Empirical Methods for Industrial Organization

1
Empirical Methods forIndustrial Organization

• Eugenio J. Miravete
• University of Pennsylvania CEPR
• Portuguese Competition Authority
• This version March 10, 2004

2
Applications I
3
Road Map

• Multimarket contact.
• Evans and Kessides Airlines.
• Parker and Röller Cellular Telephone.
• Collusion price discrimination.
• Bresnahan American Automobile in the 1950s.
• Verboven European Automobile in the 1990s.
• Mergers, variety and market power.
• Nevo Breakfast Cereal.
• Ivaldi and Verboven The Volvo-Scania case.
• Goolsbee and Petrin Cable and Satellite TV.

4
Evans-Kessides

• Evans, W.N. and I.N. Kessides (1994)
• Living by the Golden Rule Multimarket Contact
  in the U.S. Airline Industry. Quarterly Journal
  of Economics, 109, 341-366.
• Golden Rule Airlines refrain from initiating
  aggressive pricing actions in a given route for
  fear of what their competitors might do in other
  jointly contested routes.
• This is a reduced form, panel data model of the
  1000 largest city-pair routes between 1984 and
  1988.
• Fares are higher in city-pair markets served by
  carriers with extensive overlapping of their
  networks.

5

• Airline industry.
• This is an ideal market to test for the existence
  of multimarket contact and its effects
• Firms compete on different city-pair markets.
• Markets are precisely defined.
• The hub-spoke system conveys important cost
  advantages to the carrier operating the hub.
• Cost difference of carriers across routes.
• Different spheres of influence in each
  geographical region.
• Market power of carriers may depend on their
  airport dominance.
• There is a significant difference in the number
  of operating carriers and growth rate across
  routes.
• Abundant data (publicly available).
• Data Bank of the US Department of
  Transportations Origin and Destination Survey.

6

• Model.
• Pijt Average fare charged by airline i on route
  j in time t.
• Xijt Carrier based characteristics (vary over
  time and market).
• Route market share.
• Airport market share average of airline market
  share at the the airports of the endpoints of the
  route.
• Fraction of direct flights (as opposed to flow
  passengers).
• Fraction of round-trip tickets.
• Vector of airlines dummy variables.

7

• Zjt Route based characteristics (vary over
  time).
• Route Herfindahl index (clearly endogenous).
• Mileage.
• Measures of multimarket contact
• Average route contact.
• Average revenue contact.
• µi Airline fixed effect.
• ?t Time fixed effect.
• ?j Route fixed effect.
• Some issues
• Are price movements driven by demand or by costs?
• Authors emphasize the robust estimates obtained
  by controlling for the unobserved heterogeneity.

8

• Indices of multimarket contact.
• At any year t, there are n routes and m airlines.
• On route j there are only fj carriers offering
  the service.
• Define Dij1 if airline i operates on route j.
• The number of routes serviced by both airlines k
  and l is

9

• The average route contact for route j is then
• For monopolized routes ARCj0.
• ARCj will be positively correlated with firm size
  as larger carriers overlap their networks more
  frequently.
• Objections
• This measure of route contact weights all routes
  served by two carriers equally.
• ARCj does not account for its economic magnitude,
  that is in relation to the revenues or profits
  generated in each route.

10

• Define Rij the percentage of airline is revenues
  that come from route j.
• The revenue-weighted measure of the importance of
  contacts between airlines k and l on different
  routes is
• The average revenue contact for route j is then
• Observe that the magnitude of the contact is not
  only a matter of the size of the route but also
  of the market conduct of the firms.

11

• Descriptive statistics.

12

13

• Fixed effects estimation.
• Most of the sample variation is due to
  differences in unobserved market-specific
  characteristics.
• The data also include time-series and fixed
  effects capture systematic differences in prices
  through route, airline, and year.
• This is a period of tremendous consolidation in
  the industry.
• Multimarket contact doubled between 1984 and
  1988.
• Since fixed effects is a within (route)
  estimator, the large within-route change in
  multimarket contact should identify pricing
  effects.

