Embed Size (px)
Citation preview
Global giants and local stars:Multinational brand amalgamation∗
Vanessa Alviarez† Keith Head‡ Thierry Mayer§
February 28, 2019
PRELIMINARY AND INCOMPLETEPlease read the updated version at this URL:
https://blogs.ubc.ca/khead/files/2019/02/global_giants.pdf
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 portfolio? For some products it ap-pears that one of the most important motivations is to overcome home bias by consumers.In other words, only using the MNC’s set of home-country brands would limit the mar-ket share (and hence profits) to be obtained in foreign markets. Another rationale is tofind new markets and distribution channels for the existing portfolio of their globally ori-ented brands. The third question is whether the increasing concentration of brands underthe ownership 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 greater 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 portfolio appears to be a primarymotivation 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’ markups.
This paper also relates to the literature on the sources of firm heterogeneity. In par-ticular, Hottman et al. (2016) 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. (2016))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 sketchesthe framework we have in mind for the process of multinational brand amalgamation.The final stage of this framework, oligopoly price or quantity setting with CES demand,is the subject of section 4. There we also show how we back out brand-market appeal, costand markups. Section 5 quantifies the relative importance of appeal and cost of acquiredforeign brands in explaining the observed differences in firms’ performance. Section 6estimates home bias and distance effects for both appeal and costs. To quantify effectsof changing ownership patterns on concentration and consumer welfare, Section 7 gener-ates counterfactual prices and market shares under a forced brand divestment scenario.Section 8 concludes.
2 Data and Relevant Facts
Table 1 provides a preview of the type of information in our data and also motivatesthe title of our paper. Diageo was formed in 1997 as a merger of Grand Metropolitanand Guinness. It dramatically expanded its portfolio of spirits brands when took overthe brands of the failing Seagram company in 2001. On its website Diageo distinguishesbetween “Global Giants” and “Local Stars.” We operationalize the concept of the formeras brands that are sold in many countries and achieve high market share world wide andthe latter as brands sold in few markets but which achieve very high market share in their
4
country of origin. While the table shows all of Diageo’s Global Giants, it selects 7 of theLocal Stars to illustrate the range of countries represented. In keeping with Diageo’s ownfocus, 12 of the 14 brands shown are spirits.
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 coun-tries. Within each combination of product category, market, and year GMID lists salesfor all brands above a threshold market share, which the documentation lists as 0.1%.It ispossible to recover total sales because GMID also provides sales for “Private Label” and“Others.” Where needed for clarity, we refer to brands with sales broken out individuallyas “listed brands.” GMID also allows us to track any changes in ownership, at the brandlevel, occurring over the period 2006–2016, which is a unique feature of this dataset.3
In this paper, we concentrate our attention on beverage categories for which GMIDprovides sales volume as well as value information, making it possible to calculate unit
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
prices at the brand level in each country-year pair. These categories are Beer, Spirits, Wine,Coffee, Bottled Water, Carbonates, Concentrates, Juice, and Sports and Energy Drinks.Table 2 shows that each category comprises hundreds of firms and there are thousands ofbrands in most categories. As we will discuss in the following section, we have identifiedheadquarter countries for most firms and determined the origin of most listed brands.
Table 2: Firms and their brands in the GMID beverage data
Category Brands Firms Headquarter Origin MarketsCountries Countries
Beer 2229 455 76 91 77Cider 317 191 44 31 59Spirits 2927 869 82 98 76Wine 1692 779 59 65 54Coffee 812 542 76 79 77Water 1293 821 81 93 77Carbonates 1047 475 77 78 77Concentrates 655 506 79 69 72Juice 1347 864 84 92 77Energy 513 375 73 64 77
This dataset is better suited for measuring market concentration than data from theeconomic census or firm-level databases such as Compustat and Orbis. This is becauseit overcomes two main limitations inherent to databases relying on firm’s revenue. First,firm’s revenue includes exports to other markets and excludes imports. Thus it doesn’tmeasure sales in the market in question. To the extent that imports comprise productsof foreign firms, this will lower concentration in the market; but, imports carried out bylarge firms with little or no local production can actually increase concentration relative tomeasures based on domestic shipments. Similarly, including exports could significantlyupward bias concentration when the exporters are multinationals that use the local mar-ket as export platform. Our data overcomes these issues, as we construct measures ofconcentration by looking at brand-level sales in a market without regard to where thegoods being sold were sourced. Second, most of the research on rising concentrationonly uses data pertaining to the US market, whereas our data source considers up to 77countries per category.4
4One of these markets, Venezuela, only has reliable price data through 2013.
6
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. Forbrands that have changed ownership, USPTO also keeps an assignment dataset thattracks chronologically the owners of a given trademark.5
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-
5Firms 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
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.
Table 3: Sources of brand origins
Category Manual Wikipedia Trademark registries UnknownChecks Infobox Summary WIPO USPTO counts sales
Beer 0.53 0.06 0.10 0.12 0.04 0.15 0.03Carbonates 0.09 0.02 0.01 0.43 0.14 0.32 0.03Spirits 0.31 0.03 0.04 0.23 0.08 0.30 0.08Water 0.68 0.01 0.01 0.00 0.02 0.28 0.04Wine 0.23 0.06 0.06 0.25 0.07 0.33 0.13Note: fractions of listed brands, except last column which is the fraction of sales by listedfirms (excludes “Others” and “Private Label”)
Table 3 displays the origins of brands for the five categories where we have identifiedorigins for two thirds or more of the brands. The main contributors of origins varies acrosscategories, with spirits and carbonates relying heavily on the World Intellectual PropertyOrganization (WIPO) and bottled water mainly based on our own internet searches (man-ual checks). Beer currently has by far the most complete data and draws more evenly onall the sources. The data collection process is ongoing and future drafts of this paper willinclude the other five beverage categories.
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.
8
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-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.6
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 The Brand Amalgamation Problem
Why do firms acquire brands from other firms, as opposed to relying on the brands theycreate in-house? Moreover, why would the brand be worth more to the brand buyer thanit was to the brand seller? If we think of the assignment of brands to firms as the outcomeof an auction, we want to know the systematic determinants of the identity of the highestbidder. Ideally we would have a model that specifies the fraction of brands originatingin given country that end up being owned by firms headquartered elsewhere. Anothermoment of interest is the fraction of the sales that multinational firms derive from brandsthey have acquired abroad.
