ABA Brownbag Series Market Definition Patrick DeGraba U.S. Federal Trade Commission* December 12, 2016 *The views expressed in this presentation are my own and do not necessarily represent those of the Federal Trade Commission or any individual Commissioner.

1

Overview • What is Market Definition? • Why Do Market Definition? • Product Market Definition: – Hypothetical Monopolist Test – Targeted Customers

• Geographic Market Definition • Market Definition in Some Cases

2

What is Market Definition? • Ultimate goal of merger analysis is competitive effects. Will market power be created and exercised? • Useful to separate products likely to significantly constrain the merging firms’ prices from those that will clearly not significantly constrain price. • Market Definition is an over inclusive first approximation of those products that will likely constrain prices. 3

What is Market Definition? • Terrible Use of the English Language • There is no “actual market.” • How can you define something that does not exist? • Like “defining” Washington metropolitan area

• We might be better served with a term such as “substitutes triage” or “likely competing products delineation.” • Back of the envelope competitive effects 4

Why Do Market Definition? • Need to have a set of competing products to calculate market shares. – Mergers of firms with high market shares in highly concentrated market most likely to lead to higher prices. • Quick way to eliminate obviously non-problematic mergers • High concentration establishes prima facie case

• Might help organize an investigation. • Provide a starting point to be used in a more comprehensive competitive effects analysis, though competitive effects analysis often includes firms outside the formal market definition

• Section 7 cases seem to require market definition. • “Line of business” equated with market definition 5

Why Market Definition is Controversial • Economic analysis does not require market definition to obtain competitive effects results. • Merger guidelines- “[competitive effects] analysis need not start with market definition.” – Evanston- Judge found effects results compelling even though he disagreed with FTC market definition.

• Market definition can wrongly supplant competitive effects. • The “zero – one” effect – All competitors in a market often treated equally. » Hershey - “…these 19 other hospitals within a 65 minute drive of Harrisburg provide a realistic alternative that patients would utilize.” (DCD p10) – Competitors not in the market are ignored completely

• A market defined too broadly can short circuit competitive analysis – Whole Foods

• A market defined too narrowly can short circuit competitive analysis – Lundbeck 6

How to do Market Definition • Intuitively - Find a set of products to which “enough customers that matter” might switch if the prices of the products under investigation increase. – Market definition is driven by what consumers view as close substitutes.

• There are two dimensions – Product Market – based entirely on customer willingness to substitute among products with different attributes – Geographic Market – Based either on customer location or supplier location

7

The Hypothetical Monopolist Test: Product Market • The Hypothetical Monopolist Test uses an iterative process to “define” a market – Begin with a product sold by each of the merging firms as a candidate market. – Ask if a hypothetical monopolist that controlled this product could profitably impose a “small but significant and nontransitory” increase in price” (SSNIP) above the benchmark (often prevailing) level. • If yes, stop. This product is a market. • If no, go to the next step.

– Expand candidate market by adding the product to which the most sales divert, and return to the SSNIP question.

8

The Hypothetical Monopolist Test: Product Market Illustrated • •

Product “A” is a candidate product market. Raising product A’s price by a SSNIP is unprofitable because of significant substitution to product B and some substitution to C. Now consider A and B as the candidate market. Raise A’s and(/or) B’s prices by a SSNIP. Substitution to C (and others) does not make the price increase unprofitable. Then, A & B comprise a relevant product market. C is not in the market.

• • • •

2

1 C A B Consumer substitution to B defeats price increase.

C

A B

Insifficient consumer substitution to C to defeat price increase. A and B comprise a market.

9

SSNIP test that raises all prices is over-inclusive • When the price of product A increases, the price of substitutes will typically increase as well, but not by as much. – Understates the prices of products not in the candidate market. • The typical increase in price makes the substitutes less of a competitive constraint than it would be if its price remained unchanged. • The SSNIP test keeps the price of substitutes not in the candidate market unchanged. • Thus, the SSNIP overstates the competitive constraint these substitutes would impose if their prices were allowed to adjust to a price increase in the candidate market.

– Can overstate the prices of products in the candidate market. •

If the price of merging products increases 5%, close substitutes would increase < 5%

•

Having the price of all products in candidate market increase by 5% overstates customers willingness to “leave” the candidate market and substitute to products outside the candidate market, compared to their willingness to leave the candidate market if the merging parties raised their prices and the rest of the market responded with increases of less than 5%.

