Problems in the Use of Econometrics in Antitrust Matters

1
Problems in the Use of Econometrics in
Antitrust Matters
Presented at The Conference Board The 2003
Antitrust Conference Antitrust Issues in Todays
Economy New York March 18, 2003
Jonathan M. Jacobson Akin Gump Strauss Hauer
Feld, LLP 590 Madison Avenue New York, NY
10022 [phone] jjacobson_at_akingump.com
2
Overview

• Goals of econometric analysis in antitrust
• Econometrics in the courts
• Econometrics at the agencies
• Where we go from here
• List of post-1990 cases

3
Goals of antitrust econometrics

• Econometrics is the application of mathematical
  and statistical techniques to economic analysis
  of data and issues
• Use of econometric techniques in the study of
  antitrust economics has expanded rapidly in
  recent years
• The most prevalent econometric tool is multiple
  regression analysis

4
Goals of antitrust econometrics

• Sound econometric analysis should improve the
  quality of antitrust decision-making by
• Providing bases for quantification in a field
  where substantiality matters
• Providing bases for decisions to be informed by
  conclusions from hard data, rather than uncabined
  opinion or generalized presumptions
• Have our experiences been consistent with these
  goals?

5
Econometrics in the courts

• Lexis Westlaw antitrust databases yielded about
  100 reported antitrust decisions since 1990
  mentioning econometrics or regression analysis
• Econometrics was significant in about 30
• Of these, only 14 (less than half) involved
  antitrust merits (as opposed to damages or class
  certification)

6
Econometrics in the courts

• Cases fall into four main areas
• Expert efforts to develop an econometric formula
  to determine the fact of injury on a classwide
  basis to support class certification
• Regression analyses used to compute damages
• Expert testimony that the data were or were not
  consistent with a horizontal conspiracy
• Econometrics used to determine cross-price
  elasticities for market definition and/or to
  simulate the price effect of a merger

7
Econometrics in the courts

• Class certification
• Big battleground is whether impact (and sometimes
  violation) can be proven on a classwide basis
• Now routine for the class expert to say that he
  or she can develop a regression analysis that
  will demonstrate impact on a classwide basis
  notwithstanding differences among class members

8
Econometrics in the courts

• Class certification
• Some of these efforts are legitimate, at least in
  principle
• Cases involving price-fixing of commodity
  products sold in a national market
• Impact can be determined by comparing actual
  prices with econometrically determined
  competitive price (calculated based on unaffected
  regions or time periods)
• Wal-Mart v. Visa (comparison of actual discount
  rate for offline debit cards against discount
  rate for pin-based debit cards)

9
Econometrics in the courts

• Class certification
• Other cases are more problematic
• Cases such as Linerboard where the expert says
  only that that a regression equation can be
  developed and that data for the relevant
  variables are available without actually
  indicating what the equation or formula is
• We deem experts conclusion of common impact
  to be significant because it was supported by
  charts and trade publication studies.
• An extreme case is Heerwagen v Clear Channel, now
  pending, involving alleged monopolization of
  rock concerts the class expert promises to
  sort out over 30 variables that vary by
  geographic area, performer, and class member to
  yield a single coefficient for impact and damage
  estimation

10
Econometrics in the courts

• Damages computations
• Standard area for regression analysis
• Clearly appropriate subject for econometrics, yet
  some of the most egregious errors have been made
  here
• Conwood case - 1.05 billion award
• Conwood expert noted that, in those states
  ("foothold states") in which Conwood enjoyed a
  pre-1990 market share of 15 or more, Conwood
  grew in share from 6.5 to 8.1 between 1990 and
  1997
• He argued that, but for Defendants exclusionary
  acts, Conwood's market share would have grown by
  these same amounts in non-foothold states

11
Econometrics in the courts

• Damages computations - Conwood
• The regression coefficient was almost entirely
  based on the dramatic share growth from 19901997
  in a single foothold state Washington, DC
  where only 0.03 of moist snuff is sold
• In DC, Conwoods share plummeted from 21 to 7.2
  from 1984 to 1990, but then increased to 17.6 in
  1997
• If the DC outlier is omitted, damages are zero
• Following chart (courtesy Dan McFadden/John
  Harkrider) shows how DC behavior in pre-1990
  period was literally off the chart
• Each dot represents one of the 49 states or data
  points