6Multinational 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).
9
Figure 1: World Market Share of Rest of the World Brands
Note: This figure shows world market share of rest of the world brands (R), defined as those brandswhich country of origin is different from the headquarter country of their owner. World market shareof foreign brands are break down in two categories: (1) world market share of R brands in their owncountry of origin (or local market) (lavender) and (2) world market share of R brands outside thebrand’s country of origin (light gray).
10
Figure 2: World sales of the 10 largest companies
(a) Beer (b) Spirits
(c) Bottled Water (d) Carbonates
Note: This figure shows the sales (USD Bn) of the largest 10 firms in the world in four categories.Firm’s sales are break down in four categories: (1) sales of headquarter brands in the headquartercountry (dark blue); (2) sales of headquarter brands in the rest of the world (light blue); (3) salesof acquired or foreign brands in the brand’s country of origin (lavender); (4) sales of foreign brandsoutside the brand’s country of origin (light gray), and (5) brands with unknown country of origin(white).
11
We see the following benefits to the MNC obtains by acquiring foreign trademarks: (i)acquire the brand 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 distribu-tion channels to market the recently incorporated brands in other markets. Mechanisms(i) and (ii) apply most strongly to the brands that Diageo calls “Local Stars” whereasmechanism (iii) is about creating or further extending “Global Giants.”
The model we are sketching differs considerably from the “standard model” whichtraces back to work of Helpman and Markusen but whose modern version is set out inArkolakis et al. (2018). In the standard model varieties are invented in the headquartercountry. The task of the MNC is to assign their production location to countries takinginto account comparative advantage, demand, and frictions. In the proposed multina-tional brand agglomeration model, varieties potentially come from many different coun-tries. Their production locations are historically determined in some cases (Scotch, Evian)or pre-determined by production facilities (breweries, bottling plants). Thus the centraltask of this type of MNC is not so much where to produce a given set of varieties. Rather,it is to determine which varieties to own and where to sell them.
We envision a model with 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 portfolio.
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 portfolio.
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 portfolios and also expand theirinternational scope. Some firms may do much more of this than others, leading to anincrease in concentration at the world level. While the first two stages are not yet formal-ized, the third stage is outlined in the next subsection.
4 CES oligopoly model
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; and
12
understand the role of the acquired portfolio of brands in explaining the observed levelof firm heterogeneity.
4.1 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.7 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 consumersin other countries. In section 6 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.8 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)
7In 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. (2016) two-nested CESstructure, which allows for different σ’s to substitute across products offered by a given firm, σB and tosubstitute across products offered by different firms, σF . The unit of observation in our dataset are brands(e.g. Coca-Cola, Pepsi, Fanta) rather than universal product codes (UPCs) (e.g Coca-Cola Classic of 12 ozcans in a 24/Pack; Coca-Cola Bottle of 2L). While Coca-Cola Bottle of 2L (UPC No. 049000050103) could bea closer substitute for Coca-Cola Classic Coke, 96 fl oz (UPC No. 049000005486) than what Dr Pepper TEN,12 fl oz, 12 pack (UPC No. 078000103168) is; at the level of the brand, it is more plausible that consumers donot face significantly different levels of sensitivity when substituting between Coca-Cola and Sprite, bothbrands owned by Coca-Cola Co., than when substituting between Coca-Cola and Dr Pepper, where thelatter is owned by Dr Pepper Snapple Group Inc.
8Consumers 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.
13
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 firm’s portfolio (Ff ) and offers in market n (Ibn = 1):
Sfn =∑b∈Ff
Ibnsbn. (4)
The brand-level profits earned 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-marketcombination. The firm maximizes the sum of πbn over the set of brands it owns:
Πfn =∑b∈Ff
Ibnπbn, (6)
4.2 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
εfn − 1cbn, (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.9
µbn = µfn =εfn
εfn − 1, ∀b ∈ Ff . (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 expenditure
9This results comes from Hottman et al. (2016) and contrasts sharply with the case of linear demandanalyzed by Mayer et al. (2014).
14
share, Sfn, and that depends on the assumed form of oligopoly interactions. In particular,the firm-market perceived demand elasticity under Bertrand competition is given by:
εBfn = σ + (1− σ)Sfn, (9)
whereas with Cournot competition the firm-market perceived demand elasticity is:
εCfn =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 p0bn = (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.
We can therefore express the equilibrium markups for the two conduct assumptionsas
µBfn = 1 +1
(σ − 1)(1− Sfn), (11)
andµCfn = σ × 1
(σ − 1)(1− Sfn), (12)
which both converge to σ/(σ−1), the monopolistic competition benchmark when marketshares go to 0.
4.3 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. (13)
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. (2016), a tilde over avariable denotes its geometric mean over the relevant market-year (specified in its sub-
15
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). (14)
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 (14) 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 (14)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), (15)
Since c = p/µ, log delivered costs can be inferred as
ln cbn = ln pbn − lnµbn = ln pbn + ln(1− 1/εfn), (16)
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 5.
In the Cournot case, inferred costs take the simple form:
ln cbn = ln pbn + ln(1− Sfn) + ln(1− 1/σ), (17)
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)
Under Bertrand, there is a different way to express the markup that leads to a similar
16
result noted by Hottman et al. (2016).
ln(µbn − 1)− ln(µb′n − 1) = ln(1− Sf ′n)− ln(1− Sfn)
Thus under both Cournot and Bertrand there is a way to express relative markups suchthat they depend only on the ratio of the share of the market not accounted for by thatbrand and its owners other brands. These relative markup distributions are invariant tothe elasticity of demand.
