NATURAL EXPERIMENTS IN U.S. BROADBAND REGULATION Thomas W. Hazle, Anil Caliskan George Mason Universiy School of Law George Mason Universiy Law and Economics Research Paper Series 08-04 This paper can be downloaded wihou charge from he Social Science Research Nework a hp://ssrn.com/absrac_id=1093393
Naural Experimens in U.S. Broadband Regulaion Thomas W. Hazle Professor of Law & Economics Direcor, Informaion Economy Projec George Mason Universiy hazle@gmu.edu Anil Caliskan Research Economis Informaion Economy Projec George Mason Universiy acaliska@gmu.edu Nov. 29, 2007 The auhors wish o hank Kahryn Zeiler, Howard Shelanski, and seminar paricipans a Georgeown Law Cener for useful commens. They, of course, remain blameless for any remaining errors or omissions.
Absrac Nework neuraliy (NN) regulaions governing how broadband Inerne Service Providers (ISPs) package and price high-speed las mile access are being considered. Advocaes say such rules are necessary o separae ranspor and applicaion layers of he Inerne, proecing innovaion a he edge. Opponens argue ha such rules are unnecessary and could block efficien forms of verical inegraion. Eiher side poins o fuure developmens NN advocaes o ani-compeiive foreclosure by ISPs, opponens o a deerrence of invesmen due o regulaory disincenives as primary argumens. Such predicions are sharply conesed. A naural experimen, however, may yield marke evidence. U.S. residenial broadband markes have been subjec o open access rules, analogous o NN regulaion, ha have varied across ime and echnologies. These policy swiches allow empirical measuremen of consumer reacions. Using he simple meric of marke peneraion, which incorporaes boh demand and supply effecs, he relaive effeciveness of rival policy regimes can be appraised. This paper considers hree disinc regimes governing he wo leading echnologies for residenial broadband, cable modems (CM) and digial subscriber line () service. Prior o 1Q2003, CM service was unregulaed (and has remained so), while was subjec o nework unbundling mandaes ha included line sharing rules enabling rival Inerne Service Providers o access las-mile loops a incremenal cos. CM service enjoyed nearly a wo-o-one marke share advanage during his period. Following 1Q2003, when line-sharing rules were eliminaed by he Federal Communicaions Commission (FCC), effecively raising wholesale access prices, subscribership sharply increased relaive o rend and o rend and o conemporaneous CM subscriber growh. This paern coninued unabaed following furher deregulaion in 3Q2005, when he FCC classified as an informaion service. By year-end 2006, subscribership was abou 65% above he linear growh rend esablished in he regulaed pre-1q2003 era, some eigh o en million more households han prediced. This evidence, in sum, suggess open access broadband regulaion deers subscriber growh, poenially imporan empirical inpu ino he ongoing regulaory debae. 2
I. INTRODUCTION A debae is aking place over he opimal economic srucure of compuer neworks. Ciing he dominance of wo rival broadband neworks in residenial markes, ne neuraliy (NN) advocaes argue ha Inerne Service Providers (ISPs) will exploi gaekeeper posiions o decrease compeiion in complemenary markes, punishing consumers. These broadband neworks are seen o be in a posiion o exrac rens from cusomers or conen suppliers by blocking (and unblocking) he flow of e-commerce, a varian of he verical foreclosure argumen (Farrell & Weiser 2003; Economides 2007). 1 Governmen regulaion is advanced as a remedy. To ani-compeiive behavior, rules are proposed 2 o preven broadband ISPs from preferenially supplying improved access o paricular applicaions (paricularly hose owned by, or paying, he ISP). Whaever he meris of hese policies, such limis carry coss. The reduced flexibiliy afforded consumers, ISPs, caching services, and conen providers in conracing for services consrains he range of business models over which markes may opimize. The coss of regulaion are subsanial if he benefis of verical inegraion are large. Such inegraion is commonly observed in he developmen of communicaions neworks (Owen 2007; Hahn & Lian 2007). Seing aside poenial efficiency gains sacrificed by verical limis, NN rules designed o consrain ISPs prohibi wha is presumably (by revealed preference) profi-maximizing behavior. This direcly reduces incenives for invesmen in broadband neworks, and forms he crux of he argumen agains mandaed NN (Hahn & Wallsen 2006). These direc effecs may be offse in par or in whole by indirec effecs ha encourage innovaion in applicaions and hereby increase demand for Inerne access. Where he increase is sufficien in scale o dominae he coss of regulaion, ISPs will face sronger incenives o inves in nework capaciy, despie being consrained from achieving a local profi maximum. This raionale is provided in he end o end argumen supplied by Lessig & Lemley (2000), Wu (2003) and ohers. Their view posis ha ransacional and organizaional efficiencies resul from separaing ownership of basic ranspor funcions (via physical neworks connecing backbones and ISPs) from conen. Specifically, removing ISP conrol over applicaions faciliaes compeiive enry ino creaion of he laer. By removing he prospec of verical inegraion (by ownership or conrac), raffic is said o flow freely end-o-end, simulaing innovaion a he edge. This view is dispued in is porrayal of Inerne archiecure (Yoo 2004). Whaever differences exis in his lively discussion, here is a fundamenal consensus ha he gulf beween policy posiions needs o be bridged wih empirical evidence. The osensible purpose of NN rules is o increase efficiency and promoe condiions ha improve he producs consumers use. Such rules enail rade-offs, 1 2 For an elaboraion on he argumen for NN regulaion, see Wu (2003) I is no clear precisely wha NN rules would be imposed (Peha 2007). 3
