shapiro (1997)

A R T I C L E S • A N D • F E A T U R E S

S P R I N G 1 9 9 6 · 2 3

T he antitrust treatment of hori-zontal mergers by the JusticeDepartment and the FederalTrade Commission is one of

the most well developed and closely scru-tinized areas of antitrust law. Theenforcement agencies have extraordinaryexperience reviewing mergers and themerger bar is no less sophisticated. Frommy perspective as an antitrust economist,this sophistication permits mergerenforcement to be at the cutting edgewhen it comes to incorporating econom-ic learning into competition policy.

The 1992 DOJ and FTC MergerGuidelines have placed important atten-tion on the “unilateral effects” of a merg-er, i.e., the tendency of a horizontal merg-er to lead to higher prices simply byvirtue of the fact that the merger willeliminate the direct competition betweenthe two merging firms, even if all otherfirms in the market continue to competeindependently. Unilateral effects are con-trasted to “coordinated effects,” i.e., thedanger that the merger will lead to collu-sion between the merged entity and itsremaining rivals.

This article discusses some of themethods used by the Antitrust Division toanalyze unilateral effects in mergersinvolving differentiated products. Whilethe methods outlined here do not replacethe standard steps of defining marketsand measuring market shares and con-

centration, they can significantly supple-ment those steps.

To place the topic in context, it isinstructive to compare mergers withhomogeneous products to those involvingdifferentiated products. For homoge-neous products, the traditional structuralapproach of defining markets and mea-suring market shares and market concen-tration has deep roots, along with a richempirical tradition linking market struc-ture to performance. In these markets, itis both natural and appropriate to countup each firm’s unit or dollar sales, orcapacity, to measure market shares, andto make inferences about a merger’seffects based on market structure, includ-ing the HHI of market concentration. Thedanger of collusion is surely related tomarket concentration (although quanti-fying this relationship is difficult), andeconomists’ primary model of non-coop-erative oligopolistic competition amongmanufacturers of homogeneous goodsrelates market structure to performance.1

Although economists continue to debatethe empirical relationship between mar-ket structure and performance, thereexists a solid foundation for using marketstructure prominently in evaluating hori-zontal mergers involving homogeneousgoods.2

This traditional structural approachtowards merger policy, which dates backto the 1960s but has been refined as justdescribed, is less attractive for differenti-ated products. When products are highlydifferentiated, concerns about coordinat-ed effects may be secondary to concernsabout unilateral effects. And, to assessunilateral effects most accurately, it ishighly desirable to go beyond industryconcentration measures to look directly at the extent of competition between themerging brands. This is especially true ifcompetition is “localized,” i.e., if somebrands are especially close substitutes forother brands due to their product charac-teristics or image. To put this differently,at the Antitrust Division we must con-

cern ourselves with the prospect that theprices for one or more brands sold by themerging parties will rise after the merger,even if prices do not uniformly risethroughout the relevant market. Suchconcerns are not present, or are far lesspronounced, in markets for homogeneousproducts.

The danger that a merger of Brands Aand B will cause anticompetitive priceincreases for one or both of these brandsis greatest if the merging brands are“close,” in the sense that many customersusing one brand consider the other brandtheir second choice. Ultimately, unilateralanticompetitive effects are based on thefollowing logic: As the price of Brand Arises, some customers will shift fromBrand A to Brand B. Prior to the merger,these customers would be lost to the firmowning Brand A. After the merger, thissame firm owns Brand B and thus doesnot lose these customers. As a result, theprice increase is more profitable to themerged entity.

To explicate this logic, consider amerger between Brands A and B. Thelikely post-merger price increase forBrand A is driven by two variables, eachof which we have a good chance ofobserving with a proper investigation.The first I call the Diversion Ratio fromBrand A to B. This Diversion Ratio is theanswer to the following question: If theprice of Brand A were to rise, what frac-tion of the customers leaving Brand Awould switch to Brand B? The secondvariable is the Gross Margin — the per-centage gap between price and incre-mental cost — for Brand B.3 Roughlyspeaking, a valuable index of the poten-tial anticompetitive unilateral effects isobtained by multiplying the DiversionRatio by the Gross Margin. Any dangerof a unilateral price increase may be alle-viated by product repositioning, entry, orefficiencies. Nonetheless, the DiversionRatio and the Gross Margin are the keyvariables in the demand-side portion ofthe analysis.

Mergers with Differentiated Productsby Carl Shapiro

Carl Shapiro is the Deputy AssistantAttorney General in charge of economicsat the Antitrust Division, U.S. Departmentof Justice. He is currently on leave from hisposition as the Transamerica Professor ofBusiness Strategy at the Haas School ofBusiness and Professor of Economics inthe Economics Department at theUniversity of California at Berkeley. Dr.Shapiro thanks Joseph Kattan, GeorgeRozanski, Steve Salop, Greg Werden, andRobert Willig for valuable comments onthis paper.


2 4 · A N T I T R U S T

Monopolistic CompetitionTo a greater or lesser degree, virtually allmarkets involve some element of productdifferentiation.4 Even in a classic homo-geneous-goods market — such as themarket for an agricultural commodity orfor a specific chemical compound — pro-ducers often attempt to differentiatethemselves based on product quality, reliability, or customer service.

My emphasis here is on marketswhere the brands are distinct in importantand long-lasting ways. It is helpful tokeep two categories in mind: (1) brandedconsumer products, where each brand ofpens, bread, facial tissue, computer soft-ware, or cereal, is distinct; and (2) phys-ical facilities that distribute or delivergoods or services, such as supermarkets,department stores, branch banks, or hos-pitals, where the differentiation is basedon location. These two broad categoriescorrespond to the familiar tasks of prod-uct and geographic market definition.

