FTC Staff Comment Before the Federal Communications Commission Concerning Petition to Reclassify AT&T as a Non-Dominant Carrier

V940002
Before 	the 	ORIZEDFEDERAL COMMUNICATIONS cOMMISSI<OOMMISSION 	AUT"Washington, 	D.C. 	20554
In the Hatter of
Policy and Rules 	Concerning
Rates 	for 	Competitive COIIIIIlOn
carrier Services 	and Facilities
Authorizations 	Therefor
)
)
)
)
)
)
CC Docket No. 	79-252
C01IIIIlents 	of 	the 	Staff of 	the
Bureau of 	Econoaics 	of the
Federal Trade Commission
On October 	25, 	1993, 	the 	staff 	of 	the 	Bureau of 	Economics
filed 	the 	attached 	analysis 	with 	the 	Federal 	Communications
Commission 	in CC 	Docket 	No. 	93-197, 	In 	the matter of Revisions 	to
Price 	Cap Rules 	for 	AT&T. 	This 	analysis, 	conducted by Michael
Ward of 	the 	FTC's 	Bureau of 	Economics, 	assesses 	empirically the
existence 	of market 	power 	in 	the 	long distance 	telecommunications
industry.
In 	a 	separate matter, 	the 	FCC 	released on OCtober 	13, 	1993,
a 	Public 	Notice 	inviting 	interested parties 	to 	file 	comments 	on a
motion 	filed 	by AT&T 	requesting 	that 	the 	Federal 	Communications
Commission reclassify AT&T 	as 	a 	nondominant 	common carrier, 	CC
Docket 	79-252. 	This 	letter 	submits 	formally 	the 	staff 	analysis
filed 	previously to CC 	Docket 	No. 	93-197 	to CC 	Docket 	No. 	79-252.
We 	believe 	that 	this 	analysis 	addresses 	issues 	that 	may be
relevant 	to 	the 	latter docket. 	Any questions 	on 	this 	analysis
should be 	directed 	to 	Michael 	Ward 	at 	202-326-2096 	or 	to 	Timothy
Daniel, 	Assistant 	Director, 	Division of 	Economic 	Policy Analysis,
at 	202-326-3520.
ff!1!mitted.
Ronald S. 	Bond
Acting 	Director
Bureau of 	Economics
Federal 	Trade 	Commission

Market Power in Long Distance Telecommunications"
Michael R. Ward
Economist, Federal Trade Commission
September 16. 1993
ABSTRACT: 	AT&T's long distance prices are regulated in order to avoid supra-
competitive pricing. 	This rationale presumes that AT&T possesses sufficient market
power to raise prices substantially above marginal cost. 	This paper estimates firm
specific demand elasticities for AT&T and the other large long distance companies using
monthly data for 1988 through 1991 from five states. 	Lerner indices are calculated from
these elasticities and interpreted as price markups over marginal costs. 	An upper bound
for the piice markup for AT&T is estimated to be 18.6 % of price, which implies a
potential deadweight loss of less than J.26% of current long distance revenue. 	This is
likely to overestimate the current deadweight loss because competition has subsequently
increased. 	.
1l1e views expressed here are the author's and do not necessarily reflect those of the Federal Trade
CommIssion or any individual commissioner. 	I would like thank Tim Daniel, Jan Pappalardo and Steve
Parsons for comments and Dolly Howanh for research assistance.

Table of Contents
I. 	Introduction . 	,; 	.
U. 	The Long Distance TeJecommuniatioM Market 	3
A. 	Carrier A.ccess . . . . . .. 	4
B. 	Regulation of Te/tcommunications . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 	6
m. 	Empirical Methodology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 	8
A. 	Two-Ievt!l budgeting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 	8
1. 	Upper Levt!l DefNllUi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . " 	11
2. 	Lower Level DefNllUi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 	12
B. 	Econometric Issues 	. . . . . . . . . . .. 	13
J. 	1M Instrumental Variables Method 	. . . . . . . . . . . . . . . . . . . . . . .. 	14
2. 	Rtvt!rst Regressions 	16
C. 	The Lerner Inda 	16
IV. 	Data Description 	. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 	21
A. 	Upper Levt!l Estimation 	21
B. 	Lower Levt!1 Estimation 	22
J. 	DemmuJ Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 	22
2. 	Potential Measurement Errors in tM Long Distance Prices 	24
3. 	Instrumental Variables 	. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 	25
4. 	Some Tests of the Instrumental Variables . . . . . . . . . . . . . . . . . . . ., 	27
V. 	Demand Estimation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 	29
VI. 	Potential Deadweight Loss CaJcuJatioM 	. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 	31
VII. 	Condusion................................................. 	35
References 	. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 	36

I. 	Introduction
Price regulation is desirable only when its benefits outweigh its costs. These benefits include the
increase in consumer welfare and economic efficiency from reducing supra-competitive prices. while the
costs include direct administration costs as well as reductions in productivity arising from disincentives
directly caused by the regulation. 	For the telecommunications industry, there is growing evidence that
regulation has led to substantial productivity losses (Mathios and Rogers (1989,1990), Kaesmer and Kahn
(1990), Olley and Pakes (1992), Crandall (1991), Kwoka (1993), Ying and Shin (1993». 	While most
experts agree that competition has :::~reased in the long distance market, AT&T's prices are still
regulated. 	The evidence on the extent of residual market power in the industry is still unclear. 	Some
studies examining the structure (Egan and Wavennan (1991» 	and pricing behavior (Levin (1991) of the
long distance telecommunications industry conclude that competition already exists. Others interpret the
structural evidence differently (Selwyn, Cornell, Taschdjian and Woodbury (1991», or conclude titat full
implementation of fiber optic technology will render the industry a natural oligopoly that will support
supra-competitive prices (Huber, Kellogg, and Thorne (1993». 	This paper attempts to measure the
degree of market power in the small business and residential market for long distance telecommunications
by estimating the degree of substitutability between AT&T's and its rivals' service.
Before the 1984 divestiture, AT&T was thought to exert market power in two different ways.
First, it was feared that AT&T-controlled monopolies in local telephone service would subsidize otherwise
competitive long distance service, or that AT&T would discriminate against rival long distance carriers
by providing inferior access to end users (Brennan (1987». 	Second, it was feared that AT&T would
wield market power in the long distance market directly by charging supra-competitive prices. 	The first
threat did not materialize, 	as evidence strongly suggested that the subsidy flowed from the more
competitive long distance markets to the regulated monopoly local service markets (Temin and Peters

(1985a. 1985b). Kaserman. Mayo and Flynn (1990), Shin (1993». 	Nevertheless. AT&T's divesture of
its local servicp operations in 1984 was explicitly designed to remedy the subsidy and discrimination
complaints (Brennan (1987). 	As a result, it would seem that if the long distance industry could be shown
to be sufficiently competitive, there would be no compelling economic arguments for continued price
regulation of AT&T's long distance service.
This paper measures the degree of competition in the long distance telecommunications market
primarily through estimates of firm-specific demand elasticities. 	This demand elasticity indicates the
extent of a firm's loss in quantity demanded due to unilaterally raising prices - that is. the extent of the
firm's market power (Landes and Posner (1981». 	The reciprocal of the own-price elasticity, the Lerner
index, provides an estimate of the percentage price markup over marginal cost for an unconstrained,
profit maximizing firm. 	The applicability of the Lerner index to AT&T is discussed below. 	Finally,
estimates of this price-eost margin provide the basis for measuring the potential deadweight loss from
supra-eompetitive pricing.
The paper's general conclusions are as follows. 	The estimate of AT&T's short-run own-price
elasticity is -2.9, implying a short-run AT&T Lerner index of 0.337. 	Adjusting this index with an
estimate of the bias due to the use of a short-run. rather than long-run, elasticity estimate implies that
AT&T's price markup over long-run marginal cost is less than 18.6% of price. 	At this upper-bound
value, the potential deadweight loss from supra-eompetitive pricing was less than 1.26% of total revenues
for the 1988 to 1991 time period. 	Because the market likely has become even more competitive since
the period analyzed in this study, this estimate is likely to overstate the current loss. 	For example,
adjusting for AT&T's subsequent fall in market share decreases the upper-bound potential deadweight loss
estimate to 0.71 % of revenues. 	These estimates of deadweight loss are substantially less than estimates
of the efficiency gains due to past deregulatory actions.
2

ll. 	The Long Distance Telecommunications Market
The FCC's SpeciaJized Common Carrier decision in 1971 opened up the long distance market
to competition. AT&T's tough posture toward its new competitors led to many private antitrust lawsuits,
and culminated in the Justice Depanment's massive antitrust suit. 	The 1982 settlement of this suit
provided for the 1984 divestiture of AT&T's long distance operations from newly created regionaJ local
telephone companies. 	As pan of the settlement, the divested local telephone companies were obligated
to install switching equipment that aJlowed for "equaJ access" by any long distance company. 	This
allowed AT&T's competitors to introduce services that were comparable to AT&T's. 	While AT&T's
share of the long distance market's revenues in 1982 was 95%, by 1987 its market share had faJlen to
80%, and it is currently about 60% (Kwoka (1993), Statistics of Communications Common Carriers
(1992». 	By 1991, MCl's and Sprint's revenue market shares had climbed to 17% and 10% respectively,
and the two next largest firms, WilTel and Cable & Wireless, had market shares of somewhat less than
1% each. 	Since the divestiture, industry output, measured in the number of calling minutes, has nearly
tripled. 	Even though AT&T's share has declined to 60%, its output has increased by two-thirds over
1984 levels.
A long distance company operates a communications netWork that COMects 10caJ telephone
exchanges (hence, it is often caJled an Interexchange Carrier, or "IXC"). 	A long distance company's
network terminates in different local telephone companies' jurisdictions. The local telephone companies,
such as the regionaJ Bell Operating Companies and GTE, transpon telephone calls between the customers'
premises and the long distance network. 	These services are caJled "carrier access," and they currently
represent nearly 40% of aJllong distance costs. 	Because these carrier access costs play an important role
in the empiricaJ anaJysis carried out below. it is wonhwhile to describe them in some detail.
3

