Cost structure and regulation in the telecommunications industry

(USC Thesis Other)

Cost structure and regulation in the telecommunications industry

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content COST STRUCTURE AND REGULATION IN THE TELECOMMUNICATIONS
INDUSTRY

by

Torna Omar Soro

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements of the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)

December 2006

Copyright 2006 Torna Omar Soro
ii
Dedication

To my parents
iii
Acknowledgements

This work would not have been possible without the support and
encouragement of many people. First and foremost, I would like to thank my
advisor, Professor Geert Ridder for his constant support and encouragement. He
shows me how to overcome the different obstacles I encounter while working on
my dissertation. His support and input really kept me moving ahead.
I would never forget the presence of Professor Jean-Jacques Laffont during
my stay at USC. He was the one who first show me how to do research and
encourage me to attend USC economics Department. I will always be grateful to
him. I would also like to thank Professor Cheng Hsiao, Professor Roger Moon,
Professor W. Bentley MacLeod, for their helpful suggestions, comments and
constant encouragement.
I am also grateful to Professor Hill Han for not only his helpful suggestions
and comments, but also for serving as an outside member on both my qualifying
examination and final committee. My thanks also go to Professor Juan Carello,
Professor Isabelle Brocas for their comments, their support during my job search
and also serving as members of my qualifying exam. I thank Professor John Strauss
for helping me in so many ways. I thank him for encouraging me and believing in
me. I also thank Professor Jeffrey Nugent for his support and constant
encouragement. I thank all the faculty member, staff member and students at the
department of Economics for making my life comfortable throughout my doctoral
iv
study at usc. I would also like to thank all my close friends at USC and outside
USC who have also support me during the years I spent at USC.
I owed a long debt of thanks to my parents. I especially thank them for their
constant prayer and love. I thank all my brothers and sisters for the constant support
and for being there for me. Above all, I thank my mother for her long prayer, day
and night.
v
Table of Contents

Dedication ii

Acknowledgements iii

List of Tables viii

List of Figures x

Abstract xi

Chapter 1: Introduction 1

Chapter 2: Cost Structure, Subadditivity Test and Cost 7
Complementarities in the Telecommunications
Industry: Empirical Analysis.
2.1 Introduction 7
2.2 Model 9
2.3 Estimation Method1 11
2.4 Estimation method 2 16
2.5 Subadditivity and Cost complementarities 17
2.6 Data and estimation of cost function 19
2.7 Results 22
2.8 Conclusion 38

Chapter 3: Structural Regulation, Scale Economies Merger in the 40
Telecommunications Industry: An Empirical Analysis.
3.1 Introduction 40
3.2 Model 42
3.3 Econometric Model 44
3.4 Data 48
3.5 Results 51
3.6 Conclusion 57

vi

Chapter 4: Comparison of Regulatory Regimes in Telecommunications 58
Market: An Empirical Analysis
4.1 Introduction 58
4.2 Telecommunication Cost Proxy Model 61
4.3 Regulatory Regimes 64
4.4 Model 73
4.5 Result of The Optimal Regulatory Mechanism 76
4.6 Alternative Mapping for the Type() β and Effort() e 81
4.7 Conclusion 84

References 86

Appendices 89

Appendix A 89
Appendix B 91
vii
List of Tables

2.1: Summary Statistics of Local exchange Carriers 20

2.2: Full Translog Cost function Estimates 23

2.3: Simplified Translog Cost function Estimates 27

2.4: Subadditivity with No Heterogeneity model 28

2.5: Subadditivity with Heterogeneity model (Fixed Effect) 29

2.6: Subadditivity with Heterogeneity model (DIFF-GMM1) 30

2.7: Subadditivity with Heterogeneity model (DIFF-GMM2) 31

2.8: Size of Firm with Cost Complementarities without Heterogeneity 34

2.9: Size of Firm with Cost Complementarities with Heterogeneity 34

3.1: Summary Statistics of Local exchange Carriers (one output) 49

3.2: Parametric Analysis: Estimates of Pooled Data 1992-2001 53

3.3: Semiparametric Analysis of Total Cost per Output: Estimates of 55
Pooled Data 1992-2001

4.1: Expected Value of Social Welfare 77

4.2: Percentage Losses recovered between Incentive versus 78
Traditional Regulation, Quadratic Disutility Function

4.3: Percentage Losses recovered between Incentive versus 78
Traditional Regulation, Exponential Disutility Function

4.4: Percentage Losses recovered between LT (cost observed) and 79
BM (cost not observed), Quadratic Disutility Function

4.5: Percentage losses recovered between LT(cost observed) and 79
BM (cost not observed), Exponential Disutility Function
viii

4.6: Percentage losses recovered Between PC and LT or BM, 80
Quadratic Disutility Function

4.7: Percentage losses recovered Between PC and LT or BM, 81
Exponential Disutility Function

ix
List of Figures

2.1: Saving from the Existence of a Monopoly 33

2.2: Cost Complementarities 36

2.3: Size of Cost Complementarities with no Unobserved heterogeneity 37

2.4: Size of Cost Complementarities with Unobserved Heterogeneity 37

3.1: Kernel Regression 56

3.2: Natural Cubic Spline 56

x
Abstract
The U.S. telecommunications industry has been the subject of many studies since
the divestiture of AT&T in 1984 and the introduction of competition into long-
distance. In 1996, the US government introduced the telecommunication reform
(1996 Telecommunications Act) that outlines the main rules for competition in the
local telephone market. But until recently, the implementation of the 96-
Telecommunications Act has been problematic and under investigation. This
dissertation consists of three essays, each an application of econometrics and
calibration techniques to the telecommunications industry. The first essay analyzes
the subadditivity and cost complementarities issue in the U.S. telecommunications
industry. I use a dynamic panel data approach to estimate the total cost function of
the local exchange carriers. The estimated cost function is then used in the test for
subadditivity and cost complementarities. I find that the Local exchange carriers’
cost appears to be subadditive meaning that breaking up the local exchange carriers
is not efficient in terms of cost saving. I also find that the cost generated by jointly
producing local and toll calls is on average lower than not producing them jointly.
This means that we have presence of cost complementarities. The wave of merging
in the telecommunication industry may be explained by the presence of cost
complementarities. The second essay estimates semiparametrically the scale
economies in the telecommunication industry. The semiparametric model uses the
translog cost function where output enters non-parametrically. The result shows the
presence of scale economies in the local exchange carrier. The scale economies
xi
seem larger for small firms compared to big firms. This result may also explain the
wave of merging and consolidation that is going on in the U.S telecommunications
industry. The last essay compares the different regulatory models using an
engineering cost proxy function. The method used in our study is based on
generated data from a computer modeling of the local telephone network instead of
the traditional approach, which focuses on real data. Using the calibration
technique I find that the incentive regulation performs better compared to
traditional regulation mechanisms (cost plus regulation). Following the
construction of “cocomo2”, a computer software cost estimation model, I also
develop an alternative mapping for effort and efficiency.

1

Chapter 1
Introduction
1.1 Introduction
The US Telecommunications industry has dramatically evolved since the
divestiture of AT&T in 1984 when competition was introduced into the long
distance telecommunication. Since then local the telecommunications have enjoyed
a monopoly position until 1996. In 1996, congress passed a law that built up the
theoretical guideline for introducing competition into the local telecom market.
However the objective of a facility-based competition has not been reached and a
new type of service called VOIP (Voice over the net) has emerged making the
telecommunication industry more complex to regulate. The outcome today is not
what the FCC was looking for. Although a competitive environment has been set
up in the local market after the 1996 Telecommunications Act was passed, its
practical implementation is still under investigation. The wholesale rate for leasing
set by the incumbents did not encounter the approval of the potential competitors
such as MCI and AT&T. Instead of more competition, competitors that have relied
on leasing Bell lines and gear are pulling back or withdrawing as the Bells
2
recapture local customers and make big inroads in long-distance
1
. Today, the U.S
telecommunications industry is dominated by a wave of merging and consolidation.
The U.S telecommunications market has been the object of many studies. Shine
and Ying (1992) examined the subadditivity of local exchange carriers (LECs)
using a data set that consists of a pooled cross-sectional sample of 58 LECs from
1976 to 1983. They found that the LECs’ cost function is not subadditive. Wilson
and Zhou (2001) incorporated unobserved heterogeneity in Shine and Ying's (1992)
model. As noted by Wilson and Zhou (2001), there are many reasons for
incorporating unobserved heterogeneity in the estimation of cost functions
2
. Using
an unbalanced panel from 1988 to 1995, Wilson and Zhou (2001) found that the
subadditivity tests are sensitive to the specification of the cost functions. Their
results suggested that the cost function of the LECs is subadditive, meaning that
local telephone markets are natural monopolies. Beresteanu (2004) uses a
nonparametrics analysis to estimate the cost complementarities in U.S
telecommunication industry. Using a separable nonparametric specification of the
cost function, his first result shows that identification of economies of scope and
subadditivity depend on the distance of the support of the covariates from the
origin. His second result shows the presence of cost complementarities.

1
(Los Angeles Times: Business 09-10-2004)
2
Gabel and Kennet (1994) point out additional unobserved influences on costs which may bias the
results of a traditional specification of cost. The unobserved characteristics include the omission of
vertical services, interexchange services, the use of proxy variables for customer density which may
bias results for key parameters, and the measurement of capital stock. In addition, LECs operate in
different geographical areas and are subject to differing regulatory environments (Wilson & Zhou,
2001).
3
1.2 U.S. Telecommunications industry
Before 1984, the U.S. Telecommunication industry was operating in a monopoly
environment. The long distance (AT&T) and the local exchange carriers ('Baby
Bells'') were the only ones operating in the US. In 1984, the carriers were broken
up after some studies had demonstrated that the long distance was no longer a
natural monopoly. The competition in the long distance was very successful. Only
the local carriers were still regulated under the incentive regulation mechanism. In
1996, congress passed a law that built up the theoretical guidelines for introducing
competition into the local telecom market. Since the implementation of the
incentive regulation, the US Telecommunications industry has been moving toward
a structure regulation. The goal of structure regulation is to introduce changes in
the organizational structure of an industry so that it more closely approximates the
structure needed for competitive behavior.
3

Although the theoretical lines of a competitive environment have been set up in the
local with the 1996 act, the practical implementation of the competition is still
under way. The wholesale rate (for leasing) set up by the incumbents did not
encounter the approval of the potential competitors such as MCI and AT&T.
Instead of more competition, competitors that have relied on leasing Bell lines and
gear have pulled back or withdrawn as the Bells recapture local customers and
make big inroads in long-distance through merging and acquisition. At the end of
1998 and beginning of 1999, AT&T merged with Tele-Communications

3
(CEPIS/OPS)
4
Incorporated (TCI) a cable network. Its goal was to compete directly with the Baby
Bells on the local phone market. In June 2000, FCC approved AT&T's merger with
MediaOne, a cable television company. This will allow AT&T to transmit voice
over TCI's and MediaOne's cable lines. In July 2000 US West merged with the
backbone provider Qwest. In the same year, telecommunications giants GTE and
Bell Atlantic completed a $116 billion merger after obtaining approval from the
federal regulators. This makes the combined company Verizon Communications
one of the largest providers of wireline and wireless services in the US and also a
leader in Data service (World economy: International Committee of the fourth
international: ICFI). In recent years the telecommunication industry has been fueled
by the Internet growth in demand for additional access lines to high-end data
services. Today new technology such as DSL and the cable modem are coming on
line. In November 2002 Comcast Corp, one the major players in the cable business
obtained regulatory clearance from FCC to merge with AT&T Broadband. This
merging paved the way for the creation of the largest cable and Internet provider in
the country. At the beginning of 2004, the proposed merger of Cingular and AT&T
wireless received a lot of attention from consumers because of a potential in the
dominance of the regional Bell companies in both the landline and wireless.
Cingular is now part of BellSouth. Recently, AT&T began aligning itself more with
cable companies after losing an eight-year battle over leasing local phone lines and
equipment at low wholesale rate from the incumbent. After it stopped marketing
conventional residential service, AT&T Corp inked a deal to sell Internet calling
5
over Adelphia Communications Corp.'s cable lines. This agreement makes
Adelphia the preferred high-speed Internet provider to deliver AT&T's CallVantage
service to Southern California homes (Los Angeles Time Business 09/2/04).
CallVantage is AT&T's form of VOIP, which sends calls over high-speed lines
much like email. For now, there is no competition in the conventional local
telephone service. Cable modem service is the only effective competition. But for
this service to work, it requires consumers to have broadband connections, either a
cable modem or digital subscriber line, or DSL. DSL is controlled by regional Bell
Companies such as SBC Communications Inc., California's dominant local phone
service provider. The break up of the Bell monopoly seems unfeasible. The wave of
acquisition and merging has continued through 2005. In the beginning of 2005,
SBC acquired AT&T and MCI merged with Verizon in March 2005. With the
recent consolidation, mergers and acquisitions, the U.S. telecom industry is down
to five big giants, which are SBC, Verizon, Bell South Corp, Sprint Nextel and
Qwest who couldn't acquire MCI as described in figure below.
These networks will also be served to offer cable TV services. Nobody knows
where the telecommunication industry is heading and nobody is quite sure what to
make of this new era of communication services. From newspapers and radio
stations to new digital technologies in telecommunication, fewer and fewer players
are controlling larger and larger percentages of each communications medium. The
industry is full of opportunities and uncertainty for entrepreneurs. It is well known
that the confusion reigns.
6

Figure 1.1: U.S. Telecommunication History
4

4
The first part of this figure (1984 to 1998) comes form Noll Michael (1999). I complete it until
2005.
AT&T
Bell
System
AT&T
Baby Bells:
Ameritech
Bell Atlantic
Bell South
NYNEX
Pacific Telesis
SBC
US West
NCR McCaw
AT&T AT&T AT&T AT&T
Lucent
Technologies
NCR
TCI
Teleport
Pacific Telesis Pacific Telesis Ameritech
1984 1991 1994
1996
1994
AirTouch Bell Atlantic
NYNEX
US West
MediaOne
SBC SBC
1997
1998
1998
1998
Verizon
Comm
Qwest
Comm
SBC
AT&T
2000
2005
SBC/AT&T
MCI
Verizon
MCI
MCI
Qwest
Comm
Bell
South
Sprint
Nextel
7

Chapter 2
Cost Structure, Subadditivity test and
Cost Complementarities in The
Telecommunications Industry:
Empirical Analysis.

