger analysis has already connected theoretical models with strategies for empirical identification. As a result, merger simulations are now often
standard to assess the effects of a merger.1 Despite these major advances, there is no universal model for merger analysis. Industry characteristics remain important to decide on the
appropriate model application — although, in practice, a model of Bertrand competition with
differentiated goods is most often used. In addition, the intricacies of estimation lead to controversial debates of empirical results of merger simulation studies and the use of a treatment
effects approach has been proposed.2
Empirical analysis of merger simulation involving bidding markets has so far received little
attention. The advanced level of auction theory3 and its empirical modelling, though, provide
fertile ground for the application to merger analysis.4 With our research we plan to address
the following topics: First, we simulate mergers in an auction model relying on the theoretical
auction literature. Based on a model of first-price auctions, we estimate and identify the effects
of a proposed merger, i.e., market power and efficiency effects. We evaluate the mergers in our
data on construction procurement auctions with respect to the non-merger benchmark as implied
by the merger simulation model. Second, we compare the static with the dynamic specification.
1
Merger simulation as a tool for competition policy was introduced by Hausman et al. (1994) and Werden and
Froeb (1994). For an introduction to merger simulation see Davis and Garcés (2009).
2
See the discussion of treatment effects approach versus merger simulation in Nevo and Whinston (2010).
3
The vast auction literature is summarized by Klemperer (2004).
4
Among notable exceptions that combined auctions to merger analysis, in particular with respect to theoretical
considerations, are Klemperer (2005) and Werden and Froeb (2005). Hypothetical mergers in auctions are analyzed
in Li and Zhang (2013).
1
Bidders may consider the strategic consequences of their backlog and price into their bids also
the lost option value of winning today versus winning in the next auction. If we ignore this
effect, estimated cost and thus estimated markups may yield a biased assessment of mergers.
To quantify this potential bias, we build one the theoretical and empirical auction literature
and estimate a static and a dynamic first-price auction model. We then compare the effects of
mergers on bidding outcomes across models. Third, we evaluate the merger simulation model
in comparison to the actual outcome of the observed mergers, with respect to bidding behavior
and cost efficiencies.
We will model mergers within a theoretical first-price auction model. An empirical model
is used for identification and recovery of the cost distribution. This paper departs from Gugler
et al. (2014) where we analyze the effects of the recent economic crisis on firms’ bidding behavior
and markups in sealed bid auctions. Using data from Austrian construction procurement, we
estimate bidders’ construction costs within a private value auction model. We also implement
a dyanamic auction model following Jofre-Bonet and Pesendorfer (2003).
The basic model for the simulation of mergers is the standard approach of analyzing firstprice auctions (Athey and Haile, 2005; Guerre et al., 2000) and to derive marginal costs directly
from observed bids. Based on the estimates of marginal costs, we can evaluate the effect of
the merger and observe the outcome for cost efficiencies. The crucial maintained assumption
of this method is — as in any merger simulation — that the theoretical model describes the
agents’ economic behavior well enough. Experimental evidence, especially when experienced
bidders are participating in first-price auctions, shows that the equilibrium outcome is reached
rapidly (Dyer et al., 1989). For practical reasons, we impose a specific distribution on the bids
and recover the distribution of bidders’ costs using the first order condition for optimal bidding
behavior. First order conditions are derived from standard profit maximization under the rules
of first-price sealed bid auctions, i.e. that the lowest bidder wins the contract. The market
model is strongly driven by the auction design rules and narrows the discretion of the researcher
to some extent.
By obtaining structural estimates of marginal costs we can disentangle the efficiency effect
from the market power effect due to the merger. Decreases in marginal costs from before to after
the merger should be attributable to the efficiency effect, any increase in the markup should
2
be attributable to the market power effect. Because we have two independent metrics for these
effects we can disentangle them and need not rely on a measurement of the net effect.
Several interesting aspects can be addressed with our framework and data. We rely on
several actual mergers completed between firms in our sample and do not only look at a single
or a hypothetical merger. The data contain many bids of the companies both before and after
the merger, which provides a lot of information for evaluation of the merger and evaluation of
the simulation model. The merger evaluation looks at the market power and efficiency effect
using the actual bids; the ex-post evaluation of the simulation model looks at ex-post bids
compared to predicted bids that are solely based on ex-ante data (ex-post evaluation of the
simulation model is what Nevo and Whinston (2010) call “retrospective merger study”). Some
model evaluation studies have been made (see, among others, Peters (2006), Houde (2012),
Weinberg and Hosken (2013)). The authors discussing strengths and deficiencies of structural
and treatment approaches usually agree that more “retrospective study” is useful. None of the
studies so far has considered bidding markets, despite their relevance and the rich theory that
that has been developed.
2
Literature review
To analyze mergers, we follow the microeconomic literature and estimate bidders’ cost distribution from submitted bids.5 By imposing the structure of a game-theoretic model we are able to
back out firm specific costs and run counterfactuals, in our case merger simulations. We have to
assume, however, that bidders behave according to the theoretical model.6 For the empirical implementation, our starting point is Athey et al. (2011) and a parametric version of Guerre et al.
(2000) to recover the distribution of bidder costs from the observed bids. The distribution of
bidders’ costs uses the assumption that in equilibrium the estimates of the distribution function
summarize bidders’ beliefs and infererence of bidders’ costs is based on the first-order conditions of optimal auctions. The estimated distribution of bids includes auction characteristics,
firm characteristics as well as the level of competition to which firms are assumed to respond
optimally.
5
For a survey on nonparametric identification and estimation of auction models see Athey and Haile (2005).
For surveys of empirical studies of auctions see Hendricks and Paarsch (1995) or Laffont (1997).
6
Bajari and Hortaçsu (2005) provide evidence from experimental data that assesses the reasonableness of
structural estimates.
3
Within the empirical auction literature, especially highway procurement auctions have been
studied extensively. Jofre-Bonet and Pesendorfer (2003) show non-parametric identification in
repeated auction games and using a parametric model for the estimations. They find evidence
for capacity constraints in Californian highway procurement auctions. Krasnokutskaya (2011)
proposes a nonparametric estimation method to recover the distribution of bidders’ private
information when unobserved auction heterogeneity is present and finds that private information
is estimated to account for 24 percent of the variation in bidders’ costs in Michigan highway
procurement auctions. Balat (2012) looks at road construction in California and investigates the
effect of the American Recovery and Reinvestment Act on equilibrium prices paid by the U.S.
government. He extends the models by Jofre-Bonet and Pesendorfer (2003) and Krasnokutskaya
(2011) and allows for unobserved project heterogeneity and endogenous participation. His results
show that prices rather than quantities increase as a consequence of the implemented stimulus.
Further studies accounting for endogenous participation are Krasnokutskaya and Seim (2011),
Athey et al. (2011), and Athey et al. (2013).
With the results for the cost estimates we run counterfactuals to mimic the situation after
a merger. Here, we follow the large standard literature on merger simulation in models with
Bertrand competition. Merger simulation as a tool for competition policy was introduced by
Hausman et al. (1994) and Werden and Froeb (1994). Subsequent research has looked at a
variety of issues, such as alternative demand models, e.g. Nevo (2000), Epstein and Rubinfeld
(2001) or Ivaldi and Verboven (2005). Some of this work has explicitly compared different
demand models and showed how different functional forms may result in rather different price
predictions, see Froeb et al. (2003), Huang et al. (2008) and Slade (2009).
In a recent paper, Li and Zhang (2013) estimate the effect of a hypothetical merger. Their
counterfactual analyses apply a parametrical estimation of entry and value distributions to two
types of mergers. First, a merger of the two lowest-ranked bidders; second, a merger of winner
with the second-ranked bidder. These two counterfactuals are uniformly applied to all auctions
and do not trace the identity of the bidders. Bidder asymmetry enters the model in one firmcontract specific variable, which is the distance to the timber tract that is to be contracted by
the winner of the auction. After the merger, the combined entity obtains two draws from the
value distribution — one based on the shorter, one based on the longer distance to tract —
and can use the higher valuation as the basis for the bid. In their affiliated values specification,
4
a merger of the bidders ranked one and two leads on average to a decrease of 4.2 percent of
revenues for the seller. The merger of the two low bidders leads to an increase of 2.4 percent of
revenues. If the highest bidders merge, profits decrease for the merged firm; if the lowest bidders
merge, profits increase. For the IPV framework, their simulation predicts a reduction of mean
sellers revenue of almost 8 percent on average when rank one and two merge, and an increase
of sellers revenue of 3.3 percent when the low-rank firms merge. Profits increase in both merger
types in the IPV framework.
