1

THREE ESSAYS IN EMPIRICAL INDUSTRIAL ORGANIZATION AND MERGER POLICY

A dissertation presented

by

Chengyan Gu

to

The Department of Economics

In partial fulfillment of the requirements for the degree of

Doctor of Philosophy

In the field of

Economics

Northeastern University

Boston, Massachusetts

March, 2015

   

2

THREE ESSAYS IN EMPIRICAL INDUSTRIAL ORGANIZATION AND MERGER POLICY

by

Chengyan Gu

ABSTRACT OF DISSERTATION

Submitted in partial fulfillment of the requirements

for the degree of Doctor of Philosophy in Economics

in the College of Social Sciences and Humanities of Northeastern University

March, 2015

   

3

Abstract

The first chapter is titled “Endogenous Market Structure and Fixed-to-Mobile Competition

in the U.S. Telecommunications Industry.” This paper develops an empirical model to examine the

intramodal and intermodal competition effects within and between the wired and wireless sectors in 1997

and 2002, a period in which wired and wireless carriers entered the local market on a large scale

simultaneously. A two-step procedure is proposed to address the problem caused by the fact that wireless

carriers usually make entry decisions in larger geographical regions than wired carriers. The results show

that compared with national CLECs, regional CLECs and wireless carriers tend to be close competitors, as

the presence of wireless carriers substantially lowers regional CLECs' margins. The intermodal competitive

effect of wireless carriers on regional CLECs is much greater than their effect on national CLECs. It is also

found that providing wired and wireless services together in a market does not lower a CLEC's profitability.

The entry model for wireless sector shows that once the market has five wireless carriers, the next entrant

has little effect on competitive conduct.

The second chapter is titled "Predicting Merger Outcomes: How Accurate Are Stock Market Event

Studies, Market Structure Characteristics, and Agency Decision?" a paper coauthored with John Kwoka.

This paper analyzes two leading methods of predicting the outcomes of mergers, as well as the accuracy of

antitrust agencies’ decisions whether or not to challenge mergers. Drawing on actual price effects of forty

mergers, this paper develops data on market structural characteristics, on stock market event studies, and

on agency decisions. It finds that event studies systematically underpredict the incidence of anticompetitive

mergers, while market structure criteria systematically overpredict competitive concerns. While these might

seem to be opposing errors, market structure serves primarily as an initial screen, so that over-inclusiveness

may be an optimal decision approach.

The third chapter is titled "The Mergers Effects on Telephone Carriers' Efficiency: a Conditional

Difference-in-Difference Approach." This paper examines the merger effects on telephone carriers'

efficiency in the U.S. telecommunications industry over the period of 1996-2007. One limitation of

   

4

traditional difference-in-difference (DID) approach is that if the mergers are not randomly assigned, any

omitted factor that is correlated with the outcome measure as well as the probability of a merger will

generate biased estimates of the impact of merger. Without the need for any instrumental variable, this

paper uses a propensity score matching approach combined with DID method to address selection on both

observables and unobservables associated with merger formation. The results show that mergers reduce

merging carriers' incentive to innovate or invest on frontier technology, rather than lower carriers' technical

efficiency. Moreover, mergers do not speed up carriers' scale efficiency progress as the carriers promised

in the antitrust review processes.

   

5

Acknowledgements

I am grateful for the opportunity to thank all the people who generously offered their help

and support throughout my entire graduate education. First and foremost I would like to thank my

dissertation committee, John Kwoka, James Dana and Steven Morrison for their continuing advice

and guidance. This dissertation was possible only because of the insightful comment and

continuous encouragement provided by them. Especially, I would like to express my special

appreciation and thanks to my advisor, Professor John Kwoka, who has been a wonderful teacher

and mentor over the past several years. His tremendous mentorship on both research and my career

has been the most valuable asset for me to cherish.

Northeastern’s Department of Economics provided an excellent environment for developing

as an applied economist, and I am grateful to all those who taught and helped me over the years.

In particular, I would like to thank William Dickens, Maria Luengo-Prado, Jackson Osborne,

Zhongmin Wang, Gustavo Vicentini for helpful comments and discussion on various parts of this

dissertation. Each of them has made an indelible impact on both my research and training as a

researcher. I would also like to thank Neil Alper, Kathy Downey, Cheryl Fonville, and Gregory

Wassall for their generous helps and guidance during my graduate journey. In addition, I would

thank my friends and participants at various department seminars for their supportive discussions

and helpful suggestions.

Last but not least, I wish to express my deepest appreciation to my family, especially my

parents. Their support has been unconditional all these years. They have cherished with me every

great moment and supported me whenever I needed it. This dissertation is dedicated to them.

   

6

Table of Contents

 

Abstract 2

Acknowledgements 5

Table of Contents 6

Chapter 1 Endogenous Market Structure and Fixed-to-Mobile Competition in the U.S.

Telecommunications Industry

Introduction 9

Industry Background 15

Data 19

Model 24

Empirical Implementation 32

Results 37

Policy Concerns and Experiments 43

Conclusion 47

References 49

Chapter 2 Predicting Merger Outcomes: How Accurate Are Stock Market Event Studies, Market

Structure Characteristics, and Agency Decision?

Introduction 63

Qualifying Mergers 66

Prices and Predictions 68

Results of Empirical Testing 79

Conclusion 89

References 93

Chapter 3 The Mergers Effects on Telephone Carriers' Efficiency: A Conditional Difference-in-

Difference Approach

Introduction 104

Background and Data 106

Empirical Strategy 106

   

7

Results 108

Conclusion 110

References 112  

   

8

Chapter 1

Endogenous Market Structure and Fixed-to-Mobile Competition

in the U.S. Telecommunications Industry

   

9

1. Introduction

The Telecommunications Act of 1996 was intended to transform telecommunications in the U.S., both

by altering the behavior of incumbents and by facilitating the entry of new carriers. Starting in 1997, large

numbers of CLECs (Competitive Local Exchange Carriers) began entering local markets. Almost

simultaneously, the wireless market experienced large-scale entry by PCS (Personal Communications

Service) carriers. During the period of 1983-1994, the wireless industry was largely operated as a regional

duopoly, with one local wired carrier and one independent carrier in each market. This duopoly was quickly

broken by the launch of PCS auctions in late 1995. From that point on, ILECs (Incumbent Local Exchange

Carriers) have had to face competition not only from CLECs but also from their wireless cousins.

This transformation of the telecommunications industry has raised many new issues, such as

mechanism design problems in spectrum auctions, calling party pay vs. receiving party pay, local number

portability, and price discrimination problems when carriers offering bundling services.1 Among the more

important recent issues concerns the relationship between wired and wireless carriers. Are they competitors,

or are they complements to each other? Should we define them into one market or two separate markets?

Generally speaking, if mobile markets are more competitive than wired markets, this competition would

constrain market power in the fixed-line markets, or at least, the combined wired and wireless market would

be more competitive than each individually (Vogelsang, 2010).

These types of questions are of particular policy interests for many reasons. For example, if

competition by wireless carriers has successfully constrained the market power of local fixed-line carriers,

it could provide supporting evidence to justify deregulation in the fixed-line industry. Moreover, it could

shed light on certain antitrust policies. For example, if wired and wireless carriers represent a converged

                                                            1  Carriers often bundle their wired and wireless services together. For example, in 2004 BellSouth offered a

"Dial+Wireless Bundle" including local, long distance, Internet and wireless service at the fixed rate of $78.89 per

month+0.05 per minute. Qwest provided a similar, so-called "Qwest Choice DSL+Wireless Bundle" at the rate of

$107.97 per month for the first year, and $112.97 per month after that. See Crandall's (2005) summary in Tables 5-7.

   

10

market, the effect of a proposed merger between two carriers (e.g. AT&T and T-Mobile) would be smaller

and less likely to trigger antitrust concerns. A conventional method of addressing this question is to measure

consumer substitution and then assess whether it is strong enough to constrain any effort by the merged

firm from raising prices. The myriad difficulties with that approach are well-known (Whinston, 2007;

Farrell and Shapiro, 2010) and occasionally detailed consumer-level data are not available. The antitrust

agencies have come to realize that in some cases that approach is not just infeasible, but also unnecessary.

Where sufficient variation in market structure exists, the effect of eliminating a competitor can be measured

directly, 2 obviating the need to undertake two difficult inquiries-determining the market, then using

concentration changes to predict price change.

Consequently, I ask three specific questions in this study: (1) Is the wireless market competitive

enough? (2) If so, what is the competition effect of wireless carriers on fixed-line carriers? (3) Because

many carriers provide wired and wireless services simultaneously in some market, what is the effect of a

carrier providing both services jointly? To answer those questions, I construct a complete information

sequential entry model with observed and unobserved heterogeneities to estimate the effects of intramodal

and intermodal competition on carriers' profitability.

The fact that wired and wireless carriers usually make the entry decisions at different market levels

raises one econometric challenge here. Generally speaking, cities are the best approximations for CLEC

markets, as CLEC services are inherently locally focused.3 For example, CLECs have to establish their

presences in local markets to connect their business and residential customers. In contrast, largely due to

                                                            2 This approach was first used to great effect in FTC v. Staples (1997) and then FTC v. Whole Foods Markets, Inc.

and Wild Oats Market, Inc. (2007). The Horizontal Merger Guidelines 2010 place less emphasis on the process of

market definition and allow for defining the market and measuring the competitive effects simultaneously. For a

discussion of the Staples case, see Ashenfelter et al (2006). For the Whole Foods case, see Farrell and Shapiro (2010). 3 As Greenstein and Mazzeo (2006) justify, cities best approximate the markets for CLECs, since the CLECs’ services

are inherently local focused, which makes the midsize cities geographically distinct market areas. See also Goldfarb

and Mo (2011) for example.

   

11

the mobility of wireless technology, mobile carriers often make their entry decisions in larger geographical

regions, namely, CMA (Cellular Market Area4) level markets, as has been well documented by a large

number of carrier and antitrust agency files. A naïve approach to modeling the strategic interaction between

wireless and CLEC players at the CMA level would make the entry game intractable. For example, if a

CLEC competes with wireless carriers at the CMA level, it could strategically select its locations within

such a CMA market to replace the marginal wireless carrier. Its strategy space will proliferate quickly with

the number of submarkets and the number of potential entrants, making the usual approach of evaluating

all possible choice combinations to find the profit-maximizing vectors infeasible.

Instead, I tackle this problem using a two-step approach. In the first step, I employ a Bresnahan and

Reiss (1991a) model for the wireless sector, which I then use to generate a CMA-level correction term. An

ordered probit model is constructed to describe the market structures in the wireless sector across cross-

sectional heterogeneous CMA markets. In the second step, I add this correction term to the profit functions

of CLECs to address the possible endogeneity of the number of wireless carriers. The endogeneity arises

because such market structure variables are likely to correlate with the market unobservable. As is common

in the standard selection model (starting with Heckman, 1979), this term is used to adjust the correlation

between the unobservable in wireless carriers' profit functions and the error term in the CLECs' profit

functions. However, this "observed" pure CMA-level correction term cannot fully capture the city-level

unobserved profit shocks (which make the number of fixed-line carriers endogenous). As a result, I adopt

the random coefficient technique proposed by Heckman and Macurdy (1986), Villas-Boas and Winer

(1999), Train (2009) and Petrin and Train (2010). That is, I allow such CMA-level profit shocks to have

some random effects at city-level markets. Then, a method of simulated moments (MSM) estimator is used

to recover the parameters in the CLECs' profit functions.

                                                            4 The Cellular Market Areas (CMAs) were created from the Metropolitan Statistical Areas (MSAs) defined by the

Office of Management and Budget (1-305), the Gulf of Mexico (306), and Rural Service Areas (RSAs) established

by the FCC which do not cross state borders (307-734).

   

12

It is worth mentioning that the current model does have two limitations. First, it does not directly

model the strategic interaction between wireless and wired carriers (if at all) in an entry game. The profit

inequality approach proposed by Pakes et al (2011) and Bajari et al (2007) is a promising method5 because

it avoids detailing the entry process by focusing on profit inequalities rather than equilibrium choice

probabilities. However, even if we adopt this approach, we would still have to make the strong assumption

that carriers cannot benefit from swapping licenses. Second, to generate the correction term, I impose some

distributional assumptions that largely ignore the carrier-level heterogeneities in the wireless market.

Despite all its drawbacks, the current model is general enough to describe the intermodal competition

problem between wired and wireless carriers.

The analysis exploits a unique dataset I collected that covers detailed entry information for both wired

and wireless carriers from 1997 to 2002 in 1,673 midsize U.S. cities (within 292 CMA markets) with

populations between 20,000 and 1,000,000 as of the 2000 census. The results indicate that both national

and regional CLECs are trying to differentiate their services from competitors on the basis of geographical

footprint. The intramodal competitive effects on CLECs primarily come from the same-type CLECs, and

these negative effects increase over time. For intermodal competition, I find that, comparing with national

CLECs, regional CLECs and wireless carriers are closer competitors because they may target similar niches

in the market. By 1997, wireless carriers only had minor effects on national CLECs but significant negative

effects on regional CLECs' margins. Despite the shakeout in 2001, wireless carriers imposed significant

competitive pressures on both types of CLECs by the end of 2002, but the competitive effects on regional

CLECs were still much greater than the effects on national CLECs. Moreover, the significant coefficients

                                                            5 The basic idea of this matching approach is that swapping or deviations from current equilibrium will decrease

current payoff. It eliminates the unobserved heterogeneities using a double difference-in-difference framework.

Several papers have applied this approach to various settings. For example, Pakes and Ho (2013) estimate the demand

function in the hospital industry. Bajari and Fox (2013) employ this approach to investigate the FCC's C-block auction

problem. Akkus et al (2012) adopt this approach to describe the merger choice in the bank industry, and Ellickson et

al (2013) use it to model the entry problem of retail discount industry.

   

13

on the correction term and its standard deviation indicate the need to correct the endogeneity of the market

structure and the importance of including the market and firm-specific unobserved heterogeneities. The

results from the entry model for wireless sector show that the top five wireless carriers significantly decrease

mobile carriers' margins, while the 6th to 8th competitors only have minor effects on profits. Entry threshold

tests show that at least five wireless carriers are needed to keep the wireless sector sufficiently competitive.

My empirical model fits into the entry literature beginning with Bresnahan and Reiss (1987, 1991a,

1991b) and Berry (1992). The basic insight of this type of empirical entry model is that parameters in the

profit function can be inferred from firms’ revealed preferences, which should be based on a Nash

equilibrium of interacting firms’ entry decisions. Occasionally, however, such Nash equilibria are not

unique, or even worse, do not exist. Various solutions have been proposed to solve such multiple equilibria

problem (Berry and Reiss, 2007), such as focusing on the number of firms (Bresnahan and Reiss, 1987,

1991a, 1991b; Berry, 1992; Mazzeo, 2002a), placing additional orders of the entry process (Berry, 1992;

Mazzeo, 2002a; Jia, 2008; Goldfarb and Mo, 2011), assuming firms only have incomplete information on

other firms' profitability (Seim, 2006), or using the bounds identification technique (Ciliberto and Tamer,

2009). Ellickson et al (2013) use the profit inequality approach to study the entry problem in the retail

discount industry. This approach does not require that researchers solve for any type of equilibria and only

requires that the observed choices yield weakly higher expected payoffs than any feasible alternatives, while

holding the action of rivals fixed. In this study, I follow in the tradition of Berry (1992), focusing on the

equilibrium number of carriers in each market. Specifically, carriers are assumed to be post-entry symmetric

and to move in order of profitability.

This study is also related to the entry literature in the telecommunications industry. Abel and Clements

(2001) and Abel (2002) examine how regulatory regime changes would affect the number of new entrants.

Mini (2001) and Alexander and Feinberg (2004) investigate whether incumbent carriers try to deter entry

by engaging non-price strategic behaviors. Using the structural model, Greenstein and Mazzeo (2006) and

Economides et al (2008) show that product differentiation is a recipe for success when CLECs enter the

   

14

local markets. Goldfarb and Mo (2011) estimate the effects of managers’ abilities on carriers' entry

decisions. Seim and Viard (2011) examine how entry and technology adoption affect the varieties and prices

of the plans offered in the cellular market. While the above papers emphasize either the wired or wireless

sectors, my paper contributes to the literature by emphasizing the intermodal competition problem between

these two sectors. In addition, unlike previous studies such as Radini et al (2003) and Loomis and Swann

(2005), which focus on estimating the cross-elasticity of demand, my empirical entry model offers a more

straightforward way and fewer data requirements, to measure such intermodal competition effects by taking

the endogenous market structure and firm heterogeneities into account. To my knowledge, this study is the

first study to analyze the intermodal competition effects between wired and wireless sectors using an

empirical entry model.

Lastly, yet importantly, this study also complements the econometric literature regarding the control

function approach. Beginning with Heckman (1979), this approach has been widely used in labor

econometrics to address selection issues. In the IO field, several papers had adopted this method to tackle

the endogeneity problem. For example, Mazzeo (2002b), Manuszak and Moul (2008), and Zhu et al (2009)

use this approach to solve the non-randomness issues of market structure when measuring its effect on

prices or sales. Seim and Viard (2011) use the similar full information maximum likelihood approach to

solve the endogenous market structure problem. Unlike those studies, following Heckman and Macurdy

(1986) I combine the control function approach with the random coefficient model to address the

endogeneity problem caused by the number of competitors within markets: a CMA-level correction term is

generated and then a city-level random shock is employed to describe the city-level deviations from the

CMA market. Villas-Boas and Winer (1999) and Petrin and Train (2010) have already applied this approach

in a consumer demand setting. My paper complements previous studies in the sense that I fit this approach

into a market entry framework.

The reminder of this study is organized as follows. Section 2 provides background information on the

telecommunications industry. Section 3 describes the data, and Section 4 specifies the model. Section 5

   

15

explains the estimation method and identification strategy. Section 6 presents the estimation results for the

descriptive and structural analysis. Section 7 discusses the policy concerns, and Section 8 concludes.

2. Industry Background

Prior to 1996, there was little local service competition in the U.S. wired telecommunications sector,

largely due to state regulators' resistance to allowing new entrants to compete in local markets. The 1996

Telecommunications Act opened local markets to competitions by removing such regulatory entry barriers.

In addition, as a condition for gaining the right to enter long-distance markets, ILECs were required to

provide their local network facilities for new entrants. As a result, a new entrant was able to enter a local

market by following any of (or a combination of) three paths: reselling ILEC services (Resale), leasing

ILEC facilities (such as unbundling network elements and the so-called UNE-Loop and UNE-Platform), or

building its own facilities (Facilities-Based Entry).

Starting in 1997, CLECs, many of which formerly acted as competitive access providers (CAPs),

began to enter local markets. Initially, the CLECs concentrated on the resale strategy that provided 15 or

20 percent gross margins. They quickly switched to the UNE strategy after they realized that leasing UNE

loops (and building their own switches) was more profitable than the resale approach. 6 By the end of 2002,

18.9 percent of the CLECs' end-user switched access lines were provided through resale (resale accounted

for 43 percent in 1999), 55.3 percent were provided by UNEs, and the remaining 25.8 percent were served

by the CLECs' own facilities (FCC, 2003a).

                                                            6 They enjoyed a much larger wholesale discount rate as 50 percent or more, based upon FCC's forward looking total

element long-run incremental cost (TELRIC) prescription. One advantage of resale and UNEs was that it speeded up

the entry process by allowing new entrants to avoid the duplicate investments on some essential facilities. But doing

so made it harder (compared to ILECs) for them to attract high-revenue customers, as they could not provide an array

of services, such as video, local telephone and internet over a single integrated network. See Crandall's (2005)

discussion for details.

   

16

Figure 1 shows the long-run trends of the number of subscribers in wired and wireless sectors. The

number of CLEC end-user switched access lines increased gradually from 1997, whereas the number for

ILECs started to decrease in 1999. In 1997, approximately 4 or 5 million end-user switched access lines

(less than 3% of nationwide lines) were served by CLECs. By the end of 2002, this number had increased

to 24.8 million, accounting for 13.2% of the 187.5 million total end-user switched access lines (FCC, 1998a,

2003a). More strikingly, the fixed-line telephone end-user switched access lines experienced a sudden

decrease in 2001, from its peak of-192 million in June 2001 to 187.5 million at the end of 2002, which was

unprecedented in the modern history of the United States.

The explosion of wireless telephony started at roughly the same time as the deregulation of the wired

markets in 1996. Before 1994, one CMA (a total of 734 CMAs) was usually served by two carriers (one

local wired carrier and one independent carrier, with 25 megahertz for each system). However, this regional

duopoly system was quickly broken by the launch of broadband PCS (which thereafter evolved into digital

cellular services) spectrum auctions. Between 1994 and 1997, 99 PCS licenses (A and B blocks, 30 MHz

each) were assigned to 51 MTAs (Major Trading Areas), and 1990 PCS licenses (including C, D, E, and F

blocks) were awarded, with up to six PCS licenses in each BTA (493 Basic Trading Areas).7 During the

period 1998-2002, another 724 inactive PCS licenses (blocks C, D, E and F) were re-auctioned. Moreover,

the FCC allowed Nextel, which obtained most of 1554 available 800 MHz Specialized Mobile Radio

                                                            7 A lot of C, D, E and F licenses were awarded to qualified small businesses. For example, in auction 11 (D, E & F

blocks auction conducted during 08/26/1996-1/14/1997), 93 small business bidders won 598 of total 1472 licenses.

FCC allocated 30 megahertz to block C and 10 megahertz each for blocks D, E and F. Starting from 1998, PCS C

block licensees were allowed to elect to disaggregate their licenses and return 15 megahertz spectrum to the

Commission. As a result, some licensees disaggregated some of their licenses, creating certain BTAs with seven or

eight broadband PCS spectrum licenses (FCC, 2003a).

   

17

Service (SMR) licenses, to convert its spectrum from SMR service to digital cellular service. Thereafter,

each of the top 305 CMA markets could be served by up to eleven wireless carriers by the end of 2002.8

Figure 1 shows that, compared with the declining trend of the number of ILEC subscribers, the number

of wireless subscribers grew substantially year-by-year and exceeded the total number of wired users in

2004. In detail, the number of subscribers of the mobile telephone sector increased from approximately 55

million in 1997 (20% nationwide penetration rate) to 141.8 million by 2002 (49% nationwide penetration

rate). Meanwhile, subscribers' minutes of use per month had increased from 117 minutes in 1997 to 427

minutes in 2002, indicating a 53 percent annual growth rate.

Although the merger waves9 during the period of 1997-2002 consolidated the wireless sector in the

hands of several national cellular carriers, competition among cellular carriers was still fierce enough to

drive down wireless prices and raised service quality for the end user, and making wireless service a more

attractive alternative to wired service (Woroch, 2002). The mergers only increased the nationwide

Herfindahl-Hirschman Index (HHI) of concentration (for the top 25 carriers) 10 from 680 in 1997 to 1309

in 2002. Moreover, the churn rates11 were very high for most carriers during this period. For example,

                                                            8 For the rural CMA markets (428 RSA markets) which usually cross the borders of several BTAs, the number of

potential entrants in those markets could be larger than eleven. In addition, with Commission approval, licensees may

divide their spectrum into smaller amounts of bandwidth or divide their licenses into smaller geographical areas to

other entities. Consequently, there was the possibility that a CMA market would have more than eleven potential

entrants in some markets (FCC, 2003a). In reality, large carriers more often owned several licenses in a local market.  9 Such as serial mergers among AirTouch, USWest, Vodafone, PrimeCo, GTE and Bell Atlantic to form Verizon,

consolidation between AT&T and Telecorp, mergers among VoiceStream, Omnipoint, Aerial, and Powertel to form T-

Mobile, mergers among 360 Communications, CenturyTel Wireless, and Alltel to form the new Alltel Wireless, and

joint venture between SBC and BellSouth to form Cingular Wireless. 10 Here the HHI is calculated based on the top 25 wireless carriers' national market shares. Starting in 2003, the FCC

began to provide HHI based upon 172 EAs (Economic Areas). At the end of 2003, the EA-based HHI was 2151. 11 Churn refers to the number of customers an operator loses over a given period of time. Mobile carriers usually

report churn in terms of an average percent churn per month.

   

18

between 2000 and 2002, most carriers reported churn rates between 1.5 percent and 3 percent per month

(FCC, 2003b), which implies that more than 30 percent of subscribers changed providers during that year.

Unlike the price regulation in local wired market, wireless carriers enjoyed greater flexibility in pricing

(free of regulation after 1993, except for reciprocal interconnection compensation rates). Between 1997 and

2002, the real average price per minute (measured by ARPU, the Average Revenue Per Minute) for cellular

service decreased at an annual rate (compound) of 24 percent. However, the real average price per minute

for interstate calls over the traditional fixed line fell by only approximately 9 percent. Additionally, while

local telephone CPI (provided by Bureau of Labor Statistics) increased 18.5 percent from 1997 to 2002,

cellular CPI decreased by 32.6 percent during the same period.

Moreover, many wireless carriers designed service plans to compete directly with wired local

telephone carriers. For example, in May 1998 AT&T announced several new nationwide single-rate plans

that, according to AT&T, were partly targeted at individuals who travel frequently and view wireless as a

replacement for fixed-line service. Other carriers, such as Verizon (Bell Atlantic), Sprint and Nextel,

followed in the next year. In addition, some carriers offered their customers technologies and plans that

encouraged use of their mobile phones while they were at home or near home. For example, Bell Atlantic

provided a service to its limited mobility and price conscious customers called TalkAlong, which offered a

lower pricing for phone usage inside a small geographically defined local calling area. Leap, doing business

as "Cricket Wireless", allowed subscribers to make unlimited local calls and receive calls from anywhere

for approximately $30 per month, a price much lower than those offered by its wired rivals. At the end of

third quarter of 2002, 26 percent of Leap's customers did not have a wired phone at home. Other carriers,

such as Qwest, Alltel, Triton PCS, MetroPCS, and NorthCoast PCS, were all offering unlimited local

calling plans in their licensed regions. Moreover, for approximately $40-$60 per month, many carriers

offered regional or national calling plans with 500 or more anytime minutes and over 3000 night and

weekend minutes (FCC, 1998b, 2003b).