2.1 Introduction
Since the 1996 Telecommunications Act was passed, the government has tried to
put competition in the local exchange carrier market. However the objective of a
facility-based competition has not been reached. A new type of service called voice
over internet (VOIP) has emerged as a potential challenger to the traditional phone.
The outcome today is not what the FCC was hoping for. Although a competitive
environment was set up in the local market after the 1996 Telecommunications Act
was passed, its practical implementation is still under investigation. The wholesale
rate for leasing set by the incumbents did not encounter the approval of the
potential competitors such as MCI and AT&T. Instead of more competition,
competitors that have relied on leasing Bell lines and gear are pulling back or
withdrawing as the Bells recapture local customers and make big inroads in long-
distance. The U.S. telecommunication industry today is a complex one. The
8
question of whether the local telephone utility is a natural monopoly has been the
subject of many empirical studies but the question has never been fully resolved.
Many studies have been done using single output and multi-output cost functions
through different methods to control for technological changes over time. Shine
and Ying (1992) examined the subadditivity of local exchange carriers (LECs)
using a data set that consists of a pooled cross-sectional sample of 58 LECs from
1976 to 1983. They found that the LECs’ cost function is not subadditive. Wilson
and Zhou (2001) incorporated unobserved heterogeneity in Shine and Ying's (1992)
model. As noted by Wilson and Zhou (2001), there are many reasons for
incorporating unobserved heterogeneity in the estimation of cost functions
5
. Using
an unbalanced panel from 1988 to 1995, Wilson and Zhou (2001) found that the
subadditivity tests are sensitive to the specification of the cost functions. Their
results suggested that the cost function of the LECs is subadditive, meaning that
local telephone markets are natural monopolies. The main goal of this paper is to
evaluate the cost complementarities in the presence of unobserved heterogeneity
and without unobserved heterogeneity. A cost function exhibits cost
complementarities when the marginal cost of producing a good one declines as
more of good two is produced. Using a new panel data (from 1992 to 2001) and a
dynamic panel data approach, I first reevaluate the subadditivity issue. I then test

5
Gabel and Kennet (1994) point out additional unobserved influences on costs which may bias the
results of a traditional specification of cost. The unobserved characteristics include the omission of
vertical services, interexchange services, the use of proxy variables for customer density which may
bias results for key parameters, and the measurement of capital stock. In addition, LECs operate in
different geographical areas and are subject to differing regulatory environments. (Wilson & Zhou).
9
for cost complementarities using a cost function without unobserved heterogeneity
and with unobserved heterogeneity.
The remaining part of the paper is structured in the following way: Section 2.2
describes the theoretical model. Section 2.3 shows a description of the dynamic
panel data estimation approach. I give a brief presentation of the Difference
Generalized Method of Moments (DIFF-GMM). Section 2.4 presents the second
estimation method which is the seemingly-unrelated regression techniques. I
describe in Section 2.5 the subadditivity and cost complementarities test. Section
2.6 contains a description of the data. The results are presented in section 2.7.
Section 2.8 concludes.

2.2 Model
Suppose a general model of the cost function is given by the equation below:
() u Z Y P C C , , , , ω =
where C is the cost, Y is the output, P is the factor prices, Z is the observed firm
characteristics, ω is the productivity and u is the error term.
The second order Taylor series approximation of this continuous cost function
expanded jointly in the natural logarithms of the price variable of the inputs and
characteristics is the translog cost function.

10

∑∑ ∑
+ + + =
kl
lt l kt k
j
jt j it
Z Y P C ln ln ln ln
0
δ γ β β (2.1)

∑∑ ∑∑ ∑∑
+ + +
lw
wt lt lw
kg
gt kt kg
jq
qt jt jq
Z Z Y Y P P ln ln
2
1
ln ln
2
1
ln ln
2
1
τ θ η

∑∑ ∑∑ ∑∑
+ + + + +
k
it it
l
lt kt kl
jl
lt jt jl
jk
kt jt jk
u Z Y Z P Y P ω λ µ λ ln ln ln ln ln ln
where
1/
it
C is the lo g of cost of firm i at time t.
2/
it
P is the price variable. It is composed of
a/
it
PL is log of price of labor of firm i at time t.
b/
it
PK is log of price of Capital of firm i at time t.
c/
it
PM is log of price of material of firm i at time t.
3/
it
Y is the output variable. It is composed of
a/
it
TL is log of Total line of firm i at time t.
b/
it
LO is log of local calls of firm i at time t.
c/
it
TO is log of toll call of firm i at time t.
4/ Z is the operating characteristics variable. It is composed of
a/
it
DA is the variable of technology. I used the percentage of digital
access (digital access divided by electronic access)
b/
it
CO is the variable of central offices
11
c/
it
AL is the variable of average lenghth loop
5/
it
ω is the unobserved heterogeneity or productivity
6/
it
u the error term.
We use two methods to estimate the model. The first method is the dynamic panel
data approach. The second method is the iterated-seemingly-unrelated-regression
techniques.

2.3 Estimation method 1.
This method is the dynamic panel data approach. We consider the translog cost
function.
it it
it
it
u X C + + = ω β
'

∑∑ ∑
+ + + =
kl
lt l kt k
j
jt j it
Z Y P C ln ln ln ln
0
δ γ β β (2.2)

∑∑ ∑∑
+ +
kg
gt kt kg
jq
qt jt jq
Y Y P P ln ln
2
1
ln ln
2
1
θ η

∑∑
+
lw
wt lt lw
Z Z ln ln
2
1
τ
∑∑
+
jk
kt jt jk
Y P ln ln λ

∑∑ ∑∑
+ + + +
k
it it
l
lt kt kl
jl
lt jt jl
u Z Y Z P ω λ µ ln ln ln ln
where
it i t it
υ α ξ ω + + = and
it t i it
ϕ ρυ υ + =
−1 ,
with 1 < ρ
12
t
ξ is a year-specific effect,
i
α is an unobserved firm specific effect,
it
υ is the
productivity shock.
it
ω is the unobserved heterogeneity and
it
u is the usual error
term (iid).
We can rewrite
1 , − t i
ρυ as

∑∑ ∑ − − − − −
− − − − =
kl
lt l kt k
j
jt j it t i
Z Y P C
1 1 1 0 1 1 ,
ln ln ln ln ρδ ργ ρβ ρβ ρ ρυ (2.3)

∑∑ ∑∑ − − − −
− −
kg
gt kt kg
jq
qt jt jq
Y Y P P
1 1 1 1
ln ln
2
1
ln ln
2
1
ρθ ρη

∑∑ ∑∑ − − − −
− −
jk
kt jt jk
lw
wt lt lw
Y P Z Z
1 1 1 1
ln ln ln ln
2
1
ρλ ρτ

∑∑ ∑∑ − − − −
− −
kl
lt kt kl
jl
lt jt jl
Z Y Z P
1 1 1 1
ln ln ln ln ρλ ρµ

1 1 − −
− − −
it i t
u ρ ρα ρξ
Equation 2.2 can be written as:

∑∑ ∑
+ + + =
kl
lt l kt k
j
jt j it
Z Y P C ln ln ln ln
0
δ γ β β

∑∑ ∑∑ ∑∑
+ + +
lw
wt lt lw
kg
gt kt kg
jq
qt jt jq
Z Z Y Y P P ln ln
2
1
ln ln
2
1
ln ln
2
1
τ θ η

∑∑ ∑∑ ∑∑
+ + +
kl
lt kt kl
jl
lt jt jl
jk
kt jt jk
Z Y Z P Y P ln ln ln ln ln ln λ µ λ

it it t i i t
u + + + + +
−
ϕ ρυ α ξ
1 ,

13
By substituting
1 , − t i
ρυ into the above equation, the cost equation becomes:
()()()
∑ ∑ − −
− + − + − =
k
kit k kit k
j
jit j jit j it
Y Y P P C
1 1 0
ln ln ln ln 1 ln ργ γ ρβ β β ρ (2.4)
()()
∑∑ ∑ − − −
− + − +
jq
qt jt jq qt jt jq
l
lit l lit l
P P P P Z Z
1 1 1
ln ln ln ln
2
1
ln ln ρη η ρδ δ
()
∑∑ − −
− +
kg
gt kt kg gt kt kg
Y Y Y Y
1 1
ln ln ln ln
2
1
ρθ θ
()
∑∑ − −
− +
lw
wt lt lw wt lt lw
Z Z Z Z
1 1
ln ln ln ln
2
1
ρητ τ
()
∑∑ − −
− +
jk
kt jt jk kt jt jk
Y P Y P
1 1
ln ln ln ln ρλ λ
+ ()
∑∑ − −
−
jl
lt jt jl lt jt jl
Z P Z P
1 1
ln ln ln ln ρµ µ
()
it i t it
kl
lt kt kl lt kt kl
C Z Y Z Y ε α ξ ρ ρλ λ + + + + − +
− − − ∑∑
* *
1 1 1
ln ln ln ln ln
where
1
*
−
− =
t t t
ρξ ξ ξ , ()
i
α ρ α − = 1
*
, and
1 −
− + =
it it it it
u u ρ ϕ ε
The above equation is our dynamic model. By writing equation 2.4 in first
difference, we have:
it t t i t i it
X C C ε ξ π ρ ∆ + ∆ + ∆ + ∆ = ∆ −
*
, 1 ,
ln ln (2.5)
where
X is the right hand side variables of equation 2.4
14

()()
∑ ∑ − −
− + − =
k
kit k kit k
j
jit j jit j it
Y Y P P X
1 1
ln ln ln ln ργ γ ρβ β (2.6)
()()
∑∑ ∑ − − −
− + − +
jq
qt jt jq qt jt jq
l
lit l lit l
P P P P Z Z
1 1 1
ln ln ln ln
2
1
ln ln ρη η ρδ δ
()
∑∑ − −
− +
kg
gt kt kg gt kt kg
Y Y Y Y
1 1
ln ln ln ln
2
1
ρθ θ
()
∑∑ − −
− +
lw
wt lt lw wt lt lw
Z Z Z Z
1 1
ln ln ln ln
2
1
ρτ τ
()
∑∑ − −
− +
jk
kt jt jk kt jt jk
Y P Y P
1 1
ln ln ln ln ρλ λ
+ ()
∑∑ − −
−
jl
lt jt jl lt jt jl
Z P Z P
1 1
ln ln ln ln ρµ µ
()
∑∑ − −
− +
kl
lt kt kl lt kt kl
Z Y Z Y
1 1
ln ln ln ln ρλ λ
This transformation sweeps
*
i
α out of the equation 2.5. While this transformation
solves the problem of heterogeneity (the time invariant effect), it introduces another
endogeneity problem through
1 ,
ln
− t i
C
6
. In order to solve for this endogeneity
problem, Anderson and Hsiao (1982) propose the use of
3 ,
ln
− t i
C as an instrument
for
1 ,
ln
−
∆ t i
C in equation 2.5.
2 , − t i
LnC cannot be used as instrument for
1 ,
ln
−
∆ t i
C
because
2 ,
ln
− t i
C is correlated with
it
ε ∆ . This comes from the relation below:

6
() ( )
1 1
ln ln
− −
= ⇒ =
it it it it
f C f C ε ε
15

1 −
− = ∆ it it it
ε ε ε

2 1 1 1 − − − −
+ − − + − =
it it it it it it
u u u u ρ ρ ϕ ϕ
() ( ) 0 , ln 0 , ln
2 2 2
≠ ∆ ⇒ ≠
− − − it it it it
C Corr u C Corr ε
The vector of instrument used by Anderson and Hsiao can be written as
[ ]
t i t i
X C F
, 3 ,
, ln ∆ =
−
with the assumption that all variables in
t i
X
,
∆ are exogenous.
Arellano and Bond(1991) extend the Anderson and Hsiao's model by arguing that if
3 ,
ln
− t i
C is a good instrument, then
3 ,
ln
− t i
C ,
4 ,
ln
− t i
C ,……,
k t i
C
− ,
ln are also good
instruments.
By making a standard assumption on the initial condition ([][ ] 0
1 1
= =
it i it i
u H H E ϕ
for T t ,......, 3 = ), the moment condition can be written as:
[ ] 0
,
= ∆ − it s t i
H E ε for 3 ≥ s (2.7)
where
( )( )() ( )
()() ()








=
. ln ln , ln ln
, ln ln , ln ln , ln ln , ln ln , ln , ln , ln , ln
ilt ikt ilt ijt
ikt ijt iwt ilt igt ijt iqt ijt it it it it
it
Z Y Z P
Y P Z Z Y Y P P Z Y P C
H

Suitable lags of the variables in equation are used as instruments of their own first
difference. A Generalized Method of Moment (GMM) is then used.
16
2.4 Estimation methods 2

I jointly estimate the cost equation and the share equations.

