ECONOMIC ANALYSIS GROUP

DISCUSSION PAPER

                                                    

 

 

 

 

 

The proposed merger of AT&T and T-Mobile:  Are there unexhausted scale

economies in U.S. mobile telephony?

 

 

By

Yan Li and Russell Pittman

EAG 12-2    April 2012

 

 

EAG Discussion Papers are the primary vehicle used to disseminate research from economists in the Economic Analysis Group (EAG) of the Antitrust Division.  These papers are intended to inform interested individuals and institutions of EAG’s research program and to stimulate comment and criticism on economic issues related to antitrust policy and regulation.  The Antitrust Division encourages independent research by its economists.  The views expressed herein are entirely those of the authors and are not purported to reflect those of the United States Department of Justice.

 

Information on the EAG research program and discussion paper series may be obtained from Russell Pittman, Director of Economic Research, Economic Analysis Group, Antitrust Division, U.S. Department of Justice, LSB 9004, Washington, DC 20530, or by e-mail at russell.pittman@usdoj.gov.  Comments on specific papers may be addressed directly to the authors at the same mailing address or at their e-mail address.

 

Recent EAG Discussion Paper and EAG Competition Advocacy Paper titles are available from the Social Science Research Network at www.ssrn.com.  To obtain a complete list of titles or to request single copies of individual papers, please write to Kathy Burt at kathy.burt@usdoj.gov or call (202) 307-5794.  In addition, recent papers are now available on the Department of Justice website at http://www.justice.gov/atr/public/eag/discussion-papers.html.

 The proposed merger of AT&T and T-Mobile:  Are there unexhausted scale economies in U.S. mobile telephony?

 

By

 

Yan Li* and Russell Pittman+

 

 

 

 

 

ABSTRACT

 

From the beginning, the debate on the likely results of the proposed acquisition of T-Mobile USA by AT&T focused more on the claims of the parties that “immense” merger efficiencies would overwhelm any apparent losses of competition than on the presence or absence of those losses, and the factors that might affect them, such as market definition.  The companies based their “economic model” of the merger on estimates of efficiencies on AT&T’s “engineering model”, without addressing the credibility of the results of the latter in the context of the economics literature on the telecommunications sector.  In this paper we first argue that the economics literature on economies of scale (especially) and economies of density in mobile telephony suggests caution in expecting such massive cost reductions from increasing the size of an already very large firm.  We then present new econometric evidence from an international data base supporting the notion that most large mobile telephone service providers have reached the point of constant or even (rarely) declining returns to scale.

 


1. Introduction

AT&T’s proposed $39 billion acquisition of T-Mobile USA (TMU) raised serious concerns for US policymakers, particularly at the Federal Communications Commission (FCC) and the Antitrust Division of the Justice Department (DOJ), which shared jurisdiction over the deal.  Announced on March 20, 2011, the acquisition would have combined two of the four major national providers of mobile telephony services for both individuals and businesses, with the combined firm’s post-acquisition share of revenues reportedly over 40 percent, Verizon a strong number two at just under 40 percent, and Sprint a distant number three at around 20 percent.[1]

As usual in a large and complex merger deal, there were questions about how exactly to define markets, in both the product and geographic dimensions.  In the former category were the questions of both whether prepaid and postpaid services and whether individual and business (“enterprise”) services might constitute separate markets; in the latter category it was noted that the four major national suppliers competed along some dimensions at both the local and national levels, and at the local levels both their shares and the identity of additional competitors varied to some degree.  Both DOJ (in its Complaint, filed on August 31, 2011) and the staff of the FCC (in its Staff Analysis and Findings, filed on November 29, 2011) argued that the merger would be highly concentrating and anticompetitive regardless of the choices made on these more specific market definition questions.

An interesting and unusual aspect of the debate that ensued following the announcement of the merger, through the issuing of the DOJ complaint and the (negative) FCC staff report, until the companies abandoned the merger project on December 19, 2011, was that the merging firms did not seem to devote much effort to the presentation of evidence against the market definitions that implied these high levels of concentration.  The companies’ principal economic experts, Dennis Carlton, Allan Shampine, and Hal Sider (hereinafter CSS), in their Declaration filed at the FCC with the merger proposal, stated the basis of their support for the merger in very clear terms:  “We conclude that the proposed transaction will promote competition by enabling the merged firm to achieve engineering-based network synergies that increase network capacity beyond the levels that AT&T and T-Mobile USA could achieve if the two companies continued to operate independently.”  (Declaration at ¶7)  The Declaration proceeded to argue that the merged firm would continue to face competition from a variety of sources, as well as that the standard “unilateral effects” analysis expected to be engaged in by DOJ would be misleading for a variety of industry- and situation-specific reasons (Declaration at ¶9), but did not take strong issue with the basic market definition and concentration analysis that it deemed likely flow from the Horizontal Merger Guidelines of DOJ and the Federal Trade Commission.[2]  In their Reply Declaration (June 9), CSS stated the point even more starkly:  “These consumer benefits are independent of the scope of the product and geographic markets and, as a result, the precise definitions of the product and geographic markets are not central to the evaluation of the proposed transaction” (¶59).

In their FCC application and later in more detail at the agencies, the merging companies highlighted two categories of efficiencies that would result from the merger.  The first category, apparently not quantified in a single summary figure, constituted cost savings that were a primary input into the companies’ “engineering model,” which in turn was a primary input into the CSS “economic model” of the impact of the merger.  (Both models were in fact introduced subsequent to the filing of the merger application and the CSS Declaration.)  These cost savings were summarized by the companies as follows:

“[T]he transaction will enable the merged firm to create far greater capacity on the combined network than the two networks could achieve on their own by (i) creating a denser network with additional cell sites that increase aggregate capacity; (ii) increasing spectrum available to provide service by consolidating redundant GSM network control channels; (iii) increasing the efficiency of existing spectrum through ‘channel pooling’; (iv) making greater use of underutilized networks; and (v) freeing up spectrum for more spectrally efficient services and thereby expanding the number of areas in which such services will be deployed.  In so doing, the transaction will give the combined company much-needed flexibility to relieve capacity constraints by enabling it to optimize its use of spectrum on a market-by-market basis….”  (Application, at p. 42)

The second category of efficiencies, estimated in the Application at $39 billion total, “with an annual run rate on the order of $3 billion from year three forward,” included more efficient use of the cell towers owned by the two companies (including the sale of those no longer needed post-merger), as well as economies in combining retail operations, customer support, marketing, and procurement, the latter including “handsets as well as network equipment and infrastructure”) (Application at pp. 51-52).

In this paper, we seek to place these very large merger efficiency claims in the context of the cost structure of mobile telephony and other network industries.  Section 2 of the paper examines more closely the efficiencies claims made by the parties and their implications.  Section 3 considers these claims in the context of the economic literature on economies of scale and economies of density.  We suggest there that the parties’ claims of efficiencies, while certainly not outside the realm of possibility, are remarkably high for a company already as large and with as high a market share as AT&T.

We proceed to present a new set of econometric results on scale economies for 22 mobile carriers from seven countries (involving two North-American, three European and two East-Asian countries) over the time period 1998-2007, using a parametric stochastic frontier approach to identify the firms’ economies of scale.  Section 4 of the paper elaborates the methodology, including theoretical models and empirical applications.  Section 5 describes the data, and Section 6 discusses the empirical results.  Section 7 concludes with policy implications. 

2. The Efficiencies Claimed by the Merging Firms

As noted above, the public record appears to contain no single dollar figure for the value of the efficiencies claimed by the companies in the “engineering model”.  However, the companies’ Application at the FCC described them as “immense” (pp. 12 and 23), and the subsequent filing by the companies in “Opposition … to Petitions to Deny” (June 10, 2011) promised that the merger would create “immense new capacity that will provide enormous benefits to consumers” (at 1; emphasis supplied).  We can make more precise inferences concerning the magnitude of these claims of efficiencies by combining information from various sources.

First, the FCC staff report summarizes the companies’ redacted analysis as concluding that following the merger, “prices would fall between 3.8 and 9.4 percent” (Staff Analysis at ¶136).  Second, a redacted letter from AT&T counsel to the FCC, accompanied by a redacted slide presentation titled “Competitive Effects of the AT&T MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  T-Mobile Transaction”, notes that the economic model presented by CSS assumes linear demand in order to be on the conservative side regarding the pass-through of cost reductions to consumers.  Since standard economic theory holds that firms facing linear demand pass through approximately fifty percent of cost changes,[3] this by itself would suggest that the parties are claiming cost reductions of between 7.6 and 18.8 percent.

But the same presentation implies that MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  as usual in merger simulation MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  the model would predict a post-merger price increase if there were no efficiencies:  “Our analysis indicates that the cost and quality benefits are more than sufficient to counter any upward pricing pressure...” (slide 3, emphasis added).  Similarly, AT&T economist Mark Israel noted in the FCC’s “Workshop on the Economics of the Proposed AT&T MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  T-Mobile Merger” that “what really drives the results is the change in the AT&T marginal cost, and whether that’s large enough to overcome the competitive effects” (transcript p. 251; emphasis supplied).  Finally, in their Reply Declaration (June 9, 2011), CSS discuss the “Upward Pricing Pressure Analysis” calculated by Stephen Salop, Stanley Besen, Stephen Kletter, Serge Moresi, and John Woodbury and submitted on May 31 on behalf of Sprint’s challenge to the merger.  CSS report that, after making some but not all of their proposed corrections to the Sprint model, the model predicts upward pricing pressure “not very far from the threshold that is used at the Antitrust Division in determining whether GUPPI levels raise unilateral effects concerns” MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  by which a footnote explains that they mean 5 percent (¶87).

Thus the efficiencies claimed must reduce the price not from its current level but from its higher but-for level MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  adding a figure “not very far from” 5% into the mix, so perhaps 6-7%.  Furthermore, the letter and the Application strongly emphasize that the efficiencies included in this pricing analysis should be considered an underestimate of the true efficiencies to be expected from the merger MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  that the efficiencies included in the “economic model” do not, for example, include the $39 billion of efficiencies included in the second category discussed above.

Thus we may confidently infer that the parties were claiming something at least in the range of 13.6 to 25.8 percent in cost reductions from the merger, and perhaps higher.  (Recall that the merging parties characterize them as “immense”.)  We argue in the next section of the paper that this is a fairly remarkable claim for an already very large provider of mobile telephone services.

