Exclusive dealing as an entry barrier?

Exclusive dealing as an entry barrier? –
Evidence from automobiles
Laura Nurski and Frank Verboven
University of Leuven
August 2013
Automobile demand and distribution networks
The size of distribution networks
has a substantial impact on automobile demand
may partly explain the home bias and its declining role over
the past decade
Plan of talk:
First verify these claims with panel data for European countries
Then assess the role played by exclusive dealing
Automobile demand and distribution networks
Nested logit model
ln sj =s0 = xj
pj + fj (Nj ; ) +
1
ln sj jhg +
2
ln shjg +
j
Nests: segments; subnests = domestic/foreign; products:
model
Estimate for panel of 9 European countries during 2000-2009.
Account for endogenous prices and product-speci…c …xed
e¤ects j
Nested logit demand model (9 European countries,
2000-2009)
price (relative to income)
ln sj jhg
ln shjg
foreign (1/0)
foreign*trend
(1)
-1.41
(-25.09)
0.69
(18.46)
0.44
(11.24)
-0.62
(-10.56)
0.03
(5.55)
dealers/household
(2)
-1.34
(-22.02)
0.43
(9.19)
0.25
(5.55)
-0.37
(-6.46)
0.00
(0.54)
6.34
(13.34)
dealers/squared km
(3)
-1.38
(-22.04)
0.39
(8.24)
0.18
(3.65)
-0.44
(-7.43)
0.01
(1.16)
0.10
(15.46)
log(dealers)
Observations
Number of models
R-squared
(4)
-1.22
(-18.50)
0.32
(5.69)
0.15
(2.75)
-0.53
(-8.88)
0.01
(1.72)
18,029
495
0.708
18,021
495
0.590
18,021
495
0.558
0.31
(11.26)
18,021
495
0.520
The role of exclusive dealing
Main research questions
Do incumbent …rms apply exclusive dealing to soften
competition and foreclose entry by (foreign) …rms?
Or do they have an e¢ ciency motivation, i.e. raise demand
by protecting brand reputation and preventing free-riding?
What are the e¤ects of exclusive dealing on market shares,
consumer surplus and welfare?
We address these questions with an empirical analysis of the
car market in Belgium
demand model with product and spatial di¤erentiation
model of oligopoly pricing (and location decisions)
counterfactuals of a shift from exclusive dealing (ED) to
multi-branding (MB)
Outline rest of talk
Vertical restraints and exclusive dealing in the car market
Previous literature
Pro…t incentives for exclusive dealing
Empirical model
Empirical analysis and counterfactuals
Vertical restraints in the European car market
Manufacturer and dealer contracts
selective distribution = quant. and qual. restrictions
exclusive distribution = exclusive territories
exclusive dealing
= no competing brands
European Commission has tolerated these vertical restraints
through a series of block exemptions:
1985-2002
2002-2010
2010-2013
sel. distr.
AND
OR
OR
excl. distr.
AND
OR
OR
excl. deal.
up to 80%
up to 30%
up to 80%
Preliminary evidence (Belgium)
As for rest of Europe, market shares show positive correlation with
dealer network size.
Preliminary evidence (Belgium)
Exclusive dealing is common, but there is also multi-branding.
1
2
3
4
5
brand (=exclusive)
brands
brands
brands
brands
Number of dealers
1501
291
57
8
3
% of dealers
81%
16%
3%
0%
0%
75% of multi-brand dealers only sell brands of the same …rm
Most common multi-brand combinations are:
VW group (VW, Audi, Skoda, Seat, Porsche)
Fiat group (Fiat, Alfa Romeo, Lancia)
Big luxury brand + small luxury brand (BMW-Mini,
Mercedes-Smart)
Literature on foreclosure: mixed theoretical …ndings
Chicago School: Incumbent cannot pro…tably compensate
buyer to sign an ED contract and thus cannot exclude
e¢ cient entry
Post-Chicago theories: Incumbent can induce the buyer to
sign an exclusive contract under certain circumstances
Aghion & Bolton (1987); Rasmusen, Ramseyer & Wiley
(1991), Segal & Whinston (2000), Fumagalli & Motta (2006),
etc.