however, ha may enirely offse inended effecs, reducing consumer welfare. Simply saed, incenives for nework invesmen decline when nework owners lose a se of valuable propery righs, as in he case where regulaion is imposed o block cerain pricing or bundling models. Wheher a corresponding inensificaion of demand via edge produc simulaion occurs remains a fac quesion. This poin has been made by Lawrence Lessig, who has ariculaed he case for ne neuraliy regulaion. In a 2006 forum, he characerized he argumen ha such rules would burden invesmen a fallacy, and wedged his verdic on observed markeplace oucomes. Of course, his is an empirical claim. 3 The source for Lessig s conclusion was aribued o Gerald Faulhaber (2002), which evaluaed he impac of open access regulaion for deploymen of broadband services. Such rules, similar o NN policies, mandae ha a broadband nework permi rivals o lease is faciliies in order o supply compeiive reail service. Open access can be applied boh o cable modem (CM) access and o digial subscriber line () service provided over elephone neworks. In he laer insance, where i has been implemened in he U.S., i can also involve line sharing, where enrans are permied o send high-speed daa over local loops simulaneously used o make voice phone calls over sandard (circui swiched) echnology. Like NN rules, open access has been offered as a policy proecing end o end (Lemley & Lessig, 2000), eliminaing gaekeepers by yielding subscribers and applicaion vendors a choice of alernaive ISPs. Hence, in arguing for NN, Lessig offered Faulhaber s conclusion as empirical jusificaion: [I] is difficul o susain he argumen ha regulaory policy regarding open access for cable or line sharing for digial [subscriber line] service has in any way been an impedimen o broadband deploymen. If here have been impedimens, o deploymen, hey have been overwhelmingly on he supply side (p. 241). The more elling componen of his analysis, however, specified he conclusion and conained a crucial cavea. Given ha cable TV operaors were hen, as now, unregulaed (i.e., no subjec o open access mandaes), Faulhaber focused on he effec of open access on carriers, primarily incumben local exchange carriers (ILECs), of which he larges are he Bell companies. He wroe: [T]he regional Bells are deploying broadband as fas as hey can as a compeiive necessiy, and hey have been willing o suffer subsanial inernal inefficiencies o do so. I is likely ha he cos increase due o he line-sharing mandae is small compared o hese oher coss and will have no effec on he deploymen speed of digial subscriber lines. Ulimaely, his is an empirical issue, and he hard evidence of wha hese coss are 3 Key Issues in Telecommunicaions Policy, AEI-Brookings Join Cener for Regulaory Sudies (May 10, 2006), hp://www.aei-brookings.org/evens/page.php?id=155. 4
and how hey compare o he relevan marke incenives is no ye available (Ibid.). This inerpreaion moivaes furher empirical analysis. Since Faulhaber s conclusion was offered, more sysemaic deploymen daa have become available. 4 Of paricular imporance is ha wo fundamenal policy regime swiches occurred in 2003 and 2005, boh wih key implicaions for residenial broadband access suppliers. While NN has no specifically been imposed, open access regulaions have, carrying similar implicaions for consumers and producers. NN and open access are close policy subsiues, and he marke s reacion o open access serves as a proxy for he likely effecs of NN. This logic flows direcly from Lawrence Lessig s use of Faulhaber s conclusion, gleaned from implemenaion of open access obligaions on elephone carriers offering, as facual suppor for he posiion ha NN will no deer broadband deploymen. Hence, open access regulaion in U.S. broadband markes offer a poenially useful naural experimen. Tesing he deploymen effecs of regulaory changes already occurring in broadband markes yields evidence on he likely effecs of NN mandaes. Given ha he policy debae hinges on empirical verificaion, hese daa should prove crucial o he policy debae ongoing. II. BROADBAND REGULATION 1. General Approach Rules were iniially imposed on elephone company-delivered digial subscriber line () services, and hen largely removed by regulaors. Residenial broadband markes hen offer a es bed for he effeciveness of open access policies, one ha incorporaes he poenial marke power of broadband inerne service providers (ISPs). To he degree ha exising providers inefficienly resric consumers choices, regulaory consrains more easily improve oucomes. This inquiry focuses on he broadband deploymen as proxied by CM and residenial subscribership. This is he general approach suggesed by Faulhaber (2002) and elsewhere. Residenial subscribership is a simple, highly-aggregaed meric, signaling an oucome of key ineres o policy makers. While examining he effecs of regulaion on growh, general broadband rends in he U.S. and echnological developmens ha poenially affec subscribership are proxied by U.S. CM and Canadian and CM subscriberships. A number of alernaive specificaions are used 4 Ineresingly, Faulhaber has also rendered an empirical assessmen ha a relaed regulaory policy for voice (narrowband) services, unbundling local loops, was ineffecive in advancing local elephone compeiion (Faulhaber 2003). 5
