Quantifying the effects from horizontal mergers in
European competition policy
Marc Ivaldi a,c,d,*, Frank Verboven b,d,1
a
University of Toulouse, France
b
University of Leuven, Belgium
c
EHESS, Toulouse, France
d
CEPR, London, United Kingdom
Available online 23 September 2005
Abstract
This paper starts from a recent case to study how merger analysis in Europe may be potentially improved.
Starting from the geographic market definition in the Merger Decision, we formulate and estimate an
oligopoly model with differentiated products. Based on the estimates, we compare several alternative market
power tests: a hypothetical market power test, an actual market power test and a comparative market power
test. Our comparison of these alternative tests illustrates that the recently introduced empirical techniques
need not be in conflict with traditional policy practices in Europe, but should be seen as a natural extension of
commonly accepted principles. Furthermore, we emphasize the importance of our dynamic comparative
market power test in the European context. According to this test, the relevant point of comparison when
assessing a merger is not the status quo, but rather a relevant alternative merger scenario that is likely to
happen in the event the merger is rejected. In a European context, the choice is often to accept a sequence of
regionally based mergers or rather a sequence of pan-European mergers. Finally, we also emphasize the
importance of constructing confidence intervals when computing the predicted merger effects, especially if
one is interested in testing the hypothesis that a merger has no effects.
D 2005 Elsevier B.V. All rights reserved.
JEL classification: L40; L41; L13; L62; C81
Keywords: Merger analysis; Competition policy; Differentiated product markets; Nested logit models
0167-7187/$ - see front matter D 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.ijindorg.2005.08.004
* Corresponding author. Institut D’Economie Industrielle, Universite´ de Toulouse 1—Sciences sociales, Manufacture
des Tabacs—21, Alle´e de Brienne, 31000—Toulouse, France. Tel.: +33 5 61 12 85 92; fax: +33 5 61 12 86 37.
E-mail addresses: ivaldi@cict.fr (M. Ivaldi), frank.verboven@econ.kuleuven.be (F. Verboven).
1
Faculty of Economics and Applied Economics, University of Leuven, Naamsestraat 69, B-3000 Leuven, Belgium.
Tel.: +32 16 326944; fax: +32 16 326732.
International Journal of Industrial Organization
23 (2005) 669–691
www.elsevier.com/locate/econbase
1. Introduction
Merger policy has shown several interesting new developments over the past years. In the
U.S., the policy principles have been modified to incorporate recent theoretical developments in
Industrial Organization, such as the analysis of oligopoly behavior and the role of efficiencies
(see European Commission, 1997). This evolution is illustrated by the various revisions of the
Merger Guidelines. At the same time, U.S. policy practice has shown an increasing reliance on
empirical methods and bsimulation analysis.Q In Europe, policy principles have evolved more
slowly, in part because of the shorter experience with European merger cases. The principles of
market definition have been made more in line with U.S. rules in 1997 (see Federal Trade
Commission, 1997). But once the market is defined, the actual merger investigation has
traditionally been based on the concept of dominance. Only very recently, in 2002, the
Commission has proposed reforms to make its policy principles more in line with economic
thinking about oligopolistic behavior, while keeping the concept of dominance. Policy practice
in the E.U. has up to now still been largely based on the assessment of market shares and
qualitative criteria such as ease of entry and buyer power, with limited attention to econometric
analysis.
This paper starts from a recent European case, the proposed Volvo/Scania merger, to draw
lessons about how policy practice towards mergers may be improved in Europe. Taking the
geographic market definition from the Merger Decision as given, we formulate and estimate a
suitable empirical oligopoly model with differentiated products. As an initial specification test,
we verify whether our parameter estimates imply a price elasticity of total market demand that is
consistent with industry sources. We then show how the model can be used to quantify the
market power effects from mergers. We look at the merger from several different angles. First,
we propose a hypothetical market power test, which measures the extent to which hypothetical
unilateral price increases by the merging parties would be profitable. This test has the
advantages of imposing only weak assumptions on firm behavior and to be close to current E.U.
merger principles on the relevant market definition. Second, we apply an actual market power
test, which measures the actual expected price increases based on more specific assumptions
about post-merger firm behavior. The test also allows one to evaluate the effects of efficiency
claims, which is given more weight according to the recent E.U. reform. Third, we apply a
comparative market power test. This test takes a dynamic approach and takes into account that
the decision to accept or block a merger affects the subsequent merger process. According to this
test, the relevant point of comparison when assessing a merger is not the status quo, but rather a
relevant alternative merger scenario that is likely to happen in the event the merger is rejected.
There is an emerging literature on predicting merger effects.2
Baker and Bresnahan already
propose a first approach to this question in 1985. Later Hausman, Leonard and Zona (1992,
1994) advocate the use of econometric models in competition analysis.3
Werden and Froeb
(1994) calibrate (but do not estimate) a logit model to measure the effects of alternative mergers
among U.S. long distance carriers. Jayaratne and Shapiro (2000) calibrate a nested logit model to
assess the effectiveness of partial divestitures as a remedy to the possible anticompetitive effects
2
Besides mergers, econometrics and statistical methods are now applied in many, if not all, domains of competition
law. For a review, see Bishop and Walker (1999). See also the book edited by Slottje (1999).
3
Hausman, Leonard and Zona (1994) estimate a multi-level demand model instead of using a logit-type model as most
references in the econometric literature on competition policy. See also Hausman and Leonard, 1997.) A discussion of the
relative advantages of the two types of modeling is beyond the scope of this paper.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691670
of horizontal mergers. Nevo (2000) estimates a general random coefficient model to study
merger effects in the U.S. ready-to-eat cereal industry. Pinkse and Slade (2004) follow a distance
metric approach to study mergers in the brewing industry. All these papers have their own focus
of analysis, depending on the specifics of the case.
Relative to this literature our paper makes several contributions. First, we compare several
alternative possible merger tests, which differ in the extent to which they conform to
traditional policy practice in Europe. Our hypothetical market power test is clearly closest to
E.U. merger principles and may therefore be particularly useful to gain more common
acceptance by practitioners. Our actual market power test (often referred to as bsimulation
analysisQ in the literature) is somewhat further apart from the strict dominance approach used
at the time of the merger. However, after the recent reform by the European Commission the
test should without ambiguity be seen as consistent with E.U. merger principles, in particular
also regarding the possibility to explicitly treat efficiencies in the test.4
Our comparative
market power test does not seem to be included in the current merger framework, but we hope
that our analysis shows it can be a fruitful way to think about mergers, as it is a natural extension
of the previous tests. In sum, our comparison of the alternative possible merger tests illustrates
that the recently introduced empirical techniques need not be in conflict with traditional policy
practices in Europe, but should be seen as a natural extension of commonly accepted principles.
It is thus not appropriate to view bsimulation analysisQ and related tests as a drastically different
approach to merger analysis, as it is sometimes presented, but rather as a further development of
the analysis.
Second, we stress the importance of our comparative market power test, especially in the
European context (as opposed to a U.S. context where markets might be more integrated).
Specifically, our test essentially considers a sequence of two regional mergers, i.e. between firms
that are strong in the same geographic area. We compare the market power effects from this
merger sequence with the alternative of two pan-European mergers, i.e. between firms that are
strong in different geographic areas. This comparison is based on the assumption that Volvo and
Scania (who are both strong in the Nordic countries) would seek for partners from other
geographic areas, in the event the Commission blocks their proposed merger. In our application,
we find that sequential pan-European mergers have substantially less anticompetitive effects
than regional ones. We stress, however, that this finding is based on the assumption in our
analysis that the truck market is not yet an integrated market.5
Note that our approach could be
helpful for competition authorities which have to consider a forward looking analysis when
evaluating mergers, and in particular for the European authority which must take into account
how possible future mergers may affect European integration.
Finally, we stress the importance of computing confidence intervals around the estimated
merger effects. Because of the nonlinearity of the oligopoly model, it turns out that the
distribution of the estimated consumer surplus and welfare effects is not symmetric. In our
application, it is skewed to the left. This implies that our point estimates for consumer and
4
In our application the results from the actual market power test were broadly consistent with the results from a
traditional market share analysis based on the dominance approach, i.e. our predicted market power effects are the largest
in those countries where the Commission found the highest market shares. We emphasize that this is not generally true,
since it will depend on the specific parameter estimates, so that an econometric approach is desirable.