First, however, let us note in passing that the FCC staff was quite sceptical of these efficiency claims MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  labelling them “seriously flawed”, “implausible”, and “extremely sensitive to adjustments” (Staff Analysis at ¶138) MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  though on the grounds of a close examination of the engineering model on which they were based rather than on the grounds examined in this paper.  In particular, according to the Staff Analysis, the companies calculate merger efficiencies by projecting the costs of the two firms going forward independently, as demand increases and capacity levels are reached at the level of local markets, and then comparing those “but-for” costs with the costs of the combined firm under the same conditions.  The staff objects first to the companies’ extrapolation of their calculations for fifteen local markets to the totality of areas served by the two firms, as “the fifteen markets chosen by the Applicants do not seem to be particularly representative” (¶169).

More fundamentally, though, the staff believes that the methodology used to calculate costs in the but-for world “contains a serious flaw in the cell-splitting algorithm that appears to cause the model to greatly overestimate incremental costs, and the overestimate is much greater for the standalone firms than the merged firm” (¶147).  In particular, the methodology modelled by which the companies would address growing tightness in capacity “is not rational and does not reflect how any wireless provider would operate or model its business….  As a result, the Applicants significantly overstate the estimated cost savings of the proposed transaction” (¶¶175-76).

Finally, let us note that at least one component of the second category of claimed efficiencies appears to confuse private benefits with public benefits.  One of the sub-categories is labelled “cost savings … from combining the networks”, and one element of this subcategory is described as follows:  “savings from a reduction in interconnection and toll expenses as a result of switching [T-Mobile calls] to existing AT&T facilities where possible for transport.”[4]  There is no estimate reported of the value of these efficiencies, though the sub-category of which they are a part is estimated to provide efficiencies of $10 billion in net present value.[5]

The FCC staff report includes the “reduction in interconnection and toll expenses” as one of many elements of the efficiencies claims where it is difficult to determine what portion of the efficiencies should be counted as savings in fixed charges and what portion in variable charges, under the traditional thinking that savings in the latter are more likely to be passed along downstream in the form of lower prices than are savings in the former.[6]

It appears, however MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  it is impossible to be sure from the public record MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  that the FCC staff critique does not go far enough in this case.  In fact the actual amount of interconnection and toll expenses that are paid by TMU to AT&T pre-merger and would be internal to the firm post-merger should count as a transfer, not a savings in resources.  The only portion of this flow that constitutes a true efficiency comes from the fact that the internal transfer price for this service should be at marginal cost rather than something higher, in which case “double marginalization” is avoided and the merged firm would have the incentive to expand output accordingly.  Both CSS and their fellow AT&T consultants Robert Willig, Jonathan Orszag, and Jay Ezrielev note that the merger eliminates pre-merger double marginalization in this area, but neither these nor other statements sponsored by the merging firms states that the resulting merger efficiencies are only those flowing from this particular incentive for output expansion, rather than the entire volume of cost savings from the internalization of these flows.[7]

3. The Literature on Scale Economies, Particularly in Network Industries

Early, seminal texts by Marshall (1920) and Viner (1931) discussed the importance of “net internal economies of large-scale production”.  Kahn (1971) and others applied this concept to the possibility that a single firm might be a “natural monopoly … [where there] is an inherent tendency to decreasing unit costs over the entire extent of the market.”  Some commentators have applied this concept to the AT&T/T-Mobile merger.  For example, Goldfarb (2011), noted that

The mobile wireless industry is characterized by economies of scale and scope.  In a static market, it would be less costly and/or more efficient to build out and operate a single network instead of multiple networks with partially duplicative facilities; to give a single provider use of a large block of spectrum rather than giving a number of providers use of a subset of that block; and to design and mass produce a single suite of handsets rather than making handsets for smaller groups of customers using many different standards and network technologies.

This sounds reasonable, and yet it seems to suggest at least two follow-up questions.  First, are we considering here local or national economies of scale?  And second, do these economies exist for all relevant levels of demand MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  so that the firm may be a “natural monopoly” MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  or are they exhausted at some point, after which increases in output are accompanied by proportional or even greater than proportional increases in cost?

Regarding the first question, for decades now the economic literature on network industries has made the useful distinction between economies of overall system size and economies of density.  Walters (2007) summarizes the difference succinctly in the railways context:

A significant development in all of this research [in “rail cost analysis”] was refining the distinction between economies of scale and density.  The latter is the behavior of costs as output expands over a given network, whereas economies of scale focuses on the behavior of costs if the network size increases as output expands.

Similarly, Caves, Christensen, and Tretheway (1984) make the distinction in their analysis of airline costs:

“We define returns to density as the proportional increase in output made possible by a proportional increase in all inputs, with points served, average stage length, average load factor, and input prices held fixed….  We define returns to scale as the proportional increase in output and points served made possible by a proportional increase in all inputs, with average stage length, average load factor, and input prices held fixed.”  (emphasis in original)

It seems useful, indeed important, to make this distinction in the present case of the market for mobile telephony.  A significant portion of the economies claimed by AT&T for the merger seem to be some variant on economies of density, involving as they do the more efficient utilization of (especially) spectrum and cell towers in particular metropolitan areas.  However, other claimed economies are firm-wide in scope and independent of the density of local areas, including those associated with marketing, customer service, procurement, and overall company administration.

All of this raises the crucial question:  How reasonable is it to assume that under current (i.e. without the merger) conditions, AT&T and T-Mobile enjoy substantial unexhausted economies of density and size of national operations?  Recall that the fragmentary estimates made public suggest claims of at least 10-15 percent reductions in cost, and perhaps 25 percent or more.  Absent an econometric examination of mobile telephony for the US as a whole as well as for individual metropolitan areas, what can we infer from the existing literature?

The literature on at least one other network industry is not particularly supportive.  In the freight railways sector, the literature suggests that in Western Europe, the railways have reached efficient scale in terms of system size but have not exhausted all available economies of density.[8]  In the much larger and more intensely operated US freight railroads, however, the most recent study suggests that the largest companies have reached or are reaching minimum efficient scale in both system size and density (Christensen Associates, 2008).

What about mobile telephony?  Here we are hampered by both the very dynamic nature of the industry MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  so that even fairly recent data may not well reflect economies going forward MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  as well as the paucity of publicly available data, especially regarding economies of density.  One of the most knowledgeable of U.S. analysts, Ingo Vogelsang (2010), in a survey article laments the lack of convincing studies but seems to lean toward believing in constant returns to scale, even with regard to economies of density:

“The case for constant returns comes from the observation that a doubling of traffic leads to cell splitting and increases the number of cells required even in the same area, roughly doubling costs.”

Writing eight years earlier, Cave, Majumdar, and Vogelsang (2002) seem to agree:

“If the econometric models were to be relied on, there would be a strong case for pursuing infrastructure competition throughout the network.  If the engineering models were to be relied on, infrastructure competition in local areas would make sense only in very dense networks, where economies of density are exhausted….  [C]onceptually, there could exist two important ranges of natural monopoly.

In the first stage, the natural monopoly property can be weak.  In this range, economies of scale and scope are almost exhausted and sunk costs tend to be small.  In this situation, competition is likely to be beneficial, because it leads to pressure on costs, prices and innovation.  Competition is also likely to occur here, because in most telecommunications markets demand is moderately to strongly inelastic.  Thus, we can expect duplicate network investments, associated with some cost inefficiency and excess capacity, but possibly lower prices than under regulated monopoly.  This is the case of long-distance services, mobile telephony and local business services in downtown areas of industrialized countries” (pp. 30-10).

In fact the parties’ own engineering experts concede at least the possibility of the effective exhaustion of economies of density in one area; they note that “the percentage gains from channel pooling diminish as the size of the pool increases”.[9]  And all of this would be consistent with the findings of Gabel and Kennet (1991) that for the local fixed wire network, economies of density are exhausted in densely populated urban areas.[10]

Most of the existing empirical literature features observations at the firm level, with output measured as number of subscribers or, less frequently, revenues or airtime minutes.  These studies tend to find constant returns to scale or even decreasing returns to scale for the largest operators MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  i.e., generally U-shaped cost curves.  These papers include McKenzie and Small (1997), examining five US firms; Gagnepain and Pereira (2007), three Portuguese firms; Vendruscolo and Alves (2009), 38 Brazilian firms (the number declining over time); and Nam, Kwon, Kim, and Lee (2009), three Korean firms.  The only paper we have found directly related to economies of density in local mobile telephone markets is Foreman and Beauvais (1999), which uses internal, local market level data for GTE, at the time a supplier in primarily rural areas with coverage of a few urban areas in California and Florida, and finds economies of scale with respect to the number of subscribers, controlling for airtime minutes per cell site.  Even this paper finds economies for mobile telephony “below those that have been estimated for wireline technology.”

In (unsatisfying) summary, the literature suggests that it is unlikely that a firm as large as AT&T MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  and perhaps T-Mobile as well MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  is operating at a point on its overall enterprise cost curve of substantial unexhausted economies of scale.  With regard to economies of density at the metropolitan level, the little evidence available is more supportive of the presence of at least some unexhausted economies of density in some locations, though not in the most dense urban areas.  But even in that case we may justifiably ask whether if one believes the evidence of “immense” economies presented by the merging companies, one should take the next step and consider whether mobile telephony in U.S. cities is a “natural monopoly”, with declining costs throughout the relevant regions of demand?

In fact one party challenging the merger at the FCC, the Ad Hoc Telecommunications Users Committee, makes this precise point, both on its own and in an attached Declaration by economist Lee Selwyn.  They two begin by noting at least an implicit tension between the merging parties’ contention that only this merger can ease the capacity constraints faced by each firm separately, and the merging parties’ simultaneous contention that the merger could not be anticompetitive because much smaller firms like MetroPCS and Cellular South would easily expand in order to discipline any post-merger price increase.  They then go on to make the argument that if firms as large as AT&T and T-Mobile can achieve “immense” cost reductions through merging their operations, presumably AT&T, T-Mobile, and Verizon would achieve even greater economies by merging.  In other words, if the efficiencies claimed by the parties are correct, then unless (for example) economies of density in the industry are coincidentally exhausted at the point of around 40 percent of current capacity MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  and the parties make no such claim, much less provide evidence for it MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  then mobile telephony may well be a sector characterized by natural monopoly, and the FCC should rely on regulation rather than competition going forward.