4.4 Implementation issues
There are several other practical considerations in the inference of appeal, costs andmarkups. First, we do not have transaction price data. Instead we divide the value ofsales by the quantity (in liters) to obtain brand-market-year level unit values. Perhaps be-cause the brands we examine are narrowly defined or the data more carefully recorded,we have not observed the noisiness in unit values that plagues trade data sets. It is worthacknowledging that the compositional changes within brands (rises in the sales weight ofhigh-priced bar codes) could generate unit value increases even if bar-code level priceswere not changing. In practice, this concern is mitigated by the narrow way GMID de-fines brands. For example, Johnnie Walker “Red Label” listed separately from “BlackLabel” (which is 60% higher priced). Second, the GMID data is censored at market sharesof approximately 0.1%. GMID sums the sales of smaller brands in a given market liststhem in aggregates called “Private Label” and “Others.” The latter often accounts for asizeable share of total sales. When calculating the geometric means of market shares andprices, we include only the individually “listed” brands. The assumption underlying theappeal calculation is that all the firms in the geometric mean operate in a market facing acommon price index. Thus, it is perfectly fine to define the tilde variables over subsets.10
In calculating the εfn that determine markups (and hence costs) we calculate firm marketshares Sfn using as a denominator the sales of all brands—including others and privatelabel.
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 4 lists the own-price de-mand elasticities we use. We hold these σ constant over time.
10It would even be possible to choose a single to use for sn and pn such as the largest brand in the market.However, the geometric mean is likely to be more stable over time.
17
Table 4: Elasticities of Substitution, σ
Category Elasticity SourceBeer 4.3 Miller and Weinberg (2017)Spirits 2.8 Miravete et al. (2018)Wine 5.1 Aguirregabiria and Suzuki (2016)Bottled Water 2.6 Friberg and Ganslandt (2003)Carbonates 2.7 Dube (2004)
5 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 portfolio of headquarter brands (origin of the brand is the same as headquarters ofthe MNC) and how much is attributed to brands acquired abroad. Extending the single-market approach of Hottman et al. (2016), we disentangle how much of the dispersion ofsales shares in a given market n can be attributed to the inferred “appeal” and marginalcost of HQ and non-HQ brands in each firm’s portfolio.
Denote the full portfolio of brands owned by denoted firm f as Ff = Hf ∪ Rf ∪ Uf .that is the union of the sets of headquarter and non-headquarter brands. Hf brands havethe same origin as their owner’s HQ. Coca Cola and Heineken are well-known exam-ples. Rf brands were acquired (mainly) or developed in the “rest of the world.” Thesebrands, the focus of this paper, include such market-dominating examples as Smirnoff (anoriginally Russian Vodka owned by a UK-based firm), San Pellegrino (an Italian-originsparkling water owned by a Swiss company). The residual set of brands comprise the setUf unknown-origin brands. For these cases either the brand’s origin or the owner’s HQhas not yet been determined.
Recalling that Ibn is a binary variable set to one in markets where b is distributed, thereare NH
fn =∑
b∈Hf Ibn HQ brands and NRfn =
∑b∈Rf Ibn non-HQ brands offered in market
n. Adding the the NUfn brands of unknown origin the full “scope” of firm f in market n
is Nfn. Adopting the convention of Hottman et al. (2016) to use tilde over variables todenote geometric means, we define
Afn ≡
∏b∈Rf
AIbnbn
1Nfn
and cfn ≡
∏b∈Rf
cIbnbn
1Nfn
(18)
18
Plugging equation (7) into (2), we obtain the market share for each brand offered in n:
sbn = Aσ−1bn (µfncbn)1−σP σ−1n
Summing across brands in the portfolio 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 . (19)
As in the earlier calculation to back out appeal, we need to remove the price index sinceour use of market shares already eliminates market size (group-level expenditures in mar-ket n). Since it is a factor common to all firms, the trick is to difference by the log of thegeometric mean. Again extending notation from Hottman et al. (2016) to the case of mul-tiple markets, define the ∆g operator as the difference between a variable and the meanof that variable over all the firms in a product-group that are present in that market. Thenumber of firms selling in market n is given by Mn =
∑f Dfn, where Dfn is 1 if firm
f offers at least one brand in market n, that is Dfn =∑
b∈Ff Ibn. Using this notation,∆g lnSfn = lnSfn − 1
Mn
∑f Dfn lnSfn. The result is a decomposition 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
. (20)
The firm’s appeal can be expressed as a weighted geometric mean of the appeal of itsHQ, non-HQ, and unknown brands with shares NH
fn/Nfn and NRfn/Nfn as weights. By
properties of geometric means,
Afn =(AHfn
)NHfnNfn
(ARfn
)NRfnNfn
(AUfn
)NUfnNfn (21)
19
The appeal term in equation (20) can now be further decomposed into three contributions:
(σ − 1)∆g ln Afn = (σ − 1)
[∆g
(NHfn
Nfn
ln AHfn
)+ ∆g
(NRfn
Nfn
ln ARfn
)+ ∆g
(NUfn
Nfn
ln AUfn
)]
The delivered cost index for each firm, cfn can be separated intoH,R and U componentsin the same way:
(σ − 1)∆g ln cfn = (σ − 1)
[∆g
(NHfn
Nfn
ln cHfn
)+ ∆g
(NRfn
Nfn
ln cRfn
)+ ∆g
(NUfn
Nfn
ln cUfn
)]
Substituting these six brand-type appeal and cost terms back into equation (20) allowsus measure the relative importance of the rest-of-world brands in explaining firm perfor-mance. The decomposition proceeds by running a simple regression of each of the nineright-hand-side components in the brand-origin augmented version of equation 20 on thedemeaned log of firm market share, ∆g lnSfn. Because the components sum to the total,the sum of the coefficients (each equal to a covariance divided by the variance ∆g lnSfn)must sum to one, as pointed out by Hottman et al. (2016).
In the regressions described above Sfn and Nkfn (for k ∈ {H,R,U}) are directly ob-
served in the data. Equations (15) and (16) deliver µfn and cbn as functions of observedmarket share and price data given an estimate of σ and a conduct assumption. Regardingappeal, equation (14) provides ln(Abnt/Ant). The ∆g operator removes ln Ant because it iscommon to all firms for a given market-year.
