MCTM Teams

American Economic Association
Too Many Products: Decentralized Decision Making in Multinational Firms

Author(s): Catherine Thomas
Source: American Economic Journal: Microeconomics, Vol. 3, No. 1 (February 2011), pp.
280-306
Published by: American Economic Association
Stable URL: http://www.jstor.org/stable/41237178
Accessed: 01-06-2016 18:15 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to
American Economic Journal: Microeconomics
This content downloaded from 152.3.102.242 on Wed, 01 Jun 2016 18:15:18 UTC
All use subject to http://about.jstor.org/terms
American Economie Journal: Microeconomics 3 (February 2011): 280-306
http://www.aeaweb.org/articles.php ?doi=10.1257/mic.3. 1.280
Too Many Products:

Decentralized Decision Making in Multinational Firms*
By Catherine Thomas*
/ analyze country-level product ranges offered by multinational

laundry detergent manufacturers in Western Europe. Observed prod-
uct range variation across countries exceeds the optimal firm-level
response to differences in consumer preferences and retail environ-
ments. Counterfactual analysis reveals that increased product range
standardization would reduce firm costs and increase profits. These
findings are consistent with theory models of local agency, where
decentralized decision making can be the constrained optimal orga-
nizational form despite the resulting lack of coordination across
divisions. My analysis suggests that organizational structure affects
product market outcomes and firm performance. {JEL D23, F23,
L21, L25, L65)
decision making is a widely observed characteristic of large mul-

tidivisional firms. This paper conducts an empirical investigation of the con-
sequences of decentralizing product range choice within multinational consumer
goods firms. The setting is the laundry detergents industry in Western Europe. The
industry is dominated by a few large multinationals (MNCs), all of which had a sim-
ilar organizational structure for the time period of this study. While production and
research and development (R&D) were relatively centralized in the region, product-
range choices were made at the brand-country level, by brand-country managers
facing incentives related to the local performance of the brand.
The within-firm product varieties available on supermarket shelves vary sub-
stantially across different markets. Even for a given multinational detergent brand,
the composition of the product range differs in terms of the number of products
and the characteristics of the products sold in each country. Some variation in
product range is optimal for the firm due to varying local consumer preferences
and differences in the vertical relationship between the firm and the retail sector.
At the same time, economies of scale in production create incentives for product
range coordination.
♦Columbia Business School, Uris Hall 605B, 3022 Broadway, New York, NY 10027 (e-mail: cmt2122@
columbia.edu). This paper is based on my dissertation work at Harvard University. I would like to thank AiMark for
providing the data, Henkel GmbH for allowing me to visit their production facilities, and Aviv Nevo for making his
code publicly available. I would also like to thank Pankaj Ghemawat, Elhanan Helpman, Tarun Khanna, and Ariel
Pakes for their advice. This paper benefited from helpful discussions with Pol Antras, Wouter Dessein, Ray Fisman,
Amit Khandelwal, Ben Lockwood, Emi Nakamura, Michael Raith, William Simpson, Debbie Strumsky, and many
seminar participants. Financial support from Harvard Business School, Columbia Business School, and the Global
Initiative at NYU-Stern are gratefully acknowledged. All errors are my own.
ŤTo comment on this article in the online discussion forum, or to view additional materials, visit the article page
at http://www.aeaweb.org/articles.php?doi= 10. 1 257/mic.3. 1 .280.
280
VOL 3 NO. 1 THOMAS: TOO MANY PRODUCTS 281
This paper shows that the observed variation in products, offered by the same
firm in different markets, cannot be solely attributed to differences in local con-
sumer demand or retail environment. Using a structural model, I estimate the effects
on demand and firm profit of an increase in the degree of cross-country standardiza-
tion in product range. Selling fewer product varieties across a wider set of markets,
while keeping the number of products sold in each market constant, does not sig-
nificantly decrease variable profit. Increased standardization would allow the firms
to reduce the total number of product varieties manufactured in the region. At the
time of the data used in the study, Procter and Gamble (P&G) could have profitably
eliminated up to 20 percent of the Ariel brand product varieties sold across three
Western European countries, and Unilever could have ceased production of up to
30 percent of the Surf brand product varieties sold in the same markets. Firms are
manufacturing too many products, in that reducing the number of product varieties
for a given brand would reduce total costs and increase firm-level profits.
Recent developments in the organizational economics theory literature offer an
explanation for the observed excessive localization in decentralized firms. The mod-
els presented by Ricardo Alonso, Wouter Dessein, and Niko Matouschek (2008) and
Heikki Rantakari (2008) translate most directly to the empirical setting analyzed
here. These authors describe how decentralized decision making can be the optimal
organizational form in the context of centrally unobservable local conditions despite
a resulting lack of coordination. Excessive localization by decentralized decision-
makers is observed in equilibrium due to division manager agency. Applied to the
setting examined in this paper, each local manager choosing a range of products for
his local market does not consider the full impact of his choices on the costs of the
other divisions and the firm as a whole.
The empirical challenge addressed here is establishing that local product range tai-
loring is suboptimal at the firm level even in the face of the diverse market conditions
which prompted the decentralized decision making structure to begin with. To do this,
I first measure preference variation across countries, following Aviv Nevo (2001), and
then control for both demand and retail sector heterogeneity. I estimate a discrete choice
random coefficients demand system using data from four Western European markets to
estimate local demand. The results establish that underlying consumer preferences for
laundry detergent characteristics are significantly different across countries.
Focusing on the Ariel and Surf brands - the two most widely sold brands in the
data - a set of alternative product ranges that represent increased firm-level coordi-
nation are constructed for each brand, in each country where the brand is sold. An
alternative product range involves a minor product substitution, when a local product
of a given brand and format is replaced with a product that is currently being sold in
another market that is of the same brand and format, but differs slightly in pack size.
For example, at the time of the data, Ariel sells a 30 washload pack of concentrated
liquid in the United Kingdom (UK), and a 35 washload pack in Germany. Both packs
are produced in the same manufacturing plant. One alternative product range for Ariel
in the United Kingdom, hence, involves replacing the 30 washload pack with the 35
washload pack.
The analysis holds the number of products sold in each market fixed while allow-
ing firms to reduce the total number of products manufactured in the region. In
282 AMERICAN ECONOMIC JOURNAL: MICROECONOMICS FEBRUARY 2011
doing so, the approach controls for cross-country variation in the local relationship
between manufacturing firms and retailers, and any strategic decisions related to any
one firm's product line size.1 The effect on brand-level and firm-level variable profit
is evaluated for each alternative product range. When an increase in standardiza-
tion does not decrease local variable profit, and allows the firms to reduce the total
number of product varieties manufactured in the region, the firm could save the fixed
production cost of the local varieties that are no longer produced, and also benefit
from any lowering of marginal cost due to increased unit production runs.
Existing studies of multiproduct multinational firms include models which explain
the number - rather than type - of products sold by a given firm in each national
market. Examples include Costas Arkolakis and Marc- Andreas Muendler (2009), in
which differences in the size of the product range result from diseconomies of scope
in market entry, and Andrew B. Bernard, Stephen J. Redding, and Peter K. Schott
(2010), which includes one-dimensional variation in country-level demand for a
firm's products. In these papers, each firm sells its highest productivity products in
each country where it operates so that variation in product range composition comes
about only through the number of products offered. In contrast, this paper allows for
within-firm cross-country variation in the number and types of products marketed
resulting from multidimensional local heterogeneity. Allowing for heterogeneity has
two separate effects. First, the composition of the optimal product range for each
market can vary due to the differences in local conditions - the constraints imposed
by local retailers, the actions of competitor firms, and variation in underlying local
consumer preferences. Second, diverse local conditions prompt the firm to adopt a
decentralized organizational form.
The large and growing organizational economics literature on decentralization
originates in the work of Michael C. Jensen and William H. Meckling (1976).2
Much of the literature on multidivisional firms in economics and accounting takes
the presence of division-level agency as a starting point for a discussion of within-
firm hierarchies and incentives structures without any empirical evidence of its
relevance.3 This paper establishes that multidivisional firms do not align decision
making by division managers in this empirical setting, consistent with the presence
of an underlying agency problem.
A related empirical literature examines different explanations of why firms
decentralize decision making. Examples include Massimo G. Colombo and Marco
1 Rather than address whether firms are engaged in product line length competition of the type discussed in
Michaela Draganska and Dipak C. Jain (2005), or whether retailers and manufacturing firms are price discriminat-
ing within-country, for example across brands of differing perceived quality, as in Draganska and Jain (2006), the
structure of the counterfactual analysis allows for both to exist in each market.
Along with Alonso, Dessein, and Matouschek (2008), and Rantakari (2008), other theory papers relating
organizational form to information or knowledge asymmetry between local divisions and the coordinating center
include Jensen and Meckling (1995); Philippe Aghion and Jean Tiróle (1997); George Baker, Robert Gibbons, and
Kevin J. Murphy (1999); Oliver Hart and John Moore (2005); Dessein, Luis Garicano, and Robert Gertner (2010);
and Guido Friebel and Michael Raith (2010). Other strands of the theory literature on organizational form examine
the relationship between decentralization and characteristics of firm technology, such as Raghuram G. Rajan and
Luigi Zingales (2001).
3 Examples of papers in this strand of the accounting literature are: Stanley Baiman, David F. Larcker, and
Madhav V. Rajan (1995); Robert M. Bushman, Raffi J. Indjejikian, and Abbie Smith (1995); Margaret A. Abernethy,
Jan Bouwens, Laurence van Lent (2004); and Bouwens and van Lent (2007).
VOL. 3 NO. 1 THOMAS: TOO MANY PRODUCTS 283
Delmastro (2004); Daron Acemoglu et al. (2007); and Nicholas Bloom et al. (2009)
who ascribe variation in organizational form to differences in the firm-level produc-
tion function, such as the operational complexity, distance to technology frontier,
and the specific nature of internal technologies, respectively. Maria Guadalupe and
Julie Wulf (2010) examine the relationship between firm organizational form and
industry-level product market competition. A subset of this literature also estab-
lishes a causal effect of organizational form on other aspects of the firm-level pro-
duction function, such as labor skill mix in Eve Caroli and John Van Reenen (2001)
and R&D activity in Jakub Kastl, David Martimort, and Salvatore Piccolo (2008).
This paper provides some of the first evidence that the observed firm-level organiza-
tional form has consequences for product market-level outcomes that are consistent
with predictions drawn from recent theory.
The paper is organized as follows: Section I describes how decentralized deci-
sion making about the composition of local product ranges generates the prediction
that the firm will manufacture too many products overall. Section II contains a brief
description of the features of the laundry detergents industry in Western Europe that
are relevant to this study. The data used for the analysis are described in Section III.
Section IV presents the analysis of local demand conditions, the results of which
establish that consumer preferences for detergent products differ across countries.
Section V is an investigation of whether firms can profitably increase cross-country
product standardization, given the estimated differences in local demand, while con-
trolling for the local retail environment. Section VI concludes.
I. Agency in Product Range Choice
This section describes a simplified model of the local brand manager's choice of
the product range Jb >c, for brand b in country с in a decentralized firm /at any given
point in time, illustrating the agency problem associated with this organizational
structure when production is centralized. A product, j9 is defined here and through-
out the paper as a brand, format, and pack-size combination.
Firm-level profits are the sum of the revenues generated in each product market
for each brand given the range of product varieties on sale in each market, less the
centrally incurred fixed and variable production costs.4 Marginal production costs
are assumed to be constant for a given product, and there is a per period fixed cost
for each product.5 Firm profits can be written:
(1) П/=ЕЕЕ Puqu - E 'mCjE ЯМ ~ FA - Ff9