A. 	Carri~r Acc~ss
Almost aJl carrier access rates are regulated by the F( 	Jr state regulatory commissions. 	For
standard toll 	service, 	long 	distance companies 	purchase 	II switched 	access II 	from 	local 	telephone
companies. 	Switched access prices are divided into three main components: carrier common line, !(lcal
switching, and local transpon. 	Carrier common line charges are levied to cover that ponion of the local
telephone distribution plant assigned to the long distance companies for capital recovery. Local switching
charges are levied because a long distance telephone call must be switched through the local netWork, thus
tying up switching capacity that has alternative uses. 	Local transport charges are levied as a rental of the
Iine between the long distance netWork and the relevant local switch. All three of these charges are levied
per minute of use at each end of the telephone call using switched access.
The carrier common line charge is levied specifically to defray the costs of the local telephone
distribution plant, and not the costs of completing long distance calls (hence, it is sometimes called the
"non-traffic sensitive" charge), and has a long and tortured history. 	Half a century ago, the courts ruled
that, because AT&Ts long distance service used local exchange netWork loops, a ponion of the cost of
these local network loops should be recovered through long distance rates. The ponion of the local loop
assigned to long distance service steadily grew to 27% in 1982 with little relation to underlying economic
costs. 1 The FCC's 1983 Access Charge Plan formalized this cost assignment as the carrier common line
charge, which was to be levied as a per minute charge despite its fixed cost nature. 	At first, the non-
AT&T long distance companies (collectively known as the Other Common Carriers or oeCs) had inferior
connections to the local telephone companies (for example, customers were required to dial extra digits
to reach the long distance company). 	This son of access is called "non-premium access" 	and its
associated carrier common line charge was set at 45% of the premium charge paid by AT&T. 	The FCC
'Originally the proponion of .the local network loop costs assigned to long distance operations was
the fraction of calls going over long distances. 	In the early 1950s, this was less than 3 %, by 1982 it was
8.3% and in 1991 it was 14.4%.
4

has since required the local telephone companies to install equipment to provide equal access to any long
distance company asking for it. 	Equal access equipment is now in place for nearly all (91 %) long
distance company customers, and non-premium service accounts for only 3 % of acc service. 	A
subscriber common line charge of $3.50 per month for each residential line and $6.00 per month for each
business line was phased in between 1985 and 1991; this new charge gradually replaced most of the
carrier common line charge revenues. The reduction in the carrier common line charge, which has fallen
to about one-sixth its 1984 level (total switched access prices fell about 60%), accounts for much of the
price reductions in long distance service (faylor (1991».
Local switching and transpon ~"rvices are together called "traffic sensitive services" because their
costs depend on the volume of traffic. 	For both of these services, there are many different rate elements
that make the price dependent on various switching services and the distance of the transpon. In general,
however, the rates for these services have declined only slightly since divestiture. 	These rates are
generally believed to be considerably above the services' 	marginal costs. 	Cost studies have put
incremental costs (Mitchell (1990» and long-run marginal costs (Shin (1993» of switched access at one-
tenth to one-third of the switched access rate.
Long distance companies also purchase a different form of carrier access, special access, from
local telephone companies. Special access Iines are not switched by the local telephone company and are
leased by the month at rates corresponding to their capacity and distance. 	Actual usage is not metered.
For a sufficient volume of traffic, special access can represent significant cost savings over switched
access. 	Special access competes with third pany, or facility bypass, access provision. 	Long distance
companies use special access to connect directly to end users with high volumes of traffic or nonstandard
technical requirements and to connect different long distance nodes within a metropolitan area as a
substitute for the local transpon ponion of switched access.
5

Long distance company carrier access costs per minute differ due to differences in the mix of
carrier access services that the companies purchase. 	First, different local telephone companies can have
different prices, and the smaller, mor~ rural local telephone companies tend to have higher prices. 	Long
distance companies that carry a disproponionate share of the calls originating or terminating with these
local telephone companies will tend to have higher carrier access costs. 	Second, the amount of local
transpon purchased for a typical call can differ across long distance companies. 	As a long distance
company serves more urban customers, who tend to be closer to its local network connection, or adds
more local network connections, its average local transpon distance and corresponding charge per minute
tend to fall. 	Third, switched access costs will depend on the amount of lower priced, non-premium
access purchased by a long distance company; however, this cost difference is becoming negligible due
to declines 1J0th in non-premium usage and the carrier common line charge.
B. 	Regulation oj Telecommunications
States regulate prices for intrastate services (local services, intrastate carrier access and intrastate
long distance), and the FCC regulates interstate services (interstate carrier access and interstate long
distance). 	While the FCC regulates AT&T prices directly, the accs and the third pany-access providers
are not regulated directly, although the accs file prices with the state PUCs and the FCC similar to tariff
filings of regulated firms.
Price-eaps 	are 	increasingly 	replacing 	rate~f-retum 	as 	the 	form 	of 	regulation 	in 	the
telecommunications industry. 	Generally, price-eap regulation allows the regulated firm to charge any
price below a regulated price-eap that is periodically adjusted to reflect changes in exogenous cost factors
(Liston (1993». 	The FCC decided to move to price-eap regulation for AT&T in May 1989. 	Federal
price-cap regulation of local telephone companies' interstate access rates was implemented in 1anuary
1991. 	While price-eap regulation of local telephone companies has also been introduced by various
6

PUCs, these mechanisms typically contain explicit profit "sharing" provisions if the actual rate-of-return
exceeds a predetermined limit (Braeutigam and Pa.nzer (1993». 	Profit sharing mechanisms could also
be implicit in the FCC form of price-<:ap regulation. 	In fact, high profits due to the increased efficiency
of British Telecom under price-cap regulation in the U.K. and prices below the price-cap led to a revision
of the price-cap formula toward a more binding cap (Kwoka (1993».
The specific form of price-cap regulation adopted for AT&T divided services into three "baskets"
depending on the perceived level of competition in the service. 	"Basket I" includes residential and small
business services, international services and operator assisted and calling card services; "Basket 2" is
limited to 800 number services; and "Basket 3" contains all remaining services, principally those offered
to large businesses. 	Each basket has its own price-<:ap. and sometimes a floor, that increases with
inflation and decreases with a productivity factor and "exogenous" changes in costs, mainly carrier access
charges. 	As services have been shown to be competitive. they have been removed from price-cap
regulation. In October 1991, the FCC permitted contract carriage for AT&T's offerings to large business
customers. 	However. AT&T must publish a summary of the contract and offer the same price to
similarly situated customers (Griboff (1992». 	Similarly, almost all 800 services were removed from
Basket 2 in May 1993 with the introduction of 800 number ponability.2 	This study uses prices from
Basket I, thought to be less competitive than the other baskets.
2800 number portability allows a customer to keep its 800 telephone number when it changes long
distance 	companies. 	Since 	companies 	make 	investments 	specific 	to 	a 	telephone 	number
(e.g.,advenisements, printed material), switching costs arise if the number must be forfeited. 	See also
Kaserman and Mayo (1991).
7

Ill. 	Empirical Methodology
This study anempts to measure the degree of market power as the difference between price and
marginal cost. 	In theory, a profit maximizing firm not facing price regulation will set prices such that
the price markup over long-run marginal costs, its Lerner index, is equal to the absolute value of the
inverse of the firm's own-price demand elasticity. 	Thus, the degree of market power can be inferred
from estimates of this elasticity. 	However, the data and methodological requirements for the estimation
of firm specific demand elasticities are extremely demanding. 	This section describes the methodology
employed to estimate demand elasticities for AT&T and the accs and to infer price markups over
marginal cost.
A. 	Two-lew:l budgeting
Long distance service can be somewhat differentiated across firms. 	For high-volume business
customers and those who use data transmission services, there may be substantial variation in product
attributes. 	Even for residential and small business customers, dif(erences in, for instance, perceived
quality, customer service, and billing systems could render long distance services heterogeneous across
firms. 	Finally, some carriers offer lower quality, lower priced. non-premium service. although its
popularity has diminished considerably. 	Despite the foregoing differences. it is not unreasonable to
aggregate different firms' long distance service into a single market that is distinct from outside goods.
In this way, it is possible to disentangle the decisions regarding the amount of long distance service to
consume and the firm chosen to provide it.
A two-level budgeting approach is used to estimate the demand system (Deaton and Muellbauer
(1980». 	The upper level determines the industry-wide demand for long distance. while the lower level
determines how demand is allocated among the various tirms in the market. 	This approach is used
8

because it accords with perceptions of the long distance market, and it paniaJly separates the effects of
generaJ inaustry orice reductions from price differences among firms within the industry. 	Real prices
for long distance service fell by about 50% between 1984 and 1991. 	Differences among firms' prices
were much smaJler throughout this period. Two-level budgeting is aJso supponed by the observation that,
while real total expenditures on aJl toll service increased by only 3.5 % between 1984 and 1991, the
market shares of individuaJ firms changed considerably.
With two-level budgeting. consumers are assumed to aJlocate funds across different broad
commodities (level one) and then distribute the aJlocated funds among the specific goods within the
commodity group (level two). 	In thp upper-level, demand for aggregate long-distance service is
determined based on a price index for the industry.
Q;:' = Q;:'(pLD,p~,pSIt,Y)
where Y represents total income and pLD, pLOC, and ps. represent the prices of long distance service, 10caJ
telephone service and telephone sets respectively. 	The total expenditures on long distance service is
determined as a function of price,
yLD ;: P LD Q!:(P w,P ~,pSlt, Y)
where ~ represents income aJlocated to the long distance commodity group in the upper level.
In the lower level, consumers choose which long distance carrier's service to purchase based on
their relative prices and )"D, the amount of income budgeted to long distance service.
(1)
Equation (1) looks Iike a traditionaJ demand equation where quantity demanded is a function of prices
and .. income." 	TraditionaJ demand elasticities are caJculated by differentiating equation (1) with respect
to firm I's price, pl. 	The paniaJ derivative introduces two terms because firm l's price is implicit in
long distance expenditures,
9