5
As will be discussed below, our paper does not directly analyze the issue of whether markets are indeed integrated.
Instead, we base ourselves on the conclusions of the Commission, which conducted an institutional and empirical
analysis of the geographic market definition.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 671
welfare losses are somewhat higher than their expected value.6
We conclude that empirical
merger analysis should be careful in reporting confidence intervals around the point estimates of
the predicted merger effects, especially if one is interested in testing the hypothesis that the
merger had no significant effects.
The outline of the paper is as follows. Section 2 reviews the merger process in the E.U. and
the U.S. to put our analysis in context. Section 3 summarizes the key aspects of the Volvo/Scania
merger decision. Section 4 introduces the model and discusses the empirical results. Section 5
discusses the results from our merger analysis and alternative tests. Section 6 concludes.
2. The merger process in practice
2.1. General aspects of the merger process
The process of evaluating mergers follows various stages. A first stage usually asks whether
the notified merger falls within the jurisdiction of the merger authority. In the European Union
this includes the question whether the merger has a European dimension. If this is not the case,
then the investigation is left to the individual country or countries.
The second stage considers the definition of the relevant market. The U.S. 1997 Horizontal
Merger Guidelines explicitly state that bthe market definition focuses solely on demand
substitution factors, i.e., possible consumer responses.Q Supply substitution factors do not fall
under the market definition process, but are considered elsewhere. Since 1982, the Guidelines
specify their approach more precisely, based on the hypothetical market power test, or SSNIPtest.
This principle states that the relevant market is a group of products and a geographic area
such that a hypothetical, profit-maximizing firm would impose a bsmall but significant and nontransitory
increase in prices.Q The profitability of such a price increase clearly depends on the
extent of demand substitution. In a recent Notice (OJ C 372 on 9/12/1997), the European
Commission has also set out its principles for market definition. Much in the spirit of U.S.
practice, the focus is on demand substitution factors, with a reference to the SSNIP-test. bSmall
but significantQ price increases are specified to be in the range of 5–10%. The attitude towards
supply substitution factors is ambiguous. While it is stated that supply substitution constraints
are taken into account at bthe assessment stage of competition analysis,Q it is also stated that they
may be taken into account when defining markets.
The third stage constitutes the actual merger investigation. In the U.S., this stage involves an
analysis of market concentration and an investigation of potential adverse competitive effects,
including the ease of entry.7
The anti-competitive effects from the mergers may be explicitly
balanced against beneficial effects, in particular the presence of efficiencies. In the European
Union, the merger investigation at the time of the Volvo-Scania Case focuses on an investigation
of anti-competitive effects, without explicitly assessing beneficial effects, such as potential
efficiencies. The investigation amounts to assessing the presence of dominance. A dominant
position is found when a firm (or a group of firms) would be able to behave to an appreciable
extent independently of its competitors and consumers. Assessing dominance is based on criteria
6
More precisely, our point estimates refer to the median of the consumer and welfare losses, and these are lower (in
absolute value) than the expected value.
7
Willig (1991) shows that, bfor a market where the structure of demand was well represented by the logit model, (. . .)
analysis of merger would be accurately based on market shares (. . .).Q This remark is particularly acute to shed light on
our analysis.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691672
such as the joint market share of the merging firms, the strength of the remaining competition,
and potential competition. While the Commission’s recent reform proposals keep the concept of
dominance, they introduce more scope for more economic reasoning about market power effects
in oligopoly, and for a more explicit consideration of efficiencies.
Neven et al., (1993) review the European merger decisions during the early nineties. They
argue that the Commission has displayed little confidence in using the joint market share of the
merging firms as the sole criterion. Their only consistent finding is that a joint market share of
less than 25% is cleared within one month of notification. They also argue that the bstrengthQ of
the remaining competitors is most often used to complement information on the joint market
shares. Finally, they observe the importance attached to buyer power and entry possibilities.8
2.2. Differentiated products in U.S. merger analysis
During the stage of the actual merger investigation the issue arises how markets with
differentiated products should be treated. The European Merger Regulation at the time of the
Volvo-Scania case does not consider this issue explicitly, in contrast to the U.S. Merger
Guidelines. The Guidelines discuss the extent of product differentiation as one key
determinant to identify a lessening of competition through unilateral effects (as opposed to
coordinated interaction).9
According to the Guidelines, the merging firms may find substantial
price elevations profitable, depending on the extent to which the lost sales will be diverted to the
product of the merging partner. If the merging firms’ products are close substitutes relative to
other products, the diversion will be particularly strong, making unilateral price increases more
profitable. Interestingly, the Guidelines treat the observed market shares as an important
indicator. According to the Guidelines, a significant share of consumers would be adversely
affected if, among other things, the merging firms have a combined market share of at least 35%,
and if a significant share of consumers of one merging firm’s product regards the others as a
second choice. Nevertheless, the Guidelines also state that market shares may both understate or
overstate the competitive effects of concern.
Shapiro (1996) is more explicit about how the US antitrust agencies may conduct the actual
merger investigation when products are differentiated. He distinguishes between four
(pedagogical) steps in assessing the unilateral effects from a merger between brands bAQ and
bB.Q The first step is devoted to the measurement of the diversion ratio, which provides the
fraction of sales lost by brand bAQ that are captured by brand bB.Q Equivalently, it is the ratio of
the cross-price elasticity of demand for bBQ over the own price elasticity of demand for bA.Q At
the second step, based on the diversion ratio and merging firms’ current mark-ups, one calculates
the post-merger price increase, assuming no synergies and no rival responses.10
The third step
attempts to account for the effects of price and product responses by the rival firms. Finally the
fourth step evaluates the potential presence of synergies that could reduce marginal costs.
8
More recently, Neven and Ro¨ller (2002) have also looked at the determinants of merger decisions in the E.U. They
provide a framework within which they investigate whether the Commission has pursued different objectives from those
that it has been assigned.
9
The other determinant is the presence of capacity constraints.
10
For example, under a constant elasticity demand, when firms are symmetric and sell only one product before merger,
Shapiro recalls that the predicted price increase is equal to ( p*Àp)/p =mD /(1Àm ÀD), where p and p* are the premerger
and post-merger prices, D is the diversion ratio and m is the pre-merger markup.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 673
Shapiro discusses that the practical implementation of these steps depends on data
availability. If there are detailed data on sales and prices, an oligopoly model with differentiated
products can be estimated or calibrated to simulate the post-merger prices. The first two steps are
thus effectively combined; the diversion ratio is implicitly subsumed in the analysis. If no
detailed data are available, Shapiro proposes to adopt the above steps and obtain the required
information on markups and diversion ratios in other ways. For example, the diversion ratio can
be directly computed through survey data or company documents on the consumers’ first and
second choices. In other cases, one may apply the result that, under certain assumptions, the
market share of brand bBQ relative to the market share of all brands except bAQ may also be
considered as a first proxy for the diversion ratio.11
3. The Volvo–Scania merger
The proposed merger between Volvo and Scania was notified to the European Commission
on 22 September 1999 (Case No COMP/M. 1672). A Commission Decision on 15 March 2000
declared the merger incompatible with the Common Market and the functioning of the EEA
Agreement. In light of our analysis, we now summarize the relevant aspects of the investigation
by the European Commission, as discussed in the Commission’s Decision.12
We restrict
attention to the analysis of the effects in the heavy trucks market.13
3.1. Market definition
The analysis of the relevant product market is short, and does not refer to the SSNIP-test.
The truck market is classified in three categories: light duty trucks (less than 6 tons), medium
duty trucks (5–16 tons) and heavy duty trucks (more than 16 tons). The heavy truck market is
further subdivided into two segments: rigid trucks and tractor trucks. Rigid trucks are integrated
trucks, from which no semi-trailer can be detached. Tractor trucks are detachable. While it is
recognized that rigids and tractors may not be fully substitutable, the overall conclusion is that
the category of heavy trucks constitutes the single relevant market. Light and medium duty
trucks are thus not included. No mention is made as to whether second hand trucks are part of
the relevant market.
The analysis of the relevant geographic market is more detailed. Despite the views of the
merging parties, the Commission concludes that the national markets constitute the relevant
geographic market in the regions most affected. Several arguments are used to support this view.