14

15

• Results
• Route effects account for between 47 and 73 of
  sample variance of airline prices.
• External contacts are significant and very
  important.
• Prices tend to be higher in routes where carriers
  have extensive contact.
• Fares are 5.1 higher when comparing routes on
  the upper and lower quartiles of ARC.
• Round-trip tickets are 32 less expensive than
  one-way.
• Market power is correlated to airport presence
  rather than route share. This may be evidence of
  network externalities.
• Most of the benefit of competition is achieved
  with a duopoly.
• Fares decrease by 11 with a second carrier.
• Fares only decrease by 1 when going from 4 to 5
  carriers.

16
Parker-Röller

• Parker, P.M. and L.H. Röller (1997)
• Collusive Conduct in Duopolies Multimarket
  Contact and Cross-Ownership. RAND Journal of
  Economics, 28, 304-322.
• This is a static, conjectural variations,
  empirical oligopoly model that analyzes the
  duopolistic US cellular industry in its infancy,
  from 1984 to 1988.
• Product and technology are homogeneous.
• Markets are well defined.
• The data is suitable to
• Validate the specification as it includes
  monopolistic markets as well as duopolies.
• Evaluate the effect that multimarket contact has
  on the ability to collude in a structural model.

17

• Some facts.
• United States, 2001
• Market penetration of 45 with 136 million
  subscribers.
• Sales 60 billion.
• Employment 200,000 direct jobs.
• United States, 1988
• 1.6 million subscribers.
• Sales 2 billion.
• Employment 9,000 direct jobs.
• Average monthly bill 95.00.

18

• Market Definition.
• Technological constraint Scarce radio spectrum.
• Solution
• Service areas divided in small cells served by
  its own low-powered transmitter. It allows this
  frequency to carry a different call in a
  non-adjacent cell.
• A mobile telephone switching office maintains a
  continuous transmission when customers move to a
  cell that uses a different frequency.
• FCC design of the early US cellular market
• Define 305 non-overlapping markets
  SMSAs.
• Wireline license fixed line carriers in that
  area.
• Non-wireline license any other US citizen or
  company.

19

• SMSA.
• It includes a central city or urbanized area of
  at least 50,000 people.
• It also includes the county containing the
  central city and other contiguous counties with
  strong economic and social ties to the central
  city.
• US Census (1990)
• 76 of the population.
• 16 of the land.

20

21

• Tariff Plans.
• Cellular Price and Marketing Letter, Information
  Enterprises
• Pricing plans information reported by firms
  between December of 1984 and 1988.
• Price plans are typically two-part tariffs with
  quantity discounts.
• The number of plans varies from 1 to 6.
• Plans normally include a peak-load component and
  airtime allowance.
• Focus
• Retail market.
• Peak period billing for 500 minutes.

22

• Market size.
• Cellular Business, various issues, 1984-1988
• Cell sites, number of antennas.
• Start-up date.
• Remarks
• Output level is not directly observable.
• Each cell site represents between 1,100 to 1,300
  subscribers.
• In a sample of 87 markets in 1987, the
  correlation between number of cells and
  subscribers was about 0.92.
• Market shares of competing firms are not known
  (except for about 20 markets in 1987).

23

• Factor Prices.
• US Department of Labor, US Department of Energy
  BOMA Exchange Report, various issues, 1984-1989
  and Statistical Abstracts of the US, 1989
• Wages per employee for the cellular industry.
• Price of Kilowatt-hour of power.
• One period lagged prime lending rate.
• Average monthly rent of square foot in each
  market.
• Other operating costs per-square foot index of
  office building cost. It accounts for cleaning,
  repair and maintenance, administration,
  utilities, local taxes, security, office payroll
  and other leasing expenses.

