Global giants and local stars: Multinational brand ...

Global giants and local stars:Multinational brand amalgamation∗

Vanessa Alviarez† Keith Head‡ Thierry Mayer§

January 23, 2019

PRELIMINARY AND INCOMPLETEPlease read the updated version at this URL:

https://www.dropbox.com/s/siixdfft8nqd8wo/global_giants.pdf?dl=0

Abstract

In the standard theory, multinational corporations (MNCs) take blueprints devel-oped at home and move production abroad to exploit comparative advantage, bypasstrade costs, or achieve economies of scale. An altogether different mode of expansionabroad occurs when firms expand by acquiring brands created in other countries.The purpose of such multinational brand amalgamation is to exploit complementar-ities between the firm’s original and acquired products. We identified the country oforigin for 1000s of beverage brands using trademark registries and web-scraping. Us-ing data on the volumes and values of brand-level sales in 79 countries, we invert thesales equation of a multi-product oligopolist to back out appeal, cost, and markupsfor each brand-market combination. This allows us to distinguish the effect of bor-ders and distance on preferences separately from the effects on cost. The estimatesthen inform counterfactual experiments to determine the effects of changing owner-ship patterns on concentration and consumer welfare.

∗This research has received funding the Centre for Innovative Data at UBC. Very helpful comments werereceived during seminar presentations at the Universities of Texas and Michigan during the fall of 2018.†Sauder School of Business, University of British Columbia, [email protected]‡Sauder School of Business, University of British Columbia, CEPR, and CEP, [email protected]§Sciences Po, Banque de France, CEPII, and CEPR, [email protected]

1

1 Introduction

The dominance of enormous multinational corporations (MNCs) is a feature of modernmarket economies that has long attracted attention. The standard notion of how multi-nationals become so large is by achieving economies of scale in a set of core productsdeveloped at headquarters. The German car maker BMW is a good illustration. One corebrand accounts for the majority of the parent company’s sales and market value.1 Anotherroute to large size that the literature has not explored is the amalgamation of products de-veloped in multiple countries through mergers and acquisitions. Auto examples includeFiat-Chrysler and Renault-Nissan.2 Acquisition of foreign brands can dramatically alterthe position of firms in foreign markets. Diageo, owner of such well-known spirits brandsas Johnnie Walker, Smirnoff and Tanqueray, had a market share of less than two percentin India in 2013. Then it bought the local Group United Breweries, and now it has a 39%market share in this economy.

The goal of this paper is to answer three questions inspired by the above observa-tions. First, how widespread is the phenomenon of multinational brand amalgamation?Second, what economic mechanisms motivate firms to outbid brand-owners from othercountries in order to add their brands to the existing roster? For some products it appearsthat one of the most important motivations is to overcome home bias by consumers. Inother words, only using the MNC’s set of home-country brands would limit the marketshare (and hence profits) to be obtained in foreign markets. Another rationale is to findnew markets and distribution channels for the existing roster of their globally orientedbrands. The third question is whether the increasing concentration of brands under theownership of a small number of dominant firms can be expected to reduce consumerwelfare, especially in the target markets. To the extent that independent domestic brandswill now be priced to minimize competition with the brands of the MNC this seems verylikely. However, the MNC may also use the production and distribution assets of thenewly acquired firm to bring in its own brands, or use its own distribution channels tomarket the recently incorporated brands, leading to more variety at lower cost.

Going back to the seminal papers on the economic theory of multinationals (Markusen(1984); Helpman (1984)), the literature assumes that firms create their products at head-quarters and then manufacture them abroad. This assumption continues in recent work(surveyed in Antras and Yeaple (2014)). Arkolakis et al. (2018) explicitly identify the firm’scountry of origin as “the source of the idea.” From this perspective, cross-border acquisi-

1Interbrand calculates the value of the BMW brand to be $41 billion in 2018 compared to recent marketcapitalizations of BMW AG at $53 billion.

2BMW itself owns the UK brands Mini and Rolls-Royce.

2

tions of existing firms are just an alternative means to greenfield investment for a firm toexpand its overseas production capacity. This implicitly assumes that the acquirer com-pany does not retain the portfolio of brands of the recently acquired firm, but instead,that it expands production only through its “headquarter” brands. While the use of ac-quisitions for expanding production of a given product seems relevant in some industries(cement, for example), the beverage firms we study only rarely drop brands after acquir-ing them. Instead, adding distinct brands to the MNC’s roster appears to be a primemotivation for the acquisition.

The unique dataset assembled in this paper identifies each brand’s country of origin,that is the country in which the brand was originally developed. We do so by combininginformation from trademark registries and Wikipedia articles, supplemented by addi-tional web sources. Using corporate databases we also determine the headquarter coun-try of the current owner of the brand. This information allows us to assess the extent towhich cross-border acquisitions are implemented as an instrument to suppress existingcompetition, (“buy to kill”, Cunningham et al. (2018)), or as a tool for expanding marketshare (“buy to keep”). Since our data do not permit us to use the production functionapproach of De Loecker and Warzynski (2012) to calculate markups, we rely on an esti-mated oligopoly model with a constant elasticity of substitution demand in order to inferthe impact of cross-border brand acquisitions on firms’ markup.

This paper also relates to the literature on the sources of firm heterogeneity. In par-ticular, Hottman et al. (2016a) find that more than two-thirds of the observed differencesin firms’ performance are due to differences in firms’ appeal (i.e. quality or taste), faroutweighing the importance of differences in product scope and marginal cost. We con-tribute to this literature by quantifying the extent to which a firm can improve its appeal inforeign markets through the acquisition of local brands that have strong appeal. We alsoexamine whether the acquisition of foreign brands impacts multinationals’ appeal acrossmarkets, and whether this phenomenon varies across consumer product categories.

This paper joins the recent efforts to understand the drivers of market concentration(surveyed in Van Reenen (2018)). The current debate has revolved around whether con-centration is rising due to lax anti-trust enforcement or the success of superstar firmsGrullon et al. (2018); Autor et al. (2017). This paper provides grounds for an alterna-tive explanation in which market concentration is the result of cross-border acquisitionof brands as a firms’ attempt to overcome home bias and geographic specificities. Ourdataset is better equipped to measure market concentration as it overcomes two mainlimitations. First, instead of using firms’ output—which includes exports and excludesimports—we calculate measures of concentrations from brand-level sales in a given mar-

3

ket without regard to where the goods were sourced. Second, we depart from the focuson the U.S., and measure concentration across 79 countries.