Unilateral Competitive Effects The predominant approach taken byeconomists studying markets with differ-entiated products is to model the firms as independently setting the prices ofeach of their brands. As usual in eco-nomics, we assume that each firm seeksto maximize its own profits. This methodof analysis fits perfectly with the “uni-lateral effects” portion of the MergerGuidelines (§ 2.2).

It is fair to say that economic analysisof differentiated-products mergers at theDivision typically focuses on unilateraleffects, unless there are structural factorsfacilitating collusion following the merg-er or there is a history of collusion in theindustry. This emphasis represents a sig-nificant shift in a fairly short period oftime.

Economic analysis regarding unilat-eral effects is more amenable to quantifi-cation than is economic analysis of thedangers of collusion. Ultimately, we aretrying to measure the added incentive toraise price caused by the merger.Employing a combination of game-theo-retic and econometric methods, we nowhave the capability to estimate consumerdemand using industry data and, basedon these demand estimates, to derive spe-cific predictions regarding post-mergerprices.5 This contrasts somewhat with

the analysis of the dangers of collusion.While we are fairly confident in listingand analyzing factors that facilitate orhinder collusion, including market struc-ture, there is no single accepted methodof quantifying the increased likelihood of collusion attendant to a merger.

Our ability to predict unilateral effectsbased on demand patterns is only as goodas the available data. Very often we mustmake do with qualitative evidence orrather rough estimates. However, evenwhen the data are limited, the theory ofmonopolistic competition can providesome very helpful predictions based onpremerger Gross Margins and marketshares, which often can be measuredusing data routinely collected during thesecond request process. These predictionswill need to be checked against the viewsof industry participants, company docu-ments, and other information sources, butthey do constitute a vital part of mergeranalysis.

A final caveat: the analysis belowapplies when the firms independently setuniform prices for their branded prod-ucts. To the extent that firms engage inprice negotiations on a customer-by-cus-tomer basis or engage in other forms ofprice discrimination, the analysis must be modified or replaced with anotherapproach.

Estimating Post-Merger Prices:Four StepsEconomists in the agencies and thoseworking for private parties largely agreeon what to look for to estimate unilateralcompetitive effects, although they maydiffer in specific implementation steps.We seek to estimate the post-mergerBertrand equilibrium in prices, account-ing for the new market structure in whichsome brands are jointly owned that hadpreviously been independent and for thenew cost structure of the merged entity.6

The analytical steps in this exercisecan be described succinctly in a mannerconsistent with the methodology of theMerger Guidelines: If a significant pro-portion of consumers considers the merg-ing firms’ products to be their first andsecond choices (at premerger prices),then the merged entity will have an incen-tive to impose a nontrivial price increasefollowing the merger. Unless productrepositioning or entry would render such

a price increase unprofitable, and unlesssynergies imply that the price increasewill not in fact raise profits, the mergerwill injure consumers and be anticom-petitive.

There are many valid ways to carryout this analysis. As much as anything,the method chosen depends upon theavailable data. Generally, however, I findit useful to structure my thinking aboutunilateral effects in differentiated productmarkets in terms of the following foursteps.7 Consider a merger between brandsA and B, and focus attention on the dan-ger that the price of Brand A will be ele-vated after the merger:8

(1) Consider a price increase forBrand A of, say, 10 percent. Try to mea-sure what fraction of the sales lost byBrand A due to this price increase wouldbe captured by Brand B. I call this frac-tion the Diversion Ratio (from Brand A toBrand B).

(2) Based on premerger Gross Mar-gins and the estimated Diversion Ratio,calculate the post-merger price increase,assuming no synergies or rival supplyresponses.

(3) Try to account for any likely andtimely changes in prices or product offer-ings by nonmerging parties, includingproduct repositioning and entry.

(4) If there are credible and docu-mented synergies that lower marginalcosts, reduce the predicted post-mergerprices accordingly. This step can lead toa predicted price decrease.

The first two steps capture thedemand-side analysis; the third step isthe supply-side analysis; the final stepaccounts for possible efficiencies. If thesesteps indicate that the merged entitywould find it optimal to impose a signif-icant price increase following the merger,then the merger is very likely to be anti-competitive.

The Diversion RatioThe concept underlying the DiversionRatio is easily comprehensible: “If youraise your price, what fraction of yourlost customers will turn to your rival(now merger partner)?”

The Diversion Ratio is a close cousinof the cross-elasticity of demand betweenthe merging brands.9 If the unit sales ofthe merging brands are equal prior to themerger, the Diversion Ratio from Brand


S P R I N G 1 9 9 6 · 2 5

A to Brand B is equal to the cross elas-ticity of demand from Brand A to BrandB, divided by the own-price elasticity ofdemand for Brand A.10 Suppose, forexample, that Brand A has an own-priceelasticity of demand of 2.0 (i.e., a 1 per-cent increase in the price of Brand Acauses unit sales to decline by 2 percent),and that the cross-price elasticity ofdemand from Brand A to Brand B is 0.5.If the two brands’ premerger unit salesare equal, then the Diversion Ratio fromBrand A to Brand B is (0.5)/(2.0) or 25percent. Stated differently, one-quarter ofthe unit sales lost by Brand A if its pricerises are captured by Brand B.

In some cases, the Diversion Ratiofrom Brand A to Brand B will be closelylinked to Brand B’s market share. In par-ticular, if all sales lost by Brand A arecaptured by other brands in the market,and if all brands are “equally close” toeach other, then the Diversion Ratio fromBrand A to Brand B may be stated assB/(1-sA), where sA and sB are the brands’respective market shares.11 In the morerealistic situation where some customerssubstituting away from Brand A switch toproducts outside the market entirely, thisDiversion Ratio will be proportionatelylower. For example, if 20 percent of thecustomers lost by Brand A leave the mar-ket entirely, the Diversion Ratio fromBrand A to Brand B will instead be(0.8)sB/(1-sA).