This equation can be expressed in elasticity form by multiplying both sides by pi I (t,
(2)
The second term in equation (2) above is the elasticity conditional on the amount of income budgeted to
long distance, .,.,fj' The first term represents the "income" effect due to changes in the amount of income
allocated to long distance service from changes in the price of good 1. 	The first part of the first term.
~ a~ I ~ a~. is analogous to an "income" elasticity for the good, E;. 	If income elasticities do not
differ much across long distance companies. the value of this lower-level income elasticity will be close
to one. 	The second part of the first term. pi a~ I ~ api. can be modified as follows:
where wi is firm l's share and ,.,w is the industry level demand elasticity. 	Putting the components
together yields.
10

_ (1 + 	Llh 	1 C 	C
1111 - 	11}w E 1 + 1111' 	(3)
Equation (3) has an economically intuitive interpretation. The first term measures the "income"
effect while the second measures the elasticity holding budgeted income constant. 	Funher, the first term
can be decomposed into three effects: pJ on pLD. pLD on ,tW, and ,tW on Q'. 	First, an increase in pJ
holding all other prices constant, will increase pLD by the share of the total market that good I represents.
w'. 	Second, since ,tW is the product of pLD and r.:fD, a one percent increase in pLD will increase ,tW one
percent directly and decrease ,tW by the upper-level elasticity due to a movement along the upper-level
demand curve. 	Finally, an increase in ,tW will increase the quantity of good I demanded according to
the lower-level income elasticity.
1. 	Upp~r Uvt/ D~mand
The industry demand elasticity for equation (3) is estimated from monthly time series data. Long
distance quantity is estimated to be a function of its own price, the prices of local telephone service and
telephone sets, income. a time trend and monthly seasonality dummy variables,
logQrLD = 'lU> 10gP,LD + 1jUc logPr
Uc + 'lScfJogP,s.. + E logIncom~r
12
+ 4 0 time + E 4 f mollth,' + (dt'
t -I
(4)
, The quantity of long distance service demanded is expected to fall as price increases, with rF measuring
the industry elasticity. Since local telephone service and telephone sets are complements to long distance
service, increases in their prices are expected to lead to a fall in the demand for long distance service,
implying that rf« and 77s.t should both be negative. 	The coefficient on income. E, is intended to measure
the income elasticity for long distance telephone service and is expected to be positive. 	The prices of
PBXs, computers and modems, which can also be thought of as complements to long distance telephone
11

service, tended to decline over the sample period. 	Sir::e price series are unavailable for these products.
a time trend is mtroduced to capture this shift in demand, and the coefficient on that time trend. 00>
should be positive. 	Month dummy variables are intended to account for the seasonality of the demand
for telephone calling.
2. 	Lower Level Demand
Budget share regression equations can incorporate the assumptions of two-level budgeting, 	If the
lower-level demand elasticities are constant in the relevant range and only two carriers exist, then
equation (1) becomes,
Assuming E7 is one,3 then by adding log pi to and subtracting the last term from both sides, we get.
Since tu' is total industry expenditure. this is the budget share equation. 	Firm level constrained
elasticities are estimated by regressing a firm's revenue market shares against the price of its own service
as well as that of its rivals, month seasonality dummy variables and state dummy variables,
Equation (5) is estimated for AT&T and an aggregation of Mel and Sprint. 	Changes in the relative
prices within the industry are expected to induce brand switching causing '1~i to be negative and T/~j to be
)It is lik.ely that income elasticities do not differ much across long distance companies, implying that
the value of E: is one. 	Data limitations make its estimation problematic. 	Since quantity is the dependent
variable and it is implicit in total long distance expenditures, errors in variables problems could render
estimates biased.
12

positive. 	Long distance calling demand exhibits sensitivity to seasonal changes and including month
dummy variables allows for differences in shifts in demand across brands. 	State dummy variables are
intended to account for idiosyncratic differences in supply and demand across states.
B. 	Econometric Issues
One of the criticaJ assumptions underlying regression analysis is that all of the regressors are
uncorrelated with the error term. 	If the statistical independence assumption is violated, the parameter
estimates are biased and inconsistent. 	In demand est~mation, a frequent cause of statistical dependence
is observed prices incorporating the in:'uences of both supply and demand. That is, prices and quantities
are determined at the intersection of supply and demand curves. 	For instance, industry price, one of the
regressors in equation (4), and the firms' prices in equation (5), will depend on the level of output
through supply relationships. 	The error terms, "', and IJ."" 	represent the effects on output that are not
explained by the modeled demand relationship. Thus, because both the prices and the error terms depend
on the output level, they are likely to be correlated.
This problem, of endogenous explanatory variables, can also be thought of as pan of the broader
problem of "measurement error." that is. of the divergence between the data being observed and the
variables being modeled. 	For example. in demand estimation, the relevant price variable is the price that
would prevail if the demand curve did not shift. 	To the extent that the observed price is affected by a
shifting demand curve. it is measured with error. 	Measurement error in explanatory variables implies
correlation between the explanatory variables and the error term, resulting in biased coefficient estimates.
The direction and magnitude of the bias is a function of various coefficients and correlations between
variables.' 	In equation (5), coefficient bias can result from measurement error in both the own and
4Measurement error in a variable will tend to bias its coefficient toward zero. 	Measurement error
in a different explanatory variable will tend to bias a coefficient in the direction of the product of the
(continued... )
13

competitor prices. 	Measurement error in the own price will tend to bias the coefficient toward zero.
Measurement error in competitor prices will also tend to bias own-price elasticity estimates toward zero
(since the correlation between own and competitor prices and the cross-elasticity are both positive and
own-price elasticity is negative). 	Likewise, measurement error in these prices will tend to bias cross-
elasticities down (toward zero).
Two general methods of dealing with measurement error are instrumental variables and reverse
regressions. 	The instrumental variables method brings other information to bear in order to recover
estimates that are consistent. 	This method attempts t:J purge endogenous explanatory variables of their
correlation with the error term. 	Reverse regressions, on the other hand, simply attempt to put bounds
on the magnitude of the bias. 	These regressions switch the dependent and independent variables to
generate estimates biased above and below the true parameter value.
1. 	1h~ InstTlllUnraJ Variablu M~thod
Since price and quantity are jointly determined, observed prices represent a mixture of demand
and supply relationships. 	Measures of the demand elasticity from direct regression techniques on
observed prices will yield biased and incon.;istent estimates because the price is endogenously determined.
Disentangl ing demand relationships from supply relationships empirically requires a technique that can
distinguish between shifts in the supply curve (movements along the demand curve) and movements along
the supply curve (shifts in the demand curve). 	Price and quantity pairs associated solely with shifts in
the supply curve, for instance, will trace out a demand curve whose slope (or elasticity) can be estimated
without bias. 	One method for identifying shifts in the supply curve is to use variables that represent the
4( ... continued)
correlation between the two variables and the coefficient of the variable with measurement error. 	The
magnitude of the bias will be a function of these various correlations and coefficient values (Maddala
(1988) pp. 388-391).
14

cost of prod""·ion. The price level that is predictf'd by these variables would not depend on demand. but
instead would reflect only changes in the cost of pro~uction. In this manner. the predicted price becomes
independent of the error term and the resulting demand estimates are unbiased. 	In such an application
of the instrumental variables technique, the variables representing the cost of production are called the
instrument set.
The ability of instrumental variable methods to obtain meaningful demand parameters depends
on the ability to find suitable instrumental variables for the endogenous price. 	The two general
requirements are that the instrumental variables be independent of the error term in the demand equation
and that they be correlated with the endogenous variable. 	First, instrumental variables that are themselves
functions of the output level will create interdependence between the predicted price and the output level
and, thus, between the predicted price and the error term. 	This reintroduces the problem for which
instrumental variables were sought in the first place. 	Second, correlation between the instrumental
variables and price insures that they "explain" some (\f the variation in the price. 	That is, they must
represent enough of the shifting in the supply curve to provide significant movement along the demand
curve. 	Better predictions of the shifts in the supply curve provide more precise (Le., smaller variance)
estimates of the shape of the demand curve.
A test of bias due to errors in variables can be conducted when instrumental variables are
employed. The Durbin-Wu-Hausman test (Hausman (1978» compares the parameter estimates from two
different specifications of a regression model. 	If the estimates are sufficiently different (in a statistical
sense), the specification that relies on the stronger assumptions regarding the data is rejected. 	In the
present context, the assumption that the errors in the variables do not lead to biased estimates (implicit
in ordinary least squares (OLS) results) is stronger than the assumption that they might (implicit in
instrumental variables results). A Durbin-Wu-Hausman test can also compare parameter results from two
different instrumental variable specifications where the instrument sets are different.
15

2. 	Rtvtrst Rtgrtllions
It is possible to place bounds on the true parameter by reversing the direction of the regression.
Regressing Yon X when both are measured with error yields a coefficient biased toward zero. Similarly.
regressing X on Y will also yield a coefficient biased toward zero. 	However, the reciprocal of the
coefficient of Y from the second regression provides an alternative estimate of the coefficient of X from
the first regression. 	This reciprocal will be biased upward and provides an upper bound on the true
parameter.
Iplim PI < IPI < Iplim l/y I
... 	...
where {j is the estimated coefficient of X in the direct regression, 'Y is the estimated coefficient of Y in
the reverse regression ard {j is the true parameter value. 	This procedure generalizes to multivariate
regressions and generates the set of estimates that bound the true parameter value (Klepper and Leamer
(1984)). 	Parameter estimates from reverse regressions are maximum likelihood estimates, and the set
of parameter values bounded by these estimates contains the true parameters. 	As discussed in the
previous section, instrumental variable methods, in principle, yield consistent parameter estimates. 	Yet,
because instrumental variable methods may still yield biased coefficient estimates if the instruments
themselves are functions of output, reverse regressions can provide additional information about the size
of any remaining bias.
C. 	1ht LLrntr Inda
The reciprocal of the own-price elasticity, the Lerner index, provides an estimate of the
percentage price marlcup over marginal cost for an unconstrained, profit maximizing firm:
16