We adopt the multi-product oligopoly model with a constant elasticity of substitution(CES) demand (Atkeson and Burstein (2008); Edmond et al. (2015); Hottman et al. (2016a))as our workhorse for estimation and counterfactuals. We start by inferring brands’ desti-nation specific and time variant appeal, lnAbnt, as well as, delivered marginal costs, ln cbnt,using the demand equation and price formulation derived from the model. In order toback out measures of a brand’s global quality, ln Ab, and marginal production cost, ln cb,we recover the brand fixed effect from a regression of lnAbnt and ln cbnt on the determi-nants of frictions between the brand’s country of origin (b)—and the headquarter countryof the company (c)—and the destination market (n), including home effects (both brandand corporate), distance, common language and regional trade agreements. Armed withan estimation of how changes in corporate headquarters of a brand affect consumer de-mand and firm costs, we construct a counterfactuals. In particular, we consider a policyof forced divestiture of domestic brands owned by foreign firms.

The remainder of the paper is organized as follow. Section 2 describes the data andpresents statistics on the extent of multinational brand amalgamation. Section 3 intro-duces a quantitative framework to study the effects of cross-border brand acquisitions.Section 4 quantifies the relative importance of appeal and cost of acquired foreign brandsin explaining the observed differences in firms’ performance. Section 5 estimates the effectof borders and distance on brand’s preferences and cost. Section 6 performs counterfac-tual experiments to determine the effects of changing ownership patterns on concentra-tion and consumer welfare. Section 7 concludes.

2 Data and Relevant Facts

Table 1 provides a preview of the type of information in our data and also motivates thetitle of our paper. Diageo was formed in 1997 as a merger of Grand Metropolitan andGuinness. It dramatically expanded its roster of spirits brands when took over the brandsof the failing Seagram company in 2001. On its website Diageo distinguishes between“Global Giants” and “Local Stars.” We operationalize the concept of the former as brandsthat are sold in many countries and achieve high market share world wide and the latteras brands sold in few markets but which achieve very high market share in their countryof origin. While the table shows all of Diageo’s Global Giants, it selects 7 of the Local Starsto illustrate the range of countries represented. In keeping with Diageo’s own focus, 12of the 14 brands shown are spirits.

4

Table 1: A selection of Diageo brands

Global Giants

Origin: UK UK UK Russia Jamaica Ireland Ireland# Markets: 68 21 28 64 43 57 30rank (world): 2nd 30th 46th 1st 12th 24th 21st

Local Stars

Origin: Brazil India Turkey Venezuela Australia Canada Kenya# Markets: 2 2 2 4 1 3 1rank (origin): 6/44 1/47 1/51 2/44 5/119 5/87 1/14Note: Rank of Global Giants is out of 1681 spirits brands (first 6 columns) and 1567 beer brands (7th column). Rankof Local Stars shown relative to number of brands offered in the origin country.

Market shares and prices

Data on value and/or volume sales at the brand level is obtained from Passport GlobalMarket Information Dataset (GMID) from Euromonitor, which records information for83,000 brands owned by 46,000 companies, across 153 product categories, in 79 countries.The consumer product categories included in the GMID dataset account, on average, for20% of household final consumption expenditure. GMID also allows us to track anychanges in ownership, at the brand level, occurring over the period 2006–2016, which is aunique feature of this dataset.3

In this paper, we concentrate our attention on three main aggregate beverage cate-gories (alcoholic, hot and soft drinks), for which we have volume as well as value in-formation, making it possible to calculate unit prices at the brand level in each country-year pair. These categories include: Beer, Spirits, Wine, Coffee, Bottled Water, Carbon-ates, Concentrates, Juice and Energy Drinks. It is worth noting that this dataset is betterequipped to measure market concentration, as it overcomes two main limitations inher-ent to databases relying on firm’s revenue. First, firm’s revenue includes exports to othermarkets and excludes imports, potentially biasing concentration measures. To the ex-

3Most M&A datasets record changes in ownership at the firm level without providing explicit informa-tion about which divisions or brands are involved in the transaction. The M&As observed in GMID dataare corroborated with Zephyr, a comprehensive database of merges and acquisitions from Bureau van Dijk.

5

tent that imports comprise products of foreign firms, this will lower concentration in themarket; but, imports carried out by large firms with little or no local production can actu-ally increase concentration relative to measures based on domestic production. Similarly,including exports could significantly upward bias concentration when the exporters aremultinationals that use the local market as export platform. Our data overcomes theseissues, as we construct measures of concentration by looking at brand-level sales in amarket without regard to where the goods being sold were sourced. Second, most of theresearch on rising concentration only uses data pertaining to the US market, whereas ourdata source considers 79 countries.

Brand origins

A novel feature of the dataset assembled in this paper is the identification of the coun-try of origin for each brand. We use web-scraping techniques to identify the origin ofbrands from online information provided in Wikipedia, and two government trademarkregistries, the World Intellectual Property Organization (WIPO) and the United StatesPatent and Trademark Office (USPTO). We design a classification algorithm based in theinformation available in these three sources to identify each brand’s country of origin.This allows us to separate the headquarter country of the current owner from the countryin which the brand was originally created.

There are two core challenges in classifying brand’s country of origin. First, we needto assess that our search results correspond to the brand we are indeed looking for (e.g.Searching for “Desperados”, the beer brand from France, returns Desperados, a bandfrom Spain). Second, we need to select amongst multiple “nominations” in those cases inwhich more than one country of origin is returned from Wikipedia, WIPO and USPTO.

Within Wikipedia’s website, we search for each brand name combined with a helperword that refers to the narrow category where the brand belongs to (e.g. Beer, Wine, etc.).We tabulate the information in Wikipedia’s “Summary” (introductory paragraphs), andin Wikipedia’s “Infobox” (tabular data located at the top right of the Wikipedia page),for Wikipedia pages in six languages: English, Spanish, German, Italian, Portuguese, andtranslating the helper word accordingly. From the Wiki summary we extract the firstcountry and nationality mentioned. We also recover the country information available inWiki Infoboxes, under the following field categories: origin, created, founded, location,headquarter, nationality, produced.