o es he general hypohesis ha broadband regulaion (as in open access rules) increase efficiency. 2. Three Regimes for Regulaing Three idenifiable regulaory regimes have governed. As CM service has been consisenly unregulaed (no open access mandaes) from incepion in he mid- 1990s, he episodic variaion of regulaion enables empirical analysis comparing v. CM oucomes. Regime I: Regulaed wih Line Sharing Obligaions (pre-2003) Cable TV operaors began offering CM services in 1995 (Rosson 2006, p. 6) wihou any obligaion o share nework infrasrucure wih rival service providers. A cable sysem operaor such as TCI or AT&T (which purchased TCI in 1997) could offer broadband Inerne access o reail cusomers via verical inegraion, or via an exclusive conrac wih an ISP such as @Home. Alernaively, an independen ISP such as Earhlink could negoiae an agreemen o sell Inerne access o reail cusomers using he cable operaor s nework, bu such deals were unregulaed, A campaign requesing mandaory wholesale access, a reasonable erms and condiions, quickly maerialized, led by America Online (AOL) and GTE (a large local elephone carrier laer acquired by Verizon) (Esbin 1998, 2000). These effors were unsuccessful. 5 A 1999 FCC repor concluded ha access regulaion would risk deerring invesmen in he rapidly evolving marke (FCC 1999). The FCC laer classified CM access as an inersae informaion service, caegorically exemping i from commoncarrier or open access obligaions a he federal, sae, or local level. The U.S. Supreme Cour upheld he FCC s deerminaion in June 2005 (Brand X, 2005). In conras o unregulaed CM services, services were regulaed from heir incepion in he mid-1990s. In paricular, incumben local exchange carriers (ILECs) supplying faced hree major obligaions. Firs, under he Compuer III regime, elephone companies were required o provide he broadband ransmission componen of services on a common-carrier basis (FCC 2005, pp. 19-20). Second, under he Commission s unbundling rules, elephone companies were mandaed o provide he copper loops used o provide service on an unbundled basis. 6 This enabled compeiive local exchange carriers (CLECs, or dclecs for hose enrans specializing in daa services) o place heir swiches in a elephone company s cenral office and o connec ha equipmen o an unbundled loop supplying services o end users. Third, he FCC s line sharing rules required elephone companies o lease jus he high- 5 The AOL/Time Warner merger, consummaed in early 2000, imposed unique hird pary access obligaions. The provisions required he merged firm o offer AOL Broadband only afer permiing wo independen ISPs o uilize Time Warner Cable infrasrucure. The rules did no regulae wholesale prices, nor did hey regulae Time Warner s Road Runner broadband ISP. 6 47 C.F.R. 51.319(a)(1). 6
frequency porion of he loop ( HFPL ) used o provide services (FCC 1999). Federal and sae regulaions hen se he price for he HFPL far below he price for an unbundled loop as a whole, subsanially reducing dclec coss. 7 Regime II: Parially Deregulaed (ex-line Sharing) In February 2003, he FCC eliminaed line sharing rules. 8 This mean ha, in order o supply service o cusomers over an ILEC s lines, dclecs would have o pay for he enire local loop or srike a commercial agreemen wih he carrier o share a loop. The raionale for he reform was ha, wih lessened nework sharing obligaions, elephone carriers would inves more heavily in bringing broadband services o residenial cusomers (Crandall 2005, p. 127). FCC member (laer Chairman) Kevin Marin voed for he measure, noing ha compeiion--no regulaion--is he bes mehod of delivering he benefis of choice, innovaion and affordabiliy o consumers." 9 The policy swich was criicized by ohers, however, as a measure ha would unleash marke power, slowing broadband deploymen. For consumers, i's a bleak day, according o he Consumer Federaion of America. There are abou 40 million subscribers now in he Unied Saes, and oday's decision will likely choke off any fuure growh, one consumer group said. 10 Hence, a naural experimen was creaed. As developmens in he residenial broadband markes have played ou, subscriber growh rends across he regime swiches can now be observed, shedding ligh he poenial effecs of regulaion. Regime III: Unregulaed Cable/Unregulaed In Augus 2005, he FCC eliminaed he Compuer III rules wih respec o services. This removed he remaining open access regulaions when Inerne connecions are bundled wih ranspor. Wih he Commission deermining ha fell under Tile I of he Communicaions Ac, broadband Inerne access became reaed as an informaion service exemp from common-carrier regulaion. 