where qj,c is the quantity of product j sold in market с and pj,c is its local price. mCj
is the marginal cost and Fj is the fixed cost associated with production of product
4 The major firms in the laundry detergents industry dominate many consumer products industries. Each firm
here is assumed, for simplicity, to manufacture and sell product varieties of one product category.
If, instead, economies of scale were modeled as decreasing marginal costs within product variety, the agency
costs associated with decentralization would be greater in magnitude since the production runs of each product
variety are lower under decentralization.
variety./. Product-level fixed costs are assumed to be independent of the number and
identity of the countries in which the product is sold. The firm may incur overhead
costs and brand-level fixed costs in each country where the brand is sold; these are
summed over brands and denoted Ff. The decision about which brands to sell in each
country, and brand-level fixed costs, are assumed to be fixed throughout this paper.
The focus here is on the composition of Jbc, given that brand b is present in
country c. Each of the large firms in this industry has responded to the challenge
posed by variation in local demand conditions by decentralizing some aspect of
decision making to the country level. At the time of the data used in this study,
firms employed brand-country managers who were responsible for setting prices
and choosing the range of product varieties sold under the brand name in each local
market. The brand-country managers faced local incentives, tied to the profits of the
brand in their respective markets.6 The manager's objective function can be written:
(2) 7Tb,c = E (PU - ku)qj,c.
Rather than observing the marginal production cost for each product variety, mcj9
the manager observes a unit cost kjc which is the per unit transfer price for the prod-
uct in question. Conditional on the product range, Jb >c, and the product-level transfer
prices, kjiÇ9 the local manager will set prices to maximize equation (2). Holding fixed
the total number of products sold under the brand in each country, тгЬс is assumed to
be maximized under the observed product range Jbc.
To illustrate the agency problem in product range choice arising from the decen-
tralized organizational structure which is the focus of this study, it is useful to restrict
attention to the question of which one of two alternative products is chosen by the
brand-country manager to complete the product range of a given size. One of the
products, /, is a standardized product and will be manufactured by the firm in any
case since it will be sold in another country market. The other product j, if chosen,
will be a local product, manufactured and sold only in country c. None of the other
country marketing divisions will find it profitable to include product j in their own
product range. Under the assumption that the brand-country manager in country
с chooses the product range to maximize equation (2), product range Jbc, which
includes product у , will be observed rather than product range Abc, which includes
product у 'if:
(3) E Puqu - E Puqu > E he®* - E h^-

№ь,с j€AbjC jeJbiC jeAb¿
6 The empirical implications of this model arise from the fact that local managers do not fully incorporate the
effects of their product choices on firm-level costs. A number of different organizational structures could lead to
this outcome. I focus here on a model based on localized incentives since this relates most closely to features of the
organizational structure in the major firms at this time. Costly communication between local brand managers could
generate similar empirical implications.
This inequality demonstrates that the locally-tailored product j will be chosen

whenever the increase in revenues with the local product relative to the standardized
product outweighs the increase in variable costs, as seen by the local manager.
The firm, however, would prefer the local product j to satisfy a more stringent
criterion to be included in country c's product range. If the product range decision
were centralized, and the decision-maker had access to all the relevant information,
j would be chosen only if the left-hand side of inequality (3) exceeded: Fj +
E ej mcjclhc ~~ Е-€Л mcjclj^ The firm would manufacture and sell product
j in market с only when the increase in revenues from including j rather than /
exceeded the incremental marginal costs and the fixed costs incurred from produc-
ing ]. Specifically, the possibility arises that the brand-country manager will market
product range Jbc when the product range Ab c would generate greater firm-level
profits whenever:
(4) F] + E mcj1hc - E mcj4U > E киЯм - E kj,cVU.

J£Jb,c J€Ab,c J€Jb,c J€Ab,c
While firm-level transfer pricing methods are often fairly opaque, especially at the
product level, the first part of the Web Appendix shows that expression (4) may well
hold under the two most frequently employed methods of transfer pricing: market
price transfer pricing and full cost transfer pricing. If the difference in revenues
between Jbc and Abc, as given on the left-hand side of inequality (3), lies between
the values of the left- and right-hand sides of inequality (4), the firm will manu-
facture too many products as a consequence of decentralizing the product range
choice decision. That is, reducing the total number of product varieties manufac-
tured within the region would increase firm profit.
This inference guides the counterfactual analysis described in Section V. To pro-
vide evidence that observed product ranges reflect the agency problem described in
this section, it is sufficient to show that the firm would be at least as well off under
a more standardized product range, reducing total production costs. The laundry
detergents industry has some features which facilitate this empirical comparison, but
it also presents some challenges. These will be discussed in the following sections.
II. The Laundry Detergents Industry in Western Europe
Laundry detergent, sometimes referred to as heavy-duty detergent, typically

refers to all washing machine detergents whose primary purpose is the laundering of
clothes and other textiles (and excludes fabric softeners and handwashing products).
Data from four countries are used in this paper: the United Kingdom, Germany,
Italy, and Spain. Anecdotal evidence suggests that consumer preferences vary across
these countries. Explanations for cross-country differences refer to persistent coun-
try-specific laundry habits, reflect differences in washing machine technology such
as front- versus top-loading machines, and are also related to the legacies of differ-
ent brand histories in each country.
The industry is dominated globally and especially within Europe by a small
group of large MNCs. A firm-level analysis in this concentrated industry represents
Table 1 - Summary of MNC Activities in Each Country
UK Germany Italy Spain

Number of brands detailed in Europanel product-level dataset 20 20 19 20
Of which: Number of MNC brands* 7 9 14 13
Of which: Number of international MNC brands** 3 2 4 5
Typical per period number of distinct products detailed in dataset* * * 1 80 1 34 1 54 135

Of which: Number of MNC products 89 72 120 92
Of which: Number of MNC products for international brands 66 42 21 20
Of which: Number of international brands products also on sale 14 4 17 20
elsewhere
* Brands manufactured and sold by P&G, Unilever, Henkel, and Reckitt Benckiser.
** Brands sold in at least one other country in dataset.
*** A product is a particular brand, format and pack size combination.
Source: Europanel dataset.
a relatively large fraction of the entire market in each country.7 Table 1 presents
some summary statistics about MNC activities in each country at the time of the
data used in this study. MNCs sell both country-specific and international brands.
Even for international brands, only a fraction of all products manufactured under
each brand are sold in more than one country. This is consistent with Christian
Broda and David E. Weinstein (2008) who find that a large fraction of cross-border
price differences at the industry level can be attributed to product range composi-
tion effects for a broad range of industries. Throughout Europe, the bulk of branded
MNC detergents are sold through supermarkets and hypermarkets. Even in Italy
and Spain where smaller retail formats are relatively prevalent, 87 percent and 84
percent, respectively, of laundry care product sales are made through supermarkets,
hypermarkets and discounters (Euromonitor 2003b, 2003c). Recent years have seen
market-share growth in Europe for private label detergents, competing mainly on
price.8
The UK market tends to have the greatest product diversity, and relatively recently
introduced format varieties - liquids, tablets, and capsules - are widespread across
MNC brands and private labels. Liquids and concentrated formats were first seen in
the 1980s. The tablet format was first introduced by Unilever in the United Kingdom
in 1998, and was rapidly copied by other firms so that it was widespread in Europe by
1999. Capsules were introduced by P&G in the United Kingdom in 2001. The 2003
Euromonitor report on Germany (Euromonitor 2003a) emphasizes that German
consumers tend to be price sensitive due to both the widespread presence of hard
discounters and also to the recessionary macroeconomic environment at that time.
German consumers are also said to use less product per household, mainly because
they are more careful about precise dosage due to environmental considerations.
Regarding Italian preferences, Euromonitor (Euromonitor 2003b) goes so far as to
7 The Enterprise Directorate General of the European Commission, and others, refer to these firms collectively
as "The Big Soapers" (European Commission Enterprise DG 2001 ) . P&G and Unilever together have a joint market
share of between 20 percent and 72 percent in each country in the data.
8 Supermarkets tend to sell both MNC branded detergents and own label products. Hard discounters (for exam-
ple, Aldi and Lidi) tend to sell only their private label varieties.
remark that products marketed as environmentally friendly are "unpopular." As of