P 	- Me 	1	L=_, __I=
t 	P. 	I 	I	I 	flu
This condition is derived from the first order conditions that equate marginal revenue to marginal cost.
For larger elasticities (in absolute terms), the marginal revenue curve is closer to the demand curve and
the profit maximizing price is closer to marginal cost. 	In this way, the demand elasticity indicates the
extent to which the firm can unilaterally raise prices without suffering a large loss in quantity. 	This is
the extent of the firm's market power. 	The application of the Lerner index to estimated demand
elasticities requires assumptions regarding: the degree to which competitors respond to price changes of
rivals, the effect of regulation on AT&~'s ability to set prices above marginal costs; and the bias in the
implied marginal cost due to the shon-run, rather than long-run, nature of the estimated demand
elasticities.
The demand elasticity described in equation (3), used to infer the Lerner index, represents the
effect of a change in a firm' s price on its quantity when competitors' prices are held constant. 	This is
the appropriate elasticity if the firm assumes that its competitors' prices wiJI remain unchanged when it
changes its own price, as would be true with the Benrand conjecture for a differentiated product industry
(Carlton and Perloff (1990), pp. 272-276, 308-310). 	If, however, the firm conjectures that changing its
price will induce rivals to adjust their prices (independent of a common change in costs), then this is not
the appropriate elasticity. In the extreme case of perfect collusion, price changes by one firm correspond
to equal price changes by all others, and the relevant firm-specific elasticity is the industry demand
elasticity .
We assume that AT&T will conjecture that its competitors' prices will not change in response to
its own price change. 	First, the non-AT&T carriers ("Other Common Carriers" or "OCCS") have ample
capacity with which to expand output. 	While AT&T's share of fiber optic capacity was 41 % in 1992
(Kraushaar (1993)} its share of output was 60% by 1992. 	This implies that the OCCS have even more
17

capacity 	~x~ ...1Sion than does AT&T. 	Second, the average OCC customer is likely to be significantly
more pric~ ela..~•.; than the average AT&T customc:r. 	The average OCC customer demands more than
twice the calling volume as demanded by the average AT&T customer.' 	Moreover. since most OCC
customers have switched from AT&T at some time. they have revealed themselves to consider OCC
service to be a closer substitute for AT&T service than do existing AT&T customers. 	Third, besides
being more price sensitive. OCC customers can choose among a large number of non-AT&T long
distance companies whose services are likely to be perceived as good substitutes for each other. 	Over
500 OCCs other than MCI and Sprint compete in the interstate long distance market (Statistics of
Communications Common Carriers (1992». 	While most of these firms' operations are confined to
reselling service supplied over the facilities of other long distance carriers, nine firms operated facilities
in more than 45 states by 1991 (Statistics of Communications Common Carriers (1992)). 	Moreover. the
combined market share of these other OCCs is greater than that of Sprint and the growth in their
combined market share since 1988 was greater than that for MCI or Sprint.6 This suggests that an OCC
faces the prospect of customers switching to any of a large number of potential competitors if it aneptpted
to raise its price in response to an AT&T price change. 	Founh. the telecommunications industry exhibits
characteristics which tend to impede collusion, either tacit or explicit. 	In addition to the large number
of firms, collusive behavior is more difficult to enforce in industries with rapidly changing technologies
and, consequently. changing market shares (Stigler (1964». 	Long distance telecommunications has
experienced an accelerating pace of innovation. 	Technology innovations include microwave and fiber
~e ratio of a firm's total minutes supplied to the number of its customers represents an index of
calling volume per customer. 	The calculated calling volume index for OCC customers is 2.7 times that
for AT&T customers in December, 1987 and 2.0 in June, 1992 (Statistics of Communications Common
Carriers (1992».
6Market shares in 1991 were 15.0%,9.7% and 13.1 % and 1988 market shares were 10.3%.7.2%
and 8.0% for MCI. Sprint and all other OCCs respect~vely (Statistics of Communications Common
Carriers (1992».
18

optic transmission, asynchronous transfer mode (A TM) and frame relay switching, and software defined
network (SDN) and b~dwidth-<>n-demand data communications. 	Consumer related innovations include
magnetic strip calling cards. optional calling plans and "EasyReach 700" service. 	As long distance
companies adopt these innovations to varying degrees, they become bener able to serve different niches
of the market and collusive arrangements become more difficult to enforce.
Inferring a price-cost markup from a Lerner index requires the assumption that marginal revenue
is equated with marginal cost. 	However, this assumption may not hold for firms, such as AT&T, that
face price regulation. 	With a constrained price, in the shan-run the quantity demanded will exceed the
unconstrained profit maximizing level and marginal cost will exceed the unconstrained marginal revenue
curve. 	Thus, the price-cost margin under regulation would be smaller than the price~ost margin derived
from the profit maximizing assumption implicit in the Lerner index. 	However, since the finn may be
able to respond to regulation, the divergence between marginal revenue and marginal cost will be
diminished for two reasons. 	First, if the finn expects price to be constrained into the future, it can
reduce marginal cost to the lower marginal revenue level by reducing investments that maintain quality.
Second, if the firm expects price to be unconstrained for some amount of time in the future, the relevant
marginal revenue is a weighted average of the constrained and the unconstrained marginal revenues,
where the weight is the probability that the constraint will be binding. 	In the first case, a sub-<>ptimal
level of quality is chosen and in the second, the relevant constraint is the expected price~ap over periods
when it is binding.
In fact, regulation does not appear to have greatly constrained AT&T's prices, at least since price-
caps have been in place. 	For Basket I services, AT&T's price was at its cap only about one-third of the
time that price-eap regulation was in effect. 	Figure 1 shows that the price cap is more likely to be
binding just after large changes in the cap brought about by large changes in regulated carrier access
19

prices. 	I am told that this result could be a reflectio'l of regulatory delay in reviewing AT&T price
changes. 	Thus, even one-third is likely to overstate the fraction of time that the price-eap was binding.
If the estimated elasticities represent less-elastic short-run demand, then the implied Lerner indices
will overstate the actual long-run price-eost margin (see figure 2). 	There are two reasons to expect the
estimates to represent short-run elasticities. First, the elasticity estimates are generated with monthly data
on price and quantity. In this market, consumers are not likely to fully adjust to a price change until after
a year Of so. 	Short-run demand is less elastic because price information is conveyed imperfectly to
consumers through advertisements and experiences with monthly bills. Consumers respond to new price
information when they become aware of the price change, in some cases months after it actUally occurred.
Notably, studies estimating long-run industry demand curves for long distance telecommunication often
allow current price changes to affect current quantity as well the quantity demanded for a year or so into
the future.(Taylor (1980), Taylor and Taylor (1993». 	Second, the OCC Lerner index values, if
construed as long-run estimates, would represent economic rents, i.e., profits from the exercise of market
power. 	While it is possible that oces are earning rents, it is likely that most of their implied price-eost
margin stems from the upward bias due to the estimation of sh9rt-run, and not long-run, demand
elasticities.
A measure of the AT&T Lerner index bias generated by the estimation of short-run demand
. elasticities, instead of long-run elasticities, can be derived from the estimated OCC Lerner index. 	Under
I
the assumption that long-run OCC firm-specific demand is nearly horizontal, (that is, the OCCs are
virtually textbook competitive firms), 	their long-run price-eost margin 	is very small. 	The FCC
considered "competition in business services to be thriving," from which can be inferred that the FCC
considered AT&T's ability, let alone the OCCs' ability, to set prices above marginal cost for business
services to be negligible (Federal Communications Commission (1991). 	Since some of these business
customers are long distance resellers. the ability of the accs to set prices significantly above marginal
20

costs to residential customers appears quite limited. In this case, almost all of the estimated acc Lerner
index represents a measure of the bias. 	If the amount of the bias is the same for AT&T and the aces.
then the actual price-rost margin for AT&T can be approximated by,
the difference between the estimated price-cost margins for AT&T and the aces.
IV. 	Data Description
This section describes the basic data employed to estimate long distance telephone demand
relationships described by equations (4) and (5). 	The upper level demand estimation, equation (4), uses
national time series data for the period July 1986 to August 1991. The lower level demand estimations,
equation (5), uses monthly data for the years 1988 to 1991 for five states. 	Instrumental variables are
employed to identify structural demand parameters in both estimations. 	This section describes the data
used, focusing primarily on the lower level estimation.
A. 	Upp~r Lev~/ Estimation
For the upper level demand estimates, equation (4), national data were collected from various
sources on output, prices and income. 	The total number of minutes of interstate calling is used as the
industry output. 	These data come from National Exchange Carrier Association (NECA) reports to the
FCC. 	The CPI prices for interstate long distance and local service, and the PPI for telephone sets were
used as price variables. 	Per capita personal income was used as the measure of income. 	All prices and
income data are deflated by the CPI for all goods and services.
21