Both WIPO and USPTO maintain public databases with information on trademarkregistrations, such as date of the application, registration, and expiration date, status(pending, active or inactive), owner’s name and address and trademark description. For

6

brands that have changed ownership, USPTO also keeps an assignment dataset thattracks chronologically the owners of a given trademark.4

We designed a regression/voting classification method to assign a country of originto each brand in our dataset. Our method begins with a large set of nominations and abinary dependent variable indicating nominations that are correct according to our own“manual classification.” The latter is in most cases a Google search conducted by one ofthe authors. We then regress the match variable on the following explanatory variables:(1) “string distance” between the brand name in GMID and the search output; (2) loca-tion of GMID brand name in search output; (3) presence of keywords associated with thecategory (e.g. Vodka for spirits, Lager for beer) in Wikipedia summary or USPTO trade-mark description; (3) field in which the country was detected: origin, created, founded,location, headquarter, nationality, produced; (4) language: English, German, Spanish,French, Italian, Portuguese; (5) tax haven dummy, and (6) date of the trademark applica-tion/registration. The fitted value of this regression on other brands gives a probabilitythat each nomination is correct. We then choose among the nominations based on theirpredicted probabilities, giving greater weight when more than one source of informationfavours the same nomination.

Corporate headquarters

The headquarter country of each company in GMID dataset is obtained by combiningOrbis from Bureau van Dijk, the historical Directory of Corporate Affiliations from Lexis-Nexis, and Capital IQ. These datasets provide detailed information about ownershipstructure of the firm, as well as information on their affiliates’ location, sectors of opera-tion, sales, assets, operating profits and employment. Matching the name of each brand’sowners in the GMID dataset with the names of firms in the Orbis, Lexis Nexis, and Capi-tal IQ datasets, we identify the headquarter country of each brand’s owner in our dataset,with a careful treatment of companies incorporated in tax haven locations.

World market share of Non-HQ brands

Brand amalgamation is a widespread practice by MNCs. Figure 1 shows the world mar-ket share of non-headquarter brands—defined as those which country of origin differsfrom the headquarter country of their owner–in alcoholic, hot and soft drinks categories.The figure also distinguishes how much of these brand’s sales take place in brand’s coun-

4Firms are not enforced to report changes in ownership to USPTO, but failing to do so will precludethem from defending their trademark against infringement attempts by new registrants.

7

try of origin (share of non-HQ brands in local market) and abroad (share of non HQbrands in non-local markets). In the beer category, non-headquarter brands account formore than 60% of world market share; and for eight out of the ten beverage categories, theworld market share of foreign brands is 25% or more. For beer, cider and bottled watersales of non-HQ brands in their home market (lavender bar) represent more than half oftheir world market share; whereas in wine and carbonates most of world market share offoreign brands are due their sales outside the brand’s country of origin.5

Figure 2 displays the sales of the ten largest firms worldwide in four selected cate-gories. Each bar stacks four components of total sales: (1) sales of headquarter (home)brands in the owner’s headquarter country (light blue); (2) sales of headquarter brandsoutside the headquarter country (dark blue); (3) sales of acquired (foreign) brands in thebrand’s country of origin (lavender); and (4) sales of foreign brands outside the brand’scountry of origin (light gray). The largest firms among the top 10 sellers in the Beer, Spir-its and Bottled Water categories, rely relatively more on sales of acquired brands, particu-larly in the brand’s country of origin. Carbonates show a different pattern in which firms’revenue is concentrated in sales of headquarter brands, at home and abroad.

3 Modeling Framework

3.1 Sketch of the Brand Amalgamation Problem

There are two key questions the model must address. First, why do firms acquire brandsfrom other firms, as opposed to relying on the brands they create in-house? Second, whyare the creators of brands willing to relinquish their creations? If we think of the assign-ment of brands to firms as the outcome of an auction, we want to know the systematicdeterminants of the identity of the highest bidder. The model should specify the fractionof brands originating in given country end up being owned by firms headquartered else-where. Another moment of great interest is the fraction of the sales of multinational firmsderived from brands it acquired, both in the headquarter country and abroad.

We see the following benefits to the MNC obtains by acquiring foreign trademarks:(i) acquire their reputation, especially in the brand’s country of origin, mitigating homebias effects; (ii) use the production and distribution assets of the newly acquired firm tobring in the MNC’s own products, or (iii) use the MNC’s own production and distribution

5Multinational brand amalgamation is important in industries other than beverages. The US-basedsnack foods maker Mondelez, is the second largest chocolate seller in the world but it achieves 81% of itschocolate sales via brands originating in Britain (Cadbury), Belgium (Cote d’Or), Switzerland (Toblerone,Milka), Norway (Freia, Marabou), and Greece (Lacta).

8

Figure 1: World Market Share of Non-headquarter Brands

Note: This figure shows world market share of non-headquarter or foreign brands, defined as thosebrands which country of origin is different from the headquarter country of their owner. World marketshare of foreign brands are break down in two categories: (1) world market share of foreign brands intheir own country of origin (or local market) (lavender) and (2) world market share of foreign brandsoutside the brand’s country of origin (light gray).

9

Figure 2: World sales of the 10 largest companies

(a) Beer (b) Spirits

(c) Bottled Water (d) Carbonates

Note: This figure shows the sales of the largest 10 firms in the world in four categories. Firm’s sales arebreak down in four categories: (1) sales of headquarter brands in the headquarter country (light blue);(2) sales of headquarter brands outside the headquarter country (dark blue); (3) sales of acquired orforeign brands in the brand’s country of origin (lavender); and (4) sales of foreign brands outside thebrand’s country of origin (light gray).

10

channels to market the recently incorporated brands in other markets.The model we intend to develop has three stages.

1. MNCs bid to acquire the brands of target firms. The highest bidder is the MNC thatgenerates the highest profits with the targeted brands added to its roster.

2. MNCs decided in which market to offer each of their brands, taking into accountthe distributional assets available in each market as a consequence of their currentbrand roster.

3. Prices and market shares determined in each market as the Nash Equilibrium in aBertrand or Cournot oligopoly structure.

Over time, MNCs will increasingly diversify their brand rosters and also expand theirinternational scope. While the first two stages are not yet formalized, the third stage isoutlined in the next subsection.

3.2 CES oligopoly prices and market shares

We interpret the data from the lens of a quantitative framework in order to: disentanglethe brand and firm market share in its components of appeal, cost and markups; andunderstand the role of the acquired portfolio of brands in explaining the observed levelof firm heterogeneity.

