Transcript The Role of Econometric Analysis in Antitrust

Mergers among firms that
manage revenue:
The curious case of hotels
Luke Froeb
Vanderbilt University
May 17, 2008 (10:20am)
“New Perspectives on Competition Policy“
Truland, IIOC, Arlington, VA
• “…an economist is
somebody who sees
something happen in
practice and wonders if it
will work in theory."
2
Joint work
• Arturs Kalnins
– School of Hotel Administration, Cornell University
• Steven Tschantz
– Mathematics, Vanderbilt University
3
Summary of Findings
• Empirical Finding: Hotel in-market mergers
– Relative to in-market non-merging;
• increase capacity utilization 3%; reduce price 1%
– Relative out-of-market merging
• increase capacity utilization 3%; same price
• Theoretical Mechanisms:
– Post-merger information sharing
– Post-merger referrals to sister hotels
– Post-merger loyalty to merged hotels
• Antitrust Policy: short run gain from merger,
“call-arounds”
4
Talk Outline
• Empirical Finding
• Revenue management heuristics
• Can we find a theory to explain the finding?
– Post-merger information sharing
– Post-merger referrals to sister hotels
– Post-merger loyalty to merged hotel
• Antitrust Policy
– Mergers
– “Call arounds”
5
Data
• Texas Comptroller of Public Accounts.
– Owner, address, rooms, quarterly revenue.
– entry and exit dates
– ownership transfer
• Smith Travel Research (proprietary)
– 1999Q2 -2005Q3, self-reported
• Larger, brand-affiliated (82%) hotels
– average price per room-night (Price)
– room-nights sold (Quantity)
6
“10th Closest” Local Merger Area
10th closest unit
Unit changes
to “green”
ownership;
increases HHI
of local area
“Green”
owner’s other
unit
Descriptive Statistics
Definition of
“local area”
Hotels in
mergers
Num. of
rooms
Occupancy
Price
(ADR)
10 closest
20 closest
25 closest
30 closest
40 closest
50 closest
All of TX
51
79
91
99
111
135
889
110
120
116
117
120
120
121
66.35%
66.58%
66.18%
65.95%
66.26%
65.74%
65.08%
$64.22
$66.83
$66.02
$65.80
$66.15
$66.62
$64.51
Non-merging
868
98
61.68%
$59.00
8
Fixed-Effects Regressions
• Data
– 196 Texas hotel mergers (889 hotels) from 1999-2005
– Which increase local HHI
• Effects
– Hotel dummies
– Year X Type dummies
• Type s: urban, suburban, small town, highway, airport and resort
– AR(1)
• Owner characteristics
– First year of new owner
– Experience of owner
– Number of other hotels
9
Regression: Mergers Increase Q
Dependent Variable: utilization
rate
Hotel that Merged Locally
(raises HHI within merger area)
Hotel that Merged Distantly
(raises HHI of state, not merger
area)
Hotel within Area of Merger
(but did not participate in the
merger)
First Year of New Owner
Log Count of Owner’s Hotels
Log Years Owner in Business
F test; Ho: Local = Distant
F test; Ho: Local = Within Area
Local area definitions (closest #)
10
.018+
(.011)
.003
(.002)
20
.019*
(.008)
.002
(.002)
25
.020*
(.008)
.002
(.002)
30
.016*
(.008)
.002
(.002)
40
.012+
(.007)
.002
(.002)
50
.013+
(.007)
.003
(.002)
.001
(.006)
-.002
(.004)
-.002
(.004)
.000
(.003)
.006+
(.003)
.014**
(.003)
-.024**
(.003)
-.003
(.003)
.008**
(.003)
-.024**
(.003)
-.003
(.003)
.008**
(.003)
-.024**
(.003)
-.003
(.003)
.008**
(.003)
-.024**
(.003)
-.003
(.003)
.008**
(.003)
-.024**
(.003)
-.003
(.003)
.008**
(.003)
-.023**
(.003)
-.003
(.003)
.008**
(.003)
2.01
1.98
3.62+
5.05*
4.78* 2.77+
6.51** 3.72+
1.37
.550
2.12
.020
10
Regression: Mergers reduce Price
Dependent Variable: price (avg.
rev.)