Since the Diversion Ratio plays a cru-cial role in this analysis, in differentiated-product mergers the Antitrust Divisionwill invariably want to know the bestestimate of the Diversion Ratio based onthe available evidence.

Estimating Unilateral Competitive Effects in PracticeSteps 1 and 2: Demand-Side Analysis. Inpractice, the demand-side analysisdepends heavily on the availability ofdata.12 Two recent merger cases at theAntitrust Division that led to consentdecrees illustrate what can be done withdetailed data.

The merger of Interstate BakeriesCorporation and Continental BakingCompany involved the first and thirdlargest bakers of fresh bread in the United States.13 The Division concludedthat the deal would substantially lessencompetition in the production and sale

of white pan bread in five regional markets. The proposed Final Judgmentorders Interstate and Continental to divestcertain white bread brands in each geographic market. Our competitive con-cerns, as well as the ultimate relief, weregreatly informed by our economic analy-sis of the merger’s anticompetitive uni-lateral effects.

The Kimberly-Clark acquisition ofScott Paper Company posed a threat tocompetition in the markets for babywipes and facial tissues.14 Again, ouranalysis was significantly informed byeconometric analysis of the demand forthe various brands in these markets, fromwhich we were able to estimate likelyprice increases following the acquisition.

In both of these merger investigations,we had access to excellent data on pricesand unit sales derived from checkoutscanners at retail locations. These dataprovided information regarding competi-tion both from other brands and from private label products. In the bakeriesmerger, we focused on the brands of pre-mium white bread sold by Continentaland Interstate. In the baby wipes market,we focused on competition betweenKimberly-Clark’s Huggies brand of babywipes and Scott’s two brands of babywipes, Baby Fresh and Wash-a-ByeBaby; together the merging parties con-trolled over 50 percent of the sales in this$500 million market. In the facial tissuemarket, we estimated the cross-elasticitybetween Kimberly-Clark’s Kleenexbrand and Scott’s Scotties brand; togeth-er, the two parties controlled nearly 60percent of the sales in this $1.34 billionmarket.

In these two mergers we were able,with considerable work and making var-ious assumptions about the structure ofdemand, to estimate a complete model ofdemand for the various brands. Thesemethods subsume the Diversion Ratioconcept I stressed above. The calibratedmodel of consumer demand derived fromthe supermarket scanner data was thenused, in a high-tech version of Steps 1and 2 above, to predict the likely post-merger price increases for the variousmerging brands.15

In this prediction exercise, weassumed that all firms in the industry setprices independently after the merger tomaximize profits. The predictions of the

model at this point do not account forproduct repositioning, entry, or synergies.In all cases, however, this demand-sideanalysis did take account of competitionfrom private-label products as well asbranded goods. In the bread merger, thecomputer model predicted price increas-es in the 5 to 15 percent range forContinental’s and Interstate’s premiumwhite pan breads in the Los Angeles andChicago areas. In baby wipes, we esti-mated that there would have been sub-stantial price increases following themerger. Price increases were also pre-dicted in facial tissue, especially for thesmaller Scotties brand.

While econometricians dream aboutthis type of “high-tech” analysis, in real-ity the data are rarely available to do thistype of full-blown simulation analysiswith assurance. If econometric estima-tion of elasticities based on transactionsdata is not possible, relevant consumersurvey data may still be available todirectly estimate the Diversion Ratio.Survey data are not as good as actualtransactions data, but can still be reliableif valid sampling procedures were usedand if the results are not overly sensitiveto the framing of the questions. If reliablesurvey data are also unavailable, it stillmay be possible to use company docu-ments and other qualitative informationregarding consumers’ brand preferencesto estimate Diversion Ratios.

The firm’s market shares can be veryhelpful in estimating Diversion Ratios ifnone of the brands in the market are espe-cially “close” to or “distant” from eachother. As noted above, Diversion Ratioswill be proportional to market shares inthis case. For example, consider such amarket in which Brand A has a 25 per-cent share and Brand B has a 15 percentshare. Suppose also that very few cus-tomers of Brand A would reduce theiroverall purchases in the market if BrandA were to raise its price; instead thesecustomers would by and large pickamong the other brands. In this case, theDiversion Ratio from Brand A to BrandB is 20 percent.16 The Diversion Ratiowill be lower, to the extent that some ofBrand A’s customers reduce their totalpurchases in the market when the price ofBrand A rises. If half of the customersdropping Brand A leave the market alto-gether, or if customers switching to other


2 6 · A N T I T R U S T

brands in the market reduce their pur-chases by half when switching away fromtheir favorite Brand, the Diversion Ratiofrom Brand A to Brand B will only be 10percent.

Under this analysis, if the mergingbrands are similar in characteristics, or ifthe merging brands have large shareswithin a broader product category, theDiversion Ratio is likely to be high. Notein particular that the Diversion Ratio islikely to be high for a brand that is merg-ing with a dominant brand: the large mar-ket share of the dominant brand makes itlikely that customers switching awayfrom the smaller brand will divert to thedominant brand rather than elsewhere.On the other hand, if the merging brandsare usually sold to different types of con-sumers, or through different channels, orif consumers’ preferences are such thatthey can easily substitute to a broad rangeof products (e.g., gifts instead of premiumpens, or breakfast foods instead of ready-to-eat cereals), the Diversion Ratio islikely to be lower, other things equal.