Our paper: Entrant may compensate incumbent not to sign
an exclusive contract with the buyer.
Literature on foreclosure: scarce empirical evidence
Laboratory experiments
Smith (2007), Landeo & Spier (2009), Boone, Müller &
Suetens (2013)
Empirical evidence on prices
Slade (2000): ED raises prices, lowers welfare
Sass (2004): ED raises prices, consumption, welfare
Empirical evidence on foreclosure
Sass (2004): no, since ED more common in larger markets
Asker (2005): no, since ED does not raise rival costs
Ater (2010): yes, since ED reduces total sales
Pro…t incentives for ED: overview
A move from exclusive dealing to multibranding with new entrant:
intensi…es competition, leading to lower prices
reduces demand because of reputation loss
=> these aspects are usually emphasized in theoretical work
increases demand because of increased spatial availability
single incumbent: MB can only attract new customers
multiple incumbents: MB can also steal business from other
incumbents –> strong individual incentive, but not necessarily
a joint incentive for MB
may imply …xed cost savings since entrant can make use of
incumbent’s network
=> these aspects may also be important in practice
Empirical framework: overview
To account for these various e¤ects, we estimate a spatial
product di¤erentiation demand model, derive …xed cost
bounds and perform counterfactuals
We use combination of
micro-data: car model sales by zip code and sex
aggregate data: dealer locations, and distribution of other
demographics at zip code level (income, household size,
urbanization, etc.)
Empirical framework: data
We combine four data sets
Car sales: by model, town, sex (Febiac)
Car characteristics: price, horsepower, length, fuel e¢ ciency
(JATO)
Dealers: location and all brands sold at each location
(Spectron Business solutions)
Consumer demographics by area code: men, women, income,
household size, age, immigrants and urbanization (StatBel)
Nr. of obs.: 573,888 = 588 (towns) x 488 (models) x 2 (sexes)
Summary statistics
Variable
Sales
- incumbents
- entrants
Distance (km)
- incumbents
- entrants
Multibrand (0/1)
- incumbents
- entrants
Mean
0.7
1.1
0.4
11.7
7.1
14.7
0.4
0.3
0.4
Std. Dev.
10%.
Median
2.0
0
0
3.3
0
0
1.5
0
0
Dealer characteristics
12.1
2.3
8.4
5.2
1.7
6.0
14.2
3.3
10.9
0.5
0
0
0.5
0
0
0.5
0
0
90%
2
3
1
# Obs.
573,888
223,440
350,448
24.2
13.7
29.8
1
1
1
573,888
223,440
350,448
573,888
223,440
350,448
Summary statistics
Variable
Price (/GDP per cap)
Horsepower (in kW)
Fuel e¤ciency (liter/km)
Length (in cm)
Population(10 3 )
Men(10 3 )
Women(10 3 )
Mean income(10 3 )
Hh size
Age
Immigrants (%)
Urbanization
Mean
Std. Dev.
10%.