Ex-post merger analysis moved in parallel with merger simulation, and mainly aimed to
evaluate the relevance or effectiveness of competition policy towards mergers. Early work focused
on mergers in major industries, such as airline markets (Borenstein, 1990; Kim and Singal, 1993),
banking (Focarelli and Panetta, 2003) and gasoline (Hastings, 2004; Hastings and Gilbert, 2005;
Hosken et al., 2011). Ashenfelter and Hosken (2008) take advantage of scanner data to assess
mergers in five different branded goods industries. They find moderate but significant price
effects in the range of 3 to 7 percent. Using different methodological approaches, all these
studies find some evidence for price increases after mergers in the industry. This points to a
preponderance of market power effects over efficiency effects.
The discussion about the analysis of the effects of mergers takes a prominent role in the
discussion of structural versus treatment approaches in Nevo and Whinston (2010). While Nevo
and Whinston (2010) acknowledge that a quasi-experimental analysis of treatment effects of
mergers can have advantages in certain circumstances, they also see the need for estimation
based on simulation. The main challenge for direct causal analysis of mergers in the treatment
approach is to use data to describe a counterfactual world in which the merger did not occur.7
In particular, it is not obvious to find firms that are good comparisons yet at the same time not
affected by the merger. Nevo and Whinston (2010) see the additional problems that mergers may
be endogenous and — in an anti-trust context — that it is hard to find past similar mergers to
guide anti-trust decision making on a given current merger. The difficulty for analysis based on
simulations is the requirement to estimate several aspects of the market: first, a demand system
has to be specified and appropriate instrumental variables have to be found; second, a model
of market conduct (e.g. Bertrand-Nash price-based competition with product differentiation)
has to be postulated. The substitution matrix that results from the estimated demand system
7
Nevo and Whinston (2010) in particular mention Hastings (2004) as being very careful in that respect.
5
together with the assumptions on market conduct identifies marginal costs from first order
conditions; finally, industry outcome is simulated with and without the merger.
The question of why firms merge is a perennial one not only in economics but also in business
administration and finance. In a recent survey, Mueller (2012) lists the following reasons of
mergers: (1) mergers may be just another form of investment (Jovanovic and Rousseau, 2002);
(2) mergers occur in the context of Coase’s theory of the firm and are efficiency-enhancing
solutions to market failures (Weston, 1970; Williamson, 1970), (3) mergers are attempts to
eliminate competition and increase market power (Stigler, 1950). These three sets of theories
can all be characterized as neoclassical in that they assume that managers are maximizing
profits and thus that mergers increase profits. Further reasons for mergers are: (4) behavioral
theories posit other goals for managers, like empire building (Mueller, 1969), or hypothesize
that managers are gripped by irrational impulses out of hubris (Roll, 1986). These theories do
not imply that mergers increase profits, but that they are likely to destroy shareholder wealth.
(5) Hostile takeovers might be solutions to managerial failures. Poorly managed companies get
taken over and their managers replaced (Manne, 1965; Marris, 1964).
3
Data and industry background
This section provides information on the data and organization of procurement auctions in
Austria. We also describe the mergers that have taken place in the years 2006–2009, the time
span of our sample.
3.1
Data, mergers and summary statistics
For the empirical analysis, we combine several sources of data to complement the main data on
procurement auctions in the construction sector.8 An Austrian industrial construction company
provided data containing bids, competitors and auction characteristics. These data cover all
auctions in both building and heavy construction during the period 2006 to 2009, where this
company took part either as the parent company itself or through a subsidiary. According to
the company, the database covers more than 80 percent of all auctions which must be con8
The construction sector accounts for 7 percent of GDP on average between 2006 and 2009 (in Germany this
share is 4.2 percent, in the USA 4.4 percent; OECD STAN data set).
6
ducted according to the Austrian Public Procurement Law.9 Thus our sample covers nearly the
population of all public procurement auctions in Austria during the four years 2006 to 2009.
Within this period, these data reflect on average 14 percent of Austria’s total construction sector. Additional data is obtained from matching the bidding firms to Bureau van Dycks Amadeus
database, which contains nearly the population of companies in Austria. Matches are possible
by company names and postal codes. Almost all bids were made by firms that are well known
in Austria and matching is not a major concern. The population of Austrian construction firms,
as reported by the Austrian Central Bureau of Statistics, consists of 4,796 firms on average for
2006 to 2009 (building: 3,958, heavy: 838). Therefore, about one third of all construction firms
submitted bids in the procurement auctions.
For our sample of firms we have bidder characteristics such as number of employees, sales and
assets. In addition, we measure transportation costs by the distance of a firm to a construction
site. Based on the construction sites’ and bidders’ postal codes, we used Microsoft’s Bing
Maps to calculate the driving distances for all bidders to the project sites corresponding to
the auctions.10 Backlog used by a firm at a point in time is calculated as in Jofre-Bonet and
Pesendorfer (2003). Every project is added to the backlog when a firm wins it. Projects are
linearly worked off over their construction period, which releases capacity. Every firm’s backlog
obtained is standardized by subtracting its mean and dividing by its standard deviation. See
Table 1 for the exact variable descriptions.
Table 1 about here
Auction characteristics are the technical estimate, the subsector (building, heavy construction) and variables describing the auction procedure. We further know the actually participating
bidders and cumulative characteristics of other participating firms such as the sum of distances
or the sum of backlogs. We also derive measures for potential bidders, applying restrictions
like counting only firms that submit a bid for any heavy construction auction in the sample as
9
Austria’s public authorities (federal and regional government, social security institutions, and the like) are
subject to the Federal Public Procurement Law (“Bundesvergabegesetz”). In principle, there are no lower limits
for the applicability of the Public Procurement Law. Upper limits are in effect, but only to enforce EU-wide
announcement rules for larger projects (in construction, currently 4.845 million euros).
10
If a firm has multiple plants we still can only calculate the distance between the headquarter and the construction site as we do not have any information on plants. Only for larger firms we know their subsidiaries and
can calculate the distances accordingly.
7
potential bidders for any other heavy construction auction, and constraints relating auction size
and distance to historical bidding behavior of each firm.
On average 7.5 firms take part in Austrian procurement auctions in the construction sector
(see Table 2)). The population of Austrian construction firms, as reported by the Austrian
Central Bureau of Statistics, consists of 4,796 firms on average for 2006 to 2009 (building:
3,958, heavy: 838). This suggests a highly competitive market environment. But there are a
few large construction companies, too,11 and there is differentiation both in the product and the
spatial dimension. Our sample contains bids from 1,655 different companies — so one third of all
construction firms submitted bids in the procurement auctions. The firms in our sample, with a
few exceptions, are mostly medium sized companies with around 50 employees. The mergers we
observe in the sample period — the data contain ten mergers where both acquirer and target
submit bids — similarly involve small to medium sized firms. In such an environment behavioral
hypotheses on mergers do not appear to be the main driving forces, as e.g. empire building or
hubris, nor are there hostile takeovers in our sample. Thus, we are left with the neoclassical
hypotheses, of which the efficiency versus market power motives play the central role. Firms
are likely to merge in the Austrian construction sector because they expect to increase their
efficiency via economies of scale and scope and/or because they expect to be able to reduce
competition.
Table 2 about here
Over the sample period, our data contain ten transactions where both the acquirer and the
target have records as bidders in the procurement auctions. To identify the transactions, we
have used three sources: Thomson Reuters’ SDC Platinum Database, the case list of the DG
Competition (ec.europa.eu/competition) and the case list of the Austrian Competition Authority
(www.bwb.gv.at). The acquirers are major players in the construction sector. They submit
about a third of the 30,000 bids in our sample. The targets submit 250 bids. Six of the ten
transactions are full acquisitions, three acquisitions of majority interest and one is an acquisition
of partial interest. We will further enrich the sample of mergers by complementary search in
Austrian trade and economics periodicals to also identify smaller transactions that might be
overlooked by our main sources.
11
The largest four firms have approximately 40 percent market share within our sample of procurement auctions.
8
Table 3 shows an overview of the mergers involving one or both sides of the transaction
within the sample period. Several horizontal mergers fall within our sample period. There
are nine full acquisitions, five acquisitions of majority interest and two acquisitions of partial
interest. Upstream mergers concern firms which do not compete in the auctions and are not
considered in our analysis. In most cases in the list of mergers, the target retains bidding as an
entity under its old name.
Table 3 about here
Auction data reveal if acquirer and target were direct competitors before the transaction,
as measured by their actual bidding behavior. Table 4 shows this for the mergers we observe in
our data.
Table 4 about here
3.2
Organization of Procurement Auctions
In Austria, public authorities (federal and regional government, social security institutions, and
the like) and private companies active in a “sectorial activity” (provision of water, mail services,
energy, or traffic) or when their their activity is regulated (e.g. entry regulations) have to oblige
the Federal Public Procurement Law (PPL, “Bundesvergabegesetz”). The main purpose of this
law is to stimulate competition and safeguard equal treatment of all potential bidders. The law
stipulates some restrictions on bidders, but they do not, in our judgment, affect competition, but
are intended to forestall opportunistic behavior (e.g. the requirement of a 5 percent bid bond).