12
Econometrics in the courts
Change in Conwood Market Share from 1984 to 1990
13
Econometrics in the courts

• Horizontal conspiracy cases
• Sound subject for econometrics
• Cannot tell us whether there has or has not been
  a conspiracy, but can tell us
• Whether data are consistent/inconsistent with
  collusion given supply and demand factors
• Whether pricing is consistent with pricing in
  analogous competitive markets
• Whether changes in prices are attributable to
  changes in costs or something else

14
Econometrics in the courts

• Horizontal conspiracy cases
• Econometric results have generally been
  superfluous
• Where the case is thrown out on summary judgment,
  the court finds some reason to discredit the
  plaintiffs econometrics. E.g., Blomkest.
• Where the case has survived, econometrics has
  been just one of the prongs used to sustain the
  result. E.g., HFCS.
• In no case has econometrics been the difference
  in finding that an issue of fact has been (or not
  been) raised

15
Econometrics in the courts

• Merger/Market Definition cases
• Econometric issues have been addressed in several
  litigated merger or market definition cases
• Staples
• Swedish Match
• NY v. Kraft
• Kodak (consent termination)
• Moore v. Wallace
• Heinz (district court)

16
Econometrics in the courts

• Staples/Swedish Match
• In both cases, Judge Hogan acknowledged but
  disregarded extensive econometric evidence on
  market definition, the key issue in each case
• He was far more comfortable relying on high-level
  internal company planning and strategy documents,
  and was particularly skeptical of econometric
  results inconsistent with the documents

17
Econometrics in the courts

• Staples/Swedish Match
• Nevertheless, in Staples, Judge Hogan credited
  FTCs econometrically determined elasticities in
  rejecting defense arguments about pass-through of
  efficiencies
• And in Swedish Match, he credited FTCs
  econometrics in sustaining unilateral effects
  argument as alternative ground

18
Econometrics in the courts

• Kraft
• Judge Wood extensively analyzed and credited Dan
  Rubinfelds econometrically determined own and
  cross-price elasticities in rejecting the states
  market definition and unilateral effects
  arguments
• Note that Fred Kahn served as court-appointed
  expert
• Kodak
• District Court and Court of Appeals credited
  Jerry Hausmans cross-price elasticities in
  determining that the geographic market for film
  was worldwide
• But even though econometric analyses were
  credited in both cases, they were not essential
  to either decision
• One reason they were credited was consistency
  with other evidence

19
Econometrics in the courts

• Moore v. Wallace
• Court rejected Hausman regression showing that
  the presence of a bid from Moore lowered Wallace
  margins significantly
• Inconsistent with non-econometric evidence
• Court credited Sumanth Addankis market analysis
• Although not discussed, regression had egregious
  omitted variable problem (size of bid)

20
Econometrics in the courts

• Heinz
• Court of Appeals rejected District Court opinion
  that relied on defendants econometric proof of a
  lack of cross-price elasticity at consumer level
• Relevant competition was for retailers
• Court might also have faulted analysis for
  disregarding coupons and, more importantly,
  competition on quality and product choice at
  retail level

21
Econometrics in the courts

• Summary
• Econometrics often found dispositive on class
  certification and damages issues
• With frequent dangerous and arguably wrong
  outcomes
• Without much understanding of the underlying
  analysis
• On these issues, courts are more willing to
  accept uncritically the econometricians black box

22
Econometrics in the courts

• Summary
• Econometrics appears (to date) never to have been
  dispositive on issues relating to antitrust
  merits
• Relied on as secondary backstop in analyzing
  sufficiency of evidence supporting a finding of
  collusion
• This makes good sense since econometrics can
  never prove or disprove a conspiracy
• Relied on in merger/market definition cases only
  as minor, secondary support after actual outcome
  determined by other evidence

23
Econometrics at the agencies

• Very different from in-court experience
• Agencies understand the econometrics and cannot
  be fooled by a magic black box
• But staff economists desire to perform demand
  estimations, merger simulations, critical loss
  analyses, etc., can be so strong as to detract
  from sound analysis based on data less
  susceptible of precise quantification
• Can result in false positives and false negatives

24
Econometrics at the agencies

• Key issues
• Data limitations
• Effects not capable of capture in numeric values,
  such as choice or quality
• Inconsistency of results with industrial facts
  contemporaneous strategic and planning
  documents and testimony
• Agency management have been alert to the problems
• See, e.g., Scheffman/Coleman paper on Use of
  Econometrics

25
Econometrics at the agencies

• Data limitations Scanner data
• Temptation to try econometrics on retail scanner
  data runs very high, but problems abound
• Several of the issues are addressed in D. Hosken,
  D. OBrien, D. Scheffman M. Vita, Demand System
  Estimation and its Application to Horizontal
  Merger Analysis (FTC Working Paper 2002)