First, there are substantial price and markup differences across countries. Second, the models and
technical configurations differ considerably, because of local consumer preferences and national
technical requirements (e.g., the cab crash test in Sweden). Third, the selective and exclusive
distribution system links the sales and after-sales services. The importance of profits from aftersales
service may therefore induce dealers to charge higher prices to foreign customers. Finally,
there are large variations in market shares across countries.
11
For example, if brand bAQ has a market share of 25% and brand bBQ a market share of 15%, one could compute the
diversion ratio as 15/(100À25)=20%. This is an exact measure if (i) all brands compete symmetrically, and (ii) no
consumers quit consuming, i.e. market demand is inelastic.
12
For the complete Decision see http://europa.eu.int/comm/competition/mergers/cases/decisions/m1672_en.pdf.
13
There was a separate analysis of the effects in the markets for buses and coaches.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691674
3.2. Assessment
The Commission Decision explicitly describes its methodology for assessing the creation or
strengthening of a dominant position. It uses traditional market power proxies, i.e., market
shares, supplemented by qualitative factors such as customer purchasing power and the
likelihood of entry. The investigation is limited to the five countries where the creation of a
dominant position is found (Sweden, Norway, Finland and Ireland), or where this is found to be
likely (Denmark).
For each of the five countries analyzed, the Commission takes the market shares as the
starting point of the assessment. Table 1 gathers the market shares of the seven truck
manufacturers. The table shows that the joint market share of Volvo and Scania is the largest
in precisely the five countries where dominance is found (in the 49–91% range). This reveals
that the Decision attached a high weight to the merging firms’ joint market share in
assessing dominance. In fact, from the borderline countries, namely Denmark and Portugal,
one may infer that the critical market share for concluding dominance is around 44–49% in
this case.
In its market share analysis the Commission also stresses that the merging firms’ joint market
share have remained stable, and showed no tendency to decline. Finally, the Commission points
out the large difference between the joint market share of the merging parties and the market
share of the largest remaining competitor in most of the five countries.
The Commission supplements its market share analysis with qualitative factors. First, the
extent of brand loyalty and the customer structure is considered. For most of the five
investigated countries, the Commission finds indicators of considerable brand loyalty and of a
dispersed customer structure, with the large majority of transport companies owned by small
operators. The Commission concludes that there is little customer purchasing power to
compensate for the increased market power by the merging firms. Second, the likelihood of
entry is assessed. Entry costs are calculated to be high, especially in light of the small size of
Table 1
Market shares per country in 1998
Daf Daimler Iveco Man Renault Scania Volvo
AU 5 18 6 34 4 16 12
BE 17 18 6 11 8 17 23
DK 4 18 7 10 3 30 29
FI 0 10 4 3 18 31 34
FR 8 16 8 5 38 9 14
GE 5 42 6 26 2 9 8
GR 3 36 2 12 3 17 24
IR 13 9 8 6 3 27 22
IT 4 16 41 6 9 12 12
LU 15 28 8 14 10 15 11
NE 33 12 3 9 3 23 16
NO 4 9 2 12 1 32 38
PO 14 12 7 6 17 19 25
SP 9 19 20 8 19 16 13
SW 2 6 0 0 1 46 45
UK 18 9 9 7 6 19 18
EEA 11 21 11 13 12 16 15
Source: Commission Decision, based on the Notification.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 675
the markets and the low population density. The cab crash test in Sweden is mentioned as an
additional entry barrier.
4. The econometric model
We specify an oligopoly model for the European heavy truck market that starts from the
market definition adopted by the Commission. We thus assume geographically segmented
markets and product differentiation in two segments: tractors and rigids. The specification
closely follows the exposition of the nested logit model in Berry (1994), as extended to
multiproduct firms by Verboven (1996).
4.1. Consumers and demand
A typical consumer is here a freight transportation company. There are N potential
consumers, who may either buy one of J products (heavy trucks), j =1,. . . ,J, or otherwise
choose the outside good 0, e.g. a medium duty truck, a second hand truck, or another transport
mode.14
The nested logit model classifies the products into G groups, and one additional group
for the outside good. Products within the same group are closer substitutes than products from
different groups. We consider two groups for heavy trucks: rigids (R) and tractors (T). The tree
on Fig. 1 depicts the consumer choice set. Note that each group (buying a rigid truck or a tractor,
or employing another transportation mean) corresponds to the use of a different logistic chain.
The utility to consumer i from purchasing product j is given by:
uij ¼ dj þ fig þ 1 À rð Þeij: ð1Þ
The first term, dj, is the mean valuation for product j, common to all consumers. It depends
on the price of product j, jp, a vector xj of observed characteristics of product j, and an error term
nj reflecting unobserved characteristics:
dj ¼ xjb À apj þ nj; ð2Þ
where aand b are parameters to be estimated.
The second and the third term in (1), fig and eij, are random variables reflecting individual i’s
deviation from the mean valuation. The term fig is consumer i’s utility, common to all products
14
Note that in this industry, each manufacturer produces one model per type of truck, which can come under many
variants.
Truck models
Rigids Tractors Outside good
... ...
Truck models
Fig. 1. The choice set.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691676
belonging to group g, whereas the term eij is consumer i’s utility, specific to product j. The
parameter r lies between 0 and 1 and measures the correlation of the consumers’ utility across
products belonging to the same group. If r =1, there is a perfect correlation of preferences for
products within the same group; so these products are perceived as perfect substitutes. As r
decreases, the correlation of preferences for products within same group decreases. If r =0 there
is no correlation of preferences: consumers are equally likely to switch to products in a different
group as to products in the same group in response to a price increase. In this case, we have the
standard logit model in which products compete symmetrically.15
Each potential consumer i chooses the product j that maximizes utility. To compute the
probability that a consumer chooses product j, the nested logit model assumes that the random
variables fig and eij have distributions such that fig and fig +(1Àr)eij have the extreme value
distribution. Normalizing the mean utility level for the outside good to 0, i.e., d0 =0 the
probability sj that a potential consumer chooses product j is given by the following formula:
sj ¼
exp dj= 1 À rð Þ
À Á
Dg
D1Àr
g
1 þ
PG
g¼1 D1Àr
g
; ð3Þ
where Dg is defined by:
Dg ¼
X
kaGg
exp dk= 1 À rð Þ½ Š: ð4Þ
For the model to be consistent with (random) utility maximization, a has to be positive and r
has to lie between 0 and 1. At the aggregate level, the choice probability sj coincides with the
market share of product j. The total quantity sold of product j, qj, is simply given by the probability
that a potential consumer chooses product j times the total number of potential consumers N:
qj ¼ sjN: ð5Þ
The net consumer surplus, CS, measures the attractiveness of the set of J +1 products in
monetary terms, after subtracting the price consumers have to pay. It is given by the expected
value of the maximum of utilities. Using the assumptions of the nested logit model, the net
consumer surplus CS equals (see, e.g., Anderson et al., 1992):
CS ¼
1
a
ln 1 þ
XG
g¼1
D1Àr
g
!
: ð6Þ
4.2. Pre-merger and post-merger pricing
Each firm f produces a set Ff of products. Its profits are given by the sum of its operating
profits for each product minus fixed costs K. The operating profits for product j equal the total
sales of product j times the operating margin, i.e., the price pj minus the (constant) marginal cost
cj for product j. Thus firm f Vs profits are:
pf ¼
X
jaFf
pj À cj
À Á
qj À K: ð7Þ
15
In this case, the merger analysis can be based on market shares exclusively. See footnote 4.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 677
Producer surplus is simply the sum of these profits across firms. Total welfare is the sum of
producer surplus and consumer surplus, defined earlier.
Firms choose the prices of their products to maximize profits, given the prices set by the other
firms. Each firm trades off two effects when considering an increase in price by one unit: (i) it
increases profits proportional to the current sales level of the firm, (ii) it reduces sales, which
lowers profits proportional to the current markup. The multiproduct firm takes into account that
the lost sales on one product may be partly compensated by increased sales on its other products.
The importance of this effect depends on the above discussed diversion ratio between the
products, i.e., the fraction of the sales lost due to a price increase that is recaptured by the other
products owned by the firm.