2003, capsules were not available in Italy but, following the introduction of the new
Marseille (Marsiglia) soap format by a small independent firm, the MNCs offered
some of their main brands in this format.
The major firms in this industry are viewed primarily as brand managers and mar-
keters, and there is relatively little publicly available information about the supply
side. Of particular relevance to this study is the fact that between 2000 and 2003,
a brand manager for a given brand in a country was accountable to both a country
manager and a regional brand manager, but decisions within brand tended to be
made at the country level. Brand-country managers' incentives were related to the
local performance of the brand. Both brand-country profit and market share were
considered to be important measures of local performance. Production and other
functional divisions were more centralized, typically at the regional level.
Nonetheless, there is substantial anecdotal evidence that product variety-level
economies of scale, either in terms of product-level fixed costs or declining within-
product marginal costs, are significant for detergents and in the production of con-
sumer goods in general.9 An industry wide initiative, described in Ralph Drayer
(1999), details how firms in the industry together developed a program during the
1990s called "Efficient Consumer Response," one of the four aims of which was
the "Efficient Assortment" strategy which aimed to reduce "duplication of SKUs
while maintaining the optimal product assortment to meet customers' needs." 10 The
increasing prevalence of multilingual packaging in many world markets is indirect
evidence that firms find it worthwhile to consolidate production lines to produce one
variety for multiple markets when feasible.
In 1999, P&G embarked on its "Organization 2005" global restructuring plan. One
of the stated aims was to invest in "cost reduction through the standardization of pro-
cedures resulting in better economies of scale" (Euromonitor 2003e).1 1 Drayer (1999)
describes how P&G aimed to improve supply chain management of the "proliferation
of product, pricing, labeling, and packaging variations." He notes that "the bloating
of the supply chain with product, together with the proliferation of product variations
related to promotions, increased manufacturing costs." In 2000, Unilever launched its
"Path to Growth" restructuring program. A further phase of restructuring - Unilever
2010 - was launched in 2004. The goals of these programs included a more stream-
lined brand portfolio - moving from 1,600 brands in 1999 to a target number of 400 by
the end of 2004 - and improved global scale, procurement, and media. Euromonitor
(2003f) reports that the goal was to "transform Unilever from 'hopelessly local' as
9 Thomas (2006) estimated some bounds on feasible product level fixed costs under the assumption that mar-
ginal costs could be estimated from prices and firm-level first order conditions. This analysis was based on the
assumption that the size of the product range was optimal given marginal costs. The inferences made are to be
treated with caution since they ignore decentralized price setting which is characteristic of these firms. Nonetheless,
the lower bound on the product level fixed cost in each market (extrapolating from the data based on the total num-
ber of households in each country) was estimated to be at least 70,000 euros per 8- week period.
10One example of the results of this program is from P&G's Head and Shoulders shampoo brand in the United
States. Steven M. Cristol and Peter Sealey (2000) note that in the early 1990s, "P&G reduced formulas and packag-
ing variations until the company's total number of hair care SKUs were cut by half." An SKU (stock keeping unit)
corresponds to the definition of a product used in this paper.
1 1 See Mikolaj Jan Piskorski and Alessandro L. Spadini (2007) for a detailed discussion of the evolution of
P&G's organizational structure.
management described it, into a coherent 'multi-local-multinational' organization that

can get the balance right between the economic benefits of global scale and the mar-
keting advantages of optimal tailoring to local customers."
Regulations across the European Union (EU), that have come into force since
1992, mean that there are now fewer differences in chemical composition of deter-
gent product across countries. An EU Commission report from 1991 stated that
"the detergent industry contains large volumes of substantially similar products"
(Enterprise DG, EU Commission 1991). Beata Javorcik, Wolfgang Keller, and
James Tybout (2008) comment that economies of scale in detergent production
led P&G to consolidate its US production plants from 14 to 4 between the late
1970s and 2005. As of 2003, P&G manufacture of detergents in Europe was fairly
geographically concentrated. Liquid products and liquid capsules for all the firm's
brands were manufactured in Amiens, France. All powder tablets for all brands were
manufactured in Mechelen, Belgium. By far, the majority of powder products was
manufactured in London, United Kingdom.12 At this time, Unilever's key detergent
manufacturing plants in Europe were at Port Sunlight and Warrington in the United
Kingdom, with some production facilities in Spain and Italy. The production plants
tended not to be country-specific, and the smaller plants in Southern Europe fre-
quently produced the product varieties sold in the United Kingdom.
III. Data
The laundry detergent product data used in the paper come from a panel for each
of the four countries, covering the years from 2000 to 2003. They are proprietary
data collected by the market research firm Europanel which surveys a representative
set of households on their shopping purchases and consumption habits.13 The UK
panel consists of 15,000 households, the German panel of 12,000, the Italian panel
of 5,000, and the Spanish panel of 6,000. Consumers in each country record their
purchases and this information is then aggregated to the product level. The data are
available for each four-week period over the three years.
Each brand-format pack-size level observation includes the physical quantity of
product purchased and the total amount spent on the product in a four- week period.
Products included in each panel are all products on sale at that time for each of the
top 20 brands in a country, where the top 20 is as defined at the end of the three years
in question.14 Also given in the data is the total quantity purchased and amount spent
on the entire category of detergent in each time period. In the demand specification
12 P&G also manufactured some powders in Italy and Spain for brands sold only in those countries. Powders
tend to have the highest weight to value ratio, so transportation is relatively expensive for these formats.
3 Europanel is a joint venture between the UK marketing firm TNS and the German marketing firm GfK. The
company is careful to ensure that each panel is representative and participants are compensated in non-monetary
ways so as not to distort their purchasing behavior. More information on their sampling methods is available at
www.europanel.com. Since all household detergent purchases are included in the data, the product level data is
comprehensive.
The ranking of brands by market share is more or less constant in each country between 2000 and 2003.
Germany is an exception to this since P&G introduced a new brand - Meister Propper - towards the end of the time
period. For most months in Germany, then, the dataset contains product level detail for only 19 brands. New private
label brands were also introduced in the data in some countries. In Italy, there appears to be a data issue with the P&G
brands Bolt and Ace. The data record them as having identical sales at the product level over the time period. Since
in the following section, each household is assumed to purchase either a top 20 brand
product or some other product in each time period. The top 20 brands typically make
up between 70 percent and 96 percent of the total quantity of detergent sold in each
country, as recorded in the data. Hence the data include product-level information
for a large fraction of the total market in each country together with an accurate
measure of the total size of each market.
A key feature of the product-level dataset is that it demonstrates very clearly the
extent to which the product range choices made by brand managers differ across coun-
tries. There are many cases where a product appears in one market that is similar but
slightly different to a product sold elsewhere. Since production is centralized, this
fact is, in itself, suggestive of some production inefficiency and serves to motivate the
product-level analysis in Section V. The existence of very similar products on sale at
any one time also provides candidate products to include in alternative product ranges
when assessing the impact on firm-level profits of increased standardization. These
products are already being manufactured and all the fixed costs of development and
production are sunk, so switching existing products in and out of the product range for
any one country will not affect total fixed costs at the firm level.
Two smaller miscellaneous dataseis are also used in the paper. The first is a
conversion table of each format of laundry detergent from weight and volume mea-
sures, given in metric units for each country in the data, to the actual number of
washloads of detergents contained in each pack. This is used to create a standard-
ized quantity unit.15 Second, average monthly exchange rates for 2000 to 2003 are
used to convert the UK data from pounds into euros. These data came from the
website www.x-rates.com (x-rates.com 2003).
IV. Estimating Local Conditions
Establishing that observed product ranges are suboptimally localized from the
firm's perspective requires that the underlying variation in local conditions be mea-
sured and then taken into account. The product-level data about consumer purchases
in each country are used to estimate local consumer demand in a discrete choice
random coefficients framework incorporating both product characteristics and
unobserved household attributes, based on Daniel McFadden (1974); Steven Berry,
James Levinsohn, and Ariel Pakes (1995); and following Nevo (2001). 16 As in Irene
Brambilla (2007), allowing demand to vary across countries permits a relatively
detailed study of the complexities of management practices within these multina-
tional multiproduct firms.
A market is defined as an eight-week period in each country, where two four-
week time periods are aggregated to form each market as described in the Web
this is likely to be an error, the product level data on one of the brands is discarded, effectively, it is included in the
outside good. The Italian data thus detail only 19 brands at the product level.
15 There is an EU regulation to the effect that the number of washloads worth of detergent in each pack must
be printed on the outside of the pack. This information was gathered during a series of store visits in the United
Kingdom and Germany in the summer of 2005.
16 The Web Appendix contains a more detailed description of the estimation process and a description of the
instrumental variables strategy.
Appendix. The data contain information from 19 markets for the United Kingdom,
22 for Germany, 19 for Italy, and 19 for Spain. The conditional indirect utility of
household i in country с from using one washload's worth of detergent product j at
time r, rather than a detergent product that is not from one of the top 20 brands, is
made of up of three parts, öjAc, /¿/¿¿c, and e^u :
(5) W/Дс = SjAc(XjAc,WjAc,pj,t,c,A^c^l)