Merely asserting that there are numer-ous products to which consumers couldsubstitute, and thus the Diversion Ratiomust be low, ignores the importance ofdiverse consumer preferences and doesnot replace this step of the analysis. Noris a merger immunized merely becausethe merging brands are not next-closestsubstitutes, as some parties claim, anymore than a merger is immunized mere-ly because the merged entity still facessome post-merger competition.

To complete the demand-side analy-sis, the estimated Diversion Ratio can beused, along with premerger GrossMargins and perhaps other industry mea-sures, to give a rough prediction of thepost-merger price increases for the merg-ing brands. The principle here is that high Gross Margins and high DiversionRatios suggest large post-merger priceincreases. The tricky part is that the actu-al calculation of the post-merger priceincrease depends upon the specific shapeassumed for the demand curve.

A simple formula can be derived forthe post-merger price increase if one iswilling to assume that consumer demandfunctions exhibit constant elasticity overthe relevant range of prices. Very oftenwhen economists estimate demand usingdata, they employ such constant-elastic-

ity demand functions. With constant-elas-ticity demand, and assuming that the twomerging brands are symmetric prior tothe merger, the merged entity’s profit-maximizing percentage price increase ismD/(1–m–D).17 Here m is the premergerGross Margin and D is the DiversionRatio between the two merging brands.

For example, suppose that the pre-merger price is $100, and the cost perunit is $60, so the premerger markup, m, is 40 percent, not uncommon at all for differentiated products. If we assume

a Diversion Ratio of 0.2 (i.e., 20 percentof the sales lost when the price of Brand A goes up are captured by Brand B), then the optimal post-mergerprice increase in percentage terms is(0.4)*(0.2)/(1–0.4–0.2) = 0.2: a 20 per-cent price increase would maximize profits.

This formula must be used with greatcaution because it relies on several strongassumptions, as I have noted. To theextent that the elasticity of demand for abrand rises as the price of that brand rises,

Diversion Ratio Example

The importance of the Diversion Ratio and the Gross Margin can be seen with a fewnumerical examples. Consider a situation in which Brands A and B each have pre-merger prices of $100 and premerger unit sales of 1000. Suppose that each brand has

a marginal cost of $60, so the premerger Gross Margin is ($100–$60)/$100 or 40%. Priorto the merger, Brand A makes profits (often, contribution to fixed costs) of $40 per unit times1000 units, or $40,000. Consistent with premerger optimal pricing by Brand A, suppose thata 10% price increase by Brand A would lead to a 25% reduction in sales, to 750 units. At thishigher price, Brand A could earn profits of $50 per unit times 750 units, or $37,500. Sincethis is less than the $40,000 figure, prior to the merger Firm A did not find it profitable to setthe higher price.

How would a merger between Brands A and B change this calculation? Suppose that theDiversion Ratio from Brand A to Brand B is 30%. In other words, of the 250 units lost by BrandA due to the price increase, 30%, or 75 units, are diverted to Brand B. The merged entity wouldtake into account the additional profits earned by Brand B from these customers when con-sidering raising prices from $100 to $110. Assuming the price of Brand B also rises to $110,these 75 diverted sales generate profits of $50 per unit, or $3750 in total. Adding these to the$37,500 from the premerger calculation, the post-merger profits per brand at the elevatedprices are $41,250. The merger has made it profitable to raise prices 10% above premergerlevels. In this example, a 10% price increase will be profitable after the merger if and only ifthe Diversion Ratio is at least 20%.*

Compare this example to one in which the premerger Gross Margin is smaller. Specifically,suppose now that the marginal cost is $80, so the premerger Gross Margin is only($100–$80)/$100 or 20%. The premerger profits on Brand A are now only $20 per unit or$20,000. Now, consistent with optimal premerger pricing, suppose that a 10% price increasewould result in a 50% loss of sales, to 500 units. Prior to the merger, a 10% price increasewould reduce profits to $30 per unit times 500 units or $15,000, some $5000 less than couldbe earned at a price of $100.

What about after the merger? Now a 30% Diversion Ratio would mean 30% of the 500lost units, or 150 units, would be captured by Brand B; at a $30 profit margin per unit, thisadds $4500 in profits. But this $4500 of extra profit is still not enough to make up for the lossof $5000 in profits on Brand A. With the smaller Gross Margin in this example, a largerDiversion Ratio would be necessary to make a 10% price increase profitable.

*In these examples, I ask only whether a 10% price increase would generate more profits than premerger prices.Below, I discuss how to calculate the optimal post-merger price increase. Fortunately, with linear demand, themaximal profitable price increase is exactly twice the optimal price increase. So, if a 10% price increase is prof-itable, it will be optimal to raise prices at least 5%. Throughout this article, I use 10% price increases purely forillustrative purposes.


S P R I N G 1 9 9 6 · 2 7

the constant-elasticity-of-demand calcu-lations will overestimate the post-mergerprice increases. This overestimation canbe significant, especially if the formula asstated generates a large percentage priceincrease.18 Certainly it is desirable to findsupport for the constant-elasticityassumption in documents, in the testimo-ny of industry participants, or in the databefore using this formula or variants on it.