Demand

Consumers have CES preferences over brands offered in country n, and they substituteacross brands with an elasticity σ, regardless of the identity of the firm that owns thebrand.6 The appeal of a brand, Abnt, is market, n, and time, t, specific. First, having abrand appeal that is market-dependent allows the model to capture the fact that a brandcan be popular in one country (very often its origin), but be less attractive to consumers

6In our framework consumers face the same elasticity of substitution across brands offered by differentfirms than across brands offered by the same firm. This departs from Hottman et al. (2016b) two-nestedCES structure, which allows for different σ’s to substitute across products offered by a given firm, σB andto substitute across products offered by different firms, σF . The unit of observation in our dataset arebrands (e.g. Coca-Cola, Pepsi, Fanta) rather than universal product codes (UPCs) (e.g Coca-Cola Classic of12 oz cans in a 24/Pack; Coca-Cola Bottle of 2L). While Coca-Cola Bottle of 2L (UPC No. 049000050103)could be a closer substitute for Coca-Cola Classic Coke, 96 fl oz (UPC No. 049000005486) than what DrPepper TEN, 12 fl oz, 12 pack (UPC No. 078000103168) is; at the level of the brand, it is more plausible thatconsumers do not face significantly different levels of sensitivity when substituting between Coca-Cola andSprite, both brands owned by Coca-Cola Co., than when substituting between Coca-Cola and Dr Pepper,where the latter is owned by Dr Pepper Snapple Group Inc.

11

in other countries. In section 5 we estimate how much of the variation in brand’s appealacross countries can be explained by a brand being particularly appealing to consumersin the same country as where the brand was originally designed, this is we estimate theimportance of “home bias.” Second, having a time-variant appeal allow us to capturechanges in taste, if any, that consumers experience when a brand changes owners afteran acquisition.7 For simplicity, we suppress the time subscripts for the rest of this sec-tion. Let utility of the representative consumer in market n have a constant elasticity ofsubstitution:

Un =

[∑b

(Abnqbn)σ−1σ

] σσ−1

. (1)

Given the CES structure of demand, the market share conditional on brand b servingmarket n is:

sbn = (pbn/Abn)1−σP σ−1n , (2)

where pbn is the price of brand b in market n, and Pn is the market price index whichaggregates over all the brands offered in the market n, as indicated by Ikn:

Pn =

[∑k

Ikn(pknAkn

)1−σ] 1

1−σ

. (3)

The total market share of firm f in market n, Sfn, is obtained by aggregating the marketshares of all the brands the firms owns (Mfb = 1) and offers in market n (Ibn = 1):

Sfn =∑b

MfbIbnsbn. (4)

The brand level profits earn by firm f in market n is:

πbn = qbn(pbn − cbn) = sbnYn(pbn − cbn)

pbn= sbnLbnYn, (5)

where qbn denotes the quantity sold in market n, cbn is the marginal cost of deliveringbrands to market n, Yn denotes total expenditure in market n in the relevant productcategory, and Lbn ≡ (pbn − cbn)/pbn is the Lerner index relevant in that brand-market

7Consumers could change their valuations of the brand because they perceived an actual change in thequality of the product after the acquisition, or simply because their valuation of the brand also factors intheir perception of its owner.

12

combination. The firm maximizes the sum of πbn over the set of brands it owns:

Πfn =∑b

MfbIbnπbn, (6)

Markups for different conduct assumptions

First order conditions on this profit function for the firm yields equilibrium prices for eachbrand and destination:

pbn = µfncbn =εfn

1− εfncbn, (7)

where µfn is the markup of firm f , which is the same across all brands the firm offers inorder to satisfy the first-order conditions of the profit maximization problem with a CESdemand system.8

µbn = µfn =εfn

εfn − 1, ∀Mfb = 1. (8)

Since each firm internalizes the effects of its pricing decision on the market price index,they perceive a variable elasticity of demand, εfn, that is decreasing on firm expenditureshare, Sfn, and that depends on the assumed form of oligopoly interactions. In particular,the elasticity firm-market perceived demand elasticity under Bertrand competition the isgiven by:

εfn = σ + (1− σ)Sfn, (9)

whereas with Cournot competition the firm-market perceived demand elasticity is:

εfn =σ

1 + (σ − 1)Sfn. (10)

Because the price elasticities depend on market shares in a highly non linear way, it iswell known that there are no closed-form solutions for equilibrium prices. Equilibriumcan however be obtained via fixed point iteration, starting with a guess of prices (suchas the monopolistic competition price vector p0

bn = (σ/(σ − 1))cbn). At each step, marketshares are obtained for a given set of markups, which then imply a new set of optimalmarkups, new market shares, until convergence to unique price and market share vectorsis reached.

8This results comes from Hottman et al. (2016b) and contrasts sharply with the case of linear demandanalyzed by Mayer et al. (2014).

13

Inferred appeal, costs, and markups

The CES oligopoly model can be put to to work to back out appeal, costs, and markupsfrom the observed market shares and prices, combined with an estimate of σ. FollowingKhandelwal et al. (2013) we manipulate equation (2) to obtain,

ln sbnt + (σ − 1) ln pbnt = (σ − 1) lnAbnt + (σ − 1) lnPnt. (11)

In order to isolate appeal as a function of observables, we need to eliminate the priceindex. Since it is an nt variable, it is common to all brands in a given market year andcan therefore removed through demeaning. As in Hottman et al. (2016b), a tilde over avariable denotes its geometric mean over the relevant market-year (specified in its sub-script). So long as we have an estimate of σ we can express inferred appeal as a functionof observables:

ln(Abnt/Ant) =ln(sbnt/snt)

σ − 1+ ln(pbn/pnt). (12)

Only relative Abnt within a market-year can be identified since multiplying all the Abnt bya scalar would not change the equilibrium market shares conditional on prices.

Equation (12) is equivalent to the regression approach of Khandelwal et al. (2013)equation (7) except that they aggregate over multiple sectors (and therefore include sectorfixed effects), whereas we calculate appeal within each category of goods. Equation (12)is also equivalent to a logged version of Redding and Weinstein (2018) equation (17).

Log markups can be obtained from taking the log of equation (8).

lnµfn = − ln(1− 1/εfn), (13)

Since c = p/µ, log delivered costs can be inferred as

ln cbn = ln pbn − lnµbn = ln pbn + ln(1− 1/εfn), (14)

where εfn is obtained by applying either the Bertrand or Cournot formulas shown inequations (9) or (10). These formulas rely only on Sfn and on an estimate of σ. Becauseonly firm-level market share matters for markups under CES, the ratio of cbn to its geo-metric mean (or to the inferred cost of any other brand owned by the same firm) is inde-pendent of the conduct assumption. This fact will be used in the decomposition of firmperformance in section 4.