Pre-merger and post-merger pricing follow a similar logic. One simply needs to
appropriately reinterpret the set of products Ff owned by the merged firm. If there are no
other changes due to the merger (such as cost synergies), then the merged firm will always have
an incentive to raise prices. This is because it takes into account the effect of a price increase on
the sales of its merging partner. The magnitude of the price increase will depend on the
diversion ratio.
More formally, a multiproduct Nash equilibrium is given by the system of J necessary firstorder
conditions derived from Eq. (7), i.e. for each j:16
X
kaFf
pk À ckð Þ
Bqk
Bpj
þ qj ¼ 0: ð8Þ
The marginal cost for each product j is constant and depends on a vector wj of observed
characteristics of product j, and an error term xj reflecting unobserved characteristics:
cj ¼ exp wjc þ xj
À Á
; ð9Þ
where c is a parameter vector to be estimated.
4.3. Specification and estimation
We estimate the demand Eq. (3) and the pre-merger pricing Eq. (8), using the expressions for
the mean utility (2) and marginal cost (9). The parameters to be estimated are a, r, b and c. The
observed variables are prices, pj, sales, qj, and the characteristics, xj and wj, influencing the
mean valuation and marginal cost. The total number of potential consumers N is assumed to be
known. The econometric error terms are the unobserved characteristics, nj and xj. They enter
nonlinearly in both the demand Eq. (3) and first-order conditions (8). The system of pricing Eq.
(8) can be inverted and log-linearized to obtain a solution for xj. To linearize the demand Eq.
(3), we follow the transformation procedure proposed by Berry (1994). We estimate the
transformed demand and the pricing equations simultaneously using nonlinear three-stage least
squares. This estimator takes into account cross-equation parameter restrictions and possible
correlation between the error terms nj and xj. It also takes into account the endogeneity of prices
and sales through instruments.
16
In the following we assume that a Nash equilibrium exists. Caplin and Nalebuff (1991) prove existence in a general
discrete choice model, assuming single product firms. Anderson and de Palma (1992) prove existence for the nested logit
model with multiproduct firms, assuming symmetry.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691678
To estimate the model we use a panel of 16 countries in the E.E.A. over 2 years (1997 and
1998). Prices are list prices of a base model. Sales are total sales for the model range.17
The
characteristics vectors xj and wj contain the same exogenous variables: horsepower, a dummy
variable to denote btractor,Q a set of dummy variables to measure country-specific effects, a set of
dummy variables to measure firm-specific effects, and an interaction dummy variable to indicate
whether the product is produced by a domestic firm.18
To account for endogeneity of prices and
sales, we use the following variables as instruments in the demand and pricing equation: the sum
of horsepower of all competing products in a country per year, and the sum of horsepower of all
competing products in a group (tractor or rigid group).
The total number of potential consumers N is set equal to the average total sales in the
country over 1997–1998, multiplied by a potential market factor 1+r. We consider two
scenarios: r =0.5 and r =0.3. In other words, we assume that the total potential market is either
50% or 300% larger than the average annual sales. We return to this assumption below.
Most parameter estimates for the characteristics (in the vectors c and b) are significant with
the expected sign, and robust whether r =0.5 or r =0.3 is assumed. For example, the estimates of
the firm-specific fixed effects show that a firm with a higher marginal cost also tends to have a
higher mean valuation. Domestic firms also receive a higher mean valuation. They experience
only an insignificant cost advantage over foreign firms.19
Horsepower has a positive and
significant effect on marginal cost; it has a negative but insignificant effect on the mean
valuation by consumers. To interpret this, note that the horsepower variable may not only
capture truck performance. It may also reflect unmeasured maintenance and operating costs for
truck drivers, which do not affect marginal cost, but negatively affect the mean valuation. As a
result, the effect of horsepower on marginal cost is unambiguously positive, whereas its effect on
the mean valuation may be positive or negative.20
Most relevant for our purposes are the estimates of the parameters a and r, which determine
the estimated price elasticities and markups, conditional on the potential number of consumers.
Table 2 shows the estimates for the two assumed values of the potential market factor (r =0.5
and r =0.3). Note that these estimates satisfy the necessary restrictions for the nested logit model
17
These data have been obtained from a survey run by the Merger Task Force under the conduct of the authors. The
values of market shares obtained by that way are very close to the ones presented in Table 1.
18
More specifically, DAF is a domestic firm in the Netherlands, MAN and Daimler are domestic firms in Germany,
Renault is a domestic firm in France, Iveco is a domestic firm in Italy, and Volvo and Scania are treated as domestic firms
in the Nordic countries.
19
A cost advantage for domestic firms may occur for example because of lower transportation costs.
20
The literature on the automobile market usually finds a significant and positive effect of horsepower on the mean
valuation term. First, this literature may better control for maintenance costs in the mean valuation term, see, e.g. Berry,
Levinsohn and Pakes (1995) who include fuel efficiency. Second, consumers of cars may put a higher weight on the
performance of a car relative to the implied higher maintenance costs.
Table 2
Estimates of main parameters of interest
Potential market factor
r =0.5 r =3.0
Estimates Standard error Estimates Standard error
a 0.312 0.092 0.280 0.094
r 0.341 0.240 0.304 0.240
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 679
to be consistent with random utility maximization. Consumers respond to a price increase by
reducing demand (a z0). The hypothesis that trucks within the same group (rigid or tractor) are
perfect substitutes can be rejected, since r is significantly less than 1. The hypothesis of
symmetric competition between trucks from the same group and trucks from different groups
cannot be rejected, yet the 95% confidence interval for r suggests that there is segmentation
between segments. The estimated marginal costs cj and mean valuations dj implied by the
estimates are positive for all products, and usually of a reasonable order of magnitude.
4.4. Discussion of the empirical results
Within the limited time frame we had to conduct the empirical analysis, there was not much
room for a detailed sensitivity analysis. We focused the sensitivity analysis to the assumed value
of the potential market factor r. This is important since there is a close correspondence between
the value of r and the level of the aggregate price elasticity, and hence the consumers’ outside
substitution possibilities. As r increases, the outside good becomes an important substitute for
new trucks, so that the price elasticity of total market demand increases.21
While r can thus in
principle be estimated, just as one can estimate the price elasticity of total market demand, in
practice this would require information for a sufficiently large time horizon. Since our productlevel
data set covers only two years, estimation of r proved difficult. Our solution to this
problem was to consider two alternative scenarios for r, compute the implied price elasticities of
total market demand and confront these with supplementary information from other studies on
price elasticities.
The first scenario assumes that r =0.5, i.e., the potential market size is 50% larger than the
actual average market size during 1997–1998 (measured by country-average annual shipments
over the two periods). The second scenario assumes r =3.0, i.e., the potential market size is four
times larger than the actual market size. The implied price elasticities of market demand are
displayed in Table 3. In the first scenario, the implied price elasticity mostly varies between
À0.5 and À0.6, depending on the country. In the second scenario, the elasticities of market
demand increase to the range of À1.0 and À1.5. The reason for the lower elasticities of market
demand in the first scenario is that in that case relatively few potential consumers are assumed to
actually not buy a truck (only 50% on average), hence consumers view the outside good as a
relatively poor substitute compared to the second scenario.
To check the plausibility of our estimates, the elasticities as implied by the model may be
confronted with elasticities from existing evidence. The European Commission provided us their
estimates obtained from industry sources, reporting that the price elasticity is around À0.9 for
heavy trucks above 16 tons, and À0.4 for trucks above 24 tons. These numbers are thus of a
comparable order of magnitude as the elasticities implied by our independently chosen
scenarios. Our second scenario (i.e. r =3.0) is the more conservative one: it assumes that
consumers substitute more easily out of new trucks when all prices increase, so that the
estimated anticompetitive effects from a merger will be smaller. We therefore decided to use the
second scenario in our subsequent computations on the likely merger effects.
To obtain a further judgment on the reliability of our results, we now summarize the main
criticisms that were raised by the Volvo economists in response to our analysis. One main
21
For example, in the simple logit model (r =0) the price elasticity of market demand, measured at identical prices p, is
equal to Àa[r /(1+r)]p.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691680
comment was that our analysis is based on information on list prices rather than actual
transaction prices. While this is a valid criticism, data on discounts were not available to us at the
time of estimation. We were however informed that discounts tend to vary more across markets
or time, than across individual consumers. This implies that at least part of this data problem is
controlled for by our inclusion of market and time fixed effects.