If demand instead takes a linear form,the elasticity rises as the price rises, mak-ing the optimal post-merger priceincrease smaller. In this case, an alterna-tive formula can be derived: the optimalpost-merger percentage price increasewith linear demand (again with premerg-er symmetry between the two brands) isgiven by mD/2(1–D). This formula isquite different from the earlier one. Usingthe same numerical example as above,with a premerger Gross Margin of 40 per-cent and a Diversion Ratio of 20 percent,the post-merger price increase would be“only” 5 percent.19

I am keenly aware that the two for-mulas presented above give quite differ-ent predictions. This observation does nottake away from my main point — thatDiversion Ratios and Gross Margins arekey variables to explore in a mergerinvestigation involving differentiatedproducts — but should serve as a warn-ing. The fact is, the profit-maximizingpost-merger price increase is sensitive toconsumers’ ability to substitute awayfrom the merging brands as prices rise.20

It is simply not possible, and one shouldnot expect, to fully predict price changeson the basis of two numbers, theDiversion Ratio and the Gross Margin,alone. One way or another, however, theDiversion Ratio (or its close cousin) andthe Gross Margin will be an integral partof the analysis.

Finally, even if data are limited andprecise predictions of post-merger priceincreases are difficult to make with con-fidence, these formulas still are a usefulstarting point in gaining a sense of thelikely magnitude of any post-mergerprice increase, if full-scale demand esti-mation is not possible.21 And the numer-ator in both of these expressions, mD, i.e.,the product of the Gross Margin and theDiversion Ratio, is a very useful quickguide to the likely anticompetitive dan-gers, even if we cannot predict the

post-merger price increase confidentlybased on these two variables alone.

In some cases, we can measure GrossMargins quite well, but may be uncertainabout the Diversion Ratios. In this situa-tion we can use either of the two formu-las above to ask how large the DiversionRatio must be in order for the optimalpost-merger price to be at least 10 percent(say) above premerger levels. Using thelinear demand formula just given, thepost-merger price increase will be at least 10 percent if and only if theDiversion Ratio is at least 0.2/(m+0.2). Ifthe premerger Gross Margin is 40 per-cent (m=0.4), prices will go up at least 10 percent if the Diversion Ratio is atleast one-third. Again, the larger theGross Margins, the lower the DiversionRatio necessary to raise anticompetitive concerns.

Steps 1 and 2 invariably will lead tothe interim prediction that prices will riseafter the merger, if indeed Brands A andB compete with each other. After all, amerger between rival brands does elimi-nate competition between those brands,which in and of itself leads to higherprices.22 But we do not condemn all hor-izontal mergers, of course. To begin with,we recognize that this incentive to raiseprice is very slight for some mergers,even horizontal ones. Steps 1 and 2should detect this, in the form of a verysmall predicted price increase. But thereare two more important reasons why hor-izontal mergers often are not anticom-petitive. First, the true price increase maybe far smaller than predicted by Steps 1and 2, or negligible, because rivals mayrespond to defeat any price increase.Second, the merger may reduce costs.Steps 3 and 4 are where these crucial con-siderations come into play.

Step 3: Product Repositioning andEntry. If the merging brands are “close”in attributes, the Diversion Ratio is like-ly to be high, and Step 2 will suggest asignificant price increase. In preciselythis situation, however, it may well payfor a rival firm to reposition its brand, orintroduce a new brand, closer to themerging brands. And this threat couldwell deter the price increase in the firstplace. Alternatively, a de novo entrantcould locate its brand near Brands A andB if prices of these brands were abovecompetitive levels. Very often in differ-

entiated-product markets, brands enterand exit with some regularity, and exist-ing products may be repositioned eitherthrough design changes or revised mar-keting strategies. As a general rule, the“farther” a brand must be moved to com-pete more effectively with the mergingbrands, the less likely it is that such amove would in fact occur in response toa post-merger price increase. As notedgenerally in the Merger Guidelines, thegreater the sunk costs associated withproduct entry or repositioning, and thelonger such supply responses take, theless likely they are to deter or defeat ananticompetitive price increase.

Notwithstanding the foregoing, rivals’responses do not necessarily reduce theprofitability of a post-merger priceincrease. Game-theoretic analyses ofpricing competition with differentiatedproducts indicate that rivals will typical-ly find it optimal to raise their prices inresponse to higher prices set by the merg-ing firms. Accounting for these accom-modating responses tends to increase, notdecrease, the predicted post-merger priceincrease.

Merging parties in consumer-goodsindustries may be tempted to argue thatbrand name is unimportant, but theyshould be cautious in doing so. Suchclaims are not credible if the partiesthemselves have made substantial invest-ments in brand equity, or if the deal priceitself reflects substantial brand equity.Attempts to downplay the importance ofbrand names are particularly problemat-ic if new brands historically have found itdifficult to gain a secure foothold in themarket.

In the recent bread merger theDivision concluded that brand nameswere very important. The evidence alsoshowed that a brand that achieved successin one region might not meet with accep-tance in another area. Furthermore, newentry required significant sunk invest-ments in brand promotion, as well asinvestments in a route delivery network.For these reasons and more our concernswere not alleviated by the prospect ofproduct repositioning or entry.

In the facial tissue and baby wipesmarkets, we also found that entry takessignificant time and expense. The babywipes market had unique aspects thatwarranted an especially close look at the


2 8 · A N T I T R U S T

dynamics of entry: the most likelyentrants appeared to be firms with well-established brand names in related con-sumer products, and entry is surely madeeasier by the fact that there is relativelyrapid turnover in demand in this market,since individual babies, and even parentswith two or more children, soon outgrowthe need for wipes. Nonetheless, entryinto this market has proven difficult, andthe prospect of entry did not alleviate ourconcerns.

In both cases, as in all mergers, theDivision analyzed whether a priceincrease, or a reduction in quality, wouldremain profitable, after accounting forrivals’ supply-side responses. The histo-ry of brand entry, exit, and positioning,and the associated costs, is relevant forthis portion of the analysis.