11 This essenially pu services on regulaory pariy wih CM service. The raionale, as wih he line-sharing deregulaion, was ha reducing nework sharing mandaes would improve operaor incenives o deploy, upgrade, and marke broadband access. Once again, proponens of regulaion argued ha he FCC s deregulaory seps would have negaive consequences for broadband cusomers and he Inerne generally. 12 7 8 47 C.F.R. 51.319(a)(1)(i). FCC, FCC Adops News Rules for Nework Unbundling Obligaions Of Incumben Local Phone Carriers (Feb. 20, 2003); hp://hraunfoss.fcc.gov/edocs_public/aachmach/doc-231344a1.pdf. 9 Ben Charny, FCC Loosens Broadband Rules, CNET NEWS.COM (Feb. 20, 2003), hp://news.com.com/fcc+loosens+broadband+rules/2100-1033_3-985313.hml. 10 Ibid. The CFA aribued he slowdown in growh o higher prices and fewer choices for consumers in high-speed services. 11 Federal Communicaions Commission (Sep. 23, 2005), op ci. 12 Andrew Jay Schwarzman of he Media Access Projec was quoed as saying: This is a bad day for he Inerne. I hink i means higher prices and less compeiion and hreaens he growh of he Inerne. 7
III. EMPIRICAL OUTCOMES The premise for open access rules is ha consumer welfare will increase when cable modem (CM) and digial subscriber line () nework owners are legally consrained o permi rival ISPs o serve reail broadband subscribers via heir faciliies. Via such independen, non-inegraed ISPs, end users may obain unfeered access o Inerne conen and applicaions, as ISPs unaffiliaed wih he underlying broadband nework have no business ineres in inefficien blocking. This no only parallels he heoreical argumen for Ne Neuraliy (NN), i operaes as a similar policy inervenion, seing regulaory erms for broadband service providers by mandaing user access o verical services. In eiher case, common carrier regulaion supersedes marke forces (Owen 2007). Indeed, in he policy debae open access and NN are porrayed as subsiues. Advocaes for NN frequenly base heir argumen for regulaion on he eliminaion of s open access regime. For insance, THE NEW REPUBLIC ediorialized for NN legislaion in June 2006 on he grounds ha, during he previous year, he FCC exemped elecoms ha provide Inerne connecions from [open-access] resricions, dealing a blow o boh enrepreneurship and poliical discourse." 13 Similarly, Vinon Cerf, co-developer of he IP/TCP proocol and Chief Technology Evangelis for Google, recommends implemenaion of NN due o eliminaion of open access. Cerf said ha he hough hings were beer before 2005 when broadband providers were conrolled by common carriage rules ha prevened providers from discriminaing in erms of wha raffic was carried. I proeced he Inerne, he said. 14 The parallel naure of hese policy approaches allows empirical evaluaion of NN prior o is adopion. Herein, we examine how broadband subscribership responds o changes in broadband regulaions, esing he implicaions of open access rules on he evidence yielded by subscriber choices. We, in paricular, es wheher he U.S. subscriber growh decreases wih deregulaion afer conrolling for he concurren U.S. CM and Canadian and CM peneraions. The main focus is on he effecs of he line sharing deregulaion in 1Q2003, bu we also provide an analysis of he 3Q2005 deregulaion even hough our daase conains few observaions from his second deregulaory period. The pre-deregulaion rends are examined as well. FCC Reclassifies as Daa Service - Week of 9/20/05, Norhern Ligh; hp://www.cenerformarkeinelligence.com/analysviews/20050920-weeklyrepor.hm (visied April 5, 2007). 13 The Ediors, Open Ne, THE NEW REPUBLIC (June 19, 2006); hp://www.nr.com/doc. mhml?i=20060626&s=ediorial062606. 14 Wayne Rash, Ne Neuraliy Advocaes Face Off, eweek (July 17, 2006); hp://www.eweek.com/aricle2/0,1895,1990357,00.asp. 8
If open access promoes efficiency, he following hypoheses canno be rejeced: Period/Regime/Null Predicion (1) pre-1q2003: CM unregulaed, regulaed wih line sharing Predicion: subscriber growh will exceed CM subscriber growh. (2) 1Q2003-4Q2006: line sharing eliminaed 1Q2003 Predicion: subscriber growh will decline from rend. (3) 3Q2005-4Q2006: classified informaion service 3Q2005 Predicion: subscriber growh will furher decline from rend. 1. Growh of vs. Cable Modem Prior o 1Q 2003 15 While and CM echnologies were developed a roughly he same ime, unregulaed cable companies expanded he availabiliy and peneraion of heir services more aggressively han regulaed elephone companies. By year-end 1999 CM dominaed he emerging residenial broadband marke: residenial and small business lines oaled jus 0.29 million, while CM subscribers numbered 1.40 million. CM coninued is dominance hrough year-end 2002, when i served 11.34 million, double he number of lines (5.53 million) according o he FCC. See Figure 1. [Inser Figure 1 here] While facors unrelaed o regulaion may help explain CM s dominance, hese daa -- absen furher evidence are inconsisen wih he hypohesis ha open access promoes broadband deploymen. Moreover, some Wall Sree analyss concluded ha regulaory facors did play an imporan causaive role in he relaively quick deploymen of cable modem services. According o Blake Bah of Lehman Brohers: The reason ha he cable companies really sepped up heir invesmen in 1997 and beyond was hey were no regulaed, hey weren forced o open up heir neworks. There were muliple revenue sreams ha hey could address. They could price he services however hey waned. 16 Invesor behavior provides furher evidence of he regulaory linkage. In a sudy of 29 evens which, beween Jan. 1998 and Oc. 2000, significanly impaced he 15 Research on he broadband race under his regime has been evaluaed in Bilingmayer & Hazle (2002), which is summarized here. 16 Quoed in Adam Thierer, Broadband and he Markes: Perspecives from he Invesmen Communiy, Cao Insiue Policy Forum (July 24, 2001), p. 15; hp://www.cao.org/evens/ranscrips/010724e.pdf. Indusry analyss rouinely cied a more favorable regulaory climae for cable operaors vs. elephone carriers when assessing broadband markes. See, e.g., Jupier Research, US Broadband Household Projecions: Tier Services and Pricing o Drive Demand (Feb. 3, 2003). 9