10

SSNIP test is never under-inclusive • Choose a candidate market. – If candidate market is too narrow, then the test tells us SSNIP is not profitable – So candidate market expands marginally – Expansion stops when the smallest market is reached

• A “significant” competitor may not be included in the market – A competitor may be noticeable but not big enough to make a SSNIP unprofitable and so is correctly excluded from the market

• SSNIP won’t find an overly broad candidate market 11

Critical Loss and the SSNIP • SSNIP test can be done by comparing something called critical loss to something called predicted loss. • Critical loss is a percentage that we calculate from the size of the SSNIP and the profit margin of the candidate market. – Relatively straightforward – Parties will argue over size of profit margin

• Predicted loss is estimated from data and/or documentary evidence. – Evidence used can be different for each case – More disagreement among economic experts 12

Critical Loss and the SSNIP • A price increase (including a SSNIP) implies – sales of some units will be lost – remaining units will be sold at a higher price

• Unit sales loss reduces profits • Higher price increases profits • The number of lost units that causes the profit reduction to just equal the profit increase from higher prices from a SSNIP is called the “Critical Loss.” 13

Critical Loss and the SSNIP • Critical Loss (CL) is the number of lost units such that the profit reduction from lost sales just equals the profit increase from a higher price on the remaining sales. %CL = • • • •

• •

%Δp M + %Δp

Where M =

p - mc p

Derivation Profit Increase = Δp(q – Δq) Profit reduction = (p – mc)Δq CL = Δq such that the Added Profit just equals the Lost Profit Δp(q – Δq) = (p – mc)Δq qΔp CL = Δq* = p + Δp - mc Divide both sides by q and both fraction terms by p to get %CL 14

Critical Loss and the SSNIP • Examples with a 5% SSNIP and 5%, 20% and 45% margins • Increasing margins imply decreasing critical losses

%CL =

%CL =

%CL =

5 5+5 5 20 + 5 5 45 + 5

= 1/2 = 50%

= 1/5 = 20%

= 1/10 = 10%

15

Critical Loss and the SSNIP • The “Predicted Loss” (also called the “Actual Loss”) is the number of lost unit sales that the hypothetical monopolist is predicted to lose due to the price increase. • The predicted loss is estimated from data and documentary evidence (See later slides). • If predicted loss is greater than the critical loss, then the SSNIP is not profitable and the candidate products are not a relevant market. • If predicted loss is less than the critical loss, then the SSNIP is profitable and the candidate products are a relevant market. • A key issue, then, is how to reliably estimate the Predicted Loss. 16

Critical Loss and Market Definition • Predicted Loss estimates are based on data that allow us to estimate diversion among products or price effects. – Nestle/Dreyer’s (2003) – super premium ice cream • Looked at retail data and observe entry by Dreyer’s resulted in lower prices of other super premium ice cream. • Could calculate diversion between super-premium and premium ice cream base on retail data.

– Sysco (2015) – broadline foodservice distribution • looked at bidding data and inferred the next best alternative to one of the merging parties would be if that party raised it prices by a SSNIP (FTC v Sysco; M&O at 36 https://www.ftc.gov/enforcement/cases-proceedings/ftc-v-sysco-usfholding-corp-us-foods-inc

) 17

How Should We Interpret High Margins? • High margins mean the critical loss will be small: • Remember: %CL =

%Δp M + %Δp

• But high margins also suggest that price elasticity is low, which means predicted loss will be small. • M = 1/Eown where Eown is elasticity of demand • While not dispositive this creates an inherent tension from a claim of low critical loss but high predicted loss. • O’Brien and Wickelgren (2003) and/or Katz and Shapiro (2002) 18

Critical Loss and Market Definition • SSNIP/CL analysis by itself is rarely dispositive. Courts will also check that such results are consistent with other evidence of market definition. • Data may not always be available to perform a SSNIP test. 1.

Interviews with market participants 1. 2.

2. 3.