14

Table 2: Elasticities of Substitution, σ

Category Elasticity SourceBeer 4.3 Nathan and Weinberg (2017)Spirits 2.8 Miravete et al. (2018)Wine 5.1 Aguirregabiria and Suzuki (2016)Bottle Water 2.2 Broda and Weinstein (2006)Carbonates 2.7 Dube (2004)Concentrates 3.4 Broda and Weinstein (2006)Juice 3.4 Broda and Weinstein (2006)Coffee 3.7 Broda and Weinstein (2006)Energy Drinks 1.8 Broda and Weinstein (2006)

In the Cournot case, inferred costs take the simple form:

ln cbn = ln pbn + ln(1− Sfn) + ln(1− 1/σ), (15)

which implies that, under Cournot, an increase in σ shifts the inferred log costs for everybrand up by the same amount. Another way to put this result is that relative markupsof brands owned by different firms under Cournot depend only on the market shares ofthose firms (and not on σ):

lnµbn − lnµb′n = ln(1− Sf ′n)− ln(1− Sfn)

A key requirement for extracting appeal, markups, cost from price and quantity datais an estimate of σ for each product group. In this version of the paper we mainly drawestimates of the elasticity of substitution from the industrial organization literature, withthe remaining estimates taken from the trade literature. Table 2 lists the own-price de-mand elasticities we use.

4 Contribution of foreign brands to firm performance

In this section we disentangle how much of the dispersion in firm’s sales is due to thefirm’s roster of headquarter brands (origin of the brand is the same as headquarters ofthe MNC) and how much is attributed to brands acquired abroad. We denote the set of boriginating in the firm’s headquarter (HQ) country as Ro

f and the set of brands acquiredin non-HQ countries as Rø

f . Extending the single-market approach of Hottman et al.(2016b), we disentangle how much of the dispersion of sales shares in a given market n

15

can be attributed to the inferred “appeal” and marginal cost of o and ø brands in eachfirm’s brand roster.

Let the roster of brands owned by denoted Rf = Rof ∪ Rø

f , that is the union of thesets of headquarter and non-headquarter brands. Recalling that Ibn is a binary variableset to one in markets where b is distributed, the number of o and ø brands offered in eachmarket are given by Nfn =

∑b∈Rf Ibn, N o

fn =∑

b∈RofIbn, and Nø

fn =∑

b∈RøfIbn. Adopting

the convention of Hottman et al. (2016b) to use tilde over variables to denote geometricmeans, we define

Afn ≡

∏b∈Rf

AIbnbn

1Nfn

and cfn ≡

∏b∈Rf

cIbnbn

1Nfn

(16)

Plugging equation (7) into (2), we obtain the market share for each brand offered in n:

sbn = Aσ−1bn (µfncbn)1−σP σ−1

n

Summing across brands in the roster that are offered in market n, we calculate firm’saggregate market share, Sfn. Then, multiplying and dividing by Nfn, cfn, and Afn, wecan express firm market share as

Sfn = P σ−1n

Nfn

(Afn

)σ−1

(cfn)σ−1 (µfn)σ−1

1

Nfn

∑b∈Rf

Ibn

(cfncbn

Abn

Afn

)σ−1 . (17)

Again extending notation from Hottman et al. (2016b) to the case of multiple markets,define the ∆g operator as the difference between a variable and the mean of that variableover all the firms in a product-group that are present in that market. Letting Ωn denote theset of firms offering at least one brand in market n, the total number of entrants is the car-dinality of that set, | Ωn |. Thus, for Sfn, for example, ∆g lnSfn = lnSfn− 1

|Ωn|∑

f∈ΩnlnSfn.

The key benefit is to eliminate n-level factors, namely P σ−1n , since our use of market shares

already eliminates market size (group-level expenditures in market n). The result is a de-

16

composition of market share into five determinants:

∆g lnSfn = (σ − 1)∆g ln Afn︸ ︷︷ ︸Appeal

+ ∆g lnNfn︸ ︷︷ ︸Scope

− (σ − 1)∆g lnµfn︸ ︷︷ ︸Markup

− (σ − 1)∆g ln cfn︸ ︷︷ ︸Marginal Cost

+ ∆g ln

1

Nfn

∑b∈Rf

Ibn

(cfncbn

Abn

Afn

)σ−1

︸ ︷︷ ︸Cost-appeal covariance

, (18)

The firm’s appeal can be expressed as a weighted geometric mean of the appeal of its HQand non-HQ brands with shares N o

fn/Nfn and Nøfn/Nfn as weights.9

Afn =(Aofn

)NofnNfn

(Aøfn

)Nøfn

Nfn . (19)

The delivered cost index for each firm can be separated into o and ø factors in the sameway, allowing us to decompose the appeal and marginal cost components into the contri-

9To obtain equation (19), start by grouping HQ and non-HQ brands separately:∑b∈Rf

Ibn lnAbn =∑b∈Ro

f

Ibn lnAbn +∑b∈Rø

f

Ibn lnAbn

Then multiply and divide by the number of brands of each type, and divide both sides by Nfn:

1

Nfn

∑b∈Rf

Ibn lnAbn =

(No

fn

Nfn

1

Nofn

) ∑b∈Ro

f

Ibn lnAbn +

(Nø

fn

Nfn

1

Nøfn

) ∑b∈Rø

f

Ibn lnAbn

From equation (16), it follows that ln Afn = 1Nfn

∑b∈Rf

Ibn lnAbn, so we can make substitutions to obtain

ln Afn =No

fn

Nfnln Ao

fn +Nø

fn

Nfnln Aø

fn,

where Aofn and Aø

fn are the geometric means of the HQ and non-HQ brand appeals offered in each market.