   

19

By 2002, ILECs' end-user switched access lines had fallen to 162.7 million from 181.3 million in 1999,

a decrease of approximately 18.6 million. Meanwhile, CLECs reported an increase of approximately 16.6

million lines, and the number of mobile subscribers increased by approximately 54.7 million (Figure 1). In

addition, the usage patterns of fixed-line and wireless services changed sharply after 1997. For example,

the minutes of use per month for cellular subscribers increased from 117 in 1997 to 427 in 2002, while the

minutes of use per month for residential wired end-users decreased from 149 to 90 during the same period.12

The average monthly household telecommunications expenditure for wired and wireless services were 83.8

percent and 16.2 percent, respectively in 1997, but the percentage for wireless service increased to 41

percent by the end of 2002.

3. Data

The data used in this study consists of information about wired and wireless carriers in 1,673 midsize

U.S. cities in 1997 and 2002 with population between 20,000 and 1,000,000 in Census 2000. During this

period, both CLECs and PCS carriers began to enter the local market on a large scale, and numbers of

mergers drastically reshuffled the market structure in both sectors. In addition, there were notable changes

in regulatory rules and investors’ opinions about the opportunities in the telecommunications industry.

The identities of CLECs are collected from 1997 and 2002 CLECs Report (Edition 9 and 17), and

CIOC Report (Edition 2), acquired from New Paradigm Resources Group, Inc. (NPRG). These reports

cover the detailed profile for every CLEC on its history, financial matrices, ownership, state certifications,

and local markets served at the city-level. I define potential entry as whether the CLEC owns the

certification in a state, and entry as whether the CLEC provides voice or data services in the city. The

identities of ILECs are collected from Local Calling Guide website, provided by Telmetrics, Inc., and

                                                            12 This number includes the intraLATA (Intrastate and Interstate), InterLATA (Intrastate and Interstate), and

International toll minutes, and data for business users are not available. See FCC (2005) for details.

   

20

Phillips Telephone Industry Directory (1998 and 2000). I use the former to collect the rate-center-level

ILECs’ information,13  and then track the ownership change of ILECs using the Phillips Directory.

The identities of wireless carriers are assembled from several sources. First, the information about

licensees for each market is gathered from FCC’s Universal Licensing System (ULS). The BTA and MTA

level licensees are matched up to CMA markets using FCC's market-identity crosswalk. Next, I use FCC’s

Antenna Structure Registration (ASR) system and Antenna Search website, provided by General Data

Resources, Inc., to track the presence of wireless carriers in each market.14 While ASR only provides the

information about registered mobile tower, the latter source provides the detailed information about both

registered and not registered towers, such as the planned mobile towers, location, constructed year, and

history of each tower (ownership change), which are good supplements to FCC's database.

The continuous mergers between 1997 and 2002 complicate the equity relationship among carriers.

For example, AT&T's tower sometimes operated under the name of “TeleCorp” or “ABC Wireless” after

its merger with TeleCorp Holding Corp, Inc. in February, 2002, and Verizon's tower used the name of

“Cellco” or “AirTouch” after series of mergers among GTE, Bell Atlantic, PCS PrimeCo, AirTouch, etc.

To track these ownership changes, I use FCC's Annual CRMS Competition Reports (3rd-8th) and Telecom-

munications Provider Locator to check each carrier’s identity, type and affiliation relationship for each year.

Note that carriers with a close relationship (such as joint venture or affiliation) are treated as one effective

                                                            13 A rate center is a geographic area used by a LEC to determine boundaries for local calling, billing, and assigning

phone numbers. Typically, a call within a rate center is a local call, while a call from one rate center to another is a

long-distance call. As Phillips Telephone Industry Directory was only published to 2000, here I use the data of 2000

as an approximation for 2002. Fortunately, the ownerships of ILECs in our sample do not change tremendously except

for some big mergers, such as GTE and Bell Atlantic merged in 2000 to Verizon, which are tractable. 14 Both FCC’s ASR system and Antenna Search website provide data about mobile tower and antenna at the city-level

market. Only some antennas and mobile towers are required to register in the FCC if they met some criteria. For

example, most antenna structures that are taller than 60.96 meters (200 feet) above ground level or that may interfere

with the flight path of a nearby airport must be cleared by the Federal Aviation Administration (FAA) and registered

with the FCC.  

   

21

competitor in each market. For example, in some markets where both of SBC and BellSouth owned licenses,

they are treated as one potential entrant because they operate jointly through their joint venture, Cingular

Wireless. Similarly, Alamosa PCS, AirGate, and US Unwired are treated as one firm, since they are

affiliates of Sprint and operate under Sprint's brand name.

To further verify the wireless carriers' information above, I go to each licensee's website for those two

years (at the end of that year or at the beginning of next year if the former is not available) through the

Wayback Machine. In most instances the carrier's footprint information is provided directly through its

websites or annual reports. For example, Sprint provides a detailed list of its markets on its website in 1997

and Western Wireless lists 18 CMA markets (and other RSA markets) in its 2002 annual report. In some

other cases, where only the coverage maps are available, I carefully compare their coverage maps with

FCC's market maps (CMA, BTA and MTA maps) to get the precise market list for each carrier. Combined

with the licensees and mobile tower information, I have a unique, complete list of market entrants

information for the top 305 CMAs. Unlike the CLEC market, for the wireless sector, I define potential entry

as whether a carrier owns a license in a CMA market, and define entry as whether such carrier provides

service in a CMA market.

Data on local markets’ demographic characteristics are obtained from 2000 Census.15 To be a qualified

city market, two criteria should be satisfied: (1) The city's population should be between 20,000 and

1,000,000. While the small cities might not be able to support the entry of CLECs and wireless carriers,

super cities often contains several submarkets in which it is hard to define the strategic interaction among

entrants. (2) It should be within the boundaries of the top 305 CMA markets. Here I only focus on the top

305 CMAs as the other 428 RSAs are quite different from MSA markets in their characteristics. For

example, the top 305 CMA markets cover 77% of total population in the U.S., whereas the 428 RSAs only

                                                            15 Due to the fact that data are not available for each place annually, without loss of generality, I assume that the

changes of demographic characteristics are not large enough to affect carriers’ entry decision drastically.

   

22

cover the remaining 23%. Another reason is that the top 305 FCC CMA markets are consistent with

boundaries of MSA markets used by Census. RSA markets usually cross several BTA markets, which

makes the number of (potential) entrants incomparable with those top 305 CMA markets. Finally, 1,673

cities within 292 CMA markets are left in the sample.

Table 1 describes the 1,673 midsize cities. Following Greenstein and Mazzeo (2006), I include a

MSA_10 dummy to indicate a city is in a top ten MSA and hypothesize that CLECs could share costs among

nearby cities. In some cases, a small city within a large MSA may be less expensive or more attractive for

a carrier to serve than a completely isolated city of equal size. A RBOC dummy, that equals one if incumbent

is a Regional Bell Operating Company, is included to see whether RBOCs are more willing to facilitate

entry as the regulatory rules required (Mini, 2001). The year-count variable, Regulation stringency, is

collected from Abel and Clements (1998). It equals to 0 if the regulatory regimes of rate freeze and price

cap have never been imposed, 1 if they have been used for between one and four years, and 2 if more than

four years (though 2000). A higher category of regulation stringency indicates a more opened attitude

toward competition, which might lower the entry costs for the CLECs.

Table 2 describes 292 CMA markets. The county-level demographics data are collected from Census

2000, and then aggregated to CMA level. Besides the traditional demographics variables such as population

and income, I include urban ratio and mean travel time to work (workers age 16+) to capture other local

demand heterogeneities. While the former could capture the extra benefits or cost saving effect when a

carrier serves a more urbanized area (the capacity utilization ratio might be very different between urban

area and rural area), the latter variable could measure the additional value of a mobile phone to long distant

drivers, which could affect carriers' choices of plan variety (see also Busse, 2000; Miravete and Roller,

2004; Seim and Viard, 2011). I also include the median house value to capture the cross-sectional variation

in land costs which might affect carriers' build-out costs of mobile tower.

In an industry with significant network effect, carrier size is one of the most important firm-level

characteristics which is valued heavily by customers (shown by Greenstein and Mazzeo, 2006). To capture

   

23

such type of heterogeneity and correspondingly different entry behaviors, I further classify CLECs into two

groups: A CLEC is defined as a regional carrier if it operates in less or equal to ten geographically adjacent

states. A national carrier is recorded if it covers more than ten states. In some cases, carriers that operate in

less than ten states but with non-contiguous footprint are grouped as national carriers as well.16 Table 3

shows the descriptive statistics of national and regional CLECs for 1997 and 2002. On average, national

CLECs are older than regional players, and more likely to be publically owned with larger market shares.

The patterns of subsidiary relationship are mixed due to the exits, bankruptcies, and mergers occurred

during the period of 1997-2002.

Table 1 and Table 3 show that on average a city market had 0.3 and 1.2 CLECs in 1997 and 2002, and

a CMA market had 3.3 and 5.6 wireless carriers in 1997 and 2002 separately. To further investigate entry

behavior, I count the frequency of market structure by sector shown in Table 4. In 1997, only 261 of 1,673

cities (15.6 percent) had one or more CLECs, while this number increased to 552 (33.0 percent) in 2002.

Similarly, 96 of 292 CMA markets (32.87 percent) had two or less wireless carriers in 1997, but only 5

markets (1.71 percent) had a duopoly structure by the end of 2002. In 1997 the most frequent market

structures of wireless sector were duopoly and triopoly (79 markets, 54.11 percent), whereas in 2002 the

most common market structure was one with at least five wireless carriers (230 markets, 78.77 percent).

Table 5 displays the joint frequency distribution of market structure described by the numbers of

national and regional CLECs. By 1997, national CLECs had entered 187 markets (11.18 percent) and this

number had increased to 433 (25.88 percent) by the end of year 2002. Similarly, regional CLECs had

entered 165 markets (9.86 percent) by 1997 and 339 markets (20.26 percent) by 2002. Moreover, Table 5

                                                            16 Greenstein and Mazzeo (2006) use the same criterion to classify national and regional CLECs, and provide a detailed

comparison about those two types of CLECs' product characteristics, such as location overage, target customers, menu

of services, service quality, etc. In addition, they do test the robustness of this classified criterion. They indicate that

the findings for this national/regional classification are much stronger than other kinds of classified criteria. Here, I

just follow their criterion.

   

24

shows that regional CLECs are more likely to enter in markets with fewer national CLECs, and vice versa

for the national CLECs. This pattern provides a snapshot of the geographical differentiation strategies

adopted by national and regional CLECs.

Lastly, there were 12 CLECs (12.77 percent) that provided wireless services at some markets by 1997,

such as AT&T, MediaOne (a subsidiary of USWest), TDS (through its US Cellular and Aerial Wireless

brands), Logix (a subsidiary of Dobson Communications which owned the Dobson Cellular brand), Cox

and TCI (through their joint venture with Comcast and Sprint), and so on. Starting in 1997, carriers began

the efforts to market their fixed-line and wireless service together. For example, in May 1997 USWest

bundled its wireless service to its fixed-line-based services, which offered consumer a service that gave

them a single telephone number for when they were at home, the office, or away from either. Other carriers

were adopting the similar strategy to offer such so called "one-stop-shopping" and an integrated bill for

many services as fixed-line, wireless, internet access and paging (FCC, 1998a). By 2002, there were 14

CLECs (15 percent), such as AT&T, Alltel, D&E Systems, NTELOS, SBC, Qwest, and Verizon, that

offering wireless services at some markets.

4. Model

Because CLECs and wireless carriers make their entry decisions at different market levels, modeling

equilibrium choices at the CMA-level by allowing for strategic interaction between CMA-level wireless

players and their submarket CLEC players is unmanageable in the current setting.

Instead, I focus on the intermodal competition effect wireless carriers have on CLECs and, thus, the

partial equilibrium at the city-level market. A two-step procedure is employed to correct the endogenous

market structure problems in wired and wireless sectors. In the first step, I construct an ordered probit entry

model for the wireless sector to generate a correction term, which is added to the CLEC profit function to

correct the possible endogeneity caused by the number of wireless carriers. In the second step, to fully

   

25

capture the city market level unobserved heterogeneities a city market level profit shock term is included

to reflect the deviation from the CMA level market correction term.

4.1. Specification of Profit Functions

As usual, I assume that the market equilibrium in the CLEC sector will be determined by a two-stage

oligopoly game in which CLECs make their entry decisions at the first stage17 and then compete with each

other at the second stage if they enter the market. Unlike previous studies, in the current setting CLECs

must take the competitive effect from wireless carriers into account when they make their entry decisions.

In addition, I allow a CLEC to offer two types of services within a market, which might bring about two

opposite effects on the carrier itself: business stealing and economies of scope. To capture these effects, a

national CLEC i's and a regional CLEC j's profit functions can be specified as follows:

     ln( 1) ln( 1) ln( 1)N N N N R W Other W Ownim m i NN m NR m NW m NT m imX W N N N D , N

mi , m

     ln( 1) ln( 1) ln( 1)R R R N R W Other W Ownjm m j RN m RR m RW m RT m jmX W N N N D , R

mj , m

Here, Nm denotes the set of all national carriers and R

m denotes the set of all regional carriers in

market m. mX is a vector of the observed characteristics of market m, which affect the carrier's profitability

and are independent of the unobserved error terms m and im (or jm ). W is a vector of a carrier's nationwide

characteristics, and it may differ for national and regional players. NmN and R

mN are the equilibrium numbers

of national and regional carriers in the market. W OthermN is the number of wireless carriers except for carrier i

(or j) itself, and W OwnD is a dummy taking a value of one if a CLEC also provides wireless service in that

market.

                                                            17 In such an experiment, both incumbent carriers and other potential entrants are assumed to make an effective “entry”

choice in each period, as the “exit” could be viewed as an opposite option for the incumbent carriers. Such mechanism

could be justified as a post-entry, “War of Attrition” game. See also Berry (1992). 

(1)

   

26

The competitive effect from the same-type rivals (intramodal competition) on national CLECs is

captured by ln( 1)NNN mN and ln( 1)R

NR mN .18 While the former presents the competitive effect of national

CLECs on national CLEC i, the latter captures the competitive effect of regional CLECs on national CLEC

i. Similarly, the competitive effects from national and regional CLECs on regional CLEC j are captured by

ln( 1)NRN mN and ln( 1)R

RR mN . The intermodal competitive effects from other wireless carriers are presented

by ln( 1)W OtherNW mN and ln( 1)W Other

RW mN , and the net effects of the business stealing effect and the within-

market economies of scope effect when a national and a regional CLEC provide wireless service are

captured by W OwnNTD and W Own

RTD . The log form implies that the profitability will decline with the number

of rivals in the market m, but at a decreasing rate, which is flexible enough to capture the asymmetric

competition effects on different types of competitors.

The last two components of the profit function m and im (or jm ) are profit shocks that are known to

the carriers but not observed by the econometrician (a complete information game). In detail, m is the

market-specific error term which will affect all carriers operating in market m, and im (or jm ) is an

idiosyncratic shock to national CLEC i (or regional CLEC j). For computational convenience, im and jm

are assumed to be i.i.d (independent and identically distributed) across markets and carriers and to follow

the standard normal distribution. Note that it is the market-specific error term m that makes the number of

national CLECs, regional CLECs, other wireless carriers and dummy variable W OwnD endogenous in the

profit function. To separate this term's effects on wireless and wired sectors, I further decompose this error

term into two components, a CMA-level profit shock and a city-level profit shock.

                                                            18 All competitive effects are calculated using the formula Ln(N+1) to avoid Ln0 for some markets without any

effective wired or wireless competitors.

   

27

In detail, I first introduce a model to describe the observed market structure in the wireless sector and

then use it to generate a correction term as the CMA-level profit shock. Following Bresnahan and Reiss

(1991a), I set up a latent profit function for the wireless carriers in CMA market k (where kN mobile

carriers are present) as follows:

2

( , , , , ) ( , , , )kN

k k k k k k k n kn

Z N V Z N Z

where Z is the vector of market characteristics that affect the profitability, and the vector contains

the demand and cost parameters to be estimated. In this profit function, V could be viewed as a carrier's

variable profits and k is a profit shock (or fixed cost) for all carriers in this market, which is observed by

all carriers but is not observed by the econometrician. Furthermore, I assume that this unobserved factor is

independent of the vector Z and is i.i.d. across markets by following a standard normal distribution.19

As is common in many previous studies, I assume that the wireless carrier's variable profits decrease

with the number of wireless competitor in a reduced form. That is, the parameter n measures the change in

variable profits when the nth carrier enters. To guarantee a Nash equilibrium, I assume n is negative, i.e.,

profits always decrease with the entry of an additional carrier. In this Nash equilibrium, carriers will enter

the market until no additional carriers could profitably enter. As such, if we observe that *N carriers enter

a market, we could infer the lower and upper bounds on the unobserved error term (here, profit shock) using

the following two inequalities:

*( , , , ) 0k kV Z N and *( , 1, , ) 0k kV Z N

Then, the probability of observing *N carriers in market k can be written in the following way:

* * *Pr( ) [ ( , , , )] [ ( , 1, , )]k k kN N V Z N V Z N

                                                            19 Note that by doing so I scale the wireless carrier's profit into a standard normal distribution as well.

(2)

(3)

(4)

   

28

Here, is the cumulative distribution function (CDF) of a standard normal distribution according to

our distributional assumption on k . Then, the log-likelihood function can be constructed as follows:

* *

1 1

ln [ ( , , , )] [ ( , 1, , )]MaxK N

kj k k k kk j

I V Z N V Z N

where kjI is an indicator variable that takes a value of one if the number of wireless carriers (in market

k) equals j, and M axN is the largest number of operating wireless carriers in the sample. K is the number of

CMA markets.

Note that there are several drawbacks of the current model. First, the model assumes that carriers have

the same variable profits and profit shock (or fixed cost) in the same local market. While the i.i.d. standard

normal distribution assumption of the error term largely simplifies the log-likelihood function into an

ordered probit form, it also largely assumes away the firm-level heterogeneities and possible correlation

across markets. Second, this model implicitly assumes that there are a large number of potential entrants in

each market, which should be qualified in wireless markets because the scarcity of the spectrum might limit

the number of competitors.20

In reality, several mechanisms alleviate this problem: (1) The secondary market of the spectrum

enables carriers to obtain licenses in a certain market. A wireless carrier could disaggregate its spectrum by

bandwidth or geographical areas to other carriers.21 (2) Partially due to the network effect, the wireless

                                                            20 To avoid spectra being concentrated in the hands of several large carriers, the FCC set a spectrum cap for each

market. For example, the FCC required that no entity could control more than 45 MHz of cellular, broadband PCS, or

SMR spectra in a market. However, this cap restriction was repealed on January 1, 2003. See FCC (2003b). 21 More often in reality, they disaggregated their spectra and swapped with each other to improve their capacity within

a market and coverage over different markets given approval from the FCC. For example, on February 20, 2001,

Verizon completed its acquisition of 20 unbuilt 10 megahertz PCS licenses in six states from Alltel. In May 2001,

Cingular Wireless and VioceStream (T-Mobile) swapped 5 BTA and MTA level licenses in New York, St. Louis,

Detroit, Los Angeles and San Francisco. On November 3, 2000, AT&T and Sprint exchanged licenses for 10 megahertz

blocks of PCS in markets across GA, NC, NM, OH, TN, TX (Sprint gaining) and CA, FL, TX, UT, WA (AT&T

gaining). See also FCC, Annual CRMS Competition Reports (6th, 2001), pp.16 for further details.

(5)

   

29

sector provides a textbook example of a "Winner-take-all" industry. Some factors, such as heavy usage for

long-distance calling by consumers, expensive roaming fees and incompatibility and standards wars among

carriers drive the carriers to merge with each other, to acquire more licenses from small carriers and FCC

auction and to build their own nationwide footprints (Crandall, 2005). By 2002, the top six wireless carriers

accounted for approximately 80 percent of all mobile subscribers. (3) Third, the FCC’s spectrum auctions

had squeezed out the extra profits of wireless carriers, especially for those fringe, small regional carriers.

By the end of 2002, large numbers of small carriers activated in the 1995-1997 broadband PCS auctions

had exited markets by returning/selling their licenses or directly affiliating to large carriers.

Despite its drawbacks, the current model has some attractive features, such as the simplicity of

likelihood function and its direct connection to theory. Most important, it can connect the error termk in

wireless profit functions with the market-level error term m in CLECs' profit functions. Note that the

possible endogeneity is caused by the correlation between the number of other wireless carriers (and the

W OwnD dummy) and market-specific unobservable m . To correct such endogeneity,22 I further impose a

distributional assumption on those two error terms. In detail, I assume k and k (by abusing notation k

here, to be explained below) following a joint normal distribution as follows:

2

2

0~ ,

0k

k

N

                                                            22 Compared with traditional instrumental variables approach, the control function approach has several advantages

here. For example, it can provide a direct test of endogeneity and can provide consistent estimation when the

instruments variables are unavailable. Most importantly, it is more efficient (given the control function is correctly

specified) when the endogenous variable enters the outcome function (at the second stage) nonlinearly, such as the log

of number of wireless competitors in the current setting. Wooldridge (2010) already shows that the control function

estimator can be more precise even if we have few functions of the first stage residuals (pp.268-271). In addition, the

control function approach can be extended to a nonparametric framework. See Newey, Powell and Vella (1999).

(6)

   

30

where 2 and 2

are the variances of these two error terms. By assumption, 2 1 here. is the

correlation between these two errors and, thus, in relative terms can translate the impact of specific error

k on k . is a scale factor. Then, following Heckman (1979), Mazzeo (2002b) and Manuszak and Moul

(2008), a correction term can be generated and then added to the CLECs' profit functions:

[ ( , , , )] [ ( , 1, , )]

( , , , )[ ( , , , )] [ ( , 1, , )]

k kk k

k k

V Z N V Z NZ N

V Z N V Z N

where and are the probability distribution function (PDF) and CDF of the standard normal

distribution. Then, CLECs' profit function can be written in the following way:

     ln( 1) ln( 1) ln( 1)N N N N R W Other W Ownim m i NN m NR m NW m NT k k imX W N N N D e ,

     ln( 1) ln( 1) ln( 1)R R R N R W Other W Ownjm m j RN m RR m RW m RT k k jmX W N N N D e ,

Here, is a parameter to be estimated, and the term k is the so-called "Inverse Mill's

Ratio" (IMR) by Heckman (1979). It reflects the probability that [ , ]W Other W Ownm mE N D is not equal to zero.

( , , , )k k k ke Z N is the left residual after adjusting the correlation between the number of wireless

carriers and market-specific unobserved profit shock, and is i.i.d. normal with conditional mean zero.

To this point, I have used the small k to indicate a CMA market and the small m to denote a city

market. Here, a pure CMA level correction term cannot fully capture the city-specific profit shock. Thus,

by following Heckman and Macurdy (1986), Villas-Boas and Winer (1999), Train (2009) and Petrin and

Train (2010), I allow such a CMA level profit shock to have random effects23 (around its mean) at the city-

level market. To do so, I replace the CMA level residual ke with a random error component

m to capture

the city-specific heterogeneity. m is an i.i.d. standard normal random term, and is its standard deviation

                                                            23 Actually, aside from the random effect of the city-level profit shocks, a more flexible random coefficient model

could be specified here. Heckman and MaCurdy (1986) discuss a general random coefficient selection model with

several empirical applications.

(7)

(8)

   

31

to be estimated. The measured magnitude of can reveal the importance of unobserved city-specific

heterogeneity. In addition, I allow this correction term and city-specific shock to have different effects on

national and regional CLECs' profits. Lastly, their profit functions could be written as:

        ln( 1) ln( 1) ln( 1)N N N N R W Other W Own N N

im m i NN m NR m NW m NT k m imX W N N N D ,

Nmi , Kk , m k

        ln( 1) ln( 1) ln( 1)R R R N R W Other W Own R R

jm m j RN m RR m RW m RT k m jmX W N N N D ,

Rmj , Kk , m k

As a result, the CLEC profit functions include three error terms: k is the CMA level profit shock that

is common to all cities within this CMA k ( K indicating the set of CMA markets in the sample); m is a

city-level profit shock that is common to all potential entrants in this city; im is a pure firm-level shock

reflecting unobserved firm-specific heterogeneity. , , , , are parameters to be estimated.

4.2. Setup for the Entry Game

To obtain a unique Nash equilibrium for the CLEC sector, several assumptions should be made for

current two-stage game. First, I assume that (1) an additional market participant always decreases the profits

of existing firms, regardless of type, and firms can be ranked in an order of profitability that does not change

as the set of entering firms changes (Bresnahan and Reiss, 1991b; Berry, 1992). In this case, a pure strategy

Nash equilibrium exists: carriers enter in the order of descending profitability until the last carrier receives

a negative or zero expected profit.

Next, to avoid the multiple equilibria problem, I focus on the equilibrium number of carriers N for a

certain market by following Bresnahan and Reiss (1991b) and Berry (1992). That is, I assume that (2)

carriers are symmetric post-entry. This assumption implies that rivals’ competitive behaviors, which may

(9)

   

32

fall within multi-dimensions as price and quality, can be summarized as a single number-N, and carriers

with various characteristics will be viewed as the same once they enter the market.24

Moreover, I assume that (3) the most profitable (or efficient) entrant always moves first, which can

further refine the Nash equilibria to a unique Subgame Perfect Nash Equilibrium (SPNE).25 In addition, I

employ four alternative assumptions regarding the order of entry. The first is that (I) national CLECs move

first, with more profitable CLECs moving before other CLECs. The second is that (II) regional CLECs

move first, with more profitable CLECs moving before other CLECs. The third one is that (III) most

profitable carriers move first, regardless of their type. Lastly, (IV) incumbents move first, with more

profitable incumbents (regardless of carrier type) moving before other incumbents and more profitable

potential entrants moving before other potential entrants. 26

In the telecommunications industry, it has long been argued that the incumbent carriers can enjoy

various first-mover advantages, such as large sunk costs, vast scale economies and increasing advertising

expenditures. In this case, the last assumption might be more reasonable if the more profitable incumbents

benefit substantially from such a first-mover advantage over latecomers.