∑∑ ∑
+ + + + =
kl
l l k k
j
j j
Z Y P t C ln ln ln ln
0
δ γ β τ α (2.8)

∑∑ ∑∑ ∑∑
+ + +
lw
w l lw
kg
g k kg
jq
q j jq
Z Z Y Y P P ln ln
2
1
ln ln
2
1
ln ln
2
1
τ θ η

∑∑ ∑∑ ∑∑
+ + +
kl
l k kl
jl
l j jl
jk
k j jk
Z Y Z P Y P ln ln ln ln ln ln λ µ λ

∑ ∑ ∑
+ + + +
l
l l
k
k j
j
j j
u Z t Y t P t ln ln ln ω ψ ρ
The application of Shephard’s Lemma yields the cost shares of factors as:
∑ ∑ ∑
+ + + + =
l
i l il
k
k ik
h
h ih i i
u Z Y P S ln ln ln σ ϕ η β
where
i
u is an error for the ith share equation.
The above cost equation and the share equations are jointly estimated by iterated-
seemingly-unrelated-regression techniques (Zellner iteration technique).
17
2.5 Subadditivity and Cost Complementarity
Cost economies of scope, subadditivity test and cost complementarities are used
most of the time in economic literature to study the market structure in the industry.
Economies of scope and subadditivity have been widely used in applied works
related to the telecommunications industry. In this article we focus on subadditivity
test and cost complementarities using a new unbalanced data set of the U.S.
telecommunications industry from 1992 to 2001.

2.5.1 Subadditivity test
A cost function of a firm is subadditive if the cost incurred by breaking up a
monopoly is greater than the cost generated by a monopoly. In analytical terms, a
cost function ()
2 1
,Y Y C is subadditive
7
if and only if:
()( )
2 1 2 1
, , Y Y C Y b Y a C
i i
i
>
∑
n i ,......, 1 = and 0 ,
2 1
≥ Y Y
1
1
=
∑
=
n
i
i
a , 1
1
=
∑
=
n
i
i
b
By testing the above condition on three outputs which are
a
Y (access lines) and
t
Y (toll calls) we have:
()
t t l l a a
Y Y Y Y β β β , ,
1
=
()
t t l l a a
Y Y Y Y ) 1 ( , ) 1 ( , ) 1 (
2
β β β − − − = where =
t l a
β β β , , (0.1, 0.2,….,0.9).

7
See Evans and Heckman (1984) for more details.
18
Zero outputs are rule out. The hypothetical cost savings (or increases) owing to
breaking up monopolies are computed as:
()
() ( ) ()
()
100 , ,
2 1
Y C
Y C Y C Y C
Sub
t l a
− +
= β β β
S > 0 implies subadditivity which means that the percentage cost increases from
breaking up a monopoly. S < 0 implies superadditivity which is the percentage cost
saving from replacing a monopoly with multiple-firm production.

2.5.2 Cost Complementarities test
A cost function exhibits cost complementarities when the marginal cost of
producing a good one declines as more of good two is produced () 0 /
2 1
< ∆ ∆ Q MC .
I test the condition on two outputs, which are local calls and toll calls. A firm's cost
function, C, has cost complementarities if C is submodular
8
. This means that:
( ) () [ ] ( ) () [ ] ( ) () [ ] Y C Y Y C Y C Y Y C Y C Y Y C − Ω + + − Ω + ≤ − +
'
2
'
1
'

or
( ) () [ ] ( ) () [ ] ( ) () [ ] 0
' '
2
'
1
≥ − + − − Ω + + − Ω + Y C Y Y C Y C Y Y C Y C Y Y C (2.9)
for any β ∈ ' Y and 0
'
> Y such that 1 0
1
≤ Ω ≤ , 1 0
2
≤ Ω ≤ and 1
2 1
= Ω + Ω .

8
g is submodular if (-g) is supermodular (Beresteanu 2004)
A function g: R R
n
→ is supermodular if for all
() ( ) ( ) () () (). ' , max ' , min ' , ' , x x g x x g x g x g R x x
n
+ ≤ + ∈ This inequality is equivalent to
() ( ) ()() () () [] () () () () [] ' , min ' , max ' , min ' ] ' , min [ x x g x x g x x g x g x x g x g + ≤ + + +
This means that the change in the function when several arguments are increased separately is less
than the change resulting from increasing all the arguments together. See Topkis 1998, for more
detail.
19
In the above equation, () () Y C Y Y C − + ' represents the increase in cost when output
is increased jointly by ' ' '
2 1
Y Y Y Ω + Ω = . The following formula
( ) () [ ] ( ) () [ ] Y C Y Y C Y C Y Y C − Ω + + − Ω +
'
2
'
1
represents the sum of two independent
increases in cost when the production is increased from Y to '
1
Y Y Ω + and from
Y to '
2
Y Y Ω + .
Cost Complementarities exist if:
()
()() []()() [] []
() () [] []
0 100 1
'
' '
' ,
2 1
≥






−
− +
− Ω + + − Ω +
=
Y C Y Y C
Y C Y Y C Y C Y Y C
Y Y Comp (2.10)

2.6 Data and estimation of cost function
2.6.1 Data source
The data set consists of an unbalanced panel of 49 local exchange carriers (LECs)
over 1992-2001. It covers the period after the divestiture or breakup of AT&T and
the Baby Bells. The primary data source is the Statistics of Communications
common carriers (SOCC), published annually by the FCC and the Electronic Armis
Filing System (ARMIS). The ARMIS was initiated in 1987 for collecting financial
and operational data from the largest incumbent local exchange carriers. The
Electronic ARMIS filing System is a web-based application that allows users with
web browsers and Internet access to obtain ARMIS data in various formats. The
source includes data on the major LECs assets, liabilities, revenues, expenses, plant
statistics, and output. The sample comprises 49 LECs and 480 observations. 480
20
are the total number of observations for all carriers. Table A.1 in appendix contains
the name and years of operation of the local exchanges carriers (LECs).
Total cost, labor price, capital price, material, investment are measured in
thousands dollars. Local calls, toll calls and access lines are measured in thousands
(counting). Local loop circuit length is measured in miles.

Table 2.1 Summary Statistics of LEC data
Variables Obs Mean Std.dev Minimum Maximum
Total Cost 458 1,755,866 2,566,294 73,426.71 12,500,00
0
Price of Labor 458 53.324 9.422 35.532 90.627
Price of Capital 458 0.146 0.033 0.049 0.236
Price of Material 458 0.070 0.029 0.0213 0.1932
Local Calls 458 10,080,000 10,080,000 156,486 99,300,00
0
Toll Calls 458 1,999,560 2,907,434 36,673 16,800,00
0
Access lines 458 4,118,401 6,375,641 118,586 39,200,00
0
Interlata Billed Access Mn 458 13,300,000 20,000,000 332,445 110,000,0
00
Technology proxy
(%digital access)
458 0.129 0.123 2.74e-07 0.714
Number of central offices 458 432.520 471.588 23 1967
Local loop circuit lenght 458 25,200,000 50,800,000 12789 300,000,0
00
Material 458 685,809.9 994,078.1 28,156.16 4,943,190
Investment 458 201,832.1 437,554.2 557.449 4,278,906

21
2.6.2 Variable construction
I follow the method Shin and Yin (1992) used to compute the total cost, price of
capital, price of labor and output. The total cost (TC) is given by operating
expenses minus depreciation. For capital expenditures, we used the gross
communications plant, which includes plant in service (central office equipment,
cables, buildings, etc.) and plants under construction. From the Statistics of
Communications Common Carriers, I use the total telecommunications plant in
service as capital expenditure. I obtain a real capital stock by dividing the gross
capital communication plant by the 20 year average communications equipment
implicit price deflator. This price deflator is available at the National Income and
Product Accounts. I finally convert the real capital stock to current dollars. The
price of labor (PL) is the compensation per employee. The price of capital (PK) is
capital expenses divided by the average number of access lines. The price of
material (PM) is obtained by dividing the cost of materials by the aggregate number
of calls. The output variable is the average number of access lines or telephones
(TL) and the output usage variables consist of local calls (LO) and toll calls (TO).
Local calls are calls within the local area. The local area is the geographical area
within which subscribers can call each other without incurring tolls (extra charges
for a call). Toll calls and long distance calls are calls that go outside the local
exchange. In my data, the toll calls are the sum of intra - local Access and
Transport Area (Intra-LATA) toll call and the inter-local Access and Transport
Area (Inter-LATA) toll call. Intra-LATA toll calls are calls that are made outside
22
the local calling area but inside the LATA and are carried by the LEC. Intrastate
calls are calls made within the state but outside the LATA. Interstate calls are calls
made from one state to another. Both Intrastate and Interstate are part of Inter-
LATA toll calls and require a long distance carrier. The DA is the percentage of
access line that is digital.
It represents the observed technology. All prices are in real terms. They are deflated
by the gross domestic product price index (GDP-PI).

2.7 Results
I report the results obtained from the estimation in Table 2.2 and Table 2.3. Table
2.2 contains three models. They all use the translog cost function. The first model
specification in Table 2.2 does not control for unobserved heterogeneity. It is
similar to the model used by Shin and Ying (1992). Their data set covers the year
1976 to year 1983. At that time the Bell operating companies (``Baby Bells'') were
an integral part of the Bell system and AT&T was the sole provider of long
distance. A dummy variable was used to denote the Bell operating companies. I did
not include a dummy for Bell operating companies in our model. The second model
in Table 2.2 is a fixed effect approach. It extends the first model by using firm
dummy variable to control for unobserved heterogeneity. This model is similar to
Wilson and Zhou (2001)’s model.

23
Table 2.2: Full Translog Cost Function Estimates

24
The two first models use Zellner's iterative seemingly unrelated regression
technique (Zellner, 1962). The cost function and share equations are jointly
estimated by iterated-seemingly-unrelated-regression techniques that converge to
maximum likelihood estimator. The cost function and the share equations are
described in the appendix. The third model in Table 2.2 shows the estimates of our
dynamic panel data approach. In the three models, the price of labor (PL), price of
capital (PK) and price of material (PK) have the expected (positive) sign and are
statistically significant at 5% except for the price of capital in model 2 (with
unobserved heterogeneity: fixed effect) where the parameter is significant at 10%.
The outputs total line (TL), local calls (LO) and toll calls (TO) affect positively the
cost function in the three models. They are statistically significant at 5% except for
the toll calls in the model without heterogeneity (first model). In the two first
models, the variable that denotes the digital access (DA) has a positive and
insignificant effect on the cost. In the dynamic model, the use of digital switching
(DA) equipment rather than electronic or analog equipment reduces the production
cost but not in a significant manner. Digital and electronic technologies offer many
advantages. They increase network capability and reduce the cost of capacity and
access (Correa 2003). All the studies using US data such as Shin and Yin (1992),
Wilson and Zhou (2001) and Beresteanu (2004) had focused on electronic
technologies. The reason is due to their data set covering the period before 1995.
The central offices (CO) parameter has a negative sign for both non-heterogeneity
and heterogeneity model. But it is only significant in the first model (non
25
heterogeneity). In our DIFF-GMM model, the coefficient of the central offices
(CO), although positive, is not significant. This means that marginal effects of the
number of central offices on local exchange carriers' cost are negligible. The
average loop length decreases the production cost in the three models. This variable
is only significant at 5% in the first model.
In table 2.3, we use a simplified translog cost function. We use only the interaction
between the price of input (PK, PL, PM) and the output (TL, LO, TO). The
characteristics variables (al, da, co) enter linearly. This simplified translog function
was first used by Yatchew (2000) in estimating an electricity distribution cost
function for Canada. We report the estimates for the model without heterogeneity,
the model with heterogeneity, first-differenced instrument variable model
(Anderson and Hsiao, 1982) and the dynamic panel data model (difference-GMM).
We show only part of the estimates in Table 2.3. This simplified translog cost
function offers the advantage of having fewer regressors compared to the full
translog used by earlier studies. The prices of the three inputs are significant at 1%
in the unobserved heterogeneity models while only the price of capital and price of
material are significant. The output variables (TL, LO, TO) are positive and
significant at 5% for the unobserved heterogeneity models. The only outputs that
are significant are the total line (TL) and local call).
Focusing on the DIFF-GMM model, the Sargan Test does not reject the null
hypothesis that over-identification restrictions are valid. This means that our model
has valid moment restriction. The null hypothesis of no first order autocorrelation
26
in the difference residuals is rejected, but it is not possible to reject the null
hypothesis of no second-order autocorrelation. According to Arellano and Bond
(1991 and 1998) the first differencing introduces MA(1) serial correlation when the
time-varying component of the error term in levels is serially uncorrelated. So it's
required that the test for first -order autocorrelation rejects the null while the test for
second-order autocorrelation fails to reject the null hypothesis of no correlation.
The GMM is only consistent when the second-order correlation is not significant.
The conclusion is that the Sargent Test, m1 and m2 supports the validity of the
DIFF-GMM estimator in Table 2.3.

27
Table 2.3 Simplified Translog Function Estimates

28
The Table 2.4, Table 2.5, Table 2.6 and Table 2.7 below show respectively the
subadditivity test for model 1 (with no heterogeneity), model 2 (with
heterogeneity), the DIFF-GMM model with full translog cost function and the
DIFF-GMM with the simplified translog cost function.