24

• Demand.
• US Department of Commerce Census Bureau and The
  Cellular Market Data Book, EMCI, Inc.
• Market population and growth.
• Time trend.
• Extension of SMSA (for population density).
• Average commuting time.
• Number of potential business establishments
  (services such as business, health care,
  professional and legal, construction,
  transportation, finance, insurance and real
  estate).

25

• Other issues.
• Market specific characteristics
• Index for majors (holding more than 5 of
  markets).
• Market configuration Bell vs. Independents
  carriers.
• Cross-ownership (partners in any other market).
• Age of the market since the first firm started
  operating.
• Lead time of the wireline relative to the
  non-wireline.
• Multimarket contact indicator.
• Regulation (State level)
• No regulation (50 of markets).
• Voluntary tariff reporting to public utility
  commissions (26).
• Mandatory tariff reporting to public utility
  commissions and regulatory approval of any tariff
  increase (24).

26

• Relevant features
• Multi-year license awarded over time in 10 tiers
  of 30-65 markets.
• Wireline awarded by comparative hearing.
• Nonwireline awarded mostly by lottery.
• Entry after the license was granted always
  occurred. There were no plans to grant new
  licenses
• We can rule out strategic monopoly pricing. Entry
  deterrence was not feasible in equilibrium.
• Preemption during duopoly can also be ignored.

27

• Model.
• Demand semi-logarithmic.
• Supply quadratic costs.
• Profit maximization conjectural variations.
• Simultaneous estimation of demand, costs, and
  ?ts.
• The existence of a monopoly period is used to run
  an specification test.

28

29

30

• Results
• Downward sloping demand.
• Absolute price elasticity (at simple means)
  2.5.
• Population and business have a similar effect on
  demand.
• Fast market growth, about 8.5 a month.
• Increasing marginal cost.
• CRS are rejected. This is a necessary condition
  for collusion through multimarket contact.
• Behavioral parameter.
• The industry is more collusive than
  non-cooperative duopoly.
• The price-cost margin is about 35.

31

• Explaining collusion.
• In order to allow for some variation of the
  conjectural variations parameter, a fourth
  equation is added and estimated simultaneously to
  the previous three
• There are four specifications of this general
  formulation
• RBOCs vs. Independent companies.
• Identity of cellular carriers.
• General effect of regulation.
• Multimarket contact and other market
  characteristics.

32

• RBOCs vs. Independent companies
• BELLBELL, INDBELL, BELLIND (always
  wireline-nonwireline) colluded similarly to the
  average collusive behavior.
• INDIND index of collusion is significantly
  higher.
• This occurs in 22 of markets.
• Identity of carriers. In decreasing order of
  contribution to ?ts
• Independents Rest, GTE Mobilnet, Contel
  Cellular, McCaw Communications, (Century
  Cellular).
• RBOCs Bell Atlantic Mobile, Pactel Mobile
  Access, US West Cellular, BellSouth Mobility,
  (Ameritech Mobile), (Nynex Mobile), (Southwest
  Bell Mobile).
• The behavior of firms between parenthesis is
  competitive.

33

• Effect of regulation
• Low regulation, that is, reporting tariffs
  voluntarily, facilitates collusion.
• Multimarket contact and other market
  characteristics
• Positive effect on collusion
• Cross-ownership.
• Multimarket contact.
• Negative effect on collusion
• Market age. There is a downward trend in prices.
• No effect.
• Lead time of the first relative to the second
  entrant.
• Duopoly leads frequently to more collusive
  behavior than in monopoly. Where monopolists
  trying to gain market share?

34

35
Bresnahan

• Bresnahan, T.F. (1987)
• Competition and Collusion in the American
  Automobile Industry. Journal of Industrial
  Economics, 35, 457-482.
• This paper is an attempt to document a price war
  in the US automobile market in 1955.
• Mild macroeconomic expansion.
• Auto sales increased by 45, never matched
  earlier or later.
• Model of short run equilibrium with vertically
  differentiated products.
• Short run prices and quantities are set as well
  as product lines.
• Price competition will drive price close to
  marginal cost for those products for which a
  close, competitive substitute exists.
• Under collusion, the markup will not depend on
  whether the substitute product is sold by some
  competitor or the firm itself.