The price of long distance is treated as endogenous, so instrumental variable techniques are
necessary. 	Inst" ,-,ental variables that are availablE' are the PPI indices for transmission and digital
switching equipment and the wages of telecommunications workers. 	All of these represent prices of key
inputs into the production of long distance services and, thus, should represent shifts in the supply curve.
They will be correlated with the quantity demanded only to the extent that the long distance industry
represents a significant portion of the total demand for the individual factors and these markets have
upward (or downward) sloping supply curves. 	While the long distance industry does account for a large
fraction of these equipment markets, there is no evidence on the slope of the supply curves.
B. 	Lower Levt.' Estimation
The principal data used in the estimation of lower level demand, equation (5), are interstate
carrier access usage and expenditure information for AT&T and the Other Common Carriers (OCCs).'
The carrier access usage information available is for sy·itched and special access purchased by AT&T,
MCI, Sprint and an aggregation of the other long distance companies. 	These data span five states and
each month from January 1988 through December 1991 for a total of 240 observations. 	Interstate toll
service is the focus of this study both because these data were the most accessible and because this is one
of the most imponant segments of the market.
I. 	D~mand Variab/~s
A number of relevant variables are available. 	For each state, month and long distance company,
the variables available are the number of minutes and dollar expenditure on interswe switched access and
the number of lines and dollar expenditure on interstate special access. 	Since these data are obtained
"These data are used under a nondisclosure agreement with Southwestern Bell Telephone Co. and are
not publicly available.
22

from billi~ ..J :.1formation, they should reflect actual purchases accurately. The number of switched access
minutes measures tht: quantity of long distance service aemanded. 	The quantity is the sum of both the
number of outgoing and incoming minutes. 	Long distance companies have increasingly moved their
larger customers to special access and facilities bypass. 	If special access usage grew faster for the oces
than for AT&T, market shares based on switched access alone would underestimate the acc market
penetration and will tend to bias own-price elasticities downward. 	However, the market share based
solely on switched access more accurately represents Basket 1 services. 	Dividing interstate switched
access expense by interstate switched minutes yields an average price for switched access per minute.
Dividing interstate special access expense by the interstate number of lines yields an average price per
special access line.
The price of long distance service for different firms is of key importance to the estimated results,
and is the weakest data element. 	Long distance price variables were constructed from price infC'rmation
in tariffs filed at the FCC. 	AT&T, MCI and Sprint su""mit rate schedules to the FCC when they change
their rates. These schedules list prices by time of day (day, evening and night), first or additional minute
and distance of the call (there are twelve different mileage bands). 	Since the quantity variable aggregates
calls over all of these dimensions, the relevant price is a weighted average over all of these dimensions.
However. the appropriate weights can· only be approximated ~d some assumptions must be made
regarding the relative use along these dimensions.
Time of day. duration and distance weights were computed using intraLATA toll information and
some simplifying assumptions. 	The average duration and the fraction of calls by day, evening and night
were available for local telephone toll service by state. 	Applying these weights to the interstate data
assumes comparability between the shoner distance intraLA TA and the longer distance interstate calling
patterns. 	The relative weights for the separate mileage bands were computed in an admittedly ad hoc
way. 	A so called "gravity" model of telephone traffic flows was employed to calculate the expected
23

number of calls flowinl between two points as propJrtional to the product of :: .• "mass" of the two
locations divided by the square of the distance between the two locations. 	The expected flows were
calculated for each of the 3187 counties in the U.S. to each county of the states in the dataset using the
counties' geographic center to compute distances and its 1991 population for its mass. 	The individual
county flows for a state are aggregated into the mileage bands.' 	The price is averaged over mileage
bands using the aggregated state flows as weights.
2. 	Potential M~surement Errors in the Long Distance Prices
Due to the many assumptions inherent in the construction of the price variable. it is appropriate
to check it against other sources. 	One price series available for comparison is the AT&T price index
reponed to the FCC. 	As part of the filing requirements for price-cap regulation. AT&T has reported
price indices for different baskets of services since April of 1989. The price index for Basket I Services
represents an independent measure of the long distance prices constructed above. 	Figure 3 displays the
constructed long distance prices for AT&T and the aces in Texas and the Basket 1 index for AT&T. 9
It appears as though the constructed AT&T price closely follows the Basket I index. 	The gradual
deviation between the two series may be accouilted for by smaller price reductions in the international
and calling card caBs. which are included in the Basket 1 price, index but excluded from the constructed
price.
Discounts from posted prices are another potential problem with the constructed prices. Programs
like AT&T's Reach Out America. MCl's Friends and Family and Sprint's Most tend to offer a percentage
discount off the posted prices. 	If discounted prices vary more than posted prices. then changes in posted
'Bob Evett and Equifax National Decision Systems are gratefully acknowledged for the use of these
data.
~e other states in the sample yield similar results. 	Texas was chosen because it represents about
half of all the long distance in the sample.
24

prices will underc:tate actual price changes and esti-nated price elasticities will overstate true price
elasticities. 	Figure 4 confirms that AT&T's Residential and Small Business and Reach Out America
indices diverge slightly over time. 	However, 	the slight divergence suggests that this source of
measurement error is small. 	The same is likely to be true for the acc price.
Another measurement error in the ace price results from the exclusion of firms other than the
two largest, MCI and Sprint. 	Price data were not collected for these other firms because the largest
represents less than I % of industry revenues and because quantity information was available only for the
aggregation of these firms. 	The combined market share of these firms has grown from about 7 % to 13 %
between 1988 and 1991. 	To the exter.. chat these firms' services are substitutes for AT&T's and these
smaller firms prices are uncorrelated with MCls' and Sprints', AT&T's estimated own-price elasticity
should be biased toward zero. However, the small differences between the MCI and Sprint prices relative
to their difference from the AT&T price suggests that the aecs' prices are highly correlated with each
other and that this bias should be small.
3. 	Insrrum~nraL VariabL~s
As noted above, inconsistent coefficient estimates due to errors in variables can be overcome with
the use of suitable instrumental variables. 	The constructed firm-specific long distance prices are likely
to contain errors because of both supply and demand simultaneity and the assumptions inherent in their
It
construction. Instrumental variables are required to predict supply prices and purge measurement errors.
Moreover, in order to estimate firm-specific price elasticities, firm-specific instruments are needed so that
shifts in the supply curve for a particular firm are independent of shifts in the supply curves for other
firms. 	Factor prices could be suitable instrumental variables if they are unique to each firm. 	If,
however, a factor represents a commodity good to all firms (e.g., raw materials), then changes in the
factor price are likely to induce similar shifts in supply for all firms in the industry.
25

The avail:\ble firm-specific factor prices include the cost of capital and the average prices of
carrier access. These factor prices are added to the set of instrumental variables used in the upper level.
Measures of the cost of debt are derived from Moody's yield to maturity calculations on outstanding debt
for each of the largest three long distance companies. 	A bond's yield to maturity is deflated by the yield
to maturity for a similarly lived government bond in order to adjust for changes in expected inflation.
Finally, a firm's outstanding bonds are aggregated into a single yield to maturity using their face value
as weights.' 	The cost of capital can be expected to be unique to each firm, correlated with price and
uncorrelated with the error term in the demand equation. 	First, the cost of capital depends largely on
the riskiness of the firm borrowing the funds. 	In the long distance market. AT&T's capital costs reflect
a relatively risk free firm, while Mel pays a relatively high interest rate on its junk bond debt. 	Second,
since the firms in this industry are relatively capital intensive. changes in their cost of capital are likely
to induce relatively large price changes. 	Third, the cost of capital is likely to be independent of the
output level and, thus, the error term.
While carrier access prices may be both unique for each long distance company and highly
correlated with long distance price, they may also be correlated wi~ output and, thus, the error term.
Firms differ in their carrier access purchases mainly because of the degree to which they integrate into
the distribution of telephone calls. 	When a sufficient volume of calling for a long distance company
originates or terminates in a panicular area, the long distance company will extend its network into the
area and thus reduce its purchase of access from the local telephone company. 10 Thus, average switched
access prices will depend on the location and the calling patterns of a long distance company's customers.
10Long distance companies typically terminate their networks near the centers of metropolitan areas.
Calls to and from outlying areas are transponed to the long distance network by local telephone company
at a charge ttat increases with distance. 	When the volume of traffic for the outlying area develops
sufficiently, a long distance company will extend its network to the area. 	This occurs when the cost
savings from reduced expenditures on local telephone company transpon is greater than the cost of
extending the network.
26

More geographically concentrated customers and calls will lead to more backward integration, lower
expenditures for switched access and lower total costs. 	Likewise, average special access prices tend to
be lower in areas of more densely located customers and calls because less costly, higher capacity lines
can be used. 	Changes in carrier access rates are highly correlated with long distance prices, explaining
much of the decreases in prices (faylor (1991). 	However, there is evidence that carrier access prices
fall as more is demanded (parsons and Ward (1993». 	Thus, using carrier access prices as instruments
may reintroduce correlation between long distance prices and the error term in the demand equation.
4. 	SOIM Tests ofthe Instrumental Variables
While the focus of this paper is on demand estimation, one way to evaluate the effectiveness of
the demand instruments is to estimate supply relationships. 	This analysis can also provide an estimate
of the effect of price-cap regulation on AT&T's costs. 	Price-caps, which replaced rate-of-return as the
regulatory scheme for AT&T midway through the sample, may provide AT&T stronger incentives to
reduce costs (Liston (1993)). 	In general, the supply curve is expected to be quite elastic given the
relatively large fixed costs relative to variable costs in the industry'" 	Supply relationships are estimated
with the firm specific data using the equation,
5
logP" = ,¥cap" + elogQh 	+ L ~llogw~· + L PEstatt~ + Vh'
IEL 	E-I
(6)
Cap is a dummy variable whose value is one during the time that AT&T was regulated under the price-
cap regime as opposed to the rate-of-return regime. 	The Mis, the factor input prices, are the PPI indices
for transmission and digital switching equipment, the wages of telecommunications workers, the yield to
IIln fact, 	Huber, et al. (1993) contend that the technology is developing into that of a natural
oligopoly due to scale economies in transmission.
27