Except for very small projects the public authority must choose between an “open procedure”
or a “restricted procedure with publication of a contract notice”. Such projects have to be made
public and documents must be freely accessible. In our sample, 85 percent of the auctions are
“open”, and only 2 percent are of the “restricted” type. The remaining 13 percent were not
publicly announced. Instead, selected qualified companies were invited to submit bids.
Contracts are awarded by first-price sealed-bid auctions. There is no explicit reserve price –
the seller has the right to withdraw the auction when the winning bid is “contra bonos mores”
(offending against good morals). In court, the seller would proof this by providing a cost
estimate that is based on standard commercial and professional principles and has to show that
9
the winning bid is far higher than this estimate. Each project is described in a procurement
catalogue. The seller defines the components and quantities needed for the construction project.
The bidder provides a unit price for each component. The final price in the auction is then the
sum of prices for the various components multiplied with the quantity. Bidders who want to
participate in an auction have to proof their commercial and professional abilities before being
admitted to the bidding process. On the letting day, all bids are unsealed, ranked, and the low
bidder wins the auction. After the lowest bid has been identified all bidders are informed. If no
bidder objects the decision within 10 days, it is final.
3.3
Descriptive analysis
Some target firms are infrequent bidders, as Table 3 reveals. We assume that only segments
where a specific target operates have an influence on the acquirer’s outcome and restrict the
acquirer’s effect to those segments defined by targets’ bidding history. Note that the Tables
report the difference before to after the merger separately for each of the acquisitions. Target
results appear only if the target remains as separate identity after the transaction.
In two steps of descriptive data summaries, we compare several variables before and after
the merger. First, Table 5 shows the difference of each of three selected acquirers and their
respective targets to their pre-merger levels.
In an oligopoly model, usually all firms in the market are affected by the merger. The second
descriptive part, based on regressions run for Table 6, add the bids from firms of auctions where
acquirer and target did not bid to the sample. In Table 6 we report the effect on acquirers and
targets compared to all other bids, but also the difference for bids of firms that bid together
with the merging firms. The estimated equation is
Variableijt = β0 + β1 1 [Firmi ] + β2 1 [Postt ] + β3 1 [Auctionj ]
+ β4 1 [Firmi × Postt ] + β5 1 [Postt × Auctionj ] ,
and β4 again measures the difference for the merging firm compared to the other firms in the
auction, while β5 captures the difference of firms in auctions where the merger takes place versus
firms in auctions where the merging firms did not participate.
10
The differences to a merging firm’s own history (Table 5) shows a significant positive difference for the number of employees of acquirer 2 and 4; for target 7; and a negative difference
for acquirer 10. Backlogs are significantly higher for all acquirers and one target. Bids and
winning bids are significantly lower for acquirers 7 and 10. Among the two firms with reduced
bids, the number of bidders is only higher for acquirer 10, but additionally for acquirers 5 and
9. Acquirer 9 is the only one facing lower money-left-on-the-table than before its acquisition. A
lower number of bids shows up for acquirer two.
Table 5 about here
The top panel of Table 6 shows the effects when the differences are measured against the
other participants of the affected auctions and unaffected auctions are included in the regressions.
Again, an auction is affected only if it falls within one of the segments where the target firm
is active, either before or — if the target exists after the merger — after the merger. Some of
the results for the differences hold up for employees and backolgs in the comparision of merging
firms to their fellow bidders. Differences in all other variables disappear. In the bottom half of
Table 6, several variables turn out to be significant at the auction level, i.e. comparing auctions
with mergers to auctions where the specific acquirer or target does not bid. Bids are lower in
most cases and with a similar pattern to winning bids and, in some cases, the pattern carries
over to money-left-on-the-table. Backlog is now significantly lower in some cases, significantly
higher in ohters. Employees and the number of bids display fewer significant coefficients.
Table 6 about here
4
Merger simulation in auctions
Regressions in the previous section do not give a clear picture on the competitive changes for
merging firms and their competitors. We proceed by the description of the theoretical and
empirical model that we use to evaluate the merger with respect to the two main variables of
interest: markups as a measure of the market power effect, and backed-out costs as a measure
of efficiency effects.
In the widely applied model of Bertrand competition with product differentiation the effect
of a merger is measured by decreasing the number of competitors by one and assuming that the
11
same products earlier offered by two firms are now produced by one firm. Product repositioning
or entry of new firms is usually not considered. In a bidding market the reduction of one bidder
may not completely capture the effect of merger. Firms may alternately participate in auctions
or always participate in the same auctions. Depending on their past behavior the effect of the
merger will differ — even if there is no repositioning or no entry. By estimating bidder specific
cost we can also take efficiencies into account.
Our empirical setting focuses on procurement auctions, where demand comes from the government. We are aware of the possibility to model the government as an additional player to
represent decision making on the demand side. For the basic model, we will neglect this and
treat the government’s demand decisions as unaffected by potential changes in price setting
after the merger. We are also aware that we treat firms’ decision to merge exogenously. This
choice, however, might be influenced by past, current or anticipated future market conditions
unobserved to the researcher (Nevo and Whinston, 2010). To model the choice of firms to merge
explicitly, a dynamic model and enough observed mergers are warranted.
This section describes the theoretical model, the econometric model of the distribution of
bidders’ construction costs and the hypotheses for our empirical analysis.12 We also describe our
data and give summary statistics. To estimate the distribution of bidders’ costs, we implement a
static and dynamic first-price auction. In the latter model, we account for the strategic effect of
capacity constraints following Jofre-Bonet and Pesendorfer (2003). Firms price into their bids
also the lost option value of winning today versus winning later. The static model does not
account for this additional effect: we attribute some of the lost option value to the (technical)
cost of the project and therefore expect to underestimate markups.
4.1
Bidding behavior
We adopt a standard first-price sealed bid auction model to describe the bidding process.13 We
consider an auction for a single contract. The set of bidders is denoted with N = 1, ..., N .
Bidders are assumed to be risk-neutral and their identity to be known.14 Bidder i learns her
12
In this preliminary version, we here heavily draw on Gugler et al. (2014).
For an introduction to auction models in general and first-price sealed bid auction models in particular, see
Krishna (2009).
14
This is a commonly made assumption in the empirical auction literature. From personal conversations with
firms we inferred that it is also suitable for our environment. Before submitting their bids, bidders in these
auctions usually know who is potentially capable to complete the construction and is likely to bid.
13
12
private cost for the project, ci , and bids in the auction. Bidder i’s cost, ci , is an independent draw
from a distribution Fi with continuous density fi and support [c, c] ⊂ R+ . Bidders independently
submit bids and all bids are collected simultaneously. The contract is sold to the bidder with
the lowest bid, provided that her bid is not higher than the seller’s secret reserve price r, and
the winner receives her bid for the contract.15 A bidding strategy bi = bi (ci ; N ) specifies i’s bid
as a function of her cost, the number of bidders and their identity.
Bidder i has cost ci and maximizes her expected profits
πi (ci ; N ) = maxb≤r (bi − ci )
Y
(1 − Gj (b; N )),
(1)
j∈N \i
where the lowest bidder wins the contract and makes a profit equal to bi − ci and all other
bidders make zero profits. For the estimations, it is helpful to follow Guerre et al. (2000) and
write bidder i’s expected profit as a function of Gj instead of Fj , where Gj (b; N ) = Fj (b−1
j (b; N ))
16 As is standard in the literature,
is the probability that j will bid less than b and b−1
j (b; N ) = cj .
we focus on Bayesian Nash equilibria in pure bidding strategies. The first order condition for
i’s bidding problem is then
X
gj (bi ; N )
1
,
=
bi − ci
(1 − Gj (bi ; N ))
j∈N \i
(2)
which provides the basis for estimating bidders’ cost distributions. There is no explicit solution
to (2), but the first order conditions, together with the boundary conditions that bi (c; N ) = c for
all i uniquely characterize optimal bidding strategies. In equilibrium, the bidders use a markup
strategy and bid their values minus a shading factor that depends on the equilibrium behavior
of opponents (Maskin and Riley, 2000).17 ,18
15
In Austrian procurement auctions, the seller has the right to withdraw the auction only when the winning
bid is “contra bonos mores” (offending against good morals). In court, the seller would have to proof this by
providing a cost estimate that is based on standard commercial and professional principles and be able to show
that the winning bid is far higher than this estimate. We interpret this as a secret reserve price that is hardly
ever binding. In principle, we could model a secret (random) reserve price and estimate the distribution of sellers’
valuation as in Li and Perrigne (2003). In our sample, we however do not observe any rejected bids to obtain an
estimate for that distribution.
16
Note that – anticipating our empirical specification – the asymmetry across bidders in our paper comes from
variation in observable firm characteristics such as distance to the construction site or number of employees.
17
The conditions for a unique equilibrium are provided in (Lebrun, 2006; Maskin and Riley, 2003).