Table 2.4: Subadditivity with No Heterogeneity model

29

Table 2.5: Subadditivity with Heterogeneity model (Fixed Effect)

30

Table 2.6: Subadditivity with Heterogeneity model (DIFF-GMM1)

31

Table 2.7: Subadditivity with Heterogeneity model (DIFF-GMM2)

For each of the 458 observations, I compare the cost of the 365 hypothetical
9

output vector pairs to the cost of the monopoly firm's output. I compare the results
of our subadditivity tests of the three output specifications, which are access lines,
local call and toll calls. I process the data and compute the statistics for each year.
The results are qualitatively similar for all the years. For all the cases the statistics
S is greater than zero meaning that we have subadditivity
10
. For instance, in 2001 in
Table 2.4, we have 10950 configurations and the results shows that in all theses
cases, a single firm is always efficient compared to two firms. The minimum and

9
The 365 hpothetical outputs vector comes from the following computation: (9*9*9)/2 =365.
10
S > 0 implies that the percentage cost increases from breaking up a monopoly.
32
the maximum savings from having a monopoly is respectively 62.102% and
297.921% with an average value of 115.168%. This means that a single firm can
always produce at a lower cost than two or multiple firms. The four models exhibit
subadditivity cost function. These results are qualitatively consistent with findings
by Roller(1990a,b) and Wilson, Zhou (2001) who also used a model that accounts
for the heterogeneity issue. That's not the case with the work of Shin and Yin
(1992), which ignores the heterogeneity issue. They found that the local exchange
carriers exhibit superadditive cost. By taking into account the heterogeneity issue in
our translog cost function, our study shows that the local telephone markets are
natural monopolies. So breaking up those local exchange carriers by allowing some
potential entrants to share the central offices and other facilities is not a good
option. These results may explain why the FCC was having difficulties handling
the wholesale price issue and other difficulties encountered by the new entrants into
the local market. For instance AT&T pull back from the local telephone market.
Figure 2.1 shows the magnitude of the savings from the existence of a monopoly in
the three models. The model without heterogeneity has the biggest saving followed
by the model with heterogeneity. The two DIFF-GMM models have the smallest
saving among the four models.
33

Figure 2.1: Saving from the Existence of a Monopoly

Our second result focuses on the complementarities issue.
The following statistics give an idea about the size of cost complementarities in the
telecommunication industry. Table 2.8 and table 2.9 give the size of cost
complementarities for each combination (458).
34

Table 2.8: Size of firms with cost complementarities without Heterogeneity

Table 2.9: Size of firms with cost complementarities
with Heterogeneity (FD2SLS)

I summarize the results for various levels of local and toll call usage. The statistics
are positive for both small and large carriers meaning that there is a presence of
cost complementarities. The number in percentage is the proportion of firms that
exhibit cost complementarities. This result suggests that the different carriers can
35
take advantage of the cost complementarities by merging or extending their
production to other services such as cable TV services. The costs generated by
jointly producing local and toll calls are on average lower than not producing them
jointly. The second result shows that the cost complementarities decrease when we
move from a cost function without unobserved heterogeneity to a cost function
with unobserved heterogeneity. The carriers can increase their production without
generating additional cost if they use the local network. This means that it will be
efficient in terms of great return from complementarities if a bigger local network
via merging is used. The wave of recent merging of MCI\Verizon (in March 2005)
and SBC\AT&T can prove our points. With the recent consolidation, mergers and
acquisitions, the US telecom industry is down to five big giants: SBC, Verizon,
Bell South Corp and Sprint Nextel and. They also planned to use their network to
offer cable TV services. Figure 2.2 shows the percentage of firms that exhibit cost
complementarities. It’s an average number coming from table 2.8 and table 2.9 and
36

Figure 2.2: Cost complementarities

Figure 2.3 and Figure 2.4 show the size of cost complementarities for different
fraction of Beta under the presence of unobserved heterogeneity and without
unobserved heterogeneity. We can see that the size is well above 50% meaning that
cost complementarities exist in the U.S. telecommunications industry.
37

Figure 2.3: size of cost complementarities with no unobserved heterogeneity

Figure 2.4: size of cost complementarities with unobserved heterogeneity

38

2.8 Conclusion
The U.S. telecommunications industry has been the subject of many studies since
the divestiture of AT&T in 1984 and the introduction of competition into long-
distance. In 1996, the US government introduced the telecommunication reform
(1996 Telecommunications Act) that outlines the main rules for competition in the
local telephone market. But until today, the implementation of the 96-
Telecommunications Act has been problematic and under investigation. The
wholesale rate (for leasing) set up by the incumbents did not encounter the approval
of the potential competitors such as MCI and AT&T. Instead of more competition,
we realized that competitors that have relied on leasing Bell lines and gears, are
pulling back or withdrawing as the Bells recapture local customers and make big
inroads in long-distance. I reexamine the natural monopoly and cost
complementarities issue using a model that includes an unobserved heterogeneity. I
then use a dynamic panel data approach to estimate the total cost function of the
local exchange carriers. I use a new recent unbalanced panel data set that covers the
period between 1992 and 2001. The estimated cost function is used in the test for
subadditivity and cost complementarities, which are two characteristics of cost
structure. I find that the Local exchange carriers cost appears to be subadditive
meaning that breaking up the local exchange carrier is not efficient in terms of cost
saving. The regulated incumbent local exchange carriers are natural monopolies.
39
The second result concerns the complementarities issue. I find that the cost
generated by jointly producing local and toll calls is on average lower than not
producing them jointly. Besides, the cost complementarities decrease when we
move from a cost function without unobserved heterogeneity to a cost function
with unobserved heterogeneity. The carriers can increase their production without
generating additional cost if they use the local network. This means that it will be
efficient in terms of great return from complementarities if a bigger local network
via merging is used. The wave of recent merging of MCI/Verizon and SBC/AT&T
can prove my point. With the recent consolidations, mergers and acquisitions, the
US Telecom industry is down to five big giants, which are SBC, Verizon, Bell
South Corp, Sprint Nextel and Qwest. They also planned to use their network to
offer cable TV services. The findings are in line with the actual configuration of the
U.S. telecommunication market.
40

Chapter 3

Structural Regulation, Scale
Economies and Merger in the
Telecommunications Industry: An
Empirical Analysis.

3.1 Introduction

The US Telecommunications industry has dramatically evolved since the
divestiture of AT&T in 1984 when competition was introduced into the long
distance telecommunication. Since then local telecommunications have enjoyed a
monopoly position until 1996. In 1996, congress passed a law that built up the
theoretical guideline for introducing competition into the local telecom market.
However the objective of a facility-based competition has not been reached and a
new type of service called VOIP (Voice over the net) has emerged making the
telecommunication industry more complex to regulate. The outcome today is not
what the FCC was looking for. Although a competitive environment was set up in
the local market after the 1996 Telecommunications Act was passed, its practical
implementation is still under investigation. The wholesale rate for leasing set by the
incumbents did not encounter the approval of the potential competitors such as
MCI and AT&T. Instead of more competition, competitors that have relied on
41
leasing Bell lines and gear are pulling back or withdrawing as the Bells recapture
local customers and make big inroads in long-distance
11
. The U.S.
telecommunications industry today is complex. Today, the U.S telecommunications
industry is dominated by a wave of merging and consolidation.
This paper explains this wave of merging and consolidation by estimating the
presence of scale economies. The paper estimates semiparametrically the scale
economies in the US Telecommunication industry using updated panel data (from
1992 to 2001). But instead of using directly a semiparametric function, I begin with
a parametric function that includes an unobserved unobserved heterogeneity
(productivity). I mimic the adjusted translog cost function used by Yatchew (2000).
He uses a semiparametric (difference) approach to analyze the scale economies in
the distribution part of Canadian electricity.
I use the translog cost function to estimate the scale economies. In our final cost
function, the output enters nonparametrically while the remaining variables enter
parametrically. Section 3.2 shows the theoretical model. In section 3.3 I describe
the econometrics model and the estimation technique. Section 3.4 contains a
description of the data. The results are presented in section 3.5. Section 3.6
concludes.

11
(Los Angeles Times: Business 09-10-2004)
42
3.2 Model
Suppose a general model of the cost function is given by the equation below:

() u Z Y P C C , , , , ω =

where C is the cost, Y is the output, P is the factor prices, Z is the observed firm
characteristics, ω is the productivity and u is the error term.
The second order Taylor series approximation of this continuous cost function
expanded jointly in the natural logarithms of the price variable of the inputs and
characteristics is the translog cost function.

∑∑ ∑
+ + + =
kl
lt l kt k
j
jt j it
Z Y P C ln ln ln ln
0
δ γ β β (3.1)

∑∑ ∑∑
+ +
kg
gt kt kg
jq
qt jt jq
Y Y P P ln ln
2
1
ln ln
2
1
θ η
+ +
∑∑
lw
wt lt lw
Z Z ln ln
2
1
τ
∑∑
jk
kt jt jk
Y P ln ln λ

∑∑ ∑∑
+ + + +
k
it it
l
lt kt kl
jl
lt jt jl
u Z Y Z P ω λ µ ln ln ln ln
where
1/
it
C is the log of cost per unit of output () () Y TC C
it
/ log = .

2/
it
P is the price variable. It is composed of
43
a/
it
PL is log of price of labor of firm i at time t.
b/
it
PK is log of price of Capital of firm i at time t.
3/
it
Y is variable for the output variable.
4/ Z is the operating characteristics variable. It is composed of
a/
it
DA is the variable of technology. I used the percentage of digital
access (digital access divided by electronic access)
b/
it
CO is the variable of central offices
c/
it
AL is the variable of average length loop
d/
it
M is the variable for material.
e/ Dum96 is the dummy variable for structural regulation (it takes the value
0 before 1996 and 1 after 1996)
5/
it
ω is the unobserved heterogeneity or productivity
6/
it
u is the error term.
I suppose that the unobserved heterogeneity
it
ω is potentially correlated to some
explanatory variables such as output (Y) and even employment and capital.
44

3.3 Econometric Model

Instead of using the full translog cost function, I use a simplified translog cost
function that is similar to the one used by Yatchew (2000) in the estimation of the
electricity distribution in Canada. He uses a translog cost function that takes into
account only the interaction between the prices vector and the output. I mimic his
cost function. The advantage of this simplified translog cost function is that it
decreases the number of regressors. The characteristics variable Z that includes
central office, average loop and digital access enter the cost function without
interacting with other variables.

∑∑ ∑
+ + + =
kl
lt l kt k
j
jt j it
Z Y P C ln ln ln ln
0
δ γ β β (3.2)

∑∑ ∑∑
+ +
kg
gt kt kg
jq
qt jt jq
Y Y P P ln ln
2
1
ln ln
2
1
θ η

∑∑
+ + +
jk
it it kt jt jk
u Y P ω λ ln ln
it it it r it y x it
u M Y X C + + + + = ω β β β
'

I denote by M the material used as proxy for the unobserved heterogeneity
12
.
()
it it t
Y f M , ω = (3.3)

12
I follow the methodology closed to Olley, Pakes (1996) and Levinsohn, Petrin (2000). I take their
approach as our starting point. Their approach is applied to a single production function. Olley,
Pakes (1996) focus on a different market and on production functions. My work focuses on the
telecommunication industry and on cost functions.
45
I assume that this equation is strictly increasing in
it
ω (for every Y). By inverting
equation 3.3 we get
()
it it t it
Y M f ,
1 −
= ω (3.4)
The first key assumption in this model is strict monotonicity. I assume investment
or material must be strictly monotonic in
it
ω . Monotonicity is required for the
inversion. This will allow us to invert out
it
ω and completely remove the
endogeneity problem.
The unobserved productivity variable ω is expressed as a function of observables
and this helps us to control for ω in estimation.
()
it it it t it y x it
u Y M f Y X C + + + =
−
,
1 '
β β
where X is the vector of prices and operating characteristics. Y is the output vector.
I can rewrite the above equation as:
()
it it it t x it
u Y M X C + + = ,
'
ψ β
where
() ( )
it it t it y t
Y M f Y Y M , ,
1
0
−
+ + = β β ψ (3.5)
For simplicity, I assume that i() Y M
t
, ψ is separable and
() () () Y M Y M
t 2 1
, ψ ψ ψ + = where ()
it r
M M β ψ =
1
.
I can rewrite the cost per capita as
()
it it r x it
u Y M X C + + + =
2
'
ψ β β
I can summarize our basic econometric specification as:
46
2
22
2
11 2 1
log
2
1
log
2
1
log log log PK PL PK PL C
it
β β β β + + + =
()( )( ) PK Y PL Y PK PL
Y Y
log log log log log log
2 1 12
∗ + ∗ + ∗ + β β β
()
it it it Y
DA AL CO PK Y log log log log log
5 4 3 2
β β β β + + + ∗ +
()
it it M
u Y M + + + log log
2
ψ β

I use the Robinson's (1988) method to remove the nonparametric part. I then
construct estimates of the conditional moments vector () Y C E log / log ,
() Y X E log / log , () Y M E log / log .
I used the Nadaraya (1964)-Watson (1964) kernel
13
to compute those conditional
moments. I can rewrite

()( ) ( )() Y Y M E Y X E Y C E
r x
log log / log log / log / log
'
ψ β β + + =

where
() 0 log / = Y u E and () ()() Y Y Y E
t t
log log / log ψ ψ = .

I estimate nonparametrically the conditional moments. I then subtract the estimates
of the conditional moments from the observed data. This gets rid of the
nonparametric part. Finally I estimate the parameters with our new equation.