17

bution of HQ and non-HQ brands:

∆g lnSfn =

Firm’s Appeal︷ ︸︸ ︷(σ − 1)

[∆g

(N ofn

Nfn

ln Aofn

)]︸ ︷︷ ︸

Appeal of HQ Brands

+ (σ − 1)

[∆g

(Nøfn

Nfn

ln Aøfn

)]︸ ︷︷ ︸

Appeal of non-HQ Brands

+ ∆g lnNfn︸ ︷︷ ︸Scope

−

Firm’s Marginal Cost︷ ︸︸ ︷(σ − 1)∆g

(N ofn

Nfn

ln cofn

)︸ ︷︷ ︸

MC of HQ Brands

− (σ − 1)∆g

(Nøfn

Nfn

ln cøfn

)︸ ︷︷ ︸

MC of non-HQ Brands

+ ∆g ln

1

Nfn

∑b∈Rf

Ibn

(cfncbn

Abn

Afn

)σ−1

︸ ︷︷ ︸Cost-appeal covariance

− (σ − 1)∆g lnµfn︸ ︷︷ ︸Markup

. (20)

Hottman et al. (2016b) exploit properties of OLS to decompose the cross-sectional varia-tion of firm market shares. The method is to regress each of the components in the aboveequation on the demeaned log of firm market share, ∆g lnSfn:

(σ − 1)∆g

(Nkfn

Nfn

ln Akfn

)︸ ︷︷ ︸

Appeal of k ∈ o, ø

= βAk ∆g lnSfn + εAkfn,

∆g lnNfn︸ ︷︷ ︸Scope

= βN∆g lnSfn + εNfn,

−(σ − 1)∆g

(Nkfn

Nfn

ln ckfn

)︸ ︷︷ ︸

Marginal cost of k ∈ o, ø

= β ck∆g lnSfn + εckfn,

∆g ln

1

Nfn

∑b∈Rf

Ibn

(cfncbn

Abn

Afn

)σ−1

︸ ︷︷ ︸Cost-appeal-cov

= βV ∆g lnSfn + εVfn,

−(σ − 1)∆g lnµfn︸ ︷︷ ︸Markup

= βµ∆g lnSfn + εµfn,

Where βAo + βAø + β co + β cø + βN + βV + βµ = 1, since the coefficients from regressing each

18

yi on X add to one,∑

i βi = 1, whenever X =∑

i yi.10

In the regressions described above Sfn, N ofn, and Nø

fn are directly observed in the data.Equations (13) and (14) deliver µfn and cbn as functions of observed market share andprice data given an estimate of σ and a conduct assumption. Regarding appeal, equa-tion (12) provides ln(Abnt/Ant). The ∆g operator removes ln Ant because it is common toall firms for a given market-year.

Table 3: Variance Decomposition (Beverage Categories)

Decomposition Appeal Average MC ScopeHome Foreign Home Foreign

Beer 0.323 0.143 -0.114 -0.130 0.405(0.012) (0.008) (0.013) (0.010) (0.004)

Bottled Water 0.350 0.317 0.006 0.027 0.168(0.012) (0.011) (0.008) (0.008) (0.005)

Carbonates 0.321 0.149 -0.008 -0.028 0.370(0.008) (0.008) (0.004) (0.003) (0.005)

Spirits 0.405 0.445 -0.012 -0.333 0.340(0.026) (0.033) (0.021) (0.020) (0.008)

Wine 0.571 0.218 -0.116 -0.090 0.205(0.017) (0.013) (0.026) (0.024) (0.006)

Table 3 presents the decomposition results for several beverage categories, averagedacross countries. The value of β is a measure of how much of the variation in the distri-bution of firm sales can be attributed to appeal, marginal cost, and brand scope. The esti-mated coefficients show a great deal of heterogeneity across beverage categories, but yetsuggesting a dominant role for demand-side variables—appeal and scope—in accountingfor observed differences in firm’s performance. In particular, the variance decompositionindicates that from 46% to 84% of the overall size distribution can be attributed to thesedemand shifters, with a large component coming from the appeal of firm’s roster of non-HQ brands. In Spirits, more than half of the firm’s sales variation due to appeal is comingfrom non-headquarter (ø) brands (0.44/0.84); followed by a 48% for Bottled Water, 28%of Wine, 31% of Beer, and 32% of Carbonates.

10This result is obtained using the definition of the OLS estimator and the formula for the variance of asum in DeGroot and Schervish (2002), p. 220:

βk =cov(X, yk)

var(X)=

cov(∑

i yi, yk)

var(∑

i yi)=

∑i cov(yi, yk)∑

k

∑i cov(yi, yk)∑

k

βk =∑k

∑i cov(yi, yk)∑

k

∑i cov(yi, yk)

= 1

19

Figure 3: Contribution of Foreign Brands to Appeal(Average over Beverage Categories)

Figure 3 depicts, for each country and averaged across beverage categories, the esti-mated betas together with the 95% confidence intervals, for the overall contribution ofappeal, βA (blue dots), and the contribution of acquired or “foreign” brands, βAø (reddots). There is considerable variation in the relevance of appeal in explaining the firmsales across different countries, but for most of them the appeal of foreign brands countfor a sizable fraction of the variation attributed to firm’s appeal.

5 Estimation of the determinants of appeal

We estimate the following specification over the pooled of brands in all ten beveragecategories, across all countries and years:

lnYbnt = FEb + FEnt + Brand’s origin frictions(o)nt + HQ frictions(hq)nt,

20

where Y can be appeal (A), cost (c), or markup (µ); the cost-markup decomposition de-pends on whether Bertrand or Cournot conduct is assumed. Frictions include a homedummy, distance, and common language; where brand’s origin frictions pertain to thebrand’s country of origin relative to the destination market, and headquarter frictionspertain to the country where the brand’s owner is headquartered, also in comparisonwith the destination market. Finally, FEb and FEnt represent brand and destination-yearfixed effects, respectively.

Table 4: Pooled regressions explaining appeal, cost, and markupsln sbn lnAbn ln cbn lnµbn

mkt. share Appeal Bertrand Cournot Bertrand CournotBrand originshome 0.980a 0.315a -0.179a -0.260a 0.051a 0.128a

(0.097) (0.068) (0.028) (0.031) (0.008) (0.017)ln distance -0.178a -0.052a 0.051a 0.058a -0.006b -0.014a

(0.027) (0.020) (0.011) (0.011) (0.003) (0.005)common language 0.163a 0.010 -0.071a -0.081a 0.005 0.011

(0.063) (0.051) (0.025) (0.026) (0.006) (0.011)home (HQ) 0.136c 0.024 -0.040 -0.065b 0.015b 0.040b

(0.077) (0.053) (0.026) (0.029) (0.007) (0.016)ln distance (HQ) -0.033 -0.026 -0.000 0.003 -0.003 -0.008c

(0.022) (0.016) (0.009) (0.009) (0.002) (0.005)common language (HQ) 0.066 0.062 0.021 0.028 -0.007 -0.012

(0.052) (0.042) (0.022) (0.024) (0.006) (0.012)Observations 142413 138153 138139 138139 144579 144579R2 0.717 0.642 0.986 0.984 0.888 0.846

While the pooled regressions shown in Table 4 hide cross-category heterogeneity, onaverage, home-origin brands have huge advantages. Since exp(0.98) ≈ 2.66, home in-creases market share by 166%. The largest impact comes on the taste side (home bias).In particular being a home brand raises demand equivalent to a 37% price change. Costadvantages of home brands are also substantial, especially under Cournot, and brandsfrom nearby countries also have appeal and cost advantages, which a combined distanceelasticity (0.11), a bit higher than the 0.088 obtained by Head and Mayer (2018) for cars.Brands sell somewhat better in their HQ country, even holding brand origin constant,mainly a cost effect with a 4–7% advantage.