Other comments on the empirical analysis related to the need for further sensitivity analysis.
First, the demand side was estimated separately, without imposing cross-equation restrictions
from the supply side (exploiting the assumption of pre-merger Nash equilibrium). This produced
imprecise results for several demand parameters, which has also been observed in other studies
where the demand equation was estimated separately with data on only a limited number of
periods. Second, the model was estimated after excluding some of the countries. When only a
few countries were excluded the results remained robust. Excluding up to half of the countries,
however, generated rather imprecise parameter estimates.
Finally, there were comments on the interpretation of our parameter estimates, and the
implications for the estimated price cost margins. One remark was that it is not possible to reject
the hypothesis that the segmentation Parameter r is equal to zero, implying symmetric
substitutes. However, our interpretation of the estimated r is that the data are insufficiently
informative to make precise predictions on the extent of substitution within and between nests.
Despite this, we are able to reject the hypothesis that brands within a nest are perfect substitutes
(r =1). This is an important finding since it would otherwise not be possible to rule out the
possibility that the merger has insignificant price effects. (This is because a merger has no effects
when products are perfect substitutes and the merger does not create a monopoly).
Another remark related to the price–cost margins or the Lerner indices as implied by our
estimates. Market average Lerner indices range from .346 in Italy to .563 in Sweden (with
standard deviations in the .1–.2 range depending on the country). These numbers are higher than
the average gross margin of .3 as cited by the Volvo experts. Our response to this criticism was
that the margins predicted by the model should not necessarily coincide with margins that are
Table 3
Price elasticity of total market demand
Potential market factor
r =0.5 r =3.0
Austria À0.49 À1.53
Belgium À0.49 À1.11
Denmark À0.47 À1.02
Finland À0.38 À0.98
France À0.44 À1.17
Germany À0.53 À1.52
Greece À0.28 À0.63
Ireland À0.34 À1.05
Italy À0.61 À1.63
Luxembourg À0.41 À0.94
Netherlands À0.59 À1.54
Norway À0.56 À1.14
Portugal À0.46 À1.21
Spain À0.44 À1.22
Sweden À0.44 À0.96
United Kingdom À0.56 À1.27
Note: The price elasticity of total market demand in this differentiated product model is defined as the percentage change
in total demand for trucks when the prices of all trucks increase by 1%.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 681
directly computed from accounting cost data and are affected by the rules for allocating fixed
costs in particular. For example, in times of overcapacity the short run economic marginal cost
may be rather low. Furthermore, in the truck market sales are linked to after-sales services;
manufacturers may take the future profits from these after sales services into account when
setting their prices.
5. Merger analysis
We now show how the econometric model can be used to assess merger effects. We begin we
a discussion of the results from our first test, a hypothetical market power test. This test is
informative since it requires no assumptions about firm behavior after the merger. It also relates
to the spirit of the SSNIP-test. Next we consider an actual market power test, which uses more
precise assumptions about firm behavior after the merger to obtain more precise predictions
about price effects and effects on consumer surplus and total welfare. This test also allows one to
account for possible efficiencies realized through the merger. Finally, we consider an extension
to our original analysis. We take a dynamic perspective to account for the effect of a merger
prohibition on subsequent mergers in the market.
5.1. Hypothetical market power test
Our first test is a hypothetical market power test. This test computes the profitability of
unilateral and nontrivial price increases by the merging firms. More specifically, we consider
price increases by 5%, 10% and 25%. The elements of this test are defined as follows:
– Profitability. The profitability of a price increase depends on the diversion ratio and the premerger
markups. The greater is the diversion ratio between the two merging firms’ products,
the more the merging partners can recapture each other’s lost sales from the price increase.
The larger are the markups, the more the partners gain from the recaptured sales.
– Unilateral price increase: The rival firms are assumed not to respond to the price increase,
e.g., by partially raising their prices as well.
– Nontrivial price increase: A small price increase after the merger would necessarily be
profitable. It has a negative but negligible effect on the product’s own profits, and a positive
and non-negligible effect on the profits of the merging partner’s profit.22
In contrast, a
sufficiently large price increase eventually becomes unprofitable: The negative own-profit
effect becomes substantial and at some point outweighs the positive profit effect of the
merging partner. We consider nontrivial price increases of 5%, 10% or 25%.
Note that the test is related to the previously discussed SSNIP-test, yet the focus is somewhat
different. The SSNIP-test asks how many firms are needed to make a given price increase
profitable, for the purpose of defining the relevant antitrust market. Our hypothetical market
power test asks whether the two merging firms can profitably raise prices by alternative amounts.
The purpose is here to examine the potential of increased unilateral market power.
22
This is because the negative own-profit effect is a second order effect, since firms are already maximizing their own
pre-merger profits. The positive effect on the merging partner B’s profit is a first order effect, proportional to the premerger
markup of firm B.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691682
Since our hypothetical market power test relates to the SSNIP test, it has the advantage of
being familiar to antitrust practitioners. Furthermore, the test does not require specific
assumptions about firm behavior after the merger. Hence, the test can be applied whether or
not coordinated effects are a source of concern. This is important since there is no obvious way
to select a plausible post-merger equilibrium when coordinated effects are possible. It is, of
course, necessary to decide what constitutes a nontrivial price increase. This problem is however
of a similar nature as deciding what constitutes a nontrivial price increase when defining the
relevant market.
Table 4 shows the (normalized) percentage profit changes accruing to Volvo and Scania,
when both firms would unilaterally raise the prices of all their products by 5%, 10% and 25%.
A robust finding is that a hypothetical price increase by 5% is profitable in almost all
countries. Only in four countries would Volvo and Scania’s joint operating profits (slightly)
decrease after this hypothetical price increase. The highest profit increases are found in Sweden,
Norway, Finland, Ireland and Denmark.
A different picture emerges for a hypothetical price increase by 10%. On the one hand, such a
price increase is unprofitable for nine countries. On the other hand, for markets where the price
increase is profitable, it is frequently more profitable than the 5% increase. This is most notably
true in Sweden, Norway, Finland and Denmark. This conclusion extends to the considered large
price increase of 25%. Such a price increase is only profitable in Sweden, but the profit increase
is larger.
One may also interpret the results in Table 4 in terms of the market definition based on the
SSNIP test. Following the rule that the relevant market is the minimum number of firms that can
profitably raise prices by 5%, the merging firms by themselves constitute the relevant market in
twelve out of the sixteen countries. Modifying the rule to a 10% price increase (the upper bound
in the European Notice), the merging firms still constitute the relevant market in seven out of the
Table 4
Hypothetical market power test
Price increase Profit change of merging firms from alternative price increases
5 10 25
Austria À0.70 À5.96 À35.09
Belgium 1.05 0.49 À8.63
Denmark 1.63 2.09 À2.25
Finland 2.51 2.98 À4.89
France 0.18 À1.40 À13.86
Germany À0.23 À2.79 À19.70
Greece 1.39 À0.02 À14.49
Ireland 2.12 1.70 À10.02
Italy À1.14 À7.63 À41.79
Luxembourg À0.07 À1.51 À11.86
Netherlands 0.77 À2.47 À26.70
Norway 2.74 3.58 À2.37
Portugal 1.16 À0.12 À13.37
Spain 0.23 À2.05 À18.65
Sweden 2.95 4.91 5.67
United Kingdom 1.28 0.49 À11.04
European Union 1.00 À0.49 À14.32
Note: Profit effects are based on the parameter estimates of scenario 2 (r =3.0). The numbers are normalized such that the
average percent profit change in the European Union under a 5% price increase is equal to one.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 683
sixteen countries. Hence, for these countries our analysis implies there is a merger to monopoly,
in the traditional antitrust sense.
We note that the ranking of the countries based on the hypothetical market power test is not
inconsistent with the Commission’s ranking in terms of the merging firms’ joint market shares.
One notable difference occurs in the top 5, where Denmark and Ireland switch places. However,
the similarity between the results from our hypothetical market power test and the Commission’s
traditional market share analysis is not true in general. In our application, it follows in part from
the fact that the estimate of our segmentation parameter r was relatively low (though imprecisely
estimated). In general, it is only possible to find out after the estimation of the model and the
application of the hypothetical market power test whether the results are similar to the traditional
market share approach. Furthermore, the hypothetical market power test has the advantage of
using a more consistent rule in determining the cut-off points at which a merger would be
considered as potentially harmful.