5. Empirical Implementation

5.1. Estimation

Given the present assumptions on the order of entry and the large number of potential entrants in each

city market, there is no simple way to write down a likelihood function for all potential Nash equilibria

(which would vary with the identities of participants in each market). Thus, I turn to use the method of

                                                            24 Berry (1992) shows that given the negative incremental effect of new entrants, all pure strategy Nash equilibria

involve a unique number of entering firms. See Berry (1992) for details. 25 Doing so could give me an additional set of "firm-level" moment conditions, as I can determine each carrier's entry

decision under such assumptions on entry order. 26 Here, incumbents stand for the operating carriers observed in the data set, except for the ILECs, and the potential

entrants indicate that those carriers have licenses but have not yet entered the market.

   

33

simulated moments (MSM) estimators proposed by McFadden (1989) and Pakes and Pollard (1989) to

estimate the parameters in the profits function 0 , , , , KR (K is the dimension of the vector of

coefficients to be estimated). Here, the basic insight of MSM estimator is trying to minimize a set of

covariance conditions to match observed equilibrium number of carriers to the expected equilibrium

number of carriers predicted by the specified entry games. The following moment condition is assumed to

hold at the true value0 in the DGP (data-generating process):

0[ ( , )] 0,mE h X

where is a 1K vector. ∈ is a 1L vector of moment conditions that measure the difference

between observed and predicted market structures, and L is the dimension of the vector for the moment

conditions. The vector X includes all relevant observables as the potential endogenous regressors and

instrumental variables.  

When the number of moment conditions is larger than the number of coefficients to be estimated

(L>K), a MSM estimator ˆ will be chosen to make a weighted quadratic form 1 ˆ( , )mmM h X as close to

zero as possible. Specifically, a MSM estimator will minimize the objective function:

        '

1 1

1 1ˆ ˆ( ) ( , ) ( , )M M

M m M mm mQ h X W h X

M M

where h is a simulated estimation of the true moment conditions, and MW is an L L symmetric,

positive semidefinite weighting matrix. Also, 0P

MW W , where0W is a finite symmetric positive definite

matrix. In this case, the MSM estimator is a root of the first-order conditions ( ) / 0MQ and is consistent

for0 . Assuming the L K gradient matrix of the moment condition (3) as

0

'0 [ / | ]G E h , then the

distribution of MSM estimator could be shown as following:

         1 ' 1 ' ' 10 0 0 0 0 0 0 0 0 0 0 0

ˆ( ) [0,(1 )( ) ( )( ) ]dMSMM R GW G GW S W G GW G

(10)

(11)

(12)

   

34

where0 0

'0 | ( , ) |mS E hh V h X is the variance-covariance matrix of the moment condition at the true

value of0 , and R is the number of simulation. A two-step procedure is implemented to get the optimal

MSM estimator: at the first step, an identity matrix could be used as the suboptimal weighting matrix; then

the consistent estimator of S at the first step can be used to construct the weighting matrix in the second

step 1ˆMW S . As a result, the optimal MSM estimator is asymptotically normally distributed with mean

0 ,

and the asymptotic variance has a formula as 1 1 ' 1 10 0 0(1 )( )M R G S G .27

5.2. Identification

There are two vectors of parameters to be identified: the parameters { , } in wireless carriers' profit

functions and the parameters , , , , in CLECs' profit functions.

For the wireless sector, the identification of demand and cost parameters is straightforward: the

interaction between variations in the market characteristics and the variations in entry decisions by the same

carrier in different markets can identify the market characteristics' effects on carriers' profits. The

competitive parameters can be identified by matching and comparing similar markets with different market

structures, and/or different markets but with the same numbers of wireless carriers. After controlling for

the market size, the differences in the number of wireless entrants and the other market characteristics can

provide information about the impact of competition, and the incremental competitive effect from extra

carriers can be inferred from the variations in the market structure.

For the CLEC sector, besides the variations in (i) the CMA-level market characteristics, I have three

additional sets of variations in the dataset: (ii) city-level market characteristics, (iii) firm characteristics and

(iv) the number of potential entrants in each market. Similarly, market level demand and cost parameters

                                                            27 I use the nongradient Nelder-Mead "simplex" optimization search routine. The shuffled Halton sequence is applied

to improve the coverage and efficiency (Train, 2009). Here, R=150 times (R=300 when calculating variance). For

detailed discussion of MSM estimation procedure, see Berry (1992) and Jia (2008).

   

35

can be identified through the variation of market characteristics. Variations in firm characteristics can

provide two items of useful information: first, when combining with market characteristics, they can explain

why the same carrier enters different markets; second, they can explain why different carriers enter the

same market. That is, these variations provide an additional set of moment conditions to identify the

competition parameters. A carrier's characteristics can affect another firm's entry decision through the first-

order condition of its profit maximizing function.28 For example, a large carrier (with a reputation as a price

killer) would be daunting to small carriers considering market entry.

The variation in the number of potential entrants in each market is another important source for

identifying the coefficients on competitive effect (Berry and Tamer, 2006). On the one hand, given the post-

entry symmetric assumption, such variation can directly determine the equilibrium number of actual

entrants in each market. On the other hand, such variation can reveal the magnitude of competition effect

within a market to some extent. For example, considering two identical markets, one with a small number

of potential entrants and one with a large number of potential entrants, if the entry probability decreases

from market one to market two, the magnitude of this decrease could provide information on the magnitude

of such a negative competition effect (Goldfarb and Mo, 2011).

The offering of a wireless service by a CLEC may affect other CLECs' entry decision and may affect

itself through two opposite effects: a business stealing effect (when competing for the same customers) and

a within-market scope economies effect (e.g. sharing market information, facilities, or marketing jointly).

To identify its effect on other CLECs and its effect on itself separately, I divide the intermodal competitive

effect into two parts: the competitive effect from other wireless carriers and the net effect of business

stealing and scope economies effects.

                                                            28 Note that I assume carriers are post-entry symmetric. Thus, it could only affect other carriers' pre-entry profits

function in the current setting.

   

36

One empirical challenge here is to separate the supply-side substitutability or complementarity from

the demand-side correlation. For example, a carrier providing two services frequently when entering

markets could be the result of the supply-side within-market scope economies effect; it could also be caused

by unobserved local market demand heterogeneities. This problem has been well addressed by Gentzkow

(2007) under a consumer demand estimation setting. Following Gentzkow (2007), here I use CMA-level

market characteristics, such as travel time, as the exclusion variables. If such a net effect is positive, within-

market scope economies will dominate, and an increase in travel time will not only increase the profitability

of its wireless service but also increase the probability of offering a wired service. Similarly, if such a net

effect is negative, the probability of offering a wired service would decrease. On the other hand, if the net

effect is zero or the observed pattern of offering two-services together is due to demand-side correlation,

then the probability of offering a wired service will not change. As a result, combined with the correction

term generated from the entry model for the wireless sector, the intermodal competition parameters are

formally identified.

In summary, I construct two sets of moment conditions to identify 34 parameters for each year. The

market level predicted errors (the differences between the model predicted and observed value of the

number of national and regional CLECs) are interacted with (ii) market characteristics variables and (iv)

the number of potential national and regional CLECs. In addition, following Berry (1992), I interact them

with the number of potential entrants serving the surrounding markets within such a CMA.29 Also, given

the SPNE, I create a set of firm-level moment conditions. In each market, I pick up one national CLEC and

one regional CLEC with the largest nationwide market share,30 which might be thought as the "most likely

                                                            29 By doing so, I assume that CLECs are more likely to enter geographically adjacent markets, which is consistent

with the observed patterns in the data set.  30 Because I do not have the market-level market share information for each carrier, here I select the carriers using

their nationwide market share. Any arbitrary exogenous rule is appropriate for estimation here. See also Berry's (1992)

Appendix. When a market has no national (regional) potential CLECs, I use the second largest regional (national)

CLECs to construct such moment conditions.

   

37

entrants" in such market. The predicted errors, the differences between observed entry decisions and

predicted decisions, are then interacted with (ii) city-level market characteristics, (iii) own-firm

characteristics and other firms' characteristics. Here, with some arbitrariness, I use the sum of other potential

CLECs' market shares squared and the mean of other potential CLECs' distances to their headquarters.

Additionally, I interact these errors with the CMA-level population (in log) and mean travel time to work.

6. Results

6.1. Some Descriptive Results

Table 6 presents the parameter estimates from the ordered probit model in which the number of

wireless carriers in a CMA market is a dependent category variable and the wireless carrier's profit is a

latent variable. Due to the reduced form profit function and normalization of the error term, it is difficult to

compare and explain the estimates' magnitudes across time periods. As such, I only focus on their signs,

significances and relative magnitudes within the same period. As expected, the market size-population-is

an important determinant of the number of wireless carriers in a CMA market. Income has a positive but

insignificant effect on wireless carriers' profits. Carriers are more likely to enter a more urbanized market

and a market with a longer mean travel time to work. High land cost decreases carriers' profits, but this

effect is not significantly different from zero.

More interesting, the estimates indicate that the presence of 2nd to 5th carriers significantly lowers

carriers' profits, while the 6th, 7th and 8th entrants' effects are close to zero. The third entrant had the largest

negative effect on profits (which was roughly larger than the second entrant) in 1997, and the fourth player

had the largest negative effect in 2002. Note that the coefficients on these competitive variables should be

interpreted with caution, as these estimates denote the relative profit changes when an additional carrier

   

38

enters the market (for example, the estimate on N_3 shows the difference between duopoly and triopoly

profits), relative to carriers' profits in a benchmark market in a specific year.31

Table 7 presents the results from OLS regressions in which the numbers of operating national and

regional CLECs in city markets are counted as dependent variables. The numbers of operating CLECs are

positively correlated with the market size variable (population), while local income only has mixed effects

on the numbers of entrants. Additionally, the estimates show that both types of carriers are less likely to

enter a city market within the top ten MSAs. Consistent with many previous studies, a CLEC's entry is more

likely in markets where the regulatory commissions have a more friendly attitude toward competition and

where the incumbent firms are RBOC. Both effects increase during the sample period.

In addition, the numbers of operating CLECs are strongly associated with the numbers of potential

entrants. Although the effects of the number of potential national CLEC entrants are not significant, the

number of potential entrants at the CMA level has a good explanatory power on the number of operating

CLECs. The estimates also show that the number of operating regional CLECs is negatively correlated with

the number of potential national carriers but is positively correlated with the number of potential regional

CLECs, which hints at the possibility that regional and national carriers are targeting different markets.

Moreover, the number of wireless carriers is negatively associated with the number of regional CLECs. Its

effect on national CLECs was close to zero in 1997 but was significant in 2002.

6.2. Results from the Simulated Estimator

Note that both observed market heterogeneities and firm heterogeneities can explain the positive

correlation between the number of entering carriers and market size. Now, I extend the analysis by taking

the unobserved market and firm heterogeneities into account. Tables 8 and 9 provide results from the

                                                            31 According to Table 6, the benchmark market was a monopoly in 1997, and in 2002 the benchmark was a duopoly.

Note that if the truncated market structure is due to sample selection issues, the coefficients on N_2 could either

underestimate or overestimate the true competition effect in 1997, similar to the parameter of N_3 in 2002. In this

case, a Tobit model could better recover the parameters.

   

39

simulated estimator for 1997 and 2002. The first model in these tables assumes that national CLECs move

first, with the more profitable CLECs moving before the other CLECs. Similarly, the second model assumes

that regional CLECs move first, with the more profitable carriers moving before the less profitable carriers.

The third model assumes that the most profitable carriers move first, regardless of their type, whereas the

last model assumes that incumbents move before the new participators.

Table 8 presents the simulated results for 1997. Similar to the OLS regression results above, a large

market can attract more carriers of either type, but such attractiveness is much greater for national than

regional CLECs. Generally, income per capita is positively associated with CLECs' entry decisions,

although its significance varies with different model assumptions. As expected, entry into an urbanized area

is more profitable than entry into a relatively "rural" area. Moreover, the urban ratio variable appears to

have a relatively greater impact on the national than regional CLECs. One possible explanation is that

CLECs can enjoy cost-saving effects (such as economies of scale) when serving an urbanized region. The

estimates on the MSA_10 dummy variable indicate there was no geographic scope economy for either

national or regional CLECs in 1997. Similarly, regulatory and RBOC dummy variables only affect carriers'

profits in an insignificant way.

For those observed firm characteristics, the results show that a firm's age and subsidiary dummy

variable only have minor effects on CLECs' profits, if at all. However, the negative signs on the "private"

dummy variable indicate that publicly-owned carriers are more likely to enter a market than privately owned

carriers. This result represents the different risk preferences of privately and publicly owned companies.

Another important variable with market-level variation is the distance to carrier's headquarters (HQ), which

may capture the different expansion patterns of national and regional CLECs or the cost dis-advantage

when CLECs enter a distant market. As expected, CLECs are less likely to enter a remote market, and such

distance effect on regional CLECs is much larger than its effects on national CLECs. Moreover, the

relatively smaller value of the constant term for regional CLECs indicates that a regional CLEC is more

likely to enter before a national CLEC.

   

40

Next, consider the intramodal and intermodal competition effects. The estimates here show that the

intramodal competitive effects on CLECs are primarily from same-type CLECs.32 For example, the number

of national CLECs has a significantly negative effect on national carriers (-0.714 for model (I)-National

First), but its effect on regional CLECs is not significantly different from zero (-0.080). A similar pattern

is observed for the competitive effects from regional CLECs. One possible explanation is that national and

regional CLECs might differentiate their geographic coverage intentionally, which could mitigate direct

competition from each other. These findings are consistent with the observed entry patterns in Table 5. In

addition, the results show that national CLECs face much fiercer competition from their counterparts, and

the competitive effect within the group of regional CLECs is comparatively smaller (0.714>0.474). These

findings suggest that the footprints of national CLECs may overlap each other (for example, they may all

focus on the central city of a large MSA), whereas regional carriers have succeeded in differentiating

themselves from other regional CLECs in geographic coverage.

More interesting is that the estimated parameters before the intermodal competition variable, namely,

the number of other wireless carriers, show that the presence of other wireless carriers lowers the profit

margins of those fixed-line carriers. In addition, the intermodal competitive effect on regional CLECs is

much larger than its effect on national CLECs. In 1997, the competitive effect by other wireless carriers on

regional CLECs was -3.603, while its effect on national CLECs was -0.284 (not significantly different from

zero). One possible reason is that regional CLECs and wireless carriers may target and compete directly on

same type of customers within similar/overlapped geographic areas. For example, while national and some

regional CLECs focus on business customers in the central city within a MSA, other regional CLECs may

also be interested in residential customers in the surrounding suburbs, which may also be the targeted niches

of wireless carriers.

                                                            32 My results here (intramodal competitive effects) are consistent with the findings of Greenstein and Mazzeo (2002)

in which they separate the competition effects of first and second national and regional competitors.

   

41

Moreover, for regional CLECs the competitive effect from wireless carriers appears to be larger than

the competitive effect from other regional CLECs. This finding implies that while regional CLECs could

isolate themselves from competition by other CLECs through differentiation in geographic coverage, they

must face the unavoidable competitions from the wireless sector. All of these findings suggest that wireless

carriers are close competitors with regional CLECs, and they could provide good constraints on regional

CLECs' market powers, if at all. More strikingly, the estimates show that the provision of mobile service

by a CLEC in one CMA market neither increases nor lowers its margin in a city-level market. One possible

reason for this finding is that the business stealing effect of providing a wireless service is largely offset by

the within-market scope economies effect.

The significant coefficients of Omega ( R ) provide direct evidence of the importance of correcting

the endogeneity caused by the number of other wireless carriers. A positive sign on the correction term (for

example, R =0.401 in Model-(I)) indicates that some CMA-level unobserved, desirable (if they are

positive) heterogeneities could attract more mobile carriers and more regional CLECs, which is consistent

with the findings above that regional CLECs and wireless carriers tend to be close competitors. Additionally,

for both CLEC types, the standard deviation variable (Sigma) is statistically significant, and the null

hypothesis that the standard deviation equals zero can be rejected at the conventional level of confidence.

These results confirm my assumption that the CMA-level correction term does not fully capture the city-

specific unobserved profit shocks and the need to control for such heterogeneities.

Table 9 presents the results from same analysis using 2002 data. Similar to 1997, entry is more likely

in a market with a larger size of population with higher income. All else being equal, the positive and

significant coefficient on the urban ratio variable indicates that CLECs could benefit from entering into a

more urbanized area. Contrary to the results in 1997, in 2002 serving a suburb tended to decrease regional

carriers' margins (-0.609 in model (I)), while its effect on national carriers was close to zero. Moreover, the

estimates on the RBOC dummy variables show that in 2002 entry was more likely to occur in markets

   

42

where the incumbent firm was a RBOC. These findings are consistent with the fact that under section 271

of the 1996 Act, local Bell companies accommodated entry (such as quicker agreement, less litigation, and

more favorable access price) more than did other ILECs (Mini, 2001).

For these firm characteristic variables, being a subsidiary of a large company tends to decrease a

CLEC's margins, but this effect is not significant at any conventional levels of significance. Similar to the

findings in 1997, private firms are less aggressive when making entry decisions, and the carrier's age has

mixed and insignificant effects on its profits.33 Again, CLECs are less likely to enter a market that is far

away from their headquarters. Also, note that such a negative effect on national carriers was roughly the

same as the effect in 1997, but the distance effect on regional carriers became much larger in 2002. For

example, in model (I) the distance effect was -0.260 in 1997 and had decreased to -0.656 by 2002. All of

these findings indicate that regional carriers appear to become more reserved when choosing markets.

The coefficients on the numbers of competitors are of particular interest here. Again, the results show

that the intramodal competitive effects on CLECs are predominantly from same-type CLECs. For example,

national CLECs only have significant competitive effects on their counterparts, and regional CLECs only

have significant effects on other regional CLECs. Next, the estimates also indicate that the competition

among national CLECs and among regional CLECs become more fierce during the sample period, which

is consistent with the data that the average number of CLEC entrants increased from 0.33 in 1997 to 1.15

in 2002 and with the hypothesis that margins decrease with the number of competitors. Similarly, the

competition appears to be much stronger among national than regional CLECs.

Another important variable is the intermodal competitive effect from the other wireless carriers. Note

first that such effects on national CLECs became significantly negative in 2002, whereas in 1997 wireless

carriers only had minor effects on national CLECs. This result implies that in 2002 even national CLECs

                                                            33 These mixed results might be explained by the fact that in 2002 a portion of wired carriers with financial hardships

were acquired by other companies, and their ages were recalculated using the acquired companies' ages.

   

43

were affected by the presence of wireless carriers. In addition, the estimates indicate that wireless carriers

have a much larger competitive effect on regional than national CLECs, which confirms the findings in

1997 that regional CLECs and wireless carriers appear to be close competitors. Compared with the

estimates in 1997, the coefficients in 2002 are roughly smaller (-2.465>-3.603 for model (I)). One possible

explanation for this result is that by 2002 wireless carriers had developed more advanced and successful

customer segmentation and price discrimination tools, which alleviated the competitive pressure within the

mobile sector and its intermodal competitive effects on the fixed-line sector.

Lastly, yet importantly, the positive but insignificant coefficient on the dummy providing wireless

service indicates that the business stealing effect is largely offset by the within-market scope economies

effect. Although this estimate is not different from zero, it at least suggests that a carrier is not hurt by

providing both services in a local market. Once again, the significant coefficients on the correction term

(Omega) indicate the necessity of handling the endogeneity problem caused by the number of other wireless

carriers. The negative sign on the national CLECs' correction term and the positive sign on the regional

CLECs' correction term hints at the possibility that some CMA-level unobserved heterogeneities, which

may attract more wireless carriers and regional CLECs, tend to decrease the entry probability of national

CLEC. Lastly, the significant estimates on the two error components suggest that city-level market

unobserved heterogeneities are vital factors in determining the carriers' profits.

7. Policy Concerns and Experiments

7.1. How Many Carriers are Needed to Keep the Wireless Sector Competitive?

Using the estimates from Table 6 and breakeven conditions, I calculate the entry threshold and

threshold ratio. The entry threshold is the market size necessary to support N carriers in a market measured

by population in thousands. For example, Table 10 shows that by 1997 the population necessary to support

three mobile carriers was 652,000, while by 2002 only 298,000 people were required to support three

carriers. Similar patterns can be found for all of the other market structure types. In general, the entry

   

44

thresholds in 2002 are smaller than the entry thresholds in 1997. There are two possible reasons for this

observation. One is the increase in the mobile penetration rate. By 2002, the national penetration rate of

mobile telephones had increased pronouncedly, from 20 percent in 1997 to 49 percent in 2002 (FCC, 2003b).

Another reason is related to wireless carriers' pricing strategies. By 2002, wireless carriers had developed

several more advanced customer segmentation and price discrimination tools, which successfully increased

their revenue growth.

The entry threshold ratio measures how the level of competition changes with the number of firms.

For example, the population size should increase by sn (per firm entry threshold) to support one more nth

entrant and another sn+1 for the (n+1)th entrant. Then, the threshold ratio sn+1/sn can show the change in

competitive nature when the (n+1)th firm enters the market. Under the homogenous firm assumption, this

ratio should be equal to one if entry does not change competitive conduct (for example, perfect competition

or collusion). On the other hand, s∞/sn shows how quickly oligopoly profits approach competitive profits.

Table 11 also shows that entry threshold ratios decline with the number of entrants. In addition, when N>5,

these ratios are close to one, as s5 roughly equals s6 in 1997 and s6 , s7 , and s8 in 2002. Figure 2 indicates

that the ratio (s6/sn) is roughly equal to one when the market has five carriers. Both findings identify the

fifth carrier as the marginal player to keep a market competitive.

Note that in 2002 the population necessary to support four carriers in a CMA market was 1,115,000,

which was larger than the mean value of the CMA's population (about 733,230) in the sample. This finding

provides direct evidence explaining why some carriers who adopted the "nationwide coverage" strategies

had fallen into financial troubles by 2002. Moreover, considering that Sprint, Nextel and T-Mobile often

played the role of 5th competitor (varying over markets, see Table 11) in 2002, the merger with the top five

providers tended to reduce competition sharply. This findings can shed some light on recent (proposed)

merger cases. For example, the proposed merger between AT&T and T-Mobile in 2011 was blocked by the

DOJ because it would effectively eliminate one of the top four nationwide carriers, while the merger

   

45

between T-Mobile and MetroPCS (a large carrier and a small carrier) were less likely to trigger antitrust

concerns.

7.2. The Effect of Intermodal Competition and Fixed-to-Mobile Joint Venture

The results from the simulated estimation are intriguing in the sense that they capture the intermodal

competition effect from the wireless sector on the fixed-line sector, especially after taking the endogenous

market structure into account. One advantage of such a model is that the reduction in competition by

eliminating an effective competitor could be directly read from the estimates in the profit function.

Additionally, as mentioned by many previous studies, another advantage of this structural model is that

when conducting policy experiments, it can provide more precise "out-of-sample" predictions than the

simple model without considering the interaction among firms' entry decisions (Berry, 1992). Here, to

further exploit the marginal effect of additional wireless competitors (in the spirit of derivatives with respect

to certain variables), I use the results from model (IV)-Incumbent Moves First, which has the smallest

objective function value, to predict the effect of changes in the number of wireless competitors.

Before making such predictions, it is worthwhile to check the prediction power of the current model.

Table 12 shows the model's goodness of fit for the Incumbent Moves First model for both years. While the

first and fourth columns give the sample mean, the second and fifth columns list the mean value predicted

by the model. The results in Table 12 indicate that, although there is some noise, the model acceptably in

fitting the data generating process. Moreover, the correlations in the third and sixth columns measure the

correlation between the number of carriers predicted by the model and the observed number of carriers in

each market, and a high value indicates a good fit of the model to the data. Overall, the current model,

Incumbents Move First, explains the data well.

Figure 3 displays the predicted means of national and regional carriers under different assumptions of

the number of wireless carriers. For each counterfactual, given the estimates, I fix the number of wireless

carriers and resolve the entire entry game to obtain the new equilibrium numbers of CLECs. For the first

pillar in the graph, for example, to separate the intermodal effect on different type of carriers, I assume the

   

46

number of wireless carriers for national CLECs is equal to 0, but I keep the number of wireless carriers for

regional CLECs the same as before. Figure 3 shows the incremental effect in an average market when an

additional wireless carrier enters the market. For example, the entry by the first wireless carrier drove out

approximately 0.06 national CLECs in 1997, but it replaced 0.25 national CLECs in 2002. More strikingly,

compared with national CLECs in either year, the presence of the first wireless carrier replaced four regional

CLECs, and the second wireless carrier drove out another two regional CLECs. Compared with the

decreasing pattern of national CLECs, the number of regional CLECs decreases more sharply with the

number of other wireless carriers.

Combined with the findings in Tables 8 and 9 (which show that providing a wireless service only has

an insignificant effect on CLECs' margins), these results could shed light on antitrust and regulatory policies,

especially during the digital convergence era, with a blurred boundary between traditional markets and

industries. The current results provide evidence justifying deregulation in the fixed-line sector: if the

wireless carriers could constrain ILECs in a similar way as CLECs, the regulators should not worry too

much about market power by the wired carriers. The current model can address another policy concern

related to the effect of joint ventures between fixed-line and wireless carriers. For example, a CLEC could

benefit from forming a joint venture (or merge directly) with a wireless carrier34. In detail, the current results

show that this type of coordination/joint venture can not only facilitate the CLECs with more market

segmentation and price discrimination tools but also effectively reduce the number of intramodal and

                                                            34 For example, one of the motivations behind Alltel's merger with 360 Communications, Inc. ("360") in 1998 was

that it enhanced both companies' ability to offer integrated product bundles (FCC, 1998b). In a recent antitrust case

challenged by the DOJ (August 16, 2012), Verizon and several cable companies, including Comcast, Time Warner,

Cox and Bright House Network, reached a commercial agreement to sell each other's "quad-play" services in certain

parts of the country. Consequently, in some overlapped markets, such as New York City, Philadelphia, and

Washington, D.C., Verizon Wireless' retail outlets sold two competing quad-play offers: one including Verizon

Wireless service and a Cable firm's fixed-line, video and Internet services, and another including Verizon Wireless

service and Verizon's FiOS service (a bundle of Verizon's voice, video and Internet services).