possibiliy of open access mandaes for CM services, Bilingmayer & Hazle (2003) found ha eigh evens advancing open access rules produced negaive reurns for cable modem providers, bu no posiive reurns for he overall Inerne Index. 17 On he conrary, evens defined as sebacks for open access produced high abnormal reurns (adjusing for conemporaneous movemens in he overall marke) for cable ISPs such as Excie@Home and posiive reurns for he Inerne Index, a porfolio conaining a large number of firms selling Inerne-relaed services. See Table 1. These resuls, suggesing ha invesors did no expec open access o simulae innovaion and growh, are likewise inconsisen wih he null. [Inser Table 1 here] 2. Growh Before 1Q2003 vs. Growh Afer 1Q2003 subscribership experienced a sharp increase in rend afer he FCC s decision o end line sharing. During his period, subscribership rose from 7.14 million in he firs quarer of 2003 o 25.14 million by he fourh quarer of 2006. A he same ime, CM subscriber growh coninued a a roughly consan linear pace, suggesing he growing peneraion was due o facors specific o raher han he broadband marke generally. Looking only a CM and, he CM share of residenial broadband was 64% in 1Q2003, wih a 36%. By 4Q2006, CM share had fallen o 54%, wih rising o 46%. 18 Figure 2 displays and CM subscribership from 1Q1999 hrough 4Q2006. I includes he linear rend esablished prior o he line sharing deregulaion (exrapolaing he 3Q2000 hrough 1Q2003 rend). By 4Q2006, peneraion exceeded his prederegulaion forecas. By a dramaic 65%. 4Q20006 CM subscribers, meanwhile, exceeded rend (using he same forecas mehod) by 11%. [Inser Fig. 2 here] The ime series of he residuals from regressing peneraion on CM subscribership, 1Q1999 hrough 4Q2006, illusraes he growh in subscribership. This approach shows ha, following deregulaion, growh sharply increased relaive o he general broadband marke as proxied by CM peneraion. See Figure 3. [Inser Fig. 3 here] 17 Reurns were calculaed for 1-day and 3-day periods surrounding open access even announcemens. Repored reurns are abnormal, Nasdaq-adjused. 18 U.S. broadband subscribership daa from 1Q1999 hrough 4Q2006 are colleced from wo sources. Legg Mason and CM daa cover he period from 1Q1999 hrough 1Q2006. Kagan provides and CM daa from 1Q05 hrough 4Q06. In order o merge he daa from differen sources, he Kagan daa for 2Q2006 and 4Q2006 are normalized wih Kagan/Legg Mason mulipliers calculaed and averaged over he five overlapping daa poins, 1Q2005-1Q2006. 10
In order o beer undersand he change in growh afer aboliion of line sharing in 1Q2003 and o es he hypohesis ha open access rules for promoe efficiency, a simple regression model is consruced, ln g β + ε, 19 (I) US, US, CM = 0 + β1 ln g + β 2 ln + β 3Swich1 which predics he logarihm of he rae of subscriber growh in period as a funcion of a Swich 1 dummy, conrolling for he concurren CM growh rae and a ime rend. The Swich 1 dummy is se o one for he periods including and afer 1Q2003, and zero oherwise. Conemporaneous CM subscriber growh is used as a proxy for general broadband marke rends. I is clear ha CM peneraion is no enirely exogenous, as cable TV operaors are likely o respond o compeiive moves made by providers (and vice versa). However, his endogeneiy biases he es in favor of he null, diluing relaive growh gains. Table 2 shows he OLS regression resuls. [Inser Table 2 here] The coefficien for he Swich 1 dummy is posiive and saisically significan. Based on he regression resuls, he line sharing deregulaion is esimaed o increase he rae of subscriber growh by 13.3%. 20 Robusness ess of he OLS regression based on Model I indicae a poenial firsorder auocorrelaion problem. The Breusch-Godfrey LM es, a furher es for firs-order auocorrelaion, obains a p-value of 0.01 and as a resul, he null hypohesis of no serial correlaion is rejeced. The Prais-Winson esimaor correcs for he firs-order auocorrelaion. The resuls are lised under Model I in Table 2. Wih his ransformaion, he Durbin-Wason d-saisic improves from 1.08 o 2.03. The coefficien for he Swich 1 dummy, on he oher hand, remains saisically significan and posiive, alhough lower in value. In his model, deregulaion seems o have increased he growh rae of subscribership by 8.6%. One explanaion for he esimaed effec of he line sharing deregulaion migh be ha here were changes in indusry economics independen of he regime swich ha ipped growh o favor. For ha, concurren evidence from he Canadian broadband marke is examined. If facors specific o echnology are helping o accelerae deploymen, he same rend should be apparen in he Canadian broadband race. Canada is no only geographically proximae o he U.S., i feaures similar income levels and highly developed elecommunicaions infrasrucure. I is also srucured similarly wih respec o broadband, where compeiion is primarily beween CM and service. 19 x s s Growh rae of x in period is compued by he formula, g = +1 x, where s is he number of x x s subscribers in period. 20 The percenage increase in he rae of subscriber growh when he Swich 1 dummy akes he value ' β 3 3 of one is compued by he formula, 100β = 100( e 1). See Halvorsen and Palmquis (1980). x x 11