maturity on corporate bonds. and the avera~~ price- far switched and special access. 	The instruments
for quantity are income and month dummy . ..riablt.. I
Estimation results for equation (6) are reponed in table 1. 	First. as expected. supply curves
appear to be flat. 	Still. coefficient estimates not significantly different from zero are likely due to the
meager instruments available for the quantity demanded. 	Indeed. the negative coefficients for minutes
of use could be a demand relationship picked up because both output and income, an instrumental
variable, are correlated through a time trend in both variables. 	Second, there is evidence that the
movement toward price-eap regulation lowered AT&T's costs. 	The price-eap coefficient is negative for
both AT&T and the accs but is significant for AT&T only. 	This conforms to the hypothesis that price-
cap regulation is a more efficient form of regulation for long distance telephone service ard with other
empirical results (Mathios and Rogers (1989. 1990), Kaestner and Kahn (1990». 	Since the accs are
not subject to regulation, the only effect of price-eap regulation on their prices would be through more
vigorous competition with a more cost-efficient AT&T.
The third result is that the industry-wide and firm-specific factor input prices are relatively good
explanatory variables for the price of long distance service. 	Increases in the prices of inputs common
to all firms -- switching equipment. transmission equipment and labor - increase the output price for both
AT&T and the accs. with coefficient magnitudes and confidence levels differing across firms. The yield
to maturity on corporate bonds. a measure of the cost of capital, is significant only for the accs. 	This
result is reasonable since much of MCl's debt is in the form of junk bonds whose prices are more
variable than the bonds issued by AT&T. 	Note also that the estimated coefficients of switched and
12Price-eap regulation of AT&T could render its supply price censored at the price-eap, suggesting
that Tobit estimates of equation (8) are more appropriate. 	As mentioned above, for Basket I services.
AT&T's price was at its cap about one-third of the time that price--eap regulation was in effect. 	AI<;o.
periods in which the cap was binding are possibly the result of regulatory delay. 	Since, in these c
the price cap may actually be a floor and not a ceiling, attempts to account for censoring are not reported.
28

special access prices are aJl positive and significant. 	The estimated standard errors are quite smaJl.
indicating that these input prices should be good proxies for firm-specific shifts in supply, which are
needed in the demand estimation.
V. 	Demand Estimation Results
The results of the demand estimations are presented 	in this section. 	In the upper level
regressions, the results are quite similar to estimates presented in other research. 	SpecificaJly. the
commonly accepted 	industry deman~ elasticity of -0.65 cannot be rejected. 	In the lower 	level
regressions, firm-specific demand elasticities for both AT&T and the oces are found to be sensitive to
assumptions regarding the errors in the variables. 	Carrier access prices can be rejected as suitable
instrumental variables due to their endogeneity with output. 	The elasticity estimates resulting from the
restricted instrument set (Le., one without carrier access prices) provide a smaJler range of estimates.
Lower bound estimates of shon-run own-price demand elasticities are -2.9 for AT&T and -6.6 for the
accs. 	While consumers are not likely to fully adju$t to price changes until after a year or so, these
estimates indicate that demand is fairly elastic even in the shon-run.
Estimation results for the industry level demand estimation, equation (4), are reponed in table
2. 	Columns (I) and (2) exclude the time trend and provide evidence that local service and telephone sets
are strong complements with long distance services. 	Columns (3) and (4) include a time trend to capture
an exogenous shift in demand (e.g., the growth of data and facsimile transmissions). The resulting cross-
price elasticities are smaJler and are now insignificant at standard levels. 	Where Hausman, Tardiff and
Belefante ( 1993) find significant cross-elastic effects from long distance prices on 10caJ service penetration
rates. the converse cross-elastic results are mixed here. 	Estimated own-price elasticities are significant
29

and similar to those reponed elsewhere (Taylor and Taylor (1993). Taylor (1980». 	In fact, 	Ule
commonly accepted value of -0.65 can only be rejected in column (1).
Table 3 reports two-stage leiaSt squares (2SLS) regression results for the lower level demand.
equation (5). under various assumptions regarding the errors in variables. 13 	Because Durbin-Wu-
Hausman tests always reject OLS results in favor of instrumental variables results. only the latter are
reported. 	The left panel uses all of the instruments and the right panel uses all instruments except the
carrier access prices. 	Since carrier access prices can be correlated with the quantity demanded. results
that rely on them may still be biased. 	In each panel. the results from the direct estimation of the marlcet
share regressions. as well as the results from the estima~ion of me reverse price regressions. are reported.
Coefficient estimates from the reverse regressions are not directly reported. but are used to compute
parameter e~timates comparable to those obtained from the direct regression. 	The top panel reports the
results for AT&T demand and the bottom panel reports those for the aces.
Coefficient estimates 	vary widely depending on the errors in variables assumptions made
regarding the data. 	When all of the instruments are used. the estimate on AT&T's own-price elasticity
varies from -1.16 to -5.87. 	Bias toward zero is confirmed by reverse regression estimates that are larger
(in absolute terms) than those from the direct regression. 	The range of coefficient estimates shrinks to -
2.15 to -4.68 when the instrument set excludes carrier access prices. 	Likewise. the range of own-price
estimates for the OCCs collapses from between -3.09 and -13.95 using all instruments to between -5.73
and -9.27 using the restricted instrument set. 	These results are consistent with the conjecture that carrier
access prices are unsuitable as instruments. 	Indeed, in all cases. Durbin-Wu-Hausman tests reject the
hypotheses that the coefficient estimates are the same using either instrument set.
l)While applying a Tobit regression to the first stage to account for possible censoring at the price-eap
does change the estimated coefficients in the first stage. the results for the second stage are virtually
unchanged.
30

Un("n"ditional own-price elasticities are now constructed using equation (3). 	These elasticities
and their associated L~rner indices are reported in l41.ble 4 assuming an industry demand elasticity of -0.65
and revenue market shares, of 0.647 for AT&T and 0.269 for the oces.l.e For example. AT&T's own-
price coefficient of -2.15 from the direct regression using the restricted instrument set yields an estimated
own-price elasticity of -2.92 and a Lerner index of 0.337. 	Under the same assumptions. the OCC own-
price coefficient of -5.73 translates into an own-price elasticity of -6.64 and the Lerner index is 0.151.
That is. AT&T's price markup over marginal cost is estimated to be about 33.7 % of its price. while the
OCe's is 15.1 %.
VI. 	Potential Deadweight Loss Calculations
As noted above, firm-specific demand elasticities imply price marlcups over marginal cost via the
Lerner index. 	Also noted above, the estimated elasticities and their associated Lerner indices are likely
to represent shon-run, and not long-run relationships. 	This implies that the estimated marlcup over
marginal cost is biased upward. To calculate long-run elasticities and Lerner indices for AT&T, the size
of the bias introduced by short-run demand elasticities must be accounted for. 	Under the assumption that
the OCCs are earning no economic rents at all, their Lerner index represents a measure of this bias for
the OCCs. IS 	If the amount of the bias is the same for AT&T and the oces, then the actual price-eost
margin for AT&T can be approximated by the difference between the estimated price-eost margins for
AT&T and the oces. 	This provides an upper bound estimate of the price marlcup over long-run
marginal cost of 18.6 % (33.7 % - 15.1 %) using the estimated results from the direct regression and the
14I'J'hese shares do not add up to one because the OCC share excludes long distance companies other
than MCI and Sprint.
Ulf the OCCs are in fact earning some economic rents, then this methodology will tend to overstate
the Lerner index bias caused by the estimation of short-run demand elasticities.
31

restricted insuument set. 	The lower bound estimate from the AT&T price rever" ~ ~egression is 8.5 %
(18.3% -9.8%).
The potential deadweight loss from supra-competitive pricing can be calculated as the area
bordered by the demand curve. marginal cost curve and the current output (figure 2). 	The economic
profits, output times the difference between price and long-run marginal cost, could also be considered
losses to society if that amount is expended to perpetuate the entry barrier creating the rents (posner
(1975». 	.This is likely not the case here since regulators are taking significant steps to introduce
competition to the long distance market. 	Given this regulatory predilection, the return to this son of rent
seeking by AT&T should be quite low. 	If so, the dead weight loss can be calculated simply as DWL =
J~~ /Q(p) - Q(P,)J dp, assuming constant marginal costs. 	With constant elasticity, the inverse demand
is given by Q(P) 	= 	Qo(PIP,)". 	The resulting deadweight loss as a fraction of current revenue is
DWLlREV = /J-(l-L)'+"J!(1 +.,J} - L where L is the Lerner index.
While firm-specific elasticities are appropriate for generating estimates of the Lerner index, the
industry elasticity is the relevant elasticity for calculating the deadweight loss. 	If the OCCs were able
to match the hypothetically lower AT&T prices set at long-run marginal cost, then the deadweight loss
applies to the whole industry. 	This would be the case if AT&T were providing a price umbrella over
an industry in which all firms had similar nondecreasing cost functions. 	If instead. AT&T can operate
at lower long-run marginal cost than the oces, then it would capture the entire market with prices at
long-run marginal cost. 	In this case the firm demand elasticity is the industry demand elasticity. 	Table
5 reports the ratio of potential deadweight loss to revenue for various price-marginal cost margins
assuming a demand elasticity of -0.65.
The relevant deadweight loss from table 5 depends on which values of long-run marginal cost are
used. 	Reducing the largest estimate of AT&T's shon-run price-cost margin (0.337) by the size of the
bias derived from the OCC Lerner index (0.151) yields a long-run Lerner index of 0.186 and a
32