Model characteristics
0.9
0.6
0.4
94.9
45.2
51
5.8
1.5
4.3
436.3
43.2
374.0
Household demographics
Median
90%
# Obs.
0.7
85
5.4
440.6
1.4
150
7.5
485.4
488
488
488
488
17.8
28.1
4.0
11.4
32.3
588
8.7
13.7
2.0
5.6
15.8
588
9.1
14.4
2.1
5.8
16.6
588
24.6
2.5
52.7
5.7
5.3
3.5
0.2
1.3
6.7
3.0
20.2
2.3
51.6
1.0
2
24.4
2.5
52.8
3.2
5
29.3
2.6
54.0
14.6
9
588
588
588
588
588
Empirical framework: spatial demand model
Utility of consumer i for buying model j in market t
uijt
= xjt
i
+
i pjt
+ dij +
jt
+
ijt
= xjt + pjt + jt +fxjt ; pjt g Di + fxjt ; pjt g
{z
|
{z
} |
ijt ( i ;D i ;d ij )
jt
Probability that i chooses product j:
Pr ijt ( i ; Di ; dij ) =
1+
exp(
PJ
k =1
jt
+
exp(
i
+ dij +
}
ijt ( i ; Di ; dij ))
kt
+
ikt ( i ; Di ; dij ))
Market share of product j:
Z
sjt = Pr ijt ( ; D; d)dF ( ; D; d)
ijt
Empirical framework: pricing
Pro…t of …rm f who produces subset of models j 2 Ff
X
(pj cj ) sj (p; d) L
f =
j 2F f
Bertrand-Nash equilibrium in prices p (d) solves
qj (p; d) +
X
k 2F f
(pk
ck )
@qk (p; d)
@pj
Remark: as if manufacturers are vertically integrated with
dealers (non-linear contracts)
Empirical framework: dealer entry
Derive bounds on …xed entry cost for an exclusive dealership
lower than variable pro…t drop when dealer would be removed
higher than the variable pro…t increase when dealer would be
added
Account for selection issue as in Eizenberg (2010)
Use these to compute the maximum …xed cost savings below
which ED is pro…table
A move from exclusive dealing...
... to multibranding
From ED to MB
We consider various MB agreements and compute
Pro…t incentive: change in variable pro…ts
Welfare e¤ects:
change in incumbent and entrants’market share
change in consumer surplus
change in total welfare
Estimation: Method of Simulated Moments
Macro moments
Product characteristics xjt , j = 1; : : : ; J, are exogenous,
uncorrelated with jt . This implies:
G 1( ) =
1 XX
TJ t
jt (
)zjt
j
where zjt includes product j’s own characteristics xjt and sums
of other products’characteristics, and jt is
jt (
)=
jt (s
obs
; )
xjt
pjt :
Estimation: Method of Simulated Moments
Equate predicted micro moments to observed moments in the data
Average consumer demographics Di :
G 2( ) =
=
pred
(
D1
obs
D1
XXX
t
i
)
(qijt
Prijt ( )Lit )Di1
j
Covariance between consumer demographics Di and product
characteristics xj .
Average dealer characteristics dij .
Variance of mean product characteristics accross submarkets.
Results from spatial demand model
variable
const
- const
- const
- const
- 1
urban
age
hhsize
price
- princ
- 2
income
hp
- hp
- 3
age
length
- length
- length
- 4
df
immigrants
coe¢ cient
23.02
-0.63
-0.13
-7.39
-9.42
std.err.
9.55
0.10
0.18
0.84
1.62
t-value
2.41
-6.06
-0.71
-8.84
-5.81
-6.01
0.86
-0.39
0.61
0.11
0.10
-9.80
7.58
-3.85
4.47
-0.00
0.04
1.72
0.03
0.22
2.61
-0.11
0.16
6.29
-12.48
-0.22
3.15
2.44
0.54
0.04
0.47
2.58
-23.05
-5.46
6.65
Results from spatial demand model
variable
fuel
- fuel df
- fuel urban
- 5
coe¢ cient
-8.55
-0.27
0.14
-1.27
std.err.
0.84
0.22
0.03
0.17
t-value
-10.16
-1.23
5.50
-7.62
distance
-0.14
0.01
-17.24
multibrand
-0.04
0.02
-2.80
year dumies
origin dumies
yes
yes
Results from spatial demand model (cont’d)
Mean utility parameters: price and fuel consumption have
negative impact, horsepower has positive impact
Distance & multibrand: negative impact
Observed consumer heterogeneity:
high income households less price sensitive
women prefer smaller cars
city households less sensitive to fuel e¢ ciency
city households more likely to choose outside good
Unobserved consumer heterogeneity:
strong heterogeneity, regarding outside good, price, size, fuel
e¢ ciency
Counterfactuals
Multibrand agreements
between 8 incumbents and 24 smaller entrants
involves shift in distance matrix and multibrand dummy
“one-way access”: entrants at incumbents’locations (if
pro…table), but not vice versa
Two types of counterfactual
single multibrand agreements: unilateral incentives?
joint multibrand agreements: collective incentives?