13
I use gauss to estimate the conditional moments
47
() ( ) () ( ) () u Y M E M Y X E X Y C E C
r x
+ − + − = − log / log / log / log log
'
β β

I know that u is the conditional mean independent of the variable inputs, so no-
intercept ordinary least square estimation can be used to obtain estimates of the
parameters. In Robinson's approach, the dependent and explanatory variables have
been constructed. So they depend on the local least squares estimates. This
situation makes the OLS standard errors biased for our application. I use the
bootstrap to estimate the standard error. The primary goal of the nonparametric
method has been to provide statistical tools that work in complex situations without
imposing unrealistic or unverifiable assumptions about the data-generation
mechanism.
Returning to the nonparametric specification, I remove the estimated parametric
effect from the dependent variable and analyze the non-parametric effect. I get:
()
i i i i
y f x c υ β + = −
ˆ
(3.6)
I then apply conventional kernel or spline estimation methods to the
pairs( )
i i i
y x c ,
ˆ
β − . I can now analyze the scale economies through the generated
graph.
48
3.4 Data
3.4.1 Data source
The data set consists of an unbalanced panel of 49 local exchange carriers from
1992 to 2001, ie after the breakup of ATT and the Baby Bells. The primary data
source is the Statistics of Communications common carriers (SOCC), published
annually by the FCC and the Electronic Armis Filing System (ARMIS). The
ARMIS was initiated in 1987 for collecting financial and operational data from the
largest incumbent local exchange carriers. The Electronic ARMIS filing System is
a web-based application that allows users with web browsers and internet access to
obtain ARMIS data in various formats. The source includes data on the major
LECs assets, liabilities, revenues, expenses, plant statistics, and output. The sample
comprises 49 LECs and 480 observations.
49

Table 3.1: Summary Statistics of LEC data
Variables Obs Mean Std.dev Minimum Maximum
Total Cost 458 1,755,866 2,566,294 73,426.71 12,500,000
Price of Labor 458 53.324 9.422 35.532 90.627
Price of Capital 458 0.146 0.033 0.049 0.236
Operating Revenue
(Output)
458 2,337,844 3,391,754 111,580.1 1.77e+07
Interlata Billed Access Mn 458 13,300,000 20,000,00
0
332,445 110,000,000
Technology proxy
(%digital access)
458 0.129 0.123 2.74e-07 0.714
Number of central offices 458 432.520 471.588 23 1967
Local loop circuit lenght 458 25,200,000 50,800,00
0
12789 300,000,000
Material 458 685,809.9 994,078.1 28,156.16 4,943,190
Investment 458 201,832.1 437,554.2 557.449 4,278,906
DA = Percentage of Digital Access.

The use of panel data provides major benefits over data sets consisting of a single cross
section or time series for econometric estimation in at least four areas (Hsiao 2000). First,
the panel data increases degrees of freedom and reduces problems of data multicollinearity,
second it helps to identify economic models and discriminate between competing
economic hypotheses, and third it eliminates or reduces estimating bias. Finally, it provides
micro foundations for aggregate data analysis.
50
3.4.2 Variable construction
I follow the method of Shin and Yin (1992) to compute the total cost, price of capital, price
of labor and output. . For capital expenditures, we use the gross communications plant,
which includes plant in service (central office equipment, cables, buildings, etc) and plants
under construction. In the Statistics of Communications Common Carriers, we use the total
telecommunications Plant in service as capital expenditure. I obtain a real capital stock by
dividing the gross capital communication plant by the 20 year average communications
equipment implicit price deflator. This price deflator is available at the National Income
and Product Accounts. I convert the real capital stock to current dollars. The price of labor
(PL) is the compensation per employee. The price of capital (PK) is capital expenses
divided by the average number of telephones or access lines. The total cost (TC) is given
by operating expenses minus depreciation. The DA is the percentage of access line that is
digital
14
. Instead of taking the average number of access lines or telephones (TL), local
calls (LO) and toll calls (TO) as the output variables, I use the operating revenue (Y). The
local area is the geographical area within which subscribers can call each other without
incurring tolls (extra charges for a call). Toll calls and long distance calls are calls that go
outside the local exchange. The toll calls are the sum of intra - local Access and Transport
Area (Intra-LATA) toll call and the inter-local Access and Transport Area (Inter-LATA)
toll call. IntraLATA toll calls are calls that are made outside the local calling area but
inside the LATA and are carried by the LEC. Intrastate calls are calls made within the state
but ouside the LATA. Interstate calls are calls made from one state to another. Both
Intrastate and Interstate are part of InterLATA toll calls and require a long distance carrier.

14
The whole world is going digital. Telecommunication systems are increasingly employing digital
switching and digital transmission media. In Telecommunication, the technology has moved from
mechanical to electronics and it's now heading to a digitized environment.
51
In our case it represent the observed technology. Shin , Yin (1992) and Wilson, Zhou
(2001) used the percentage of lines that are electronic.
All prices are in real terms. There are deflated by the gross domestic product price index
(GDP-PI).

3.5 Results
Our cost equation in more extensive form is the transformed translog cost function.

2
22
2
11 2 1
log
2
1
log
2
1
log log log PK PL PK PL C
it
β β β β + + + =
()( )( ) PK Y PL Y PK PL
Y Y
log log log log log log
2 1 12
∗ + ∗ + ∗ + β β β
()
it it it Y
DA AL CO PK Y log log log log log
5 4 3 2
β β β β + + + ∗ +

it it it d Y it M
u Dum Y M + + + + + ω β β β 96 log log

I use a proxy to control for the unobserved heterogeneity. The derived translog cost
function has a semiparametric form. The output enters nonparametrically and the
remaining variables take a parametric form.
2
22
2
11 2 1
log
2
1
log
2
1
log log log PK PL PK PL C
it
β β β β + + + =
()( )( ) PK Y PL Y PK PL
Y Y
log log log log log log
2 1 12
∗ + ∗ + ∗ + β β β
()
it it it Y
DA AL CO PK Y log log log log log
5 4 3 2
β β β β + + + ∗ +
()
it it M
u Y M + + + log log
2
ψ β
52
Where () () Y TC C C
it it
/ log = is the log of cost per output from firm i at time t,
it
Y is the
log of output variable which is operating revenue,
it
PK is the log of capital price,
it
PL is
the log of labor price.
it
M is the log of material,
it
CO the log of number of central offices,
it
AL is the log of average loop length,
it
DA is the log of the percentage of access lines
that are digital, Dum96 is dummy variable for structural regulation. it takes the value 0
before 1996 and 1 after 1996 which is the period after the establishment of the
telecommunication act in 1996.
Table 3.2 shows the estimates of a parametric translog cost function. I have the results for
OLS estimation, within and total (GLS) estimation and dynanic panel data estimation. In
the dynamic estimation, I only focus on first difference instrument variable. In the three
models, the price of labor (PL) and the price of capital (PK) have the positive sign and the
estimates appear to be significant. The output (y) has a negative sign and appears to be
significant. This means that the output negatively affects the total cost per output (average
cost). As the output increases, the cost per output decreases. The increase of the production
and capacity appears to be beneficial for the firm. This means that, there is a presence of
scale economy when using a parametric functional form on the US data. The coefficient of
the structural variable (Dum96) appears to affect negatively and significantly the cost
function. This means that the passing of the telecommunication act in 1996 by the US
congress, has led the telephone companies to operate more efficiently.
In table 3.3, I describe the estimates of our semiparametric cost function. I use the translog
cost function. I use gauss program to compute the conditional moments. I bootstrap the
standard error given the reason mentioned above.
53
Table 3.2: Parametric Analysis: Pooled data1992 to 2001
Panel data Dynamic Panel
data
Variable OLS Within Group Total FDIV
Pl 0.2426 0.1261 0.1509 0.1773
0.0289 0.0245 0.0257 0.0242
Pk 0.1120 0.0965 0.1205 0.1580
0.0248 0.0302 0.0280 0.0388
Y -0.1825 -0.5493 -0.2817 -0.6984
0.0248 0.0403 0.0226 0.0571
1/2
2
Y
0.0063 -0.0492 -0.0039 -0.0375
0.0035 0.0222 0.0088 0.0321
2
2 / 1 pl
0.3579 0.4765 0.5222 0.2998
0.1842 0.1254 0.1344 0.1137
2
2 / 1 pk
0.0850 0.0492 -0.0159 0.0265
0.0883 0.0696 0.0711 0.0783
pl*pk 0.2947 0.0204 0.0455 0.1492
0.0975 0.0647 0.0687 0.0578
Y*pl 0.0808 0.0316 0.0439 0.0245
0.0152 0.0121 0.0127 0.0120
Y*pk 0.0581 -0.0334 -0.0020 -0.0200
0.0134 0.0162 0.0141 0.0184
Dum96 -0.0265 -0.0082 -0.0250 -0.0192
0.0062 0.0068 0.0064 0.0090
Mat 0.2295 0.2608 0.2372 0.3145
0.0156 0.0129 0.0129 0.0143
Bam -0.0079 0.0698 0.0650 0.0734
0.016 0.0169 0.0169 0.0233
Da 0.0016 0.0006 -0.0002 0.0046
0.0015 0.0012 0.0012 0.0018
Co -0.0048 -0.0086 -0.0091 0.0014
0.0049 0.0050 0.0051 0.0060
Loop -0.0124 -0.0022 -0.0080 0.0060
0.2295 0.0033 0.0032 0.0035

The price of labor (PL), the price of capital (PK) and the material's variable (Mat)
have the positive sign and the estimates appear to be significant at 5%. The
54
structural variable, which is dum96 has a negative and significant (10%) effect on
the cost per output. As in the parametric case, the adoption of the
telecommunication act in 1996 has led telephone companies to operate more
efficiently. Using the translog cost functions, I analyze nonparametrically the effect
of an increase of output on the cost per output. Figure 3.1 and figure 3.2 show the
nonparametrics representation of the scale economies. Figure 3.1 shows the kernel
regression method and figure 3.2 shows the natural cubic spline method. The two
graphs are almost similar. The vertical axis of figure 3.1 and 3.2 shows the average
cost. The horizontal axis shows the output. The two graphs show a decrease in the
cost per output for an increase in output. As the output increases the average cost
decreases. The downward sloping of the graph shows the decrease of the cost per
output as the output increases. This is the sign of the presence of scale economies.
The scale economies are larger for small firms compared to big firms. This result
may explain the wave of merging and consolidation we have experienced and
continue to experience in the telecommunication industry.
55

Table 3.3: Semiparametric Analysis of Total cost per output
from 1992 to 2001

.Semiparametric Analysis of Total costs per output from 1992 to 2001
I assume () Y ψ is the nonparametric part.
Full Translog Model
Variable Coefficient. Standard Error
Pl 0.2003 0.0275
Pk 0.0868 0.0224
2
2 / 1 pl
0.4604 0.1712
2
2 / 1 pk
0.1106 0.1212
pl*pk 0.2838 0.1324
Y*pl 0.0626 0.0181
Y*pk 0.0379 0.0181
Dum96 -0.0151 0.0058
Mat 0.2096 0.0208
Bam -0.0212 0.0165
Da 0.0006 0.0017
Co -0.0090 0.0046
Loop -0.0140 0.0034
RootMSE 0.0653
R-Square 0.3897

56

Figure 3.1: Kernel Regression

Figure 3.2: Natural Cubic Spline

57
3.6 Conclusion
The goal in this paper has been to estimate the scale economies in the U.S
Telecommunications industry under weak functional assumptions. The paper uses a
new panel data set that covers the period between 1992 and 2001. The
semiparametric model uses the translog cost function where output enters non-
parametrically. The output is the operating revenue. The remaining variables enter
parametrically. Our semiparametric estimation method is based upon Robinson's
techniques. I also use OLS, within (fixed effect), GLS and first difference IV to
estimate a parametric translog cost function. The data shows the presence of scale
economy when using a parametric cost function. As the output increases the
average cost decreases. This results also hold for the semiparametric specification
of the cost function where our output enters nonparametrically. The downward
sloping of the graph shows the decrease of the cost per output as the output
increases. This is the sign of presence of scale economies. The scale economies are
larger for small firms compared to big firms. This result may explain the wave of
merging and consolidation that is going on in the U.S telecommunications industry.

58

Chapter 4

Comparison of Regulatory Regimes in
Telecommunications Market: An
Empirical Analysis.

4.1 Introduction
Government regulation of economic activity has undergone dramatic change during
the past quarter century. This has been marked by regulatory retrenchment in the
United States and by large-scale privatization of government-owned enterprises in
Africa and other developing countries. Reforms have ranged from dismantling the
regulatory apparatus in sectors such as transportation and energy, to substantially
relaxing regulatory restrictions in industries such as financial services, and to large-
scale restructuring of markets and regulatory mechanisms as in telecommunications
and electric utility industries around the world (Gasmi et al 2002). The
telecommunications industry is an industry in transition. The long-distance sector
of the industry has been successfully opened to competition in many countries but
the local exchange portion of the industry has seen relatively little actual entry. As
the long distance telecommunications systems have been restructured to promote
competition, the regulation of remaining local monopoly segments and the terms
under which competitors can interconnect with these monopoly providers continue
59
to raise a variety of new, complex issues that have been difficult to be solved by
theory and traditional empirical analysis alone. Schmalensee (1989) has presented
some simulation results in which cost and disutility of effort functions are given
specific functional forms. His objective was to compare analytically the
performance of linear regulatory mechanisms, which include cost-plus and price-
cap regulation as extreme cases. Through his simulation studies he reaches the
conclusion that, when there is no uncertainty, price-cap regimes are often optimal.
Building on the simple model proposed by Schmalensee (1989), Gasmi, Ivaldi and
Laffont (1994) used simulation techniques to analyze and compare various
regulatory schemes. Their analysis focused essentially on the measurement of the
trade-off between rent extraction and incentive for efficiency. Econometrics has
also contributed to the debate on regulation of public utilities by producing a set of
tools for evaluating economies of scale and scope. Various methodological
attempts have been made to use firm-level data to estimate production and cost
functions. The translog cost function is one of the most used cost functions. It has
been popularized by Christensen et al (1973). Its degree of flexibility makes it one
of the most favored specifications by economists. Some problems arose with theses
methods regarding the controlling of the effect of technological change when
measuring economies of scale. The lack of data was also a serious problem. Shin,
Ying (1992) and Wilson, Zhou (2001) circumvented some of the data problems by
constructing a large panel data set which they used to estimate a translog cost of the
local exchange industry function. They examined the subadditivity of local
60
exchange carriers (LECs). Wilson and Zhou (2001) found that estimates of scale
economies are dramatically affected by the treatment effect of unobserved firm
heterogeneity. But still none of these data types proves to be completely suitable
for proper policy analysis. Time series data on a representative firm are inherently
retrospective and data on a cross section of firms raise a different type of policy
problem due to the heterogeneity of the sample ( Gasmi, Kennet, Laffont and
Sharkey (2002)). These approaches also do not take into account the effect of the
regulatory environment. We should have an interaction between cost and regulation
because according to the new theory of regulation, costs are affected by regulation.
Gabel and Kennet (1994) point out additional unobserved influences on costs that
may bias the results of a traditional specification of cost using past data. We can
see that one common problem pointed out by those works is the problem of data
and the unobserved heterogeneity in the telecommunication industry. Applied
econometrics drawn heavily on past data in an industry with such a high speed of
evolution in technology might not be appropriate.
The work by Gasmi et al (2002) departs from the above works by addressing some
of the most significant challenges facing telecommunications regulators and policy
makers with an innovative approach that combines an engineering cost model of
local telecommunications networks and the modern regulatory theory. Cost proxy
models based on the computer modeling of local telephone network may be more
suitable for policy advice and constitute a powerful tool for the empirical analysis
of regulation. Our article tries to push forward this line of research by incorporating
61
into the analysis different disutility of effort functions, different cost functions to fit
the data and an alternative mapping for the arguments (effort and efficiency) in the
regulation cost function. We mirror the work of Gasmi et al (2002) by making use
of a data set from computer model of the cost of local telephone network
(LECOM). We consider a local exchange carrier industry that possesses the
characteristics of a monopoly. We compare different regulatory mechanisms (in
term of expected welfare) applied to a telecommunication firm.
Section 4.2 gives a description of the different telecommunication Cost proxy
models. The section 4.3 covers the different regulatory regimes. Section 4.4 gives a
detail of the model. Section 4.5 contains the results of the optimal regulation, the
stability of the optimal regulatory scheme by using alternative disutility of effort,
cost functions and alternative mapping for effort and efficiency. Section 4.6
concludes and section 4.7 covers the appendices.