Hotel that Merged Locally
(raises HHI within merger area)
Hotel that Merged Distantly
(raises HHI of state, not merger area)
Hotel within Area of Merger
(but did not participate in the merger)
First Year of New Owner
Log Count of Owner’s Hotels
Log Years Owner in Business
F test; Ho: Local = Distant
F test; Ho: Local = Within Area
Local area definitions (closest #)
10
-.866
(.693)
-.918**
(.107)
.537
(.391)
-.553**
(.214)
.709**
(.162)
-.467**
(.165)
20
-1.291*
(.540)
-.892**
(.110)
.256
(.253)
-.542*
(.214)
.725**
(.162)
-.466**
(.165)
25
-1.301*
(.514)
-.890**
(.110)
.271
(.227)
-.536*
(.214)
.727**
(.162)
-.463**
(.165)
30
-1.162*
(.496)
-.874**
(.111)
1.13**
(.216)
-.521*
(.214)
.730**
(.162)
-.473**
(.165)
40
-1.181*
(.472)
-.855**
(.112)
.818**
(.201)
-.517*
(.214)
.714**
(.162)
-.470**
(.165)
50
-1.104*
(.447)
-.851**
(.112)
.883**
(.194)
-.510*
(.214)
.729**
(.162)
-.474**
(.165)
.010
3.210+
.510
7.18**
.590
8.38**
.310
19.2**
.430
16.5**
.290
18.2**
11
12
Talk Outline
• Empirical Finding
• Revenue Management Heuristics
• Which theory can explain the finding?
– Post-merger information sharing
– Post-merger referrals to sister hotels
– Post-merger loyalty to merged hotel
• Antitrust Policy
– Mergers
– “Call arounds”
13
Canonical Rev. Management Problem
• Firms set price before demand realized
• Fixed capacity, (big fixed or sunk costs, small
marginal costs)
• Q=Min[demand(price), Capacity]
• Price to fill ship, hotel, parking lot
– Max{revenue} Max{profit}
14
Rev. Mgt. pricing models:
minimize expected pricing errors
• Cost of over-pricing is unused capacity
– Q(P-MC) [Could have sold more]
• Cost of under-pricing is excess demand
– P(Q) [Could have charged more]
• Optimal P minimizes E[error costs]
– Prob[over-pricing]*Cost[over-pricing] +
Prob[under-pricing]*Cost[under-pricing]
15
Typical Profit Curve
with a Rounded Peak
profit
3500
3000
2500
2000
1500
1000
500
price
60
80
100
120
140
16
Non-binding capacity constraint:
Under-pricing errors more costly
profit
3500
3000
2500
2000
1500
1000
500
price
60
80
100
120
140
17
Expected profit curve:
avoid under-pricing
profit
3500
3000
2500
2000
1500
1000
500
price
60
80
100
120
140
18
Binding capacity constraint:
Over-pricing errors more costly
profit
3500
3000
2500
2000
1500
1000
500
price
60
80
100
120
140
19
Expected profit curve:
avoid over-pricing
profit
3500
3000
2500
2000
1500
1000
500
price
60
80
100
120
140
20
It takes a lot of uncertainty to make a
noticeable difference
Vanderbilt University
21
Early merger model:
CompetitionMonopoly
• Merger 
monopoly competition
• No effect if
capacity
constrained
Price
MC
– Dowell (1984)
Quantity
MR
22
Game-theory
merger models:
Parking lots
• J. E’metrics (2003)
• Constraints on
merging lots
attenuate price
effects by more
than constraints
on non-merging
lots amplify them
• Accounts only for
“original” not
“reflected”
demand
• Certainty
equivalence
23
Rev. Mgt. Merger Heuristics
• Unilateral effect for unconstrained hotel:
– Increases under-pricing error costs because a
decrease in price steals share from sister hotels
• Info sharing: fewer pricing errors
– Fewer over-pricing errors  higher utilization
• Referrals: reduce under-pricing error costs
– Hotel can refer over-booked customers to sister
hotel
• Loyalty: reduces under-pricing error cost
– Increases future demand for hotel “network.”
– Role of merger?
24
25
Talk Outline
• Empirical Finding
• Revenue Management Heuristics
• Which theory can explain the finding?
– Post-merger information sharing
– Post-merger referrals to sister hotels
– Post-merger loyalty to merged hotel
• Antitrust Policy
– Mergers
– “Call arounds”
26
Post-merger information sharing
• Our hotel participates in call-arounds regularly,
daily at 8am, 6pm, and 11pm. We will ask [for all
proximate properties] availability, rate, number of
arrivals, and how many rooms are left to sell.
Hotels that are not among the Midway Hotel
Center [i.e., not operated by the same
management company] participate as well, but
front desk attendants will give false information
because they are too lazy or don’t care enough to
give accurate numbers.
– Hampton Inn, Chicago Midway Airport.
27
Post-merger info-sharing
• Analogous to the difference between
– expected profit maximization (uncertainty); and
– deterministic profit maximization (no uncertainty)
• Fewer over-pricing errors  higher utilization
– Price can be higher or lower.
• Can we illustrate this effect in a game
theoretic context?