Table 5 presents the decomposition results for several beverage categories, with theregressions pooled across countries. These results assume Bertrand conduct, but this onlyaffects the cost and markup terms. This can be seen in Table 7 in the Appendix whichreports the decompositions under the Cournot conduct assumption. The contributions ofappeal and the dispersion term do not depend on the conduct assumption. The value ofestimate is a measure of how much of the variation in the distribution of firm sales canbe attributed to appeal, marginal cost, scope and covariation between cost and appeal(“Disp.” in the table).
The estimated coefficients show a great deal of heterogeneity across beverage cate-gories, but yet suggesting a dominant role for demand-side variables—appeal and scope—in accounting for observed differences in firm’s performance. In particular, the variancedecomposition indicates that from 83% to 111% of firm market share variation can be at-tributed to these demand shifters. Non-HQ Rf brands account for 29–46% of the appealof known brands. The second panel of table 5 follows HRW in first-differencing each of
20
Table 5: Variance Decomposition
Appeal Scope Cost µ Disp. # obs.Category Hf Rf Uf Hf Rf Uf
All GMID firms, Cross-sectionalBeer 0.26 0.14 0.06 0.37 0.24 -0.13 0.02 -0.13 0.17 9144Water 0.27 0.23 0.20 0.16 0.06 0.09 -0.07 -0.04 0.10 8688Carbonates 0.26 0.16 0.11 0.34 0.12 -0.02 -0.04 -0.09 0.15 6073Spirits 0.32 0.28 0.18 0.33 -0.01 -0.28 0.08 -0.02 0.11 18750Wine 0.41 0.17 0.19 0.23 0.01 -0.00 -0.06 -0.02 0.07 11655Pooled 0.31 0.21 0.15 0.30 0.08 -0.12 0.01 -0.05 0.12 54310Low Income 0.26 0.24 0.20 0.28 0.10 -0.07 -0.06 -0.07 0.12 28252High Income 0.36 0.17 0.08 0.34 0.05 -0.18 0.09 -0.04 0.13 26058
All GMID firms, GrowthBeer 0.25 0.19 0.08 0.28 0.25 -0.03 -0.00 -0.13 0.12 7823Water 0.32 0.28 0.30 0.10 -0.00 0.01 -0.03 -0.02 0.05 7504Carbonates 0.25 0.39 0.17 0.13 0.04 -0.02 -0.02 -0.02 0.08 5204Spirits 0.35 0.42 0.19 0.16 0.09 -0.25 -0.00 -0.01 0.05 16377Wine 0.42 0.24 0.35 0.11 0.10 -0.18 -0.07 -0.01 0.04 10278Pooled 0.32 0.31 0.21 0.16 0.11 -0.12 -0.02 -0.04 0.07 47186Low Income 0.28 0.36 0.22 0.15 0.06 -0.07 -0.02 -0.05 0.07 24473High Income 0.40 0.21 0.19 0.19 0.20 -0.22 -0.03 -0.02 0.07 22713
Firms with Sfn > 0.005, Cross-sectionalBeer 0.10 0.09 0.03 0.47 0.29 -0.03 0.02 -0.20 0.24 5978Water 0.28 0.25 0.15 0.22 0.04 0.10 -0.12 -0.07 0.14 6888Carbonates 0.26 0.08 0.09 0.45 0.17 -0.04 -0.07 -0.14 0.20 4785Spirits 0.09 0.08 0.01 0.57 0.28 -0.28 0.09 -0.05 0.21 11148Wine 0.02 0.09 0.05 0.36 0.42 -0.10 0.09 -0.04 0.11 7138Pooled 0.15 0.11 0.06 0.44 0.24 -0.08 0.00 -0.11 0.19 35937Low Income 0.14 0.12 0.07 0.41 0.23 0.01 -0.04 -0.13 0.18 19914High Income 0.16 0.11 0.04 0.47 0.25 -0.22 0.06 -0.07 0.20 16023
Firms with Sfn > 0.005, GrowthBeer 0.03 0.12 0.05 0.44 0.42 -0.02 0.02 -0.24 0.19 5033Water 0.33 0.23 0.21 0.16 0.01 0.07 -0.04 -0.04 0.08 5921Carbonates 0.20 0.26 0.19 0.31 0.10 -0.06 -0.06 -0.07 0.13 4094Spirits 0.22 0.23 0.12 0.33 0.17 -0.15 -0.01 -0.03 0.12 9576Wine 0.46 0.24 0.25 0.20 -0.15 -0.25 0.22 -0.03 0.05 6193Pooled 0.20 0.20 0.13 0.32 0.18 -0.06 0.02 -0.11 0.13 30817Low Income 0.18 0.19 0.17 0.29 0.13 0.07 -0.02 -0.13 0.12 17028High Income 0.24 0.21 0.06 0.38 0.29 -0.36 0.11 -0.07 0.15 13789Hf is the set of headquarter brands,Rf is the set of brands acquired in other countries andUf is the set of brands where origin is uncertain. Conduct is Bertrand.
21
the variables. Thus it is decomposition of the causes of growth in market share. Demand(appeal plus scope) accounts for 79% to 112% of growth firm market share variation.Non-HQ (Rf ) brands account for 36–62% of the appeal contribution to growth of knownbrands. The bottom two panels of table 5 consider only firms that achieve a market shareof half a percent or more in a market. This lowers the number of observations by aboutone-third. The most notable difference in the results is that scope rises substantially inrelative importance. For these larger firms non-HQ brands continue to explain a highshare of appeal effect but headquarter costs are also increasingly important.
6 Estimation of the determinants of appeal
We estimate the following specification over the pooled set of brands in all our beveragecategories (g), across all countries n and years t (the appendix reports separate regressiontables for all categories):
lnYbgnt = FEbg + FEgnt + Brand’s origin frictionsbn + HQ frictionsfnt,
where Y can be market share, sbnt appeal (Abnt), cost (cbnt), or markup (µfnt); the cost-markup decomposition depends on whether Bertrand or Cournot conduct is assumed.Frictions include a home dummy, distance, and common language; where brand’s originfrictions pertain to the brand’s country of origin relative to the destination market, andheadquarter frictions pertain to the country where the brand’s owner is headquartered,also in comparison with the destination market. Brand origins are time-invariant by def-inition whereas the owner f of a given brand b can vary over time. Finally, FEbg andFEgnt represent brand and destination-year fixed effects, respectively. Note that those arespecific to a category: Countries can have different price indices for different goods, andif brands operate in different categories, they are allowed to have different appeal, costsand markups in each of those.