36

• FACTS
• Table I. Automobile market.
• Quality adjusted price reduction.
• Significant increase in sales.

37

• Table II. Macroeconomic variables.
• Mild expansion of demand.
• Low interest rate may increase the demand for
  durable goods.

38

• Model
• Demand.
• Individual unit demand linear in prices.
• Constant marginal rate of substitution between
  automobile quality and all other goods.
• Consumers with taste parameter v for quality, x,
  and money not spent on autos, Y-P, reach the
  utility level
• Variable v is the willingness to pay for auto
  quality.
• E is the shadow price of the outside option.
• Parameter g is the quality assessment of the
  outside good.

39

• Aggregate demand.
• We need to find the cut-off types as a function
  of prices and qualities

40

• The demand for product i is the probability that
  the valuation of consumers falls within the
  interval vhi, vij
• Demands are only affected by immediate
  substitutes.
• The smaller the differences in qualities,
  (xi-xh), the closer the cross-price derivative
  d/(xi-xh) is to the own price effect d/(xj-xi)-
  d/(xi-xh) in absolute value.
• Costs.
• Marginal cost is increasing and convex in quality

41

• Industry conduct.
• Only neighboring products have any
  interdependence of demand, i.e., their producers
  can be either cooperating or competing.
• Where Hij1 if firms producing products i and j
  cooperate and Hij0 otherwise.
• Example
• Firm 1 sells products 2, 4, 5.
• Firm 2 sells product 3.
• Prices P2 and P3 are very close to marginal cost
  as they are almost perfect substitutes.
• Products 4 and 5 are large cars, only produced by
  GM.
• Products 1 and 2 are smaller cars produced by GM,
  Ford and Chrysler

42

43

44

• Quality and other data.
• Vector Z includes the length, weight, horsepower,
  number of cylinders and a hardtop dummy to
  indicate type of body.
• These characteristics together with the above
  specification of quality rank models from top to
  bottom of the market.
• In addition the model includes production
  (instead of sales or registration) from various
  sources and list prices of models with standard
  options.
• Alternative specification Hedonic price
  (non-strategic).

45

• Results
• Models are non-nested (Cox test).
• 1954 The hedonic model is rejected as well as a
  model where firms price as single-product
  producers.
• 1954 and 1956. Only collusion is not rejected
  against any other model.
• 1955 Nash competition is not rejected.
• Table IV summarizes the results for each year for
  the non-rejected specification.
• Parameters are stable and the differences in year
  1955 are captured by changes in the behavioral
  assumptions!

46

47

• Some other results
• Negative cylinder effect.
• It may be due to colinearity with horsepower.
• It may capture the effect of the fall in the cost
  of horsepower after the introduction of the V-8
  engine.
• Lower endpoint, g.
• Used and new cars are not close substitutes.
• The lower quality index for a new car is 9.
• The collusive outcome for 1955 would have led to
  a 0.5 increase in the average 1954-1956 price
  and only to an increase in output of 25.5
  instead of a 40.

48

49
Verboven

• Verboven, F. (1986)
• International Price Discrimination in the
  European Car Market. RAND Journal of Economics,
  27, 240-268.
• This paper addresses the issue of geographic
  market segmentation across Europe in the
  automobile industry (1990).
• The pricing game corresponds to a Bertrand
  oligopoly with differentiated products.
• Markups will depend on
• Own price elasticity of different models.
• Existence of an import quota.
• Collusive behavior (difficult to identify).

50

• Feature There is important price dispersion of
  prices of identical models across Europe.

51

• Arbitrage costs.
• Policy issues.
• Regulation 123/85 on selective and exclusive
  dealer distribution system if price differences
  are not excessive.
• They were exempted from the application of
  Article 85.1 of the Treaty of Rome on integration
  of the market and promotion of competition.
• This exemption was extended for another 10 years
  in 1996.