We can back out measures of brand level quality—defined as the destination- andtime-invariant component of appeal—and marginal production cost. The method is toregress the inferred lnAbn and ln cbn on “friction” determinants and recover the brandfixed effects. An interesting question is whether “quality pays.” The idea is that mak-

21

ing higher quality brands requires higher production costs. If the elasticity of productioncosts with respect to quality exceeds one, then higher quality brands will have lower mar-ket shares and, presumably, lower profits. Under Bertrand competition, we find quality isexpensive in the beer industry with an elasticities of 0.43, but very cheap for carbonates,with an estimated elasticity of 0.2. Since both elasticities are well below one, we infer thatquality is profitable in these industries.

Figure 4: How higher quality affects costs

Coefficient: .43

-3-2

-10

12

ln c

ost

-3 -2 -1 0 1 2log quality

Coefficient: .2

-4-2

02

4ln

cos

t

-6 -4 -2 0 2 4log quality

(a) Brewing Beer (b) Carbonates

6 Counterfactual exercises

6.1 Exact hat algebra for costs and appeal changes

We consider counterfactuals that change supply conditions as well as those in whichhome bias is eliminated. Market share in the counterfactual is s′bn = sbnsbn. We representpolicies affecting the cost of supply brand b to market n as cbn, and preference changesas Abn. At this stage, we hold the extensive margin of market entry at the brand levelconstant. Then , the change in brand-level market share is:

sbn =

(µfncbn

AbnPn

)1−σ

=(µfncbn/Abn)1−σ∑

k Iknskn(µknckn/Akn)1−σ, (21)

22

with the change in country’s price index under CES demand given by:

Pn =

(∑k

Iknskn(µknckn/Akn)1−σ

) 11−σ

. (22)

Aggregating (21) to firm-level market share changes yields:

Sfn =

∑bMfbIbnsbnsbn

Sfn. (23)

The adjustment of the firm-level markups is given by:

µfn =1

µfn

εfnεfn(εfnεfn − 1)

=εfn(εfn − 1)

(εfnεfn − 1). (24)

Recalling the Bertrand own-price elasticity is εfn = σ+ (1− σ)Sfn, its proportional changein the counterfactual is:

εfn = [σ + (1− σ)SfnSfn]/εfn. (25)

Recalling the Cournot own-price elasticity is εfn = σ/[1 + (σ − 1)Sfn], its proportionalchange in the counterfactual is:

εfn = σ[1 + (σ − 1)SfnSfn]−1/εfn. (26)

The algorithm initializes with sbn = 1. Then it aggregates to the changes in firm-level market shares Sfn using equation (23). Using data on initial firm market sharesSfn and an estimate of σ, we calculate the initial εfn vector assuming Bertrand (equa-tion 9) or Cournot (equation 10). Next, endowed with Sfn and εfn vectors, we use eitherequation (25) or (26) to obtain εfn. This plugs into (24) to obtain proportional changes inmarkups. Combined with the stipulated changes in distribution cost, cbn, and/or pref-erences, Abn, and the firm-level market share data, equation (21) has all the informationit needs to update the vector of brand-level market shares. A fixed point iteration is rununtil the vector of brand-level market share changes stabilizes. The resulting sbn is thesame as the one obtained by solving for the equilibrium, sbn, before and after the frictionchange and taking the ratio (up to machine accuracy).

23

6.2 Forcing the divestiture of foreign-owned, domestic-origin brands

Suppose instead of shocking frictions, we change the ownership matrix, Mfb. Specificallyover the last decade, many brands have been acquired by large multinationals. The coun-terfactual stipulates a new matrix of ownership relationships, M′fb. The “hat” equationsof the previous section hold, except now firm-level market share changes depend on thenew ownership matrix:

Sfn =

∑bM′fbIbnsbnsbn

Sfn. (27)

So far as we know, this is the first application exact hat algebra to merger analysis. Giventhe very low information requirements (just market shares, prices and σ must be known)this approach seems attractive as compared to methods that involve solving the fullmodel and thus require Abn and cbn which are generally unknown.

We are interested in calculating the cumulative impact on concentration and consumersurplus from all the brand acquisitions by MNCs. One way to proceed is to start fromthe status quo and imagine a policy compelling each foreign-owned firm to sell off thebrand it owns that originate in the local market. We do this by creating counterfactualowners for these brands, where each brand becomes a firm. Then, we use the Bertrandand Cournot versions of the CES oligopoly model to predict market shares and markupsin this counterfactual.

Figure 5 shows that the multinational brand amalgamation in the beer industry hashad strong impacts on both concentration and consumer welfare. Reversing this processas in our divestment experiment would lower concentration index by 2000 points in manycases, bring concentration back down to close or under the EU thresholds for high con-centration. The price index for beer would fall by as much as 63% (South Africa, Cournot)but the price changes vary substantially across markets. In general Cournot conduct leadsto much larger price index effects, echoing a result by Hottman et al. (2016b). These re-sults are obtained by applying the model. They are conservative results in the sense thatthe divested brands are allocated to single firms whereas in reality brand amalgamationoften took the form of buying brands from multiple independent original owners. To as-sess the model we plan to look directly at price changes following major acquisitions thatoccurred during the time-frame of our data.

6.3 Aggregate welfare changes

In terms of welfare analysis, we can write the resulting changes for both consumer surplusand firms’ profits. For the consumer, the compensating variation CV is the difference in

24

Figure 5: Effects of compelling foreign owners to divest domestic beer brands

(a) Conduct assumption: Bertrand

0.0 0.2 0.4 0.6 0.8

010

0020

0030

0040

0050

0060

0070

00

Share of foreign−owned, domestic origin

Mar

ket c

once

ntra

tion

(Her

finda

hl in

dex)

US

CN

JP

BR

GB

DE

MXKR

ESAU

RU

CA

FR

ZA

IT

2016 dataCF divestment

(#) %change in price index

high concentration (EU guidelines)(−1)

(−1)

(0)

(−18)

(−1)

(0)

(−5)(−6)

(−3)(−5)

(−3)

(−9)(−4)

(−38)

(−1)

(b) Conduct assumption: Cournot

0.0 0.2 0.4 0.6 0.8

010

0020

0030

0040

0050

0060

0070

00

Share of foreign−owned, domestic origin

Mar

ket c

once

ntra

tion

(Her

finda

hl in

dex)

US

CN

JP

BR

GB

DE

MXKR

ESAU

RU

CA

FR

ZA

IT

2016 dataCF divestment

(#) %change in price index

high concentration (EU guidelines)(−5)

(−4)

(0)

(−37)

(−5)

(0)

(−22)(−23)

(−11)(−15)

(−11)

(−25)

(−14)

(−63)

(−4)

25

expenditure for initial (0) and counterfactual (1) prices, holding utility constant at theinitial level.