+ V>ijAc(Wj,t,c>PjAc,Vhe2) + €lVAc,
where Xjtc is a matrix of zeros and ones constructed to represent the brand and
format of producty, together with a constant term. wjAc is the pack size of the product
in units of the number of washloads found using the format conversion table. pj,t,c
is the price per washload in euros.17 Д£/лс captures any unobserved time-specific
deviation in average consumer utility obtained from the product relative to the mean
valuation of the brand-format-pack size combination, v, represents unobserved
attributes for each household i.
SjAc is the mean utility experienced from each product, common to all house-
holds. It is given by:
where the parameters ßc, 7C, and ac together make up the vector 61? and are country-
specific so that the mean preference for each brand and format can vary across
countries. ¡iiit,c is a household-level, mean zero heteroskedastic deviation from each
country's mean product-level utility which measures the effects of the interactions of
unobserved household attributes with the product characteristics being pack size and
price in the model. This term varies across households within a country. It is given by:
№u = ^^;i%c + VpViiPiu,
the parameters aw and ap make up the vector 02. The magnitudes of the interaction
coefficients do not vary across countries. Each household has the option to purchase
the outside good (a product that is not one of the top 20 brands) if it obtains greater
utility from doing so than from purchasing one of the products from one of the top
20 brands. Instrumental variables are used to address potential price endogeneity.
Table 2 presents a subset of the parameter estimates for the vector 61# The brand-
country level estimated coefficients for the P&G brand Ariel and the Unilever brand
Surf are shown in the table. These two brands will be the focus of the remain-
der of the paper since, as mentioned in the introduction, they are the two brands
17 The UK data is converted to euros using the mean exchange rate between 2000 and 2003. Converting to euros
using each time period's exchange rate would introduce a fluctuation in price to the UK data that is not experienced
by UK consumers.
Table 2 - A Subset of Estimated 0l Coefhcients, Full Model with Instruments
Variable UK Germany Italy Spain

Unilever, Surf 1.09*** 0.40*** ÕÍ9
(0.17) (0.11) (0.23)
P&G, Ariel 2.07*** 0.71*** 0.13 1.45***
(0.27) (0.19) (0.16) (0.22)
Liquid 1.11*** 1.04*** 0.43*** 0.00
(0.26) (0.12) (0.09) (0.11)
Tablet 2.40*** 1.12*** -0.07*** 0.59***
(0.41) (0.17) (0.21) (0.23)
Capsule 3.14*** -0.60**
(0.51) (0.27)
Gels/Other 0.69***
(0.19)
Liquid soap Marsiglia 0.63***
(0.15)
Concentrated -1.37*** -0.45*** -1.20*** -2.44***
(0.31) (0.13) (0.25) (0.39)
Pack-size coefficient -0.02* 0.01 -0.01 0.01***
(0.01) (0.01) (0.01) (0.00)
Price coefficient -23.66*** -30.21*** -19.17*** -28.10***
(5.06) (5.60) (5.27) (6.42)
Constant-term coefficient 1.14 -2.07** - 1 .84 -1.81**
(1.70) (1.04) (1.13) (0.74)
Notes: These estimated brand (format) coefficients can be interpreted as the additional utility the average household
obtains by using this particular detergent brand (format) for one washload of laundry rather than one of their coun-
try's private label brands (standard powder). The last three variables are the estimated coefficient associated with
pack size (number of washloads), price per washload (euros), and each country's private label brands (the constant
term). The second row contains the standard error estimate for the coefficient given above.
***Significant at the 1 percent level.
** Significant at the 5 percent level.
♦Significant at the 10 percent level.
aKnown as Bio-Presto Surf in Italy.
that are sold most widely across the four countries studied. The country-specific
mean brand preference parameter estimates tend to be positive and significantly
different from zero, suggesting that the average consumer in the relevant country
gains posctive utility from using one of these particular MNC brands, rather than a
domestic or private label brand in the top 20 country brands to do one washload of
laundry. Consumers in each country tend to value other formats more highly than
standard powder, holding price, pack size, and brand constant. The two exceptions
to this are that the typical Spanish consumer prefers standard powder to capsules,
and the typical Italian consumer prefers powders to tablets. Consumers in the United
Kingdom value newer formats - tablets and capsules - particularly highly, on aver-
age. Consumers in all countries, and particularly in Spain, attach a large discount to
concentrated formats. This is consistent with anecdotal evidence in the Euromonitor
reports.18 The average UK consumer exhibits a slight dislike of larger pack sizes, all
else equal, and the average Spanish consumer prefers a larger pack size. As antici-
pated, the typical consumer in each country dislikes higher prices.
18 Industry sources suggest this could be because consumers are particularly likely to "overdose" with con-
centrated products, using more product than is recommended per washload. This means that, per washload, these
formats appear relatively expensive to consumers.
Table 3 - Wald Test Statistics of Differences in Coefficient Estimates across Countries
Cross-Country Differences
UK-Germany UK-Italy UK-Spain Germany-Italy Germany-Spain Italy-Spain
Ariel 33.83 51.28 941 6'62 13.92 31.32
Surf - 12.89 11.63 - - 1.05
Liquid 0.12 5.63 28.90 15.40 57.26 7.01
Tablet 12.25 23.80 40.25 15.40 6.24 3.61
Concentrated format 13.82 0.53 12.86 12.68 34.84 15.78
Pack size 24.32 4.80 12.84 13.93 0.47 17.48
Price 6.14 4.33 3.36 16.43 0.94 19.23
Constant term 17.81 16.51 7.52 0.47 0.26 0.00
Note: Evaluated against an F distribution, where the critical values for F(l, 1000) are:
10% 2.71
5% 3.85
2.50% 5.04
1% 6.66
In the full model specification, including the random coefficients, which fits the
data better than the version of the model which excludes these coefficients, aw is
estimated to be -0.002 with a standard error of 0.04, suggesting that there is limited
random variation in pack size preference across households in each country. The
estimated ap coefficient representing the effect of the interaction of an unobserved
household attribute and product price is 12.889 and highly significant (the stan-
dard error is 3.14). This implies that there is significant variation in price sensitivity
within a country across households.
Wald test statistics for the significance of the differences in the estimated rela-
tive preferences across the countries are given in Table 3. Ariel was sold in all four
countries in the data, and Surf was sold in the United Kingdom, Italy, and Spain over
the three years in question. The estimates in Table 2 and Table 3 reveal that Ariel
is significantly more highly valued in the United Kingdom than in Germany, Italy,
and Spain; significantly more highly valued in Spain than in Germany and Italy;
and significantly more highly valued in Germany than in Italy, in each case relative
to a local non-MNC brand. Surf is significantly more highly valued in the United
Kingdom than in Italy and Spain, but the null hypothesis of no significant difference
in relative Surf brand value between Italy and Spain cannot be rejected.
Turning to relative format preferences, there is often a significant difference
in the relative preference of the average consumer for various formats over stan-
dard powder across all pairs of countries. While consumers in all countries have a
negative preference for concentrated formats, on a per washload basis, consumers
in the United Kingdom, Italy, and Spain have, on average, a significantly greater
dislike for concentrated formats than do consumers in Germany. Spain also has a
stronger negative preference for concentrated formats than does each other country.
All country-pairs other than Germany and Spain exhibit a significant difference in
their average preferences for larger pack sizes. Similarly, all country-pairs other
than Germany and Spain are significantly different in the average consumer's price
sensitivity. German consumers are most price sensitive on average, and Italian con-
sumers are the least price sensitive on average. Last, the Wald test statistics for
significant difference in the constant term reveal that consumers on average in the
United Kingdom have a greater relative preference for products in one of the top 20
brands that are not MNC brands. This is consistent with UK consumers attaching a
relatively high value to domestic supermarket own-brands.
As in Nevo (2000), the own price elasticity for goody allows for more flexibility
in substitution patterns than would a logit demand system.19 Similarly, the own size
elasticity for goody at time t within a given country - how sensitive product demand
is to pack size - is given by:
, v 5sit wit wit 1 v^
where 7, = 7C + o„viX is the household-level size elasticity. Own-size elasticities

are of particular interest in this study, since the counterfactuals involve changing
individual product-level pack size within country. Furthermore, the full model allows
the change in market share for product j in response to a change in pack size of any
other product - the cross-size elasticity - to be a function of household preferences
for the product-level characteristics of each product. The demand system is, hence,
flexible enough to allow households to be more likely to switch to similar brand-
format-size products in response to a given pack size change.20
The results presented in this section attach parameter values to local consumer
preferences and reveal that there is significant variation in preferences across coun-
tries for observable product characteristics. Together with the actions of local and
multinational competitor and retailing firms in each country, consumer prefer-
ences determine the local conditions faced by each multinational firm in each of its
national markets. Variation in local conditions, across markets and within markets
over time, suggests there are significant incentives for a multinational firm to sell
a locally tailored product range and to organize its activities so as to be informed
about local conditions. The estimated parameter values also permit an evaluation of
heterogeneous predicted changes in product-level demand following changes to the
composition of the product range choices available in each market.
V. Increasing Product Range Coordination across Countries
A. Motivation
Table 1 shows that the multinational brands market a range of local and standard-
ized products in each market. For example, in the United Kingdom, there are 66
products sold in a typical time period under one of the three multinational brands
available in the United Kingdom, but only 14 of these varieties are sold in another
^Product-level price elasticties reflect household-level price elasticities, given by a¡ equal to ac + opvi2.
Of the a¡ coefficients estimated for each household i in each country in this study, only 1 percent, 3 percent, 7 per-
cent, and 2 percent in the United Kingdom, Germany, Italy, and Spain, respectively, are positive. This suggests the
model does a relatively good job of measuring household-level dislike of high prices.
20 The Web Appendix contains more detailed information on estimated own price, own size, cross price and,
cross-size elasticities.
294 AMERICAN ECONOMIC JOURNAL: MICROECONOMICS FEBRUARY 201 1
market during the same time period. This section analyzes the relative profitability
at the firm level of small changes in Ariel and Surf's product ranges in the direction
of increased product-range standardization across countries. Jbc denotes the product
range on sale in April and June of 2003 for each brand b in each country c. Taking
each brand-country pair in turn, same-brand, same-format products on sale in
another country were found for ally G Jb¿. This allowed the construction of alterna-
tive product ranges for brand b in country c' in each alternative, one product y E JbiC
was replaced with a close substitute product, denoted/, that was already being sold
in another market. An alternative product range is denoted Ab c?x
Increased product-range standardization increases firm profits if selling Abc rather
than Jbc reduces total costs by more than it decreases revenues. Following the nota-
tion in Section I, the firm could profitably increase product range standardization
across countries for brand b if:
(7) E Puqu - E mcuQ* > E Puqu - E тс„<И* ~ Fj-