18
If there were only one bidder, which we do not observe in our data, the equilibrium strategy had to be
adjusted.
13
4.2
Estimation of Bidders’ Costs
To illustrate the econometric model and the structural estimation of the distribution of bidders’
construction costs, we use sealed-bid data to estimate the parameters of the theoretical model
as a function of auction and bidder characteristics making use of the approach developed by
Guerre et al. (2000). They suggest to estimate the distribution of bids in a first step and to
recover the distribution of bidders’ costs in a second step by using the first order-condition for
optimal bidding behavior. With the estimates of the distribution of bidders’ costs at hand, we
then calculate markups.
Let X denote the set of auction characteristics and Z denote the set of bidder characteristics.
We assume that both sets of characteristics are known to the econometrician and the bidders.
Such an assumption precludes the existence of characteristics unknown to the econometrician,
but known to bidders. We believe that due to the detailed data set including backlog, distance
and firm size, this assumption is plausible. We model N , the set of bidder identities, with
number of bidders N and cumulative characteristics of the other bidders that participate in the
auction.19 The list of variables denoting X , Z and N is given in Table 1.
Bidders also know their private cost ci . We denote the distribution of bidders’ costs as
Fi (·|X , Z), and assume that bidders’ costs are independent conditional on (X , Z). Given these
assumptions, one can write the distribution of bids as Gi (·|X , Z, N ).
Distribution of Bids. The first step of Guerre et al. (2000)’s approach to obtain an estimate
for the distribution of bidders’ cost is to estimate the distribution of bids. Their approach is
very general and allows the non-parametric identification and estimation of the distribution of
bidders’ cost. Here, we adopt a parametric approach. Conditional on the observable auction
and firm characteristics (X , Z), and the set of bidders identities N , the joint distribution of bids
in a given auction is the distribution Gi (·|X , Z, N ). We specify the Weibull distribution as the
distribution of bids:
Gi (bi |X , Z, N ) = 1 − exp
−
bi
λi (X , Z, N )
ρi (X ,Z ,N ) ,
19
(3)
In the empirical analysis, our N also includes the count of active construction firms described earlier as a
measure of potential bidders. Herewith, we account for differences between potential and participating bidders,
although bidders’ entry decisions are for simplicity not formalized in the theoretical model that describes our
main specification.
14
where λi (X , Z, N ) is the scale and ρi (X , Z, N ) is the shape of the Weibull distribution. We
parameterize the scale as λi (X , Z, N ) = λ0 +λX X +λZ Z+λN N and the shape as ρi (X , Z, N ) =
ρ0 + ρX X + ρZ Z + ρN N . We estimate the parameters of the model, (λ, ρ), by maximum
likelihood estimation.
Distribution of Costs. Assuming bidders behave as predicted by the model, the distribution
Fi (·|X , Z) is identified from the distribution of observed bids. The advantage of this approach
is that no differential equation has to be solved and no numerical integration has to be applied.
The estimation of bidders’ costs is directly derived from identification and can be expressed as
follows:
ĉi = φi (bi ; X , Z, N ) = bi − P
1
ĝj (bi |X ,Z ,N )
j∈N \i 1−Ĝj (bi |X ,Z ,N )
,
(4)
where Ĝj and ĝj are estimates of the distribution and the density of bidder j’s costs. Bidder
i’s estimated cost are a function of her equilibrium bid and the joint distribution of her rivals’
equilibrium bids. If we observe all bids and bidder identities, then the asymmetric independent
private values model is identified (Campo et al., 2003; Laffont and Vuong, 1996; Li and Zhang,
2013) and it is then straightforward to construct an estimate of φi given the estimates of Gi and
gi , i.e. Ĝi and ĝi .20,21,22 With the pseudo sample of bidders’ costs, ĉi , we are able to calculate
the distribution of bidders’ costs as
F̂i (ĉ|X , Z) = Ĝi (φ−1
i (ĉi ; X , Z, N )|X , Z, N ).
(5)
Finally, we assume a unique equilibrium or that the selected equilibrium is the same across
observations. If this is not case, we would not match the distribution characterizing a bidder’s
beliefs in a given auction, as the observed distribution of opponent bids would be a mixture of
those in each equilibrium.
20
Actually, Li and Zhang (2013) show identification in the affiliated value model with endogenous entry. For a
general discussion on identification in first-price auctions, see Athey and Haile (2005).
21
Note that in our case Gi also depends on firm characteristics, Z. Thus, we have to calculate Gj for all bidders
j in equation (4). This is different to other applications such as Athey et al. (2011) and Krasnokutskaya (2011).
In these applications there are at most two groups of bidders and Gj is equal among these groups.
22
As the reserve price in Austrian procurement auctions is secret, we do not need to account for it in the
estimations. Note that we observe all bids, thus there is also no selection in the data.
15
4.3
Dynamic model
Our main specification does not consider that firms may price into their bids also the lost option
value of winning today versus winning later. Thus, we may underestimate markups in both periods and only obtain a lower bound for the estimated drop in markups in the crisis. We illustrate
the econometric model of bidders’ dynamic decision based on the estimated distribution of their
construction cost following Jofre-Bonet and Pesendorfer (2003).23 They estimate a dynamic
game in which bidders play sequential equilibria and use symmetric Markovian strategies, i.e.,
given the state variables bidders are symmetric and follow the same bidding strategy once they
are in the same state.
Let Q = (X ,Z ′,N ) with Z ′ = Z \ (si , s−i ), where si is the backlog of bidder i and s−i the
backlog of all other firms.24 We define the hazard function of bids submitted by bidder i as
ĥ(.|Q, si , s−i ) =
ĝ(.|Q, si , s−i )
1 − Ĝ(.|Q, si , s−i )
(6)
,
where Ĝ and ĝ are the estimated distribution function and the estimated density function of
our main specification. Using (6), the first order condition of optimal bids yields the following
equations for privately known cost:
ĉi = φ(b|Q, si , s−i , h, V ) = b − P
−β
X
j∈N \i
1
j∈N \i ĥ(b|Q, sj , s−j )
ĥ(b|Q, sj , s−j )
P
l∈N \i ĥ(b|Q, sl , s−l )
[Vi (ω(Q, s, i)) − Vi (ω(Q, s, j))],
(7)
where V is the value function and ω the transition function. Equation (7) provides an explicit
expression of the privately known cost that involves the bid, the hazard function of bids and
the value function. Once we know the value function and the transition function, we are able to
back out cost from bids.
The transition function ω describes how the backlog evolves over time and is exactly defined
as in Jofre-Bonet and Pesendorfer (2003). The key idea to obtain an estimate for the value
23
For an implementation of a dynamic first-price auction model with endogenous entry and unobserved heterogeneity, see Balat (2012).
24
Note, in the econometric specification of the static model, the backlog is an element of Q.
16
function V is to draw a random sample of states and numerically solve the explicit expression
for the value function derived in Jofre-Bonet and Pesendorfer (2003) for every bidder on this grid
and approximate it by a polynomial outside the grid. We draw 200 contracts randomly. For the
quadratic polynomial, we run a robust regression including firm fixed effects. Our simulation
sample includes 398 bidders. As a robustness check, we also run the static specification with
this (reduced) sample and obtain markups that are very close to the original ones.
4.4
Expected effects of a merger
After a merger, the incentive to shade bids above one’s cost increases. A bid is equal to the
expected cost of the competitors conditional on the competitors’ costs being less than the bid. If
there is less competition in the sense that there are fewer actual bidders in an auction, the markup
goes up, because the expected conditional cost of competitors decreases after the absorption of
one of the competitors. This implies that bidders’ expected rents go up in equilibrium and gives
the first hypothesis:
Hypothesis 1: The Lerner index of merging firms goes up post merger, i.e. there is a
market power effect due to the merger.
Costs of the combined entity can decrease after the merger. This efficiency effect drives the
bids down after the merger. In general, both effects can be present at the same merger, thus
our our second hypotheses is:
Hypothesis 2: The marginal costs of merging firms go down post merger, i.e. there is an
efficiency effect due to the merger.
The net externality on the market depends on the relative strength of these two effects, but
standard oligopoly models suggest a positive net effect: under quantity competition or price
competition with differentiated goods, the merged entity finds it — absent substantial efficiency
gains — optimal to reduce its production (Deneckere and Davidson, 1985; Farrell and Shapiro,
1990). In reaction, the competitors expand their output, but by a lesser amount than the insiders
decrease their output. In the new equilibrium the competing firms sell a higher quantity at a
higher price, which clearly is profitable. Farrell and Shapiro (1990) consider the possibility that
the merged entity experiences efficiency gains through economies of scale or learning. Using
simple Cournot examples, they show that these efficiency gains would have to be relatively
17
large to make the merger unprofitable for outsiders. In Bertrand oligopolies with differentiated
products (Deneckere and Davidson, 1985) mergers are profitable for both insiders and outsiders,
but again the free-riding outsiders benefit more than the merging firms. Thus from both kinds
of standard IO oligopoly models (quantity competition and price competition with differentiated
products), we would infer a positive externality on the profits of the rivals in the relevant market
concerned by the merger, as long as no substantial efficiency gains are achieved.