Step 4: Synergies. If a post-mergerprice increase is profitable, even afteraccounting for rivals’ responses as wellas consumer substitution, the merger islikely to be anticompetitive. Consumerslikely will be harmed by the combina-tion, unless it truly offers substantial effi-ciencies that lower incremental costs.Reductions in incremental costs can off-set the incentive to raise price, since themerged entity, like any firm, will have anincentive to set a lower price, the lowerare its incremental costs.

To be relevant, the cost savings musttruly stem from synergies specific to themerger. If one firm alone can achievelower costs by expanding its scale ofoperations, that should occur throughcompetition, not merger. Furthermore, ifthey are to benefit consumers, the syner-gies typically must lower incrementalcosts.23

Market Definition and Market SharesIn my experience the main battlefield inlitigated merger cases is market defini-tion. However, any attempt to make asharp distinction between products “in”and “out” of the market can be mislead-ing if there is no clear break in the chainof substitutes: if products “in” the marketare but distant substitutes for the mergingproducts, their significance may be over-stated by inclusion to the full extent that their market share would suggest;and if products “out” of the market havesignificant cross-elasticity with the merg-

ing products, their competitive signifi-cance may well be understated by theirexclusion.

Since we need to define markets toidentify the lines of commerce affectedby the merger, it may be tempting forcounsel to argue for a very broad marketif there is no clear break in the chain ofsubstitutes. This may be inconsistent,however, with a careful application of theMerger Guidelines, which can lead to amarket boundary between very similarproducts. In the pens case, testifying onbehalf of Gillette, I agreed with theDivision that one could define a marketboundary based on price, even thoughpens were sold at a continuum of prices.Judge Lamberth accepted this principle inhis opinion.24 In the case of geographicmarkets the mere fact that there are gasstations or supermarkets all over LosAngeles does not necessarily imply thatthe geographic markets for these prod-ucts are as broad as the entire L.A. area.Nor is there a single market for all food,despite the fact that it is difficult to drawboundaries between food groups.

On the other hand, an anticompetitivemerger cannot be disguised by arguingfor a very narrow market so as to quar-antine the merging parties in separate“markets.” Suppose that Brands A and Bpropose to merge, but Brand X is situat-ed between them. Suppose further that amerger between Brands A and X wouldlead to at least a 5 percent price increase,and likewise for a merger betweenBrands B and X. Defense counsel mightbe tempted to argue that A and X form amarket, and that B and X form a market,but that A and B are not in the same mar-ket. Still, if the Diversion Ratio betweenA and B is significant (albeit smaller thanbetween A and X or between B and X),the merger of A and B could well harmconsumers. In this case, it is appropriatefor the Division to use a price increase of more than 5 percent in defining themarket, as suggested in the MergerGuidelines.

Once markets are defined, using theGuidelines approach there remains theissue of how to use the firms’ marketshares in differentiated-products markets.As the Guidelines point out, market sharenumbers must be interpreted in conjunc-tion with evidence about the proximityof the merging brands. If the merging

brands are close together, the DiversionRatio is likely to be high, and any givenlevel of market concentration is moretroubling. The reverse is true if the brandsare distant.

If all the brands in the market areroughly equidistant from each other, thenthe market shares of various brands willbe proportional to their Diversion Ratios,making emphasis on the two brands’market shares very appropriate. TheGuidelines look to both the sum of themarket shares of the merging brands (tosee if this sum exceeds 35 percent in eval-uating unilateral effects) and to the prod-uct of the market shares (which willreflect the number of consumers whoregard the merging brands as their firstand second choices, and determines theincrease in the HHI).

As illustrated above, if the brands areequidistant from one another, informa-tion about market shares can be com-bined with a measure of the overall mar-ket elasticity of demand to estimate theDiversion Ratio between the mergingbrands. This estimated Diversion Ratiocan then be combined with informationabout premerger margins to give at leasta rough estimate of the profit-maximizingpost-merger price increase.

In arguing for a broad market, a com-mon tactic is to calculate a “critical elas-ticity” of demand for a group of productsbeing considered as a market,25 and thenargue that the true elasticity is above thiscritical level, making a 5 percent priceincrease unprofitable. This method mustbe used with great caution in the contextof differentiated products, to avoid atleast two pitfalls. First, there is no reasonto restrict attention to a uniform priceincrease of 5 percent for the purposes ofmarket definition if a single firm control-ling the entire product category wouldfind it optimal to increase the prices ofdifferent brands by different amounts.Second, care must be taken to ensure thatthe claimed “market” elasticity is consis-tent with information about each brand’sown elasticity of demand and the cross-elasticities of demand among the prod-ucts in the category. Remember, the“market” elasticity will be lower than theindividual brand elasticities of demand,and significantly so if the DiversionRatios are large. If each brand sells at ahigh markup, this is strong evidence of a


S P R I N G 1 9 9 6 · 2 9

low price elasticity for each brand, whichis inconsistent with a high “market” elas-ticity of demand. If the premergermarkups are large and the DiversionRatios among the brands are large, claimsof a large “market” elasticity of demandare not credible.

Unilateral Effects in the CourtsIn addition to the growing use of theapproach described here by AntitrustDivision economists, courts have alsoincorporated this type of analysis intotheir reasoning. For example, in the penscase, Gillette, Parker Pen, and other firmssuch as Cross, Schaeffer, and MontBlanc, offered fountain pens, roller ballpens, ball point pens, and mechanicalpencils at a continuum of prices. I em-ployed the methods sketched out here to develop my testimony on behalf ofGillette, although we lacked data onprices and quantities for premium pens toeconometrically estimate the DiversionRatio. This same general methodologywas employed by George Rozanski in hisanalysis for the Antitrust Division.Although Gillette and the Division agreed

it was appropriate to focus on unilateraleffects, we disagreed about how to inter-pret the evidence of substitution betweenfountain pens and roller ball and ballpoint pens, and on the difficulty of prod-uct repositioning.