4.2 Telecommunication Cost Proxy Model.
In recent years, engineering process models have been developed as an alternative
to more traditional econometrics and accounting approaches to cost measurement.
The procedure is to examine in great detail the engineering production process in
order to uncover the main properties of its cost structure. The engineering process
models offer a more detailed view of cost structures than using econometric with
past data. Besides, engineering models are better suited for modeling forward-
62
looking (long run) costs since they rely much less on historical data. This process
has been used in many industries such as electric power generation. We can cite the
work of Smith (1957) in trucking. Chenery (1949) discussed the general
methodology of the engineering production function. This first class of models has
been termed in the literature as ``process models''. The second class of engineering
optimization models closed in general spirit embeds a full-blown optimization of
the process being studies that the first class of models does not have. These second
classes of models are written in high-level computer language. Cost proxy model
(forward-looking economic cost) is used by FCC as a basis for determining
universal service support levels, cost-based access charges, pricing for
interconnection and unbundled network elements. Forward-looking economic
computer based cost models help regulatory authorities estimate the forward-
looking cost of network facilities and services without having to rely on detailed
cost studies done by incumbent local exchange carriers. A cost proxy model
conceptually consists of a set of more or less detailed descriptions of the
technological processes underlying the cost function of a representative firm in the
industry. We can outline a few of them.
The first one is the Interconnection Cost proxy model. This model originally
developed for the European Commission is now used to set interconnection charges
in several European countries, including Austria, France, Denmark and the Czech
Republic. The HAI model was developed by both AT&T and MCI. The
63
Benchmark Cost proxy Model (BCPM) was sponsored by US West, Sprint and Bell
South.
The second model is the Hybrid Cost Proxy Model (HCPM). This model has been
developed by the FCC. It has been built from both the HAI and the BCPM. A
complete description of HCPM can be found in Bush, Kennet et al (1999). HCPM
consists of several independent modules. The first module is a clustering algorithm
that groups customer locations into neighborhood serving areas. The second
module is a cluster interface that computes the area and line density of each cluster
and assigns individual locations to cells in a grid structure. The last module is a
loop design that uses network design algorithms to connect the grid cells in each
serving area to a central office switch. The HCPM also includes a modules that can
be used to compute the cost of switching and interoffice transport. The different
modules of HCPM are written in high level compiled programming language
(Visual Basic). The computer program of HCPM reads relevant input data, perform
the calculations relevant to a portion of the local network, and print output reports
for use in succeeding modules.
The last model is the local Exchange Cost Optimization Model (LECOM). It was
developed by Gabel and Kennet (1991). It has almost the same features as HCPM
but less complex. The model has been written in Pascal-Turbo. The role of
LECOM is to generate local exchange cost data by examining in great detail the
engineering production process. A LECOM city consists of three regions of
varying population density: a central business district, a mixed commercial and
64
residential district, and a residential or rural district. Serving areas in LECOM
always have a uniform population distribution within them, and they are always
rectangular in shape. Feeder plants run from the serving area interface (SAI) to the
central office. A description of the cost optimization is done in the model.

4.3 Regulatory Regimes
This part describes the different regulatory mechanisms. We give a finer
classification of traditional and incentive regulation regimes. We can divide them in
three groups. The first group is called the traditional regulation, the second is the
incentive regulation and the third is structural regulation. The traditional regulation
is composed of Cost-plus regulation (with reimbursement of observable costs and
no additional transfers). The incentive regulation is composed Laffont-Tirole's
model (LT), Barron-Myerson's model (BM) and price cap's model (PC). Laffont-
Tirole's model is an optimal regulation with incomplete information, cost
observability and transfert. Barron-Myerson's model is a regulatory regime with
incomplete information, no cost observability and with transfers. Price-cap
regulation (PC) is an incentive regulation with no cost observability and no
transfers. We also have price-cap regulation with sharing of earnings with cost
observability and one-way transfers (PCT). Since the telecommunication act in
1996, the U.S. telecommunications market is heading toward a structural
regulation. The U.S. Telecommunications industry is moving toward a Structure
65
Regulation after implement the incentive regulation in a long period. The goal of
structure regulation is to introduce changes in the organizational structure of an
industry so that it approximates more closely the structure need for competitive
behavior.

4.3.1 Traditional regulation
A / Cost-plus regulation
This regulation is coupled with reimbursement of observable costs and no
additional transfers. Under this scheme, the firm that is fully reimbursed for it costs
has no incentive to minimize them. Effort is minimal and efficiency low. In this
case, the regulator is then able to eliminate information rents. The firm utility is
given by () e ψ − and the profit-maximizing firm can be assumed to choose the
minimum level of effort () 0
min
= e , which gives zero disutility. The regulator in
this case imposes a production level () β q that solves
() () 0 , 0 , ) ( )) ( ( = − β β β β q C q q P
The firm is required to produce () β q that maximizes expected social welfare.
() () () () []() β β β β β
β
β
d f q C q S
∫
−
− , 0,

66

B / Rate of return regulation (RORR)
The rate of return regulation can be viewed as a form of cost-plus regulation
because prices tend to be set to allow the regulated firm to recover its realized
operating costs plus a fair return on investment. Rate of return regulation is often
popular when operating costs are rising because it provides the regulated firm with
a convenient vehicle for securing higher prices to cover unavoidable cost increases
(AI, Sappington: 2002). Full knowledge is assumed about the costs. In practice,
prices often are determined in a more stepwise fashion. More precisely we can
divide the process into three basic steps: (1) the firm's costs are reviewed, and cost,
which are supposed unnecessary, are eliminated. (2) A rate-of -return judged to be
fair for the firm is specified. (3) Prices and their structure are set to generate enough
revenue to cover costs and provide a fair rate of return.
Let’s write the profit of a regulated firm as

T d rK wL R − − + − = π

where L is a variable input such as labor and K is the capital. The prices of L and K
are w and r, respectively, and the firm takes theses prices as given. R is the revenue.
d and T are respectively the depreciation of the capital stock and the firm's tax bill.
The firm's objective will be to maximize equation 4 with a constraint imposed by
the regulator, which is
67
s
D K
T d wL R
≤
−
− − −

where D is the sum of all past period's depreciation. The left-hand side is the rate of
return, and s is the allowed rate of return specified by the regulator. The numerator
is the net operating income and the denominator is the firm's rate base. This
constraint required the firm's rate of return to be no greater than a specified allowed
rate of return
15
.

4.3.2 Incentive regulation
Incentive regulation regime have been implemented to foster network
modernization, reduce operating costs, lower prices, streamline the regulatory
process, and enable regulated suppliers to cope with growing competitive pressures
(AI, Sappington 2002). We can distinguish four main incentive regulatory regimes:
Price cap, Price cap plus transfer, Laffont Tirole model and Baron Myerson model.

15
Many problems occur with the rate of return regulation. The first one is the allowable cost. The
firm may have an incentive to exaggerate its cost by claiming that operating cost (wL) are greater
than what they actually are. The firm can also incur costs that are not necessarily in the best interest
of consumers (goodwill advertisements and not providing information). The second problem is
related to the lack of incentive to reduce cost. Because the firm is not residual claimant for it cost, it
has no incentive to reduce its operating cost as long as those costs can be passed on to the consumer.
The last problem encountered when using a RORR is related to the Rate-base determination and the
allowed return through the difficulty in measuring the firm's rate base.
68

A / Price-cap
Price-cap (pure) regulation is a regulation without cost observability and without
transfer. The firm is residual claimant for its cost. That means the firm bears the
full financial consequences of its performance in the market place during the
specified period of the plan. We tried to give a detail of this regulation scheme
following Gasmi, Kennet Laffont and Sharkey (2002). Let's write the profit
maximization of the firm's problem as

() () ()() e p q e C p pq
p e
ψ β − − , , max
,

where q(p) is the demand function and () β
M
p the profit-maximizing price
associated with the optimal effort level.
The first-order conditions gave us

() ( ) ()() e p q e C e
e
ψ β ψ − − = , ,
'

and
() ()
()
() dp p dq
p q
p q e C p
q
/
, , − = − β or
η
1
=
−
p
C p
q

69
A regulatory body solves
() () () () []() β β β β β
β
β
d f q C q S
p
∫
−
− , 0 , ( max
() () () () () () () () () []








− −
∫
*
*
, , , max
β
β
ψ β β β β β
M M M
p
p q p q e C p q S

under the participation constraint of the least efficient firm,

() () () () ( ) () () ( ) 0 , , , ,
* *
≥ − − p q e p q p q e C p q p β ψ β β

For 0 <
qq
C , the constraint above is binding. We can solve for p from the
participation constraint to obtain the optimal price-cap ( ) p p
M
=
*
β .

B / Laffont-Tirole's model (1986): Cost ex-post observable
Let's consider a single-product regulated firm with aggregate cost function

() q e C C , , β =

where β is a technological parameter or firm's type, with 0 >
β
C , e is its
managers' cost-reducing effort, with 0 <
e
C , q is the firm's output level, with
70
0 >
q
C . () q C E , , β is the effort required for a firm of typeβ to produce q at cost C.
In Laffont-Tirole model, the cost is ex post observable. For accounting matter, they
suppose that the cost is reimbursed by the regulator to the firm and the revenue
generated by the sale of outputs is received by the regulator.
The regulatory program is given by:

() ()( ) ()() ()() ()() [] () {}() β β β λ β β ψ β β β λ β
β
β
d f U q e q e C q V
p
∫
− + + − , , 1 ( max

under the participation constraint and incentive compatibility constraints given by

() 0 ≥ β U for all β in [ ] β β,
and
() () ( ) () () ()() () β β β β β β ψ β
β
q q e C E e U , , , ,
'
− =

The rent is
() () ( ) () () ()() () db b q b q b e b C b E b e U
∫
=
β
β
β
ψ β , , . ,
'
.
In this model we have two instruments to decrease the information rent. The first
instrument is related to the use of effort. By decreasing () β e we can decrease
71
() () b e
'
ψ . The second instrument involves
β
E by using more sophisticated
distortions involving () β q and therefore pricing when
β
E is a function of () ⋅ q .

C / Baron-Myerson (BM 1992): Cost not observable ex-post
In their model, BM derived the optimal solutions when aggregate costs of the
regulated firm are not observable ex-post. The cost is not observable and therefore
cannot be reimbursed. The regulator's program
16
becomes:

() ()
() ()( ) ()() () () ()() [] () {}() β β β λ β β ψ β β β λ β
β
β
d f U q e q e C q V
U q
∫
− + + −
⋅ ⋅
* *
,
, , 1 ( max

subject to
() 0 ≥ β U for all β in [ ] β β, or () 0 = β U

and

() () ()() ( ) β β β β β q q e C U , , ,
*
− =

16
The first order condition is () ( )
()
()
q q
C
f
F
C q V
β
β
β
λ λ + + = 1
'
or for linear pricing
()
() p
C
f
F
p
C p
q q β
β
β
λ
λ
η λ
λ
+
+
+
=
−
1
1
1
.
72
The rent is
() () ()() () db b q b q b e b C U
∫
=
β
β
β
β , , ,
*
.

In this model, without cost observability we have lost one instrument and the
distortion of the pricing rule away from Ramsey pricing is now the only way to
decrease the information rent. From the rent equation we can see that only a
distortion of ) (⋅ q can affect this information rent. In the case of Laffont-Tirole's
model we have two instruments to decrease the information rent.

() () () ()() ()() ()
∫
=
β
β
β
ψ β db b q b q b e b C b E b e U , , , , '

The first instrument is related to the use of effort. By decreasing () β e we can
decrease () () b e ' ψ . The second instrument involves
β
E . By using more
sophisticated distortions involving also () β q and therefore pricing when
β
E is a
function of () ⋅ q . We can see that by using the observability of cost, Laffont-Tirole's
model have more instruments to decrease the information rent compared to BM
model. So by using the observability of cost, Laffont-Tirole's model has more
instruments to decrease the information rent compared to BM model.

73
4.4 Model
This section describes the engineering cost proxy function. We start with a simple
description of the engineering model and describe the inner working of the model
algorithms. Then we show how to use the data generated from the engineering
model into the different regulatory regimes.