[Inser Fig. 4 here] Figure 4 displays Canadian and CM sub growh and projecions based on he 3Q2000-1Q2003 rend. The rend is exrapolaed hrough 1Q2005 since he Canadian broadband daa afer 1Q2005 are provided by a differen source and here is no overlap wih which o normalize he pos-1q2005 daa. 21 As seen in he figure, boh and CM subscriber levels in 1Q2005 are lower han he (1Q2003) projecions. However, he decrease is less for han for cable modem, signaling ha indusry economics migh have urned in favor of in he pos-2002 period. We herefore include he Canadian and CM subscribers as conrol facors when predicing U.S. subscriber growh in Model II, esimaed as: US, US, CM Canada, Canada, CM g = β + β ln g + β ln g + β ln g + β ln + β Kagan+ β Swich + ε 0 1 2 3 4 5 6 1 ln (II) Again, he independen variable of ineres is he Swich 1 dummy. The differen sources for he Canadian broadband daa are also conrolled via a dummy, Kagan, se o one for he periods afer and including 2Q2005. Resuls are displayed under Model II in Table 2. The coefficien for he Swich 1 dummy is posiive and saisically significan. The line sharing deregulaion is esimaed o have increased he rae of subscriber growh by 10.2%. The Cook-Weisberg es for heeroskesdasiciy performed as a robusness check on he OLS regression for Model II leads o a p-value of 0.02. Therefore, he null hypohesis of consan variance can be rejeced. However, a subsequen esimaion of Model II wih robus sandard errors indicaes ha he coefficien for Swich 1 is sill saisically significan. Resuls are displayed under Model II in Table 2. As an alernaive approach, he daa are here esed for evidence of a srucural break in 1Q2003, following he procedure in Mankiw, Miron, and Weil (1987). The regression model o be esimaed, = β ln + ε, (III) US, US, CM ln g 0 + β1 ln g + β 2 is a modified form of Model 1. The Swich 1 dummy is no included since he model will be esimaed for 1Q1999-4Q2002 and 1Q2003-4Q2006 sub-samples separaely. These will be he unresriced regressions o be used in an F es for srucural break. The resriced regression is Model III esimaed using he enire sample, from 1Q1999 hrough 4Q2006. 21 Canadian broadband subscribership daa are, similarly, colleced from wo sources. Legg Mason and CM daa sar from 1Q2005 and goes back o 4Q1999. Kagan daa exend from 2Q2005 o 4Q2006 whereas Kagan CM daa covers he period from 3Q2005 o 4Q2006. The missing CM daa poin in 2Q2005 is esimaed as he average of 1Q2005 and 3Q2005. 12
An iniial OLS esimaion of he resriced regression on Model III shows, again, a firs-order auocorrelaion problem. The Durbin-Wason d-saisic is 0.55. 22 Therefore, Model III is esimaed using he Prais-Winson. Table 3 presens he resuls. [Table 3 here] The F saisic for n = 3 1 and n = 25 2 [WITH] degrees of freedom equals 4.98, which exceeds he criical value of 2.99 (95% confidence level). The null hypohesis of no srucural break in 1Q2003 can be rejeced by he daa. Model IV is he refined form of Model II, where Canadian broadband daa is included. The model, ln g US, US, CM Canada, Canada, CM = 0 + β1 ln g + β 2 ln g + β 3 ln g + β 4 ln + β 5 β Kagan + ε, (IV) is esimaed using Prais-Winson o address he firs-order auocorrelaion problem found in he OLS esimaion of he model over he enire sample. 23 Again, he resriced regression is Model 4 esimaed over he enire sample whereas he unresriced regressions are esimaed for 4Q1999-4Q2002 and 1Q2003-4Q2006 sub-samples separaely. Table 3 shows he resuls. The F saisic for n = 5 1 and n = 17 2 degrees of freedom is 3.24, which exceeds he criical value of 2.81 (95% confidence level). The null hypohesis of no srucural break in 1Q2003 can be rejeced afer conrolling for Canadian broadband rends as well. As one migh suspec, he increase in peneraion following he line sharing deregulaion in 1Q2003 is accompanied by reail rae reducions. According o Bernsein Research, hese price cus were iniiaed by SBC in May 2003, wih oher elcos quickly following (Bernsein Research, 2006). Cable TV operaors, while leaving nominal prices relaively sable for cable modem service, responded by aggressively increasing daa speeds. 24 A price series showing esimaed U.S. average monhly reail raes for and CM service is consisen wih his explanaion. See Fig. 5. [Inser Fig. 5 here] 3. Growh Before 3Q2005 vs. Growh Afer 3Q2005 In Augus 2005, was furher deregulaed when he Inerne access service was declared by he FCC o be an informaion service no subjec o common carriage regulaion. Figure 6 exends Figure 2, projecing broadband subscriber rends as of 2Q2005 and showing acual levels recorded hrough 4Q2006. 22 23 24 These regression resuls are available from he auhors. Again he OLS regression resuls are available from he auhors upon reques. Bernsein Research, Broadband Updae: Seasonaliy Remains a Quesion Bu Growh Coninues Unabaed; The Figh Turns o Ne Addiions (July 7, 2006). 13