5.2. Actual market power test
The above test assessed the potential of increased market power, by looking at the
profitability of price increases without making detailed assumptions about post-merger firm
behavior. The advantage of the test are that it relates the practitioners’ familiar SSNIP-test, and
that it can be used without making specific assumptions about post-merger firm behavior.
However, the test is conservative in that it underestimates the profitability of joint price increases
in two respects. First, it considers a percentage price increase that is the same for all products of
the merging partners. In practice, the merging firms will typically find it optimal to increase the
prices of their products by different amounts. The profitability of optimal price increases is thus
typically larger. Second, the test ignores responses by competitors. In practice, competitors may
respond to a price increase by also raising their prices. This is most obviously the case if the
merger triggers collusive responses by the other firms; in this case price matching could be
complete. Yet also if the rivals behave non-cooperatively, one may expect positive – albeit
incomplete – price responses.
An alternative approach is the actual market power test, which imposes more specific
assumptions about firm behavior after the merger. In our application we assume that firms
continue to set prices non-cooperatively after the merger, hence we assume that there are only
unilateral effects in the merger. In principle, one could consider an indeterminate number of
alternative coordinated effects equilibria. Our discussion above, however, emphasized it may be
better to use a hypothetical market power test if one is not sure about how coordinated effects
would materialize after the merger.23
To implement the actual market power test, we numerically compute the post-merger Nash
equilibrium, with a modified ownership structure, and compare this to the pre-merger Nash
equilibrium. This amounts to solving a system of 14 pricing and demand equations, to obtain the
equilibrium prices and quantities of the 14 products.24
We do not consider the possibility that
behavior shifts from non-cooperative before the merger to something more collusive after the
merger.
23
Nevertheless, if one is sure about the exact coordinated equilibrium after the merger, an analysis based on simulation
analysis would predict stronger market power effects, since it accounts for collusive reactions by the rivals.
24
Recall that there are 7 firms and 2 groups of products.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691684
5.2.1. No efficiencies
In our initial analysis we computed the effects of the merger on prices, consumer surplus
and total welfare assuming that the merger does not entail any efficiencies. As is well known,
in this case the merger will necessarily increase prices and reduce consumer surplus. Hence, to
appropriately interpret the results from a simulated merger absent efficiencies, the policy
maker should apply a reasonable btolerance levelQ regarding price increases absent
efficiencies. This tolerance level may for example be based on a general presumption
regarding the average size of efficiencies expected from mergers; see Fisher and Lande
(1983), and Ro¨ller er al., (2001) for a discussion on the general presumptions approach.
We first looked at the effects of the merger on the individual prices in the various
countries. Generally speaking, Volvo and Scania are predicted to raise their prices by more
than 10% in several countries, in particular the Nordic countries and Ireland. It is interesting
to note that the predicted price responses by the competitors are positive, but quite small.
Most firms in most countries respond by negligible price increases or price increases by less
than 1%, even in countries where the merging firms raise their prices by substantial amounts
of more than 10%.
To summarize the results about the mergers’ price effects, one may look at the estimated effects
from the merger absent efficiencies on consumer surplus. Decreases in consumer surplus may be
viewed as increases in an industry price index. The results are shown in the first column of Table 5.
There are two countries for which the estimated decline in consumer surplus is greater than 10%:
Norway (À10.71%) and Sweden (À17.77%). For an additional three countries the estimated
decline in consumer surplus is greater than 5%: Denmark (À6.02%), Finland (À8.44%) and
Ireland (À7.00%). For all countries, the 95% confidence intervals of the consumer surplus effects
are negative, which is unsurprising given the significantly negative estimate for the parameter a
Table 5
Actual market power test Welfare analysis with and without cost efficiencies
No cost efficiency of 0% 5% cost efficiency
Change in Change in
Consumer surplus Total welfare Consumer surplus Total welfare
Austria À1.09 [À1.10,À0.44] À0.35 [À0.36,À0.09] 0.02 [À1.09,1.70] 0.01 [À0.37,0.58]
Belgium À3.05 [À3.11,À1.11] À1.16 [À1.16,À0.33] À1.41 [À2.85,1.17] À0.56 [À1.12,0.48]
Denmark À6.02 [À6.22,À2.04] À2.48 [À2.49,À0.82] À4.25 [À5.83,À0.26] À1.79 [À2.38,À0.11]
Finland À8.44 [À8.82,À3.06] À3.51 [À3.55,À1.23] À6.93 [À8.52,À1.99] À2.93 [À3.46,À0.87]
France À1.04 [À1.05,À0.43] À0.34 [À0.35,À0.08] 0.10 [À1.06,1.84] 0.04 [À0.36,0.62]
Germany À0.62 [À0.64,À0.23] À0.19 [À0.19,À0.03] 0.24 [À0.66,1.63] 0.08 [À0.21,0.48]
Greece À3.46 [À3.73,À0.86] À1.30 [À1.32,À0.35] À2.06 [À3.41,0.60] À0.80 [À1.25,0.21]
Ireland À7.00 [À7.12,À3.03] À2.87 [À2.87,À1.03] À5.15 [À6.77,À1.17] À2.16 [À2.77,À0.49]
Italy À1.03 [À1.05,À0.40] À0.33 [À0.34,À0.07] À0.02 [À1.07,1.63] 0.00 [À0.36,0.51]
Luxembourg À0.59 [À0.60,À0.18] À0.19 [À0.19,À0.04] 0.24 [À0.66,1.69] 0.08 [À0.22,0.50]
Netherlands À2.85 [À2.94,À1.03] À1.02 [À1.02,À0.27] À1.30 [À2.67,1.19] À0.48 [À1.00,0.43]
Norway À10.71 [À11.55,À2.99] À4.68 [À4.88,À1.34] À8.95 [À11.07,À1.89] À3.96 [À4.72,À0.86]
Portugal À3.34 [À3.39,À1.34] À1.27 [À1.28,À0.39] À1.78 [À3.18,0.81] À0.70 [À1.25,0.34]
Spain 1.39 [À1.40,À0.57] À0.48 [À0.49,À0.12] À0.20 [À1.39,1.60] À0.07 [À0.50,0.59]
Sweden À17.77 [À20.97,À4.80] À8.05 [À9.13,À2.10] À15.89 [À20.13,À3.48] À7.23 [À8.85,À1.58]
UK À3.77 [À3.86,À1.36] À1.46 [À1.46,À0.43] À2.13 [À3.57,0.65] À0.85 [À1.41,0.28]
Note: The calculations are based on the parameter estimates of scenario 2 (r =3.0). Confidence intervals are in
parentheses, based on a bootstrapping procedure as described in the text.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 685
and the estimate of r significantly different from 1.25
A careful look at the confidence intervals
gives more interesting results. The confidence intervals are skewed: the upper bound of the
confidence interval (in absolute value) is closer to the point estimate than the lower bound. For
example, in Sweden the 95% confidence interval around the point estimate of À17.77 is
[À20.97,À4.80]; in Austria the 95% confidence interval around the point estimate of À1.09 is
[À1.10,À0.44]. Furthermore, the expected values of the consumer surplus effects are lower (in
absolute value) than our reported point estimates. The skewness is the result of the nonlinearity
of consumer surplus in the parameters a and r. As a way of comparison, we also constructed
confidence intervals for the expected price increases per product (not reported in Table). The
skewness was found to be considerably less pronounced. For example, the point estimate of the
price increase for a Volvo tractor in Sweden was 9.98, with a confidence interval of [3.46,23.78];
for the same model in Austria it was 1.88, with a confidence interval of [0.42,4.25].
For completeness Table 5 also reports the changes in total welfare (i.e. the sum of producer
and consumer surplus), although it is interesting to note that the European Commission
essentially did not pay much attention to this. The expected welfare effects of a merger are a
priori ambiguous even when the merger does not entail efficiencies, because of a possibly
beneficial output reallocation effect.26
In our application, however, the welfare effects (and the
confidence intervals) are negative for all countries.