4.4.1 Step one: Cost proxy model
The total cost minimization procedure performed by LECOM is written as:

() () S y x FC S y x DC C
d d
S y x
r s f d
, , , , , , min
, , , , , ,
τ τ
τ τ τ τ
+ = (4.1)
()( ) S y x TC S y x SC
d d
, , , , , , τ τ + +

Where C is total cost composed of distribution cost (DC), feeder cost (FC),
switching cost (SC), and trunk cost (TC) .The set of control variables comprises
f f d d
T T ∈ ∈ τ τ , , and
r r
T ∈ τ which are vectors of technologies available for,
respectively, distribution, feeder, switching, and interoffice trunk, S, which is the
number of switches employed, and x and y which are S-dimensional vectors
representing the horizontal and vertical coordinates of the switches in the map
17
. To

17

d
T consists of only copper wire.
f
T consists of analog large switches, analog small switches,
digital hosts, digital remotes, and digital stand-alone switches.
r
T consists of only one type of
digital interoffice trunk.
74
illustrate the inner working of the LECOM algorithms, consider the problem of
simply optimizing over the number of switches.

()() ()
()()
()()
L M L K L M L K L
T M L K T M L K T
M L K CCS M L K S
D L L P P P VC L P P P FC
D R CCS R P P P VC S CCS R P P P FC
CCS S P P P V CCS S CCS S P P P FC S C
⋅ ⋅ + +
⋅ ⋅ + ⋅ +
⋅ + ⋅ =
; , , ; , ,
, ; , , , ; , ,
, ; , , , ; , ,
(4.2)

Where C(S) is the total cost expressed only as a function of S, the number of
switches.
L
D is the average loop length (depends on S). R is the number of
interoffice trunks (depends on S).
L
FC is the fixed cost of loops for a given city
size (exogenous cost parameter).
L
VC is the variable cost per unit of distance of
loop plant (exogenous cost parameter).
S
VC is the variable switching cost per
hundred busy-hour calling seconds (exogenous cost parameter).
T
VC is the variable
cost per unit of distance of trunk plant (exogenous cost parameter). L is the number
of loops (exogenous demand parameter). B represents hundreds of busy-hour
calling seconds (exogenous demand parameter)
75

4.4.2 Step 2: Approximate of the engineering cost function

We generate some data from the cost developed in step one. The procedure for
generating the data is described in appendix.
In order to evaluate the different regulatory regimes we need to compute the cost
function that appears in the maximization problem
18
.
The best solution is to use directly the data generated from the engineering model
in the different regulation models. But because of tractability issues, we fit the data
into some different econometrics cost functions.

() Q PL PK C C , , = (4.3)
() Q PL PK C , , can take different forms(parametric, semiparametric or
nonparametric). We then construct a mapping between the arguments in the
econometrics cost function and the arguments in the regulatory cost function.
() q e C C , , β = (4.4)

18
The cost function is a critical component of the optimal regulatory mechanism. In the case of
Laffont-Tirole's model below, we can see that even if it's possible, it will be hard to directly put the
engineering cost in the regulator's problem. For pratical reasons we have to fit those data in some
smooth cost functions under some assumptions on the new approximate cost function which is
() ()
() ( ) ( ) () () ()() ()() []
∫
+ + −
⋅ ⋅
β
β
β β ψ β β β λ β q e q e C q V
e q
, , 1 { ( max
,

+
()
()
() ()
() ( ) ()
() () ()
() β β
β β β
β β β
β ψ
β
β
λ
β
d f
q e C
q e C
e
f
F
e
}
, ,
, ,
'
76

4.5 Result of the Optimal Regulatory Mechanism
The first part of this section will make use of the price of labor and price of capital
as respectively proxies for effort and type. The second part will use alternative
mapping for the type () β and effort () e from the literature of software cost
estimation models.

4.5.1 Impact of the choice of the cost function and disutility
function on the regulatory regime.

In this section we analyze the effect of the choice of a particular cost function on
the rank of the regulatory schemes, the welfare losses recovered, and the gain
associated with observability of cost and the regulator's ability to use transfer. We
rank in table 4.1 the different regulatory regimes. The complete information (CI)
has the highest expected welfare, followed by Laffont-Tirole's model (LT), Baron-
Mayerson's model (BM), price cap plus transfer (PCT) and price cap (PC). Cost
plus (C+) has the lowest expected welfare. The Cost-plus transfer (C+T) has an
expected welfare greater than the price cap plus transfer and lower than BM
regime.
77

Table 4.1: Expected value of social welfare

Consumers' welfare, firm's rent and rate of return (%)

W E
β
CW E
β
U E
β
ROR E
β

CI 630.19 630.19 0.00 0.00
LT 628.52 623.64 4.88 0.16
BM 628.32 622.07 6.24 0.21
PCT 616.96 612.14 6.46 0.21
PC 616.58 610.12 4.83 0.15
C+T 624.75 624.75 0.00 0.00
C+ 612.13 612.13 0.00 0.00

Table 4.2 and table 4.3 compare the incentive and the traditional regulation using
all the cost function. Except for the modified Leontief cost function, other cost
functions do not have any effect on the rank of the regulatory mechanism. When
using a modified Leontief cost function, we have a higher expected welfare in the
BM model compared to the LT model.

Table 4.2: Percentage Losses recovered / between Incentive
versus Traditional Regulation

(Quadratic disutility function) /C+ (Traditional regulation)
Incentive FthCost Cubic Tslog Homoth
Regulation Cost+ Cost+ Cost+ Cost+
PC 23.15 22.99 24.62 24.22
PCT 25.45 25.29 26.76 26.40
BM 89.36 89.45 89.63 89.50
LT 90.72 90.80 90.76 90.67

78

Table 4.2: Continued

(Quadratic disutility function) /C+ (Traditional regulation)
Incentive Quad Poly4 GMF GLeont Average
Regulation Cost+ Cost+ Cost+ Cost+ Cost+
PC 23.51 25.15 18.51 9.81 22.23
PCT 25.83 27.02 20.54 15.09 24.67
BM 89.13 90.80 90.26 90.54 89.99
LT 90.48 91.04 90.72 84.86 90.18

Table 4.3 Percentage losses recovered between Incentive versus traditional regulation

(Exponential disutility function)
Incentive FthCost Cubic Tslog Homoth
Regulation Cost+ Cost+ Cost+ Cost+
PC 33.80 33.04 35.02 34.76
PCT 39.30 39.12 40.22 40.06
BM 91.34 91.41 91.55 91.46
LT 96.05 96.08 96.03 96.01

Table 4.3: Continued

(Exponential disutility function)
Incentive Quad Poly4 GMF GLeont Average
Regulation Cost+ Cost+ Cost+ Cost+ Cost+
PC 37.60 34.80 31.25 27.62 33.56
PCT 39.85 39.86 35.79 31.71 38.24
BM 91.18 92.04 92.18 92.54 91.71
LT 95.96 95.99 96.08 96.02 96.03

We realize that more than 33 percent (91 percent for LT and BM) of the welfare
loss associated with traditional regulation can be recovered by switching to
incentive regulation under the exponential disutility function assumption. This is
true no matter which cost function you choose to fit the data with. The need to
create incentives for cost minimization has been one of the main driving forces
79
behind recent regulatory reforms in developed (Gasmi and al 2002) and developing
countries. Let's measure the gains associated with observability of cost. LT and BM
may be interpreted as an indication of the value of good auditing procedures. We
assume that accounting manipulations can be detected. Using our table 4.4 and
table 4.5 below, we notice that the regulator can recover on average 15.98 percent
of the loss associated with the BM regulatory mechanism. This number increases
when we use an exponential disutility function.

Table 4.4: Percentage losses recovered between LT(cost observed) and
BM(cost not observed), Quadratic Disutility Function

Quadratic disutility function / BM (Non auditing)
FthCost Cubic Tslog Homoth
Auditing (LT) BM BM BM BM
LT 12.67 12.71 10.94 11.18

Table 4.4: Continued

Quadratic disutility function / BM (Non auditing)
Quad Poly4 GMF GLeont Average
Auditing (LT) BM BM BM BM BM
Auditing (LT) 12.39 25.75 4.71 37.48 15.98

Table 4.5: Percentage losses recovered between LT(cost observed) and
BM(cost not observed), Exponential Disutility Function

Exponential disutility function / BM (Non auditing)
FthCost Cubic Tslog Homoth
Auditing (LT) BM BM BM BM
LT 54.37 54.35 52.93 53.26

80

Table 4.5: Continued

Exponential disutility function / BM (Non auditing)
Quad Poly4 GMF GLeont Average
Auditing (LT) BM BM BM BM BM
Auditing (LT) 54.38 49.69 49.87 46.61 51.93

Table 4.6 and table 4.7 show the gap in percentage of welfare loss recovered
between PC, LT model and BM model. Table 4.6 uses a quadratic disutility
function and table 4.7 uses an exponential disutility model.
The big gap in percentage of welfare loss recovered between PC (without transfer),
LT and BM (with transfer) shows the consequences of constraining the regulator
from using direct transfers to the firm.
19

Table 4.6: Percentage losses recovered Between PC and LT or BM,
Quadratic disutility function

Quadratic disutility function / PC Regulation Without transfers
Regulation FthCost Cubic Tslog Homoth
With transfers PC PC PC PC
LT 87.92 88.04 87.75 87.70
BM 86.16 86.30 86.24 86.15

Table 4.6: Continued
Quadratic disutility function / PC Regulation Without transfers
Regulation Quad Poly4 GMF GLeont Average
With transfers PC PC PC PC PC
LT 87.55 88.03 88.61 83.22 87.35
BM 85.79 87.71 88.05 89.51 86.99

19
The constraint is usually imposed for political reasons or by fear of capture (Gasmi and al 2002).
81

Table 4.7: Percentage losses recovered Between PC and LT or BM
Exponential disutility function

Exponential disutility function / PC Regulation Without transfers
Regulation FthCost Cubic Tslog Homoth
With transfers PC PC PC PC
LT 94.03 94.09 93.88 93.88
BM 86.92 87.06 87.00 86.92

Table 4.7: Continued
Exponential disutility function / PC Regulation Without transfers
Regulation Quad Poly4 GMF GLeont Average
With transfers PC PC PC PC PC
LT 93.87 93.86 94.30 94.50 94.05
BM 86.57 87.79 88.63 89.69 87.57

4.6 Alternative Mapping for the Type() β and
Effort() e .

Gasmi et al (2002) chose an unintuitive choice of the variables PK and PL to
represent technology and effort. They view technology as something embodied in
the capital stock of a representative firm and effort as a function of the labor input,
which they interpret as the efficiency price of labor. The use of the price of labor
and capital respectively as proxy for effort and efficiency are not entirely
compelling and clearly call for more work. We need to find another mapping from
the arguments in the regulation cost function,β , e to those in the engineering cost
function
L K
P P,.
Many model generation processes tried to formulate the equation of effort that
allows one to estimate the cost, effort, and schedule when planning a new software
82
development activity. I will focus on the recent one, which is the Constructive Cost
model (COCOMOII). COCOMO II was developed by the computer science
department at the university of southern california. It’s a model that allows one to
estimate the cost, effort, and schedule when planning a new software development
activity. It's a software cost estimation model.
Effort is modeled as a function of size and a set of effort multipliers, which account
for differences in hardware constraints, personnel quality and experience, use of
modern tools and techniques, and other significant parameters. This model uses a
set of 17 multiplicative effort multipliers and a set of 5 exponential scale factors to
adjust for project, platform, personnel, and product characteristic Chulani, Boehm
and Steece (1998).

[]
∏
=
× × =
17
1 i
i
B
EM S A Effort
where
∑
=
× + =
5
1
01 . 0 01 . 1
j
j
SF β

A is a multiplicative constant. It is used to calibrate the model locally for a better fit
and it captures the linear effects of effort in projects of increasing size. S is the size
of the software project measured in terms of KSLOC (thousands of Source Lines of
Code, Function Point, or Object Points). EM is effort multiplier. SF is scale factor.
We can build a similarity between the above work and our cost function.
83
In our case we can write the effort level as:

[]
∏
=
× × =
k
i
i
B
EM Output A e
1

For simplification purposes, we will suppose A = 1, B = 1 and the effort
multiplicator will be the inverse of the price of labor (pl). We end up with:

PL
Output e
1
× = (4.5)

The sizes, which is in our case the output and price of labor are very good
predictors of effort.
Regarding the proxy for technology, let's mirror the methodology above. The type
of a firm can be written as a function of the number of lines per consumer but also
the price of capital. A firm with a high number of lines per consumer is supposed to
be a high type firm compared to a firm with low number of lines per consumer.
β is a function of number of line per consumer and price of capital.
We can write β as:

PK Output× = β (4.6)

where output is the number of line per consumer.
84
So equation 5 and equation 6 represent the variables for effort and technology.
Using the alternative mapping for the type and effort we evaluate the impact of the
choice of the cost function and disutility function on the regulatory regime.

4.7 Conclusion
We used a local exchange cost proxy model (LECOM: computer model) to
generate local exchange cost data by examining in great detail the engineering
production process. Those data are then synthesized through standard statistical
estimation techniques. When moving from a disutility function of the quadratic
form to an exponential form, the rent decreases significantly. This result reveals the
sensitivity of the rent to the disutility function we choose. Using different cost
functions to fit the data do not change the rank of the different regulatory regimes.
The complete information case has the higher expected welfare followed by LT's
model, BM's model, PCT, PC, and Cost plus model. We also computed the welfare
losses recovered, and the gain associated with observability of cost and the
regulator's ability to use transfer. We realized that more than 33 percent (91 percent
for Laffont-Tirole's model and Baron-Myerson's model) of the welfare loss
associated with traditional regulation could be recovered by switching to incentive
regulation. This is true no matter which cost function you choose to fit the data
with. We measured the gains when using observability of cost associated with
Laffont-Tirole's model. This model may also be interpreted as an indication of the
85
value of good auditing procedure under the assumption that accounting
manipulations can be detected. Comparing LT and BM, we found that the regulator
can recover in average 15.98 percent of the loss associated with the BM regulatory
mechanism. This number increases when we use an exponential disutility function.
The big gap in percentage of welfare loss recovered between price cap (PC) and LT
or BM (with transfer) shows the consequences of constraining the regulator from
using direct transfers to the firm. Many extensions can be done here. The first
extension is to use a semiparametric and nonparametric model to fit the data. The
second extension is to be able to directly use the data generated by the engineering
model into the regulatory models. The third extension is to simulate a competition
market in the developing countries using the cost proxy mode
86

References

ACKERBERG AND CAVES (2003): “Structural Identification of Productions
function.” mimeo Department of Economics, UCLA

ARELLANO, M, AND S. R. BOND (1991): “Some Test of Specification for Panel
Data: Monte Carlo Evidence and Application to employment Equations,” Review of
Economic Studies.