While the pooled regressions shown in Table 6 hide cross-category heterogeneity, onaverage, home-origin brands have huge advantages. Since exp(1.048) ≈ 2.85, home in-creases market share by 185%. 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,
22
Table 6: Pooled regressions explaining appeal, cost, and markupsBertrand Cournot
ln sbn lnAbn ln cbn lnµbn ln cbn lnµbnhome 1.048a 0.307a −0.181a 0.042a −0.250a 0.110a
(0.115) (0.074) (0.033) (0.008) (0.035) (0.017)distance −0.147a −0.010 0.053a −0.002 0.058a −0.007
(0.032) (0.022) (0.012) (0.003) (0.013) (0.005)common language 0.161b −0.014 −0.084a 0.006 −0.093a 0.015
(0.070) (0.054) (0.029) (0.005) (0.029) (0.010)home (HQ) 0.180c 0.060 −0.062b 0.015b −0.087a 0.040a
(0.095) (0.062) (0.031) (0.006) (0.033) (0.015)distance (HQ) 0.001 0.006 −0.009 0.002 −0.009 0.002
(0.027) (0.019) (0.011) (0.002) (0.011) (0.004)com. lang. (HQ) 0.122b 0.073 0.010 −0.003 0.011 −0.004
(0.058) (0.046) (0.026) (0.005) (0.027) (0.010)
Observations 104,947 104,947 104,947 104,947 104,947 104,947R2 0.699 0.641 0.970 0.901 0.968 0.872Standard errors in (), clustered by origin-market dyads. Fixed effects at the brand-product and market-year-product dimensions included in each specification. HQvariables defined with respect to brand owner’s headquarter country. Significancelevels: 1% (a), 5% (b), and 10% (c).
mainly a cost effect with a 4–7% advantage.We can back out measures of brand level quality—defined as the destination- and
time-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.” There is already anempirical literature supporting the intuition that making higher quality brands requireshigher production costs.11 If the elasticity of production costs with respect to qualityexceeds one, then higher quality brands will have lower market shares and, presumably,lower profits.12 Figure 3 shows the relationship between brand fixed effects for appeal andcost for three categories under the Bertrand conduct assumption. Quality is expensive inthe wine industry with an elasticity of 0.7, but relatively cheap for carbonates, with anestimated elasticity of 0.26. Beer, as well as Spirits and Bottled Water (the last two shownin Figure 5 in the appendix) exhibit intermediate cost-quality elasticities.13 Since all the
11See, for example, Kugler and Verhoogen (2011) and Crozet et al. (2012).12The presumably can be removed under monopolistic competition. With oligopoly, profits and sales are
not strictly proportional but we conjecture they are still monotonically related.13Sorted by the cost of quality elasticity, the estimates for Bertrand and Cournot (respectively) are gener-
ally very close to each other: Wine (0.70, 0.71), Spirits (0.52, 0.51), Beer (0.45, 0.37), Carbonates (0.26, 0.21),
23
estimated cost of quality elasticities are well below one, we infer that quality is profitablein these industries.
Figure 3: How higher quality affects costs
(a) Wine (b) Beer (c) Carbonates
−4 −3 −2 −1 0 1 2
−3
−2
−1
01
2
log appeal of brand b
log
cost
of b
rand
b
Coefficient: 0.70
−3 −2 −1 0 1
−3
−2
−1
01
log appeal of brand b
log
cost
of b
rand
b
Coefficient: 0.45
−6 −4 −2 0
−3
−2
−1
01
23
log appeal of brand b
log
cost
of b
rand
b
Coefficient: 0.26
Each point is a brand fixed effect for appeal (Table 6, column 2) and its associated fixedeffect for cost (Table 6, column 3). Scatter plots for other goods in appendix.
7 Effects of multinational brand amalgamation on consumers
We want to quantify how much harm to competition has been caused by the process ofmultinational brand amalgamation. We are interested in calculating the cumulative im-pact on concentration and consumer surplus from the brand acquisitions by MNCs. Thereare many ways one could approach this. One way to do this would be to reverse all themergers that occurred over our period. As we only have a decade of data, this approachwould miss many important acquisitions occuring before 2007. Also it would includemergers of firms within the same country. To emphasize the cross-border aspect of brandamalgamation, we consider a scenario in which national governments forces MNCs todivest themselves of all their locally originating brands. Those brands are absorbed intonew created entity (e.g. “former AB-Inbev brands”, “former Heineken-owned brands”).Then we use the Bertrand and Cournot versions of the CES oligopoly model to predictmarket shares and markups in this counterfactual. This allows us to construct new con-centration indexes in the hypothetical divestment scenario. Moreover, we can examinethe reduction in the price index the new ownership arrangement would be expected to
and Bottled Water (0.24, 0.21).
24
generate.
7.1 Exact hat algebra for mergers (with potential costs and appeal changes)
The counterfactual stipulates a set of brand portfolios for each firm which we denote asF ′f . Firm market shares adjust to new ownership sets and the changes in brand marketshares entailed by rearranging ownership and therefore altering first-order conditionsfor pricing. So far as we know, this is the first application of exact hat algebra (EHA)to merger analysis. Given the very low information requirements (just market shares,prices and σ must be known) this approach seems attractive as compared to methods thatinvolve solving the full model and thus require data on the levels Abn and cbn which aregenerally unknown. With EHA, only changes Abn and cbn need to be specified and theycan be obtained from the appeal and cost regressions of the previous section.