Prices rise (here: bids) when the market power effect outweighs the efficiency effect. Ex-ante,
we do not know which effect dominates, and this is what is tested via hypotheses one and two.25
5
Estimation results and counterfactual analysis
Table 7 shows the estimation results of the determinants of the bid distribution. Column (1)
shows the estimates for the scale parameter λ, and column (2) for the shape parameter ρ.
The standard errors are shown in parentheses below. As expected, the estimated effect of the
log of the number of bids is negative. More bidders increase the competitive intensity in the
auction. We use the log of the number of bidders to account for the diminishing influence of an
additional bid as the number of bidders grows. Own backlog and the backlog of the other bidders
measures firms’ capacity constraints. The estimated coefficients of both backlog measures are
positive. If their own backlog is higher, firms bid higher since they have higher opportunity
costs; and firms strategically bid higher when they know that their competitors have higher
backlogs, and therefore higher costs. We also observe that new contracts in the construction
sector increases firm’ bids implying that a higher level of utilization in the industry let firms bid
more aggressively.
One of the main determinants of bidders’ behavior is the engineer estimate. It is the cost
estimate that engineers of the firm providing the data present before the auction and in our model
it mainly describes the heterogeneity of the projects. We find that the coefficient significantly
25
The main underlying assumption for hypothesis one is that a decrease in the number of bidders yields higher
procurement prices and higher bidder revenues. This is a standard competition argument, but need not always
be true. In a model with common values, informational asymmetries such as the “winners’ curse” may offset the
above argument. For example, Somaini (2011) and Hong and Shum (2002) provide evidence of interdependent
valuations in the construction industry. However, even in an independent private values model as we apply here,
bidder asymmetries or costly bidding may also lead to the effect that more bidders could lead to reduced price
competition. Cantillon (2008) show that the composition of strong versus weak bidders is the relevant factor. Li
and Zheng (2009) show that it is possible that as the number of competitors increases bidding may become less
aggressive.
18
different from zero and positive. We may also interpret the coefficient as a markup on cost. The
log of the number of employees, our firm specific size measure, has a significant positive and
diminishing influence on the bid level. We also observe that firms react to the travel distance of
their competitors. The larger it is, the higher the bids. We also find that the average distance
of the bidders in the auction is significant. We interpret these as strategic variables in a limit
pricing context: the further competitors are away from the project site, the higher their costs.
When competitors are further away, a firm can submit higher bids. Besides that, some control
variables are included: same postal code of firm and project site, same district and same state;
open format auctions; heavy construction auctions; and bidder is a general contractor.
Using the first order conditions (4) from the static model and (7) from the dynamic model, we
back out cost and estimate their determinants using a Weibull model for each. The estimated
cost functions are depicted in Table 8. We find similar, but somewhat quantitative different
estimates for the two cost models. We find that backlog and new contracts in the construction
increase firms’ cost. The driving distance to the construction site increases cost at a diminishing
rate. We also find that larger firms have lower cost. Other controls such as same postal code of
firm and project site, same district and same state; open format auctions; heavy construction
auctions; and if the bidder is a general contractor also influence firms’ cost.
The structural models allow us to analyze some particular aspects of how the mergers affect
bidding. We analyze the supposed channels of merger effects by reporting two experiments.
The experiments apply to a merger of the firms ranked first and second. Mean effects can be
interpreted as the average merger effect for all firm pairs ever first or second. Table 9 contains
the effect for all auctions: after reporting observed static and dynamic markups and winning
bids in the first row, its second and third row show the effect of two experiments.
Experiment 1: The number of bidders decreases by one. The first-ranked bidder maintains
the same cost level, but can bid the predicted bid — after the reduction in the number of bidders
— of the second-ranked bidder.
Experiment 2: Again, the number of bidders decreases by one. From the following cost
shifters, the combined firm can choose the minimum: backlog, distance, location. For the new
predicted bids, bidders react to the changed number of bidders and to the changed cost structure.
While the first experiment evaluates the market power effect, the second experiment in-
19
troduces a change of costs and a reaction of the competitors to the changed bidder and cost
structure. The basic, traditional idea is that the merged firm has two cost draws and can choose
the minimum. The cost draw can depend on cost determinants, as applied recently in Li and
Zhang (2013), who use hauling distance as their firm-specific cost shifter.
Table 9 shows a very large effect for the first experiment on winning bids. The bid increases
by 8.5%, which hurts the seller, while markups go up by 4.9% for the static model and by 4.8%
for the dynamic model markups benefiting the winning firm. Cost reductions – which might be
implemented only a while after the merger is completed – from the second experiment result in
an average bid that is 1.4% lower than the observed bids. Static markups of the winners are
6.0% higher relative to the markups based on observed bids and 7.0% higher in the dynamic
case. The results show that cost efficiencies would more than offset the market power effect and
benefit the seller. Firms profit by even higher markups. However, while the market power is
immediate, the effect from cost efficiencies may materialize only later.
For comparison, we also provide the counterfactual results for those auctions only where
both the acquirer and target bid and, therefore, take place before the merger. We show these
results in Table 10.
6
Ex-post analysis
With the results from the structural auction model we can now apply the regression analysis of
difference comparisons to markups and cost estimates. Table 11 compares markups and costs to
a merging firm’s own history. Measuring the merger effect relative to a firm’s own history clearly
ignores changes in auction characteristics or other factors that effect all bidders within auctions.
A comparison of the change simply on the firm level is informative in comparison to the results
below, where we will account for the new competitive environment that all bidders experience
by the introduction of auction-level comparison. Table 12 measures the effects compared to the
competitors in the auctions (Top panel), and also to the auctions where merging firms did not
participate (bottom panel).
Markups only once change significantly in the within-firm historical comparison of Table 11:
for acquirer 2 and the static model. Backed-out costs are lower after transaction 7. Comparing
the effects on markups and costs to the other firms in the auction, all firm-specific effects
20
disappear (top panel of Table 12). For the total effect of each transaction, we look at the whole
auction: a merger treats not only the merging firm in an auction, but also the competitors,
who face a different environment. Auction level effects can be seen in the bottom panel of
Table 12. Except for transaction 5, all of the auction level effects on participants’ markups are
significant and positive, both in the static and dynamic markups. Costs are also lower relative
to all other auctions when markups are significantly higher, with the exception of acquirer 9.
The transactions have not brought about a positive effect specifically to the acquirer, or target,
but to all participants of affected auctions.
Table 11 about here
Table 12 about here
These results are in line with a model where the merger affects all firms in the same oligopoly
market — here: auction — and the merged firm does not experience cost reductions. Only when
the difference is made in comparison to auctions not directly affected through the participation
of a merging party, markups differ. All significant differences are positive. The results are never
significantly negative.
7
Conclusions
In this paper, we analyse mergers in bidding markets. We utilize data from procurement auctions
in the Austrian construction sector and estimate models of first-price sealed-bid auctions. Based
on estimated cost and markups, we run merger simulations and disentangle the market power
effect from potential cost efficiencies. In addition, we compare static and dynamic models of
first price auctions. Finally, we compare the outcomes of the merger simulations with the effects
of actual mergers.
Our main results are that while mark-ups increase post merger, so does efficiency, i.e. costs
go down. We find this pattern in both the simulated counterfactual results as well as after the
actually observed mergers. The net effect on the price (i.e. the bid) in both cases appears to
be close to neutral. Thus, while consumer surplus does not appear to be increased, producer
21
surplus and therefore overall welfare seems to go up post merger. In future research, we plan to
analyze the likely effects of entry.
References
O. Ashenfelter and D. Hosken. The effect of mergers on consumer prices: Evidence from five
selected case studies. Working Paper 13859, National Bureau of Economic Research, 2008.
S. Athey and P. Haile. Nonparametric approaches to auctions. in J.J. Heckman: The Handbook
of Econometrics, Volume 6, Elsevier Science, 2005.
S. Athey, J. Levin, and E. Seira. Comparing open and sealed bid auctions: Evidence from timber
auctions. Quarterly Journal of Economics, 126(1):207–257, 2011.
S. Athey, D. Coey, and J. Levin. Set-asides and subsidies in auctions. American Economic
Journal: Microeconomics, 5(1):1–27, 2013.
P. Bajari and A. Hortaçsu. Are structural estimates of auction models reasonable? evidence
from experimental data. Journal of Political Economy, 113(4):703–741, 2005.
J. Balat. Highway procurement and the stimulus package: Identification and estimation of
dynamic auctions with unobserved heterogeneity. Working Paper, 2012.