Buyers Sellers

Depositions (investigational hearings) Documents – Business and Marketing Plans – Sales reports – Internal and third-party industry studies – E-mail discussions of pricing and competition

19

Targeted Customers Product Market Definition • Products sold to an identifiable group of customers can be a separate market if: – Seller can charge that group of customers a different price than other customers – No arbitrage between this group and other groups

• Targeted customers is one of the most important ideas in the merger guidelines given court’s reliance on market definition – Targeted customers allows analysis closer to competitive effects analysis • •

Can rule out tangential competitors that are not significant Can focus on specific aspects of competition for specific customer groups

• Recent examples: – Sysco – Broadline foodservice distribution to national customers – Whole Foods – Core customers

20

Geographic Market • An area outside of which products in the product market do not compete significantly enough with products inside the area. • Grocery stores in Rockville, Md don’t compete with those in DC

• Typically important if customers purchase in a limited area. – Pinnacle/Ameristar • Local markets St Louis and Lake Charles • Exclude Las Vegas.

– Evanston Northwestern Health • Determine which hospitals near merging hospitals are good substitutes

• SSNIP test can be applied to geography just like products. • Start with merging location and raise the price. • If SSNIP not profitable add a competitor and try again.

• SSNIP is replacing Elzinga-Hogarty in hospital mergers 21

Geographic Market • Targeted customers. – If national firms can discriminate by location, each location could potentially by a separate market. – Things to consider • If price is national but promotions are local, each locality could be a market • If non-price attributes like service quality are chosen on a local basis • If two national competitors have different local competition in different cities – FCC uses local geographic markets for mobile service (AT&T-T-Mobile staff report para 32-34 https://transition.fcc.gov/transaction/att-tmobile.html ) • Local casinos might offer promotions by zip code.

• In many cases geographic component adds nothing – Arch Coal – ~10 mines and all of the electricity producer that used that coal were identified. What help is geography?

22

Geographic Market • Elzinga-Hogarty test is not valid for hospitals. – E-H test developed in the 1970s to analyze commodity flows https://www.justice.gov/atr/chapter-4-competition-law-hospitals#2a1

• Predates SSNIP test development

– Intuitively - Draw a circle on a map • If most people living inside the circle buy inside the circle… • …and very few people form outside the circle buy inside the circle, • then the circle is a geographic market

– THIS TEST CAN PRODUCE MARKETS THAT ARE MUCH TOO BIG • Used for years to defeat challenges to hospital mergers. • Elzinga testified in In re Evanston Nw. Healthcare Corp., 2007, that his own test gave overbroad markets in hospital settings. – –

Elzinga and Swisher 2011 Inter J of Econ of Bus. Capps, Dranove, Greenstein, and Satterthwaite, NBER April 2001

• One major problem is that it was designed for homogeneous goods but hospitals tend to be heterogeneous • Not designed for customer flows 23

FTC v Sysco (2015) • Merger between two largest and only national broadline foodservice distributors in the U.S. • FTC asserted a market for targeted broadline foodservice distribution to national customers. (FTC v Sysco M&O p 18) – Customers with large geographic presence would not switch their purchases to a patchwork of smaller service in response to a SSNIP

• Economic expert performed a SSNIP test – Based on margin calculations determined critical loss to be 50% (M&O 35) – Calculated predicted loss to be much lower than 50% (M&O 35-37) • Looked at bidding and RFP data to determine how customers would divert to another broadliner if the price of one broadliner were to increase by 5%. • Looked at Linc database maintained by U.S. Foods services and found much less than 50% of customers would switch to non-broadline services in response to a SSNIP. (M&O 37)

24

FTC v Sysco (2015) • Two conditions for targeted customers – Ability of sellers to price discriminate – No arbitrage

• Judge found FTC expert analysis “persuasive” – Had reservations because data was not from ordinary course documents but rather assembled for the proceeding and were not necessarily complete – Found it important that results of the data analysis were consistent with ordinary course documents and business practices •

“the determination of the relevant market in the end is 'a matter of business reality-[ ] of how the market is perceived by those who strive for profit in it” (DCD p39)

25

FTC v Whole Foods (2007) • Merger between Whole Foods and Wild Oats • Two largest organic grocery chains • Market Definition – FTC - Premium Natural Organic Stores (PNOS) – Defendants Conventional Supermarkets, were also in the market. (Circuit Court Decision (CCD) 14)

• District Judge - Because there are marginal customers that would switch to other stores there was not PNOS market. The larger market unlikely to have antitrust harm (CCD 2, 19) • Circuit Court – Viewed this case as a targeted customer case. (CCD 20)

26

FTC v Whole Foods (2007) • Circuit Court found – competition between PNOS stores and other supermarkets for dry goods – no such competition for perishable goods between PNOS and conventional supermarkets (ACD 17-18) – Competition between Whole Foods and Wild Oats (and other PNOS) did affect prices of perishable goods – “Core” PNOS customers bought primarily perishable goods – Thus they could be price discriminated against – Thus the cluster of goods they purchased constituted a submarket.