   

47

intermodal competitors, leading to a reduction in competition and an increase in CLECs' margins. In

addition, the results consistently show that regional CLECs have greater incentives to adopt such strategies

than national CLECs.

8. Conclusion

In this study, I examine the intramodal competition within the CLEC sector and the intermodal

competition effects wireless carriers impose on CLECs. To solve the problem that CLECs and wireless

carriers make their entry decisions at different market levels, I propose a two-step procedure that firstly

generates a CMA-level correction term with an ordered probit entry model and then adds this correction

term to the CLECs' profit functions to correct the endogeneity problem caused by the number of wireless

carriers. To capture the city-level market-specific and firm-specific heterogeneities, a random coefficient

model is constructed and estimated through the MSM estimators.

The results show that the intramodal competitive effects on CLECs are primarily from same-type

CLECs, and national and regional carriers tend to differentiate themselves from each other in their

geographic footprints. Compared with national CLECs, wireless carriers and regional CLECs appear to be

close competitors, as the presence of wireless carriers significantly lowers regional CLECs' margins. The

intermodal competitive effect of wireless carriers on regional CLECs is much larger than their effect on

national CLECs. Moreover, in 1997 the wireless carriers only had a minor and insignificant effect on

national CLECs, but by 2002, they significantly decreased the national CLECs' profits. It is also found that

providing wired and wireless services together in a market does not lower a CLEC's profitability.

Furthermore, I find it necessary to correct the endogeneity caused by the number of wireless carriers and

find it important to include market-specific and firm-specific unobserved heterogeneities. The empirical

entry model for the wireless sector shows that the top five wireless carriers significantly decrease existing

wireless carriers' profits, while the 6th to 8th competitors only have minor effects on carriers' margins. The

   

48

entry threshold tests consistently show that at least five wireless carriers are needed to keep the wireless

sector sufficiently competitive.

The current model does have several limitations. For example, it does not directly model the strategic

interaction between wireless and fixed-line carriers. Additionally, the current model makes distributional

assumptions that largely assume away the firm heterogeneities in the wireless market. A more general

model that takes such heterogeneities into account could be proposed in the future.

   

49

References

Abel, J. 2002. “Entry into Regulated Monopoly Markets: The Development of a Competitive Fringe in the

Local Telephone Industry,” Journal of Law and Economics, 45(2), 289-316.

Abel, J., and Clements, M. 1998. “A Time-Series and Cross-Sectional Classification of State Regulatory

Policy Adopted for Local Exchange Carriers: Divestiture to Present(1984-1998),” Columbus, OH:

The National Regulatory Research Institute.

Abel, J., and Clements, M. 2001. “Entry under Asymmetric Regulation,” Review of Industrial Organization,

19, 227-242.

Akkus, O., Cookson, J.A., and Hortacsu, A. 2012. "The Determinants of Bank Merger: A Reveal Preference

Analysis," Working Paper, University of Chicago.

Alexander, D., and Feinberg, R. 2004. “Entry in Local Telecommunication Markets,” Review of Industrial

Organization, 25, 107-127.

Ashenfelter, O., Ashmore, D., Baker, J., Gleason, S., and Hosken, D. 2006. "Empirical Methods in Merger

Analysis: Econometric Analysis of Pricing in FTC v. Staples," International Journal of the

Economics of Business, Vol.13(2), 265-279.

Bajari, P, Benard, L., and Levin, J. 2007. "Estimating Dynamic Models of Imperfect Competition,"

Econometrica, Vol. 75, 1331–1370.

Bajari, P. and Fox, J. 2013. "Measuring the Efficiency of an FCC Spectrum Auction," American Economic

Journal: Microeconomics, Vol. 5, 100-146.

Berry, S. 1992. “Estimation of a Model of Entry in the Airline Industry.” Econometrica, 60 (4), 889–917.

Berry, S., and Reiss, P. 2007. “Empirical Models of Entry and Market Structure”, in M. Armstrong and R.

Poster (eds.) Handbook of Industrial Organization, Volume III, Amsterdam: North-Holland.

Berry, S., and Tamer, E. 2006. “Identification in Models of Oligopoly Entry,” in Blundell, Newey and

Persson(eds.) Advances in Economics and Econometrics, Ninth World Congress, Vol.2, 46-85,

Cambridge University Press, New York, NY.

Bjorn, P., and Vuong, Q. 1984. “Simultaneous Equations Models for Dummy Endogenous Variables: A

Game Theoretic Formulation with an Application to Labor Force Participation,” Social Science

Working Paper, California Institute of Technology.

Braeutigam, R. 1979, “Optimal Pricing with Intermodal Competition.” American Economic Review, 69,

38-49.

Bresnahan, T., and Reiss, P. 1987. “Do Entry Conditions Vary across Markets?” Brookings Papers on

Economic Activity, 3, 833-881.

   

50

Bresnahan, T., and Reiss, P. 1991a. “Entry and Competition in Concentrated Markets.” Journal of Political

Economy, 99 (5), 977–1009.

Bresnahan, T., and Reiss, P. 1991b. “Empirical Model of Discrete Games,” Journal of Econometrics, 48,

57-81.

Busse, M. 2000. "Multimarket Contact and Price Coordination in the Cellular Telephone Industry," Journal

of Economics and Management Strategy, 9, 287-320.

Ciliberto, F., and Tamer, E. 2009. “Market Structure and Multiple Equilibria in Airline Markets,”

Econometrica, 77(6), 1791-1828.

Crandall, R. 2005. Competition and Chaos: U.S. Telecommunications Since the 1996 Telecom Act.

Brookings Institution Press, Washington, D.C.

Economides, N., Seim, K., and Viad, V. 2008. “Quantifying the Benefits of Entry into Local Phone Service.”

RAND Journal of Economics, 39 (3), 699-730.

Ellickson, P.B., Houghton, S., and Timmins, C. 2013. "Estimating Network Economies in Retail Chains:

A Revealed Preference Approach," RAND Journal of Economics, Vol. 44(2), 169-193.

Farrell, J., and Shapiro, C. 2010. "Antitrust Evaluation of Horizontal Mergers: An Economic Alternative to

Market Definition," BE Journal of Theoretical Economics: Policies and Perspectives, 10(1), 9.

Federal Communications Commission. 1998a, 2003a. “Local Telephone Competition,” Washington, D.C.

Federal Communications Commission. 1998b-2003b. “Commercial Mobile Radio Services (CMRS)

Competition Report (3rd-8th Annual),” Washington, D.C.

Gentzkow, M. 2007. "Valuing New Goods in a Model with Complementarity: Online Newspapers,"

American Economic Review, 97(3), 713-744.

Goldfarb, A., and Xiao, M. 2011. “Who Thinks about the Competition? Managerial Ability and Strategic

Entry in US Local Telephone Markets.” American Economic Review, 101(7), 3130-3161.

Greenstein, S., and Mazzeo, M. 2006. “The Role of Differentiation Strategy in Local Telecommunication

Entry and Market Evolution: 1999-2002.” Journal of Industrial Economics, Vol.LIV(3), 323-349.

Hausman, J. 2002. “Mobile Telephone,” In: Cave, M., Majumdar, S., and Vogelsang, I. (eds). Handbook

of Telecommunications Economics, Volume I: Structure, Regulation and Competition. Amsterdam:

North-Holland, 563-604.

Heckman, J. 1979. "Sample Selection Bias as Specification Error," Econometrica, 47, 153–161.

Heckman, J., and MaCurdy, T. 1986. "Labor Econometrics," in Zvi Griliches and Michael D. Intrilligator

(Eds.), Handbook of Econometrics. Elsevier Science, NY.

Jia, Panle. 2008. “What Happens When Wal-Mart Comes to Town: An Empirical Analysis of the Discount

Retailing Industry,” Econometrica, 76(6), 1263-1316.

   

51

Loomis, D. and Swann, C. 2005. “Intermodal Competition in Local Telecommunications Markets,”

Information Economics and Policy, 17, 97-113.

Manuszak, M., and Moul, C. 2008. "Prices and Endogenous Market Structure in Office Supply

Superstores," Journal of Industrial Economics, Vol.LVI(1), 94-112.

Mazzeo, M. 2002a. “Product Choice and Oligopoly Market Structure.” RAND Journal of Economics, 33,

221–242.

Mazzeo, M. 2002b."Competitive Outcomes in Product-Differentiated Oligopoly." Review of Economics

and Statistics, 84(4), 716-728.

Mini, F. 2001. “The Role of Incentives for Opening Monopoly Markets: Comparing GTE and BOC

Cooperation with Local Entrants,” Journal of Industrial Economics, 49(3), 379-414.

Miravete, E. and Roller, L. 2004. "Competitive Nonlinear Pricing in Duopoly Equilibrium: The Early U.S.

Cellular Telephone Industry." Working Paper, University of Pennsylvania.

New Paradigm Resources Group. 1998, 2003. CLEC Annual Report. Edition 9th, 17th. Chicago, IL.

New Paradigm Resources Group. 2003. CIOC Annual Report (2002). Edition 2nd. Chicago, IL.

Newey, W.K., Powell, J.L., and Vella, F. 1999. “Nonparametric Estimation of Triangular Simultaneous

Equation Models,” Econometrica, 67, 565-603.

Pakes , A., and Ho, K. 2013. "Hospital Choices, Hospital Prices and Financial Incentives to Physicians",

Working Paper, Harvard University.

Pakes, A, and Pollard, D. 1989. “Simulation and the Asymptotic of Optimization Estimators,”

Econometrica, 57(5), 1027–1057.

Pakes, A., Porter, J., Ho, K., and Ishii, J. 2011. “Moment Inequalities and Their Application,” Working

Paper, Harvard University.

Petrin, A., and Train, K. 2010. “A Control Function Approach to Endogeneity in Consumer Choice Models,”

Journal of Marketing Research, Vol. XLVII , 1–11.

Phillips Publishing Company, 1998, 2001. Phillips Telephone Industry Directory. Phillips Publishing

Company, Potomac, MD.

Reiss, P. 1996. “Empirical Models of Discrete Strategic Choices,” American Economic Review, Papers and

Proceedings, 86(2), 421-426.

Rodini, M., Ward, M., and Woroch, G. 2003. “Going Mobile: Substitutability between Fixed and Mobile

Access,” Telecommunications Policy, 27, 457–476.

Seim, K. 2006. “An Empirical Model of Firm Entry with Endogenous Product-Type Choices,” RAND

Journal of Economics, 37(3), 619-640.

   

52

Seim, K., and Viard, V. 2011. “The Effect of Market Structure on Cellular Technology Adoption and

Pricing,” American Economic Journal: Microeconomics, 3(2), 221-251.

Train, K. 2009. Discrete Choice Methods with Simulation(Second Edition). Cambridge University Press,

Cambridge, UK.

Villas-Boas, J. M., and Winer, R. 1999. “Endogeneity in Brand Choice Models,” Management Science, 45

(10), 1324–1338.

Vogelsang, I. 2010. “The Relationship between Mobile and Fixed-line Communications: A Survey,”

Information Economics and Policy, 22, 4-17.

Whinston, M. 2007. “Antitrust Policy Toward Horizontal Mergers”, in M. Armstrong and R. Poster (eds.)

Handbook of Industrial Organization, Volume III, Amsterdam: North-Holland.

Wooldridge, J. 2010. Econometric Analysis of Cross Section and Panel Data (2nd Edition). The MIT Press,

Cambridge, MA.

Woroch, G. 2002. “Local Network Competition,” In: Cave, M., Majumdar, S., and Vogelsang, I.(eds).

Handbook of Telecommunications Economics, Volume I: Structure, Regulation and Competition.

Amsterdam: North-Holland, 641-716.

Zhu, T., Singh, V., and Manuszak, M. 2009. “Market Structure and Competition in the Retail Discount

Industry,” Journal of Marketing Research, Vol. XLVI, 453-466.

   

53

Figure 1. Number of Wired and Wireless Subscribers (in thousands), 1997-2004

Source: FCC, Trends in Telephone Service, 2005. The numbers of CLEC end-users switched access lines are not available in

FCC’s reports for 1997 and 1998, and the data here are approximated by using FCC (1998a) and FCC (2001) Table 4.

   

54

Table 1. Descriptive Statistics by City Market (N=1,673)

Variable Definition Mean Std. Dev. Min Max

Population Population in thousands 66.82 89.70 20.03 951.46

Income Income per capita in thousands 23.50 8.71 7.29 77.09

Urban Percentage of urban population 98.90 2.99 55.90 100.00

MSA_10 Dummy equals one if city in a Top Ten MSA 0.34 0.48 0 1

REG Regulation stringency 1.77 0.57 0 2

RBOC Dummy equals one if incumbent is a RBOC 0.81 0.40 0 1

CLECS_97 Number of operating CLECs in 1997 0.33 1.00 0 10

CLECS_02 Number of operating CLECs in 2002 1.15 2.72 0 24

Table 2. Descriptive Statistics by CMA Market (N=292) Variable Definition Mean Std. Dev. Min Max

Population Population in thousands 733.23 1576.59 54.54 16134.17

Income Income per capita in thousands 20.42 3.44 9.90 38.35

Urban Percentage of urban population 80.02 12.04 42.00 99.61

MHV Median house value in thousands 107.49 46.90 42.80 422.60

Travel Mean travel time to work (16+) 22.52 3.53 15.14 35.94

WLESS_97 Number of operating wireless carriers in 1997 3.29 1.31 1.00 6.00

WLESS_02 Number of operating wireless carriers in 2002 5.55 1.41 2.00 8.00

Table 3. Descriptive Statistics by National and Regional CLECs 1997 2002

National Regional National Regional

Variable Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

Age 10.83 25.48 5.17 12.18 11.93 20.77 8.47 14.38

Subsidiary 0.39 0.50 0.30 0.46 0.21 0.41 0.48 0.50

Private 0.22 0.43 0.83 0.38 0.41 0.50 0.78 0.42

Share 4.70 11.04 0.20 0.58 2.66 6.97 0.11 0.19

Total 18 76 29 64

   

55

Table 4. The Frequency Distribution of Market Structure by Sector

Number of Carriers

1997 2002

CLECS WIRELESS CLECS WIRELESS

Count Percent Count Percent Count Percent Count Percent

0 1412 84.4 - - 1121 67.01 - -

1 135 8.07 14 4.79 247 14.76 - -

2 58 3.47 82 28.08 89 5.32 5 1.71

3 25 1.49 76 26.03 53 3.17 25 8.56

4 19 1.14 52 17.81 34 2.03 32 10.96

5 9 0.54 56 19.18 24 1.43 61 20.89

6+ 15 0.89 12 4.11 105 6.28 169 57.88

Total 1673 100 292 100 1673 100 292 100

Table 5. The Joint Frequency Distribution of Market Structure by National and Regional CLECs

A. 1997 Data

Regional CLECS

National CLECS 0 1 2 3 4+ Total

0 1412 71 3 0 0 1486

1 64 40 3 0 0 107

2 15 15 4 0 0 34

3 7 9 3 1 0 20

4+ 10 9 4 1 2 26

Total 1508 144 17 2 2 1673

B. 2002 Data

Regional CLECS

National CLECS 0 1 2 3 4+ Total

0 1121 102 12 3 2 1240

1 145 31 13 4 5 198

2 46 30 6 3 5 90

3 7 14 7 5 2 35

4+ 15 50 18 11 16 110

Total 1334 227 56 26 30 1673

   

56

Table 6. Entry Model for the Wireless Sector

1997 2002

Variable Estimate Std. Error Estimate Std. Error

CMA_LNPOP 0.795* 0.098 0.944* 0.094

CMA_LNINC 0.777 0.553 0.099 0.556

CMA_URBAN -0.008 0.007 0.022* 0.007

CMA_TRAVEL 0.049* 0.023 0.007 0.023

CMA_LNMHV -0.275 0.282 -0.019 0.285

N_2 -1.218* 0.309 - -

N_3 -1.265* 0.173 -0.975† 0.559

N_4 -1.025* 0.204 -1.246* 0.289

N_5 -0.977* 0.223 -0.962* 0.242

N_6 -0.146 0.371 -0.147 0.205

N_7 - - -0.150 0.180

N_8 - - -0.160 0.315

Log-Likelihood -891.38 -770.28

Observations 292 292

Note: Asterisk (*) denotes significance at the 5% confidence level and (†) denotes significance at 10%

confidence level.

   

57

Table 7. OLS Regression for Number of CLECs 1997 2002

National CLECs Regional CLECs National CLECs Regional CLECs

Estimate Std. Error Estimate Std. Error Estimate Std. Error Estimate Std. Error

Log (Population in Thousands) 0.631* 0.050 0.223* 0.024 1.690* 0.124 0.553* 0.043

Log (Income in Thousands) 0.054 0.043 0.077* 0.024 0.391* 0.147 -0.011 0.041

Urban Ratio -0.004 0.003 0.001 0.001 0.004 0.007 -0.007 0.004

City in a Top 10 MSA -0.309* 0.062 -0.130* 0.036 -0.395* 0.153 -0.104† 0.059

Regulatory Stringency 0.105* 0.031 0.032* 0.015 0.232* 0.075 0.130* 0.037

RBOC-Incumbent 0.080* 0.036 0.062* 0.020 0.324* 0.102 0.124* 0.038

No. of Potential National CLECS -0.011 0.009 -0.015* 0.005 0.002 0.013 -0.026* 0.008

No. of Potential Regional CLECS -0.008† 0.005 0.005* 0.002 -0.020 0.017 0.030* 0.006

No. of Potential Entrants_CMA 0.031* 0.007 0.031* 0.004 0.038* 0.016 0.051* 0.007

No. of Wireless Carriers 0.305 0.282 -0.597* 0.171 -0.345* 0.597 -2.892* 0.315

Inverse Mills Ratio -0.035 0.044 0.070* 0.025 -0.096 0.083 0.287* 0.037

Constant -2.608* 0.502 -0.201 0.277 -7.546* 1.344 4.081* 0.656

R-Squared 0.376 0.208 0.326 0.326

Observations 1673 1673 1673 1673

Note: Dependent variables are the numbers of national and regional CLECs. Asterisk (*) denotes significance at the 5% confidence level and (†) denotes significance

at 10% confidence level.

   

58

Table 8. Simulated Estimation Result-1997

(I) National First (II) Regional First (III) Most Profitable First (IV) Incumbent First

National Regional National Regional National Regional National Regional

Log (Population in Thousands) 1.547* 1.055* 1.531* 1.017* 1.427* 1.183* 1.536* 1.119*

(0.079) (0.084) (0.092) (0.080) (0.082) (0.101) (0.071) (0.092)

Log (Income in Thousands) 0.202 0.278 0.225* 0.257† 0.320* 0.379* 0.193* 0.238*

(0.125) (0.157) (0.094) (0.125) (0.080) (0.145) (0.087) (0.091)

Urban Ratio 0.060* 0.020* 0.063* 0.025* 0.070* 0.016† 0.057* 0.019*

(0.016) (0.005) (0.011) (0.009) (0.004) (0.009) (0.005) (0.006)

City in a Top 10 MSA -0.059 0.164 -0.050 0.160 -0.090 0.172 -0.055 0.152

(0.357) (0.183) (0.384) (0.401) (0.185) (0.244) (0.175) (0.168)

Regulatory Stringency 0.095 0.023 0.139 0.030 0.083 0.023 0.081 0.022

(0.112) (0.139) (0.200) (0.164) (0.076) (0.090) (0.087) (0.133)

RBOC-Incumbent 0.060 0.057 0.054 0.056 0.029 0.050 0.062 0.062

(0.328) (0.194) (0.181) (0.280) (0.095) (0.233) (0.253) (0.196)

Constant -14.456* -3.048* -14.175* -2.457* -14.328* -3.966* -14.146* -3.808*

(1.174) (0.565) (0.920) (0.622) (0.415) (0.864) (0.577) (0.670)

Log (Age+1) -0.241 0.058 -0.316 0.054 -0.258 0.115 -0.292 0.082

(0.145) (0.147) (0.537) (0.372) (0.149) (0.163) (0.169) (0.204)

Subsidiary 0.035 0.028 0.036 0.027 0.030 0.056 0.030 0.029

(0.760) (0.380) (0.320) (0.355) (0.107) (0.505) (0.254) (0.225)

Private -0.632* -0.473* -0.614* -1.287* -0.830* -1.247† -0.735* -0.585*

(0.108) (0.145) (0.199) (0.497) (0.198) (0.603) (0.094) (0.218)

Log (Distance_to_HQ+1) -0.101† -0.260* -0.106† -0.384* -0.108* -0.206* -0.108† -0.253*

(0.057) (0.059) (0.049) (0.141) (0.043) (0.057) (0.057) (0.115)

Log (No. of National CLECS+1) -0.714* -0.080 -0.909* -0.320 -0.650* -0.236 -0.691* -0.243

(0.273) (0.105) (0.296) (0.611) (0.268) (0.193) (0.210) (0.302)

Log (No. of Regional CLECS+1) -0.346 -0.474* -0.404 -0.337* -0.267 -0.635* -0.374 -0.464†

(0.240) (0.153) (0.340) (0.118) (0.189) (0.285) (0.301) (0.242)

Log (No. of Other Wireless Carriers+1) -0.284 -3.603* -0.064 -3.497* -0.250 -3.031* -0.276 -3.095*

(0.232) (0.360) (0.554) (0.302) (0.214) (0.612) (0.243) (0.349)

D_Providing_Wireless_Service 0.109 0.009 0.172 0.016 0.181 0.015 0.100 0.009

(0.855) (0.384) (0.572) (0.315) (0.442) (0.337) (0.386) (0.477)

Omega -0.140 0.401* -0.114 0.429* -0.086 0.303* -0.133† 0.300†

(0.238) (0.092) (0.067) (0.130) (0.091) (0.115) (0.072) (0.113)

Sigma 0.644* 0.632* 0.745* 0.602* 0.617* 0.797* 0.855* 0.568†

(0.243) (0.115) (0.341) (0.144) (0.319) (0.356) (0.242) (0.207)

Value of the Objective Function 166.5 127.9 184.8 102.6

Observations 1,673 1,673 1,673 1,673

Note: Asterisks (*) denote significance at the 5% confidence level and (†) denotes significance at 10% confidence level.

   

59

Table 9. Simulated Estimation Result-2002

(I) National First (II) Regional First (III) Most Profitable First (IV) Incumbent First

National Regional National Regional National Regional National Regional

Log (Population in Thousands) 1.264* 1.049* 1.113* 1.267* 1.297* 1.203* 1.049* 1.205*

(0.107) (0.254) (0.165) (0.093) (0.061) (0.173) (0.074) (0.186)

Log (Income in Thousands) 0.401* 0.333 0.499* 0.259* 0.416* 0.306† 0.421* 0.231

(0.104) (0.336) (0.136) (0.094) (0.088) (0.160) (0.090) (0.171)

Urban Ratio 0.091* 0.023* 0.080* 0.012† 0.093* 0.028* 0.080* 0.012

(0.008) (0.003) (0.014) (0.007) (0.006) (0.007) (0.007) (0.014)

City in a Top 10 MSA -0.056 -0.609* -0.122 -0.576* -0.054 -0.566* -0.068 -0.417*

(0.167) (0.220) (0.350) (0.122) (0.190) (0.196) (0.133) (0.108)

Regulatory Stringency 0.086 0.186 0.111 0.144 0.117 0.151 0.067 0.233*

(0.110) (0.216) (0.241) (0.112) (0.282) (0.155) (0.099) (0.114)

RBOC-Incumbent 0.299* 0.554* 0.376† 0.459* 0.301* 0.512* 0.386† 0.607*

(0.119) (0.265) (0.196) (0.222) (0.128) (0.209) (0.179) (0.242)

Constant -13.743* -0.622* -15.129* -0.764* -14.209* -0.600* -14.633* -0.769*

(0.450) (0.238) (1.679) (0.211) (0.776) (0.272) (0.508) (0.275)

Log (Age+1) 0.008 -0.225 0.052 -0.265 0.062 -0.225 0.068 -0.254

(0.126) (0.166) (0.130) (0.173) (0.066) (0.412) (0.083) (0.211)

Subsidiary -0.025 -0.381 -0.025 -0.296 -0.028 -0.299 -0.028 -0.193

(0.372) (0.554) (0.568) (0.299) (0.346) (0.567) (0.558) (0.233)

Private -0.132* -0.231* -0.183* -0.235* -0.116* -0.249* -0.114* -0.237*

(0.042) (0.032) (0.021) (0.025) (0.030) (0.067) (0.019) (0.032)

Log (Distance_to_HQ+1) -0.104† -0.656* -0.093† -0.767* -0.094† -0.771* -0.097* -0.756*

(0.059) (0.093) (0.048) (0.094) (0.049) (0.065) (0.045) (0.142)

Log (No. of National CLECS+1) -1.479* -0.366 -1.612* -0.287 -1.742* -0.416 -1.462* -0.239

(0.169) (0.362) (0.448) (0.382) (0.167) (0.307) (0.336) (0.316)

Log (No. of Regional CLECS+1) -0.336 -0.849* -0.213 -0.692* -0.242 -0.428* -0.296 -0.512*

(0.301) (0.312) (0.385) (0.310) (0.277) (0.207) (0.316) (0.112)

Log (No. of Other Wireless Carriers+1) -0.341* -2.465* -0.269† -1.912* -0.411* -2.791* -0.302* -2.429*

(0.094) (0.288) (0.148) (0.404) (0.156) (0.323) (0.103) (0.766)

D_Providing_Wireless_Service 0.297 0.360 0.247 0.366 0.230 0.291 0.259 0.466

(0.179) (0.885) (0.422) (0.733) (0.135) (2.301) (0.211) (2.335)

Omega -0.101† 0.189† -0.105† 0.180* -0.127† 0.249* -0.157* 0.172*

(0.054) (0.106) (0.060) (0.064) (0.064) (0.109) (0.050) (0.037)

Sigma 0.995* 0.826* 0.822* 0.961* 1.387* 0.917† 0.983* 1.001*

(0.150) (0.301) (0.197) (0.214) (0.145) (0.466) (0.186) (0.320)

Value of the Objective Function 111.9 98.3 125.3 93.3

Observations 1,673 1,673 1,673 1,673

Note: Asterisks (*) denote significance at the 5% confidence level and (†) denotes significance at 10% confidence level.