– if we ignore over-booked customers
28
Game theoretic model: Poisson
arrivals on top of logit choice
model
• Poisson arrival process
with mean µ
• On top of n-choice
logit demand model
• Implies n independent
arrival processes with
means (siµ)
29
Sampling Uncertainty vs. Parameter
Uncertainty
• Gamma(α, β) prior on
unknown mean arrivals
– Conjugate to Poisson
• Each firmi observes
fraction βi (common
knowledge), and gets a
private signal αi
successes.
• Firm’s posterior
information characterized
by Gamma(α+αi, β+βi) on
unknown µ
Vanderbilt University
30
Nash Equilibrium
• Optimal price maximizes expected profit as a
function of own signal, pi(αi)
• Expectation over all possible signals and all
possible quantities
31
32
Talk Outline
• Empirical Finding
• Revenue Management Heuristics
• Which Theory can explain finding?
– Post-merger information sharing
– Post-merger referrals to sister hotels
– Post-merger loyalty to merged hotel
• Antitrust Policy
– Mergers
– “Call arounds”
33
Post-merger referrals to sister hotels
• We do refer, and referrals account for a substantial part
of our sales. We first refer to the properties owned by
our same owner. These are our sister hotels. But if our
sister hotels are full, we will refer to other nonaffiliated hotels. We get very few referrals from hotels
that are not our sister hotels because most of our
competitive set are chains that have sister hotels of
their own that they refer to. We do get other referrals
occasionally and these are the people [the other hotels]
we refer to when sisters are at full occupancy.
– General Manager, Hotel Lombardy, Washington, DC
34
Post-merger referrals
to sister hotels (cont.)
• In 2000, Hilton bought Promus Hotels (4
brands and 1,700 hotels)
– “After the acquisition, … when there wasn't a
room available in the Hilton … [we would] … crosssell them to the Embassy Suite or Double Tree
Hotel in Times Square. And at last count, starting
in 2000, we run on an annual basis about US $400
million in cross-sell revenue.”
35
Referral Demand Model
• First choice (“original”) demand for 1
• Overflow demand from 12
• Total demand for 1:
– Integration over four states : both, neither, one
– Referrals matter if one of hotels is constrained.
36
Referral Model Results
• Unilateral merger Effect
– Price goes up, Quantity goes down
37
Merger Q as % of pre-merger Q
Market share
0.46
0.45
0.44
0.43
pre merger
0.42
post merger
0.41
1
2
3
4
Referral
delta
38
Merger Price effects as % of premerger price
Price
120
119
pre merger
118
post merger
117
116
1
2
3
4
Referral
delta
39
40
Talk Outline
• Empirical Finding
• Revenue Management Heuristics
• Which Theory can explain the finding?
– Post-merger information sharing
– Post-merger referrals to sister hotels
– Post-merger loyalty to merged hotel
• Antitrust Policy
– Mergers
– “Call arounds”
41
Repeat business and customer loyalty
• This [walking guests] is particularly important
because hotels are always wary of walking guests
to a property they may not win them back from!
– Manager, Mandarin Oriental, Washington DC
• We take the viewpoint that referrals are
good. When we receive a walked guest this is
viewed as a new customer. We do everything to
make them a regular guest of the property.
– Revenue Manager, Jurys Washington Hotel,
Washington, DC
42
Model of Loyalty Demand
• Pre-merger demand for customer who visited
choice 1 last period.
• Post-merger demand for customer who
visited choice 1 last period (loyalty accrues to
merged hotel)
43
Demand Recursion Equations
(to compute steady state demand)
• Pre-merger
• Post-merger
44
Loyalty Results
• Usual unilateral effect:
– Price goes up, Quantity goes down
45
Talk Outline
• Empirical Finding
– Hotel mergers reduce P and increase Q
• Revenue Management Heuristics
• Which theory can explain the finding?
– Post-merger information sharing
– Post-merger referrals to sister hotels
– Post-merger loyalty to merged hotel
• Antitrust Policy
– Mergers
– “Call arounds”
46
Antitrust Policy: Mergers
• Parking, cruise lines, hotel/casinos, hospitals
• In short run, empirical results suggest a short
run gain
– Consistent with info-sharing;
– not consistent with referrals or loyalty
• In long run, with capacity adjustment, mergers
may be anti-competitive, but
– Entry
– Product repositioning
47
Do “Call-arounds” = Collusion?
• Investigation of high-end Paris hotels followed TV show.
– Ritz employee explained (on-camera) how regularly
exchanging data helped each hotel analyze competitors.
• Competition Council: “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”
– Fines from $65,000 to $292,000 for the Crillon
• Over-deterrence?
– EU managers now afraid to share info.
• Hotel exec’s : call-arounds used for forecasting
– And “to bring more people to the area and to
maximize hotel utilization”