2011.George.Mason.MergerTools

Report
Choosing Among Tools for
Assessing Merger Effects
Mason AG Education Program
Advanced Policy Institute on Antitrust Economics
26 June 2011
George Mason Law School
Luke Froeb
Vanderbilt University
[email protected]
Acknowledgements
• Gregory Werden, US Dept of Justice
• coauthor
• Michael Doane, Competition Economics, LLC
•
Consulting partner
• Forbes Belk, Competition Economics, LLC
• Research Assistant
“Take-aways”

Models help agencies figure out:
– (i) What matters
– (ii) Why it matters
– (iii) How much it matters.

Finding a model that can describe significant
features of competition is the hard part
– Once that is done, the rest is easy

Do the best with the information you have, and
with where you are in the investigation
Outline

Do we need more than shares and
concentration?
 Beware experts bearing models
– Particularly those with “UPPI”

Every merger is different
– Coated Recycled Board
– Super-premium ice cream
– Parking, Cruise Lines, Paris Hotels

Conclusions
US data: # of Transactions
5000
4500
Number of HSR Transactions
4000
3500
3000
2500
2000
1500
1000
Year
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
1999
1998
1997
1996
1995
1994
1993
1992
1991
1990
1989
1988
Dashed line represents estimated values
1987
1986
1985
1983
1982
1981
0
Notes:
1984
Threshold changed in January 2001
500
US data: Investigation Rate
Second Requests as a Percentage of Total Filings
5.00%
4.50%
4.00%
3.50%
3.00%
2.50%
2.00%
1.50%
1.00%
0.50%
Threshold changed in January 2001
Notes:
0.00%
Dashed line represents estimated values
1994
1995
1996
1997
1998
1999
2000
2001
2002
Year
2003
2004
2005
2006
2007
2008
2009
2010
FTC Merger Challenges,96-03
90
80
Number of Markets
70
60
50
40
30
20
10
0
2 to 1
3 to 2
4 to 3
5 to 4
6 to 5
7 to 6
Significant Com petitors
Enforced
Closed
8+ to 7+
FTC Merger Challenges,98-07
120
Number of Markets
100
80
60
40
20
0
2 to 1
3 to 2
4 to 3
5 to 4
6 to 5
7 to 6
8 to 7
Significant Competitors
Enforced
Closed
9 to 8 10 to 9
10+
Can shares or concentration
predict merger effects?
Empirical search for a “critical”
concentration ratio was fruitless
 Price-concentration regressions
 For differentiated products mergers:

– No clear line between “in” and “out” of market
– Shares not necessarily good proxies for loss of
competition following merger
How well does ∆HHI predict
effects of Bertrand mergers?

Small
∆HHI  small MERGER EFFECT
BIG ∆HHI  BIG prediction error
Using Models in Enforcement
• Analysis of models provides a solid foundation for
enforcement concerns about unilateral merger effects.
• Analysis of models clarifies the precise nature and
determinants of unilateral effects in particular settings.
• Application of models to cases permits a fact-based,
quantitative assessment of unilateral merger effects.
• Models tell us
• 1 what matters, 2 why it matters, 3 how much it matters
Aren’t models built on unrealistic
assumptions?

Behind every competitive effects analysis is
an (implicit) economic model.
– Make the model explicit
– Force economists to make analysis a
transparent “map” from evidence to opinion

Every model makes unrealistic
assumptions
– Key question: does model ignores significant
features of competition that bias predictions?
– “fit” criterion of Daubert
How do we assess model reliability?

No methodology has been shown to predict
effects of real mergers
–
–
–

No coordinated effects theory,
No unilateral effects theory,
No market concentration theory.
Subject each significant modeling choice to:
–
–
If it matters, have a justification or do sensitivity
analysis and “bound” your conclusions.
Werden, Gregory J., Luke M. Froeb, and David T.
Scheffman, A Daubert Discipline for Merger
Simulation, Antitrust, 18:3 (Summer, 2004) pp. 89-95.
Does modeling sway decisionmakers at agencies?

Merger simulation is a standard
methodological tool
– No tool is definitive.
– Used to organize evidence, not to substitute for it.

First used in 1994 in US v. IBC
– Expert declaration published in Int’l J. Economics
of Bus. with five other examples from real cases.

Use in litigated cases
– Lagardere; Oracle/Peoplesoft;
Doesn’t simulation always predict a
price increase?