26
Econometrics at the agencies

• Data limitations Scanner data
• Biggest problems sometimes ignored
• Price may not be the only important competitive
  dimension
• E.g., huge sales boost for Coca-Cola in mid-1990s
  from introduction of 20-oz. contour bottle
• If its not in the data, it doesnt exist (or
  doesnt matter)
• Many retail sales are not captured at all by
  Nielsen/IRI, while others are captured poorly
• Nielsen/IRI cover supermarkets, drug stores, some
  convenience stores, some mass merchandisers
• Do not pick up Wal-Mart (!)
• Do not pick up vending machines, sports and music
  arenas, newsstands, many others

27
Econometrics at the agencies

• Data limitations Scanner data
• Aggregation issues in estimating elasticities
• Channel-by-channel?
• Store-by-store?
• MSA-by-MSA?
• Different brands?
• Different package sizes?
• Coupons not captured well, if at all, but may be
  critical
• Relevance of retail data in analyzing wholesale
  level competition
• Consumer purchases for home inventory rather than
  consumption

28
Econometrics at the agencies

• Data limitations Scanner data
• In many contexts, displays are far more important
  in driving sales than price, but data on displays
  are often not available or inadequate
• Nielsen/IRI capture displays broadly, but
• Measure display size, if at all, only on one day
  of week
• Cannot distinguish more favorable in-store
  locations from less favorable
• Do not capture shelf space location,
  shelf-talkers, etc.
• Matching of critical promotion elements ad,
  display, price, coupons, shelf-talkers poor to
  non-existent

29
Econometrics at the agencies

• Data limitations Scanner data
• Best data are typically weekly, but correct test
  for own or cross-price elasticities is a
  nontransitory change (e.g., one year)
• Nontransitory changes affect consumers very
  differently from week-to-week changes
• Consumers may stock up at low prices in response
  to week-to-week changes
• Using yearly data does not fix the problem
• Too much happens apart from price over a year,
  such as advertising campaigns, design changes

30
Econometrics at the agencies

• Data limitations Scanner data
• With all these problems, unsurprisingly there can
  be bizzare results
• Chairman Muris reported findings of no
  cross-elasticity between Coke and Pepsi
• Results are often at odds with other evidence in
  case
• Agencies have access to company data through HSR
  filings and subpoenas, which can be much more
  reliable than syndicated retail sales data

31
Econometrics at the agencies

• Data limitations Scanner data
• Given these issues, it is fair to conclude that,
  until our ability to overcome these problems has
  changed drastically for the better
• No case should be brought based primarily on
  scanner data econometric results
• But, equally, the agencies should never decline
  to bring a case because of an absence of
  supporting scanner data econometrics

32
Econometrics at the agencies

• Econometric results can still be informative
• Data from other sources, such as party/nonparty
  data obtained by subpoena, may be sufficiently
  reliable and plentiful to generate useful
  econometric results
• Even scanner-based econometric results can help
  corroborate other data
• For example, if results show a high cross-price
  elasticity, and are robust, this may be important
  as secondary support for a unilateral effects
  argument
• But, if based on scanner data, a lack of
  econometric support should rarely, if ever, be
  viewed as proof of an absence of adverse effects

33
Where do we go from here?

• Accept the fact that some decisions about
  substantiality may have to be based on judgment
• Often, objective answers can be obtained from
  non-econometric data
• Remember that price is not the only relevant
  competitive variable
• The fact that a decision may depend on judgments
  that cannot be quantified mathematically should
  be neutral
• Should not favor or disfavor a decision to bring
  a case, or to decide a case in a particular way

34
Where do we go from here?

• Econometric analysis should not be used as a
  stand-alone replacement for good old-fashioned
  document review and interviews, which remain a
  useful way to analyze the market at issue
• Econometric results are most useful when
  consistent with other evidence in the case
• But dubious when inconsistent with other evidence
• David Scheffman has emphasized these points
  effectively at the FTC
• Econometric results might provide a tie-breaker
  when results are truly robust and other evidence
  is equivocal

35
Where do we go from here?

• In the courts, greater resources should be
  devoted to understanding econometrics
• Dan Rubinfelds suggestion of court-appointed
  experts to assist in analysis of econometric
  issues should be considered more often
• Used to good effect in NY v. Kraft
• Would demystify the black box
• Would be especially useful in class certification
  decisions

