Calibrated Economic Models

Download Report

Transcript Calibrated Economic Models

Quantitative Analysis
Of Competitive Effects
For Antitrust
Day 1
Luke Froeb
Owen Graduate School of Management
Vanderbilt University
April, 2003
Case Studies Showing How
Modeling Is Used in Antitrust
WorldCom and Sprint
Branded Consumer Product
Carnival and Princess
Luke Froeb
Owen Graduate School of Management
Vanderbilt University
WorldCom and Sprint Merger:
Background
 Merger
scrutinized by
– U.S. Department of Justice
– Federal Communications Commission
– Interested third parties like Bell Atlantic
 Overlap
in residential long distance service
 Regulatory restrictions keeping local phone
companies out of market were soon to fall
WorldCom and Sprint Merger:
Methodology
Estimate consumer
choice model
(demand)
 Estimate/Calibrate
Firm Model (supply)
 Simulate “ownership”
effect of merger

Share
Price
World 16.5
Com
Sprint 6.9
16.5
AT&T 56.4
17.6
Others 20.1
14.4
17.9
WorldCom and Sprint Merger:
Simulated Merger Effects

Demand estimation from bill “harvesting”
– Inelastic demand
– Predicted price increases for merging firms
WorldCom: 5.4%

Sprint: 8.9%
Calibration from WorldCom’s margin
– Small margins imply more elastic demand
– Predicted price increases for merging firms
WorldCom: 2.2%
Sprint: 5.1%
WorldCom and Sprint Merger:
Mergers in a Post Entry World
 Does
entry occur in response to merger?
 Entry by incumbent local exchange
carriers (“Baby Bells”)
– State level “experiments” show 25% share
of long distance
– What is merger effect in post entry world?
 Baby
Bell entry cuts the industry average
price effect in half
Branded Consumer
Goods Merger
 This
is a real case that must be kept
confidential, so numbers are disguised
 Entrant
gained 25% share in two years
 Proposed
 Only
to purchase rival brand
3 brands in “high end” segment
 Aggregate
segment elasticity 1.5
Branded Consumer Goods Merger:
Brand Elasticity, Prices, Shares
A Price B Price C Price
Price Share
A Quantity 2.21
0.44
0.28
$20
35%
B Quantity 0.39
2.16
0.28
$20
40%
C Quantity 0.39
0.44
2.32
$20
25%
Branded Consumer Goods Merger:
Findings
 Estimated
demand implies brands are
fairly good substitutes for one another
 Predicted industry price increase of 4%
 Merging
and 8%
 12%
brand price increases of 5%
and 18% marginal cost reductions
required to offset price increases
Is Merger Prediction Consistent
with Entry Experience?
 Incumbent
brands did not reduce price
in response to entry with a 25% share
 Implies entrant is bad substitute
 Yet,
we get a significant price increase
following merger
 Implies entrant is good substitute
 Is
postmerger price increase consistent
with no incumbent entry response?
How to Answer Question
 Calibrate
model to observed data
 “Undo”
entry by raising price of recent
entrant until zero quantity
 Compare
price changes of remaining
(incumbent) brands
 Entry
effects are reverse of “undoing”
entry effects
Consumer Goods Merger:
Entry Model
Prices of incumbent
brands barely change
 Quantity drops
significantly
 Entrants steal
quantity, but do not
affect price
 In general, other firms
do not much affect
price

A
B
ΔP –0.85% –0.95%
ΔQ –10.3% –9.98%
Consumer Goods Merger:
How Can We Test this Prediction?
 Where
from?
did entrant’s quantity come
 65% from “outside” option

Includes lower priced brands
 35% from incumbent brands
 Is
this consistent with entry data?
 If not, modify model
2002 Cruise Line Merger:
Introduction
Carnival (largest) and Royal Caribbean
(second largest) each bid for Princess (third
largest)
 Capacity constraints and “perishable” service

– big fixed costs, small marginal costs

Key strategy is “revenue management”
– Price to match uncertain demand to available
capacity, i.e. to “fill the ships”
Merger of Parking Lot Operators
Central Parking acquired Allright
 Two largest parking lot operators in US
 Pricing: “Is lot full by 9am?”

 If “no,” then reduce price
 If “yes,” then raise price
This profit calculus unchanged by merger
 No merger effect if lots are full
 But Justice Department opposed merger

 Asked for 74 lot divestitures in 18 cities
Cruise Lines vs. Parking Lots

Similar strategies: filling ships vs. filling lots

There is no uncertainty about parking
demand, but does that make a difference?

Theories considered by FTC
 Fill-the-ship pricing is unaffected by merger
 No quantity effect, but low elasticity consumers
pay more

Were theories correct? Magnitude of effects?
Conclusions Based on Formal Model
of Revenue Management

Two merger effects
– Ownership effect raises price
– Information sharing effect raises or lowers price
But always increases quantity
Both effects small and disappear as
uncertainty decreases
 Confirms basic intuition from parking lot
merger, i.e. firms price to fill the ships, and
this profit calculus is unchanged by merger

Differentiated Products Mergers
What if we can’t estimate demand?
Modelling choices and trade-offs
Accounting for efficiencies
Luke Froeb
Owen Graduate School of Management
Vanderbilt University
Model Based Methodology
 Specify
and estimate a model
– Consumer model (demand)
– Firm model (supply)
 Use
model to simulate counterfactual
scenario
– Mergers, collusion, damages
Example and Questions
To “Test” Approach

Logit demand curve
– What about other forms?

Price setting competition
– What about product, promotion, placement?
– What about auctions, quantity setting, vertical
arrangements?