   

60

Table 10. Entry Threshold Estimates for Wireless Sector Market Entry Threshold Per firm

With N Population (1,000) Entry Threshold Ratio (sN+1/sN)

Carriers 1997 2002 1997 2002

N=2 133 - 3.28 -

N=3 652 298 2.72 2.80

N=4 2,366 1,115 2.73 2.22

N=5 8,089 3,087 1.00 0.97

N=6 9,721 3,606 - 1.00

N=7 - 4,227 - 1.04

N=8 - 5,005 - -

Table 11. Top 10 Wireless Carriers by Year 1997 2002

FIRM SHARE FIRM SHARE

AirTouch 12.08 Verizon 23.08

AT&T 10.88 Cingular 15.56

SBC 9.82 AT&T 15.44

Bell Atlantic 9.68 Sprint 10.49

BellSouth 9.11 Nextel 8.16

GTE 8.11 T-Mobile 7.04

Ameritech 5.74 Alltel 5.40

360 Com 4.67 US Cellular 2.91

US Cellular 3.09 Cricket 1.07

Nextel 2.30 Western 0.85

Note: Affiliates' shares are added to parent firms' shares.

Figure 2. Industry Ratios of s6 to sN by N

s6 /sN

   

61

Table 12. Model's Goodness of Fit for Incumbent Moves First Game

1997 2002

Number of Sample Mean Model Mean Correlation Sample Mean Model Mean Correlation

National CLECs 0.2140 0.2011 0.746 0.8153 0.7953 0.862

Regional CLECs 0.1148 0.1153 0.734 0.3317 0.3401 0.799

Note: Correlation here means the correlation between the number of carriers predicted by the model and the observed number of carriers in each market, and a high

value indicates a good fit of the model to the data-generating process.

Mea

n N

um

ber

of

CL

EC

s

Mea

n N

um

ber

of

CL

EC

s

Number of Other Wireless Carriers Number of Other Wireless Carriers

Mea

n N

um

ber

of

CL

EC

s

Number of Other Wireless Carriers Number of Other Wireless Carriers

Mea

n N

um

ber

of

CL

EC

s

Figure 3. Predicted Means of National and Regional CLECs

     

62

Chapter 2

Predicting Merger Outcomes: How Accurate Are Stock Market Event Studies,

Market Structure Characteristics, and Agency Decision?

An essay co-authored with John Kwoka

     

63

1. Introduction

Merger analysis, in contrast to most areas of antitrust enforcement, is largely an exercise in prediction.

Under the provisions of the Hart-Scott-Rodino Act, the Justice Department (DOJ) and Federal Trade

Commission (FTC) in the U.S. must be notified of a prospective merger meeting certain size criteria, and

then given a specified period of time in which to evaluate the merger for possible competitive problems and

perhaps to file a legal challenge. This process is unlike enforcement actions brought against cartels or against

dominant firm behavior, where typically the alleged anticompetitive practice has already occurred and the

remaining questions involve determination of causation and measurement of the extent of harm. In the case

of mergers, except for the occasional challenge to a consummated merger, the antitrust agency must be able

to predict the outcome of an event that differs to varying degrees from all past experience and then, if it

decides to challenge the merger, to convincingly explain the basis for that prediction to a court of law.

The task of prediction has generally been met by some combination of economic theory, empirical

evidence, and past experience. At least since 1968, the prevailing approach has been summarized in the

Horizontal Merger Guidelines issued by the Justice Department and later jointly with the Federal Trade

Commission. The first Guidelines articulated a rigorous structural standard--one that relied predominately

on market shares and concentration as the basis for predicting the outcomes of mergers and the likelihood

of an antitrust challenge. Subsequent revisions of the Merger Guidelines altered those strict numerical

standards and also broadened the analytical framework beyond market structure. Thus, market share and

concentration thresholds were relaxed, and additional factors–notably, entry conditions and possible

efficiencies–were more fully integrated into the analytical process. With the issuance of the 2010 Merger

Guidelines, this more expansive and eclectic analytical approach reached its zenith. The new Guidelines

explicitly state that “merger analysis does not consist of uniform application of a single methodology”, but

rather the application of a “range of analytical tools.”1

                                                            1 Horizontal Merger Guidelines, Aug. 2010, p. 1.

     

64

Throughout, many observers have been concerned about the predictive power of a largely structural

standard, which relies on market concentration, shares, and entry conditions, for merger enforcement.

Concerns have been raised about both its overall error rate and its supposed tendency toward incorrectly

forecasting anticompetitive effects from competitively benign mergers (Type I errors). Prodded as well by

the insistence of the courts and enabled by advances in economics and finance, some have advocated

alternative or supplementary methods of prediction. Over the past 25 years, one leading contender has been

stock market event study methodology. These event studies rely on evidence from changes in the stock

prices of the merging firms and their rivals when the prospective merger becomes publicly known, to infer

possible competitive problems with the merger.

The objective of this paper is to evaluate the relative accuracy of predictions based on stock market

event studies with those based on market structure variables, and also to compare both of those with the

predictions implied by the actual decisions of the antitrust agencies with respect to these same mergers. Since

antitrust agencies utilize market structure characteristics as well, perhaps, as event studies in their own

determinations, these determinations are not entirely independent of the other predictions. But the agencies

do not simply engage in a mechanical application of any single technique, so the accuracy of their predictions

serves as a measure of the value added from their investigative process and analytical approach.

In order to accomplish these objectives, the actual outcomes of a significant number of mergers must

be known. Here we exploit the growing literature on merger retrospectives. As will be discussed below,

retrospective studies analyze the actual outcomes of mergers—focusing on price—typically using the

difference-in-differences approach to control for other influences. The findings of these studies represent a

relatively consistent set of data on the known outcomes of these mergers. Then for each of these mergers,

we implement the two predictive methods. With respect to stock market event studies, we conduct entirely

new studies of the relevant stock price movements at the time that each merger in our data set was announced.

In order to test the market structure approach, we utilize two approaches. One is based on the structural

criteria in the Horizontal Merger Guidelines, the second on a model that empirically identifies the relevant

     

65

market structure variables to the actual outcomes for the mergers. And finally, the actual decisions rendered

by the relevant antitrust agency on each merger in our data base are recorded.

The accuracy of the predictions from stock market event studies, from market structural characteristics,

and from agency decisions themselves are the all compared to the known outcomes of each merger as

determined in the corresponding retrospective study.2 As will be detailed in the following sections, the

evidence supports three major conclusions. First, event studies prove to be poor predictors of the actual

outcomes of mergers, failing to correctly identify anticompetitive mergers in more than three-fourths of

cases (large Type II errors). Statistical tests confirm that predictions based on premerger stock price

movements of the relevant firms are statistically independent of actual post-merger price changes. Various

possible reasons for this are discussed and tested.

Second, conventional market structure factors--concentration and related measures--are in fact equally

weak predictors of individual merger price changes, but their errors are in the opposite direction. That is,

they over-predict the frequency of anticompetitive outcomes by a similarly large margin so that strict

reliance on market structure would indeed result in large Type I errors. As we later note, however, and

efficient merger screening rule for enforcement purposes might well involve over-inclusiveness to ensure

that all possibly anticompetitive mergers are initially brought to the agencies’ attention, after which a

detailed analysis of such mergers is undertaken.

And thirdly, actual agency decisions whether or not to challenge these mergers are examined and reveal

that, while not without error, these decisions in fact involve fewer errors than reliance on either event studies

or purely structural criteria. This result suggests significant value added by the antitrust agencies in their

determinations.

The paper proceeds as follows. The next section discusses the mergers and the data sets that comprise

                                                            2 An additional method of prediction—merger simulation—cannot be incorporated into the present study due to the

considerable data requirements of simulation. 

     

66

the foundation for this study--retrospective studies, event studies, market structure variables, and agency

decisions. Discussion of these data sets will illuminate the debates over merger prediction. Section 3

describes the new stock market event studies that are performed in this research in somewhat greater detail,

including the sample collection procedure, estimation methodology, and correction for sample selection

bias. Section 4 presents the empirical results--the estimation results for event studies and the key

comparisons of their predictive power with that of market structure criteria and agency decisions. Section 5

concludes.

2. Qualifying Mergers

The key data constraint in this study is reliable information about actual price changes resulting from a

sufficient number of mergers to permit valid inferences. A growing amount of such information has become

available in the form of merger retrospectives. These studies isolate the effect of a merger on product price

(less often, other outcome variables), holding other factors constant. While the alternatives of structural

estimation and other techniques are sometimes employed, the most frequent methodology has been use of

the so-called difference-in-differences (DID) methodology.

The DID approach compares the price change for the product arguably affected by the merger (say, Pt2

- Pt1: “t” for treatment group) with the price change for some otherwise comparable product not affected by

the merger (Pc2 - Pc1: “c” for control group). With a proper choice of the “otherwise comparable” control

product, the difference in these differences

δ = (Pt2 - Pt1) – (Pc2 - Pc1) (1)

measures the effect of the merger without the need for elaborate structural modeling.3 Operationally, DID

is often implemented using pooled cross-section data, so that all that is required is estimation of an equation

                                                            3 The use of price as the performance variable avoids the difficulty of disentangling market power and efficiencies: a

price increase implies an adverse net effect on consumers.

     

67

of the following form:

∗ , (2)

here TREAT is a dummy variable that equals one for the treatment group observations, POST denotes the

post-merger time period (again, with a dummy variable), and the coefficient on TREAT*POST measures the

difference in the dependent variable between the premerger and postmerger periods for the treatment group,

relative to the control group as the base case. Additional control variables are typically required, but their

presence does not affect the logic of the DID methodology.

Kwoka (2013) has recently surveyed hundreds of candidate merger retrospective studies in the

literature. For purposes of quality and consistency, various criteria are used to screen out unusable estimates

of post-merger price changes. The criteria are as follows: (1) The retrospective study must have examined

a purely or substantially horizontal transaction. Some "partial" mergers in the form of joint ventures and

airline code-sharing agreements are included as well, since these are often thought to have effects

qualitatively similar to those from pure mergers, but true vertical mergers are omitted, as they raise different

economic and policy issues. (2) Since policy takes place at the level of the individual merger or transaction,

studies that report only the average outcome for groups of mergers4 are excluded. Those studies may be

informative about the effects of mergers, but the outcomes cannot be matched to merger-specific policies. (3) Given

the focus on domestic policy, only transactions involving U.S. companies and markets are included. (4) The

study must use a recognized technique such as difference-in-difference that meets modern standards of

research design.5 (5) Lastly, the study must appear in a peer-reviewed journal in economics or related

discipline, or in a respected working paper series such as those of the NBER, FTC, or DOJ.

                                                            4 For example, Kim and Singal (1993) report the average price effect for fourteen airline mergers. Such studies may

shed some light on the mergers collectively, but they do not permit matching the individual transactions to policy actions

and hence are not included for current study 5 In fact, all the studies in our data base employ some kinds of difference-in-difference analysis. A few offer alternative

approaches such as structural modeling as well as difference-in-difference estimates.

     

68

This screening process results in some 53 usable estimates of the price effects from qualifying mergers

and other transactions. Data limitations associated with other data sets described below cause the loss of

several additional observations. The final data base used throughout this study consists of forty observations

on horizontal transactions analyzed in the retrospectives literature. These are listed in the Appendix Table,

along with various characteristics of each that will be described below.

3. Prices and Predictions

For each of these qualifying mergers, data are compiled or developed on the actual post-merger price,

on the prediction from stock market event studies that are conducted here and could have been conducted at

the time of the merger, on the predictions from the Horizontal Merger Guidelines and from characteristics of

the pre-merger market structure, and on the actual agency decisions or actions with respect to the mergers.

The following subsections explain the sources for each of these.

3.1. Postmerger Prices from Retrospectives

We extract or develop from each retrospective study a single summary measure of the magnitude of the

estimated post-merger price change. The protocol for obtaining that measure is as follows: First, if the

merger retrospective reported only one estimate or the author identified a single central estimate that was

recorded as the finding of the study. In most cases, however, multiple results are reported in a study of a

single merger, due, for example, to alternative model specifications, different levels of aggregation, or a

multiplicity of products. In these cases the second step to the protocol involved taking the average of estimates

across the key non-nested and non-duplicative results, that is, omitting secondary estimates and those that

might overlap.

In most cases this process yields a single estimate for each merger in each study. In a few instances,

however, the same merger has been studied in more than one retrospective. In these cases the third step of

protocol was to take the average across the studies of the single point estimate (calculated as just described)

for that merger. This procedure results in a single overall value for each merger that represents the best

     

69

information about its outcome in the literature, while avoiding judgments and subjectivity in the recorded

price outcomes.

Estimates of the price outcomes are available for forty mergers. All are derived from a well-accepted

methodology, carefully screened, and rendered in a consistent form. While these will ultimately be used as

the basis for evaluating the accuracy of predictions, they are of some interest themselves. We note, for

example, that the mean price change for all these transactions is an increase of 5.65 percent, implying that

on average these transactions were anticompetitive. The average, however, obscures considerable variation

in the results. As shown in Table 1, the price changes range from a decrease of 7.20 percent to an increase

of 29.4 percent. The median change is an increase of 2.47 percent. While that is more modest, fully 31 of

the forty cases are found to result in price increases, with only nine decreases. Moreover, nine of the forty

mergers result in increases in excess of ten percent. These results will represent the benchmark against

which various methods of prediction will be compared.

3.2. Predictions from Event Studies

The second set of necessary data consists of predictions of the competitive effects of these transactions

based on stock market event studies. Here we review the basis for this approach and then describe its

application in this study.

3.2.1 Background

The theory underlying stock market event studies is well known: According to the efficient market

hypothesis, stock prices capture all available public information about a firm’s financial prospects. Thus, a

merger that is likely to increase market power can be expected to raise the stock prices of the merging

companies relative to some benchmark group. By itself, examining merging companies’ stock prices is

insufficient since a merger that promises increased efficiencies will also raise those firms’ profits and stock

price. To distinguish market power from efficiencies, advocates argue that one should examine the stock

prices of rivals to the merging companies. An efficiency enhancing merger should cause rivals’ stock prices

to fall since those rivals would be facing a more formidable merged competitor. But if rivals’ stock prices

     

70

were to rise, the inference is that all firms in the market would benefit from the merger, an outcome consistent

with enhanced market power.

This argument formed the basis for the view that stock market event studies could and should be relied

upon for predicting merger outcomes. Empirical work beginning with Fama et al (1969) and in the merger

context with Eckbo (1983) and others was interpreted as support for this position. For example, Eckbo and

Weir (1985) compute the abnormal returns for 269 merging firms and their rivals, both at the times of the

merger announcement and at the time of any antitrust challenge. They concluded that the results supported the

efficiency explanation for these mergers and that the traditional approach to identifying competitive problems

with mergers often erred. “All but the ‘most overwhelming large’ mergers should be allowed to go forward”

(Eckbo and Weir, 1985).

Much debate followed these initial findings, with considerable attention to possible limitations and

other interpretations of the finding of abnormal returns (Eckbo, 1985; MacKinlay, 1997): Among the

generally recognized technical limitations are the following: For merging firms that are large and diversified,

it may be difficult to identify abnormal returns to a merger affecting one product. Further, to the extent that

rivals are small or have thinly traded stocks, their stock price effects may be difficult to estimate reliably. In

addition, information about mergers and other events often leaks out prior to the official announcement, so

that a narrow focus on the announcement date may erroneously imply the absence of effect. And

furthermore, stock returns logically reflect not just the merger but also the likelihood of an antitrust

complaint and its likelihood of prevailing.

While some of these concerns can be addressed by careful modeling,6 certain issues of interpretations

                                                            6 The information leakage issue is now commonly addressed by using various event windows or by using the date at

which abnormal returns first appear as the exact date around which to construct the window. A conventional fix for the

confounding effect of other events is to adjust the measured stock price change for those probabilities (the Heckman

procedure), or alternatively to use announcements of events associated with the antitrust review process as additional

date points for analysis.

     

71

may be more problematic. Some have noted that the announcement of a planned merger might highlight

other attractive merger opportunities within the same industry, thereby causing those rivals’ share prices to

increase for reasons other than expected anticompetitive effects. On the other hand, a merger that increases

the likelihood or effectiveness of exclusionary or predatory conduct by the merged firm would diminish the

profit outlook for its rivals. That might cause a decline in the latter’s share prices but not due to any efficiency

from the merger itself.

A long-running and vigorous debate between advocates and critics of event studies has ensued.7

Largely lacking in this debate has been actual evidence concerning the reliability of event studies as a method

for merger evaluation. The reason has not been a dearth of event studies but rather a lack of reliable measures

of the actual post-merger firm performance to which the predictions of event studies could be compared.8

There appears to be only one paper in the literature that utilizes true merger outcome data to draw the

appropriate comparison with event study predictions. McAfee and Williams (1988) examine a merger shown

in a previously published retrospective study to have resulted in substantially higher product price. They

conduct their own after-the-fact stock market event study for that merger and find no indication that the

merger was predicted by the stock market to be anticompetitive. They conclude that at least for that merger,

reliance upon a premerger event study would have resulted in an incorrect prediction and policy conclusion.

Two other studies deserve brief mention in this context. Warren-Boulton and Dalkir (2001) examine

the proposed merger of Staples and Office Depot in 1996. Comparing the price increase implied by a

                                                            7 MacKinlay reviews this debate. See, for example, Werden and Williams, 1989a, 1989b; Eckbo, 1989). Some have

continued to utilize event studies for merger analysis (Prager, 1992; McGuckin et al, 1992; Schumann, 1993; Mullin

et al, 1995; Warren-Boulton and Dalkir, 2001; Hosken and Simpson, 2001; Duso et al, 2006, 2007, 2010) while others

dispute the underlying propositions and their implications (Cichello and Lamdin, 2006; Cox and Portes, 1998). 8 A few papers investigate the correlation between post-merger accounting profit and premerger share price changes

(Healy, et al (1992), Kaplan and Weisbach (1992), Sirower and O'Byrne (1998), Duso et al (2010)). Finding such a

correlation, they conclude that event studies may be useful tools in merger analysis, but in reality these studies only

imply that share price and accounting profitability are similarly affected by mergers.

     

72

premerger event study to the price increases predicted by other techniques introduced into the trial record,

they conclude that event studies would have predicted the product price effect. Beverly (2007) examines

four mergers investigated by the UK Competition Commission. Based on the stock prices of prospective

merging parties and their rivals, she finds “few cases where the stock price movement was large enough to

be conclusively linked” to events in the Commission’s investigatory time line. Clearly some past literature

has expressed caution about reliance on stock market event studies.

3.2.2 Data and Estimation

A major contribution of this study is to replicate the McAfee-Williams approach--comparing

predictions from event studies to actual outcomes based on retrospective studies--for the forty transactions

in our retrospectives data base. As noted, however, these stock market event studies were not in fact

performed in advance of these transactions, and so we here undertake the event studies that could have been

performed at the time of announcement for all forty transactions. For reasons previously detailed, we focus

on changes in rivals’ stock prices.

Rivals are identified according to the following procedure: First, in a substantial majority of cases

rivals are identified in the retrospective studies or in the FTC or DOJ docket sheets accompanying the filing

of cases. For example, Ashenfelter and Hosken (2010) list the competitors together with their market shares

for the five mergers that they study. FTC Docket No. 9150 provides detailed market share information of

rivals for the merger between Weyerhaeuser and Menasha in 1980. Secondly, for airline industry mergers,

we record the locus of competitive concerns for each, whether that be routes, hubs, regions, or national

scope. We then use Department of Transportation DB1A (B) and T-100 data to identify as rivals those

alternative carriers that comprise a significant percent of revenue passenger miles at the same level of

disaggregation. For mergers involving many routes, this may result in multiple rivals insofar as different

carriers match up with the merging carriers on various routes.

Thirdly, in other cases, a similar procedure was used, though typically less complex (and occasionally

less precise) than that for airline mergers. For journal mergers, in addition to the published retrospectives,

     

73

information on rivals in the biomed field can be found in the compilation of the Institute for Scientific

Information, and for legal publications in Svengalis' Legal Information Buyer's Guide & Reference Manual

(2005). For the petroleum industry, upstream rivals for different geographic regions are found in the Energy

Information Administration's annual Refinery Capacity Reports. Rivals to the merging railroads are defined

as competitors with overlapping routes. For the New England bank merger, we rely on the Federal Reserve

Bank of Boston's Banking Structure in New England 1996-1999 and 1999-2001 (Report 75 and 76) for

detailed lists of banks by region.

The resulting list of rivals is included in the Appendix Table. Relevant stock market data are then

extracted from the Center for Research in Security Prices (CRSP) database as follows: We first identify the

announcement date of the proposed merger as the day on which news of the upcoming transaction first appeared

in either the Wall Street Journal or the New York Times.9 Next, information about the stock prices of the

merging parties and of the firms identified as their rivals is compiled. Rivals that were unsuccessful bidders as

well as other firms that were actively involved either as a target or as a bidder in another transaction within

the relevant period were excluded from the rival sample. The unavailability of stock price data for certain firms

(notably, nonprofit hospitals) caused a reduction of the sample to some forty transactions.

Our methodology is that pioneered by Eckbo and Wier (1985) and used by McAfee and Williams

(1988). For the ith merger in our sample, we compute abnormal returns to the rival firms at the time of the

announcement of the merger proposal, and separately for those firms subject to antitrust challenge, at the

time that the formal antitrust complaint was filed. Abnormal returns at any point in time are obtained by

estimating the coefficients of the following regression equation:

(3)

                                                            9 In fact, there are two early cases for which we cannot identify the event dates through either the Wall Street Journal

or the New York Times. For the Xidex/Scott Graphics (1976) case, we use the date showed on FTC's complaint docket

as the event date, and for Xidex/Kalvar (1979), we use the event dates identified by McAfee and Williams (1988).

Our results for the latter case are consistent with the outcomes in their study.

     

74

here ri denotes the daily continuously compounded returns to the equal-weighted portfolio of rival firms,

and rm is the daily continuously compounded returns to the value-weighted CRSP market index. The term

d is a dummy variable that takes on the value of unity for days in the event window and zero otherwise so

that the estimate of represents the average daily abnormal return to the portfolio of rival firms. The term

is the daily random error which is assumed to be independent of rm and d, serially unrelated, and

normally distributed with mean zero and constant variance. Note that pooling the returns of the rival firms

associated with a given merger to create one equal-weighted industry portfolio allows us to avoid any

contemporaneous correlation of returns across firms in the same industry.

The relevant timeline for event studies involves two different periods—the estimation period and the

event window. The estimation period is the total number of days of stock market data that are analyzed,

typically beginning from 120 to 250 days before the event to ensure no overlap (MacKinlay). We use 200

days as the estimation period. The event date is the relevant date for testing for abnormal stock price

movements and usually is simply the announcement date of the proposed merger. This date is defined as

day zero (0). To capture any prepublication leakage of relevant information, five event windows of varying

lengths are employed, denoted (-20, 10), (-10, 5), (-3, 3), (-1, 1) and day zero (0). These describe, for

example, an event window from 20 days before til 10 days after. A dummy variable d takes on the value

of zero for the comparison period (-200, -20), and takes on the value of one for each day in an event window.

For the event windows that lie inside (-20, 10), we delete the returns outside the event window but inside (-

21, 10). For example, for the event window (-3, 3), the returns in the periods (-20, -4) and (4, 10) are dropped.

In a few cases the announcements for a merger (usually in the airline industry) is close in calendar time

to other merger announcements. Although we exclude all firms that are actively involved in another takeover

either in the event window from the portfolio of rivals, this situation could still obscure the true effect of a

merger. As such, we use the method adopted by Knapp (1990). An additional dummy variable is added to

isolate the overlapping effects. For example, if two mergers have some rivals in common, and the first

     

75

proposed merger is announced 20 days before the second merger announcement, then a dummy taking on

the value of one in the period (1, 10) is added for the first merger, and a dummy taking on the value of one

in period (-30, -10) is added for the second.10

The average daily abnormal returns (AR) are calculated in this manner for the rival firms for each

merger in the data base. The total abnormal effect of a merger announcement can be computed by

multiplying by l--the length of the event window. For example, for the period of (-20, 10), a total

abnormal return equals 31 . It is these variables that represent the predictions of the stock market for

each of the mergers in our sample.

3.2.3 Correction for Sample Selection Bias

One issue with this sample is that the mergers and other transactions are not randomly selected from

the pool of all mergers, or all proposed mergers, or all deterred mergers. Most obviously, the abnormal return

to the rival firms resulting from a merger announcement can be viewed as a mixed expectation about future

benefits (that is, the market power effect vs. the efficiency effect), and uncertainty about whether the

government will prevent the merger from taking place.11 A higher post-merger HHI and a larger Delta (the

change in HHI) increases the expected future gains but also drives up the probability of a merger being

challenged by the antitrust agencies. This leads to the classical problem of sample selection bias. As a result,

we follow Eckbo (1985) in using Heckman’s two-step estimator to correct for such selection bias.

If we define as the predicted probability of a merger been challenged by antitrust agencies, then the

adjusted abnormal return (AAR) can be written as:

, (4)

                                                            10 Robustness tests show that the coefficients change only slightly when these dummy variables are removed from the

regression. 11 Due to the size of our sample, here we do not distinguish the actions enforced by the agencies. A larger sample

would also permit use of a multinomial model to predict the probabilities of different types of enforcement actions,

such as opposition. Imposition of conditions, or market structure (divesture) and conduct remedies.

     

76

where 1 is the predicted probability, conditional on the merger proposal, that the ith merger will be

consummated finally. To derive this probability, a dummy variable--CHALLENGED--is constructed,

taking on the value of one if the proposed merger was challenged by the FTC or DOJ, and then it is regressed

on the market characteristics variables. Let be a vector of industry-specific characteristics associated

with the ith merger at the time of merger proposal, and let be a binary variable such that

'1, if 0

0, i i

i

Xy

otherwise

(5)

where is a parameter vector and is a random variable. Assuming has the logit density function, the

probability can be written as a linear function of the merger characteristics

          '

' 1( 1) ( )

1 ii i i i X

p prob y prob Xe

                                                                                       

(6) 

The logit estimation results confirm that the probability of challenge increases in a highly concentrated

post-merger market with a high Delta and substantial entry barriers. These results are used to adjust the

apparent abnormal return for the probability that the merger might be challenged. The resulting Adjusted

Abnormal Return (AAR) variable is the principle outcome variable analyzed below.12

3.3. Predictions from Market Structure Variables

The third data set required for this study consists of market structural characteristics for each merger,

in particular, pre-merger HHI, post-merger HHI, entry conditions, and the number of direct competitors.