The FCC staff analysis does not appear to address this point.  CSS, in their Reply Declaration, simply state that in that case every industry in which merging parties claim economies of scale or scope must be a natural monopoly MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  hardly a serious response, given the traditional relevance of the concept of natural monopoly in infrastructure sectors such as telecommunications.  Certainly overall it seems that the economic analysis that uses the engineering model and its associated efficiencies claims in support of a precompetitive outcome to this merger is implicitly assuming a very specific, and arguably unlikely, state of the world regarding the existence of unexhausted economies of firm size and metropolitan area density in the provision of mobile telephony, with no econometric support.

In the next section we describe our own analysis of economies of scale in mobile telephony at the firm level, using the most recent data available.

4. Methodology

In this section, we describe briefly the measure, in the literature, used for economies of scale. We then focus on the frontier model used in this present study for the evaluation of firm scale efficiency as our preferred measure of scale economies.[11]  

The most convenient measure for economies of scale in the literature is the return to scale (RTS),[12] which is, by geometrical point of view, defined as measuring how accurately the distance in input space reflects the distance in output space.[13]  In our case with multiple inputs, we require to look at ray average productivity increases (decreases) with an increase in the overall inputs scale when RTS exceeds (falls short of) unity.[14]

The other measure documented in the recent literature is named scale efficiency (SE), which measures the ray average productivity at the observed inputs scale on the production frontier relative to the maximum ray average productivity attainable at a vector of inputs bundle (x) characterized by constant return to scale (CRS). The SE is equal to one, indicates the firm at the most productive scale size as defined by Banker et al. (1984). That is also the only point where the CRS prevails, and SE is equal to the overall RTS.  Elsewhere SE is below unity regardless of whether scale (inputs) elasticity is greater than or less than one (Ray, 1998).  Given the multiple inputs application in this study, we prefer the later measure of scale economies, using SE, which can be estimated using stochastic frontier approach.

Given a panel dataset we have, this study considers a true fixed effects stochastic production frontier model specified for panel data as proposed by Greene (2004, 2005):[15]

                        y it = α i +β' x it + v it u it ,i=1,2,...,N;t=1,2,...,T MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamyEamaaBaaaleaacaWGPbGaamiDaaqabaGccqGH9aqpcqaHXoqydaWgaaWcbaGaamyAaaqabaGccqGHRaWkcqaHYoGycaGGNaGaamiEamaaBaaaleaacaWGPbGaamiDaaqabaGccqGHRaWkcaWG2bWaaSbaaSqaaiaadMgacaWG0baabeaakiabgkHiTiaadwhadaWgaaWcbaGaamyAaiaadshaaeqaaOGaaiilaiaaysW7caWGPbGaeyypa0JaaGymaiaacYcacaaMe8UaaGOmaiaacYcacaGGUaGaaiOlaiaac6cacaGGSaGaaGjbVlaad6eacaGG7aGaaGjbVlaadshacqGH9aqpcaaIXaGaaiilaiaaysW7caaIYaGaaiilaiaac6cacaGGUaGaaiOlaiaacYcacaaMe8Uaamivaaaa@66AD@ ,                               (1)

where y it MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamyEamaaBaaaleaacaWGPbGaamiDaaqabaaaaa@38FE@  is the (logarithm) output of the i-th firm in the t-th year; x it MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamiEamaaBaaaleaacaWGPbGaamiDaaqabaaaaa@38FD@  is a vector of (logarithm) inputs of the i-th firm in the t-th year; β MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqOSdigaaa@378E@  is a vector of unknown parameters to be estimated; α i MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqySde2aaSbaaSqaaiaadMgaaeqaaaaa@38A6@  is the firms’ individual specific effects capturing all time invariant effects including unobserved heterogeneity, v it MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamODamaaBaaaleaacaWGPbGaamiDaaqabaaaaa@38FB@  is the random error term which is assumed to be i.i.d. (independent and identically distributed) with N(0, σ v 2 ) MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamOtaiaacIcacaaIWaGaaiilaiabeo8aZnaaDaaaleaacaWG2baabaGaaGOmaaaakiaacMcaaaa@3D34@  distribution and independent of u it MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamyDamaaBaaaleaacaWGPbGaamiDaaqabaaaaa@38FA@   MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  the inefficiency term that vary freely through time with N( μ it , σ u 2 ) MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamOtaiaacIcacqaH8oqBdaWgaaWcbaGaamyAaiaadshaaeqaaOGaaiilaiabeo8aZnaaDaaaleaacaWG1baabaGaaGOmaaaakiaacMcaaaa@404C@  distribution.

We firstly conduct three likelihood ratio (LR) specification tests to identify the most favourable functional form for our parametric stochastic frontier analysis (SFA) model (see Table 1 for the test results).  The first model specification test is for the Cobb-Douglas functional form versus the Translog functional form.[16] The second test is related to time effects that indicate the existence of significant technical change.[17]  The last test is to identify the distribution of the inefficiency error term.[18]  According to our LR test results, the preferred empirical frontier model for our study is a translog stochastic production function with the inclusion of time effects, and normal-truncated normal distribution in the inefficiency errors.  The SFA model is specified as:

                            ln y it = α i + k=1 3 β k ln x kit + β τ T+ 1 2 { k=1 3 j=1 3 β kj ln x kit ln x jit + β ττ T 2 } + k=1 3 β kτ ln x kit T + v it u it ,( i=1,2,...,N;t=1,2,...,T ) MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=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@C49A@ ,                    (2)

The more expanded form can be written as:

           lnOutpu t it = α i + β 1 ln L it + β 2 ln M it + β 3 ln K it + β 4 T + 1 2 β 11 (ln L it ) 2 + 1 2 β 22 (ln M it ) 2 + 1 2 β 33 (ln K it ) 2 + 1 2 β 44 T 2 + β 12 lnL l it n M it + β 13 ln L it ln K it + β 23 lnM l it n K it + β 14 ln L it T+ β 24 ln M it T+ β 34 ln K it T+ v it u it ,( i=1,2,...,N;t=1,2,...,T ) MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=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@4546@ ,       (3)

where the one output and three inputs of the i-th firm in the t-th year are all measured in quantities and in logarithm form (Lit, Mit and Kit denote the quantities of the labour input, material input and capital input of the i-th firm in the t-th year, respectively). The time trend variable, T, is included to capture the technical change which may affect the location of production frontier.

To calculate firm SE scores, we exploit the output-oriented measure of scale efficiency developed by Ray (1998) from an empirically estimated single output and multiple inputs translog production function.[19]  The output-oriented SE for the i-th firm in the t-th year can be calculated using

                                                   S E it ( x it )=exp{ ( 1 ε it ( x it ) ) 2 2β } MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaam4uaiaadweadaWgaaWcbaGaamyAaiaadshaaeqaaOWaaeWaaeaacaWG4bWaaSbaaSqaaiaadMgacaWG0baabeaaaOGaayjkaiaawMcaaiabg2da9iGacwgacaGG4bGaaiiCamaacmaabaWaaSaaaeaadaqadaqaaiaaigdacqGHsislcqaH1oqzdaWgaaWcbaGaamyAaiaadshaaeqaaOWaaeWaaeaacaWG4bWaaSbaaSqaaiaadMgacaWG0baabeaaaOGaayjkaiaawMcaaaGaayjkaiaawMcaamaaCaaaleqabaGaaGOmaaaaaOqaaiaaikdacqaHYoGyaaaacaGL7bGaayzFaaaaaa@5359@ ,                                           (4)

where εit(xit) is the overall RTS of the i-th firm in the t-th year, estimated using equation (5) described below; β= k=1 3 j=1 3 β kj MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqOSdiMaeyypa0ZaaabCaeaadaaeWbqaaiabek7aInaaBaaaleaacaWGRbGaamOAaaqabaaabaGaamOAaiabg2da9iaaigdaaeaacaaIZaaaniabggHiLdaaleaacaWGRbGaeyypa0JaaGymaaqaaiaaiodaa0GaeyyeIuoaaaa@4794@ , assumed to be negative to ensure S E it ( x it )1 MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaam4uaiaadweadaWgaaWcbaGaamyAaiaadshaaeqaaOGaaiikaiaadIhadaWgaaWcbaGaamyAaiaadshaaeqaaOGaaiykaiabgsMiJkaaigdaaaa@408F@ .  The SEit scores are between zero and one, where a value equals one indicating the firm at the most productive scale size.  Below we demonstrate the calculation of the RTS from the specified translog production function.

Given the preferred translog production function, which is non-homothetic and imposes no restrictions on production technology, the elasticity of the output with respect to the inputs is a function of the inputs and time trend variable.[20]  That is for any given inputs level, the RTS is ε it ( x it )= k=1 3 ln y it ln x kit MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqyTdu2aaSbaaSqaaiaadMgacaWG0baabeaakmaabmaabaGaamiEamaaBaaaleaacaWGPbGaamiDaaqabaaakiaawIcacaGLPaaacqGH9aqpdaaeWbqaamaalaaabaGaeyOaIyRaciiBaiaac6gacaWG5bWaaSbaaSqaaiaadMgacaWG0baabeaaaOqaaiabgkGi2kGacYgacaGGUbGaamiEamaaBaaaleaacaWGRbGaamyAaiaadshaaeqaaaaaaeaacaWGRbGaeyypa0JaaGymaaqaaiaaiodaa0GaeyyeIuoaaaa@52BE@ , where yit is the output and xkit is a vector of three inputs. Thus, the RTS for the i-th firm in the t-th year can be calculated directly from the empirically estimated parameters in the production function (4) using

                                        ε it ( x it )= k=1 3 ( β k + j=1 3 β kj ln x jit + β kτ T ) MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqyTdu2aaSbaaSqaaiaadMgacaWG0baabeaakmaabmaabaGaamiEamaaBaaaleaacaWGPbGaamiDaaqabaaakiaawIcacaGLPaaacqGH9aqpdaaeWbqaamaabmaabaGaeqOSdi2aaSbaaSqaaiaadUgaaeqaaOGaey4kaSYaaabCaeaacqaHYoGydaWgaaWcbaGaam4AaiaadQgaaeqaaaqaaiaadQgacqGH9aqpcaaIXaaabaGaaG4maaqdcqGHris5aOGaciiBaiaac6gacaWG4bWaaSbaaSqaaiaadQgacaWGPbGaamiDaaqabaGccqGHRaWkcqaHYoGydaWgaaWcbaGaam4Aaiabes8a0bqabaGccaWGubaacaGLOaGaayzkaaaaleaacaWGRbGaeyypa0JaaGymaaqaaiaaiodaa0GaeyyeIuoaaaa@5FCA@ .                                          (5)       

Detailed data information on mobile carriers’ operation output and inputs are described in the next section.