J€Ab¿ J€Ab,c j€h,c №ь,с
As shown in Section I, division manager agency means inequality (7) can hold even
when the manager has found it optimal to choose the product range Jbc. Because
the transfer price in place affects the observed product range, Jb c, assumed optimal
from the point of view of the manager, the transfer pricing mechanism does not play
a role in expression (7). Production costs are relevant pieces of information, but the
intra-firm transfer prices, which influence local manager decisions, are irrelevant
since they net out of firm-level profit.
The analysis will rely on the assumption that competitor firms do not change their
own product ranges in response to a product switch for the focal. This assumption is
made more reasonable by the fact that the changes in product characteristics under
each substitution are small. The counterfactual design outlined in the following
subsection also takes any constraints imposed by the local retail environment into
account since the number of products and approximate shelf space requirements are
held constant for each brand under each substitution.22
B. Counterfactual Design
Several intermediate steps are required to establish whether expression (7) holds
for each observed product range, Jb c, and the alternative product ranges Ab c, using
the available data. Each step is outlined here:
First, a new product range means the consumer choice set contains a different
set of product characteristics. Using the demand system estimated in Section IV,
21 This is only one of many ways to define an alternative product range. The counterfactual design here involves
making only one substitution at a time. Alternative specifications could switch pairs of products in and out, or make
larger changes to the observed product range.
22 One additional assumption required is that firms entail no switching costs in making each substitution.
Looking back at the data for the entire three year period, it can be seen that firms often switch products over time
within each market. The single product substitutions simulated here constitute relatively small changes in product
range when compared to the usual changes between two adjacent eight-week periods in the data.
the quantity demanded of each product can be predicted and then summed over all
brand products in the alternative portfolio for any given vector of product prices. The
demand system contains a brand-format-size deviation from mean product demand
specific to that period that is observed by the firm and not by the researcher, denoted
Ä£/,c in equation (5). When estimating demand under each alternative product range,
it is assumed that the unobserved product shock is specific to the brand and format
and not the pack size, so that A£/,c = Д£/>- This assumption is reasonable if demand
shocks reflect local advertising campaigns, which are rarely specific to a product pack
size and are more often tied to the product characteristics of the brand and format. The
analysis does not require any assumption to be made about the relative magnitudes of
unobserved product shocks for the same product sold in two different countries.
The validity of the assumption that Д£/,с = Д£/,с is likely to have a limited effect
on the findings, since the estimated Д£/,с for products that are sold in each market
are very small. Moving from the tenth to the ninetieth percentile of the distribution
of Д£/,с in each country, across all products and time periods, leads to an increase in
utility that is equivalent to a change in pack size of no more than one thousandth of
a washload in each country. Moreover, regressing these error terms on a full set of
brand-time period and format-time period indicator variables, explains 33 percent
of the variation in Д£/,с in the United Kingdom, 27 percent in Germany, 32 percent
in Italy, and 20 percent in Spain. That is, the error terms are indeed relatively similar
for products of the same brand and format in any time period.
Second, a change in product range may lead division managers to alter product-
level prices. Nonetheless, the prices currently observed remain within the firm's
choice set.23 Holding competitor prices constant, the firm conducting the switch can
generate brand revenues that are at least as high with newly optimized prices as it
can with the current product-level prices, with product/ being priced at the same
per- washload price as the product taken out,/ This implies that:
E Puqu - E Puqu > E Puqu - E Puqu,

jeAbiC jeJb,c jzAbj jeJb,c
where p]%c denotes the the new optimal price for each product j in the alternative
product range and pj,c is the price currently observed for product/ For expression
(7) to hold, it is thus sufficient that the following condition holds:
(8) E Puqu - E Puqu > E ™Cj,cqu - E тс^У* ~ Fj-

jeAb¿ jeJb,c j€AbtC j£h,c
Third, turning to the cost side, the product-level fixed cost of the standardized
product/ does not appear in expression (8) since it is a sunk cost. The product-level
23 Since price setting is decentralized and division managers observe transfer prices rather than marginal costs
when setting price, using observed prices to estimate marginal costs is problematic in this setting. It would require
assuming prices are set to maximize firm profits in order to analyze whether or not product range choices are profit
maximizing under similarly decentralized decision-making.
fixed cost Fj of the local product,/ does appear and is unobserved. However, for any
positive fixed cost, expression (8) will hold whenever:
(9) E Puqu - E Puqu > E **&№ ~ E тс^Чз*-

J€Ab,c J£Jb,c J€Ab,c J€Jb,c
By looking for counterfactuals in which inequality (8) is satisfied in the absence of

any product-level fixed cost, it is certain that the inequality will hold in the presence
of a positive fixed cost for product/ The two intermediate steps, between expres-
sions (7) and (9), have made it more difficult to establish that expression (7) holds
for any given alternative product range.
Finally, rather than deriving estimates of product-level marginal costs, it is
assumed here that the average marginal cost for the demand in the old and new port-
folios is the same. There are two possible objections to this assumption. First, the
product coming into the market may have a different marginal cost than the product
going out. This presents a concern for the test design only if the new product's actual
marginal cost is higher than the marginal cost of the product leaving the market
since this would lead to the overestimation of variable profits under the alternative
product range. The two products are, however, manufactured in the same produc-
tion facility and identically produced in each case right up to the point where they
are packaged in different sized containers. It seems likely that the washload-level
marginal costs of these products will be similar. The second reason is that the real-
location of consumer demand across products in the new product range may shift
demand towards higher or lower marginal cost products. Under the counterfactual
design used, though, the product substitutions are close enough that the realloca-
tion of demand among other brand products prompted by the substitution is small.
For example, in the United Kingdom, each product substitution changes the quan-
tity demanded of each other product by a maximum of 5 percent and most of the
changes are less than 1 percent.24
If it is assumed that average marginal cost is equal for each pair of product range
choices, then inequality (9) can be rewritten as:
(10) E Puqic - E РмЧи > mei E <ü* ~ E «A

J€Ab,c J€Jb,c 'j€Ab,c J€Jb,c I
In sum, expression (7) will hold whenever expression (10) holds, under the assump-
tions that average marginal costs are the same for Jbc and Ab c, and that the unob-
served shock to demand for product j is equal to the unobserved demand shock for
product/ in country c.
24 Thomas (2006) overlooks the implications of intra-firm transfer pricing and uses observed prices and the
demand system to back out estimates of marginal costs for each product. While these are more correctly termed
estimates of the transfer prices seen by the local division manager, these product level cost estimates are relatively
similar across countries for the same product when it is sold in more than one country. This offers some support for
the assumption that product-level marginal costs are not significantly different for similar products.
С. Implementation in the Data
If revenues increase and quantities fall under the new product range, it is clear
that the left-hand side of condition (10) is positive and the right-hand side is nega-
tive, meaning condition (10) is satisfied. If, on the other hand, quantities increase
under the new product range, rearranging condition (10) shows that it is satisfied
whenever the change in total revenue, divided by the change in quantity, exceeds
marginal cost. Although marginal costs are unobserved, it is reasonable to assume in
this differentiated products industry that the observed average price exceeds average
marginal cost for the brand in country с in each time period. In this case, the change
in total revenue when moving to the alternative product range divided by the change
in quantity will exceed marginal cost whenever it exceeds the currently observed
average price. Hence, if quantities rise under the alternative product range Abc then
expression (10) is satisfied whenever ATR/AQ > ~pjbc, where ~pJbc is the average
price for the brand with the current product range Jb %c.
The final step in the counterfactual exercise is to take the alternative product
ranges under which condition (10) cannot be rejected and ask, in each case, whether
the firm can stop making the product removed from the relevant market altogether.
This is assumed to be possible if the products switched out of country с do not appear
on sale in any of the other countries in the dataset. When a product is no longer pres-
ent in the dataset overall, the firm can cease production of the product. Since this
test was conducted setting the fixed cost of product,/ to zero, if the removed product
incurred any positive fixed cost of production, F, > 0, the firm is made strictly bet-
ter off by making the substitution to increase the degree of standardization across
countries, and reducing the total number of product varieties manufactured in the
region, since it no longer incurs these per-period fixed costs.
Table 4 contains the total revenue, quantity, and average price currently observed
for each brand in each market. The mean, median, standard deviation, and 95 percent
confidence interval constructed are given in the last four columns of the table.25'26
Confidence intervals for each estimate were found using a Cholesky decomposition
of the variance-covariance matrix for the demand system estimated in Section IV.
One thousand draws were taken from the joint distribution of the 9! and 92 parame-
ter estimates pertaining to pack size and format in the relevant market and the brand
revenues, quantities, and average price were found for each draw.27
25 The magnitude of the estimates given in Table 4 reflects the sample size of households in the product level
data rather than the total number of households in each country, but could be scaled up appropriately without chang-
ing the results.
26 The confidence intervals estimated for the Ariel brand revenues, quantities, and average price in Germany
are relatively large. This makes it difficult to infer anything about changes in these estimates under any alternative
portfolio.
27 An alternative estimate of the confidence intervals for each estimate would be to draw from a parametric boot-
strap based on the distribution of all estimated parameters, including those associated with product characteristics
that do not vary with the product substitution such as brand and price. This generates larger confidence intervals.
Hence, the conclusions derived under the confidence intervals described here - that product substitutions do not
significantly decrease variable profit - will continue to hold with the alternative approach.
Table A - Revenues, Quantities, and Average Prices for Each Brand-Country under the Current
Product Range
Size of
confidence
Actual Mean Median SE interval
Surf in the UK Revenues, euros 14,340 13,812 13,908 ЦШ 3,500

Quantities, number of washloads 90,621 87,165 88,397 9,155 29,240
Average price per washload, euros 0.16 0.16 0.16 0.01 0.02
Surf in Italy Revenues, euros 7,275 7,321 7,277 728 2,365