BARON, D., AND R. MYERSON (1992): “Regulating a monopolist with
unknown costs,”. Econometrica, 50, 911-30.

BERESTEANU, A. (2004): “Nonparametric Analysis of Cost Complementarities
in the Telecommunication Industry,” forthcoming in Rand Journal of Economics.

BLUNDELL, R. W. AND S.R.BOND (1998): “Initial Conditions and Moment
restrictions in Dynamic Panel Data Models,” Journal of Econometrics.

(2000): “GMM Estimation with
Persistent Dynamic Panel Data: an Application to Production Functions,”
Econometrics Reviews.

BUSH, C.A., V. GUPTA, D. M. KENNET, J. PRISBEY, AND W.W.SHARKEY
(1999): “Computer modeling of the local telephone network,” Federal
Communications Commission Report, Washington, DC.

CHENERY, H.B. (1949): “Engineering production functions,” Quarterly journal of
Economic, 63: 507-31

CHRISTENSEN, L. R., D. W. JORGENSON (1973): “Transcendenta

CHULANI, BOEHM AND STEECE (1998): “Calibrating Software Cost Models
Using Bayesian Analysis,” submitted for the IEEE Transactions on Software
Engineering, Special Issue on Empirical Methods in Software Engineering.

CORREA, L. (2003): “Natural or Unatural Monopolies in UK
Telecommunications?” Working paper N0 501 Queen Mary University of London.

87
GABEL, D., AND D. M. KENNET (1994): “Economies of scope in the local
exchange market,” Journal of Regulatory Economics, 6: 381-98.

GASMI, D., M. KENNET, LAFFONT AND SHARKEY (2002): “Cost Proxy
Models and Telecommunications Policy” The MIT Press Cambridge,
Massachusetts London, England.

GASMI, F., M. IVALDI, AND J.J. LAFFONT (1994) “Rent extraction and
incentives for efficiency in recent regulatory proposals,” Journal of Regulatory
Economics, 6: 151-76.

GREEN W. AND CHRISTENSEN (1976): “Economies of scale in US electric
Power Generation” Journal of Political Economy, 84, 655-676

HSIAO, C. (2003): “Analysis of panel data,” Cambridge University press

LAFFONT, J.J. (1994): “The new economics of regulation ten years after,”
Econometrica, 62: 507-37

LAFFONT, J.J., AND J.TIROLE (1993): “A Theory of Incentives in Procurement
and Regulation”. Cambridge: MIT Press.

(1994): “Using cost information to regulate
firms,” Journal of Political Economics, 94:1-49.

(2000): “Competition in Telecommunications”.
Cambridge: MIT Press.

LEVINSOHN AND PETRIN (2002): “Estimating Production function using inputs
to Control for unobservable,” Working paper

NOLL M. A. (1991): “Introduction to Telephones and Telephone Systems,”. Artech
House Boston London.

OLLEY AND PAKES (1996): “The dynamics of Productivity in the
Telecommunications equipment Industry,” Econometrica

PAGAN A., U. AMAN (1999): “Nonparametric Econometrics,” Cambridge
University Press.

RALPH (2003): “Review of ``Cost Proxy Models and Telecommunications
Policy,” Review of Network Economics

88
ROBINSON P. M. (1988): “Root-N-Consistent Semiparametric Regression,”
Econometrica, 56, 931-954

ROLLER, L.H. (1990): “Proper quadratic cost functions with an application to the
Bell system” Review of Economics and Statistics

SCHMALENSEE, R. (1989): “Good regulatory regimes,” Rand Journal of
Economics, 20: 417-36.

SHIN, R.T., AND J. S. YING (1992): “Costly Gains to breaking Up: LECs and the
baby Bells,” Review of Economics and Statistics, Vol. 74.
(1992): “Unnatural monopolies in local telephone,”
Rand Journal of Economics, 23(2): 171-83.

SMITH, V. L. (1957) “Engineering data and statistical techniques in the analysis of
production and technological change: Fuel requirements in the trucking industry,”
Econometrica, 25: 281-301.

TOPKIS, D. (1998): “Supermodularity and Complementarity,” Princeton
University Press, Princeton

VELLTURO, C.A., BERNT, E.R., FRIEDLAENDER, A.F., CHIANG, J.S.W.,
SHOWATER, M.H., (1992):“deregulation, mergers, and cost saving in class I
railroads, 1974-86,” Journal of Economics and Management Strategy, 1, 339-369.

WILSON AND ZHOU (2001) “Telecommunications deregulation and
Subadditivity costs: Are local telephone monopolies unnatural?,” International
Journal of Industrial Organization, 19, 909-930

YATCHEW, A. J. (2000) “Scale economies in electricity distribution: A
semiparametric Analysis,” Journal of Applied Econometrics, 15: 187-210

ZELLNER, A. (1962): “An efficient Method for estimating Seemingly Unrelated
regression and tests for aggregation bias,” Journal of the American statistical
association, 57, 585-612
89
Appendices
Appendix A
Summary Local Exchange Carriers
Holding Company Carriers Years with valid Data
1 BS BellSouth Telecommunications, INC YR92-YR2001
2 Qwest Comm US West Communications Inc YR92-YR2001
3 SBC SouthWestern bell Telephone Co YR92-YR2001
4 SBC Pacific Bell YR92-YR2001
5 SBC Illinois Bell Tel YR92-YR2001
6 SBC Michigan Bell Tel Co.d/b/a YR92-YR2001
7 SBC Ohio Bell Tel. Co Ameritech Ohio YR92-YR2001
8 SBC Wisconsin Bell / Ameritech Wisconsin YR92-YR2001
9 SBC Indiana Bell Telephone Ameritech YR92-YR2001
10 SBC Southern New England Telephone Co YR92-YR2001
11 SBC Nevada Bell YR92-YR2001
12 Verizon Bell Atlantic-Newyork YR92-YR2001
13 Verizon Bell Atlantic-NewEngland Telephone YR92-YR2001
14 Verizon Bell Atlantic-New Jersey, Inc YR92-YR2001
15 Verizon Bell Atlantic Pennsylvania, INC YR92-YR2001
16 Verizon GTE California Inc YR92-YR2001
17 Verizon Bell Atlantic-Virginia, Inc YR92-YR2001
18 Verizon Bell Atlantic-Maryland, Inc YR92-YR2001
19 Verizon GTE Florida Inc YR92-YR2001
20 Verizon GTE Southwest Inc YR92-YR2001
21 Verizon GTE South Inc YR92-YR2001
22 Verizon GTE North YR92-YR2001
23 Verizon GTE North West Inc YR92-YR2001
24 Verizon Bell Atlantic-Washington, D,C Inc YR92-YR2001
25 Verizon GTE Hawaiian Telephone Co. Inc YR92-YR2001
26 Verizon Bell Atlantic West Virginia, Inc YR92-YR2001
27 Verizon Bell Atlantic-Delaware YR92-YR2001
28 Verizon GTE Midwest Inc YR93-YR2001
29 Verizon Puerto Rico telephone Co YR92-YR2001
30 Verizon Contel of the South, Inc YR93-YR2001
31 United Sprint-Florida YR92-YR2000
32 United Caroline Telephone & Telegraph Co YR92-YR2000
33 United Central telephone Co YR92-YR2000
34 United United Telephone Co of Ohio YR92-YR2000
35 United United telephone Co of pennsyvania YR92-YR2000
36 United United telephone – Southeast, Inc YR92-YR2000
37 United United telephone Co. of Indiana, Inc YR92-YR2000
90
Summary of Local Exchange Carriers: Continued
Holding Company Carriers Years with valid Data
38 United United telephone Co. of New Jersey. Inc YR92-YR2000
39 United United telephone Co of Texas YR92-YR2000
40 United Central Telephone Co of Virgina YR92-YR2000
41 United United telephone Co. of the Northwest YR94-YR2000
42 United Sprint Missouri, Inc YR92-YR2000
43 Citizens Comm Cincinnati Bell telephone Co YR92-YR2000
44 Citizens Comm Frontiere Telephone of Rochester, Inc YR92-YR2000
45 Citizens Comm Alltel Georgia Communications Corp YR95-YR2000
46 Citizens Comm Lincoln / Aliant Communications Co YR92-YR2000
47 Citizens Comm Alltel Pennsylvania, INC YR95-YR2000
48 Citizens Comm Western Reserve Telephone Co., The YR95-YR2000
49 Citizens Comm Commonweath telephone telephone Co. YR92-YR2000

91
Appendix B

This part shows the procedure for generating the Data.
First we tried to describe how the data were obtained from the Local Exchange
Cost Optimization Model (LECOM).
The ranges of values respectively for the capital, labor price multipliers (PK and
PK) and the level of telephone usage (output Q) to run LECOM is used first. Usage
or total traffic intensity is measured in centi-Callseconds (CCS). One CCS equals
100 calls seconds of traffic in one hour. Thus, 36 CCS of traffic per hour means
3600 seconds of traffic every hour. 36 CCS is also equal to one Erlang, named after
the Danish engineer and mathematician, A.K.Erlang. We generate cost data for a
firm covering an area of about 57 square miles and serving 108,000 subscribers.
We face two problems in this setting. The first problem is related to the effect of
the labor on the cost. Increasing the level of effort reduces cost, whereas and
increase in the multiplier of the price of labor PL increase cost. The second
problem is related to the cost function. We can rewrite the cost function as:
()( ) Q PL PL PK C Q PL PK C , , , ,
~
0
− + = ε
Where 50 . 1
0
= PL is the maximal value of the price of labor in the grid
and e PL PL − + = ε
0
. Effort e decreases the price of labor. The total data point is
1331.

Abstract (if available)

Abstract The U.S. telecommunications industry has been the subject of many studies since the divestiture of AT&T in 1984 and the introduction of competition into long-distance. In 1996, the US government introduced the telecommunication reform (1996 Telecommunications Act) that outlines the main rules for competition in the local telephone market. But until recently, the implementation of the 96-Telecommunications Act has been problematic and under investigation. This dissertation consists of three essays, each an application of econometrics and calibration techniques to the telecommunications industry. The first essay analyzes the subadditivity and cost complementarities issue in the U.S. telecommunications industry. I use a dynamic panel data approach to estimate the total cost function of the local exchange carriers. The estimated cost function is then used in the test for subadditivity and cost complementarities. I find that the Local exchange carriers' cost appears to be subadditive meaning that breaking up the local exchange carriers is not efficient in terms of cost saving. I also find that the cost generated by jointly producing local and toll calls is on average lower than not producing them jointly. This means that we have presence of cost complementarities. The wave of merging in the telecommunication industry may be explained by the presence of cost complementarities. The second essay estimates semiparametrically the scale economies in the telecommunication industry. The semiparametric model uses the translog cost function where output enters non-parametrically. The result shows the presence of scale economies in the local exchange carrier. The scale economies seem larger for small firms compared to big firms. This result may also explain the wave of merging and consolidation that is going on in the U.S telecommunications industry. The last essay compares the different regulatory models using an engineering cost proxy function.

Conceptually similar

PDF

Essays on competition and antitrust issues in the airline industry

PDF

Growth, trade and structural change in low income industrializing economies

PDF

Vertical integration and two-sided market pricing: evidence from the video game industry

PDF

Three essays on the identification and estimation of structural economic models

PDF

Behavioral approaches to industrial organization

PDF

CEO reputation: who benefits -- the firm and the CEO?

PDF

Does mandatory adoption of International Accounting Standards reduce the cost of equity capital?

PDF

Empirical essays on industrial organization

PDF

Maternal full-time employment and childhood obesity

PDF

Pathways of local economic development: tales of cities in the United States and South Korea

PDF

Structural transformation and globalization

PDF

Essays on competition for customer memberships

PDF

Essays in retail management

PDF

Measurement error in income and consumption data and economic mobility

PDF

Market failure mentality in Japanese industrial policy: case studies of robotics and aircraft industries

PDF

Structure and regulation of lymphoid tyrosine phosphatase (LYP) in autoimmune response

PDF

Soil structure interaction in poroelastic soils

PDF

Explaining the cross-sectional distribution of law of one price deviations

PDF

Essays on congestion, agglomeration, and urban spatial structure

PDF

Essays on interest rate determination in open economies

Asset Metadata

Creator Soro, Torna Omar (author)

Core Title Cost structure and regulation in the telecommunications industry

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 09/27/2006

Defense Date 04/27/2006

Publisher Los Angeles, California (original), University of Southern California (original), University of Southern California. Libraries (digital)

Tag cost complementarities,cost structure,OAI-PMH Harvest,subadditivity test

Language English

Advisor Ridder, Geert (committee chair), Han, Hill (committee member)

Creator Email soro@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m46

Unique identifier UC151059

Identifier etd-Soro-20060927 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-10561 (legacy record id),usctheses-m46 (legacy record id)

Legacy Identifier etd-Soro-20060927-0.pdf

Dmrecord 10561

Document Type Dissertation

Rights Soro, Torna Omar

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California