5. Data

The output and inputs data used for calculating firms’ scale efficiencies are extracted from mobile carriers’ annual reports.  An unbalanced panel dataset is in turn compiled for twenty-two mobile network providers, over 1998-2007, from seven countries selected to represent different regions and major economic powers, including the USA, Canada, the UK, France, Germany, China and Korea.[21]  Table 2 lists the names of these mobile carriers, their served market regions, and the time periods observed.[22]

·         Output measures

Output value is measured by total operating revenue from the mobile service segment.  It typically consists of network service revenue (i.e. voice services, messaging and other data services) and sales of terminal equipment, but excludes interest income, disposals, capital gains/losses and dividends.[23]

Output price is used to deflate the value of output into a quantity measure that is consistent across carriers. Output price is based on the national average mobile price index, calculated for each sample country and collected from ITU. 

Output quantity is measured by total operating revenue deflated by the average mobile price index in each country in US$s at 2000 prices.

·         Input measures

(i) Labour quantity is the number of employees working in the mobile service segment.

(ii) Material cost is measured by non-personnel operating expenses in the mobile service segment.  It consists primarily of consumption of goods and merchandise, services obtained from outside suppliers, materials, cost of acquisition and maintenance of customer services, and administration.

Material price is used to deflate material cost into quantity measure that is consistent across carriers. Material price is proxied by a producer price index (PPI) for a manufacturing products basket (2000 = 100).  The PPI data are available for all sample countries in this study at the OECD online statistical database.

Material quantity is PPI deflated non-personnel operating expenses.

(iii) Capital cost is measured by the total depreciation and amortization of property, plant and equipment in the mobile service segment in each year.

Capital price is proxied by the weighted average cost of borrowings reported in carrier annual reports.[24]

Capital quantity is calculated by dividing capital cost by capital price.

Table 3 summarizes the statistics of the output and inputs (labour, material and capital) over 1998-2007 (i.e. calculated means across 22 mobile carriers by year). The first four columns present the money values in US$s, and followed by the converted quantities in the same order.  We use the quantity measures of the output and inputs in our SFA model estimating scale efficiencies for those twenty-two mobile carriers over their observation years.

6. Results

In this section, we present the results of our SFA estimation regarding scale economies for the sample mobile network companies.

Table 4 presents the SE and RTS scores for each firm across their observation years.[25]  The results suggest that on average, most mobile network operators in our sample (16 out of 22) operate in the region of constant returns to scale, with their average SE scores insignificantly different from one.[26]  The RTS estimates indicate that on average, almost all observed firms operate in either constant (i.e., the mean average RTS cross-year are equal to or very close to 1) or mildly increasing returns to scale.  For example, KTF, LG Telecom, Orange France and Rogers show an average RTS at 1.12, 1.14, 1.06 and 1.02, respectively.

In Graph 1, the firms are presented in an ascending order of firm sizes measured by their 2007 total revenue in the mobile service segment and showing their estimated SE and RTS scores.  Note that AT&T and T-Mobile USA both are among the firms that appear to be operating in the range of constant returns.[27]  There are three firms MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  Verizon Wireless, T-Mobile Germany, and China Mobile MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  operating in the range where we estimate decreasing returns to scale, with an average RTS significantly less than one (in fact below 0.9).[28]  

 

Graph 1: Average SE and RTS across firms

Graph 1:  The relationship between firm size (as measured by total revenues) and estimated or interpolated economies of scale, results from the stochastic frontier analysis.  Most firms show a measure very close to 1.0, which means neither economies nor diseconomies of scale.  However, the largest firms show measures significantly less than 1.0, suggesting that they are operating in a region of diseconomies of scale.

 

Table 5 presents, using the most recent data we have, a snap-shot of 2007 measures of the firms’ economies of scale along with the national regulatory standards for their mobile network markets.[29]  The results further emphasize that in a regulatory system featured by active competition among privatized oligopolies, the firms generally operate within the range of constant returns to scale; we find no evidence for substantial unexhausted economies of scale in mobile telecoms.  

7. Conclusion

Using stochastic frontier analysis, in this study, we investigated scale efficiency as a measure of scale economies for twenty-two mobile carriers from seven countries over the period 1998-2007.  We did not find substantial unexhausted scale economies in mobile telephony in general, especially for firms of medium to large size. Most firms we studied operate at their most productive scale size over their observation years, with a few exhibiting slightly increasing returns along with a few large operators exhibiting decreasing returns.

Regarding the proposed merger of AT&T and T-Mobile, obviously it would have been desirable to have the more localized data necessary to test for the presence of unexhausted economies of density.  The conjectures in the literature lean toward the view the in mobile telephony, as in railways, the large, dense networks have exhausted available economies of density, but the empirical evidence for this view is lacking.  Nevertheless our results concerning economies of firm size are consistent with those in the literature, suggesting that it is unlikely that T-Mobile, and very unlikely that AT&T, are currently operating in a range where large firm-level economies related to activities such as procurement, marketing, customer service, and administration would have been achievable due to the merger.  Regarding both measures, the presence of “immense” unexhausted economies for the two firms seems unlikely indeed.  On this basis (and on this basis alone), our results support the decision of DOJ to challenge the merger and the scepticism expressed by the FCC staff.

AAI, “The Effect of AT&T’s Acquisition of T-Mobile Is Likely to Substantially Lessen Competition.”  Washington, DC:  American Antitrust Institute, August 2011.

Banker, R.D., A. Charnes, and W.W. Cooper (1984), “Some Models for Estimating Technical and Scale Inefficiencies in Data Envelopment Analysis,” Management Science, 30(9): 1078-1092.

Baumol, W.J. (1976). “Scale Economies, Average Cost, and the Profitability of Marginal Cost Pricing.” In Public and Urban Economics, ed. R.E. Grieson. Lexington, Mass.: D.C. Heath/Lexington Books.

Bruner, F., K. Eades, R. Harris and R. Higgins (1998), “Best practices in estimating the cost of capital: Survey and synthesis”, Financial Practice and Education, 8(1): 13-28.

Cave, Martin, Sumit Majumdar, and Ingo Vogelsang, “Structure, Regulation and Competition in the Telecommunications Industry.”  In Cave, Majumdar, and Vogelsang, eds., Handbook of Telecommunications Economics, vol. 1:  Structure, Regulation and Competition.  Amsterdam:  North-Holland, 2002.

Caves, Douglas, Laurits Christensen, and Michael Tretheway, “Economies of density versus economies of scale:  why trunk and local service airline costs differ,”  Rand Journal of Economics 15 (1984), 471-489.

Chambers, R. G. (1988), Applied Production Analysis, Cambridge: Cambridge University Press.

Chang, Howard, David Evans, and Richard Schmalensee, “AT&T/T-Mobile:  Does Efficiency Really Count?”  CPI Antitrust Chronicle 2 (2011).

Christensen Associates, A Study of Competition in the U.S. Freight Railroad Industry and Analysis of Proposals that Might Enhance Competition.  Study for the Surface Transportation Board.  Madison, WI:  Laurits R. Christensen Associates, 2008.

Coelli, T.J. and S. Perelman (1996), “Efficiency Measurement, Multiple Output Technologies and Distance Functions: With Application to European Railways,” European Journal of Operational Research, 117: 326-339.

Dennis, Scott, and Wayne Talley, eds., Railroad Economics (Research in Transportation Economics, vol. 20).  Oxford:  Elsevier, 2007.

Färe, R., S. Grosskopf, and C. Lovell (1994), Production Frontiers, Cambridge University Press, Cambridge.

Farrell, M. (1957), “The measurement of productive efficiency,” Journal of the Royal Statistical Society, 120(3): 253-290.

FCC, Annual Report and Analysis of Competitive Market Conditions With Respect to Mobile Wireless, Including Commercial Mobile Services (14th Report).  Washington:  Federal Communications Commission, May 20, 2010.

Gabel, David, and Mark Kennet, “Estimating the Cost Structure of the Local Telephone Exchange Network.”  Report No. 91-16, National Regulatory Research Institute, October 1991.

Gagnepain, Philippe, and Pedro Pereira, “Entry, costs reduction, and competition in the Portuguese mobile telephony industry.”  International Journal of Industrial Organization 25 (2007), 461-481.

Goldfarb, Charles, “The Proposed AT&T/T-Mobile Merger:  Would It Create a Virtuous Cycle or a Vicious Cycle?”  Washington, DC:  Congressional Research Service, May 10, 2011.

Greene, W.H. (2004a), “Fixed and Random Effects in Stochastic Frontier Models,” Journal of Productivity Analysis, 23(1): 7-32.

---- (2004b), “Distinguishing Between Heterogeneity and Inefficiency: Stochastic Frontier Analysis of the World Health Organization’s Panel Data on National Health Care Systems,” Health Economics, 13: 959-980.

---- (2005), “Reconsidering Heterogeneity in Panel Data Estimators of the Stochastic Frontier Model,” Journal of Econometrics, 126(2): 269-303.

Grunes, Allen, and Maurice Stucke, “Antitrust Review of the AT&T/T-Mobile Transaction.”  Federal Communications Law Journal 64 (2011), 47-85.

Kahn, Alfred E., The Economics of Regulation:  Principles and Institutions.  New York:  Wiley, 1971.

Li, Y. and B. Lyons (2012), “Market structure, Regulation and the Speed of Mobile Network Penetration”, ESRC Centre for Competition Policy, Working Paper: 12-03.

McKenzie, David, and John Small, “Econometric Cost Structure Estimates for Cellular Telephony in the United States.”  Journal of Regulatory Economics 12 (1997), 147-157.

Marshall, Alfred, Principles of Economics (8th Ed.).  Philadelphia:  Porcupine Press, 1982 (reprint of 1920 edition).

Nam, Changi, Youngsun Kwon, Seongcheol Kim, and Hyeongjik Lee, “Estimating scale economies of the wireless telecommunications industry using EVA data.”  Telecommunications Policy 33 (2009), 29-40.

Pittman, Russell, “Options for Restructuring the State-Owned Monopoly Railway.” In Dennis and Talley (2007).

Ray, D. (1998), Development Economics, Princeton: Princeton U. Press.

Savignat, M.G. and Chris Nash, “The case for rail reform in Europe:  Evidence from studies of production characteristics of the rail industry.”  International Journal of Transport Economics 26 (1999), 201-217.