The basic sources of such information on both merging firms and rivals have been described above. Again,

merger retrospectives and agency case dockets contain not only the identities but often data on market

shares and related characteristics of the markets arguably affected by the mergers. For airline mergers

                                                            12 A reviewer has suggested that this correction might not have great power. Accordingly, we have conducted the

analysis below on the set of uncorrected abnormal returns and found much the same results. This result is consistent

with that comment, but we use the corrected data series below with the belief that may represent at least a partial

correction.

     

77

involving clearly national carriers, we use their nationwide market share to calculate HHI.13 If the concern

lies with specific local or regional markets and overlaps, we follow Singal (1996) in calculating the

percentage of an airline's revenue passenger miles (RPM) on overlapping routes.

In the case of biomedical and law journals, market shares are approximated from available data on

the number of journals in each category controlled by each publisher, and on the total number of journals in

that category as listed in McCabe (2000, 2004) and the other previously identified sources. Then market

shares are calculated as the number of relevant titles held by each firm, divided by total number of such

titles.14 For the petroleum industry mergers, information on HHI is supplied by FTC (2004) on several

merger cases and can be developed for other mergers from capacity and share data in the Energy Information

Administration’s Refinery Capacity Reports.15 Since data from the retail level for individual cities are not

available, wholesale concentration is used with the expectation that it should be correlated. For the two

acquisitions in the railroad industry, a proxy for shipments concentration at this level is given by shares of

trackage, obtained from the Surface Transportation Board's Class I Railroad Annual Reports. For a total of

16 mergers, data on market characteristics could be developed in this manner.

                                                            13 For example, Borenstein (1990) emphasizes the issue of shared hubs for the Northwest-Republic merger. In this

case, we use the market shares at the Minneapolis/St. Paul (MSP) airport to calculate HHI. Such data is collected from

Maldutis (1987) which provides the HHI for 50 U.S. largest U.S. airports for the period of 1977-1987. Note that DOT's

Form 41 data only start from 1990 on its website. 14 McCabe (2004) and American Association of Law Librarians define a set of the "480 most important" titles from

eight broad categories of commercial, print legal serials. Here we use 480 as the total number of titles, as the other

"less important" titles may not be close substitutes and will be less affected by the merger. These measures do not

include weights, and hence measure shares and concentration with error. 15 This method to calculate the market share is consistent with the method used by most of gasoline studies in our

sample. See Hosken et al (2011) for example. For some transactions involved only wholesale-level concentration

changes (such as terminals and marketing),we identify competitors through FTC (2004, 2011) which provide PADD-

level and state-level wholesale concentration estimates for each year from 1994-2010. For a few cases which have

concentration changes in several markets, we use the concentration change for the market emphasized by the studies

to make them comparable with the subsequent analysis on price changes.

     

78

A summary of these market structure characteristics is reported in Table 1. Despite some imprecision

to these market characteristics data, we shall see that they permit a useful comparison between the predictions

of stock market event studies and predictions based on market structure factors.16

3.4. Antitrust Agency Determinations

Finally, we need information about the competitive assessment of each merger by whichever U.S.

antitrust agency--the FTC or the DOJ--conducted the investigation. It should be noted, of course, that agency

determinations are likely not independent of other predictive techniques since inevitably agency assessments

rely in part on structural characteristics of the merger and its market, as well perhaps as internally conducted

event studies and other techniques. Nonetheless, to whatever extent they differ from, say, market structure-

based assessments, agency decisions do capture the incremental contribution--positive or negative--of

agency investigative techniques, analytical approaches, and judgments on each merger.

Compiling information on agency assessments is complicated by the fact that the assessments are

internal to the agency and hence confidential and unavailable for this study. We use as a proxy for those

assessments the agency’s decisions whether or not to challenge each merger. In most cases this should serve

as a good guide to the internal assessment of the merger. Divergences may occur, however, for certain

reasons. For example, the costs and/or likelihood of prevailing in a challenge may be viewed as

disproportionate to the benefits from proceeding against a competitively problematic merger. Another

possible reason would be if some non-economic consideration played an important role in the decision

whether to issue a challenge.

While bearing these caveats in mind, the agencies’ actual decisions can reasonably be taken as good

guidance for their predictions about the likely competitive effects of these transactions. Accordingly, for

each merger in the data set, a search of public records was conducted to determine whether the relevant

                                                            16 Moreover, any imprecision should, if anything, weaken the power of our test and bias the results toward the absence

of any finding.

     

79

antitrust agency in fact challenged the merger or cleared it. As commonly used by the agencies, the term

“challenge” includes not only the actual filing of a complaint but also instances in which the parties abandon

a merger in the face of agency opposition or modify it in order to secure agency clearance. In addition to any

trial record, the public records consulted include consent orders, competitive impact statements, Tunney Act

filings, agency press releases, and credible press reports. From these a complete picture of enforcement

decisions and actions is constructed.

4. Results of Empirical Testing

Having created and compiled all the necessary data, we now proceed to test the hypotheses set out

earlier. These involve comparing the predictions of (a) stock market event studies, (b) market structure

characteristics, and (c) antitrust agency decisions with the actual outcomes of mergers as established by the

corresponding retrospective study or studies, as described above. At a first approximation we classify any

merger resulting in a price increase as anticompetitive, although we also examine cases involving price

changes in excess of some minimum size and to those achieving some level of statistical significance.17 Most

of these tests are now fairly straightforward.

4.1. Abnormal Returns and Price Outcomes

We begin by testing the accuracy of stock market event studies as a method of predicting the actual

effects of mergers on prices. As described above, this involves a determination as to whether the stock prices

of rivals to the merging firms exhibit abnormal increases at the time of announcement. In the analysis just

                                                            17 This is in fact a sufficient but not necessary condition for an anticompetitive outcome in these data, since many of

these mergers were subject to challenge and attempts at remedial action by the antitrust agencies. If that action reversed

the price effect of a merger, a finding of no price change would be ambiguous, whereas in fact in this sample almost

all successfully challenged cases experience some price increases after the merger. In only two cases, the merger

between Weyerhaeuser and Menasha, and the code-sharing agreement between Northwest and Continental,

experienced some price decrease given the challenge from antitrust agencies. Also, we perform robustness checks on

this definition below, examining, for example, outcomes that are statistically significant or of some minimum size as

indicative of anticompetitive outcomes.

     

80

performed, five different event windows were examined. For present purposes, we adopt the conservative

rule that a merger is predicted to be anticompetitive if positive abnormal returns to the merging firms’ rivals

are found for any one of the five windows.

The results of this analysis are displayed in Table 2. Even by this inclusive criterion, event studies

report a total of only 9 of the 40 mergers to be anticompetitive, far fewer than the 31 determined by

retrospective studies as actually anticompetitive. Moreover, of those 9 mergers predicted to be

anticompetitive by event studies, only 7 correspond to the 31 mergers found actually to be anticompetitive

by retrospective studies. That implies that event studies made the correct determination in only seven of 31

anticompetitive cases, representing a success rate of 22.6 percent. In more than three-fourths of cases,

reliance on event studies would have resulted in incorrectly absolving the merger of competitive concerns.

On the other hand, of the nine mergers where prices did not increase, event studies would have correctly

predicted all but two to be competitively harmless. While this success rate of 77.8 percent appears to be

commendably high, it actually reflects the overall tendency for stock price event studies not to find positive

abnormal returns in either circumstance--not when returns actually are positive as well as when they are not.

Indeed, the rate at which event studies predict price increases is essentially the same for both populations--

22.6 percent vs. 22.2 percent--implying that such studies have virtually no discriminating power. We can

formally test whether their success rates (while low overall, with only 9 out of 40 correct) are the same in

the two samples. A means proportion test controls for the disparate sizes of the samples and rejects the

hypothesis of equality of the percentages at the .998 level.

This result is not a consequence of our decision rule for an anticompetitive merger or the choice of event

window, since similar conclusions obtain in all cases. We compute the simple correlations between the

observed price changes from the retrospectives literature and the abnormal return (AR) for each of the five

event windows, and then the same correlations for the adjusted abnormal return (AAR). As shown in Table

3, there is no statistically significant relationship between either abnormal returns or adjusted abnormal returns

as measures of expected profitability and ex-post measure of price change at the 5% level of statistical

     

81

significance.

This categorization is based on abnormal returns at the time that each merger was announced or

otherwise became known. A supplementary test examines stock price changes associated with the filing of

an antitrust complaint. In particular, for an anticompetitive merger, in addition to the effect at the time of the

merger announcement, the filing of a complaint by the antitrust agency should result in negative abnormal

returns to rivals of the merging parties, as their profit prospects, as well as of the merging firms, are

diminished. We further test for the accuracy of event study predictions by examining all 16 cases18 in which

either the FTC or DOJ filed an antitrust complaint, computing the average ARs to rivals of the merging firms

at the time a complaint was issued.

As before, an event study is recorded as correctly identifying an anticompetitive merger when rivals’

stock prices exhibit significant positive abnormal returns for at least one of five event windows at the time

of merger announcement, but now also when followed by significant negative abnormal returns at the time

of the complaint. By this rule, none of these 16 challenged mergers would have been correctly identified as

anticompetitive using the event study approach. Some of these mergers show negative abnormal returns at

the time of merger announcement, while others exhibit positive abnormal returns at complaint, but in

combination this rule would have resulted in the failure to correctly identify all the anticompetitive mergers

in this sample.

We test for the robustness of these results in three ways. First, we seek to ensure that these results are

not skewed by cases in which the retrospectives find price increases but only of small magnitude. While

anticompetitive in principle, these do not imply substantial market power. We therefore define as

anticompetitive only those mergers found to result in price increases of at least five percent. There are 15

such cases in these data, of which event studies correctly detect only five. This 33 percent success rate is

                                                            18 We note that this sample is defined by cases with complaints, since regardless of whether the complaint was

meritorious or not, profits of merging and rival firms are diminished in the case of an anticompetitive merger. In fact,

11 of these 16 mergers eliciting complaints were found to be anticompetitive in retrospective studies.

     

82

somewhat greater than in the overall population of mergers, but still reflects a two-thirds error rate. Of 25

mergers that are now not determined to be anticompetitive (including those with price increases up to five

percent), event studies correctly identify 21. As before, this result is due to the fact that event studies

systematically under-predict the incidence of anticompetitive outcomes.

As a second robustness check, we re-examine mergers found by retrospective studies to result in price

increases but classify as anticompetitive only those mergers whose price increases are reported to be

statistically significant. That determination proceeded as follows: Where the retrospective reports or

emphasizes one result, we record the statistical significance of that estimate. In the more typical case where

multiple estimates of varying significance are reported, we rely on the author’s final conclusion with respect

to the overall merger and record its statistical significance. Where no such guidance is offered, we

conservatively record the price change as not significant if one or more of the small number of key price

change estimates are insignificant. This process results in classifying as anticompetitive only those mergers

that clearly pass a significance test. By this criterion, 25 of the mergers result in statistically significant price

increases. Among these, only 6 were correctly identified as anticompetitive by the corresponding event

study. And of the 15 mergers that were not anticompetitive by this criterion, event studies would nonetheless

have found three of them anticompetitive. The success rates of 24 percent and 80 percent are once again in

line with those found before.

Our third robustness test examines the possibility that the high error rates exhibited by event studies on

anticompetitive mergers are due to diversification of rivals that obscures the effect on their stock prices. Of

course, for policy purposes an error due to diversification is still an error, but it is nonetheless useful to

distinguish cases where diversification is likely the cause in order to arrive at a more informed judgment

about event studies. We test this possibility in two ways. First, we examine the list of rival firms for each

merger and calculate from the sources previously identified (and in some cases other sources) whether or

not at least thirty percent of their revenues were in the market exposed to the effects of the merger. Based

on this criterion, we reclassify eight of the original 40 cases where a high degree of diversification might

     

83

prevent event studies from detecting significant movements in those rivals’ stock prices. Of the remaining

32 cases for which sufficiently undiversified rivals exist, a total of 20 mergers were found in the corresponding

retrospective study to be anticompetitive. Of these, the corresponding stock market event study correctly

detected exactly five cases. This 25 percent success rate is little different from those previously reported.

There is no evidence that diversification is responsible for the low success rate of event studies.

Our second test of diversification examines the effect of the merger on the stock price of the single

largest of the merging firms’ undiversified rivals. This procedure holds aside some smaller rivals which,

though perhaps not overly diversified, might have stock price fluctuations more subject to other influences.

This “largest competitor” analysis once again results in exactly five cases out of 20 anticompetitive

outcomes that were correctly identified by the corresponding stock market event study. There is simply no

evidence that diversification is responsible for the low success rate of event study.

A summary of all these results is given Table 4. Collectively, they make clear that the conclusions are

not fundamentally altered by limiting attention to large or significant results, or to those involving

undiversified or significantly affected rivals. The evidence simply suggests that stock market event studies

cannot and do not accurately identify anticompetitive mergers.

4.2. Market Structure and Price Outcomes

As discussed at the outset, the conventional approach to merger analysis relies--to a degree that varies

from the historically extreme to the present-day more nuanced--on the structural characteristics of the

merging parties and their market. This section tests the accuracy of this market structure-based method of

prediction and compares its success rate in determining the likely competitive effects of mergers with that

of event studies. We generate market structure-based predictions in two ways. The first method applies the

market structure standards of the Merger Guidelines to the facts of each case, using the implied likelihood of

a challenge as an indication of the agency’s prediction of the actual effect of a merger. Since the Guidelines

are often viewed as expressing too stringent a standard, we conduct a second test in which the structure-

based prediction derives from the relationship between market structure characteristics and the actual

     

84

competitive effects of these mergers as determined from an empirical analysis of the present data. As we

shall see, while these results exhibit overall consistency.

Our first method requires making a determination about the likelihood of an agency challenge based

on a comparison of the structural characteristics of the merging parties and their markets with the standards

stated in the Horizontal Merger Guidelines. The structural characteristics of each merger are calculated

from the sources of market structure data described above, which permit quantification of market shares

and concentration for merging firms and their markets. These data are then assessed against the relevant

Merger Guidelines standards for the likelihood of an agency challenge to a merger. For example, the current

Guidelines state that mergers in moderately concentrated industries, with HHI between 1000 and 1800,

“potentially raise significant competitive concerns and often warrant scrutiny” if they raise HHI by more

than 100 points. In highly concentrated industries with HHI in excess of 1800, that same language applies

to mergers raising HHI by at least 50 points. Mergers raising HHI by more than 100 points in such highly

concentrated markets, the Guidelines state, are “presumed … likely to enhance market power or facilitate

its exercise.” All other mergers are presumptively benign.

We categorize all mergers according to whether they are presumptively benign or not by the standards

of the Merger Guidelines that were operative at the time. By this criterion, of the 31 mergers found by

retrospective studies to result in higher prices, 24 of them were correctly identified as anticompetitive by

the Merger Guidelines, for a 77 percent success rate. Of the nine competitively benign mergers in the data

base, however, only four are correctly designated as presumptively harmless by Merger Guidelines criteria.

This 44 percent success rate is only about one-half that for truly anticompetitive mergers.

Before commenting on these results, we proceed to our second market structure-based method of

predicting agency action. This method relies on the actual observed relationship between the outcomes of

these mergers as determined in the retrospective studies, and a wider array of potentially important structural

explanators than those defining the presumptions of the Merger Guidelines. In this fashion we attempt to

more fully replicate the decision process utilized by the antitrust agencies, while remaining within the market

     

85

structure paradigm. The market structure characteristics now include the HHI measure of market

concentration, the change in HHI due to the merger (DELTA), a dummy variable for the existence of entry

barriers (BENTRY), and the number of significant competitors (NCOMP, and its square NCOMP2). High

collinearity among these variables limits our ability to disentangle their separate effects, but that is less

important than their overall explanatory power. Table 5 reports both their individual and their collective

explanatory power with respect to the actual observed price changes for these mergers. The first four columns

report the results for each of the variables individually while column (e) reports their collective effect.

As is evident, each of the individual variables is significantly related to the adjusted price change due to

the merger—albeit at 7.7 percent in a one-tail test of BENTRY. It is interesting to note that the best single

explanatory factor is DELTA, the increase in HHI caused by the merger. More importantly, including all

these variables simultaneously improves the overall explanatory power of the model. R-squared is now .53

and the relevant F-test confirms the statistical significance of the entire set of market structure characteristics

(F (5, 34) = 7.78, significant at 0.01 percent). Thus while this specification obscures the effects of individual

variables, it maximizes the predictive power of the full array of market structure variables.

We use this empirical relationship to generate predictions of the price effects of each merger and interpret

those as the expectation of the agencies with respect to postmerger prices. We then compare those model-

based price predictions to the actual outcomes as determined by the corresponding retrospective study. That

comparison illustrates the accuracy and reliability of this more comprehensive set of structural criteria as the

basis for predicting the actual outcomes of the mergers.19 Of the 31 mergers that were found actually to result

in price increases, this standard correctly predicts that outcome in 27 cases, for an 87 percent success rate.

This represents only the slightest gain in predictive power relative to the Merger Guidelines standard, but

both would seem to offer strong support for a structural method of prediction.

                                                            19 Since the coefficients are estimated from the observed effects, this procedure essentially answers the following

question: assuming that the agencies rely for their determinations on all structural characteristics, not just those in the

Merger Guidelines, how frequently does that implied model correctly predict the price outcome of each merger?

     

86

As before, examining the 9 mergers that retrospectives found not to be anticompetitive is also

instructive. Of these 9, this full market structure model correctly predicts only four of them to be competitively

harmless. This 44 percent success rate in identifying truly benign mergers is identical to that based on the

narrower criteria in the Merger Guidelines. These results from use of market structure criteria are summarized in

Table 6. The broad correspondence of the two methods—one based on the Merger Guidelines, the other on a more

comprehensive set of market characteristics—is apparent. Also apparent is the fact that market structure

characteristics systematically over-predict anticompetitive outcomes. Such over-prediction results in a high

success rate for mergers that in fact are anticompetitive, but a significantly lower success rate for those that

are not. We can test for the equality of success rates for market structure-based predictions in the two samples

by again employing the means proportion test. For the present sample sizes and percentages, that test rejects

equality at any significance level.

Thus, structure-based predictions would appear to err in precisely the opposite way than event studies

(which predict the same high rate of benign mergers for those that are anticompetitive as for those that are

in fact truly benign). By itself, that would indeed seem to be evidence of error, but that conclusion may

overlook an important benefit of overly inclusive structural criteria. The market structure criteria of the

Merger Guidelines represent only a preliminary screen, one that is easily understood and cheaply

implemented. But it is only a first step in an analytical process by which other information and other

analytical approaches are employed to arrive at a policy determination by the antitrust agencies. Given that,

the optimal initial screen should in fact be overly inclusive so as to be certain to identify all possibly

anticompetitive mergers at the initial stage, after which a full analysis is undertaken. By that standard, the

appropriate test is whether the market structure-based rule captures all or nearly all mergers that prove to be

anticompetitive. Indeed, that is the case here. These criteria serve to identify on the order of 90 percent of

all mergers that prove to be anticompetitive. Thus, the apparent erroneous over-inclusiveness may not be

an error at all.

4.3. Agency Decisions and Competitive Effects

     

87

This last observation focuses attention on the incremental contribution of the antitrust agencies’

investigative processes to a correct determination of the likely anticompetitive effects of these mergers. The

agencies, of course, do not simply mechanically apply market structure criteria to come to their decisions as

to whether any particular merger is likely anticompetitive or not. Rather, for those mergers not screened out

by market structure characteristics, further investigative work proceeds. This work evaluates company

documents, pricing practices, substitution patterns, and other information, and then conducts more

sophisticated analysis of the likely effects of the merger.20

The question naturally arises, then, how often the agency decision process in fact comes to correct

determinations with respect to these mergers, and whether that success rate is greater or less than that from

reliance purely on market structure criteria or, perhaps, other methodologies. Our final comparison,

therefore, is between the actual outcomes of these mergers and the decisions of the antitrust agencies whether

or not each is anticompetitive.

One potentially confounding factor is that some observed outcomes are in fact conditional on agency

challenges. A gencies are more likely to challenge mergers expected to have anti-competitive effects, and

successful resolutions of those concerns should ideally reduce or eliminate the anticompetitive outcome.

Thus, mergers for which no price increase is observed and for which a challenge occurred may not be an

unnecessary challenge to a competitively benign merger, but rather the accurate identification and successful

resolution of a merger with likely anticompetitive effect. Recording this as an erroneous challenge will

understate the accuracy of agency determinations. On the other hand, there is no ambiguity about mergers

that do result in price increases: challenges should have been made to such mergers, and if they were, the

attempted resolutions obviously failed. We conclude that associating price changes and challenges may

understate agency accuracy where price did not increase, but this confounding does not affect cases where

                                                            20 As previously noted, the 2010 Horizontal Merger Guidelines explicitly state that the analytical process is complex

and merger-specific.

     

88

prices in fact rose. As already noted, this outcome dominates the data used in this study, hence minimizing

any ambiguity.

We tabulate agency actions for mergers that resulted in the alternative price outcomes. As shown in

Table 7, among the 31 mergers with actual anticompetitive outcomes, the agencies in fact challenged 13

mergers, for a 41.9 percent “success” rate.21 While perhaps not overwhelmingly high, this rate of success is

twice that recorded for event studies. Crucially, however, with respect to the 9 mergers that in fact proved

not to be anticompetitive, agencies erred by treating only 3 of them (33 percent) as anticompetitive. This is

considerably less than the 56 percent error rate of the market structure-based tests, and indeed is exactly the

same low rate as for event studies (which, of course, achieved this low rate by systematically under-

predicting anticompetitive outcomes).

As before, we can test for the equality of the success rates for agency decisions with respect to the two

samples with the means proportion test. That test cannot reject the proposition of equality at the

conventional .95 level, but only at .914. This still suggests a difference, but not the overwhelming difference

in either of the other two tests. We therefore conclude that agencies’ internal decision processes significantly

improve on the mechanical application of market structure criteria. The wide-ranging, eclectic, and merger-

specific analytical process engaged in by the agencies in fact adds considerable value--that is, predictive

accuracy--to simple decision rules. Relative to purely market structure-based decisions, the agency decision

process substantially increases the success rate for correctly identifying benign mergers, although it displays

a lower rate of successful identification of truly anticompetitive mergers. And relative to event studies,

agencies’ decisions double the success rate of identifying anticompetitive mergers without increasing the

error rate for those that are not. But as noted, neither event studies nor market structure criteria have much

real discriminating power since their “success rates” derive essentially from their tendency to over-predict

                                                            21 Recall that this may be an underestimate to the extent that some outcomes recorded as showing no effect may be

due to successful resolution of a challenged merger.

     

89

certain merger outcomes.

5. Conclusion

Merger control has always depended on good predictive models. In practice this has largely meant

reliance on market structure criteria, although the antitrust process has never—or at least, not for a very long

time—relied simply on the numerical standards in the Merger Guidelines. Rather, mergers that seem

problematic by those standards have been subject to further evaluation by the agencies. Some observers

have advocated even less reliance on structural standards—or for that matter, on agency evaluation—and

instead the use of stock market event studies as the key predictive tool. This study is the first to compile

more than fragmentary evidence comparing the reliability of event studies and market structure criteria in

correctly identifying anticompetitive mergers. The determination of actual competitive outcomes derives

from a unique data base of 40 mergers for which well-designed retrospective studies have been reported. To

those findings are added new stock market event studies on the abnormal returns to rival firms at the time of

the merger announcement, various market structure characteristics generally believed to determine

outcomes, and the record of actual agency determinations whether or not to challenge these mergers.

The evidence is quite clear that event studies systematically underestimate the actual frequency of

anticompetitive effects from these mergers. The evidence is equally clear that market structure criteria by

themselves over-predict the frequency of anticompetitive effects from these same mergers. We have

observed, however, that over-inclusiveness may be appropriate for the initial step of agency’s evaluation

processes for determining whether mergers are in fact anticompetitive. Consistent with that, we find that

agency decisions are in fact more accurate than either event studies or purely structural criteria, balancing

errors in each direction in a fashion that demonstrates the added value of the merger-specific process of

investigation conducted by the agencies.

Overall, these results cast substantial doubt on the efficacy of reliance on any single technique as a

predictive tool for merger outcomes. In particular, enthusiasm for stock market event studies appears ill

     

90

founded. Similarly, any argument for a purely structural approach appears equally (but in the opposite

direction) incorrect. But strikingly, agency decision processes, while imperfect, significantly improve on

either simple rule for the purpose of making determinations about the likely competitive outcomes of these

mergers.

 

     

91

Studies Cited

Armantier, O. and Richard, O. 2006.Evidence on Pricing from the Continental Airlines and Northwest Airlines Code-

Share Agreement. Advances in Airline Economics: Competition Policy and Antitrust.

Ashenfelter, A., Hosken, D., and Weinberg, M.C. 2011. The Price Effects of a Large Merger of Manufacturers: A

Case Study of Maytag-Whirlpool. NBER Working Paper.

Ashenfelter, O. and D. Hosken. 2010. The Effect of Mergers on Consumer Prices: Evidence from Five Selected Case

Studies. Journal of Law and Economics, 53, 417-466.

Bamberger, G.E., Carlton, D.W., and Neumann, L.R. 2004.An Empirical Investigation of the Competitive Effects of

Domestic Airline Alliances. Journal of Law and Economics, 47(1), 195-222.

Barton, D.M., and Sherman, R. 1984. The Price and Profit Effects of Horizontal Mergers: A Case Study. Journal of

Industrial Economics, 33(2), 165-177.

Bilotkach, V. 2011.Multimarket Contact and Intensity of Competition: Evidence from an Airline Merger. Review of

Industrial Organization, 38(1), 95-115.

Borenstein, S. 1990. Airline Mergers, Airport Dominance, and Market Power. American Economic Review, 80(2),

400-404.