Variable pro…t incentives of multibranding
Peugeot & 3
Citroën & 3
Opel & 3
Renault & 3
Volkswagen & 3
Ford & 3
Audi & 3
Toyota & 3
Mean
Total
Current
joint
156.51
166.27
201.43
187.63
256.31
182.69
183.28
192.53
190.83
1526.64
Unilateral agreement
Change in variable pro…ts
big
small
joint
-0.01
0.14
0.13
0.00
0.05
0.05
-2.76
7.98
5.22
-2.12
3.12
1.00
-0.80
8.07
7.27
-1.00
1.95
0.95
-0.18
3.99
3.81
-0.88
2.37
1.49
2.49
19.92
Collective agreement
Change in variable pro…ts
big
small
joint
-2.10
0.07
-2.03
-2.29
-0.02
-2.31
-4.07
7.08
3.01
-3.91
2.36
-1.54
-2.07
6.80
4.72
-2.42
0.83
-1.59
-0.93
2.55
1.62
-1.87
0.83
-1.04
0.10
0.84
Variable pro…t incentives of multibranding
Unilateral MB agreements increase variable pro…ts, on average
by +2.49 mil.
this is because of business stealing: demand " from more
spatial availability outweights demand # from reputation loss
Joint MB agreements increase variable pro…ts much less, in
total by only +0.84 mil.
this is because of increased price competition, and no scope for
business stealing
Conclusion: maybe unilateral, but no collective incentives to
make MB agreements with entrants.
Total pro…t incentives of ED: …xed costs
Variable
Variable
Number
Variable
pro…ts
pro…t increase
of brands added at incumbents’locations
pro…t increase per brand added
Fixed costs of add. brands that rationalize ED
(as % of estimated bounds of FC of single brand)
- at lower bound (e31,990)
- at mean (e76,265)
- at upper bound (e120,540)
Unilateral
agreement
Collective
agreement
e2.49 mil.
175
e14,220
e0.84 mil.
1401
e598
>44%
>19%
>12%
>2%
>1%
>0%
Unilateral incentive for ED if …xed costs of additional brands
are at least 12%–44% of …rst brand
Collective incentive for ED if …xed costs of additional brands
are least 0–2%
=> collective incentive for ED for large range of extra …xed
costs, unilateral incentive for much smaller range
Welfare e¤ects of shift to multibranding
Current
Outside good
Inside goods
Incumbents
Entrants
78.8%
21.2%
60.8%
39.2%
Incumbents
Entrants
Total
951.7
574.9
1526.6
Total (e million)
Per hh (e)
Total
Total
change
Change due
Change due
to distance
to MB
Market shares (%)
-0.04%
-0.06%
0.02%
0.04%
0.06%
-0.02%
-1.3%
-1.4%
0.1%
1.3%
1.4%
-0.1%
Variable pro…ts (e million)
-19.7
-20.1
0.9
20.5
23.0
-2.1
0.8
2.9
-1.2
Consumer surplus
77.8
116.7
-44.3
104.2
156.3
-59.4
Welfare (e million)
78.6
119.6
-45.5
Change due
to prices
0.00%
0.00%
0.06%
-0.06%
-0.47
-0.45
-0.92
5.4
7.3
4.5
Welfare e¤ects
A ban on exclusive dealing would
shift market shares mainly due to increased spatial availability
from incumbents: –1.3%
in favor of entrants: +1.3%
bene…t consumers: 77.8 e million or 104.2 e per household
gains from increased availability: +156.3 e per household
losses from demand ine¢ ciency: -59.4 e per household
gains from reduced prices: +7.3 e per household
bene…t welfare: similar to consumer e¤ects
Conclusions
Spatial coverage has a positive impact on the demand for
cars, but consumers dislike multibranding
Pro…t incentives of exclusive dealing
stem partially from demand e¢ ciency
and also from keeping prices high
External e¤ects of ban on exclusive dealing
Positive impact on the entrants’market share
Positive impact on consumer and total welfare, mainly because
of increased spatial availability.