36
Where do we go from here?

• More broadly, continue to improve quality of the
  data our understanding of the limits of
  particular data types and our ability to see
  how, if at all, data issues can be overcome
• Already a work-in-progress at the agencies
• Avoid reliance on econometric evidence to
  disprove industrial facts
• Trust common sense
• If econometric results are at odds with common
  sense, common sense should prevail

37
Post-1990 Cases

• Class Certification
• In re Linerboard Antitrust Litigation, 305 F.3d
  145 (3d Cir. 2002)
• Wal-Mart Stores v. Visa USA, 280 F.3d 124 (2d
  Cir. 2001)
• Picket v. IBP, Inc., 2001 U.S. Dist. Lexis 22453
  (M.D. Ala. 2001)
• Paper Systems v. Mitsubishi Corp., 193 F.R.D. 601
  (E.D. Wis. 2000), revd sub nom. Paper Sys. v.
  Nippon Paper Indus. Co., 281 F.3d 629 (7th Cir.
  2002)
• In re Auction Houses Antitrust Litigation, 193
  F.R.D. 162 (S.D.N.Y. 2000)
• In re Flat Glass Antitrust Litigation, 191 F.R.D.
  472 (W.D. Pa. 1999)
• Alexander v. Q.T.S. Corp., 1999 U.S. Dist. Lexis
  11842 (N.D. Ill. 1999)

38
Post-1990 Cases

• Class Certification
• Stephenson v. Bell Atlantic, 177 F.R.D. 279
  (D.N.J. 1997)
• Lumco Indus. v. Jeld-Wen, Inc., 171 F.R.D. 168
  (E.D. Pa. 1997)
• In re Agricultural Chemicals Antitrust
  Litigation, 1995-2 Trade Cas. (CCH)  71,197
  (N.D. Fla. 1995)
• In re Potash Antitrust Litigation, 159 F.R.D. 682
  (D. Minn. 1995)
• In re Catfish Antitrust Litigation, 826 F. Supp.
  1019 (N.D. Miss. 1993)
• In re Domestic Air Transportation Antitrust
  Litigation, 137 F.R.D. 677 (N.D. Ga. 1991)

39
Post-1990 Cases

• Damages
• Conwood Co. v. United States Tobacco Co., 290
  F.3d 768 (6th Cir. 2002)
• In re Polypropylene Carpet Antitrust Litigation,
  93 F. Supp. 2d 1348 (N.D. Ga. 2000)
• In re Industrial Silicon Antitrust Litigation,
  1998-2 Trade Cas. (CCH)  72,348 (W.D. Pa. 1998)
• Rozena v. Marshfield Clinic, 977 F. Supp. 1362
  (W.D. Wis. 1997)
• Nichols Motorcycle Supply v. Dunlop Tire Corp.,
  913 F. Supp. 1088 (N.D. Ill. 1995)
• Pennsylvania v. Milk Indus. Management, 812 F.
  Supp. 500 (E.D. Pa. 1992)

40
Post-1990 Cases

• Horizontal Conspiracy
• In re High Fructose Corn Syrup Antitrust
  Litigation, 295 F.3d 651 (7th Cir. 2002)
• Blomkest Fertilizer v. Potash Corp., 203 F.3d
  1028 (8th Cir. 2000)
• City of Tuscaloosa v. Auburn Water Works Board,
  1998-2 Trade Cas. (CCH)  72,307 (11th Cir. 1998)
• Petruzzis IGA Supermarkets v. Darling-Delaware
  Co., 998 F.2d 1224 (3d Cir. 1993)
• Ohio v. Trauth Dairy, 925 F. Supp. 1247 (S.D.
  Ohio 1996)
• In re Aluminum Phosphide Antitrust Litigation,
  893 F. Supp. 1497 (D. Kan. 1995)

41
Post-1990 Cases

• Merger/Market Definition/Other
• FTC v. H.J. Heinz Co., 246 F.3d 708 (D.C. Cir
  2001), revg 116 F. Supp. 2d 190 (D.D.C. 2000)
• United States v. Eastman Kodak Co., 63 F.3d 95
  (2d Cir. 1995)
• United States v. AMR Corp., 140 F. Supp. 2d 1141
  (D. Kan. 2001)
• FTC v. Swedish Match, 131 F. Supp. 2d 151 (D.D.C.
  2000)
• FTC v. Staples, Inc., 970 F. Supp. 1066 (D.D.C.
  1997)
• New York v. Kraft General Foods, 926 F. Supp. 321
  (S.D.N.Y. 1995)
• Moore Corp. v. Wallace Computer Services, 907 F.
  Supp. 1545 (D. Del. 1995)