Calomiris, C.W. and Pornrojnangkool, T. 2005. Monopoly-Creating Bank Consolidation? The Merger of Fleet and

BankBoston. NBER Working Paper.

Gayle, P.G. 2008. An Empirical Analysis of the Competitive Effects of the Delta/Continental/Northwest Code‐Share

Alliance. Journal of Law and Economics, 51(5), 743-766.

Haas-Wilson, D., and Garmon, C. 2009.Two Hospital Mergers on Chicago’s North Shore: A Retrospective Study.

International Journal of the Economics of Business, 18(1), 17-32.

Hosken, D., Silvia, L., and Taylor, C. 2011. Does Concentration Matter? Measurement of Petroleum Merger Price

Effects. American Economic Review: Papers and Proceedings, 101(3), 45-50.

Karikari, J., Brown, S.M., and Nadji, M. 2002. The Union Pacific/Southern Pacific Railroads Merger: Effect of

Trackage Rights on Rates. Journal of Regulatory Economics, 22(3), 271-285.

Kwoka, J. and Shumilkina, E.2010. The Price Effect of Eliminating Potential Competition: Evidence from an Airline

Merger. Journal of Industrial Economics, 58(4), 767-793.

McCabe, M.J. 2004.Law Serials Pricing and Mergers: A Portfolio Approach. B.E. Journal of Economic Analysis &

Policy, 3(1) (Contributions), Article 11.

     

92

McCabe, M.J. 2002a. Journal Pricing and Mergers: A Portfolio Approach. American EconomicReview,92(1), 259-269.

Morrison, S.A. 1996. Airline Mergers: A Longer View. Journal of Transport Economics and Policy, 30(3), 237-250.

Peters, C. 2006.Evaluating the Performance of Merger Simulation: Evidence from the U.S. Airline Industry. Journal

of Law and Economics, 49(2), 627-649.

Romano, P.S., and Balan, D.J. 2011. A Retrospective Analysis of the Clinical Quality Effects of the Acquisition of

Highland Park Hospital by Evanston Northwestern Healthcare. International Journal of the Economics of

Business, 18(1), 45-64.

Schumann, L., Rogers, R.P., and Reitzes, J.D. 1992. Case Studies of the Price Effects of Horizontal Mergers. FTC

Working Paper.

Simpson, J., and Taylor, C. 2008. Do Gasoline Mergers Affect Consumer Prices? The Marathon Ashland Petroleum

and Ultramar Diamond Shamrock Transaction. Journal of Law and Economics, 51(1), 135-152.

Taylor, C. T. and Hosken, D.S. 2007. The Economic Effects of the Marathon-Ashland Joint Venture: The Importance

of Industry Supply Shocks and Vertical Market Structure. Journal of Industrial Economics, 55(3), 419-451.

Silvia, L., and Taylor, C. T. 2010.Petroleum Mergers and Competition in the Northeast United States. FTC Working

Paper.

Tenn, S.2011. The Price Effects of Hospital Mergers: A Case Study of the Sutter-Summit Transaction. International

Journal of the Economics of Business, 18(1), 65-82.

Thompson, A. 2011.The Effect of Hospital Mergers on Inpatient Prices: A Case Study of the New Hanover-Cape Fear

Transaction. International Journal of the Economics of Business, 18(1), 91-101.

Vita, M.G., and Sacher, S. 2001. The Competitive Effects of Not-for-Profit Hospital Mergers: A Case Study. Journal

of Industrial Economics, 49(1), 63-84.

U.S. GAO. 2004. Energy Markets: Effects of Mergers and Market Concentration in the U.S. Petroleum Industry.

Werden, G.J., Joskow, A.S., and Johnson, R.L. 1991. The Effects of Mergers on Price and Output: Two Case Studies

from the Airline Industry. Managerial and Decision Economics, 12(5), 341-352.

Winston, C., Maheshri, V., and Dennis, S. 2011.Long-Run Effects of Mergers: The Case of US Western Railroads.

Journal of Law and Economics, 54, 275-304.

     

93

References

Beverley, L. 2007. Stock Market Event Studies and Competition Commission Inquiries. Competition Commission:

http://www.competitioncommission.org.uk/ourrole/analysis/ topics.htm.

Cichello, M., and Lamdin, D. J. 2006.Event Studies and the Analysis of Antitrust. International Journal of the

Economics of Business, 13(2), 229–245.

Cox, A.J., and Portes, J. 1998. Mergers in Regulated Industries: The Uses and Abuses of Event Studies. Journal of

Regulatory Economics, 14, 281-304.

Duso, T., Gugler, K., and Yurtoglu, B. 2006. How Effective is European Merger Control? WZB Discussion paper SP

II, 2006-12.

Duso, T., Gugler, K., and Yurtoglu, B. 2010. Is the Event Study Methodology Useful for Merger Analysis? A

Comparison of Stock Market and Accounting Data. International Review of Law and Economics, 30, 186-192.

Duso, T., Neven, D.J., and Röller, L.H. 2007. The Political Economy of European Merger Control: Evidence using

Stock Market Data. Journal of Law and Economics, 50(3), 455-489.

Eckbo, B. E. 1983. Horizontal Mergers, Collusion, and Stockholder Wealth. Journal of Financial Economics, 11, 241-

273.

Eckbo, B. E. 1989. The Role of Stock Market Studies in Formulating Antitrust Policy toward Horizontal Mergers:

Comment. Quarterly Journal of Business and Economics, 28, 22–38.

Eckbo, B. E., and Wier, P. 1985. Antimerger Policy under the Hart-Scott-Rodino Act: A Re-examination of the Market

Power Hypothesis. Journal of Law and Economics, 28, 119–149.

Eckbo, B.E. 1985. Mergers and the Market Concentration Doctrine: Evidence from the Capital Market. Journal of

Business, 58(3), 325-349.

Fama, E. F., Fisher, L., Jensen, M.C., and Roll, R. 1969. The Adjustment of Stock Prices to New Information.

International Economic Review, 10, 1–21.

Federal Reserve Bank of Boston. 1999. Banking Structure in New England 1996-1999. Research Report 75.

Federal Reserve Bank of Boston. 2001. Banking Structure in New England 1999-2001. Research Report 76.

Federal Trade Commission. 2004. The Petroleum Industry: Mergers, Structural Change, and Antitrust Enforcement.

Bureau of Economics.

Federal Trade Commission. 2011. Gasoline Price Changes and the Petroleum Industry: An Update. Bureau of

Economics.

     

94

Healy, P., Palepu, K., and Ruback, R. 1992. Does Corporate Performance Improve after Mergers? Journal of Financial

Economics, 31, 135–175.

Hosken, D. and Simpson, J. D. 2001. Have Supermarket Mergers Raised Prices? An Event Study Analysis.

International Journal of the Economics of Business, 8, 329–342.

Kaplan, S., and Weisbach, M. 1992. The Success of Acquisitions: Evidence from Divestitures. Journal of Finance,

47, 107-138.

Kim, E.H., and Singal, V. 1993. Mergers and Market Power: Evidence from the Airline Industry. American Economic

Review 83(3), 549-569.

Knapp, W. 1990. Event Analysis of Air Carrier Mergers and Acquisitions. Review of Economics and Statistics, 72(4),

703-707.

Kwoka, J. 2013. Does Merger Control Work? A Retrospective on U.S. Enforcement Actions and Merger Outcomes,

Antitrust Law Journal, 78, 629-634.

MacKinlay, A. C. 1997. Event Studies in Economics and Finance. Journal of Economic Literature, 35, 13–39.

Maldutis, J. 1987. Airline Competition at the 50 Largest U.S. Airports since Deregulation. Salomon Brothers.

McAfee, R.P., and Williams, M.A. 1988. Can Event Studies Detect Anticompetitive Mergers? Economics Letters, 28,

199–203.

McCabe, M.J. 2000. Academic Journal Pricing and Market Power: A Portfolio Approach. Working Paper, University

of Michigan.

McCabe, M.J. 2002b.A Portfolio Approach to Print Legal Serials Pricing. Working Paper, University of Michigan.

McGuckin, R.H., Warren-Boulton, F.R., and Waldstein, P. 1992. The Use of Stock Market Returns in Antitrust

Analysis of Mergers. Review of Industrial Organization, 7, 1-11.

Mullin, G.L., Mullin, J.P., and Mullin, W.P. 1995. The Competitive Effects of Mergers: Stock Market Evidence from

the U.S. Steel Dissolution Suit. Rand Journal of Economics, 26, 314–330.

Pautler, P.A. 2003. Evidence on Mergers and Acquisitions. Antitrust Bulletin, Spring, 119-221.

Peters, C. 2006.Evaluating the Performance of Merger Simulation: Evidence from the U.S. Airline Industry. Journal

of Law and Economics, 49(2), 627-649.

Prager, R. A. 1992. The Effects of Horizontal Mergers on Competition: the Case of the Northern Securities Company.

Rand Journal of Economics, 23, 123–133.

     

95

Schumann, L. 1993. Patterns of Abnormal Returns and the Competitive Effects of Horizontal Mergers. Review of

Industrial Organization, 8, 679–696.

Singal, V. 1996. Airline Mergers and Competition: An Integration of Stock and Product Price Effects. Journal of

Business, 69, 233–68.

Sirower, M., and O’Byrne, S. 1998. The Measurement of Post-acquisition Performance: Towards a Value-based

Benchmarking Methodology. Journal of Applied Corporate Finance, 11, 107–121.

Stillman, R. 1983. Examining Antitrust Policy towards Horizontal Mergers. Journal of Financial Economics, 11, 1–

41.

Svengalis, K.F. 2005. Legal Information Buyer's Guide & Reference Manual. Rhode Island Law Press, RI: Westerly.

Warren-Boulton, F. R. and Dalkir, S. 2001. Staples and Office Depot: An Event-probability Case Study. Review of

Industrial Organization, 19, 469–481.

Weinberg, M., and D. Hosken, Evidence on the Accuracy of Merger Simulations, Review of Economics and Statistics,

2014

Werden, G. J. and Williams, M. A. 1989a. The Role of Stock Market Studies in Formulating Antitrust Policy toward

Horizontal Mergers. Quarterly Journal of Business and Economics, 28, 3–21.

Werden, G. J. and Williams, M. A. 1989b. The Role of Stock Market Studies in Formulating Antitrust Policy toward

Horizontal Mergers: Reply. Quarterly Journal of Business and Economics, 28, 39–42.

Whinston, M.D. 2007. Antitrust Policy toward Horizontal Mergers, in M. Armstrong and R. Poster (eds.) Handbook

of Industrial Organization, Volume III, Amsterdam: North-Holland.

Whinston, M.D., and Collins, S.C. 1992. Entry and Competitive Structure in Deregulated Airline Markets: An Event

Study Analysis of People Express. Rand Journal of Economics, 23(4), 445-462.

 

     

96

Table 1. Summary of the Cases

Variable Definition Mean Std. Dev. Min Max

PREHHI Pre-merger HHI 2179 1251 847 5243

POSTHHI Post-merger HHI 2972 2208 881 8954

DELTA Change of HHI 793 1079 15 4234

BENTRY Dummy=1 if entry is difficult 0.60 0.50 0 1

NCOMP No. of direct competitors 6.50 3.17 1 13

PRICE_CHANGE Observed price change in percentage 5.65 8.99 -7.2 29.4

     

97

Table 2. Counts of Successful Detection: Event Studies

Determine to be: Of Which Event Studies

Correctly Predict % Correct

Anti-competitive 31 7 22.6

Not Anti-competitive 9 7 77.8

Table 3. Correlation Coefficients for Pairwise Test

AR1 AR2 AR3 AR4 AR5

(-20 to 10) (-10 to 5) (-3 to 3) (-1 to 1) (0)

Observed Price Change -0.191 -0.040 0.177 0.012 0.031

AAR1 AAR2 AAR3 AAR4 AAR5

(-20 to 10) (-10 to 5) (-3 to 3) (-1 to 1) (0)

Observed Price Change -0.296 -0.212 -0.060 -0.108 0.010

Table 4. Results of Robust Tests

Actually

Anticompetitive

Number Correctly

Identified by Event Study

Percent

Correct

Basic Test 31 7 23

Challenges 16 0 0

5% Price Rise 15 5 33

Statistically Significant 25 6 24

Significantly Affected Competitors 20 5 25

Largest Competitor 20 5 25

     

98

Table 5. Tests of Structural Standard

(1) (2) (3) (4) (5)

POSTHHI 0.720*** 0.073

(4.842) (0.177)

DELTA 1.524*** 1.182

(5.125) (1.441)

BENTRY 11.881 4.103

(1.451) (0.623)

NCOMP -18.288*** -14.517***

(-3.009) (-2.941)

NCOMP2 1.139*** 0.915***

(2.823) (2.808)

CONSTANT -7.393 1.922 6.873 73.568*** 46.708**

R-Squared 0.382 0.409 0.052 0.204 0.534

Note: t-statistics in parenthesis, and *p<0.1, **p<0.05, ***p<0.01. POSTHHI and DELTA are adjusted by dividing 100.

     

99

Table 6. Counts of Successful Detection: Market Structure

Test 1: Using Market Structure Variables in Table 5

Determine to be: Actual Of Which Structure Correctly Predicts % Correct

Anti-competitive 31 27 87.1

Not Anti-competitive 9 4 44.4

Test 2: Using Pure HHI Standards in Horizontal Merger Guidelines

Determine to be: Actual Of Which HMG

Correctly Predicts % Correct

Anti-competitive 31 24 77.4

Not Anti-competitive 9 4 44.4

Table 7. Counts of Successful Detection: Agency Challenges

Determine to be: Actual Of Which Agencies

Correctly Predict % Correct

Anti-competitive 31 13 41.9

Not Anti-competitive 9 6 66.7

 

     

100 Appendix: Average Daily Abnormal Return to the Portfolio of Rival Firms

No Type Bidder Target Industry Agency Action Remedy Data Sources of

Rivals and Market Structure Datea

Days Relative to Proposal Announcement (Day 0)

(-20, 10) (-10, 5) (-3, 3) (-1, 1) (0)

1 Merger Northwest Republic Airlines DOJ Opposed Overruled DB1A (1986Q1); 01/24/86 0.0049 0.0130 0.0345** 0.0525** 0.1202***

Maldutis (1987) (0.740) (1.424) (2.561) (2.599) (3.491)

2 Merger Trans World Ozark Airlines Airlines DOJ Opposed Overruled DB1A (1986Q1); 02/28/86 -0.0039 -0.0025 0.0042 0.0063 0.0006

Maldutis (1987) (-1.337) (-0.634) (0.716) (0.714) (0.038)

3 Merger USAir Piedmont Airlines DOJ Cleared None DB1A (1987Q1) 02/18/87 -0.0066 -0.0085 0.0014 -0.0155 -0.0192

(-0.940) (-0.906) (0.102) (-0.763) (-0.548)

4 Merger Continental People Express Airlines DOJ Cleared None DB1A (1986Q3) 07/03/86 -0.0025 0.0016 -0.0023 -0.0039 0.0164

(-0.849) (0.413) (-0.401) (-0.452) (1.130)

5 Merger Delta Western Airlines DOJ Cleared None DB1A (1986Q1) 09/11/86 0.0051*** 0.0040* 0.0021 0.0046 -0.0027

(2.911) (1.684) (0.597) (0.822) (-0.276)

6 Code-Share

Continental America West Airlines DOJ Cleared None DB1B (1994Q1) 02/22/94 0.0006 0.0005 0.0008 0.0025 -0.0004

(0.212) (0.133) (0.150) (0.303) (-0.030)

7 Code-Share

Northwest Alaska Airlines Airlines DOJ Cleared None DB1B (1995Q3) 08/22/95 -0.0014 -0.0036 -0.0002 0.0005 0.0102

(-0.490) (-0.962) (-0.028) (0.061) (0.689)

8 Code-Share

Northwest Continental Airlines DOJ Opposed Overruled DB1B (1997Q4); 12/16/97 0.0028 -0.0015 -0.0023 -0.0050 -0.0197

DOJ (98-74611) (0.935) (-0.391) (-0.399) (-0.566) (-1.296)

9 Code-Share

Delta Continental Airlines DOJ Cleared None DB1B (2002Q3) 08/08/02 -0.0065 -0.0163*** -0.0149* 0.0022 0.0222

+Northwest (-1.360) (-2.715) (-1.710) (0.167) (0.994)

10 Merger America West USAir Airlines DOJ Cleared None DB1B (2004Q4) 04/20/05 0.0028 -0.0027 -0.0057 -0.0090 0.0085

(0.707) (-0.501) (-0.722) (-0.758) (0.416)

11 Merger P&G Tambrands Feminine DOJ Cleared None Ashenfelter and Hosken (2010) 04/09/97 -0.0008 -0.0036 -0.0054 -0.0095 -0.0485***

Hygiene (-0.397) (-1.371) (-1.373) (-1.595) (-5.109)

12 Merger Guinness Grand Spirits FTC Consent Structure Ashenfelter and Hosken (2010) 05/13/97 -0.0016 -0.0030 -0.0041 -0.0043 0.0056

Metropolitan (-0.920) (-1.306) (-1.228) (-0.841) (0.634)

13 Merger Pennzoil Quaker State Conventional FTC Cleared None Ashenfelter and Hosken (2010) 04/16/98 0.0016 0.0003 0.0011 0.0055 0.0200**

Motor Oil (0.957) (0.128) (0.328) (1.092) (2.329)

14 Merger General Mills Ralcorp RTE Cereal FTC Consent Condition Ashenfelter and Hosken (2010) 08/14/96 -0.0006 0.0015 0.0030 0.0030 0.0113

(-0.381) (0.676) (0.892) (0.580) (1.287)

15 Merger Aurora Foods Kraft Breakfast N/A N/A N/A Ashenfelter and Hosken (2010) 05/02/97 0.0050 -0.0007 -0.0032 -0.0064 -0.0046

(Log Cabin) Syrup (1.255) (-0.305) (-0.870) (-1.149) (-0.473)

16 Merger Whirlpool Maytag Home FTC Cleared None Ashenfelter et al (2011) 07/19/05 -0.0029** -0.0020 -0.0016 -0.0049 -0.0038

Appliances (-2.496) (-1.278) (-0.702) (-1.423) (-0.640)

17 Merger Wolters Lippincott Biomedical DOJ Cleared None McCabe (2000; 2002a) 05/22/90 0.0041 0.0015 0.0014 0.0057 0.0052

Kluwer Journal (1.467) (0.411) (0.253) (0.680) (0.356)

18 Merger Elsevier Pergamon Biomedical DOJ Cleared None McCabe (2000; 2002a) 03/29/91 0.0054 0.0063 0.0065 0.0233** 0.0315*

Journal (1.634) (1.418) (0.983) (2.344) (1.832)

     

101 Appendix: Average Daily Abnormal Return to the Portfolio of Rival Firms (continued) 19 Merger Thomson Shepard's Law Journal N/A N/A N/A McCabe (2002b; 2004) 11/30/95 0.0002 0.0035 0.0094* 0.0076 -0.0042

Svengalis (2005) (0.071) (0.743) (1.887) (1.289) (-0.512)

20 Merger Thomson West Publishing Law Journal DOJ Consent Structure McCabe (2002b; 2004) 10/23/95 -0.0008 -0.0013 -0.0008 -0.0031 -0.0028

Conduct Svengalis (2005) (-0.368) (-0.496) (-0.225) (-0.542) (-0.288)

21 Merger Reed Elsevier West Publishing Law Journal DOJ Cleared None McCabe (2002b; 2004) 09/13/96 -0.0015 -0.0014 0.0029 -0.0034 -0.0091

Svengalis (2005) (-0.641) (-0.443) (0.720) (-0.583) (-0.899)

22 Merger Wolters CCH Law Journal N/A N/A N/A McCabe (2002b; 2004) 11/28/95 0.0011 0.0052** 0.0053 0.0098** 0.0111

Kluwer Svengalis (2005) (0.674) (2.400) (1.623) (1.982) (1.300)

23 Merger Wolters Little Brown Law Journal N/A N/A N/A McCabe (2002b; 2004) 08/29/96 0.0016 -0.0021 0.0016 0.0019 0.0074

Kluwer Svengalis (2005) (0.513) (-0.834) (0.426) (0.335) (0.759)

24 Merger Tosco Unocal Petroleum FTC Cleared None EIA (1995); 11/19/96 -0.0006 0.0006 0.0027 -0.0034 -0.0234

Hosken et al (2011) (-0.128) (0.103) (0.329) (-0.266) (-1.091)

25 Merger UDS Total Petroleum FTC Cleared None EIA (1995) 02/25/97 0.0009 -0.0002 0.0019 -0.0099 0.0029

(0.234) (-0.031) (0.233) (-0.827) (0.143)

26 Joint Marathon Ashland Petroleum FTC Cleared None Taylor and Hosken (2007) 04/01/97 0.0001 0.0009 -0.0038 -0.0003 -0.0046

Venture (0.075) (0.398) (-1.143) (-0.062) (-0.562)

27 Joint Shell Texaco I Petroleum FTC Consent Structure EIA (1997); FTC (2004) 03/19/97 0.0000 -0.0026 -0.0033 -0.0016 -0.0356

Venture (0.002) (-0.243) (-0.205) (-0.065) (-0.842)

28 Joint Shell Texaco II Petroleum FTC Cleared None EIA (1997); FTC (2004) 03/19/97 0.0005 0.0022 0.0056** 0.0029 0.0024

Venture +Saudi Arabia (0.394) (1.358) (2.423) (0.814) (0.395)

29 Merger BP Amoco Petroleum FTC Consent Structure EIA (1998); FTC (2004) 08/12/98 0.0008 0.0024 0.0067 0.0074 -0.0141

(0.345) (0.756) (1.498) (1.092) (-1.216)

30 Merger Exxon Mobil Petroleum FTC Consent Structure EIA (1998); FTC (2004) 11/26/98 -0.0014 -0.0025 -0.0028 -0.0014 0.0138

(-0.578) (-0.745) (-0.559) (-0.181) (1.056)

31 Merger MAP UDS Petroleum FTC Cleared None Simpson and Taylor (2008) 05/25/99 0.0007 -0.0021 0.0027 0.0006 -0.0024

(0.231) (-0.517) (0.443) (0.063) (-0.148)

32 Merger Sunoco El Paso's Petroleum FTC Cleared None EIA(2003); FTC (2004) 12/31/03 -0.0012 -0.0003 -0.0005 0.0050 0.0070

Eagle Point Silvia and Taylor (2010) (-0.481) (-0.102) (-0.096) (0.658) (0.526)

33 Merger Valero Premcor Petroleum FTC Cleared None EIA(2003); FTC (2004; 2011) 04/25/05 -0.0009 -0.0018 -0.0035 0.0027 0.0087

Silvia and Taylor (2010) (-0.408) (-0.644) (-0.867) (0.439) (0.824)

34 Merger Burlington Santa Fe Rail DOJ Consent Structure STB (1996) 07/01/94 -0.0009 -0.0010 -0.0019 -0.0022 0.0105

Northern Firms' websites (-0.585) (-0.485) (-0.591) (-0.455) (1.269)

35 Merger Union Pacific Southern Pacific Rail DOJ Opposed Overruled STB (1996) 08/03/95 0.0019 0.0009 0.0023 0.0072** 0.0128**

Firms' websites (1.348) (0.512) (0.965) (2.034) (2.085)

36 Merger Weyerhaeuser Menasha Corrugating FTC Opposed Overruled FTC (Docket 9150) 09/12/80 -0.0035* -0.0026 -0.0025 -0.0035 -0.0031

Medium (-1.950) (-1.029) (-0.655) (-0.613) (-0.311)

37 Merger SCM Gulf & Western Titanium FTC Cleared None Schumann et al (1992) 07/16/83 -0.0012 0.0026 0.0127** 0.0118 0.0190

Dioxide (-0.372) (0.596) (2.005) (1.217) (1.140)

38 Merger Fleet BankBoston Banking DOJ Consent Structure Federal Reserve Bank of Boston 03/19/99 0.0013 0.0001 -0.0015 -0.0053 -0.0041

(1999; 2001) (0.633) (0.021) (-0.338) (-0.814) (-0.366)

     

102 Appendix: Average Daily Abnormal Return to the Portfolio of Rival Firms (continued)

39 Merger Xidex Scott Graphics Microfilm FTC Consent Structure FTC (Docket 9146); 06/01/76 -0.0013 -0.0025 -0.0095** -0.0117 -0.0162

Diazo Conduct McAfee and Williams (1988) (-0.552) (-0.792) (-1.997) (-1.626) (-1.295)

40 Merger Xidex Kalvar Microfilm FTC Consent Structure FTC (Docket 9146); 02/08/79 -0.0041** -0.0050** -0.0048 -0.0058 -0.0042

Vesicular Conduct McAfee and Williams (1988) (-2.447) (-2.149) (-1.386) (-1.106) (-0.462)

Note: t-statistics in parentheses, and * p<0.1,** p<0.05,***p<0.01. a. The date of merger proposal announcement. b. Northwest Airlines acquired 14% stake of Continental Airlines in this code-sharing agreement transaction.

 

 

     

103

Chapter 3

The Mergers Effects on Telephone Carriers' Efficiency:

A Conditional Difference-in-Difference Approach

 

     

104

1. Introduction

The last decade has witnessed the significant growth and consolidation process in the U.S.

telecommunications industry. During the antitrust merger reviewing process, the most frequently used

claim by the merging carriers is that merger could improve their efficiencies, such as technological

efficiencies and economies of scale (Goldman et al., 2003). Thus, it is natural to ask whether these mergers

did achieve such efficiencies improvements as the carriers promised. It is also of particular interest to know

that what factors drive carriers to merge when they are regulated on prices, more or less.

One classical approach to tackle this kind of questions is the research design based approach, such as

difference-in-difference (DID) analysis. These studies typically compare outcomes of merging firms

(treatment group) with those of nonmerging firms (control group). However, a standard DID method may

suffer from serious selection bias: if merging firms are not randomly selected, some observed and

unobserved characteristics can affect firms' merger decisions as well as their post-merger performances will

generate biased estimates of the impact of a merger. For example, those firms with financial difficulties are

more likely to be party to a merger and post-merger the new firms reduce costs and decrease prices.