Surf in Spain Revenues, euros í,108 1,030 1,045 104 350

Quantities, number of washloads 6,477 6,001 6,084 611 2,072
Ariel in the UK Revenues, euros 23,041 23,721 23,073 2,089 5,313

Ariel in Italy Revenues, euros 2,355 2,246 2,340 410 1,312

Ariel in Spain Revenues, euros 9,109 9,888 9,236 2,343 6,977

Ariel in Revenues, euros 8,952 11,007 8,880 6,759 21,066

Germany Quantities, number of washloads 90,405 137,806 89,823 122,487 368,235
Notes: This table presents estimates of the brand-country market outcomes under the product range observed in the
market. The estimates reflect the sample size of households in the product-level data rather than the total size of the
country, but could be scaled up to the country level.
D. Results
The products substituted out of each market с and the products brought in are
described in Table 5 (panel A for Surf and panel В for Ariel), noting that some products
were switched out of the market in more than one counterfactual. Each substitution is
done in turn, with replacement, and the new market demand estimated using the frame-
work described in Section IV. For each substitution, standard errors and confidence
intervals were computed around the new revenue, quantity, and average price estimates
as described in Section VC. Columns 3 and 4 of each panel show the changes in pack
size associated with each substitution. To qualify as a product to be switched into mar-
ket c, the product had to be absent from market с and differ only slightly from a product
being sold in market c. In cases where a product is a format sold only in one of the four
countries, for example, liquid soap Marsiglia in Italy, and gels and pastes in Germany,
no counterfactual substitutions were made. Tables 5 and 6 show that for each brand in
each country, product substitutions involved both increases and decreases in pack size.
The simulated effects of the product substitutions are summarized in Table 6 (and
shown in more detail in Web Appendix Table 6). Starting with Surf in the United
Kingdom, the total number of washloads of Surf product sold with each new product
VOL 3 NO. I THOMAS: TOO MANY PRODUCTS 299
Table 5 - Products to Be Introduced to Each Market, and the Products They Replace
Product appears
Pack size Pack size Source in another
Country Format taken out put in country country
Panel A. Surf products
UK Standard powder 9 5 Italy Y
Standard powder 12 8 Spain Y
Standard powder 15 23 Italy Y
Standard powder 31 41 Spain N
Standard powder 52 60 Italy N
Tablets 11 16 Italy N
Tablets 11 18 Spain N
Tablets 22 37 Spain Y
Tablets 36 26 Italy N
Italy Standard powder 5 9 UK N

Standard powder 12 15 UK Y
Standard powder 60 52 UK N
Tablets 16 14 UK Y
Tablets 16 11 UK Y
Tablets 26 22 UK N
Tablets 26 37 Spain Y
Spain Standard powder 8 12 UK Y

Tablets 18 11 UK Y
Tablets 18 26 Italy Y
{continued)
range increases slightly in some cases and decreases slightly in some cases. In cases
where quantities fall, revenues also tend to fall although, again, not significantly. When
quantities rise, the change in total revenues over the change in quantity increases in the
case of 3 of the 1 1 products, although the change in average price is not significant in
any case. Condition (10) cannot be rejected for any of the alternative product ranges.
This analysis suggests that there are substitutions which allow the firm to remove Surf
products from the UK market. The last column in panel A of Table 5 indicates whether
a switch satisfying expression (10) means that the product y is no longer present in any
market. This is the case for the 31- and 52-washload packs of standard powder and for
the 11- and 36-washload packs of tablets. If these products were no longer sold in the
United Kingdom, Unilever would cease to manufacture them, thereby saving the prod-
uct-level fixed costs, and be strictly better off. Further analysis reveals that Unilever
could remove all four of these products simultaneously and replace them with exist-
ing alternative products without significantly reducing variable profits. This implies
Unilever is manufacturing too many varieties of Surf detergent to sell in the UK market.
Table 6 reveals that substitutions of Surf products in Italy also satisfy condition
(10). Of the 10 Surf products switched out of the Italian market in turn, the last column
Table 5 - Products to Be Introduced to Each Market, and the Products They Replace (Continued)
Product appears
Pack size Pack size Source in another
Country Format taken out put in country country
Panel B. Ariel products
UK Standard powder 9 5 Italy Y
Concentrated powder 52 112 Spain Y
Standard liquid 12 44 Italy Y
Standard liquid 12 20 Spain Y
Concentrated liquid 30 35 Germany Y
Concentrated liquid 76 58 Germany Y
Tablets 17 15 Germany Y
Italy Standard powder 5 9 UK Y

Standard powder 12 16 UK N
Standard liquid 12 9 UK Y
Standard liquid 12 20 Spain Y
Standard liquid 44 61 Spain N
Spain Standard powder 4 5 Italy N

Standard liquid 11 9 UK Y
Standard liquid 20 12 UK N
Standard liquid 20 12 Italy N
Standard liquid 61 44 Italy N
Tablets 18 11 UK Y
Tablets 37 48 UK N
Tablets 37 22 Germany N
Germany Tablets 15 11 UK N
Tablets 22 17 UK N
Tablets 22 48 UK N
Notes: The final column notes whether the product taken out is still present in another market. If not, the firm can
reduce manufacturing costs by no longer manufacturing the product.
of panel A in Table 5 shows that 3 are now redundant, in that they do not appear else-
where in the dataset. The products that can profitably, and simultaneously, be removed
from the Italian market are the 5-washload pack of standard powder, the 60-washload
pack of standard powder, and the 26-washload pack of tablets. There were eight sub-
stitutions conducted for Surf in Spain for the five products marketed under this brand
at this time. Once more, condition (10) cannot be rejected for each substitution, in
that none of the switches led to significant changes in variable brand profit. Of the five
products taken out in turn, one does not appear elsewhere in the dataset. This is the
37-washload pack of tablets. These findings suggest that Unilever is also manufactur-
ing too many varieties of Surf detergent to sell in Italy and Spain.
A similar finding emerges for the P&G brand Ariel in the United Kingdom, Italy,
and Spain. As shown in panel В of Table 5, seven Ariel products currently on sale in the
United Kingdom are switched out of the market in at least one product substitution. For
Table 6 - Summary of Changes in Revenues, Quantities, and Average Prices by Brand and Country
under Alternative Product Ranges
Surf in Surf in Surf in Ariel in Ariel in Ariel in Ariel in