CVn = Yn(1− Pn) with Pn =

(∑k

Iknskn(µknckn/Akn)1−σ

) 11−σ

. (28)

The algorithm described in section 6.1 yields equilibrium changes in markups for allbrand-destination combinations, µbn, and is therefore also directly yielding the changein consumer surplus from a counterfactual change in either costs, cbn, quality Abn, orownership, Mfb.

Regarding changes in profits, the core building block is the profits made by each brandin each market, πbn. Using (5), the change in profits at the bn level then writes:

π′bn − πbn = Yn(s′bnL′bn − sbnLbn),

where Lbn = (pbn − cbn)/pbn and L′bn = (p′bn − c′bn)/p′bn denote the Lerner index of b whenselling in n before and after the counterfactual experiment. At the firm-destination levelwe have to sum those profit changes over the set of brands owned:

∆Πfn =∑b

MfbIbn(π′bn − πbn) = Yn∑b

MfbIbn(sbnLbn − 1)sbnLbn. (29)

Some of the variables can be observed (Yn, sbn), and some can be inferred from the as-sumptions on market structure. Note in particular, that since

Lbn =µfn − 1

µfn=

1

εfnand Lbn =

1

Lbn

µfnµfn − 1

µfnµfn=

1

εfn, (30)

our iterative procedure provides values for the levels and changes of Lerner indexes, Lbnand Lbn, which naturally depend on the conduct assumed. The profit change is therefore:

∆Πfn = Yn∑b

MfbIbn(sbn/εfn − 1)sbn/εfn. (31)

Finally, the firm-level profit change is given by:

∆Πf =∑n

∆Πfn. (32)

26

7 Conclusion

This paper assembles a novel dataset that identifies each brand’s country of origin andthe headquarter of its current owner to document the international expansion of MNCsthrough the acquisition of brands created in other countries. It develops a CES oligopolymulti-product model that rationalizes the mechanisms that motivate firms to outbid brand-owners from other countries in order to add their brands to its existing roster of brands,delivering an equilibrium assignment of brands to owners.

We use the model in order to infer brand’s appeal and delivered marginal cost, andestimate how changes in brand’s corporate headquarters affect consumers’ demand andfirms’ costs. Finally, we construct counterfactuals evaluating the impact of changes inbrands’ ownership matrix across countries on market concentration and consumer wel-fare.

References

Aguirregabiria, V. and J. Suzuki (2016). Estimating the effects of deregulation in the on-tario wine retail market. mimeo.

Antras, P. and S. R. Yeaple (2014). Multinational firms and the structure of internationaltrade. In Handbook of international economics, Volume 4, pp. 55–130. Elsevier.

Arkolakis, C., N. Ramondo, A. Rodrıguez-Clare, and S. Yeaple (2018, August). Innovationand production in the global economy. American Economic Review 108(8), 2128–73.

Atkeson, A. and A. Burstein (2008). Pricing-to-market, trade costs, and international rel-ative prices. The American Economic Review 98(5), 1998–2031.

Autor, D., D. Dorn, L. F. Katz, C. Patterson, and J. Van Reenen (2017, May). The fall of thelabor share and the rise of superstar firms. Working Paper 23396, National Bureau ofEconomic Research.

Broda, C. and D. E. Weinstein (2006). Globalization and the gains from variety. QuarterlyJournal of Economics 121, 541—-586.

Cunningham, C., F. Ederer, and S. Ma (2018). Killer acquisitions. mimeo.

De Loecker, J. and F. Warzynski (2012). Markups and firm-level export status. AmericanEconomic Review 102(6), 2437–2471.

27

DeGroot, M. H. and M. J. Schervish (2002). Probability and Statistics, Third Edition. PearsonEducation.

Dube, J.-P. (2004). Multiple discreteness and product differentiation: Demand for carbon-ated soft drinks. Marketing Science 23(1), 66–81.

Edmond, C., V. Midrigan, and D. Y. Xu (2015). Competition, markups, and the gains frominternational trade. The American Economic Review 105(10), 3183–3221.

Grullon, G., Y. Larkin, and R. Michaely (2018). Are US industries becoming more concen-trated? Working paper.

Head, K. and T. Mayer (2018). Brands in motion: How frictions shape multinationalproduction.

Helpman, E. (1984). A simple theory of international trade with multinational corpora-tions. Journal of Political Economy 92(3), 451–471.

Hottman, C. J., S. J. Redding, and D. E. Weinstein (2016a). Quantifying the sources of firmheterogeneity. The Quarterly Journal of Economics 131(3), 1291–1364.

Hottman, C. J., S. J. Redding, and D. E. Weinstein (2016b). Quantifying the sources of firmheterogeneity. The Quarterly Journal of Economics 131(3), 1291–1364.

Khandelwal, A. K., P. K. Schott, and S.-J. Wei (2013). Trade liberalization and embed-ded institutional reform: evidence from Chinese exporters. American Economic Re-view 103(6), 2169–95.

Markusen, J. R. (1984). Multinationals, multi-plant economies, and the gains from trade.Journal of International Economics 16(3-4), 205–226.

Mayer, T., M. J. Melitz, and G. I. Ottaviano (2014). Market size, competition, and theproduct mix of exporters. American Economic Review 104(2), 495–536.

Miravete, E. J., K. Seim, and J. Thurk (2018). Market power and the Laffer curve? mimeo.

Nathan, H. M. and C. M. Weinberg (2017). Understanding the price effects of the miller-coors joint venture. American Economic Review 85(6), 1763–1791.

Redding, S. J. and D. E. Weinstein (2018). Accounting for trade patterns. Discussion Paper12446, Center for Economic Policy Research.

Van Reenen, J. (2018). Increasing differences between firms: Market power and the macro-economy. Working Paper.

28