Conditional on survival, these firms might have decreased prices and reduced costs even more absent a

merger (Dafny, 2009).

In the literature, there are at least two kinds of methods to address this selection issue. Dafny (2009),

among others, uses an instrumental variable (IV) method to investigate the merger effect in the hospital

industry. To control for the probability that a hospital involves into a merger, she uses the distance between

two hospitals as the instrumental variable to correct the bias caused by the unobserved factors. However,

such kinds of IVs are not always available for many other industries. Another kind of approaches is so

called "propensity score matching (PSM)" method, which has been used to examine merger effects on costs

in the hospital industry (Dranove and Lindrooth, 2003). While the IV method is based upon the assumption

that selection is on unobservable, the matching estimator is based upon the unconfoudedness assumption

     

105

(selection on observables). The basic idea of this approach to compare the performance of merging firms

with the performance of those similar nonmerging firms "but-for" the merger.

In this paper, to control for the endogeneity of the merger formation, I construct a matched sample

using the propensity score matching (PSM) method. I model the probability to merge using a Probit and

match the merging firms to the similar firms that did not merge based on the predicted propensity/

probability of merging. Then, a difference-in-difference (DID) estimation is adopted to identify the merger

effects on carriers' efficiency change. This paper uses a balanced panel that includes annual data of 84 large

and midsized ILECs (incumbent local exchange carriers) in the U.S. telecommunications industry. The

results show that mergers consummated during the period of 1996-2007 reduce merging carriers' total factor

productivity growth rate (Malmquist Index) relative to the non-merging carriers. This deterioration is due

to that mergers reduce merging carriers' incentive to innovate/invest on frontier technology, rather than

lower carriers' technical efficiencies. Moreover, mergers do not speed up carriers' scale efficiency progress

as the carriers promised in the antitrust review processes.

This paper is also related to Seo et al. (2010), who compare the productivity means of merging and

non-merging carriers before and after the merger. As there is only one non-merging firm in their study, it

is natural to question that the productivity deterioration is an industry trend or a merger-specific effect. This

paper complements their study in the sense that I construct the merger counterfactual using the PSM method,

which could provide more reliable estimates of the magnitudes of merger effects. This method has been

used to examine merger effects on R&D performance in pharmaceutical industry (Danzon et al., 2007;

Ornaghi, 2009), but none has been done to investigate the merger effects in telecommunications industry.

The structure of this paper is as follows. The next section provides some background on the U.S. fixed-

line markets and the sources of data. Section 3 further describes the empirical strategy and sets out the

econometric model. Section 4 presents the results and section 5 concludes.

     

106

2. Background and Data

The data used in this note represents a balanced panel that includes annual data of 84 large and

midsized ILECs (incumbent local exchange carriers) over the period of 1996-2007. It covers 12 mergers

consummated during this period, such as the notable merger between Bell Atlantic and GTE in 2000, and

acquisition of AT&T and BellSouth by SBC in 2006. The state-level ILECs' data are collected from the

FCC's Automated Reporting Management Information System (ARMIS), and the regulation information

comes from the State Telephone Regulation Reports.

Total operating expenses (OPEX, including access, customer service, plant and labor expenses), the

number of central office switches, and the number of access lines (including the special access) are selected

as the input variables. While OPEX reflects carriers' short-run variable operating expense (Thanassoulis,

2000; Kwoka and Pollit, 2010), the latter two variables capture carriers' long-run capital expenditure. Four

output variables are the number of local calls, intra-LATA toll calls, inter-LATA toll calls and the total

operating revenues.1 Table 1 summarizes the firm characteristics.

3. Empirical Strategy

The merger effects can be identified as the differences between the observed efficiencies following

the merger and what efficiencies would have been if the carrier had not been acquired. Usually, a DID

analysis could be used to construct such counterfactual. However, this approach may suffer from serious

selection biases if the merging entities are not randomly chosen. In other words, if the decision to merger

is an endogenous process on the base of carriers' specific characteristics, some of which affect the post-

merger efficiencies, it would lead to a correlation between the merger decision and the error term in the

outcome equation.

                                                            1 Following many previous efficiency studies in telecommunications industry, such as Majumdar (1997), I use

employees, access lines, and number of central office switches as the inputs and number of local calls, intra-LATA toll

calls and inter-LATA toll calls as the outputs for 26 large ILECs (as the FCC only has the employee data for those 26

carriers). The merger effects are robust to this specification.

     

107

As a result, I use a propensity score matching with DID method proposed by Heckman et al (1997)

and Blundell and Costa Dias (2000), to control for selection based on observable firm characteristics. In

detail, let {0,1}itM be an indicator of whether firm i merged at time t , and let 1it sY be the value of a carrier's

efficiency ( Y ) at time t s if it merged and 0it sY be the value of the same carrier's efficiency if it did not

merge. The expected merger effect for firm i at time t s can be defined as:

1 0 1 0[ | 1] [ | 1] [ | 1]it s it s it it s it it s itE Y Y M E Y M E Y M

where the efficiency that firm would have had if it did not merge is unobservable. To construct such

counterfactual, I identify carriers with similar observable characteristics to merge at time t that did not

merge such that equation (1) becomes | , 1 | , 0 . is a vector of pre-

merger characteristics, which can be further summarized as an indicator of the propensity to merge,

Pr | . As a result, the difference in efficiencies can be attributed to merger status of the

carrier under the conditional independent assumption.

Given the matched carriers, a DID estimator is then used to examine the merger effects on efficiency

changes2, which can be written as following:

4 42007

, , 19970 0

( )it j i t j k i t k it it it t t itj k

Log EC O N E TW O SBC Z D IVEST YEAR

Where refers to efficiency change in log for carrier i in period t , and is the constant. Either

of the two merger terms-ONE and TWO- includes five dummy variables that take a value of 1 if the carrier

i merges in period t , 1t (i.e.1 year ago), ... , 4t . In the dataset, five state-level SBC firms experience a

third merger in 2006, and a SBC variable is added to isolate this effect. It takes a value of 1 from period t

(when the third merger occurs) to 4t for such SBC firms. Also, Sprint spun off its fixed-line business to

form a new firm called Embarq in 2006, and a DIVEST term is added that taking on a value of 1 for the

                                                            2 This kind of matching estimator has been used widely in the international trade literature. For example, Greenaway

and Kneller (2008) use the matching approach to study the entry effects on firms’ productivities in the export market.

(1)

(2)

     

108

years after such divesture. Z is a vector of control variables, including four regulation dummies (taking a

value of 1 if the carrier faces earnings-sharing, rate case moratoria, price-cap regulation or deregulation).

The efficiency changes are calculated using Malmquist Index (MI) proposed by Fare et al (1994).

Malmquist Index is an index that evaluates the productivity change of a decision making unit (DMU)

between two time periods. Following Fare et al (1992, 1994), I firstly decompose the MI into two

components under the assumption of constant returns to scale (CRS): the technical efficiency change (TE)

term measures to which degree a DMU improves or worsens its efficiency, whereas the frontier shift (FS)

term reflects the change in the efficient frontiers between the two time periods:

1/ 21 1 1 1 10 0 0 0 0 0

0 10 0 0 0 0 0

( , ) ( , )

( , ) ( , )

t t t t t t

t t t t t t

D x y D x yM TE FS

D x y D x y

, where 1 1 1

0 0 0

0 0 0

( , )

( , )

t t t

t t t

D x yTE

D x y

and 1/ 21 1

0 0 0 0 0 01 1 1 1

0 0 0 0 0 0

( , ) ( , )

( , ) ( , )

t t t t t t

t t t t t t

D x y D x yFS

D x y D x y

where , is an input distance function for DMU0 in period t relative to the production frontier

in period 1t , given its input vector and output vector . It implies by how much input quantities can

be proportionally reduced without changing the output quantities produced. M0>1 indicates productivity

gain, M0<1 means productivity loss, and M0=1 implies no productivity change from time t to 1t . Similar,

1TE (or 1 ) indicates progress (or regress) in relative efficiency (so called "Catch-up effect"), and

1FS (or 1 ) implies progress (or regress) in the frontier technology (so called "Frontier-shift effect").

The CRS assumption implicitly assumes that all carriers are operating at the optimal scale, which is

plausible in the telecommunications industry, as merging entities frequently claim that they could achieve

economies of scale after the merger (Goldman et al., 2003). Thus, I further relax the CRS assumption and

decompose MI into three terms under the variable returns to scale (VRS) assumption. Besides TE and FS,

scale efficiency change (SC) is added to investigate whether mergers have improved the carriers' scale

efficiency progress rate.

4. Results

4.1. Result from Traditional DID Approach

(3)

     

109

Table 2 shows the results using all observations in the sample. Note that the lagged effects of second

merger are summarized as TWO14 (from period 1t to 4t ).3 Here, p-value 1 of the Wald-test of the null

hypothesis that the sum of the 3 merger coefficients (ONE in t-1 to t-3) is not statistically different from

zero; p-value 2 of the Wald-test of the null hypothesis that the sum of the 4 merger coefficients (ONE in t-

1 to t-4) is not statistically different from zero.

Some ambiguous results emerge: while the first mergers have no significant effect (p-value

2=0.155>0.1) on merging carriers' productivity growth rate (MI), the second mergers significantly lower

this rate at 5% level; when the first mergers reduce frontier technology progress rate (VRS-FS, p-value

2=0.005 <0.01), the second mergers have no significant effect. One possible explanation is that these merger

effects are calculated using the poor control group even if some merger entities do not have the comparable

controls (common support problem).

4.2. Result from Propensity Score Matching Approach

The propensity score of merger is estimated using a probit regression (Table 3). The dependent

variable takes on two values, 0 and 1, depending on whether a carrier participants a merger in year t . Five

kinds of variables are included in the probit regression: firm size, past performance, competition pressure,

service quality and regulatory environments.

Following Caliendo and Kopeinig (2008), at the cost of losing some significances for some

explanatory variables, some interaction terms and quadratic terms are added to pass the balancing test. Year

dummies and brand dummies are added as well to capture the year-specific and brand-level fixed effects

which might affect the merger status. The results show that the merger likelihood increases for carriers with

fiercer local competition pressure (percentage of business access lines) and better previous performance on

                                                            3 The effect in the year of a merger is separated and is not of interest in that a merger may take place either at the very

beginning or very end of the year.

     

110

revenues growth ratio. The common support condition is also imposed to reduce the potential bias from the

mismatch between the merging and non-merging firms (Heckman et al., 1997).

Table 4 presents the balancing test results from the Capiler matching method. The acquirer and target

are matched with two carriers that have the closest merger probability within a given year (one-to-one

matching with non-replacement). Factors that might simultaneously affect the decision to merge and future

efficiencies are selected.4 Note that overall the matching procedure reduces starting unbalance well: after

matching, there are no significant observable differences in the characteristics of the merging carriers and

the matched non-merging carriers.

Table 5 displays the results using the matched sample. Similarly, p-value 1 of the Wald-test of the null

hypothesis that the sum of the 3 merger coefficients (ONE in t-1 to t-3) is not statistically different from

zero; p-value 2 of the Wald-test of the null hypothesis that the sum of the 4 merger coefficients (ONE in t-

1 to t-4) is not statistically different from zero. Note firstly that the effects of first mergers and second

mergers are consistent in their signs and significances. The results show that the first mergers lower merging

carriers' productivity growth rate (MI), which is statistically significant at the 1% level (p-value<0.01). For

example, productivity growth is 6.2 percentage points lower in the third year after merger, compared to the

control group. The reason is that under the CRS assumption, mergers reduce carriers' innovation/investment

incentives (FS) at the 1% significant level, but do not lower carriers' technical efficiency progress rate (TE,

p-value>0.1). These findings are robust when the VRS assumption is adopted. Moreover, the result shows

that the first mergers do not speed up merging carriers' scale efficiency progress (p-value=0.667>0.1).

5. Conclusion

Using a propensity score matching based DID estimation, I provide further empirical evidence that

mergers reduce merging carriers' total factor productivity growth rate relative to the non-merging carriers.

                                                            4 Following the recommendation of Caliendo and Kopeinig (2008), to pass the balancing test, some interaction terms

and quadratic terms are added.  

     

111

This deterioration is due to that mergers reduce merging carriers' incentive to innovate/invest on frontier

technology, rather than lower carriers' technical efficiencies. Moreover, mergers do not speed up carriers'

scale efficiency progress as the carriers promised in the antitrust review processes. This conclusion should

be explained with cautions, as the carriers in this sample are state-level firms and the scale efficiency

progress could be achieved at the national firm level, which is left for future research.

     

112

References

Blundell, R., and Costa Dias, M. 2000. “Evaluation Methods for Non-Experimental Data.” Fiscal Studies,

21(4), 427-468.

Caliendo, M., and Kopeinig, S. 2008. “Some Practical Guidance for the Implementation of Propensity Score

Matching.” Journal of Economic Surveys, 22(1), 31-72.

Dafny, L. 2009. “Estimation and Identification of Merger Effects: An Application to Hospital Mergers.”

Journal of Law and Economics, 52, 523-550.

Danzon, P., Epstein, A., and Nicholson, S. 2007. “Mergers and Acquisitions in the Pharmaceutical and

Biotech Industries.” Managerial and Decision Economics, 28, 307-328.

Dranove, D. and Lindrooth, R. 2003. “Hospital Consolidation and Costs: Another Look at the Evidence.”

Journal of Health Economics, 22(6), 983–997.

Fare, R., Grosskopf, S., Norris, M., and Zhang, ZY. 1994. “Productivity Growth, Technical Progress, and

Efficiency Change in Industrialized Countries.” American Economic Review, 84(1), 66-83.

Goldman, C., Gotts, I., and Piaskoski, M. 2003. “The Role of Efficiencies in Telecommunications Merger

Review.” Federal Communications Law Journal, 56(1), 87-153.

Greenaway, D., and Kneller, R. 2008. “Exporting, Productivity, and Agglomeration.” European Economic

Review, 52, 919-939.

Heckman, J., Ichimura, H., and Todd, P. 1997. “Matching As An Econometric Evaluation Estimator:

Evidence from Evaluating a Job Training Programme.” Review of Economic Studies, 64, 605-654.

Kwoka, J., and Pollitt, M. 2010. “Do Mergers Improve Efficiency? Evidence from Restructuring the US

Electric Power Sector.” International Journal of Industrial Organization, 28, 645-656.

Majumdar, S. 1997. “Incentive Regulation and Productive Efficiency in the U.S. Telecommunications

Industry.” Journal of Business, 70(4), 547-576.

Ornaghi, C. 2009. “Mergers and Innovation in Big Pharma.” International Journal of Industrial

Organization, 27, 70-79.

     

113

Seo, D., Featherstone, A., Weisman, D., and Gao, Y. 2010. “Market Consolidation and Productivity Growth

in U.S. Wireline Telecommunications: Stochastic Frontier Analysis vs. Malmquist Index.” Review

of Industrial Organization, 36, 271–294.

Thanassoulis, E. 2000. “DEA and Its Use in the Regulation of Water Companies.” European Journal of

Operational Research, 127, 1-13.

     

114

Table 1. Descriptive Statistics

Variables Definition Mean Std. Dev. Min Max

REV Total Operating Revenues (in $millions) 1087.2 1467.4 54.5 9981.0

GROWTH Revenue Growth Ratio (%, t-2 to t) -1.62 4.08 -15.27 11.15

OPEX Total Operating Expenses (in $millions) 569.7 849.9 27.8 5484.7

SWITCH No. of Total Central Switches 193.7 175.5 23.0 1630.0

ACCESS No. of Total Access Lines (in millions) 3185.1 5090.7 102.1 46800.0

LOCAL No. of Total Local Calls (in millions) 5031.4 7259.7 130.2 53500.0

INTRA No. of Total Intra-LATA Toll Calls (in millions) 159.6 493.5 0.1 5151.3

INTER No. of Total Inter-LATA Toll Calls (in millions) 878.1 1270.8 42.9 10400.0

NOREG Percentage of Nonregulated Revenues (%) 7.52 4.83 0.00 94.54

CUSTOMER Percentage of Customer Operating Expenses (%) 24.41 4.51 3.59 38.30

FIBER Percentage of Fiber Lines (%) 10.78 4.24 3.35 50.00

BUSINESS Percentage of Business Access Lines (%) 21.16 7.19 2.73 90.78

TOLL Percentage of Toll Calls (%) 19.83 8.71 5.37 70.83

CLEC No. of CLECs in Each State-Level Market 17.73 17.73 0.00 80.00

     

115

Table 2. Effects of Mergers (using all observations in the sample)

Ln(MI) Ln(CRS-TE) Ln(CRS-FS) Ln(VRS-TE) Ln(VRS-FS) Ln(VRS-SC)

ONE in t -0.003 -0.007 0.004 0.001 0.001 -0.004

(0.011) (0.011) (0.004) (0.009) (0.003) (0.004)

ONE in t-1 -0.004 0.012 -0.016*** 0.006 -0.008* -0.001

(0.010) (0.011) (0.005) (0.008) (0.005) (0.006)

ONE in t-2 0.017* 0.024*** -0.008* 0.012 -0.006 0.011*

(0.009) (0.009) (0.005) (0.008) (0.004) (0.007)

ONE in t-3 -0.048*** -0.039*** -0.010* -0.025*** -0.011*** -0.013**

(0.010) (0.009) (0.005) (0.009) (0.004) (0.006)

ONE in t-4 0.005 0.009 -0.005 0.006 -0.008* 0.006

(0.008) (0.011) (0.006) (0.008) (0.005) (0.004)

TWO in t -0.032*** -0.021* -0.010 0.002 -0.003 -0.028***

(0.012) (0.011) (0.007) (0.008) (0.005) (0.008)

TWO14 -0.013** 0.002 -0.015*** -0.005 -0.003 -0.004

(0.006) (0.005) (0.005) (0.005) (0.004) (0.004)

SBC -0.010 -0.016 0.010 -0.013 0.019 -0.011

(0.011) (0.013) (0.012) (0.009) (0.013) (0.011)

DIVEST 0.011 -0.043** 0.055*** -0.043** 0.056*** 0.001

(0.013) (0.021) (0.015) (0.022) (0.019) (0.004)

ESR -0.063 -0.057 -0.012*** -0.048 -0.006 -0.017

(0.059) (0.056) (0.004) (0.044) (0.005) (0.013)

RCM -0.021*** -0.011* -0.015*** -0.010** -0.017*** -0.000

(0.008) (0.007) (0.006) (0.005) (0.004) (0.005)

PCR -0.005 -0.002 -0.002 -0.001 -0.003 -0.001

(0.005) (0.004) (0.003) (0.003) (0.003) (0.003)

DREG 0.002 -0.005 0.008 0.001 0.006 -0.007

(0.009) (0.008) (0.007) (0.007) (0.005) (0.005)

Constant 0.008 0.003 0.005 0.000 0.004 0.005

(0.009) (0.009) (0.004) (0.006) (0.003) (0.005)

R-Squared 0.131 0.159 0.329 0.098 0.237 0.066

p-value-1 0.063 0.864 0.003 0.620 0.007 0.782

p-value-2 0.155 0.711 0.007 0.971 0.005 0.799

N 924 924 924 924 924 924

Note: Robust standard error in parentheses. Significance level: ***=1%; **=5%; *=10%. The dependent variables are efficiency

changes measured in log. p-value 1 of the Wald-test of the null hypothesis that the sum of the 3 merger coefficients (ONE in t-1 to

t-3) is not statistically different from zero; p-value 2 of the Wald-test of the null hypothesis that the sum of the 4 merger coefficients

(ONE in t-1 to t-4) is not statistically different from zero.

     

116

Table 3. Propensity Score (Probit Regression Models)

Variables (1) (2) (3)

Total Operating Revenues (in log) 0.016 -0.174 -0.961

(0.996) (1.029) (1.565)

Total Operating Revenues (in log), Squared 0.009 0.018 0.042

(0.055) (0.057) (0.085)

Revenues Growth Ratio (%, t-3 to t-1) 0.015 0.013 0.094**

(0.027) (0.027) (0.039)

Percentage of Business Access Lines (%) 0.058*** 0.059*** -0.012

(0.016) (0.017) (0.030)

% Business Lines× No. of CLECs -0.005 -0.004 0.001

(0.005) (0.005) (0.008)

Percentage of Customer Expenses (%) 0.109 0.120 0.136

(0.137) (0.140) (0.223)

% Customer Expense, Squared -0.002 -0.002 -0.002

(0.003) (0.003) (0.004)

Percentage of Fiber Lines (%) 0.004 0.006 -0.057

(0.022) (0.023) (0.054)

Earnings-Sharing Regulation - 0.231 0.713

- (0.502) (0.706)

Rate Case Moratoria Regulation - 0.552 1.073**

- (0.384) (0.522)

Price-Cap Regulation - 0.067 0.304

- (0.209) (0.299)

Deregulation - -0.336 1.063

- (0.495) (0.713)

Constant -4.778 -4.132 0.299

(4.505) (4.607) (7.428)

Year Dummies Yes Yes Yes

Brand Dummies No No Yes

N 924 924 924

Log-Likelihood -198.6 -197.2 -115.0

Note: Standard error in parentheses. Significance level: ***=1%; **=5%; *=10%. The dependent variables are dummies taking

value 1(0) for the group of merging firms (non-merging firms) in the year of the merger t. Here the merging firms include both

acquirers and targets. All the explanatory variables are measured one year before the merger.

     

117

Table 4. Balancing Test

Variable Sample Mean t-test Treated Control t p>|t|

Log total operating revenues Unmatched 9.022 8.635 3.42*** 0.001

Matched 9.064 8.755 1.20 0.232

Log total operating revenues^2 Unmatched 82.654 72.773 3.43*** 0.001

Matched 83.476 77.838 1.22 0.226

Revenues growth ratio (%) Unmatched 0.384 -1.546 4.66*** 0.000

Matched 2.081 2.019 0.09 0.929

Percentage of business lines (%) Unmatched 25.098 21.355 5.36*** 0.000

Matched 25.381 25.780 -0.40 0.689

% Business Lines× No. of CLECs Unmatched 48.169 45.683 1.03 0.303

Matched 45.326 42.992 0.38 0.708

Percentage of customer expenses (%) Unmatched 25.936 24.216 3.76*** 0.000

Matched 25.519 24.532 0.92 0.363

Percentage of customer expenses (%)^2 Unmatched 689.320 606.740 3.69*** 0.000

Matched 673.950 623.220 0.94 0.350

Percentage of fiber lines (%) Unmatched 10.421 10.503 -0.20 0.844

Matched 9.435 9.436 -0.00 0.999

Earnings-Sharing Regulation Unmatched 0.042 0.010 2.79*** 0.005

Matched 0.024 0.000 1.00 0.320

Rate Case Moratoria Regulation Unmatched 0.060 0.030 1.67* 0.096

Matched 0.094 0.045 0.89 0.378

Price-Cap Regulation Unmatched 0.608 0.662 -1.27 0.206

Matched 0.560 0.572 -0.11 0.910

Deregulation Unmatched 0.025 0.033 -0.46 0.649

Matched 0.000 0.046 -1.43 0.156

Note: Significance level: ***=1%; **=5%; *=10%.

     

118

Table 5. Effects of Mergers (control group selected using propensity score)

Ln(MI) Ln(CRS-TE) Ln(CRS-FS) Ln(VRS-TE) Ln(VRS-FS) Ln(VRS-SC)

ONE in t -0.028* -0.024 -0.001 -0.015 -0.005 -0.005

(0.016) (0.015) (0.005) (0.012) (0.004) (0.006)

ONE in t-1 -0.024** 0.001 -0.023*** -0.002 -0.017** -0.001

(0.010) (0.012) (0.008) (0.011) (0.007) (0.004)

ONE in t-2 -0.006 0.013 -0.017*** 0.011 -0.017*** 0.003

(0.010) (0.011) (0.006) (0.010) (0.005) (0.005)

ONE in t-3 -0.062*** -0.046*** -0.015** -0.030*** -0.015** -0.014*

(0.013) (0.011) (0.008) (0.011) (0.006) (0.007)

ONE in t-4 -0.009 -0.003 -0.005 -0.001 -0.012*** 0.006

(0.012) (0.015) (0.006) (0.011) (0.004) (0.005)

TWO in t -0.051*** -0.031** -0.018** -0.012 -0.004 -0.030***

(0.014) (0.014) (0.008) (0.009) (0.006) (0.009)

TWO14 -0.027*** -0.004 -0.021*** -0.009 -0.010** -0.004

(0.007) (0.007) (0.006) (0.006) (0.005) (0.003)

SBC -0.015 -0.019 0.009 -0.018* 0.020 -0.010

(0.012) (0.014) (0.012) (0.010) (0.014) (0.010)

ESR -0.079 -0.079 -0.009 -0.063 -0.007* -0.024

(0.097) (0.091) (0.006) (0.071) (0.004) (0.024)

RCM -0.011 -0.002 -0.020*** -0.002 -0.017*** -0.003

(0.007) (0.009) (0.005) (0.005) (0.005) (0.006)

PCR 0.005 0.004 0.000 0.004 -0.003 0.004

(0.005) (0.004) (0.003) (0.004) (0.003) (0.004)

DREG 0.002 -0.007 0.010 0.001 0.004 -0.008

(0.011) (0.010) (0.009) (0.009) (0.006) (0.005)

Constant 0.025** 0.017* 0.009* 0.009 0.007 0.009*

(0.011) (0.010) (0.005) (0.009) (0.004) (0.005)

R-Squared 0.223 0.207 0.346 0.126 0.250 0.122

p-value-1 0.000 0.123 0.001 0.251 0.001 0.333

p-value-2 0.000 0.192 0.004 0.293 0.000 0.667

N 550 550 550 550 550 550

Note: Robust standard error in parentheses. Significance level: ***=1%; **=5%; *=10%. The dependant variables

are efficiency changes measured in log. DIVEST variable is dropped as there are no Sprint firms in the matched

sample. p-value 1 of the Wald-test of the null hypothesis that the sum of the 3 merger coefficients (ONE in t-1 to t-3)

is not statistically different from zero; p-value 2 of the Wald-test of the null hypothesis that the sum of the 4 merger

coefficients (ONE in t-1 to t-4) is not statistically different from zero.