Constant marginal cost
– What about scale economies or capacity constraints?

Static game
– What about dynamic strategies?

Unilateral merger counterfactual
– What about coordinated effects?
Critique of Market Share Screens
With Differentiated Products
Competition does not stop at market
boundary
Shares may be poor proxies for
competitive effects
No role for efficiencies
How do you trade off a 10% marginal cost
reduction against a 400 point change in HHI?
Common Problem: Cannot Get
Reliable Demand Estimate
 Relatively
flexible functional forms
often lead to nonsensical estimates
– Goods are complements (when we know
they are substitutes)
– Inelastic demand (inconsistent with
optimization)
 Data
not up to task of estimating so
many parameters
Solution: Ask Less of Data by
Making Intuitive Assumptions

If one firm increases price, its rivals gain
quantity in proportion to existing shares
– Implies all goods are substitutes
– Implies margins proportional to shares
– Implies cross elasticities proportional to shares

These restrictive forms require less data
– Aggregate elasticity
– One brand level elasticity or margin
Replacing
Market Share
Screens



Swedish beer
merger
Aggregate
elasticity ≈ 1
Pripps margin ≈
30%
Significant
industry average
price increase
4%
1.2
Aggregat e Elast icit y

1.4
1
5%
0.8
6%
7%
0.6
2
2.5
3
3.5
Pripps 2.8% Demand Elasticity
4
Logit Model
as a Screen
WorldCom
Sprint Merger
 Aggregate
elasticity ≈ –1
 WorldCom
Margin ≈ 30%
 Relatively
small price
effects

Using Logit Model as Screen (cont.)
 Rebuttable
presumption starts dialogue
– Show products are further apart than modeled
– Show competition is more intense than modeled
 Requires
different safe harbors
– “Grant” parties a 5% MC reduction on each of
their merging products
Would have allowed WorldCom and Sprint, but not
Swedish beer
How Do you Incorporate
Merger Specific Efficiencies?
 Simulate
marginal cost reductions
 Both merging products get lower marginal cost
 Demonstrated merger specific cost reductions
 Price effect depends on inherent properties of
assumed functional form for demand
 Compute
marginal cost reductions that
keep all prices at pre merger levels
 In general, 2X to 3X times computed price increase
is enough
Which Welfare Standard?
Total vs. Consumer Welfare
 Do fixed costs count as such?
 Superior Propane: fixed cost savings but
prices went up
 Gain in profits was higher than consumer
welfare loss
 Can
fixed cost reductions be passed on
over time?
 Short run price increases vs. long run
efficiencies
Spatial Competition
Grocery Store Merger
Parking Lot Merger
Luke Froeb
Owen Graduate School of Management
Vanderbilt University
Geographic Differentiation
Retail Sector is Consolidating

In US, Wal-Mart, K-Mart, Target, Costco, and
Sears account for 60 percent of general
merchandise sales
– General merchandise is 15 % of all retail sales

Productivity advantage over smaller retailers
– Economies of scale
– Economies of purchasing
– Economies of distribution
Productivity Gains Associated with
Industry Consolidation
Retail Consolidation
Also in Europe
 In
EU, top 10 grocery stores forecast to
increase share to 50 or 60%
– In 2002, top 10 had 38%
 Wal-Mart entering
Europe
Policy Reaction to Retail
Consolidation
 FTC
challenged some retail mergers
– Blocked Kroger and Winn-Dixie
– Blocked Staples and Office Depot
 Competitive
analysis based on increase
in local (within city) horizontal market
power
– “standard” horizontal analysis
Quantitative Horizontal Analysis:
Benefit-Cost Analysis of Merger
 Goal:
effect
quantitative estimate of merger
– Necessary to weigh efficiencies against loss
of competition
 Two
methodologies
– “Natural” experiments, e.g. Staples and
Office Depot
– Model based simulations
Natural Experiments:
Staples and Office Depot

Prices in two office superstore cities found to
be 7% lower than in one office superstore city
– Is this a good metaphor for merger?
Pass-through rate (from cost to prices) was
estimated to be 15%
 This implies that a 85% reduction in costs
necessary to offset merger effect
 Is low pass through consistent with big
merger effect?
 FTC successfully challenged merger

Model Based Simulation
 Model
current competition
 Estimate model parameters
 Simulate loss of competition using
estimated parameters
 Unilateral competitive effect computed
as difference between pre and
postmerger Nash equilibria
Model Based Simulation:
Kroger and Winn Dixie

Estimate “gravity”
choice model
– Survey density in
Charlotte, NC
– Dots represent
grocery stores

Choice depends on
price, distance, and
“noise”
Premerger Equilibrium:
Share of Kroger and Winn-Dixie
Postmerger Equilibrium:
Share of Kroger and Winn-Dixie
Parking Lot
Merger



“Gravity” choice
model
Lots derive
market power
from location,
capacity
Higher prices if
– Few nearby lots
– Many nearby
consumers
– Small lot capacity
Parking Lot Merger Model:
Conclusions
Constraints on merging lots attenuate merger
effects by more than constraints on
nonmerging ones amplify them
 Merger effects poorly approximated by
shares in geographic market areas

– No bright lines between “in” and “out”
– Shares poor proxies for localized competition

Justice Department erred when it asked for
divestitures in 5 square block areas where the
merging firms account for more than 35%
 Ignored role of constraints on merging lots
Gravity Choice Models
And Merger Effects
 Merger effects “depend” on
– Location of consumers
– Location and capacity of merging firms
– Location and capacity of nonmerging firms
– Cost of travel
– Other factors affecting demand
 Weak
generalization: small price effects
if location only source of market power