S. Borenstein. Airline mergers, airport dominance, and market power. American Economic
Review Papers and Proceedings, 80(2), 1990.
S. Campo, I. Perrigne, and Q. Vuong. Asymmetry in first-price auctions with affiliated private
values. Journal of Applied Econometrics, 18(2):179–207, 2003.
E. Cantillon. The effect of bidders’ asymmetries on expected revenue in auctions. Games and
Economic Behavior, 62(1):1–25, 2008.
P. Davis and E. Garcés. Quantitative techniques for competition and antitrust analysis. Princeton
University Press, 2009.
22
R. Deneckere and C. Davidson. Incentives to form coalitions with bertrand competition. Rand
Journal of Economics, pages 473–486, 1985.
D. Dyer, J. H. Kagel, and D. Levin. A comparison of naive and experienced bidders in common
value offer auctions: a labratory analysis. Economic Journal, 99(394):108–115, 1989.
R. J. Epstein and D. L. Rubinfeld. Merger simulation: A simplified approach with new applications. Antitrust Law Journal, 69:883, 2001.
J. Farrell and C. Shapiro. Horizontal mergers: an equilibrium analysis. American Economic
Review, pages 107–126, 1990.
D. Focarelli and F. Panetta. Are mergers beneficial to consumers? evidence from the market
for bank deposits. American Economic Review, pages 1152–1172, 2003.
L. Froeb, S. Tschantz, and P. Crooke. Bertrand competition with capacity constraints: mergers
among parking lots. Journal of Econometrics, 113(1):49–67, 2003.
E. Guerre, I. Perrigne, and Q. Vuong. Optimal nonparametric estimation of first-price auctions.
Econometrica, 68(3):525–574, 2000.
K. Gugler, M. Weichselbaumer, and C. Zulehner. Competition in the economic crisis: Analysis
of procurement auctions. Europen Economic Review, forthcoming, 2014.
J. S. Hastings. Vertical relationships and competition in retail gasoline markets: Empirical
evidence from contract changes in southern california. American Economic Review, pages
317–328, 2004.
J. S. Hastings and R. J. Gilbert. Market power, vertical integration and the wholesale price of
gasoline. Journal of Industrial Economics, 53(4):469–492, 2005.
J. Hausman, G. Leonard, and J. D. Zona. Competitive analysis with differenciated products.
Annales d’Economie et de Statistique, pages 159–180, 1994.
K. Hendricks and H. J. Paarsch. A survey of recent empirical work concerning auctions. Canadian
Journal of Economics, 28(2):403–426, 1995.
H. Hong and M. Shum. Increasing competition and the winner’s curse: Evidence from procurement. Review of Economic Studies, 69(4):871–898, 2002.
23
D. Hosken, L. Silvia, and C. Taylor. Does concentration matter? measurement of petroleum
merger price effects. American Economic Review Papers and Proceedings, 101(3):45–50, 2011.
J.-F. Houde. Spatial differentiation and vertical mergers in retail markets for gasoline. American
Economic Review, 102(5):2147–2182, 2012.
D. Huang, C. Rojas, and F. Bass. What happens when demand is estimated with a misspecifed
model? Journal of Industrial Economics, 56(4):809–839, 2008.
M. Ivaldi and F. Verboven. Quantifying the effects from horizontal mergers in european competition policy. International Journal of Industrial Organization, 23(9):669–691, 2005.
M. Jofre-Bonet and M. Pesendorfer. Estimation of a dynamic auction game. Econometrica, 71
(5):1443–1489, 2003.
B. Jovanovic and P. L. Rousseau. The q-theory of mergers. American Economic Review, 92(2):
198–204, 2002.
E. H. Kim and V. Singal. Mergers and market power: Evidence from the airline industry.
American Economic Review, 83(3):549–569, 1993.
P. Klemperer. Auctions: Theory and Practice. Princeton University Press, 2004.
P. Klemperer. Bidding Markets. UK Competition Commission, 2005.
E. Krasnokutskaya. Identification and estimation in highway procument auctions under unobserved auction heterogeneity. Review of Economic Studies, 78(1):293–327, 2011.
E. Krasnokutskaya and K. Seim. Bid preference programs and participation in highway procurement auctions. American Economic Review, 101(6):2653–2686, 2011.
V. Krishna. Auction Theory. Academic Press, San Diego, 2nd edition, 2009.
J.-J. Laffont. Game theory and empirical economics: The case of auction data. European
Economic Review, 41:1–35, 1997.
J.-J. Laffont and Q. Vuong. Structural analysis of auction data. The American Economic
Review, pages 414–420, 1996.
24
B. Lebrun. Uniqueness of the equilibrium in first-price auctions. Games and Economic Behavior,
55(1):131–151, 2006.
T. Li and I. Perrigne. Timber sale auctions with random reserve prices. Review of Economics
and Statistics, 85(1):189–200, 2003.
T. Li and B. Zhang. Affiliation and entry in first-price auctions with heterogeneous bidders:
Analysis of merger effects. American Economic Journal: Microeconomics, forthcoming, 2013.
T. Li and X. Zheng. Entry and competition effects in first-price auctions: Theory and evidence from procurement auctions. Review of Economic Studies, 76(4):1397–1429, 2009. URL
http://ideas.repec.org/a/oup/restud/v76y2009i4p1397-1429.html.
H. G. Manne. Mergers and the market for corporate control. Journal of Political Economy,
pages 110–120, 1965.
R. Marris. The Economic Theory of Managerial Capitalism. Glencoe, Ill.: Free Press, 1964.
E. Maskin and J. Riley. Asymmetric auctions. Review of Economic Studies, 67(3):413–438,
2000.
E. Maskin and J. Riley. Uniqueness of equilibrium in sealed high-bid auctions. Games and
Economic Behavior, 45(2):395–409, 2003.
D. C. Mueller. A theory of conglomerate mergers. Quarterly Journal of Economics, 83(4):
643–659, 1969.
D. C. Mueller. Mergers and the market for corporate control. In D. C. Mueller, editor, The
Oxford Handbook of Capitalism. Oxford University Press, USA, 2012.
A. Nevo. Mergers with differentiated products: The case of the ready-to-eat cereal industry.
Rand Journal of Economics, pages 395–421, 2000.
A. Nevo and M. D. Whinston. Taking the dogma out of econometrics: Structural modeling and
credible inference. Journal of Economic Perspectives, pages 69–81, 2010.
C. Peters. Evaluating the performance of merger simulation: Evidence from the us airline
industry. Journal of Law and Economics, 49(2):627–649, 2006.
25
R. Roll. The hubris hypothesis of corporate takeovers. Journal of Business, pages 197–216,
1986.
M. Slade. Merger simulations of unilateral effects: What can we learn from the uk brewing industry? In B. Lyons, editor, Cases in European Competition Policy: The Economic Analysis.
Cambridge University Press, 2009.
P. Somaini. Competition and interdependent costs in highway procurement. MIT, mimeo, 2011.
G. J. Stigler. Monopoly and oligopoly by merger. American Economic Review, pages 23–34,
1950.
M. C. Weinberg and D. Hosken. Evidence on the accuracy of merger simulations. Review of
Economics and Statistics, 95(5):1584–1600, 2013.
G. J. Werden and L. M. Froeb. The effects of mergers in differentiated products industries:
Logit demand and merger policy. Journal of Law, Economics, and Organization, 10(2):407–
426, 1994.
G. J. Werden and L. M. Froeb. Unilateral competitive effects of horizontal mergers. In P. Buccirossi, editor, Advances in the Economics of Competition Law. MIT Press, 2005.
J. F. Weston. The nature and significance of conglomerate firms. St. John’s Law Review, 44:
66–80, 1970.
O. E. Williamson. Corporate Control and Business Behavior: An Inquiry into the Effects of
Organization Form on Enterprise Behavior. Englewood Cliffs, N.J.: Prentice Hall, 1970.
26
A
Appendix: Tables
Table 1: Variable descriptions
Variable
Description
Log(number of bidders) Log of number of bidders in the auction.
Backlog
Backlog variable, standardized by firm mean and standard deviation.
Backlog sum
Sum of backlog of the other bidders in the auction.
New orders
Gross inflow of new contracts, countrywide, per month (mill. 2006 euros).
Engineer estimate
Engineer cost estimate for the construction project.
Log(employees)
Log of the number of employees of a bidder.
KM
Travel distance in kilometers from a bidder’s location to the project site.
KM average
Average of all other, competing bidders’ distances between their location
and the project site.
KM sum
Sum of all other, competing bidders’ distances between their location
and the project site.
Same postal
Bidder location has the same postal code as project site.
Same district
Bidder resides in the district of the project site.
Same state
Bidder resides in the state of the project site.
Heavy construction
Dummy for auctions in heavy construction sector.
General contractor
Dummy for bidders serving as “general contractors”.
Open format
Dummy for auctions following the “open procedure”.