Every anticompetitive theory predicts
price increase
– We have safe harbours for concentration

Use simulation to organize evidence,
focus investigation, benchmark
efficiency claims, evaluate remedies.
– Can compute cost reductions that offset
price increase.
Price Competition
(Bertrand models)
• Suppose that you discover
• 1. consumers choose among alternatives on basis of price
& quality.
• 2. firms compete on the basis of price only
• 3. No entry, exit, repositioning, promotional or
advertising , capacity constraints
• Then, Bertrand model may be appropriate
• What next?
Compensating
Marginal Cost Reductions
• CMCRs are the reductions in marginal costs that exactly
offset the unilateral price effects of a merger.
• CMCRs have been used since the mid 1990s.
• Calculating CMCRs does not require assuming functional
forms: a function of own and cross elasticities ONLY.
• CMCRs can be used in a quantitative analysis or just to identify the
key determinants of unilateral effects.
Cournot CMCR
21 2
∆
=
(1 + 2 )
(1 + 2 )
• s1 and s2 are the quantity shares of the merging firms, and
e is the market demand elasticity.
• Example: if e=1, s1=s2=20%, MC have to go down by 20%
to offset incentive of merged firm to raise price
• Smaller CMCR’s with
• More elastic demand (e)
• Smaller ∆HHI
Bertrand CMCRs
   +    
1 −  
• This expression applies to single product firms, but it
generalizes to multiproduct firms.
• dij and dji are diversion ratios between merging products;
mi and mj are their margins; and pi and pj their prices.
Approximate Bertrand CMCRs
   
• This expression is essentially a first-order approximation
to the CMCRs and omits terms containing dij dji .
• If the diversion ratios are relatively low, this expression
provides a fairly good approximation to the CMCRs.
• Example: {equal prices, margin=50%, diversion=10%}
5% reduction in price of each good necessary to offset
incentive to increase price
Pricing Pressure Indicies
• Salop and O’Brien observed that the first-order impact of a
Bertrand merger on prices is determined by mj dij.
• Farrell and Shapiro proposed gross (GUPPI) and net (UPPI)
upward pricing pressure indexes scaled in monetary units.
• GUPPI = mj dij pj, is the profit recapture for one merged
product as the price of another is increased.
• UPPI = GUPPI – 10% of marginal cost.
UPPI Rescaled
    − (1 −  ) 10
• This expression is the Farrell and Shapiro UPPI after dividing
by pi to convert monetary units into a pure number.
• The first term, the GUPPI, is the approximate CMCR.
• The second term is the arbitrary 10% efficiency credit.
• BOTTOM LINE: ONCE YOU KNOW BERTRAND IS
APPROPRIATE, CHOICE OF TOOL DOES NOT MATTER
US v. Altivity and Graphic
(2008)

US DOJ challenges CRB Merger
– Altivity (35%) + Graphic (17%) of North American
capacity

Remedy
– Divest 2 plants representing 11% of capacity
Model of Harm
Nick Hill, “Mergers w/capacity closure,” DOJ
working paper
 Model: Once built, mills produce at capacity;
and merger would create incentive to close one
or more mills

– Mill shutdown  supply decrease  higher price for
remaining production
– Merger changes the usual “shut down” calculus to
make it more profitable to shut down
How modeling can help an
enforcement agency

Model tells you:
– 1. What matters
 Elasticity of demand for CRB
 Elasticity of foreign supply
– FX, transport cost, other commitments

Facility & closing costs
– 2. Why it matters
 Increases profitability of shut down
– 3. How much it matters
 Which divestitures are sufficient?
Does the model capture
significant competition?

Product Market: CRB
 Geographic market: North America
 Is CRB market operating at near capacity?
 Can model predict what we can observe?
– Pre-merger: NOT profitable to shut down
– Post-merger: profitable to shut down mill
FTC v. Nestlé and Dreyer
(2002)

Super-premium ice cream in North America
– Nestlé 36.5% + Dreyer 19.5% revenue share

Remedy: divest 3 brands to new entrant
Models help delineate markets

Question: Is super-premium a relevant
product market?
 Answer: Simulate merger-to-monopoly of
four super-premium ice cream producers

If price goes up by 5% then it is a relevant product
market
Inputs to unilateral effects analysis:
Own- and Cross-Elasticity Estimates
Tenn et al., “Mergers when firms compete
using price and promotion,” Int’l. J. Ind. Org.
Brand A
With respect to a price increase by:
Brand A
Brand B
Brand C
Brand D
-1.67
0.08
0.13
0.03
(0.06)
(0.01)
(0.02)
(0.00)
Brand B
0.20
(0.02)
-1.76
(0.06)
0.16
(0.03)
0.03
(0.01)
Brand C
0.13
(0.02)
0.06
(0.01)
-1.61
(0.06)
0.02
(0.00)
Brand D
0.16
(0.03)
0.07
29
(0.01)
0.14
(0.02)
-1.90
FTC
(0.07)
Models help interpret data

Question: how did new entrant Dreyer
obtain a 20% share without affecting
incumbent price?
– Does this mean that super-premium is not a
relevant antitrust market?

Answer: Build a model of post-merger
world, simulate exit (by raising Dreyer’s
MC), and see what happens to price
– Does incumbent pricing change?
Models help interpret data
(continued)

Question: How does promotional activity
affect merger analysis and tools that
economists use?
– What happens if we ignore promotional
activity?