UK Italy Spain UK Spain Italy Germany
Number of products in current product range 12 12 5 21 13 5 13
Number of individual product-level 11 10 8 10 12 6 6
substitutions made
Fraction of substitutions where pack size 0.55 0.60 0.50 0.50 0.42 0.67 0.50
increases
Number of substitutions where revenues 5 4 4 5 5 3 3
increase
Number of substitutions where quantities 5 4 2 5 5 2 3
increase
Number of substitutions where average price 4 3 3 7 8 2 3
increases
Number of substitutions where revenues 0 0 0 0 0 0 0
increase and quantities fall
Number of substitutions where quantities 3 3 2 4 5 2 3
increase and the ratio of the change in
revenues and the change in quantities exceeds
currrent average price
Number of substitutions where alternative does 11 10 8 10 12 6 6
not generate significantly lower variable profit
where alternative does not make sig less var
profit. That is, condition (1 1) is not violated.
Notes: For each brand in each country, this table summarizes the number of alternative product ranges which lead
to increases in revenues, quantities, and average prices in a typical eight- week period. More details of the results of
the substitutions are given in Web Appendix Table 5.
each of these switches, the alternative product range does not lead to significant changes
in brand quantities, revenues, or average prices. Table 5 indicates that two of the seven
products, once removed from the UK market, are no longer present in the dataset. The
firm can stop manufacturing the 52-washload pack of Ariel standard powder and the
48-washload pack of Ariel tablets, simultaneously replacing these products in the UK
market with other already existing products, and be strictly better off.
Relatively few Ariel products are sold in Italy, but close substitute products were
found for all products other than the liquid soap Marsiglia product. Of the four
products for which substitutions were found that did not reject condition (10), two
products were unique to the Italian market. These were the 12-washload pack of
standard powder, and the 44-washload pack of standard liquid. Of the nine Ariel
products switched out of Spain, four products were redundant in that they did not
appear elsewhere. These were the 4-washload pack of standard powder, the 20- and
61-washload pack of standard liquid, and the 37-washload pack of tablets.
Turning to Ariel in Germany, the test design here cannot provide any evidence that
the product range is too localized. Many German products are unique formats which
are relatively highly valued in that market. For example, Table 2 shows that gel for-
mats are significantly more valuable to the average German consumer than standard
powder. This means that no close substitute product from elsewhere could be found
to play the role of the alternative product/. Substitutions are conducted for the two
products where close substitutes could be found, the 15- and 22-washload packs of
tablets. Since the estimates for Ariel revenues, quantities, and average prices were
not precisely estimated in Table 4, asserting that product substitutions did not violate
condition (10) does not permit inferences to be made about the relative merits of alter-
native product ranges. However, the two products switched out of the German market
in these substitutions, still appear in the data in other markets. The 15-washload pack
of tablets is sold in Spain, and the 22-washload pack of tablets is sold in the United
Kingdom. Even if these products were removed from the German market, the firm
would continue to manufacture them and incur the product-level fixed costs.
In sum, for each case other than Ariel in Germany, a subset of products are entirely
absent in the data once they are removed from the relevant country-level product
range. Of the eight Surf products removed from each market in turn, seven products
can be eliminated from the region-level product range simultaneously. The 26-pack of
Surf tablets removed from the Italian market was used as a substitute product for the
36-tablet pack in United Kingdom (which was in turn also used a substitute product in
Spain), hence, we cannot conclude Unilever should cease production of the 36 pack in
the United Kingdom at the same time as making other reductions in the overall prod-
uct range. Similarly, for Ariel, of the eight products that could be removed from each
country in turn, a total of six could be simultaneously taken out of production. (We
cannot infer that P&G could simultaneously remove a liquid product from both Italy
and Spain, or a tablet product from both the United Kingdom and Spain.) The total
number of distinct products for Surf and Ariel across the United Kingdom, Italy, and
Spain is 23 and 31, respectively, during this time period. Hence, the analysis suggests
up to 30 percent of the total number of Surf product varieties, and up to 20 percent
of the total number of Ariel product varieties manufactured in the region, could be
removed without significantly reducing variable profits. Saving the fixed production
costs on each, or any, of these products, means each firm could be made strictly better
off by choosing more standardized product ranges in the region.
E. Discussion
One potential objection to the inferences made in this analysis is that the model
of consumer demand in each country is not estimated precisely enough to reject con-
dition (10) under any product substitution.28 To address this concern, product sub-
stitutions were conducted switching in products sold elsewhere that were not close
substitutes for the products leaving the market. Despite the fact that pair-level switches
involve only small changes to the overall product range in each country, switching in
a very different product did lead to significant changes in brand revenues, quantities,
and average prices that meant condition (10) was rejected in some cases.29 Another
28 The results in the paper rest on the credibility of the estimated demand system. Although the counterfactual
analysis involves only marginal changes in the product range available in any one country, the assessment of change in
revenues, quantities, and average prices with each alternative product range depends on how well the demand system
measures consumer reaction to this change. In addition, the demand system does not permit an evaluation of how con-
sumers in each country would react to a maior market innovation, such as a new brand or a new format introduction.
29 For example in the United Kingdom, Surf revenues, quantities, and average price with the observed product
range are: 14,340 euros, 90,621 washloads, and 0.1582 euros, respectively. Using the standard errors given in Table
4, two standard errors on either side of these point estimates give the intervals (12,175-16,505), (72,312-108,931),
and (0.147-0.169). A substitution which takes out the 36-washload pack of tablets and introduces a 5-washload
pack of standard powder gives the following point estimates, with two standard errors on either side given in
parentheses: 8,312 (6,769-9,856), 44,068 (35,955-52,181), and 0.189 (0.1845-0.194). This product substitution
set of counterfactuals was constructed where the entire product range of each brand, in
each country, was replaced with the entire brand product range for another country, at
the prices set in the source country. Under these counterfactuals, the model produced
estimates of revenues, quantities, and average prices that were more than two standard
errors away from the estimates under the observed product ranges.30 This provides
some reassurance that, other than for the Ariel brand in Germany, the demand system
estimated in Section IV captures local tastes and preferences sufficiently well to assess
the impact of the constructed counterfactuals on product-level demand.
Putting to one side the assumed structure of preferences modeled in the paper,
another possible concern is that a local brand manager does not choose a standard-
ized pack size - because this product substitution would entail marketing a product
which would either compete directly with a competitor brand, or leave the brand
without a pack in a very popular size. The set of substitutions for each brand in
each country include some that remove a pack size where there is little competi-
tion, and some that introduce a pack size where there is direct competition, and vice
versa. Both cases generate non-decreasing variable profit substitutions. The pack
sizes introduced have all been recently present in the relevant country. It is, hence,
unlikely that specific retailer preferences, or competitor activities, explain the exact
composition of the pack sizes currently offered.
If there were variation in organizational form among firms in this industry, it
would be possible to examine whether the observed relative coordination and adapta-
tion losses did indeed differ with the choice of organizational form. All of the firms,
though, tend to be broadly similar in that they decentralize product range choice to the
country level. The two firms studied in this paper do differ in their corporate histories
in the region. P&G is widely recognized as a more regionalized firm, having originally
entered Europe with a region-level strategy, and having a longer history of centraliza-
tion of certain functions such as R&D. In contrast, Unilever has expanded mostly
through acquisition, and is regarded as more localized. These corporate legacies sug-
gest there might be cross-firm differences in the extent of cross-country coordination.
While the counterfactuals conducted are not exhaustive, in that an alternative product
range was constructed only if a very similar product could be found elsewhere, the
findings do reveal some interesting cross-brand differences in excessive localization in
each country. For example, in the United Kingdom, the analysis suggests that one third
of local Surf products, 4 out of 12, could be profitably replaced while only 10 percent
of Ariel products, 2 out of 21, are redundant if profitable substitutions are made. In
Italy and Spain, the relative magnitude of the within-brand product redundancy is
leads to confidence intervals that do not overlap with the original intervals. Revenues and quantities both fall and
the average price rises, meaning that condition (10) does not hold for this substitution.
JUFor example, replacing the United Kingdom s Anel product range with the Anel product range in uermany,
and assuming that UK customers value gel and paste formats in the same way they value capsules, leads to estimates
of revenues and quantities sold in the UK that are significantly lower than the estimates revenues and quantities
with the current UK product range. Intervals constructed around each point estimate based on two standard errors
on either side are nonoverlapping for revenues and quantities. The average price is not significantly different.
Condition (10) does not hold for this alternative product range. In another example, replacing the Ariel product
range in Italy with the product range from the United Kingdom leads to significantly higher revenues, quantities,
and average prices in Italy. However, in this case, the number of Ariel products on sale in Italy increases from 5 to
21. Although condition (10) holds here, this substitution would entail a large reconfiguration of retail space, and
possibly other unobserved costs.
reversed. 40 percent of Ariel products in Italy, and 31 percent of Ariel products in