The first (and last) step in the ownership change counterfactual is to aggregate up thenew brand market shares predicted by EHA to the level of firms. Initially we set s′bn = sbn
and sum up the shares of the brands in the new ownership sets, F ′f , to yield
S ′fn =∑b∈F ′f
Ibns′bn. (22)
Using data on initial firm market shares Sfn and an estimate of σ, we calculate the ini-tial εfn vector assuming Bertrand (equation 9) or Cournot (equation 10). Recalling theBertrand own-price elasticity is εfn = σ + (1− σ)Sfn, its proportional change in the coun-terfactual under Bertrand is
εfn = [σ + (1− σ)S ′fn]/εfn. (23)
Recalling the Cournot own-price elasticity is εfn = σ/[1 + (σ − 1)Sfn], its proportionalchange in the counterfactual under Cournot is
εfn = σ[1 + (σ − 1)S ′fn]−1/εfn. (24)
The adjustment of the firm-level markups is given by substituting the conduct-based ε
and ε into the following equation:
µfn =1
µfn
εfnεfn(εfnεfn − 1)
=εfn(εfn − 1)
(εfnεfn − 1). (25)
With these markup adjustments calculated, the final step is to determine the brand-level
25
market share changes. The main cause of brand-level market share changes is the adjust-ment of markups resulting from the change in ownership. However, the method allowsfor changes in the cost of supplying brand b to market n, denoted cbn, and appeal changesas Abn. For example, consumers might like a brand better, knowing that it has returned todomestic ownership. The proportional change in brand-level market share is given by
sbn =
(µfncbn
AbnPn
)1−σ
(26)
with the change in country’s price index under given by:
Pn =
(∑k
Iknskn(µknckn/Akn)1−σ
) 11−σ
. (27)
Multiplying the market share adjustments by initial levels give the s′bn = sbnsbn needed tore-aggregate firm-level market shares in equation (22). The algorithm then iterates untilthe vector of brand-level market share changes stabilize. The resulting sbn is the same asthe one obtained by solving for the equilibrium, sbn, before and after the friction changeand taking the ratio. The advantage is that it can be calculated without knowing the levelsof all the model’s parameters.
In our data there are other brands for whom we model as monopolistically competi-tive, charging µbn = σ/(σ − 1). Hence for the “other brands” µbn = 1. We do not observesbn for them but we know their aggregate share, which we denote son.
Pn =
(son +
∑k∈listedn
Iknskn(µknckn/Akn)1−σ
) 11−σ
. (28)
For the listed brands, markups adjust, as do potentially their costs and appeal followingownership changes. The listed brand market shares adjust according to equation (26) withthe price index change given by (28). For the other brands, we assume there are no costand appeal changes. Since markups are fixed as well, other brand market share evolvesaccording to son = P σ−1
n . The outcomes of the counterfactual we examine are the changesprice indexes and in market concentration. The counterfactual level of concentration is
H ′n =∑f
D′fn(S ′fn)2 (29)
In the current version, all the original firms remain in operation in market n but severalnew firms are added.
26
7.2 Consumer impact of local brand divestments in the beer industry
Figure 4 shows that the multinational brand amalgamation in the beer industry has hadstrong impacts on both concentration and consumer welfare. Reversing this process asin 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. (2016). These resultsare obtained by applying the model. They are conservative results in the sense that thedivested brands are allocated to single firms whereas in reality brand amalgamation oftentook the form of buying brands from multiple independent original owners. To assess themodel we plan to look directly at price changes following major acquisitions that occurredduring the time-frame of our data.
8 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 portfolio 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.
27
Figure 4: Effects of compelling foreign owners to divest domestic beer brands
(a) Conduct assumption: Bertrand
0.0 0.2 0.4 0.6 0.8
020
0040
0060
0080
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
CA
RU
FR
IT
ZA2016 dataCF divestment
(#) %change in price index
high concentration (EU guidelines)(−1)
(−1)
(0)
(−18)
(−1)
(0)
(−5)(−4)
(−3)(−5)(−9)
(−4)(−4)
(−1)
(−50)
(b) Conduct assumption: Cournot
0.0 0.2 0.4 0.6 0.8
020
0040
0060
0080
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
CA
RU
FR
IT
ZA2016 dataCF divestment
(#) %change in price index
high concentration (EU guidelines)(−5)
(−4)
(0)
(−37)
(−5)(0)
(−22)(−19)
(−11)(−15)(−25)
(−12)(−14)
(−4)
(−73)
28
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.
Crozet, M., K. Head, and T. Mayer (2012). Quality sorting and trade: Firm-level evidencefor french wine. The Review of Economic Studies 79(2), 609–644.
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.
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.
Friberg, R. and M. Ganslandt (2003). Bottled water—a case of pointless trade? DiscussionPaper 4145, Centre for Economic Policy Research.
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.
29
Hottman, C. J., S. J. Redding, and D. E. Weinstein (2016). 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.
Kugler, M. and E. Verhoogen (2011, 11). Prices, plant size, and product quality. The Reviewof Economic Studies 79(1), 307–339.
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.
Miller, N. H. and C. M. Weinberg (2017). Understanding the price effects of the miller-coors joint venture. American Economic Review 85(6), 1763–1791.
Miravete, E. J., K. Seim, and J. Thurk (2018). Market power and the Laffer curve? mimeo.
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.
A Additional decomposition results
Table 7 provides the decomposition for Cournot.