Truong G., G. Partington and M. Peat (2008), “Cost of capital estimation and capital budgeting practice in Australia”, Australian Journal of Management, 33: 1-6.

Vendruscolo, Maria Ivanice, and Tiago Wickstrom Alves, “Study of the scale economy in the Brazilian mobile telecommunication sector after privatizations” (in Portuguese).  Revista Contabilidade & Financas (2009).

Viner, Jacob, “Cost Curves and Supply Curves.”  In George Stigler and Kenneth Boulding, eds., Readings in Price Theory.  Chicago:  Irwin, 1952, for the American Economic Association (from Zeitschrift für National MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaqcLbwaqaaaaaaaaaWdbiabl2==Ubaa@39D8@ konomie, 1931).

Vogelsang, Ingo, “The relationship between mobile and fixed-line communications:  A survey.”  Information Economics and Policy 22 (2010), 4-17.

Waters, William G. II, “Evolution of Railroad Economics.”  In Dennis and Talley (2007).

Weyl, E. Glen, and Michal Fabinger, “A Restatement of the Theory of Monopoly.”  Unpublished paper, June 2011, available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1851326

Table 1: Hypotheses tests for parameters of the SFPF models and final model estimation results

Null hypothesis

LogLH1

(unrestricted model)

LogLH0

(restricted model)

LR statistics

Critical value

Decision

A. H0: Cobb-Douglas model is preferred (βkj = β = βττ = 0, k, j = 1, 2, 3)

-19.137

-32.115

25.96

18.31

Reject H0, in favour

of translog model

B. H0: no technical change (time effects) (βτ = β = βττ = 0, k = 1, 2, 3)

-19.137

-28.351

18.43

11.07

Reject H0, significant

technical change

exists

C. H0: normal-half normal model is preferred (μ=0)

43.483

25.634

35.69

3.84

Reject H0, normal-

truncated normal

model is preferred

Maximum-likelihood estimates of translog normal-truncated normal model

 

 

 

 

 

 

 

 

 

 

 

 

βl

0.068

(0.107)

 

βττ

0.003

(0.002)

 

 

 

βm

0.770

(0.179)

 

βlm

-0.316

(0.095)

 

 

 

βk

0.054

(0.079)

 

βlk

0.138

(0.054)

 

 

 

βτ

-0.030

(0.038)

 

βmk

-0.018

(0.056)

 

 

 

βll

0.045

(0.039)

 

β

0.018

(0.012)

 

 

 

βmm

0.220

(0.074)

 

β

-0.023

(0.020)

 

 

 

βkk

-0.091

(0.022)

 

β

0.012

(0.009)

 

 

 

 

 

 

 

 

 

 

 

 

 

λ

6.231

(1.714)

 

 

 

 

 

 

 

σ

0.568

(0.119)

 

 

 

 

 

 

 

Note: standard errors are presented in the parentheses next to the coefficients. The first-order coefficient estimates may be interpreted as production elasticities at the sample mean because the data was mean-corrected prior to estimation.

 

Table 2: List of the names, operating markets and years for observed mobile carriers

Unit-code

Firm name

Regional market

Periods

covered

No. years

Unit-code

Firm name

Regional market

Periods

covered

No. years

D1

China Mobile             

China

98-07

10

D12

E-plus         

DE

02-07

6

D2

China Unicom             

China

98-07

10

D13

Orange France   

FR

02-07

6

D3

Vodafone UK     

UK

99-07

9

D14

Bouygues Telecom

FR

01-07

7

D4

O2UK           

UK

99-07

9

D15

SFR            

FR

98-07

10

D5

Orange UK       

UK

02-07

6

D16

SK Telecom     

Korea

98-07

10

D6

Orange FT       

--

99-07

9

D17

KTF            

Korea

98-07

10

D7

T-mobile       

--

01-07

7

D18

LG Telecom     

Korea

98-07

10

D8

T-mobile Europe 

Europe

05-07

3

D19

Sprint Nextel  

US

98-07

10

D9

T-mobile USA    

US

05-07

3

D20

Verizon Wireless

US

98-07

10

D10

O2Germany      

DE

00-07

8

D21

Rogers Wireless

Canada

98-07

10

D11

Vodafone Germany

DE

02-07

6

D22

Telus (Mobility)

Canada

99-07

9

Note: Consolidated information is used for DMU6 and DMU7.

 

Table 3: Summary statistics of output and inputs for twenty-two mobile carriers from 1998 to 2007

 

OutputQ

LQ

MQ

KQ

OutputV

LV

MV

KV

 

Mean

1998

23.13

7890.11

15.07

70.74

2217.42

291.24

1467.84

458.79

1999

34.30

9980.46

23.86

88.26

3351.74

362.25

2315.60

571.66

2000

46.94

14744.36

29.51

126.65

4694.43

468.33

2951.40

809.09

2001

61.24

18211.00

38.64

223.92

6282.31

656.59

3870.76

1321.48

2002

69.66

17336.25

41.30

537.59

7265.85

697.77

4114.11

3172.00

2003

86.20

17802.60

49.82

251.06

9183.40

852.65

5036.10

1437.22

2004

101.63

20665.05

58.69

345.41

11068.81

1033.38

6152.10

1858.38

2005

119.37

22620.09

69.49

425.45

13342.51

1191.90

7587.28

2237.05

2006

134.32

23816.05

75.05

460.50

15398.83

1511.33

8467.77

2640.58

2007

152.31

25495.64

86.14

462.32

17812.51

1672.61

9991.60

2864.78

Average annual

change rate

23.82%

14.83%

22.16%

33.47%

26.58%

21.82%

24.46%

32.34%

Note: OutputQ is the quantity of output; LQ, MQ and KQ are the quantities of labour, material and capital inputs, respectively. OutputV is the value of output; LV, MV and KV are the values of labour, material and capital inputs, respectively; all money values are measured in million (US$s).


Table 4: Summary of DEA and SFA SE & RTS estimates across units over observed years

 

China Mobile

China Unicom

Vodafone UK

O2UK

Orange UK

 

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

1998

0.999

0.943

0.87

0.716

0.955

0.884

1999

1

0.894

0.819

0.853

0.928

0.853

1

0.996

0.965

0.91

1

1.005

2000

1

0.852

0.784

0.901

0.87

0.798

1

0.998

0.974

0.978

1

1.003

2001

1

0.866

0.795

0.865

0.907

0.831

0.977

0.987

0.937

0.944

0.998

0.973

2002

1

0.811

0.753

0.942

0.871

0.799

0.992

0.991

0.949

0.985

0.961

0.892

0.991

0.994

0.957

2003

0.999

0.829

0.766

0.986

0.924

0.849

0.998

0.999

0.98

0.988

1

1

0.999

0.998

0.978

2004

1

0.799

0.744

0.997

0.925

0.849

0.992

0.993

0.953

0.997

1

0.998

0.995

0.994

0.96

2005

0.966

0.788

0.736

0.994

0.91

0.835

0.999

0.997

0.971

1

0.999

0.987

1

1

1.004

2006

0.774

0.817

0.757

0.999

0.928

0.852

0.978

1

0.989

0.999

0.991

0.95

1

0.997

1.028

2007

0.833

0.852

0.784

1

0.93

0.855

0.968

1

1.004

0.988

1

0.999

1

0.991

1.052

Mean

0.967

0.87

0.804

0.925

0.915

0.84

0.989

0.995

0.969

0.976

0.994

0.979

0.997

0.996

0.996

 

Orange FT

T-mobile

T-mobile Europe

T-mobile USA

O2Germany

 

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

1999

0.901

0.996

0.968

2000

0.858

0.988

0.941

0.143

0.999

0.981

2001

0.936

0.971

0.908

0.645

0.916

0.84

0.988

1

1.005

2002

0.888

0.969

0.905

0.71

0.84

0.775

0.886

0.928

0.852

2003

0.926

0.981

0.926

0.907

0.977

0.918

0.879

0.992

1.05

2004

0.92

0.974

0.912

0.915

0.984

0.931

0.807

0.991

1.052

2005

0.931

0.984

0.932

0.872

0.973

0.91

0.866

0.994

0.957

0.989

0.986

0.935

0.849

0.994

1.042

2006

0.819

0.99

0.945

0.887

0.972

0.908

0.904

0.992

0.952

0.931

0.987

0.938

1

0.967

0.901

2007

0.884

0.992

0.953

0.845

0.972

0.909

0.902

0.991

0.95

0.968

0.989

0.943

1

1

0.993

Mean

0.896

0.983

0.932

0.826

0.948

0.884

0.891

0.992

0.953

0.962

0.987

0.939

0.819

0.984

0.984

 

Vodafone Germany

E-plus

Orange France

Bouygues Telecom

SFR

 

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

1998

0.947

0.986

0.935

1999

0.933

0.996

0.965

2000

0.977

0.995

0.962

2001

1

0.998

0.976

0.963

0.971

0.907

2002

1

0.992

0.953

0.883

0.898

0.823

1

0.999

1.019

0.92

1

0.994

0.985

0.992

0.951

2003

1

0.997

0.973

0.879

1

1.001

1

0.995

1.039

0.984

1

0.998

0.964

1

0.991

2004

1

0.991

0.948

0.778

0.994

1.041

1

0.993

1.044

0.998

0.999

1.013

0.961

1

0.988

2005

1

0.991

0.948

0.835

0.994

1.043

1

0.984

1.068

0.998

0.994

1.04

0.999

0.997

1.028

2006

0.977

0.993

0.956

0.891

0.993

1.044

1

0.984

1.069

0.999

0.998

1.021

0.998

0.997

1.031

2007

1

0.998

0.973

0.978

0.985

1.067

1

0.949

1.124

0.99

0.995

1.04

0.998

0.983

1.072

Mean

0.996

0.994

0.959

0.874

0.977

1.003

1

0.984

1.06

0.984

0.998

1.012

0.973

0.991

0.983

 

SK Telecom

KTF

LG Telecom

Sprint Nextel

Verizon Wireless

 