Answer: Build a model of promotion +
price.
– If promotion affects elasticity, then it matters; if
not then it doesn’t
Demand, prices, and promotion level
1.None, 2.display, 3.feature, 4.both

Table 4: Elasticity Varies with Promotion

Own-price
No
Promotion
-1.62
(0.07)
Display
Only
-1.87
(0.24)
Feature
Only
-1.88
(0.15)
Feature &
Display
-2.29
(0.23)
Brand B
-1.66
(0.06)
-1.96
(0.24)
-1.94
(0.15)
-2.30
(0.22)
Brand C
-1.56
(0.07)
-1.80
(0.22)
-1.75
(0.14)
-2.24
(0.22)
Brand D
-1.80
(0.08)
-2.31
(0.28)
-2.19
(0.18)
-2.70
(0.25)
Brand A
33
FTC
Answer: promotion matters in
this case

Price-only merger models under-predict
(5% instead of 12%) the price effects of
mergers in industries where firms compete
using price and promotion
– Estimation bias: demand is too elastic
– Extrapolation bias: promotion decreases 31%
in post-merger equilibrium
34
Vanderbilt
Estimation Bias vs. Extrapolation Bias

B,C merge
Control for Promotions in:
Demand
Merger
Estimation?
Simulation?
Yes
Yes
Yes
No
No
No

% Change Price
Brand A
0.1%
0.1%
0.0%
% Change Quantity
Category
Brand B Brand C Brand D Index
8.1%
2.4%
-0.1%
2.2%
6.3%
2.0%
0.0%
1.8%
3.3%
1.1%
0.0%
0.9%
Brand A
0.7%
0.5%
0.4%
Category
Brand B Brand C Brand D Index
-13.5%
-3.4%
1.2%
-3.0%
-10.0%
-2.8%
0.7%
-2.3%
-7.4%
-2.4%
0.5%
-1.9%
Merger to monopoly
Control for Promotions in:
Demand
Merger
Estimation?
Simulation?
Yes
Yes
Yes
No
No
No
% Change Price
% Change Quantity
Category
Brand A Brand B Brand C Brand D Index
10.3%
19.9%
9.4%
17.0%
11.7%
8.9%
16.2%
8.1%
14.5%
10.0%
4.4%
7.8%
4.1%
7.2%
4.9%
35
Brand A Brand B Brand C Brand D
-13.3% -26.9% -11.9% -24.4%
-11.5% -21.6% -10.1% -20.6%
-9.3%
-15.6%
-8.2%
-16.0%
Vanderbilt
Category
Index
-15.3%
-12.9%
-10.1%
Parking lot merger

1999 Central Parking $585 million
acquisition of Allright.
 Remedy: divestitures if merged share
>35% in 4X4 block area is
– Divestitures in 17 cities
Models help economists
understand competition in
different settings

Froeb et al. (2002) criticize DOJ by arguing
that the merger would not have raised price
because there is very little uncertainty about
parking demand.
 Price to fill capacity, pre- and post-merger
– Pricing practice: “is the lot full by 9am?”

Capacity constrained  no merger effect
– How many of the lots are capacity constrained?
Model of
downtown

16 blocks
 3 lots
 Building
height
represents
demand
Cruise line merger:
What about uncertainty?

2003, the European Commission (EC) gave
their approval to Carnival's $5.5 billion
takeover of rival cruise operator P&O
Princess
– Followed UK and US approvals
Deterministic profit function w/
tightly binding capacity constraint
profit
3500
3000
2500
2000
1500
1000
500
price
60
80
100
120
140
Expected profit function (solid)
w/tightly binding constraint
profit
3500
3000
2500
2000
1500
1000
500
price
60
80
100
120
140
Paris hotels:
Trying to reduce uncertainty?

2005, six luxury hotels in Paris exchanged
information about occupancy, average room
prices, and revenue
– French competition agency: "Although the six
hotels did not explicitly fix prices, …, they
operated as a cartel that exchanged confidential
information which had the result of keeping prices
artificially high" (Gecker, 2005)
– industry executives insisted that their information
sharing was to "to bring more people to the area
and to maximize hotel utilization"
Testing for merger effects

US Price and occupancy data from Smith
Travel Research (STR).
– 32,314 U.S. hotels reported to STR the average
room-night price actually received each day, as
well as the total number of rooms available and the
number of rooms sold.
– 97 monthly observations from 2001 –2009 for each
hotel for occupancy and price.
– These 32,314 hotels represent about 95% of chainaffiliated properties in the United States and about
20% of independent hotels and motels.
Results

Relative to non-merging hotels, mergers increase
occupancy
– Gain $1700-$3300 per month for a 100-room hotel.

But only in capacity-constrained and uncertain
markets
– Mergers allow hotels to better forecast demand.

No evidence hotel mergers decrease occupancy or
raise price.
– “traditional” models would not predict this
Conclusions
• Models help. But finding the right model is hard.
• Do the best you can with what you have where you are.
• An agency should make the best possible use of the information it
has at each stage of a merger assessment.
• One size does not fit all.
• An agency should determine which of the many specialized tools to
apply by first understanding of how competition works and thus
which model, if any, fits the industry.

similar documents