Similarly, in the cereals mergerbetween Kraft and Nabisco,26 economicexperts for both sides, relying on super-market scanner data and survey evidence,spent considerable time estimating elas-ticities of demand for the purposes ofevaluating unilateral effects. JudgeWood’s opinion discussed unilateraleffects at great length, emphasizing theeconometric estimates of cross-price elas-ticity between the key merging brands,Grape Nuts and Shredded Wheat, forevaluating possible anticompetitive uni-lateral effects.

ConclusionMergers in markets with differentiatedproducts may seem a confusing area, andthe case law provides less guidance thanone might like regarding how to definemarkets to include “reasonable substi-tutes.” The Merger Guidelines, and the

consensus view among economists ofhow to analyze competition in differen-tiated-product industries, together pro-vide a consistent, valid, and reliable wayof evaluating proposed horizontal merg-ers involving differentiated products.Central to the analysis are the DiversionRatios between the two merging brands,which measure the fraction of customersof each brand that consider the mergingbrand their second choice. In caseswhere detailed price and quantity dataare available, Diversion Ratios can becalculated based on estimated elastici-ties of demand. Alternatively, theDiversion Ratios can be estimated basedon whatever pieces of evidence are avail-able, including more qualitative infor-mation.

If a significant proportion of con-sumers regard the merging brands as theirfirst and second choices, the DiversionRatios will be high, and the merger willindeed create an incentive to raise price,particularly if premerger Gross Marginsare large. This incentive can be undercutby rivals’ product repositioning, by entry,or by credible synergies.●

1 I am referring here to the Cournot model of quantity competition, whichdates back to 1838. See Joseph Farrell & Carl Shapiro, HorizontalMergers: An Equilibrium Analysis, 80 AM. ECON. REV. 107 (1990) (apply-ing this model to mergers); Asset Ownership and Market Structure inOligopoly, 21 RAND J. ECON. 275 (1990) (exploring the welfare proper-ties of this model, with applications to mergers and other transfers of assetsamong rivals).

2 In assessing mergers among suppliers of homogeneous products, the mainamendment to a structural approach urged by economics is that it is impor-tant to consider the elasticity of demand facing today’s suppliers in theaggregate. If entry is easy, or if other distinct products offer good substi-tutes, the elasticity facing today’s suppliers of the homogeneous good willbe high. And, if today’s suppliers indeed face highly elastic demand, anticompetitive effects are less likely to be significant, for any givenmarket structure. In fact, a strong case can be made that the danger of anti-competitive effects in these markets can be gauged by the ratio of the HHIto the market elasticity of demand. One of the many strong points of theMerger Guidelines is that their “hypothetical monopolist” approach tomarket definition explicitly incorporates the aggregate elasticity ofdemand into the market definition exercise.

3 It is important to measure incremental cost properly in calculating the pre-merger margin. For example, if a firm is facing capacity constraints, onemust include some capital costs in the measure of incremental cost. Also,the time frame and scale over which incremental costs are measuredshould be commensurate with the unilateral effects being studied.

4 Economists have long realized that firms selling differentiated productshave some “market power” in a technical economic sense, although typ-ically not nearly enough to rise to the level of “monopoly” power. In the1930s Joan Robinson and Edward Chamberlin developed the theory of“monopolistic competition” to describe markets in which each firm has adistinct product, but competes with several or many other firms. JOAN

ROBINSON, THE ECONOMICS OF IMPERFECT COMPETITION (1933); EDWARD

CHAMBERLIN, THE THEORY OF MONOPOLISTIC COMPETITION (1933). Bythe 1990s economists have made great progress in understanding competition with differentiated products, offering a sound foundation formerger enforcement in such industries.

5 Let me emphasize that the reliability of this analysis depends very muchon the data and other evidence upon which it is based. Ideally, the analystshould conduct a sensitivity analysis to make sure the results are not over-ly sensitive to specific simplifying assumptions that must be made.

6 Bertrand Equilibrium refers to the pattern of prices that prevails if eachfirm sets the price of its brands taking as given the prices of the otherbrands. This is also known as a “Nash Equilibrium” in prices. NashEquilibrium is the dominant method by which economists model “non-cooperative,” i.e., non-collusive, behavior among rivals. Bertrand equi-librium is not the only possible equilibrium concept, but, absent clear evi-dence to the contrary, it is a very useful workhorse.

7 I do not want to leave the impression that the actual analysis preciselytracks the four steps outlined below, rather these four steps form a con-ceptual road map.

8 This analysis will then have to be repeated for Brand B to see if its pricewill rise after the merger. If the merged entity has an incentive to raiseeither price significantly, the merger may be anticompetitive. Furthermore,if the firms own multiple brands, the four steps outlined here will need tobe repeated for each individual brand owned by either firm. With multi-ple brands, not only must the analysis be repeated, but we must keep trackof diversion to all brands owned by the merger partner when consideringthe repricing of any given brand.

9 In practice, economists often estimate the merging brands’ own- andcross-price elasticities, from which the Diversion Ratio can be calculat-ed. I find the Diversion Ratio more intuitive and easier to work with, so Iframe the discussion in terms of the Diversion Ratio.

What’s New on theInternet from theAntitrust Sectionhttp://www.abanet.org

H ave you visited the AntitrustSection on the Internet lately?

The Section’s site currently contains:

☛ a list of Section committee descrip-tions with a committee applicationand membership form;

☛ a list of Section publications anddescriptions with an order form;

☛ a calendar of upcoming Section programs; and

☛ a form for comments and questionsthat can be sent to the Section’s e-mail address:

[email protected].