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

1998

1

0.967

0.9

1

0.885

1.188

1

0.872

1.2

0.756

0.989

0.943

1

0.877

0.804

1999

0.957

0.997

0.969

1

0.887

1.187

0.907

0.922

1.153

0.762

0.972

0.91

1

0.869

0.798

2000

1

0.993

0.954

1

0.959

1.111

0.988

0.946

1.128

0.844

0.952

0.881

0.927

0.853

0.785

2001

1

0.996

0.966

1

0.966

1.1

1

0.934

1.141

0.942

0.949

0.876

0.908

0.83

0.767

2002

1

0.997

0.972

1

0.974

1.087

0.972

0.951

1.121

0.906

0.965

0.897

1

0.843

0.777

2003

1

1

0.997

1

0.981

1.075

0.766

0.974

1.087

0.916

0.975

0.914

1

0.861

0.791

2004

0.999

0.999

0.981

1

0.974

1.087

0.94

0.939

1.135

0.922

0.983

0.928

1

0.849

0.781

2005

1

1

1

1

0.974

1.088

0.883

0.947

1.126

0.885

0.987

0.938

1

0.861

0.791

2006

0.982

0.999

1.014

1

0.958

1.112

1

0.923

1.153

0.877

0.991

0.949

1

0.869

0.797

2007

0.967

0.998

1.023

1

0.926

1.149

0.814

0.894

1.181

0.791

0.996

0.964

1

0.89

0.815

Mean

0.993

0.986

0.958

1

0.949

1.118

0.927

0.93

1.142

0.872

0.975

0.918

0.984

0.86

0.791

 

Rogers wireless

Telus (Mobility)

Mean

 

SED

SES

RTS

SED

SES

RTS

SED

SES

RTS

1998

0.932

0.998

0.974

0.928

0.941

0.966

1999

0.851

0.997

0.969

0.859

0.987

0.938

0.918

0.957

0.961

2000

0.826

0.999

0.986

0.817

0.985

0.934

0.876

0.956

0.944

2001

0.829

0.998

0.974

0.63

0.964

0.897

0.914

0.953

0.931

2002

0.82

0.996

1.034

0.873

0.988

0.94

0.938

0.948

0.922

2003

0.762

0.992

1.047

0.98

0.997

0.971

0.947

0.974

0.968

2004

0.777

1

1

1

0.998

0.978

0.95

0.969

0.966

2005

0.89

0.994

1.043

1

0.999

0.987

0.953

0.97

0.973

2006

1

0.978

1.08

1

1

0.992

0.955

0.969

0.975

2007

1

0.962

1.106

0.992

1

0.998

0.951

0.968

0.998

Mean

0.885

0.992

1.014

0.906

0.991

0.96

0.939

0.963

0.959

SES is the scale efficiency estimates from SFA; SED is the DEA counterparts as a robustness check.

RTS is the overall inputs returns to scale.

RTS = 1 indicates constant return to scale (CRS); RTS < 1 indicates decreasing returns to scale (DRS); and RTS > 1 indicates increasing returns to scale (IRS).

 


Table 5: Summary firms’ SE and RTS, 2007, and national regulatory standard for mobile telecoms

Firm

Ownership

Independent regulator

Number MNOs by 2007

SE (2007)

RTS (2007)

Average annual GDP growth

China

No

2

9.24%

 

China Mobile             

state

0.852

0.78

 

China Unicom             

state

0.930

0.85

 

Korea

Yes

3

4.33%

 

SK Telecom     

private (2002)

0.998

1.02

 

KTF            

private (2002)

0.926

1.15

 

LG Telecom     

private (2002)

0.894

1.18

 

Unite Kingdom

Yes

5

2.76%

 

Vodafone UK     

private

1.000

1.00

 

O2UK           

private

1.000

1.00

 

Orange UK       

private

0.991

1.05

 

Deutschland

Yes

4

1.34%

 

O2Germany      

private

1.000

1.00

 

Vodafone Germany

private

0.998

0.97

 

E-plus         

private

0.985

1.07

 

France

Yes

3

2.15%

 

Orange France   

private

0.949

1.12

 

Bouygues Telecom

private

0.995

1.04

 

SFR            

private

0.983

1.07

 

Unite Sates

Yes

6

2.98%

 

T-mobile USA    

private

0.989

0.94

 

Sprint Nextel  

private

0.996

0.96

 

Verizon Wireless

private

0.890

0.82

 

Canada

Yes

5

3.26%

 

Rogers wireless

private

0.962

1.11

 

Telus (Mobility)

private

 

 

1.000

1.00

 

 

Note: RTS = 1 indicates constant return to scale (CRS); RTS < 1 indicates decreasing returns to scale (DRS); and RTS > 1 indicates increasing returns to scale (IRS). SE = 1 indicates the most productive scale size.

 

 


Appendix A: DEA approach to scale efficiency

In the nonparametric DEA approach, firm scale efficiency can be simply calculated by the ratio of technical efficiency (TE) scores measured under constant returns to scale (CRS) over the TE scores measured under variable returns to scale (VRS), ie. SEi = TEic/TEiv. [30]  The TE of a firm is measured relative to the TE of all the other firms in the sample, subject to the restriction that all firms are on or below the production frontier. We calculate the output-oriented TE scores by solving the linear programme (LP) problem for each firm, with (one) output, yi, and three inputs, xi (xi = x1i, x2i, x3i), under an assumption of CRS.  The LP problem is specified as:

                           Maximize w.r.t. φ i ,λ: φ i MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqOXdO2aaSbaaSqaaiaadMgaaeqaaOGaaiilaiaaysW7cqaH7oaBcaGG6aGaaGjbVlabeA8aQnaaBaaaleaacaWGPbaabeaaaaa@41E1@

                           Subject to                  s λ s y s φ i y i 0 ,s=1,2,...,N MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaabeaeaacqaH7oaBdaWgaaWcbaGaam4CaaqabaGccaWG5bWaaSbaaSqaaiaadohaaeqaaOGaeyOeI0IaeqOXdO2aaSbaaSqaaiaadMgaaeqaaOGaamyEamaaBaaaleaacaWGPbaabeaakiabgwMiZkaaicdaaSqaaiaadohaaeqaniabggHiLdGccaGGSaGaaGjbVlaadohacqGH9aqpcaaIXaGaaiilaiaaysW7caaIYaGaaiilaiaac6cacaGGUaGaaiOlaiaacYcacaaMe8UaamOtaaaa@5415@

                                                                   x i s λ s x s 0, MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamiEamaaBaaaleaacaWGPbaabeaakiabgkHiTmaaqababaGaeq4UdW2aaSbaaSqaaiaadohaaeqaaOGaamiEamaaBaaaleaacaWGZbaabeaaaeaacaWGZbaabeqdcqGHris5aOGaeyyzImRaaGimaiaacYcaaaa@4408@

                                                                   λ s 0 MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeq4UdW2aaSbaaSqaaiaadohaaeqaaOGaeyyzImRaaGimaaaa@3B4F@ ,                                                          (1)

where λs is a vector of optimal weights defined by optimization LP problem (with a constraint β' x i =1 MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqOSdiMaai4jaiaadIhadaWgaaWcbaGaamyAaaqabaGccqGH9aqpcaaIXaaaaa@3C1B@  to avoid the problem of infinite solutions), such that the efficiency measure of the i-th firm is maximized subject to the restriction that the efficiencies of all firms must be less than or equal to one, and that all weights are non-negative. In the solution, 1 φ i MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaaGymaiablQNiWjabeA8aQnaaBaaaleaacaWGPbaabeaakiablQNiWjabg6HiLcaa@3E34@ , and the value of φ i 1 MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaeqOXdO2aaSbaaSqaaiaadMgaaeqaaOGaeyOeI0IaaGymaaaa@3A76@  measures the extent to which outputs could conceivably be increased using the same inputs MathType@MTEF@5@5@+=feaagaart1ev2aqatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaacbaqcLbyaqaaaaaaaaaWdbiaa=nbiaaa@37BA@  relative to other firms in the sample. The value of 1/ φ i MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaSGbaeaacaaIXaaabaGaeqOXdO2aaSbaaSqaaiaadMgaaeqaaaaaaaa@3995@  is the output-oriented TE score for the i-th firm.[31]  It satisfies 01 / φ i 1 MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaSGbaeaacaaIWaGaeSOEIaNaaGymaaqaaiabeA8aQnaaBaaaleaacaWGPbaabeaaaaGccqGHKjYOcaaIXaaaaa@3E66@ , where a value which equals one indicates a point on the frontier and hence a technically efficient firm, referring to the Farrell (1957) definition.

Note that the above TE is measured under the assumption of CRS in which assumes all firms are operating at an optimal scale (i.e. corresponding to the flat portion of the long-run average cost curve). However, when not all firms are operating at the optimal scale, the TE scores measured by CRS model is confounded by scale efficiencies (SE). Therefore, the assumption is relaxed to a VRS model proposed by Banker, Charnes and Cooper (1984) by adding a restriction, s λ s =1 MathType@MTEF@5@5@+=feaagaart1ev2aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLnhiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq=Jc9vqaqpepm0xbba9pwe9Q8fs0=yqaqpepae9pg0FirpepeKkFr0xfr=xfr=xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaWaaabeaeaacqaH7oaBdaWgaaWcbaGaam4CaaqabaaabaGaam4Caaqab0GaeyyeIuoakiabg2da9iaaigdaaaa@3D60@ , so as to permit the calculation of TE devoid of the potential SE effects.

 

Graph 2: Average SE scores across mobile network providers (DEA estimates)

Graph 2:  The relationship between firm size (as measured by total revenues) and estimated economies of scale, results from the data envelope analysis.  Using this methodology, both the smallest and the largest firms show measures less than 1.0, suggesting that they are operating in a region of diseconomies of scale, while the middle-sized firms – about half the firms in the sample – show a measure very close to 1.0, which means neither economies nor diseconomies of scale.