30
Table 7: Variance Decomposition with Cournot conduct
Appeal Scope Cost µ Disp. # obs.Category Hf Rf Uf Hf Rf Uf
All GMID firms, Cross-sectionalBeer 0.26 0.14 0.06 0.37 0.31 -0.01 0.03 -0.33 0.17 9144Water 0.27 0.23 0.20 0.16 0.08 0.12 -0.07 -0.09 0.10 8688Carbonates 0.26 0.16 0.11 0.34 0.20 -0.01 -0.04 -0.18 0.15 6073Spirits 0.32 0.28 0.18 0.33 0.00 -0.27 0.08 -0.05 0.11 18750Wine 0.41 0.17 0.19 0.23 0.05 0.02 -0.04 -0.09 0.07 11655Pooled 0.31 0.21 0.15 0.30 0.11 -0.08 0.02 -0.14 0.12 54310Low Income 0.26 0.24 0.20 0.28 0.13 -0.02 -0.04 -0.17 0.12 28252High Income 0.36 0.17 0.08 0.34 0.08 -0.15 0.10 -0.11 0.13 26058
All GMID firms, GrowthBeer 0.25 0.19 0.08 0.28 0.27 0.10 0.02 -0.30 0.12 7823Water 0.32 0.28 0.30 0.10 0.00 0.01 -0.02 -0.04 0.05 7504Carbonates 0.25 0.39 0.17 0.13 0.06 -0.01 -0.02 -0.04 0.08 5204Spirits 0.35 0.42 0.19 0.16 0.10 -0.24 -0.00 -0.03 0.05 16377Wine 0.42 0.24 0.35 0.11 0.11 -0.17 -0.06 -0.04 0.04 10278Pooled 0.32 0.31 0.21 0.16 0.12 -0.08 -0.01 -0.10 0.07 47186Low Income 0.28 0.36 0.22 0.15 0.07 -0.02 -0.01 -0.11 0.07 24473High Income 0.40 0.21 0.19 0.19 0.21 -0.19 -0.02 -0.07 0.07 22713
Firms with Sfn > 0.005, Cross-sectionalBeer 0.10 0.09 0.03 0.47 0.39 0.18 0.04 -0.54 0.24 5978Water 0.28 0.25 0.15 0.22 0.08 0.14 -0.11 -0.15 0.14 6888Carbonates 0.26 0.08 0.09 0.45 0.29 -0.02 -0.06 -0.29 0.20 4785Spirits 0.09 0.08 0.01 0.57 0.30 -0.25 0.11 -0.11 0.21 11148Wine 0.02 0.09 0.05 0.36 0.49 -0.05 0.12 -0.19 0.11 7138Pooled 0.15 0.11 0.06 0.44 0.30 -0.01 0.02 -0.27 0.19 35937Low Income 0.14 0.12 0.07 0.41 0.29 0.11 -0.01 -0.31 0.18 19914High Income 0.16 0.11 0.04 0.47 0.32 -0.16 0.07 -0.21 0.20 16023
Firms with Sfn > 0.005, GrowthBeer 0.03 0.12 0.05 0.44 0.48 0.21 0.06 -0.56 0.19 5033Water 0.33 0.23 0.21 0.16 0.02 0.10 -0.03 -0.09 0.08 5921Carbonates 0.20 0.26 0.19 0.31 0.14 -0.05 -0.05 -0.14 0.13 4094Spirits 0.22 0.23 0.12 0.33 0.19 -0.12 0.00 -0.09 0.12 9576Wine 0.46 0.24 0.25 0.20 -0.10 -0.22 0.25 -0.12 0.05 6193Pooled 0.20 0.20 0.13 0.32 0.22 0.03 0.04 -0.26 0.13 30817Low Income 0.18 0.19 0.17 0.29 0.17 0.17 0.00 -0.29 0.12 17028High Income 0.24 0.21 0.06 0.38 0.33 -0.28 0.12 -0.21 0.15 13789Hf is the set of headquarter brands,Rf is the set of brands acquired in other countries andUf is the set of brands where origin is uncertain. Conduct is Cournot.
31
B Regression results by product group
Table 8: Explaining appeal, cost, and markups for Beerln sbn lnAbn ln cbn lnµbn ln cbn lnµbn
Bertrand Cournothome 1.545a 0.207a −0.336a 0.075a −0.476a 0.215a
(0.173) (0.067) (0.046) (0.018) (0.056) (0.040)distance −0.236a −0.067a 0.017 −0.013 0.042c −0.037b
(0.058) (0.025) (0.019) (0.008) (0.023) (0.015)common language 0.381a 0.039 −0.088b 0.012 −0.112b 0.035
(0.127) (0.053) (0.039) (0.015) (0.046) (0.031)home (HQ) 0.013 0.022 0.003 0.015 −0.033 0.051
(0.157) (0.059) (0.045) (0.017) (0.055) (0.039)distance (HQ) −0.091b −0.016 0.006 0.005 0.009 0.002
(0.040) (0.018) (0.014) (0.005) (0.016) (0.010)com. lang. (HQ) −0.213b −0.031 0.082b −0.049a 0.135a −0.101a
(0.101) (0.044) (0.035) (0.017) (0.047) (0.034)
Observations 24,913 24,913 24,913 24,913 24,913 24,913R2 0.749 0.768 0.927 0.885 0.918 0.870
32
Table 9: Explaining appeal, cost, and markups for Spiritsln sbn lnAbn ln cbn lnµbn ln cbn lnµbn
Bertrand Cournothome 0.534a 0.241c −0.061 0.005 −0.068 0.013
(0.173) (0.142) (0.068) (0.004) (0.068) (0.010)distance −0.085c 0.044 0.092a −0.002 0.095a −0.004
(0.050) (0.040) (0.020) (0.001) (0.020) (0.003)common language −0.036 −0.106 −0.086c −0.0001 −0.086c 0.00003
(0.099) (0.089) (0.047) (0.002) (0.047) (0.006)home (HQ) 0.383a 0.072 −0.178a 0.037a −0.235a 0.094a
(0.144) (0.114) (0.051) (0.006) (0.050) (0.014)distance (HQ) 0.076c 0.015 −0.031c 0.003b −0.036c 0.008b
(0.043) (0.036) (0.019) (0.001) (0.019) (0.003)com. lang. (HQ) 0.182b 0.084 −0.021 0.003 −0.027 0.010
(0.080) (0.070) (0.039) (0.004) (0.039) (0.009)
Observations 40,336 40,336 40,336 40,336 40,336 40,336R2 0.546 0.572 0.810 0.868 0.811 0.839
Table 10: Explaining appeal, cost, and markups for Wineln sbn lnAbn ln cbn lnµbn ln cbn lnµbn
Bertrand Cournothome 0.468 −0.016 −0.130 −0.001 −0.127 −0.003
(0.291) (0.145) (0.095) (0.002) (0.095) (0.010)distance −0.224a −0.057 −0.001 −0.002b 0.006 −0.008b
(0.077) (0.040) (0.028) (0.001) (0.029) (0.003)common language 0.295 0.054 −0.020 0.002 −0.030 0.012
(0.199) (0.092) (0.064) (0.002) (0.064) (0.010)home (HQ) 0.355 0.032 −0.060 0.006a −0.084 0.030a
(0.236) (0.102) (0.072) (0.002) (0.074) (0.010)distance (HQ) 0.003 0.014 0.014 −0.001 0.016 −0.002
(0.063) (0.028) (0.020) (0.001) (0.020) (0.004)com. lang. (HQ) 0.158 0.030 −0.008 −0.0004 −0.006 −0.002
(0.176) (0.073) (0.053) (0.002) (0.054) (0.010)