deadweight loss of about 1.26 % of total revenues. 	Re1ucing the smallest estimate of AT&T's shon-run
price-eost margin (0.183) by the measure of the bias derived from the OCC Lerner index (0.098) yields
a long-run Lerner index of 0.085 and a deadweight loss of about 0.25% of total revenues. 	These
estimates compare favorably with the range of the current economy-wide estimates of deadweight loss
due to market power of 0.5% to 2.0% of GNP (Scherer and Ross (1990), pp. 663-667).
If this market is currently more competitive than during the 1988-1991 period, then these
estimates will overstate the current potential deadweight loss due to supra-eompetitive pricing. 	The
evidence of a more competitive market is: l) the fall in AT&T's market share from a national average
of 67 % during the sample period to 6C';; currently, 2) the introduction of 800 number ponability, and
3) the increase in the number of foreign countries reached byOCe netWorks.
An alternative representation of the Lerner index is LAfT = ~fT/ rl."LD I + SOCC(1_~TT)J where
~fT is AT&T's market share and (Jocc is the OCC supply elasticity (Landes and Posner (1981». 	With
values of L AfT
, ~TT and ."W of 0.186,0.67 and -0.65 respectively, F 	becomes 8.95. This impHes that
the oces would be willing to increase their output by almost 90% at prices 10% higher than current
prices. 	Assuming an acc supply elasticity of 8.95, the reduction in AT&T's market share from 67%
to 60% alone would reduce the Lerner index from 0.186 to 0.142 and the potential deadweight loss from
1.26 % to 0.71 % of revenue. 	Ponability of 800 numbers and increased international access by the OCCs
will tend to increase eocc as the OCCs are better able to provide substitutes for AT&T's services. 	If, in
addition to the fall in AT&T's market share, the OCC supply elasticity has increased from 8.95 to 10.0,
AT&T's Lerner index would decrease from 0.142 to 0.129 and the potential deadweight loss would
decrease from 0.71 % to 0.58% of revenue.
To be sure, even the upper-bound estimate of the potential deadweight loss from supra-
competitive pricing, 1.26 % of long distance industry revenue, represents S696 million in 1991, no small
sum. 	The 0.25% estimate using the largest elasticities represents S138 million per year. 	However, a
33

sense of the ma:",itude of the deadweight loss can be discerned from the comparison between the benefits
and costs of pr~.;... regulation. 	Under perfect price regulation (an admittedly unattainable goal), AT&T's
prices would be equal to marginal costs and the elimination of the entire deadweight loss would be the
benefit. 	Imperfect regulation that allowed AT&T to set prices midway between marginal cost and the
profit maximizing price level would eliminate about three-quaners of the potential deadweight 10ss.16
The welfare costs due to price regulation of AT&T can only be gauged in reference to studies
of past deregUlatory actions. Mathios and Rogers (1989, 1990) and Kaestner and Kahn (1990) found that
AT&T prices were 7% lower in states that use price-eap incentive regulation compared to traditional rate-
of-return regulation. 	This effect presumably occurred. at least in pan. because the less restrictive
regulatory structure induced cost reductions. Supply estimates reponed above provide some evidence that
the movement to price-eaps from rate-of-return regulation reduced AT&T's long distance prices by
approximately 	1.6%. 	Olley 	and 	Pues 	(1992) 	find 	that competition 	spurs 	telecommunications
manufacturing plants to become more efficient. 	Ying and Shin (1993) found that the local telephone
companies' costs fell due to the divestiture of AT&T. 	Crandall (1991) estimates that telephone industry
costs would be 53.5 billion (10%) higher in 1988 than they would have been without the introduction of
competition in telecommunications. 	Kwoka (1993) estimates that each percentage point decrease in
AT&T's market share has led to a more than one-third percent (0.36%) improvement in productivity.
Since 	these 	productivity 	increases 	provide continuing 	benefits 	into 	the 	future. 	seemingly 	small
improvements quickly become substantial cost savings.
I%e resulting upper-bound estimate of the benefit from regulation would be 0.96% of industry
revenues or 5530 million per year (the lower-bound estimate would be 0.20% of industry revenues or
$104 million per year).
34

vu. 	Conclusion
This paper estimates supply and demand relationships in the long distance telecommunications
market over the 1988 to 1991 period and interprets them in terms of potential welfare losses due to supra-
competitive pricing. The estimates of industry demand elasticity are similar to those reported elsewhere.
The prices of key inputs explain much of the variation in output prices and these are taken to be valid
instruments for demand estimation. 	Estimates of firm specific demand elasticities are fairly high; lower
bounds are -2.9 for AT&T and -6.6 for the accs. 	These elasticities are best· interpreted as shon-run
elasticities. 	Long-run price-eost margins are inferred from these estimates via the Lerner index from
which the potential deadweight loss due to supra-eompetitive pricing is calculated. 	Estimates of this
potential loss appear to vary between 0.25% and 1.26% of industry revenue ($138 million to $696 million
per year).
The above analysis brings new information to the question of whether further deregulation of
AT&T is likely to be efficient. 	As stated in the introduction, the benefits of regulation are the possible
Iimiting of the deadweight loss due to supra-eompetitive pricing while the costs are inefficient production
due to regulatory distortion of profit incp.ntives. 	In the early 19805, competition may have been
insufficient to constrain AT&T prices to long-run marginal cost. 	In the intervening decade, competitive
pressures on AT&T increased substantially. 	This paper estimates that, for the 1988 to 1991 period,
competition constrained the potential deadweight loss from supra-eompetitive prices to between 0.25 %
and 1.26% of total revenues. 	Competitive pressures have continued to mount and it is likely that the
potential deadweight loss currently is smaller.
3S

References
Braeutigam, Ronald R. and John C. Panzer, "The Effects of the Change from Rate-of-Return Regulation
to Price-eap Regulation," American Economic Review 83 (May 1993) 191-198.
Brennan, 	Timothy, 	J., 	"Why 	Regulated 	Firms 	Should 	Be 	Kept 	Out 	of 	Unregulated 	Markets:
Understanding the Divestiture in U.S. v. AT&T," Antitrust Bulletin 32 (1987) 741-793.
Carlton, Dennis W. and Jeffery M. 	Perloff, Modem Industrial Organization (Scott, Forsman and
Company, Glennview, fL, 1990).
Crandall, Robert W., After the Breakup (Brookings Institution, Washington, DC, 1991).
Deaton, Angus and John Muellbauer, Economics and Consumer Behavior (Cambridge University Press,
New York, NY, 1980).
Egan, Bruce L. and Leonard Waverman, "The State of Competition in Telecommunications," in Barry
G. Cole, ed, After the Breakup (Columbia University Press, New York, NY, 1991).
Faulhuber, Gerald R., Telecommunications in Turmoil (Balinger Publishing Co.,Cambridge, MA, 1987).
Federal Communications Commission, "Competition in the Interstate Interexchange Marketplace," CC
Docket No. 90-132, Report and Order, 6 FCC Red. (1991)'.
Griboff, "New Freedom for AT&T in the Competitive Long Distance Market," Federal Communications
Law Journtll44 (1992).
Hausman, Jerry, "Specification Tests in Econometrics," Econometrica 46 (1978) 1251-1272.
Hausman, Jerry, Timothy Tardiff and Alexander Belefante, "The Effects of the Breakup of AT&T on
Telephone Penetration in the United States," American Economic Review 83 (May 1993) 178-190.
Huber, 	Peter W., 	Michael 	K. 	Kellogg and John Thorn, 	The Geodesic Network: 1993 Repon on
Competition in the Telephone Industry (Geodesic Publishing: Washington, DC, 1993).
Kaestner. Robert and Brenda Kahn. "The Effects of Regulation and Competition on the Price of AT&T
(ntrastate Telephone Service" Journtll of Regulatory Economics (1990) 363-377.
36

Kaserman, David, L. and John W. Mayo, "Competition for 800 service," Telecommunications Policy
(1991) 395-410.
Kaserman, David, L., John W. Mayo, and Joseph E. Flynn, "Cross-Substitution in Telecommunications:
Beyond the Universal Service Fairy Tale," Journal of Reguialory Economics 2 (1990).
Klepper, Steven and Edward E. Leamer, "Consistent Sets of Estimates for Regressions with Errors in
All Variables," Econometrica 52 (1984) 163-183.
Kraushaar, Jonathan M., "Fiber Deployment Update: End of Year 1992," FCC mimeo 1993.
Kwoka, John E., "The Effects of Divestiture, Privatization and Competition on Productivity in U.S. and
U. K. Telecommunications," t;.c view of Industrial Organization (1993) 49-61.
Landes, William M. and Richard A. Posner, "Market Power in Antitrust Cases." Harvard Law Review
94 (1981) 937-983.
Levin. Stanford L., "The State of Competition in Telecommunications," in Barry G. Cole, ed, After the
BreaJcup (Columbia University Press, New York, NY, 1991).
Liston, Catherine, "Price-eap versus Rate-of-Return Regulation," Journal of Regulatory Economics 5
(1993) 25-48.
Maddala. G.S., Introduction to Econometri:-s (Macmillan Publishing Company: New York, NY, 1988).
Mathios, Alan and Roben P. Rogers. "The Impact of Alternative Forms of State Regulation of AT&T
on Direct Dial Long Distance Telephone Rates," Rand Journal of Economics (1989) 437-53.
Mathios, 	Alan and Roben P. 	Rogers, 	"The Impact and Politics of Entry Regulation on Intrastate
Telephone Rates," Journal of Regulatory Economics 2 (1990) 53-68.
Mitchell, 	Bridger, 	Incremental 	Costs of Telephone 	Access and 	Use 	(Santa 	Monica, 	CA: 	Rand
Corporation, 1990).
Olley. G. Steven and Ariel Pakes, "The Dynamics of Productivity in the Telecommunications
Equipment Industry", CES Working Paper 92-2, 1992.
37