 

 

 

 


Appendix B: SFA Input Elasticity breakdown

 

Input Elasticity (returns to scale) estimates from SFA

 

China Mobile

China Unicom

Vodafone UK

O2UK

 

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

1998

0.28

0.33

0.26

0.87

0.66

-0.11

0.33

0.88

 

 

 

 

 

 

 

 

1999

0.36

0.23

0.23

0.82

0.62

-0.05

0.28

0.85

0.09

0.60

0.28

0.96

0.02

0.77

0.22

1.00

2000

0.37

0.18

0.24

0.78

0.61

-0.12

0.31

0.80

0.08

0.60

0.29

0.97

0.06

0.68

0.27

1.00

2001

0.34

0.32

0.14

0.80

0.52

0.07

0.24

0.83

0.20

0.56

0.17

0.94

0.16

0.57

0.24

0.97

2002

0.40

0.26

0.09

0.75

0.51

0.06

0.23

0.80

0.22

0.57

0.16

0.95

0.39

0.62

-0.11

0.89

2003

0.37

0.33

0.07

0.77

0.34

0.33

0.18

0.85

0.15

0.64

0.19

0.98

0.17

0.68

0.15

1.00

2004

0.41

0.28

0.06

0.74

0.34

0.35

0.16

0.85

0.23

0.66

0.06

0.95

0.17

0.70

0.13

1.00

2005

0.43

0.29

0.02

0.74

0.37

0.26

0.20

0.83

0.23

0.67

0.07

0.97

0.24

0.66

0.09

0.99

2006

0.39

0.32

0.05

0.76

0.36

0.28

0.21

0.85

0.23

0.69

0.07

0.99

0.35

0.40

0.20

0.95

2007

0.34

0.34

0.10

0.78

0.38

0.26

0.22

0.85

0.22

0.72

0.06

1.00

0.23

0.57

0.20

1.00

Mean

0.38

0.24

0.18

0.80

0.47

0.13

0.24

0.84

0.18

0.63

0.15

0.97

0.20

0.63

0.15

0.98

 

Orange UK

Orange FT

T-mobile

T-mobile Europe

 

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

1999

 

 

 

 

0.04

0.79

0.14

0.97

 

 

 

 

 

 

 

 

2000

 

 

 

 

0.04

0.72

0.19

0.94

 

 

 

 

 

 

 

 

2001

 

 

 

 

0.05

0.69

0.17

0.91

0.15

0.69

0.00

0.84

 

 

 

 

2002

0.18

0.57

0.21

0.96

0.08

0.69

0.14

0.90

0.31

0.71

-0.24

0.77

 

 

 

 

2003

0.16

0.58

0.23

0.98

0.06

0.72

0.15

0.93

0.00

0.78

0.14

0.92

 

 

 

 

2004

0.23

0.56

0.16

0.96

0.10

0.72

0.08

0.91

-0.03

0.84

0.12

0.93

 

 

 

 

2005

0.17

0.57

0.27

1.00

0.08

0.77

0.09

0.93

0.06

0.78

0.07

0.91

0.08

0.81

0.07

0.96

2006

0.15

0.58

0.29

1.03

0.05

0.84

0.05

0.95

0.10

0.71

0.10

0.91

0.14

0.73

0.09

0.95

2007

0.13

0.63

0.29

1.05

0.08

0.82

0.05

0.95

0.10

0.71

0.10

0.91

0.14

0.73

0.08

0.95

Mean

0.17

0.58

0.24

1.00

0.06

0.75

0.12

0.93

0.10

0.75

0.04

0.88

0.12

0.76

0.08

0.95

 

T-mobile USA

O2Germany

Vodafone Germany

E-plus

 

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

2000

 

 

 

 

0.38

0.07

0.53

0.98

 

 

 

 

 

 

 

 

2001

 

 

 

 

0.24

0.51

0.25

1.01

 

 

 

 

 

 

 

 

2002

 

 

 

 

0.63

0.50

-0.28

0.85

0.22

0.65

0.08

0.95

0.68

0.44

-0.29

0.82

2003

 

 

 

 

0.21

0.65

0.19

1.05

0.19

0.68

0.10

0.97

0.31

0.64

0.06

1.00

2004

 

 

 

 

0.21

0.72

0.13

1.05

0.27

0.67

0.00

0.95

0.27

0.73

0.04

1.04

2005

0.18

0.59

0.17

0.93

0.27

0.68

0.09

1.04

0.31

0.62

0.03

0.95

0.30

0.75

0.00

1.04

2006

0.19

0.55

0.20

0.94

0.66

0.22

0.03

0.90

0.33

0.57

0.05

0.96

0.34

0.67

0.04

1.04

2007

0.20

0.53

0.21

0.94

0.44

0.47

0.09

0.99

0.33

0.60

0.04

0.97

0.34

0.68

0.04

1.07

Mean

0.19

0.55

0.19

0.94

0.38

0.48

0.13

0.98

0.27

0.63

0.05

0.96

0.37

0.65

-0.02

1.00

 

Orange France

Bouygues Telecom

SFR

SK Telecom

 

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

1998

 

 

 

 

 

 

 

 

0.18

0.67

0.08

0.93

0.31

0.50

0.09

0.90

1999

 

 

 

 

 

 

 

 

0.08

0.73

0.15

0.97

0.11

0.72

0.14

0.97

2000

 

 

 

 

 

 

 

 

0.11

0.66

0.19

0.96

0.14

0.78

0.03

0.95

2001

 

 

 

 

0.30

0.16

0.52

0.98

0.21

0.51

0.18

0.91

0.20

0.76

0.00

0.97

2002

0.06

0.76

0.19

1.02

0.23

0.47

0.29

0.99

0.18

0.64

0.13

0.95

0.20

0.79

-0.02

0.97

2003

0.05

0.80

0.19

1.04

0.23

0.53

0.25

1.00

0.13

0.76

0.10

0.99

0.17

0.82

0.01

1.00

2004

0.08

0.83

0.14

1.04

0.21

0.55

0.25

1.01

0.16

0.76

0.07

0.99

0.21

0.79

-0.02

0.98

2005

0.03

0.92

0.12

1.07

0.16

0.64

0.23

1.04

0.12

0.83

0.08

1.03

0.22

0.81

-0.03

1.00

2006

0.07

0.89

0.10

1.07

0.24

0.61

0.17

1.02

0.16

0.77

0.10

1.03

0.20

0.80

0.02

1.01

2007

0.01

0.99

0.12

1.12

0.23

0.62

0.19

1.04

0.12

0.89

0.05

1.07

0.18

0.80

0.05

1.02

Mean

0.05

0.87

0.14

1.06

0.23

0.51

0.27

1.01

0.15

0.72

0.11

0.98

0.21

0.72

0.03

0.96

Lrts denotes labour input elasticity; Mrts denotes material input elasticity; and Krts denotes capital input elasticity.

RTS is the overall scale elasticity (returns to scale).

RTS = 1 indicates constant return to scale (CRS); RTS < 1 indicates decreasing returns to scale (DRS); and RTS > 1 indicates increasing returns to scale (IRS).

 

 

 


Input elasticity (returns to scale) estimates from SFA (Continued)

 

KFT

LG Telecom

Sprint Nextel

Verizon Wireless

 

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

1998

-0.09

0.94

0.33

1.19

-0.07

0.90

0.37

1.20

0.08

0.70

0.16

0.94

0.28

0.25

0.28

0.80

1999

-0.12

1.10

0.21

1.19

-0.01

0.94

0.22

1.15

0.10

0.70

0.11

0.91

0.29

0.22

0.29

0.80

2000

0.02

0.93

0.16

1.11

0.04

0.97

0.12

1.13

0.14

0.60

0.14

0.88

0.23

0.36

0.19

0.78

2001

0.03

0.93

0.14

1.10

0.08

0.81

0.25

1.14

0.13

0.59

0.16

0.88

0.27

0.34

0.15

0.77

2002

0.07

0.97

0.04

1.09

0.11

0.79

0.22

1.12

0.12

0.62

0.16

0.90

0.29

0.31

0.17

0.78

2003

0.14

0.93

0.00

1.07

0.23

0.74

0.11

1.09

0.13

0.64

0.15

0.91

0.30

0.33

0.16

0.79

2004

0.14

0.98

-0.03

1.09

0.12

0.91

0.11

1.14

0.14

0.65

0.14

0.93

0.32

0.28

0.17

0.78

2005

0.17

0.95

-0.04

1.09

0.15

0.85

0.12

1.13

0.11

0.78

0.05

0.94

0.32

0.28

0.19

0.79

2006

0.14

1.02

-0.04

1.11

0.11

0.92

0.12

1.15

0.09

0.92

-0.06

0.95

0.36

0.23

0.21

0.80

2007

0.05

1.14

-0.04

1.15

0.06

0.98

0.14

1.18

0.09

0.91

-0.03

0.96

0.34

0.26

0.22

0.82

Mean

0.05

0.99

0.07

1.12

0.08

0.88

0.18

1.14

0.12

0.69

0.10

0.92

0.30

0.29

0.20

0.79

 

Rogers wireless

Telus (Mobility)

Mean

 

 

 

 

 

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

Lrts

Mrts

Krts

RTS

 

 

 

 

1998

0.33

0.32

0.33

0.97

 

 

 

 

0.22

0.50

0.25

0.97

 

 

 

 

1999

0.35

0.26

0.36

0.97

0.40

0.29

0.25

0.94

0.18

0.56

0.22

0.96

 

 

 

 

2000

0.32

0.32

0.34

0.99

0.39

0.27

0.28

0.93

0.21

0.50

0.23

0.94

 

 

 

 

2001

0.37

0.27

0.33

0.97

0.51

0.19

0.19

0.90

0.23

0.50

0.20

0.93

 

 

 

 

2002

0.33

0.44

0.26

1.03

0.46

0.13

0.34

0.94

0.28

0.55

0.09

0.92

 

 

 

 

2003

0.32

0.51

0.22

1.05

0.41

0.17

0.39

0.97

0.20

0.61

0.15

0.97

 

 

 

 

2004

0.37

0.33

0.31

1.00

0.41

0.16

0.40

0.98

0.22

0.62

0.12

0.97

 

 

 

 

2005

0.28

0.52

0.25

1.04

0.40

0.20

0.39

0.99

0.21

0.65

0.11

0.97

 

 

 

 

2006

0.25

0.61

0.22

1.08

0.40

0.21

0.38

0.99

0.24

0.62

0.12

0.97

 

 

 

 

2007

0.21

0.66

0.23

1.11

0.41

0.23

0.36

1.00

0.21

0.66

0.13

1.00

 

 

 

 

Mean

0.30

0.42

0.29

1.01

0.42

0.21

0.33

0.96

0.23

0.58

0.15

0.96

 

 

 

 

Lrts denotes labour input elasticity; Mrts denotes material input elasticity; and Krts denotes capital input elasticity.

RTS is the overall scale elasticity (returns to scale).

RTS = 1 indicates constant return to scale (CRS); RTS < 1 indicates decreasing returns to scale (DRS); and RTS > 1 indicates increasing returns to scale (IRS).