No. of potent. bidders Number of potential bidders.
27
Table 2: Summary statistics
All auctions
mean
s.d.
Firms
Auctions
Bids
Number of firms
Employees
Total Assets (mill.)
Number of auctions
Winning bid (mill.)
Engineer estimate (mill.)
Order flows (mill.)
Actual bidders
Potential bidders
Number of bids
Bid (mill.)
Backlog
Distance (km)
Same postal code
Same district
Same state
Subsample
mean
s.d.
1,502
63.66
22.24
358.21
21.79
124
307.73
11.84
919.42
64.83
3,794
1.90
2.43
1726.93
7.56
28.86
5.85
6.48
183.10
3.08
11.60
73
8.16
6.13
1704.21
7.94
27.41
13.01
10.64
210.40
2.89
7.71
14,845
2.16
0.09
130.53
0.04
0.14
0.51
5.48
0.82
121.73
0.20
0.35
0.49
356
5.64
-0.19
162.37
0.04
0.17
0.45
11.69
0.65
140.93
0.19
0.38
0.49
Notes: Monetary values in 2006 euros. Subsample includes all auctions in which merging parties jointly
participated before they merged.
28
Table 3: List of mergers
Transaction
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Date
2006-02-01
2006-07-02
2006-07-12
2007-04-18
2007-09-21
2007-10-12
2008-02-15
2008-05-20
2008-11-07
2008-12-06
2007-01-05
2007-02-01
2008-07-15
2008-09-15
2008-11-07
2008-02-09
2008-04-08
Acquirer Bids
before
after
3
538
0
1195
801
1
123
9
1826
186
903
58
2294
2435
1826
498
539
Target Bids
before
after
193
3013
0
2356
1083
0
159
1
728
96
2648
224
1257
1116
728
451
410
29
2
18
0
19
1
0
3
1
10
101
0
3
1
8
3
0
0
Type
33
38
1
0
3
0
2
0
2
24
0
8
0
0
1
1
0
full
full
full
full
full
full
full
full
full
full
majority
majority
majority
majority
majority
parital
parital
Table 4: Acquirer and target bidding in the same auctions
Transaction
Pre
Acq.
Both
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
0
18
0
19
1
0
0
0
5
26
0
0
1
8
2
0
0
Tar.
3
520
0
1176
800
1
123
9
1821
160
903
58
2293
2427
1824
498
539
2
0
0
0
0
0
3
1
5
75
0
3
0
0
1
0
0
Notes: Acquirer, target and both are mutually exclusive.
30
Post
Acq.
Both
1
6
0
0
0
0
0
0
2
1
0
0
0
0
0
0
0
Tar.
192
3007
0
2356
1083
0
159
1
726
95
2648
224
1257
1116
728
451
410
32
32
1
0
3
0
2
0
0
23
0
8
0
0
1
1
0
Table 5: Acquirer and target before and after the merger
Merger
2
Firm
Acq.
Tar.
4
Acq.
5
Acq.
7
Acq.
Tar.
9
Acq.
10
Acq.
Tar.
Bid
121.42
(230.50)
-390.23
(320.77)
-119.82
(212.37)
-415.00
(856.50)
-945.12*
(424.04)
-2000.37
(1987.74)
112.00
(313.91)
-755.90*
(365.06)
-2429.44
(2650.10)
Backlog
Emp.
1.13*
(0.04)
-0.22
(0.17)
1.33*
(0.03)
0.81*
(0.10)
0.74*
(0.05)
-0.07
(0.09)
0.83*
(0.06)
0.50*
(0.05)
1.44*
(0.20)
1089.78*
(152.19)
-12.53
(12.62)
718.16*
(127.53)
146.47
(173.65)
-56.18
(51.13)
152.27*
(1.12)
137.64
(148.52)
-162.43*
(60.56)
69.52
(114.69)
Num.
Bids
-0.40*
(0.19)
-0.63
(0.86)
-0.02
(0.16)
0.79*
(0.36)
0.20
(0.24)
0.33
(1.36)
1.05*
(0.35)
0.79*
(0.19)
1.71+
(0.96)
Winning
bid
144.92
(216.57)
-359.38
(302.98)
-110.69
(196.09)
-206.18
(770.80)
-745.84*
(349.27)
-1597.24
(1734.60)
104.31
(284.70)
-631.65*
(311.91)
-1925.45
(2259.52)
MLOTT
0.62
(0.57)
1.92
(2.17)
-0.00
(0.47)
0.35
(0.94)
-0.58
(0.52)
0.54
(0.91)
-2.28*
(0.96)
0.15
(0.43)
-1.39
(2.21)
Notes: An asterisk indicates significant differences at the 5 percent level, a plus the 10 percent level of
a two sided t-test. MLOTT stands for Money left on the table and is expressed in percent. Bids and
winning bids in thousands of 2006 Euros.
31
Table 6: Acquirer and target, non-merging firms and unaffected auctions
Merger
Firm
Bid
Backlog
Emp.
Num.
Bids
Winning
bid
MLOTT
Acquirer/target difference in difference
2
Acq.
26.67
(321.59)
Tar.
-74.91
(1666.10)
4
Acq.
-87.55
(256.11)
5
Acq.
-38.20
(760.37)
7
Acq.
-317.96
(403.97)
Tar.
-186.61
(4739.36)
9
Acq.
-47.00
(523.43)
10
Acq.
-256.63
(353.78)
Tar.
277.16
(1485.96)
0.88*
(0.05)
-0.53+
(0.27)
0.98*
(0.04)
0.19
(0.12)
0.32*
(0.06)
-0.14
(0.76)
0.61*
(0.09)
0.11+
(0.06)
0.95*
(0.25)
854.04*
(147.57)
513.83
(942.52)
578.91*
(120.89)
513.01
(431.08)
-170.19
(228.94)
-5.57
(2683.76)
305.43
(293.02)
-92.44
(200.41)
496.73
(848.09)
0.35
(0.22)
-0.13
(1.12)
0.09
(0.17)
-0.20
(0.51)
-0.07
(0.27)
0.07
(3.20)
0.12
(0.34)
0.10
(0.24)
-0.72
(1.01)
55.78
(294.11)
-62.06
(1523.15)
-70.52
(234.16)
73.17
(695.23)
-198.61
(369.34)
23.79
(4332.56)
-14.08
(478.52)
-186.14
(323.44)
531.55
(1359.27)
0.38
(0.53)
0.29
(2.72)
0.01
(0.42)
-0.27
(1.24)
-0.52
(0.66)
-0.01
(7.74)
-0.65
(0.85)
-0.29
(0.58)
0.04
(2.45)
Difference for whole auction
2
Acq.
-319.11+
(167.65)
Tar.
-563.51
(601.85)
4
Acq.
-490.30*
(127.71)
5
Acq.
-582.65*
(288.17)
7
Acq.
-927.76*
(150.47)
Tar.
-1918.03
(1510.17)
9
Acq.
98.47
(184.98)
10
Acq.
-784.95*
(142.89)
Tar.
-2815.72*
(495.68)
-0.18*
(0.03)
-0.09
(0.10)
-0.25*
(0.02)
0.17*
(0.05)
-0.02
(0.02)
-0.38
(0.24)
-0.14*
(0.03)
0.05*
(0.02)
0.14+
(0.08)
-59.11
(76.93)
-874.40*
(340.47)
-102.26+
(60.28)
-443.21*
(163.38)
21.96
(85.27)
65.39
(855.16)
-77.34
(103.55)
43.72
(80.95)
-336.24
(282.90)
-0.96*
(0.11)
-0.19
(0.41)
-0.12
(0.09)
0.67*
(0.19)
0.16
(0.10)
0.12
(1.02)
0.18
(0.12)
0.12
(0.10)
1.80*
(0.34)
-290.82+
(153.32)
-529.36
(550.21)
-467.65*
(116.76)
-466.60+
(263.48)
-818.51*
(137.57)
-1721.02
(1380.54)
51.07
(169.11)
-711.17*
(130.64)
-2562.65*
(453.42)
1.58*
(0.27)
2.13*
(0.98)
0.79*
(0.21)
0.80+
(0.47)
-0.17
(0.25)
0.48
(2.47)
-1.59*
(0.30)
0.99*
(0.23)
-1.15
(0.82)
Notes: Results for the coefficient of “post x firm” in the top panel; for the coefficient of “post x auction”
in the bottom panel. An asterisk indicates significant differences at the 5 percent level, a plus the 10
percent level of a two sided t-test. MLOTT stands for Money left on the table and is expressed in percent.
Bids and winning bids in thousands of 2006 Euros.