52

• Other regulation.
• In 1990 there were additional legal and
  administrative difficulties to legally buy cars
  across the border.
• Standards on safety, lights, right-hand driven
  cars,
• Business practices.
• Dealers did not (and still might not) honor car
  market warranties.

53

• Industry characteristics (leading to make
  arbitrage difficult)
• Distribution network externalities.
• Highly concentrated industry on a national basis,
  in particular in France and Italy.
• Different presence of Japanese (most of
  non-European cars) models.
• This is closely related to the existence and
  level of import quotas.

54

55

• Model.
• There are F multiproduct firms operating in M
  national markets.
• A firm f sell a subset Ffm of the Jm car models
  sold in each market m.
• Sales of model j in market m, qjm(Pm) depends
  only of the vector of domestic prices Pm(P1m,
  P2m,,PJmm).
• The cost of producing a car j, Cj(qj1(P1),,qjM(PM
  )), is a function of the sales of car j in the M
  markets.

56

• Profits
• Where the wholesale consumer price is defined as
• Observable deviations such as VAT are captured by
  tm.
• Non-observable deviations such as fixed effects
  and dealer markups are captured by tm.

57

• Import restrictions
• Italy Foreign firms demand cannot exceed some
  absolute level.
• France, Germany, and U.K. Foreign firms demand
  cannot exceed a certain percentage of total
  market demand.

58

• Profit maximization
• Absolute import quota.
• Percentage import quota.

59

• Observe that multiproduct oligopolists consider
  the effect of several prices that they set.
• Economies of scope are not considered (due to
  lack of information).
• There are JJ1 J2 JN first order conditions.
• Pure-strategy equilibrium is assumed to exist.
• Parameters l represent the Lagrange multiplier of
  the corresponding import quota restriction.
• Inverting the FONCs we obtain the pricing
  equations
• Price depends on marginal costs, markup, and
  import quotas.
• Collusion also could affect the effective markups
  through different coalitions that define a set of
  Ffm prices.

60

61

• Marginal cost.
• Where
• Wjm is the vector of physical characteristics of
  car j in market m.
• Qj represents the world sales of car j in all M
  markets.
• wm is the fixed effect on marginal cost of car j
  in market m.
• wjm includes the effect of all the unobserved
  characteristics on marginal costs.

62

• Own and cross demand derivatives in Djm-1.
• Lm potential consumers in market m buy Jm car
  models or an outside good.
• Car models in market m is partitioned in Gm1
  groups.
• Groups are defined by class mini and small,
  medium, large, executive, luxury and sport.
• Each group g 1,,Gm is further partitioned in
  Hgm subgroups, h 1,,Hgm.
• Subgroups defined by country of origin (domestic
  vs. foreign).
• Each group and subgroup is defined by a common
  characteristic of car models.

63

• Identification restrictions
• Lagrange multipliers are only identified up to a
  scale factor.
• It is assumed that the value is common for all
  Japanese firms subject to the import quota in
  market m.
• Factor prices Wjm and the vector of physical
  characteristics Xjm are orthogonal to unobserved
  components wjm and xjm.
• Again, this better suited in the short run.

64

65

• Results
• Strong effect of width on willingness to pay.
• German and European built American cars are more
  valued.
• French cars are most valued in France.
• Important economies of scale.
• Significant cost disadvantage of German and
  European built American cars.
• Productivity issues? Trade restrictions? Model
  features?
• Systematic country differences in cost may
  capture different degree of competition among
  dealers.
• Binding quotas in France and Italy.
• Consistent logit structure structure (car type
  and country of origin).
• Domestic cars have small elasticities taste bias
  or networks.

66

67

• There is strong evidence of international price
  discrimination.
• Significant markup variation across countries.
• Significant markup variation across classes.
• Markups are increasing in car size. Domestic
  producers make more than one model (second degree
  price discrimination?).
• Import quota is binding.