 

 


Appendix C: Summary of firm SE and RTS over observation years

 

 

Firm

SE

RTS

Total Revenue

(2007, in bln US$s)

China Mobile

0.85

0.78

48.3

T-mobileGermany

0.95

0.88

47.7

Verizon Wireles

0.86

0.79

43.9

AT&T Mobility

1.00

0.98

42.7

OrangeFT

0.98

0.93

39.9

Sprint Nextel

0.98

0.92

34.7

T-mobileEurope

0.99

0.95

28.4

T-mobileUSA

0.99

0.94

19.3

OrangeFrance

0.98

1.06

13.7

SK Telecom

0.99

0.98

12.9

China Unicom

0.91

0.84

12.6

SFR

0.99

0.98

12.4

VodafoneUK

1.00

0.97

10.7

VodafoneGermany

0.99

0.96

10.5

O2UK

0.99

0.98

10.1

KTF

0.95

1.12

8.8

OrangeUK

1.00

1.00

8.5

Bouygues Teleco

1.00

1.01

6.5

Rogers wireless

0.99

1.02

5.1

LG Telecom

0.93

1.14

4.9

O2Germany

0.98

0.98

4.9

E-plus

0.98

1.00

4.1

Telus (Mobility)

0.99

0.96

4.0

Note: 2007 firm total operating revenues in the mobile service segment are used as a proxy of firm size.



* Norwich Business School and Centre for Competition Policy, University of East Anglia, Norwich, NR4 7TJ, U.K.  Email: li.yan@uea.ac.uk.

+ Antitrust Division, U.S. Department of Justice, Washington, DC, and New Economic School, Moscow.  Email: Russell.Pittman@usdoj.gov.  Co-author Pittman worked as a staff economist in the DOJ investigation of the AT&T/T-Mobile merger proposal.  However, this paper has been written using only public sources of information.  The views expressed are not purported to reflect the views of the Department of Justice.

 

[1] FCC (2010); AAI (2011); Grunes and Stucke (2012).  In its Complaint, DOJ emphasized not market shares but changes in the Herfindahl Hirschman Index (HHI):  “Nationally, the proposed merger would result in an HHI of more than 3,100 for mobile wireless telecommunications services, an increase of nearly 700 points.”  DOJ Complaint, at ¶25.

[2] Chang, Evan, and Schmalensee (2011) noted this as well:  “AT&T’s economists do not seem to be in significant disagreement with these product and geographic definitions” (i.e., those in the DOJ Complaint).

[3] See, for example, Weyl and Fabinger (2011).

[4] Declaration of Rick L. Moore, at ¶34.  Virtually the same language is in the parties’ merger application to the FCC, at p. 52.

[5] Declaration of Rick L., Moore, at ¶34.

[6] FCC Staff report, at ¶228.  Note that CSS (at ¶65-71) argue that this principle should not apply in this matter.

[7] CSS Reply Declaration at ¶143; Willig, Orszag, and Ezrielev Reply Declaration, at footnote 106.

[8] See the discussion in Pittman (2007), discussing, among others, the results of Savignat and Nash (1999).

[9] However, they proceed to suggest that “the vast majority” of locations served by the merging firms still “have characteristics that will permit large gains.”  Jeffrey Reed and Nishith Tripathi, “AT&T/T-Mobile:  Further Analysis of Capacity, Spectrum Efficiency and Service Quality Gains from Network Integration”, at p. 7.

[10] The authors find that those areas with unexhausted economies of density (though they do not use this term) are “slightly higher than the high end of the density found in districts dominated by single family homes” (Gabel and Kennet [1991, at 77]).

[11] Appendix A presents non-parametric data envelope analysis (DEA) results for a robustness check to our conclusion.

[12] See for example Baumol (1976).

[13] For detailed description see Chambers (1988). In other words, RTS is the ratio of the proportionate change in the output to a small proportionate change in the input quantity in the single input case or to a small equi-proportionate change in all inputs in the multiple inputs case. Specifically, three possible characteristics of production technology can be defined as follows: for a RTS exceeds unity, , the distance in input space under-estimates the distance in output space, and the production technology exhibits increasing returns to scale; for an , the production technology is embodied by constant returns to scale, and the isoquant is evenly spaced; and finally for an , the distance in input space over-estimates the distance in output space, and the production technology shows decreasing returns to scale (and of course, such returns to scale properties hold, in general, only locally).

[14] See Ray (1998). In the single-input, single-output case, average productivity increases with the input quantity when the technology exhibits increasing returns. In the multiple inputs case, the average productivity is not defined in the usual sense, and ray average productivity is introduced instead of.

[15] This model has several distinct advantages: (1) allowing technical inefficiency effects to vary freely through time; (2) dispensing with the undesirable assumption that the firm inefficiency and heterogeneity are uncorrelated with the input variables; (3) controlling the effect of unobservable & time invariant heterogeneity on measuring inefficiency. See Greene (2004ab; 2005) for a comprehensive discussion and comparison of different stochastic frontier models.

[16] H0: Cobb-Douglas specification is preferred (βkj = β = βττ = 0, k, j = 1, 2, 3). According to model estimation results, LR = 2(Loglu-Loglr) = 2(-19.13693-(-32.11519) = 25.96 > χ2(10) = 18.31. Hence, we reject H0 in favour of more flexible translog model specification.

[17] H0: No time effects in the model, i.e. no technological change (βτ = β = βττ = 0, k = 1, 2, 3). According to model estimation results, LR = 2(Loglu-Loglr) = 2(-19.13693-(-28.35059) = 18.43 > χ2(5) = 11.07. Hence, we reject H0, and find evidence of the existence of significant technological change.

[18] H0:  i.e. normal-half normal distribution. According to model estimation results, LR = 2(Loglu-Loglr) = 2(43.483-25.634) = 35.69 > χ2(1) = 3.84. Hence, we reject H0, and in favor of the normal-truncated normal distribution model.

[19] The input-oriented measure of scale efficiency is identical to the output-oriented measure, in the absence of technical inefficiency, i.e. uit = 0. In the more general case involving technical inefficiency, the input-oriented scale efficiency is then calculated using

 , where uit is the estimated technical inefficiency of the i-th firm in the t-th year using the ‘true’ fixed effects procedure; and all the other terms are the same as those defined in the output-oriented measure, Equation (4).

[20] For a homothetic production function, the marginal rate of technical substitution is homogeneous of degree zero in inputs, which requires ∑jβkj = 0, yielding a Kmenta approximation of the CES function. The production function is homogeneous of degree  if ∑kβk = , ∑jβkj = 0, and ∑kβ = 0. Linear homogeneity obtains if  = 1, i.e. constant returns to scale.

[21] Seven representative countries (i.e. six from the OECD members plus China) are selected to conduct this firm-level efficiency and productivity study, involving two North-American countries, three European countries and two East-Asian countries (the Japanese mobile carriers are not included primarily due to severe lack of data information on the inputs). The twenty-two mobile network providers cover the main mobile network operators in each selected country, based on their market share. Most selected network providers are ranked as the top twenty mobile operators by the International Telecommunications Union (ITU).

[22] The output and inputs are recorded as what happened in the individual specific operating market of each firm. For example, we record the inputs used by Vodafone in the UK mobile market for producing the output in the UK mobile market only as the inputs and output for Vodafone UK; and record the inputs used by Vodafone in the German mobile market for producing the output in the German mobile market only as the inputs and output for Vodafone Germany; and so forth. In addition, all output and inputs information used in this current study is for the mobile phone service segment only.

[23] The information on the sales of terminal equipments were only separately reported by a very few carriers in our sample. We cannot separate this value from the total operating revenue at a consistent basis across all firms. However, based on those information reported, the equipment sales only count very small proportion of the total revenue.

[24] There is a debate in measuring the price (cost) of capital. In the existing literature of telecoms efficiency and productivity study, various measures have been used, with two in particular. The first one is related to the rate of depreciation. The second one uses PPI as a deflator of capital expense. Since we have used PPI as a proxy of material price to deflate money value of material expenses into the quantity measure of material input, the second PPI proxy is thus not preferred if a better proxy is available. As for the first one, depreciation rate, is little suitable for measuring the capital price in the mobile network industry, since the depreciation rate for mobile network carriers is very high, and in many cases even exceeds 100%. Therefore, the results are highly likely to be distorted when using depreciation rate related measure of capital price in our frontier estimation.

    Despite that to the authors’ knowledge, the method of weighted average cost of capital (WACC) is a common practice used in accounting and finance to approach cost of capital (see e.g., Bruner et al. 1998; Truong et al. 2008); this approach cannot be applied in this study as a result of lack of financial and stock market information for most of sample mobile carriers. Accordingly, the closest measure  weighted average cost of borrowings (WACB)  is used as the proxy of capital price.

[25] DEA estimated SE sores for each mobile carriers over their observation years are reported in the 1st column of Table 4 for reference.  See Appendix B for the decomposed inputs elasticities in detail. 

[26] In contrast, the DEA results reported in Appendix A: Graph 2, suggest that there are relatively fewer firms (10 out of 22) at the MPSS. However, we concern the results with accumulated statistical errors given the inability of statistical underpinning of DEA. 

[27] Due to the lack of detailed information on each individual input measure, we cannot include AT&T in our frontier analysis. The AT&T’s RTS and SE scores are fitted using the information on the aggregated total costs and total revenues from the company’s annual financial reports over 2002-2007. The results summary is presented in Appendix C.

[28] See Appendix C for detail.

[29] The data information on national regulatory standards were extracted from Li & Lyons’s (2012) study.

[30] Follow the literature, we use the subscript ‘c’ denotes the CRS assumption; and the subscript ‘v’ denotes the VRS assumption.

[31] An alternative view of the optimization process is to consider the input-oriented efficiency measure, i.e. measuring the extent to which the DMU could reduce inputs to obtain the same output  again relative to the standard of other DMUs in the sample. The LP problem for input-oriented efficiency measure is specified as: Min. w.r.t. θi, λ: θi    

Subject to where the scalar θi is the input-oriented TE score for the i-th DMU, satisfying  with a value of one indicating a point on the frontier and thus a technically efficient DMU. However, the input-oriented efficiency measure provides the same value as the output-oriented efficiency measure under the CRS. In addition, it should be emphasized that the output- and input-oriented models will estimate exactly the same frontier and therefore, by definition, identify the same set of DMUs as being efficient. It is only the efficiency measures associated with the inefficient DMUs that may differ between the two methods if the VRS is assumed. Although the two measures are unequal under the VRS, nevertheless, the influences upon the efficiency scores obtained are only minor (Coelli and Perelman, 1996). This point is also confirmed by my data. Hence, in this study, in order to consist with Färe et al.’s (1994) CRS (output-oriented) Malmquist TFP measure (also used in this study), the output-oriented efficiency measure is chosen.