In Figure 2, a linear ime rend ( Trend I ) was esimaed using acual subscriber daa hrough he firs quarer of 2003. 25 A second linear ime rend, shown in Figure 6, is calculaed using observed subscriber levels during he inermediae period, 1Q2003 hrough 2Q2005. This second rend ( Trend II ) is hen projeced hrough he fourh quarer of 2006. According o his simple mehod, he increase in households ha occurs following he Aug. 2005 deregulaion amouns o abou 12% of oal projeced 4Q2006 subscribers. In oher words, subscribership again increases from rend in he period following he 2005 deregulaion. Despie he shor ime for posreform effecs o develop (jus six quarers, 3Q2005 o 4Q2006, inclusive), here are abou 2.63 million addiional households subscribing o by he end of 2006 han if he rend prior o he 2005 deregulaion (1Q2003 hrough 2Q2005) had coninued. CM growh, in conras, is jus one percen above he Trend II predicion a year-end 2006. To esimae he effecs of he Augus 2005 deregulaion on he rae of subscriber, Model I is refined o include a second dummy, Swich 2, ha is se equal o one for he periods including and afer 3Q2005, zero oherwise. Swich 1, as before, represens he regulaory swich in 1Q2003. The esimaed model is: US, US, Cable = 0 + β1 ln g + β 2 ln + β 3Swich1 + β 4Swich2 ln g β + ε. (V) Conrolling for he logarihm of he rae of concurren CM growh and ime rend as before, he logarihm of he rae of subscriber growh is regressed on wo regulaory swich dummies using OLS. The esimaed coefficiens, conained in Table 4, show ha he Augus-2005 deregulaion increased he rae of growh by 6.3% in addiion o he approximaely 12% bump observed following eliminaion of line sharing. Based on he mehodology employed in he previous secion, Model III is nex esimaed using OLS for wo sample periods, 1Q2003-2Q2005 and 3Q2005-4Q2006, o es wheher a srucural break occurs in 3Q2005. These wo regressions provide he unresriced esimaes. The resriced regression is Model III esimaed for he pos- line sharing deregulaion sample, 1Q2003-4Q2006. Resuls are displayed in Table 4. The F saisic for n = 3 1 and n = 9 2 degrees of freedom is 1.50, which fails o exceed he criical value of 3.86 (95% confidence level). Therefore, he null hypohesis of no srucural break in 3Q2005 canno be rejeced. This yields no evidence ha deregulaion slowed broadband deploymen, as increases in growh are observed. I does sugges, however, ha he incremenal effec of he 2005 regime change was no sufficienly large (in he few quarers for which daa exis) o be disinguishable from random noise. V. SUMMARY 25 This esimaion used ordinary leas squares, as do subsequen ime rend projecions. 14
The evidence in U.S. broadband markes suggess ha efficiency gains from deregulaion. Cable modem services held nearly a wo-o-one marke share advanage when carriers were mos heavily obligaed o provide open access o compeing ISPs. Once he FCC eliminaed a key provision of ha access regime, ending line sharing in a Feb. 2003 ruling, subscribership increased dramaically. By year-end 2006, subscribership was 65% higher more han 9 million households han i would have been under he linear rend esablished under open access regulaion. [Inser Table 4] growh gains are in evidence even when conemporaneous rends in U.S. cable modem subscribers and Canadian broadband subscribers (boh and CM) are conrolled for. Moreover, when furher deregulaion was graned by he FCC as of Augus 2005, growh did no decline. This robus deploymen response is inconsisen wih he view ha broadband regulaion promoes innovaion ha spurs infrasrucure invesmen or deploymen. Furher research may add complexiy o he models used herein, yielding precision o he esimaes and discovering non-regulaory facors conribuing o broadband deploymen. The naural experimen conduced by regulaors in varying regimes across echnologies and over ime no doub offer addiional empirical opporuniies. The implicaions for Ne Neuraliy are imporan. Advocaes and opponens boh predic ha he implemenaion of NN rules will fundamenally aler he value of broadband neworks and he speed a which advanced sysems are deployed. Acual evidence on such oucomes is already available in he markeplace, and can be uilized o inform he debae. 15
References. Bernsein Research. 2006. Broadband Updae: Seasonaliy Remains a Quesion Bu Growh Coninues Unabaed; The Figh Turns o Ne Addiions (July 7). Brand X. 2005. Naional Cable & Telecommunicaions Assn. v. Brand X Inerne Services, 545 U. S. 967. Bilingmayer, George, and Thomas W. Hazle. 2002. Financial Effecs of Broadband Regulaion, Chaper 11 in Rober Crandall and James Alleman, eds., BROADBAND: SHOULD WE REGULATE HIGH-SPEED INTERNET ACCESS? (Washingon, D.C.: AEI-Brookings Join Cener for Regulaory Sudies; Dec.). Crandall, Rober W. 2005 Compeiion and Chaos: U.S. Telecommunicaions Since he 1996 Telecom Ac (Washingon, D.C.: Brookings Insiuion Press). Esbin, Barbara. 