Parsons, Steven G. and Michael R. Ward, "Telecommunications Bypass and the 'Brandon Effect, '" FTC
Bureau :.: Economics Working Paper No. li9, 1993.
Posner, Richard A., "The Social Costs of Monopoly and Regulation," Journal of Political Economy
(1975) 807-27.
Scherer, F. M. and David Ross, Industrial Market Structure and Economic Performance, Third Edition
(Houghton Mifflin Company: Boston, MA 1990).
Selwyn, Peter, Nina Cornell, Martin G. Taschdjian and John R. Woodbury, "The State of Competition
in Telecommunications," in Barry G. Cole, ed., After tM Breakup (Columbia University Press,
New York. NY. 1991).
Shin. Richard. "Crors-subsidies in Telephone Pricing." unpublished mimeo. 1993.
Statistics of Communications Common Carriers. Industry Analysis Division. FCC. various years.
Stigler. George J., "A Theory of Oligopoly, " Journal of Political Economy (1964) ~1.
Taylor, Lester D., Telecommunications DeTNJJUi: A ~u~ and Critique (Balinger Publishing Co.,
Cambridge MA 1980).
Taylor. William, "Effects of Competitive Entry in the U.S. Interstate Toll Markets: An Update." filed
with the Federal Communications Commission in CC Docket No. 91-141 (1991).
Taylor. William and Lester Taylor. "Post-Divestiture Competition in the United States," American
Economics Review 83 (May 1993) 185-190.
Temin. Peter, 1M Fall of the Bell System (Cambridge University Press, New York, NY 1987).
Temin. Peter and Geoffrey Peters, "Cross-Subsidies in the Telephone Network," Willamete Law Review
21 (l985a).
Temin. 	Peter and Geoffrey Peters, 	"Is History Stranger than Theory? The Origins of Telephone
Separations," Economic History 75 (1985b).
38

Waverman. J Al'\nard, "V.S. Interexchange Competition," in Roben Crandall and Kenneth Flamm. eds.,
Changing rl.e Rules: 	Technological Chan8~. 	International Competition and Regulations in
Communication (Brookings Institution, Washington, DC, 1989)
Wenders, lohn T., The Economics of Telecommunications (Balinger Publishing Co., Cambridge, MA.
1987).
Ying, lohn and Richard Shin, "Costly Gains to Breaking Up: LECs and the Baby Bells," Review of
Economics and Statistics (forthcoming, 1993).
39

Yagure 1
AT&T's Basket 1 Price Cap & Price Index
(Nominal Prices)
100 -------------------------
99 r-
98.-
~-"t...:..:..:..:. 	.
97 ;....,
96 ~
95.-
94.-
93-
92~
91 ~
Price Cap
- 	Price Index
90 ---------------------------
May-H9 Jul·89 Oct-H9 Dec-89 Feb-90 May·90 Jul-90 Sep-9Q Noy-90 Feb-91 Apr-91 Jun-91 Sep-91 NoY-91
40

Fagure 2
Deadweight Loss from Supra-eompetitive Pricing
LR 	SR 	Esl
D 	D 	= D
p 	-----------------------
LR
MC
MC lmpIied -----
Q
41
Deadweight Loss
La
MR
SR
MR

Fllure 3
Long Distance Price Variables
(Real Prices)
Price
120 -------------------------
- Coaluuclcd AT&T Price
CODl'lruClcd acc Price
• 	Bultel 1 AT&T Price
110'-----~~~~~-- 	-----------------------------------------
100 ------------------~~-.-------------------------------
"'-
90 ----------------------------~------,~~~
1991	1990	1989
80 ---.......------------------------
1988
Year
42

Fagure 4
Long Distance Discounts
(Real Prict'S)
Price
110 -,------------------------
- 	Res. &: Small Bus.
Reach Out America
100 	----------------------------------------------------.
-------------------------.	901---------------------~~---
-.0. . 	.	.........	. 	.	.... 	'" 	.....
80 1------------------------------------------------------.
I
1991	1990
70 .......--------------------------
1989
Year
43

Table 1
Supply Price F.quatiom
2SLS Estimates - 240 Observatiom
January 1988 - December 1991
AT&T 	aee
Variable 	Price 	Price
Price Cap 	-0.016 1 	-0.007
(0.006) 	(0.007)
Minutes of Use 	-0.048 	-0.040
(0.039) 	(0.037)
Price of Switching 	0.189) 	0.117
Equipment 	(0.087) 	(0.093)
Price of Transmission 	0.368 1 	0.717 1
Equipment 	(0.112) 	(0.094)
Telecommunications 	0.301~ 	0.054
Workers Wage 	(0.151) 	(0.129)
Yield to Maturity on 	0.048 	0.503~
Corporate Bonds 	(0.404) 	(0.256)
Price of Switched Access 	0.13 II 	0.194 1
(0.022) 	(0.021)
Price of SpeciaJ Access 	0.08I1 	0.049 1
(0.020) 	(0.008)
First-Qrder 	0.789 1 	0.762 1
Autocorrelation 	(0.040) 	(0.042)
Adjusted R' 	.970 	.977
Standard errors are in parentheses and superscripts denote
significance levels for a two-tailed test if less than 10%.
Not reponed are coefficients of state dummy variables.
44

Table 2
Industry Quantity Demanded
2SLS Estimates • 63 Obstnations
June 1986 • August 1991
National 	National 	National 	National
Variable 	Quantity 	Quantity 	Quantity 	Quantity
Constant 	-1.127 	-II.781' 	-2.256 	4.737
(4.387) 	(5.533) 	(4.071) 	(4.652)
Long Distance Price 	-0.899 1 	-0.569 1 	-0.622 1 	-0.603 1
(0.085) 	(0.137) 	(0.1~3) 	(0.127)
LocaJ Service Price 	-0.849 1 	-1.024 1 	-0.221 	-0.373
(0.242) 	(0.247) 	(0.281) 	(0.323)
Telephone Set 	-0.9461 	-0.295
Price 	(0.293) 	(0.302)
Income 	1.703 1 	2.66f1 	1.499 1 	1.797 1
(0.272) 	(0.409) 	(0.243) 	(0.383)
Time Trend 	0.042 1 	0.0321
(0.014) 	(o.Ol1n
Durbin- Watson 	2.092 	2.273 	2.367 	2.398
Adjusted RZ 	.984 	.985 	.987 	.986
Standard errors are in parentheses and superscripts denote significance levels for a two-
tailed test if less than lOCI. 	Not reponed are coefficients of.month dummy variables.
45

Table 3
Lower Level Demand Equations
2SLS Estimates· 240 Observations
January 1988· December 1991
i
I
I
I
AT&T 	Alllmtruments I 	Instruments Exclude	I
Demand I
Carrier Access Prices	I
I
OCC I
Market 	AT&T	Market 	AT&T 	I 	acc	,
Variable 	Share 	Price 	Price I
Share 	Price 	Price	I
I
Regression 	Regression 	Regression : Regression 	Regression 	Regression
I,I
Own-Price 	-1.16' 	-5.87- 	-4.23- 	-2.15 1 	-4.68- 	-3.78-
(0.31 ) 	(0.49)
Cross-Price 	1.44 1 	6.09- 	5.lb- 	2.42 1 	4.91- 	4.61-
(0.31) 	(0.49)
Auto- 	0.J9' 	0.25 1 	0.25' 	0.31' 	0.25 1 0.26'
correlation 	(0.06) 	(0.06) 	(0.06) 	(0.06) 	(0.06) 	(0.06)
Adjusted R' 	.774 	.987 	.988 	.644 	.986 	.988
I
I
,,
acc 	All InstnJments 	I 	InstnJments Exclude	I
Demand I
Carrier Access Prices	I
I
Market 	AT&T 	acc: 	Market 	AT&T 	acc	I
Variable 	Share 	Price 	Pri' 	Share 	Price 	Price	ce 1
I
Reeression 	Regression 	Regression l Regression 	Regression 	Regression
I
Cross-Price 	2.47 1 	13.47· 	7.9/ 	5.14 1 	8.73- 	6.99-
(0.72) 	0.19)
Own-Price 	-3.09 1 	-13.95- 	-11. n- 	-5.73' 	-9.27· 	-8.87-
(0.71 ) 	(1.18)
Auto- 	0.61 1 	0.31 ' 	0.33 1 	0.48 1 	0.36 1 	0.37'
correlation 	(0.05) 	(0.06) 	(0.06) 	(0.06) 	(0.06) 	(0.06)
Adjusted R' 	.550 	.987 	.988 	.227 	.988 	.986
Standard errors are in parentheses and superscripts denote significance levels for a two-tailed test if
less than 10%. 	Estimates with asterisks are the implied elasticities from the coefficients of reverse
regressions. 	The underlying coefficients are always significant at the one percent level. 	Not
reponed are coefficients of state and month dummy variables.
46

Tnble 4
Unconditional Own-Price EIast£dties and Lerner Indices
Elastidties 	Lemer Indices
AT&T 	DCC 	AT&T 	OCC
AU InstnJmentai Variables
Market Share 	-1.97 	-4.00 	0.508 	0.250
Regression
AT&T Price 	-6.65 	-14.86 	0.150 	0.067
Regression
OCC Price 	-5.00 	-12.63 	0.200 	0.079
Regression
Restricted Instrument Set
Martet Share 	-2.92 	-6.64 	0.337 	0.151
Regression
AT&T Price 	-5.45 	-10.16 	0.183 	0.098
Regression
OCC Price 	-4.55 	-9.78 	0.220 	0.102
Regression
47

Table .~
Potential Deadwei&ht Lou as a Percent of Current Revenues
under Various Price Markup Assumptions
Aslumina a -0.65 Industry Demand Elasticity
P 	- Me 	DWL	0 	--
Po REV
0.350 	4.99%
0.325 	4.22%
0.300 	3.53%
0.275 	2.91 %
0.250 	2.37%
0.225 	1.89%
0.200 	1.47%
0.175 	1.10%
0.150 	0.80%
0.125 	0.55%
0.100 	0.34%
0.075 	0.19%
0.050 	0.08%
0.025 	0.02%
48