Visit us at the 44th Annual AntitrustSpring Meeting! Stop by the Internetdisplay at the Antitrust Expo and seewhat the Section’s home page has tooffer. Speakers’ papers from the 1996Spring Meeting will be posted andfuture developments are planned for thesite. We welcome your suggestions.◆


3 0 · A N T I T R U S T

10 More generally, the formula is DAB = (EAB/EA)(xB/xA), where DAB is theDiversion Ratio from Brand A to Brand B, EA is the own-price elasticityof Brand A, and xA and xB are the unit sales of Brands A and B respec-tively. Robert Willig, Merger Analysis, Industrial Organization Theory,and Merger Guidelines, BROOKINGS PAPERS ON ECONOMIC ACTIVITY:MICROECONOMICS 281 (1991), also notes that the ratio of the cross-priceto the own-price elasticity measures the share of the marginal sales of onebrand that will divert to another in response to a price increase.

11 In other words, among a group of brands that are all equally “close” or“distant” substitutes, Diversion Ratios are proportional to market shares.This is the essence of the “logit” model of demand. See Willig, supra note10, for a discussion of the role of market shares in differentiated productmarkets using the logit model. For further detail on using the logit modelto analyze mergers, see Gregory Werden & Luke Froeb, The Effects ofMergers in Differentiated Products Industries: Logit Demand and MergerPolicy, 10 J.L., ECON. & ORG. 407 (1994).

12 Of course the effects of a merger cannot be predicted based on data analy-sis alone. To be reliable, any data analysis must pass a reality check basedon business documents and the testimony of industry participants.

13 United States v. Interstate Bakeries Corp., Civil Action No. 95C-4194(N.D. Ill. filed July 20, 1995)

14 United States v. Kimberly-Clark Corp., Civil Action No. 3:95 CV 3055-P(N.D. Tex. filed Dec. 12, 1995).

15 The elasticities obtained from the econometric estimation of demandwere introduced into an industry model of the pricing of all brands to esti-mate the unilateral effects of the merger. Some of the modeling methodsemployed are described further in Gregory Werden & Luke Froeb,Simulation as an Alternative to Structural Merger Policy in DifferentiatedProducts Industries (Economic Analysis Group Discussion Paper EAG 95-2 Sept. 1995).

16 This figure is obtained by dividing Brand B’s market share of 15% intothe combined share of all the brands to which customers of Brand A turn,which is 75%. This is an application of the sB/(1-sA) formula providedabove.

17 Symmetry means that the two brands have equal unit sales and GrossMargins prior to the merger and that the Diversion Ratio from A to B isthe same as that from B to A. This formula also assumes that each merg-ing firm sells a single brand prior to the merger. The analysis is moreinvolved and the formulas much more complex if the brands are not sym-metric or if the merging firms sell multiple brands prior to the merger.

18 In fact, the constant elasticity assumption becomes logically untenable asthe sum of the Gross Margin and the Diversion Ratio approaches orexceeds unity. The dangers associated with the constant elasticity assump-tion are greater, the larger are m and D.

19 I believe that the percentage price increases predicted by the logit modeltend to be comparable to those of the linear model, although I am unawareof any comparably simple formula in the two-firm symmetric logit model.

20 The magnitude of the optimal price increase depends not only on the pre-merger Gross Margin and Diversion Ratio, but also upon how quicklyelasticity itself rises as price rises above the premerger level. In principle,this information can be obtained from the data or from other evidence ofhow consumers would respond to price changes. Incorporating suchinformation is an essential “reality check” if simple formulas like theones displayed here are to be used.

21 Although it is difficult to generalize, and the Antitrust Division certainlyis not wedded to any single approach, there are some sound reasons to useformulas based on linear or logit demand systems for these purposes inpreference over constant-elasticity systems. This is what the Division hasdone in recent investigations. As I noted above, however, econometriciansoften estimate constant-elasticity of demand systems empirically. All I cansay with confidence is that the demand system used to predict post-merg-er price increases should, as much as possible, conform to the qualitativeas well as quantitative evidence that is available.

22 Oligopoly theory generally predicts that horizontal mergers will lead to atleast marginally higher prices if they generate no efficiencies, although thetendency towards higher prices can be thwarted by product repositioningor entry. This tendency is perhaps clearest in the case of differentiatedproducts and pricing (Bertrand) competition, where rivals typically willchoose to raise prices if the merging parties do so. See RaymondDeneckere & Carl Davidson, Incentives to Form Coalitions with BertrandCompetition, 16 RAND J. ECON. 473 (1985). However, the same resultapplies with quantity (Cournot) competition, even though rivals typically increase output as the merging firms restrict output. Even with these responses, horizontal mergers under quantity competition lead to higher prices unless they generate synergies. This is the “NoSynergies Theorem” proven in Farrell & Shapiro, Horizontal Mergers,supra note 1.

23 The issue of how to measure incremental costs remains, however. Thisinvolves questions of time and questions of scale, at least. The longer thetime frame over which we are looking, the more costs tend to be variable,or incremental, rather than fixed. And some categories of costs may beincremental with respect to large customers, but not small ones. In theextreme case of one customer, such as the Department of Defense for someweapons systems, cost savings in virtually any category may be passedalong to the customer, at least to some degree.

24 United States v. Gillette Co., 828 F. Supp. 78 (D.D.C. 1993).25 The “critical elasticity” is the minimum elasticity for which a 5% price

increase would be unprofitable.26 New York v. Kraft General Foods, Inc., 1995-1 Trade Cas. (CCH)

¶ 70,911 (S.D.N.Y. 1995).

http://www.abanet.org


























shapiro (1997)

Documents