32
Table 7: Determinants of the distribution of bids
Log(Number of bidders)
Backlog
Backlog of other bidders
New contracts
Engineer estimate
Log(Employees)
Distance
Distance squared
Distance of other bidders
Average distance
Same postal code
Same district
Same federal state
Heavy construction
General constructor
Open format
Constant
Observations
Scale, λ
(1)
Shape, ρ × 1000
(2)
-16,401.60**
(3,402.064)
2,698.89**
(713.525)
1,221.21**
(232.840)
22.64**
(2.797)
1.13**
(0.004)
-430.90
(352.980)
74.58**
(19.219)
-0.15**
(0.036)
37.79**
(4.213)
-86.53**
(21.867)
-7,003.49**
(2,595.635)
2,279.17
(1,722.752)
-1,431.48
(2,104.922)
5,641.34
(2,979.309)
40,988.90**
(10,132.839)
10,142.90**
(1,253.750)
-16,772.86
(9,699.732)
0.32**
(0.086)
-0.04
(0.021)
0.03**
(0.005)
0.00*
(0.000)
0.00**
(0.000)
0.14**
(0.010)
-0.00
(0.000)
0.00
(0.000)
0.00
(0.000)
-0.00*
(0.000)
-0.16*
(0.076)
0.05
(0.038)
-0.16**
(0.051)
-0.07
(0.041)
0.92**
(0.101)
0.09
(0.052)
1.80**
(0.245)
14,845
Notes: Column (1) shows the coefficients estimated for the scale parameter λ of the Weibull distribution of bids.
Column (2) is for the shape parameter ρ. Standard errors are shown in parentheses below the parameter estimates.
∗ (∗∗) stands for significance at the 5% (1%) level.
33
Table 8: Determinants of the distribution of costs – static and dynamic model
Static model
Scale, λ
Shape, ρ × 1000
(1)
(2)
Backlog
New contracts
Engineer estimate
Log(Employees)
Distance
Distance squared
Same postal code
Same district
Same federal state
Heavy construction
General constructor
Open format
Constant
Observations
3,140.02**
(766.776)
17.35**
(2.713)
1.07**
(0.004)
-994.40**
(374.112)
93.29**
(21.735)
-0.20**
(0.038)
-7,365.59**
(2,617.645)
3,672.62*
(1,691.955)
1,489.73
(2,427.548)
3,079.74
(3,236.496)
24,332.85*
(11,078.004)
14,889.31**
(1,244.244)
-27,229.38**
(6,130.268)
0.01
(0.016)
-0.00*
(0.000)
0.00**
(0.000)
0.07**
(0.009)
0.00**
(0.000)
-0.00**
(0.000)
-0.08
(0.052)
-0.03
(0.038)
0.03
(0.044)
-0.07*
(0.030)
0.50**
(0.071)
0.19**
(0.039)
2.46**
(0.143)
14,697
Dynamic model
Scale, λ
Shape, ρ × 1000
(3)
(4)
2,885.31**
(965.580)
19.31**
(4.021)
1.05**
(0.005)
-1,152.09**
(428.537)
55.61
(28.943)
-0.12*
(0.050)
-7,150.79**
(2,758.028)
2,413.30
(2,389.599)
2,046.57
(3,088.812)
-2,314.48
(4,459.476)
15,134.19
(10,645.471)
12,998.82**
(1,728.480)
-28,300.23**
(8,270.408)
0.01
(0.022)
0.00
(0.000)
0.00**
(0.000)
0.05**
(0.012)
0.00**
(0.000)
-0.00**
(0.000)
-0.06
(0.096)
-0.06
(0.061)
0.14*
(0.060)
-0.16**
(0.041)
0.32**
(0.089)
0.07
(0.053)
2.12**
(0.192)
13,444
Notes: Columns (1) and (3) shows the coefficients estimated for the scale parameter λ of the Weibull distribution
of costs; columns (2) and (4) for the shape parameter ρ. Standard errors are shown in parentheses below the
parameter estimates. ∗ (∗∗) stands for significance at the 5% (1%) level.
34
Table 9: Estimated and counterfactual markups – all auctions
Static model
(1)
(2)
winning winning
bids
markups
Observed
2342.4
(170.6)
21.83
(0.54)
2343.6
(170.7)
23.85
(0.57)
2472.9
8.48%∗∗
26.75
4.92∗∗
2474.1
8.48%∗∗
28.64
4.80∗∗
2312.2
-1.35%∗∗
27.78
5.95∗∗
2318.5
-1.06%∗∗
30.83
6.98∗∗
Experiment 1: -1n.
Difference
Experiment 2: -1n./∆cost
Difference
Dynamic model
(3)
(4)
winning winning
bids
markups
Notes: Markups in percent. Winning bids in thousand Euro.
Table 10: Estimated and counterfactual markups – subsample
Static model
(1)
(2)
winning winning
bids
markups
Dynamic model
(3)
(4)
winning winning
bids
markups
Observed
7119.0
(2338.5)
14.50
(3.32)
7119.0
(2338.5)
14.50
(3.32)
Experiment 1: -1n.
Difference
7431.7
5.45%∗∗
18.41
3.91∗∗
7431.7
5.45%∗∗
18.41
3.91∗∗
7134.9
0.05%
18.58
4.08∗∗
7137.1
0.10%∗∗
19.05
4.55∗∗
Experiment 2: -1n./∆cost
Difference
Notes: Markups in percent. Winning bids in thousand Euro.
35
Table 11: Evaluation of markups and costs, before/after
Merger
Firm
µ
2
Acq.
2.58*
(1.28)
5.89
(4.94)
0.61
(1.03)
-0.01
(2.62)
0.55
(1.32)
2.63
(1.76)
0.16
(1.08)
-1.89
(2.25)
Tar.
4
Acq.
5
Acq.
7
Acq.
9
Acq.
10
Acq.
Tar.
Static
Cost
241.33
(291.46)
-476.16
(354.07)
-63.89
(255.02)
85.67
(1237.04)
-1799.82*
(779.69)
52.51
(352.29)
-662.81
(526.92)
-1564.96
(2551.12)
µ
Dynamic
Cost
0.82
(1.31)
5.83
(4.91)
-0.51
(1.06)
1.11
(3.52)
2.19
(1.72)
2.16
(1.92)
1.63
(1.41)
-1.87
(2.25)
268.61
(289.60)
-476.22
(354.07)
-43.64
(252.50)
38.34
(1154.84)
-1815.16*
(779.25)
42.37
(349.48)
-686.80
(526.65)
-1565.08
(2551.14)
Notes: An asterisk indicates significant differences at the 5 percent level, a plus the 10 percent level of a
two sided t-test. Costs in thousands of 2006 Euros.
36
Table 12: Evaluation, effects restricted to target segments
Merger
Firm
µ
Static
Cost
µ
Dynamic
Cost
Acquirer/target difference in difference
2
Acq.
Tar.
4
Acq.
5
Acq.
7
Acq.
9
Acq.
10
Acq.
Tar.
0.75
(1.20)
5.28
(5.92)
0.06
(0.97)
0.27
(3.76)
-1.20
(1.91)
1.54
(1.87)
-0.75
(1.56)
-0.89
(5.18)
94.05
(398.41)
-152.46
(1967.60)
-106.74
(322.19)
-25.02
(1246.35)
-452.30
(632.62)
-95.11
(621.75)
-162.78
(519.63)
977.51
(1708.03)
-0.62
(1.37)
6.01
(6.75)
-1.07
(1.10)
3.18
(4.28)
0.22
(2.17)
1.18
(2.12)
0.80
(1.78)
0.15
(5.89)
109.47
(401.99)
-167.32
(1984.62)
-88.44
(325.15)
-193.39
(1257.30)
-493.74
(638.14)
-106.65
(627.16)
-204.65
(524.16)
931.37
(1723.15)
1.78*
(0.84)
-0.86
(2.16)
3.12*
(0.61)
0.96
(1.46)
2.75*
(0.71)
3.27*
(0.71)
2.78*
(0.63)
0.19
(1.79)
-734.87*
(278.29)
-766.10
(718.63)
-539.52*
(203.36)
-150.29
(484.31)
-1816.80*
(235.83)
-40.86
(235.11)
-887.69*
(211.12)
-2749.40*
(591.29)
2.16*
(0.96)
-1.09
(2.46)
3.58*
(0.70)
-0.65
(1.66)
3.13*
(0.81)
3.56*
(0.80)
3.25*
(0.72)
-0.58
(2.04)
-766.56*
(280.76)
-772.87
(724.85)
-577.20*
(205.17)
-48.78
(488.57)
-1810.18*
(237.90)
-43.58
(237.18)
-871.87*
(213.01)
-2705.76*
(596.53)
Difference for whole auction
2
Acq.
Tar.
4
Acq.
5
Acq.
7
Acq.
9
Acq.
10
Acq.
Tar.
Notes: Results for the coefficient of “post x firm” in the top panel; for the coefficient of “post x auction”
in the bottom panel. An asterisk indicates significant differences at the 5 percent level, a plus the 10
percent level of a two sided test. Costs in thousands of 2006 Euros.
37