Spain can be replaced, while only 25 percent and 20 percent of Surf products can be
replaced in Italy and Spain, respectively. Overall, the Surf brand appears to be particu-
larly localized, in that a larger share of the total number of product varieties manufac-
tured in the region could be profitably eliminated.
VI. Conclusion
The analysis in this paper provides evidence that decentralization results in a lack of
coordination across divisions.31 Within the same detergent brand, firms manufacture
product varieties that are sold in only a subset of markets, and in many cases, in just
one national market. This paper attributes product-range variation in excess of the
optimal variation given estimated local conditions to the firm's organizational form.
Richard E. Caves (1996) points out that horizontal MNCs - firms selling similar
products in different markets - face significant challenges in organizing production,
distribution, and marketing across countries. The choice of which product varieties to
sell in each market is one important decision for the firm. Differing local conditions
determine the relative value of being multinational. If all countries required entirely
different products, there would be no advantage of economies of scale in production.
A multinational firm in such an environment might effectively operate as a collection
of independent national subsidiaries. At the other extreme, if all consumers had identi-
cal tastes across countries, and competitors and retailers also behaved identically, there
would be no incentive for local adaptation and no particular decision-making role for
local division managers. The environment faced by the two firms studied here falls
somewhere between these two extremes, and the organizational challenges this setting
presents are fairly generalizable to horizontal multinationals in other industries.
To test the prediction of excessive product range localization, it was first necessary
to allow for diverse local conditions. Most existing empirical studies about horizon-
tal multinational firms treat consumers as identical across countries. This assumption
greatly increases the tractability of these complex models, but also removes the key
motivation for decentralization - a widely observed characteristic of these firms. In
this paper, product-level data about the detergents industry allows variation in local
conditions to be parameterized using a discrete choice random coefficients demand
model. Consumer preferences for detergent characteristics are found to vary signifi-
cantly across national markets. It is also likely that the interaction between detergent
manufacturers and retailers differs across countries. To relate the firm's organizational
problem to models of the trade-off between coordination and adaptation, it is nec-
essary that local conditions are potentially different across divisions and centrally
unobservable.32 Decentralized decision making is the observed organizational form,
31 The theory models which make this prediction demonstrate how decentralization may, nonetheless, be the
preferred organizational form since centralization would incur losses from a lack of adaptation that would harm
firm profits to a greater extent. The case of Jacobs Suchard described in Dessein, Galicano, and Gertner (2010) pro-
vides one example in which an attempt to centralize decision making to achieve greater coordination incurred costs
associated with low quality communication in a situation that appears fairly analogous to the detergents industry
setting described here.
In the model presented in Alonso, Dessein, and Matouschek (2008), the two divisions' local conditions are
equal in expectation since they are both drawn from mean-zero uniform distributions.
and it seems reasonable in this industry to assume that local brand-country managers
learn or gather information about the precise, time-specific, local conditions.
Some tailoring of product range would hence appear to be optimal, but to what
extent - and how can firms figure this out? When information about local conditions
is distributed among division managers, theory predicts that agency costs will arise.
The second part of the analysis in this paper finds empirical evidence consistent
with agency in product-range choice due to the decentralization of this decision
in the United Kingdom, Italy, and Spain for the P&G brand Ariel and the Unilever
brand Surf. Controlling for variation in local demand conditions and in the local
retail environment, the product ranges marketed for the two brands are too localized.
By making small increases in the degree of product range standardization across
countries, both firms could reduce the total number of products being manufactured,
and increase firm-level profits. The results suggest that each firm is manufacturing
too many products overall. The empirical findings suggest that multinational firms
in consumer goods industries operate at a constrained optimal outcome, given the
inherent challenges of organizing firm production across markets with diverse local
conditions. The results hence demonstrate that firm structure affects product range
choice and firm performance.
REFERENCES
Abernethy, Margaret A., Jan Bouwens, and Laurence van Lent. 2004. "Determinants of Control Sys-
tem Design in Divisionalized Firms." The Accounting Review, 79(3): 545-70.
Acemoglu, Daron, Philippe Aghion, Claire Lelarge, John Van Reenen, and Fabrizio Zilibotti. 2007.
"Technology, Information, and the Decentralization of the Firm." Quarterly Journal of Economics,
122(4): 1759-99.
Aghion, Philippe, and Jean Tiróle. 1997. "Formal and Real Authority in Organizations." Journal of
Political Economy, 105(1): 1-29.
Alonso, Ricardo, Wouter Dessein, and Niko Matouschek. 2008. "When Does Coordination Require
Centralization?" American Economic Review, 98(1): 145-79.
Arkolakis, Costas, and Marc-Andreas M uendler. 2009. "The Extensive Margin of Exporting Goods: A
Firm-Level Analysis." http://www.econ.yale.edu/~ka265/research/ Arkolakis_Muendler_products_
DecO7.pdf.
Baiman, Stanley, David F. Larcker, and Madhav V. Rajan. 1995. "Organizational Design for Business
Units." Journal of Accounting Research, 33(2): 205-29.
Baker, George, Robert Gibbons, and Kevin J. Murphy. 1999. "Informal Authority in Organizations."
Journal of Law, Economics, and Organization, 15(1): 56-73.
Bernard, Andrew В., Stephen J. Redding, and Peter K. Schott. 2010. "Multi-Product Firms and Prod-
uct Switching." American Economic Review, 100(1): 70-97.
Berry, Steven, James Levinsohn, and Ariel Pakes. 1995. "Automobile Prices in Market Equilibrium."
Econometrica, 63(4): 841-90.
Bloom, Nicholas, Luis Garicano, Raffaella Sadun, and John Van Reenen. 2009. "The Distinct Effects
of Information Technology and Communication Technology on Firm Organization." National
Bureau of Economic Research Working Paper 14975.
Bouwens, Jan, and Laurence van Lent. 2007. "Assessing the Performance of Business Unit Manag-
ers." Journal of Accounting Research, 45(4): 667-97.
Brambilla, Irene. 2007. "A Customs Union with Multinational Firms: The Automobile Market in
Argentina and Brazil." http://www.econ.yale.edii/~ib55/papers/cars_mercosur.pdf.
Broda, Christian, David E. Weinstein. 2008. "Understanding International Price Differences Using
Barcode Data." National Bureau of Economic Research Working Paper 14107.
Bushman, Robert M., Raffi J. Indjejikian, and Abbie Smith. 1995. "Aggregate Performance Measures
in Business Unit Manager Compensation: The Role of Intrafirm Interdependences." Journal of
Accounting Research, 33: 101-28.
Caroli, Eve, and John Van Reenen. 2001. "Skill-Biased Organizational Change? Evidence from a
Panel of British and French Establishments." Quarterly Journal of Economics, 1 16(4): 1449-92.
Caves, Richard E. 1996. Multinational Enterprise and Economic Analysis. 2nd ed. xi, 346. Cambridge,
UK: Cambridge University Press.
Colombo, Massimo G., and Marco Delmastro. 2004. "Delegation of Authority in Business Organiza-
tions: An Empirical Test." Journal of Industrial Economics, 52(1): 53-80.
Cristol, Steven M., and Peter Sealey. 2000. Simplicity Marketing: End Brand Complexity, Clutter, and
Confusion. New York, NY: The Free Press.
Dessein, Wouter, Luis Garicano, and Robert Gertner. 2010. "Organizing for Synergies." American
Economic Journal: Microeconomics, 2(4): 77-1 14.
Draganska, Michaela, and Dipak C. Jain. 2005. "Product-Line Length as a Competitive Tool." Journal
of Economics and Management Strategy, 14(1): 1-28.
Draganska, Michaela, and Dipak C. Jain. 2006. "Consumer Preferences and Product-Line Pricing
Strategies: An Empirical Analysis." Marketing Science, 25(2): 164-74.
Drayer, Ralph. 1999. "Procter & Gamble's Streamlined Logistics Initiatives." Supply Chain Manage-
ment Review. 3(2): 32-43.
Euromonitor. 2003a. "Household Care in Germany." Euromonitor Report.
Euromonitor. 2003b. "Household Care in Italy." Euromonitor Report.
Euromonitor. 2003c. "Household Care in Spain." Euromonitor Report.
Euromonitor. 2003d. "Household Care in the United Kingdom." Euromonitor Report.
Euromonitor 2003e. "The Procter and Gamble Company." Euromonitor Report.
Euromonitor. 2003f. "Unilever Group." Euromonitor Report.
European Commission Enterprise Directorate-General. 2001. "Competitiveness of the Chemi-
cal Industry Sector in the CEE Candidate Countries." European Commission. http://web.archive.
org/web/20040324 1521 24/http://europa.eu.int/comm/enterprise/chemicals/enlargement/study/
competsectors .pdf
Friebel, Guido, and Michael Raith. 2010. "Resource Allocation and Organizational Form." American
Economic Journal: Microeconomics, 2(2): 1-33.
Guadalupe, Maria, and Julie Wulf. 2010. "The Flattening Firm and Product Market Competition: The
Effect of Trade Liberalization on Corporate Hierarchies." American Economic Journal: Applied
Economics. 2C4Ì: 105-27.
Hart, Oliver, and John Moore. 2005. "On the Design of Hierarchies: Coordination versus Specializa-
tion." Journal of Political Economy, 1 13(4): 675-702.
Javorcik, Beata, Wolfgang Keller, and James Tybout. 2008. "Openness and Industrial Response in a
Wal-Mart World: A Case Study of Mexican Soaps, Detergents and Surfactant Producers." World
Economy, 31(12): 1558-80.
Jensen, Michael C, and William H. Meckling. 1976. "Theory of the Firm: Managerial Behavior,
Agency Costs and Ownership Structure." Journal of Financial Economics, 3(4): 305-60.
Jensen, Michael C, and William H. Meckling. 1995. "Specific and General Knowledge, and Organiza-
tional Structure." Journal of Applied Corporate Finance, 8(2): 4-18.
Kastl, Jakub, David Martimort, and Salvatore Piccolo. 2008. "Delegation and R&D Spending: Evi-
dence from Italv." Centre for Studies in Economics and Finance (CSEF) Working Paper 192.
McFadden, Daniel. 1974. "Conditional Logit Analysis of Qualitative Choice Behavior." In Frontiers of
Econometrics, ed. Paul Zarembka, 105-42. New York, NY: Academic Press.
Nevo, Aviv. 2000. "A Practitioner's Guide to Estimation of Random-Coefficients Logit Models of
Demand." Journal of Economics and Management Strategy, 9(4): 513-48.
Nevo, Aviv. 2001. "Measuring Market Power in the Ready-to-Eat Cereal Industry." Econometrica,
69(2): 307^2.
Pikorski, Mikolaj Jan, and Alessandro L. Spadini. 2007. "Procter & Gamble: Organization 2005 (A)"
Harvard Business Review Case 707519.
Rajan, Raghuram G., and Luigi Zingales. 2001. "The Firm as a Dedicated Hierarchy: A Theory of the
Origins and Growth of Firms." Quarterly Journal of Economics, 1 16(3): 805-51.
Rantakari, Heikki. 2008. "Governing Adaptation." Review of Economic Studies, 75(4): 1257-85.
Thomas, Catherine Atterbury. 2006. "Essays on the Industrial Organization of Multinational Enter-
prises." PhD diss. Harvard University.
x-rates.com. 2003. Tiago Stock Consulting, www.x-rates.com.

MCTM Teams

Uploaded by

Copyright:

Available Formats

MCTM Teams

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MCTM Teams

Uploaded by

Copyright:

Available Formats

American Economic Association

Too Many Products: Decentralized Decision Making in Multinational Firms

Too Many Products:

/ analyze country-level product ranges offered by multinational

decision making is a widely observed characteristic of large mul-

I. Agency in Product Range Choice

(1) П/=ЕЕЕ Puqu - E 'mCjE ЯМ ~ FA - Ff9

(2) 7Tb,c = E (PU - ku)qj,c.

(3) E Puqu - E Puqu > E he®* - E h^-

This inequality demonstrates that the locally-tailored product j will be chosen

(4) F] + E mcj1hc - E mcj4U > E киЯм - E kj,cVU.

II. The Laundry Detergents Industry in Western Europe

Laundry detergent, sometimes referred to as heavy-duty detergent, typically

Table 1 - Summary of MNC Activities in Each Country

UK Germany Italy Spain

Typical per period number of distinct products detailed in dataset* * * 1 80 1 34 1 54 135

Source: Europanel dataset.

remark that products marketed as environmentally friendly are "unpopular." As of

management described it, into a coherent 'multi-local-multinational' organization that

IV. Estimating Local Conditions

(5) W/Дс = SjAc(XjAc,WjAc,pj,t,c,A^c^l)

Table 2 - A Subset of Estimated 0l Coefhcients, Full Model with Instruments

Variable UK Germany Italy Spain

Table 3 - Wald Test Statistics of Differences in Coefficient Estimates across Countries

, v 5sit wit wit 1 v^

where 7, = 7C + o„viX is the household-level size elasticity. Own-size elasticities

V. Increasing Product Range Coordination across Countries

(7) E Puqu - E mcuQ* > E Puqu - E тс„<И* ~ Fj-

E Puqu - E Puqu > E Puqu - E Puqu,

(8) E Puqu - E Puqu > E ™Cj,cqu - E тс^У* ~ Fj-

(9) E Puqu - E Puqu > E **&№ ~ E тс^Чз*-

By looking for counterfactuals in which inequality (8) is satisfied in the absence of

(10) E Puqic - E РмЧи > mei E <ü* ~ E «A

С. Implementation in the Data

Surf in the UK Revenues, euros 14,340 13,812 13,908 ЦШ 3,500

Surf in Italy Revenues, euros 7,275 7,321 7,277 728 2,365

Surf in Spain Revenues, euros í,108 1,030 1,045 104 350

Ariel in the UK Revenues, euros 23,041 23,721 23,073 2,089 5,313

Ariel in Italy Revenues, euros 2,355 2,246 2,340 410 1,312

Ariel in Spain Revenues, euros 9,109 9,888 9,236 2,343 6,977

Ariel in Revenues, euros 8,952 11,007 8,880 6,759 21,066

Italy Standard powder 5 9 UK N

Spain Standard powder 8 12 UK Y

Italy Standard powder 5 9 UK Y

Spain Standard powder 4 5 Italy N

Surf in Surf in Surf in Ariel in Ariel in Ariel in Ariel in

reversed. 40 percent of Ariel products in Italy, and 31 percent of Ariel products in

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.