Observations 15,125 15,125 15,125 15,125 15,125 15,125R2 0.766 0.855 0.966 0.952 0.966 0.942
33
Table 11: Explaining appeal, cost, and markups for Bottled Waterln sbn lnAbn ln cbn lnµbn ln cbn lnµbn
Bertrand Cournothome 1.565a 0.824a −0.216b 0.062a −0.285a 0.131a
(0.233) (0.151) (0.089) (0.018) (0.097) (0.037)distance −0.191c −0.074 0.045 0.0004 0.047 −0.002
(0.105) (0.067) (0.038) (0.006) (0.039) (0.014)common language 0.019 −0.087 −0.097 −0.002 −0.090 −0.008
(0.225) (0.134) (0.090) (0.012) (0.097) (0.027)home (HQ) 0.396 0.397b 0.150c −0.001 0.151 −0.001
(0.267) (0.179) (0.089) (0.016) (0.099) (0.035)distance (HQ) 0.028 0.094 0.083b −0.007 0.091b −0.015
(0.102) (0.066) (0.036) (0.006) (0.039) (0.013)com. lang. (HQ) 0.613a 0.455a 0.045 0.027b 0.010 0.062b
(0.234) (0.146) (0.082) (0.013) (0.088) (0.029)
Observations 10,443 10,443 10,443 10,443 10,443 10,443R2 0.798 0.775 0.924 0.845 0.921 0.839
Table 12: Explaining appeal, cost, and markups for Carbonatesln sbn lnAbn ln cbn lnµbn ln cbn lnµbn
Bertrand Cournothome 0.924a 0.471a −0.147b 0.058c −0.219b 0.130b
(0.275) (0.174) (0.073) (0.034) (0.086) (0.063)distance −0.279a −0.118b 0.033 0.018 0.008 0.043c
(0.093) (0.056) (0.029) (0.013) (0.034) (0.023)common language 0.820a 0.450a −0.070 0.024 −0.087 0.041
(0.199) (0.148) (0.056) (0.030) (0.063) (0.052)home (HQ) −0.090 −0.049 0.013 −0.007 0.028 −0.023
(0.292) (0.173) (0.088) (0.032) (0.099) (0.060)distance (HQ) −0.039 −0.044 −0.021 0.001 −0.020 −0.0001
(0.083) (0.054) (0.032) (0.015) (0.037) (0.026)com. lang. (HQ) −0.106 −0.053 −0.048 0.059c −0.108 0.119b
(0.222) (0.141) (0.065) (0.030) (0.074) (0.053)
Observations 14,130 14,130 14,130 14,130 14,130 14,130R2 0.691 0.666 0.827 0.767 0.817 0.791
34
Figure 5: How higher quality affects costs (other products)
(a) Spirits (b) Bottled Water
−10 −5 0
−8
−6
−4
−2
02
log appeal of brand b
log
cost
of b
rand
b
Coefficient: 0.52
−5 −4 −3 −2 −1 0 1 2
−3
−2
−1
01
log appeal of brand b
log
cost
of b
rand
bCoefficient: 0.24
Each point is a brand fixed effect for appeal (Table 6, column 2) and its associated fixedeffect for cost (Table 6, column 3). Scatter plots for other goods in appendix.
35
C Concentration and markups
A classical question in IO is how equilibrium markups and overall welfare vary withrespect to market concentration, usually measured as a Herfindhal index. In the CESoligopoly case, this is best seen with the market-level Lerner index aggregating overLerner indices of the active firms. In market n and under Cournot:
Ln ≡∑f
SfnLfn =∑f
SfnεCfn
=1
σ+
(σ − 1)
σHn =LDS + (1− LDS)Hn
=Hn + LDS(1−Hn). (30)
This shows that ∂Ln∂Hn
> 0 and ∂Ln∂LDS > 0. In other words, the aggregate markup under
Cournot in country n is systematically related to the Herfindhal index (and to the aggre-gate markup that would be relevant under Dixit-Stiglitz, LDS ≡ 1/σ).
With Bertrand conduct, the market-share weighted arithmetic mean of Lfn does notsimplify to a neat relationship with the Herfindahl index. Numerically, we can show thereis a convex relationship. Since the arithmetic mean is only one of many ways to aggregatefirm-level markups, an interesting result is that the weighted harmonic mean, denotedL]n, has a very simple relationship with Hn.
L]n ≡1∑
Sfn/Lfn=
1∑SfnεBfn
=1
σ − (σ − 1)Hn
=LDS
1− (1− LDS)Hn
. (31)
As with the Cournot case, as Hn → 0 the aggregate markup with Bertrand oligopolygoes to the monopolistic competition limit of LDS = 1/σ. And as Hn → 1, the Lernermarkup goes to one. Also as with Cournot, holding σ constant any force that increasesconcentration also increases the aggregate markup. This would include a decrease inhome bias or an increase in the variance of appeal or marginal costs.
D Aggregate welfare changes
The counterfactuals in the main text evaluate only the impact on consumers. The methodcan also be used to obtain the effect on changes in profits to get to the full partial equi-librium welfare effect of mergers. In terms of welfare analysis, we can write the resultingchanges for both consumer surplus and firms’ profits. For the consumer, the compen-sating variation CV is the difference in expenditure for initial (0) and counterfactual (1)
36
prices, holding utility constant at the initial level.
CVn = Yn(1− Pn) with Pn =
(∑k
Iknskn(µknckn/Akn)1−σ
) 11−σ
. (32)
The algorithm described in section 7.1 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, Ff .
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∈Ff
Ibn(π′bn − πbn) = Yn∑b∈Ff
Ibn(sbnLbn − 1)sbnLbn. (33)
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, (34)
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∈Ff
Ibn(sbn/εfn − 1)sbn/εfn. (35)
Finally, the firm-level profit change is given by:
∆Πf =∑n
∆Πfn. (36)
37