5.2.2. Efficiencies
In an extension of our analysis we considered the possibility that the merger creates
efficiencies, in the form of savings in marginal costs. Under these circumstances, the merger has
two effects: first, the merging firms take into account the effect of a price change on each others’
sales and profits power, inducing the merging firms to set higher prices; second, there is a cost
saving, which may induce firms to (partially) pass on these cost savings to consumers. As a
result, in the presence of efficiencies the post-merger Nash equilibrium will no longer necessarily
entail higher prices than the pre-merger equilibrium.
The second part of Table 5 summarizes the results when the merger contains efficiencies in
the form of marginal cost savings. We consider a merger with a hypothetical marginal cost
saving of 5%. This is the maximum efficiency claim that the Commission presumed to be
reasonably plausible (without however doing a very detailed analysis on these cost savings as it
was common practice before the recent reform proposals).27
The point estimates show that a 5%
cost reduction increases consumer surplus in four countries and lowers consumer surplus in the
remaining twelve countries. In fact, for five of the twelve countries with a negative point
estimate the entire 95% confidence interval for the effect on consumer surplus is negative:
Denmark, Finland, Ireland, Norway and Greece. Hence, for these countries one may conclude at
a 95% confidence level that the merger would lead to a lower consumer surplus, even when there
25
To compute the confidence intervals, we followed a parametric bootstrap procedure. We draw alternative parameter
values for a and r, assuming that they have a multivariate normal distribution with a covariance matrix equal to our
estimated covariance matrix. For these draws we compute the post-merger equilibrium and the implied changes in price,
consumer surplus and welfare. Our results are based on 1000 draws.
26
The output reallocation effect stem from the fact that a merger may cause production to shift to more or less efficient
firms. This effect is beneficial if the social value of the outsiders’ products is sufficiently high, which occurs if they have a
high market share in equilibrium. See Farrell and Shapiro (1990) for an analysis in a Cournot model.
27
The percentage cost reduction applies to the average of the marginal costs across countries. Since the estimated
marginal costs may differ across countries (because of local costs), the percentage cost reductions may actually differ
across countries.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691686
are 5% efficiencies. Incidentally, as discussed above, the European Commission found the
merger to cause dominance in precisely these countries.28
It will, of course, not generally be true
that the results from a simulation approach will coincide so closely with the European
Commission’s dominance approach, but it does illustrate that both approaches are not
necessarily in conflict with each other. As a final remark, we can have a look at the U.K. Out
of the large E.U. countries, the Commission found the effects in this country to be somewhat
borderline (using its traditional market share analysis). Based on our finding this can be
translated in a statistical sense: one cannot conclude at a 95% confidence level that consumer
surplus would decrease in the U.K. after a merger with 5% efficiencies, though one could
conclude it at a lower (unconventional) confidence level, e.g. at the 80% level.
Similar findings obtain for total welfare. The estimated welfare effect is negative for twelve
out of the sixteen countries. Furthermore, in five countries the entire 95% confidence interval for
the welfare effect is negative (the same five countries as before).
5.3. Comparative market power test
The potential and actual market power tests are consistent with current competition policy
rules because they analyze the effects of the actually proposed merger. However, a relevant
question that is currently ignored in the merger decision process is what will happen next. For
example, if claims of cost-savings in the form of returns to scale are valid, it seems reasonable to
expect that other firms with similar scale may propose further mergers. If the competition agency
continues to apply the same criteria it may be difficult to block the second merger once the first
merger has been allowed. In our application, the question of alternative sequential mergers is
especially important from an international European integration perspective. The proposed
merger between Volvo and Scania is essentially a regional one, meaning that the two merging
firms’ market shares are strongly correlated across countries: Volvo tends to have a high market
share in those countries where Scania also has a large market share (the Nordic countries) and a
low market share in countries where Scania has a low market share (and vice versa). Hence, if
this regional merger would be approved, then a second regional merger should likely to be
approved as well if the same criteria are applied. A merger authority should therefore take a
dynamic position, and also ask what will happen after the merger has been approved or blocked.
To illustrate, we consider the following realistic example. We assume that the merger between
Volvo and Scania would trigger an additional regional merger, between Renault and Iveco, who
have stronger positions in the southern parts of Europe. We compare this regional merger
scenario with an alternative scenario of two pan-European mergers. This scenario assumes that
when the Commission blocks the Volvo–Scania merger, Volvo would seek another partner,
Renault, and subsequently Scania and Iveco will merge.29
Table 6 compares the effects of the
two sequential regional mergers, Volvo/Scania and Renault/Iveco, with the alternative of two
sequential pan-European mergers, Volvo/Renault and Scania/Iveco. All four firms have
comparable European level sales, so that scale economies should be comparable.
The first column in Table 6 show that the two regional mergers (absent efficiencies) reduce
consumer surplus by more than 3% in most countries, and by more than 5% in five countries.
28
This is a coincidence since we constructed the confidence intervals after the European has made its decision.
29
The possibility that Volvo would merge with Renault if it could not buy Scania, was of course not known with
certainty at the time. But it was clear that Volvo would seek for other partners and Renault was a real possibility. In fact,
the merger was proposed and accepted shortly after the Commission’s decision to block the Volvo–Scania merger.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 687
The four firms involved in the two regional mergers effectively succeed in dividing the market,
with regional blocks being formed. The second column considers the two pan-European
mergers. Even though the same firms are involved, the merger effects are considerably smaller
for most countries. The reduction in consumer surplus is larger than 3% in only six countries,
and larger than 5% in only 2 countries. Similar conclusions can be drawn from looking at the
total welfare effects from the merger.
This example illustrates that the competition authorities should not restrict attention on the
immediate effects of a proposed mergers. It is necessary to take a dynamic perspective and ask
what would happen next, i.e. after the merger has been approved or rejected. In the context of
segmented markets, the choice may often be between accepting sequential regional or sequential
pan-European mergers.
6. Concluding remarks
Using the Volvo/Scania case as a background, we have shown how the econometrics of
differentiated products markets can be performed and applied to help European antitrust
authorities in their investigation of mergers. We hope to have shown that this technique is a
fruitful tool to evaluate the issues at stake and to account for effects that are otherwise very hard
to measure, like actions of rival firms in terms of price and strategy, cost efficiencies and
alternative merger sequences. Compared to the previous literature we made the following
contributions. We proposed a range of alternative possible tests, to emphasize that the recent
techniques do not necessarily conflict with traditional merger practice based on the dominance
concept. We specifically emphasized the importance of a comparative market power test, which
does not compare the merger outcome with the status quo but rather with a relevant and realistic
alternative merger scenario. Finally, we showed the importance of constructing confidence
intervals in conducting these tests.
Table 6
Comparative market power test Alternative merger sequences
Change in consumer surplus Change in total welfare
Two regional
mergers
Two pan-European
mergers
Two regional
mergers
Two pan-European
mergers
Austria À1.24 À0.88 À0.40 À0.27
Belgium À3.31 À1.88 À1.25 À0.66
Denmark À6.23 À2.30 À2.54 À0.82
Finland À8.93 À5.53 À3.67 À2.09
France À3.48 À4.58 À1.26 À1.74
Germany À0.72 À0.60 À0.21 À0.17
Greece À3.46 0.00 À1.30 0.00
Ireland À7.23 À2.59 À2.94 À0.91
Italy À4.10 À4.37 À1.52 À1.64
Luxembourg À1.31 À1.49 À0.42 À0.48
Netherlands À2.91 À0.78 À1.04 À0.25
Norway À10.73 À0.75 À4.69 À0.26
Portugal À4.25 À3.90 À1.57 À1.42
Spain À4.11 À4.03 À1.47 À1.43
Sweden À17.78 À0.37 À8.05 À0.13
UK À4.35 À3.00 À1.64 À1.07
Notes: The calculations are based on the parameter estimates of scenario 2 (r =3.0).
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691688
The model involves assumptions about cost conditions, demand conditions and market
equilibrium. All together they produce a model that is parsimonious in the number of parameters
to be estimated. This approach is necessary given European merger investigations provide rather
limited time to collect data and perform a quantitative analysis. However, we believe that the
model still provides a good approximation of the working of the heavy truck market and that the
assumptions are favorable to the merging firms. In other words, our chosen simplifying
assumptions tend to generate either unbiased or conservative estimates of the merger effects on
prices and welfare. We now review the role of these assumptions.