27

FTC v Whole Foods (2007) • “Diversions Ratios”: Based on analysis of Wild Oats Revenues •

FTC presentation A challenged merger of retailers: FTC v Whole Foods\ Sophia Bulgaria Sept 2009

Other PNOS

PNOS Whole Foods

PNOS

Trader Joe's

Vitamin Cottage

Conventional Sup.

Gourmet Sup.

Mass Merchants

28

FTC v Whole Foods (2007) • Wild Oat’s Response to Entry Whole Foods

Other PNOS

Trader Joe's

Margins

Gourmet Sup.

Conventional Sup.

Mass Merchants

Implied Prices

29

Evanston Northwestern Healthcare (2005) – FTC brought suit against a consummated merger of hospitals in Evanston Illinois. – FTC alleged a geographic market consisting of 3 hospitals two of which had merged. Defendants claimed a larger market including more hospitals – FTC presented direct evidence of price increases as a result of the merger. – 3 Interesting findings by Judge McGuire • Elzinga-Hogarty not appropriate for hospital geographic market because it returned overly large geographic markets. • The geographic market contained 7 hospitals, even though the plaintiffs proposed a three hospital market. • The direct evidence of anticompetitive effects were enough to overcome the larger market. 30

Penn State Hershey (2016) – Hershey, the leading teaching hospital in the Harrisburg area bought nearby Pinnacle Health System – FTC (and PA) sued, and alleged a four county market in the area (Circuit Court decision (CCD) 15)

– District court found • Market is bigger - 43% of Hershey patients are outside the 4 counties (District Court Decision (DCD) p 9); up to 65 minutes away (DCD p 10) • Hershey signed long term deals with two largest insurers, locking in prices

– Circuit Court found • Reliance on the 43% figure followed E-H and ignored SSNIP (CCD 16) • District Judge ignored 91% of Hershey customers were local (CCD 20) – Inflows don’t predict outflows, ignore product differentiation

• Contracts are not used to determine market definition (CCD 16) – Also consider that competition occurs on non-price dimensions

31

FTC v Lundbeck (2009) • “…perplexing…” market definition precluded any competitive effects analysis https://www.ftc.gov/sites/default/files/documents/cases/2010/08/110919lundbeckfindings.pdf

– Lundbeck purchased Indocin IV with retail price of $78. Treats PDA in preterm babies. Purchased the only other drug to treat PDA, NeoProfen, about to receive FDA approval. Two days later raised the price of Indocin IV to ~$1500. – After FDA approval, Lundbeck priced NeoProfen similarly. – Eight doctors testified, would not switch for small price difference. – Defendant’s expert said a hospital that would prefer lower priced drug could not “drive” doctors towards cheaper drug. (CCD 6)

32

FTC v Lundbeck (2009) • District Court Judge concluded – Doctors would not switch, implies low cross price elasticity – Low cross price elasticity implies not in same market – Not the same market implies no antitrust harm

• Low cross elasticity by itself DOES NOT imply not in same market. Must be compared to own price elasticity. – Competitive Effects (symmetric linear case) %ΔP =

MD 2(1-D)

p - mc Where M = p

and D =

εji εii

• Casual analysis of a 5% price change may not help analyze a 1300% price change. 33

Takeaways • Market definition is best viewed as an over inclusive first approximation of the important competing products. • Based on customer substitution to other goods. • SSNIP test results should be consistent with other evidence. • Can be useful – When structural analysis is appropriate. – As a way to organize an investigation.

• Pitfalls of use – Not a prerequisite for doing competitive effects and can divert attention from the competitive analysis. – Potential for all sellers in the market to be treated as equals when they are not.

• “Targeted customers” useful in focusing on competitive effects. 34