6.1. Cost assumptions
It is assumed that marginal cost is constant, i.e., independent of output. The predicted price
increases arising from the merger would be stronger if marginal costs were decreasing in
output.30
In contrast, the price effects would be weaker in the reverse case of increasing marginal
costs (a capacity constrained industry). In fact, it is rather unlikely that marginal costs are
increasing, given the cyclical nature of the market and the need for firms to invest in a sufficient
amount of capacity ex ante.
The model neither estimates fixed costs nor assesses fixed cost savings. Since the primary
interest of the competition agency is in the price or consumer surplus effects, this assumption
does not affect our analysis.
6.2. Demand assumptions
The demand side of the model is based on the nested logit model, which generates the
following substitution pattern. When the price of one product increases by 1%, it increases the
demand for the other products within the same segment by an equal percentage, whereas it
increases the demand for products in a different segment by a lower percentage. Intuitively, the
nested logit model thus imposes symmetric substitution patterns within a segment, yet it allows
for asymmetric substitution patterns across segments, while remaining parsimonious in the
number of parameters to be estimated. Moreover, the assumptions of the nested logit model
regarding the substitution patterns may be viewed as favorable to the merging firms regarding
the predicted price effects arising from the merger for two reasons. First, all products are
symmetric substitutes within a segment, whereas a separate investigation described in the
Commission’s Decision concluded that the merging firms are likely to be closer substitutes. If
this is true, our analysis would underestimate the price effects. Second, the elasticities are
increasing in prices, while in the econometric antitrust literature, one often either assumes
constant elasticities, or otherwise performs simulation analysis with a blinearizedQ model
assuming elasticities do not change. When price changes are large, the assumption of constant
elasticities or the blinearizationQ may yield a significant overestimation of the price effects.
The model is usually described as a discrete choice model, in which each consumer buys a
single truck. The model can be straightforwardly reinterpreted to describe demand behavior
whereby each consumer buys more than one truck, but still a fixed number of the same brand.
Recently, it has been shown that the discrete choice model can also be re-interpreted as a
30
This is because an increase in price implies a lower production, which in turn implies a higher marginal cost under
decreasing marginal costs. The price effect would, however, be less than proportional to marginal cost, since marginal
costs are passed on incompletely.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 689
representative consumer model, in which one consumer has a taste for diversity and decides to
purchase multiple brands of trucks (see e.g. Anderson et al., 1992, for the logit model).
When truck manufacturers are selling trucks in large amounts to transport firms, they may
find it optimal to apply discounts. Our econometric model at least partially control for discounts.
First, as discussed earlier, the model includes firm-specific and country-specific dummy
variables, thereby controlling for firm- and country-specific discounts. Furthermore, the
econometric error term in the demand equation may be interpreted as capturing product-specific
price measurement error, which we incorporated by applying instrumental variables (Berry,
1994). However the model does not allow for individual-specific quantity discounts. From a
theoretical point of view only a few studies have looked to competition with nonlinear pricing
and asymmetric information between buyers and sellers. Generally speaking, it seems fair to say
that competition may limit the feasibility of quantity discounts. Hence a merger, which reduces
the number of competitors, could create the potential for more price discriminating discounts.
This would imply a greater loss in consumer surplus, since the manufacturers can extract more to
their own gain. Note that quantity discounting also raises consumer switching costs, since it
induces consumers to stay with the same firm. Indeed, at the equilibrium, discounts must be such
that the consumer has no incentive to leave for a competing contract, in addition to be incentivecompatible
and individually rational (see Ivaldi and Martimort, 1994).
6.3. Conduct assumptions
The model assumes that the competition between firms selects a static Nash equilibrium.
However, alternative equilibrium concepts may well be more realistic in some situations. For
example, the merger may facilitate collusive behavior, whereby firms behave cooperatively
rather than non-cooperatively. Second, the merger may trigger other mergers. The price effects of
such other mergers may especially be large in a more concentrated industry.
Summing up, several assumptions can be debatable. Nevertheless, care has been taken so that
the assumptions lead to no clear bias, or in fact lead to conservative estimates of the price effects.
Furthermore, it is because we have a quantified model that one is able to assess the effects of the
assumptions. Here we have just provided an example to illustrate how quantification can help the
effectiveness of economic evidence.
Acknowledgements
We are grateful to Svend Albaek, Randy Brenkers, Claude Crampes, Francisco-Enrique
Gonzalez-Diaz, Patrick Rey, Johan Stennek, participants of the Competition Policy Seminar at
IDEI, Toulouse, the 2nd Tel-Aviv Workshop on Industrial Organization and Antitrust and the 9th
Annual WZB Conference on Industrial Organization in Berlin for their comments and insights.
Financial support from the RTN project bCompetition Policy in International MarketsQ is
gratefully acknowledged.
References
Anderson, S., de Palma, A., 1992. Multiproduct firms: a nested logit approach. Journal of Industrial Economics 40,
261–276.
Anderson, S., de Palma, A., Thisse, J.-F., 1992. Discrete Choice Models of Product Differentiation. MIT Press,
Cambridge.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691690
Berry, S.T., 1994. Estimating discrete-choice models of product differentiation. Rand Journal of Economics 25,
242–262.
Berry, S.T., Levinsohn, J., Pakes, A., 1995. Automobile prices in market equilibrium. Econometrica, 841–890.
Bishop, S., Walker, M., 1999. Economics of E.C. Competition Law: Concepts, Application and Measurement. Sweet and
Maxwell, London.
Caplin, A., Nalebuff, B., 1991. Aggregation and imperfect competition: on the existence of equilibrium. Econometrica
59, 25–59.
European Commission (1997). Commission Notice on the Definition of the Relevant Market for the Purposes of
Community Competition Law, OJC 372 on 9/12/1997.
Farrell, J., Shapiro, C., 1990. Horizontal mergers: an equilibrium analysis. American Economic Review 80 (1), 107–126.
Federal Trade Commission (1997). 1992 Horizontal Merger Guidelines (with April 8, 1997 Revisions to Section 4 on
Efficiencies).
Fisher, Alan A., Lande, R., 1983 (December). Efficiency considerations in merger enforcement. California Law
Review 71.
Hausman, J., Leonard, G., 1997. Economic analysis of differentiated product mergers using real world data. George
Mason Law Review 5 (3), 321–346.
Hausman, J., Leonard, G., Zona, J.D., 1992. A proposed method for analyzing competition among differentiated
products. Antitrust Law Journal 60, 889–900.
Hausman, J., Leonard, G., Zona, J.D., 1994. Competitive analysis with differentiated products. Annales d’Economie et de
Statistique 34, 159–180.
Ivaldi, M., Martimort, D., 1994. Competition under nonlinear pricing. Annales d’Economie et de Statistiques 34,
71–114.
Jayaratne, J., Shapiro, C., 2000. Simulating Partial Divestitures to bFixQ Mergers, Mimeo. University of California,
Berkeley. 21 pp.
Neven, D., Nuttal, R., Seabright, P., 1993. Merger in Daylight: The Economics and Politics of European Merger Control.
CEPR, London. 296 pp.
Neven, D., Ro¨ller, L.-H., 2002. Discrepancies between markets and regulators: an analysis of the first ten years of EU
merger control. The Pros and Cons of Merger Control. Swedish Competition Authority, Go¨teborg.
Nevo, A., 2000. Mergers with differentiated products: the case of the ready-to-eat cereal industry. Rand Journal of
Economics, 395–421.
Pinkse, J., Slade, M., 2004. Mergers, brand competition, and the price of a pint. European Economic Review 48 (3),
617–643.
Ro¨ller, L.-H., Stennek, J., Verboven, F., 2001. Efficiency gains from mergers. European Economy 5, 31–128.
Shapiro, C., 1996. Mergers with differentiated products. Antitrust, 23–30.
Slottje, D.J. (Ed.), 1999. The Role of Academic Economist in Litigation Support. North-Holland, Amsterdam. 240 pp.
Verboven, F., 1996. International price discrimination in the European car market. Rand Journal of Economics 27,
240–268.
Werden, G., Froeb, L., 1994. The effects of mergers in differentiated products industries: logit demand and merger policy.
Journal of Law, Economics and Organization 10, 407–426.
M. Ivaldi, F. Verboven / Int. J. Ind. Organ. 23 (2005) 669–691 691