1998. Inerne Over Cable: Defining he Fuure in Terms of he Pas, Federal Communicaions Commission, OPP Working Paper No. 30 (Aug.) 2000. The Push for Open Access o Cable Sysems: Be Careful Wha You Ask For, Dow, Lohnes & Alberson, PLLC (Washingon, D.C.). Farrell, Joseph, and Philip J. Weiser, 2003. Modulariy, Verical Inegraion, and Open Access Policies: Towards a Convergence of Anirus and Regulaion in he Inerne Age, 17 HARVARD JOURNAL OF LAW & TECHNOLOGY 85 (Fall). Faulhaber, Gerald R. 2002. Broadband Deploymen: Is Policy in he Way? Chaper 10 in Rob. Crandall and James Alleman, eds., Broadband: Should We Regulae High-Speed Inerne Access? 223 (Washingon, D.C.: AEI-Brookings Join Cener for Regulaory Sudies; Dec.).. 2003. Policy-Induced Compeiion: The Telecommunicaions Experimens, 15 INFORMATION ECONOMICS & POLICY 79. FCC. 1999. In he Maers of Deploymen of Wireline Services Offering Advanced Telecommunicaions Capabiliy and Implemenaion of Local Compeiion Provisions of he Telecommunicaions Ac of 1996, Third Repor and Order, CC Docke No. 98-147 (Dec. 9); hp://www.fcc.gov/bureaus/common_carrier/ Orders/1999/fcc99355.x.. 2005. In he Maer of Appropriae Framework for Broadband Access o he Inerne over Wireline Faciliies, Repor and Order and Noice of Proposed Rulemaking, CC Docke No. 02-33 (Sep. 23) Jupier Research. 2003. US Broadband Household Projecions: Tier Services and Pricing o Drive Demand (Feb. 3). 16
Lemley, Mark A., and Lawrence Lessig, 2000. The End of End-o-End: Preserving he Archiecure of he Inerne in he Broadband Era, 48 UCLA LAW REVIEW 925. Owen, Bruce M. 2007. The Ne Neuraliy Debae: Tweny Five Years Afer Unied Saes v. AT&T and 120 Years Afer he Ac o Regulae Commerce, Sanford Law and Economics Olin Working Paper No. 336 (Feb). Owen, Bruce M., and Gregory L. Rosson. 2003. Local Broadband Access: Primum No Nocere or Primum Processi? A Propery Righs Approach, Sanford Insiue for Economic Research Policy Paper No. 02-037 (July). Peha, Jon. 2007. Technology and he Benefis and Risks of Nework Neuraliy Requiremens, presenaion o he Federal Trade Commission Broadband Conneciviy Compeiion Policy Hearing (Feb. 13); hp://www.fc.gov/opp/ workshops/broadband/presenaions/peha.pdf. Rosson, Gregory L. 2006. The Evoluion of High-Speed Inerne Access 1995-2001, STANFORD INSTITUTE FOR ECONOMIC POLICY RESEARCH PAPER NO. 05-019 (Aug.). Wu, Tim. 2003. Nework Neuraliy, Broadband Discriminaion, 2 JOURNAL OF TELECOMMUNICATIONS & HIGH TECHNOLOGY LAW 141. Yoo, Chrisopher S. 2004. Would Mandaing Broadband Nework Neuraliy Help or Hur Compeiion? A Commen on he End-o-End Debae, 3 JOURNAL OF TELECOMMUNICATIONS & HIGH TECHNOLOGY LAW. 17
TABLE 1. ABNORMAL % RETURNS FOR INTERNET STOCK INDEX AND EXCITE@HOME AROUND OPEN ACCESS EVENTS, JAN. 1998-OCT. 2000 26 Inerne Index Excie@Home 1-day 3-day 1-day 3-day Sebacks for Open Access (n=21) Mean 1.1 1.7 7.6 8.1 Median 0.7 2.4 8.0 8.2 Vicories for Open Access (n=8) Mean -0.1-0.1-4.9-6.4 Median 0.3 0.0-4.0-6.6 Table 2. Effecs of 1Q2003 Line Sharing Deregulaion a,b Enire sample 1Q1999-4Q2006 Dependen Variable: ln US, g Enire sample including Canadian broadband daa 4Q1999-4Q2006 I I II II Consan 0.84* 0.84* 0.91* 0.91* (0.05) (0.05) (0.19) (0.33) US Cable g, -0.10-0.14 0.00 0.00 ln (0.15) (0.15) (0.33) (0.44) ln -0.28* -0.26* -0.30* -0.30* (0.02) (0.02) (0.06) (0.10) Canada, 0.01 0.01 ln g (0.05) (0.05) Canada, Cable -0.30** -0.30 ln g (0.17) (0.19) Kagan 0.06* 0.06* (0.02) (0.03) Swich 0.12* 0.08* 0.10* 0.10* (0.03) (0.04) (0.03) (0.03) 2 R 0.9482 0.9069 0.8851 n.a. Mean VIF 2.66 n.a. 7.83 7.83 Shapiro-Wilk W es 0.86 0.92 0.40 0.40 (p-value=) Cook-Weisberg es 0.58 n.a. 0.02 n.a. (p-value=) D-W 1.08 2.03 1.72 1.72 a Sandards errors are in parenhesis. b 26 Resuls as repored in Bilingmayer & Hazle 2002, p. 259. 18
TABLE 3. TEST FOR A STRUCTURAL BREAK IN 1Q2003 AFTER LINE SHARING DEREGULATION a, b, c Dependen Variable: ln US, g Enire sample 1Q1999-4Q2006 pre- line sharing deregulaion 1Q1999-4Q2002 pos- line sharing deregulaion 1Q2003-4Q2006 III (resriced) IV (resriced) III (unresriced) IV (unresriced) III (unresriced) IV (unresriced) Consan 0.84* 0.97* 0.86* 1.33* 0.05 0.04 (0.07) (0.26) (0.05) (0.23) (0.11) (0.15) US CM g, -0.17 0.42-0.08-0.44 0.93* 0.37 ln (0.15) (0.30) (0.17) (0.35) (0.36) (0.46) ln -0.24* -0.29* -0.29* -0.43* -0.01 0.00 (0.02) (0.08) (0.02) (0.07) (0.03) (0.05) Canada, -0.12* -0.03 0.31 ln g (0.04) (0.05) (0.24) Canada, CM -0.19-0.56* 0.06 ln g (0.18) (0.19) (0.18) Kagan -0.03-0.01 (0.03) (0.02) 2 R 0.8356 0.4369 0.9484 0.9243 0.5845 0.6017 RSS 0.0441 0.0236 0.0261 0.0112 0.0015 0.0009 df 28 22 13 8 12 9 Mean VIF n.a. n.a. n.a. n.a. n.a. n.a. Shapiro-Wilk W es 0.08 0.13 0.91 0.26 0.21 0.44 (p-value=) Cook-Weisberg es n.a. n.a. n.a. n.a. n.a. n.a. (p-value=) D-W 2.20 1.46 1.87 1.89 1.67 1.69 a Pre- line sharing deregulaion sample for Model IV goes back o 4Q1999 due o unavailabiliy of he Canadian broadband daa. b Sandards errors are in parenhesis. c * Significan a he 95% level. ** Significan a he 90% level. 19
TABLE 4. EFFECTS OF AUGUST-2005 DEREGULATION a, b Enire sample 1Q99-4Q06 Pos- line sharing deregulaion 1Q03-4Q06 III (resriced) 0.09 (0.12) 0.80** Dependen Variable: ln US, g Pre-2005 deregulaion 1Q03-2Q05 Pos-2005 deregulaion 3Q05-4Q06 V III III (unresriced) (unresriced) Consan 0.85* -0.02 0.43 (0.04) (0.16) (0.16) US CM g, -0.08 0.86 1.68 ln (0.13) (0.42) (0.48) (0.59) ln -0.28* -0.02 0.02-0.13 (0.01) (0.03) (0.04) (0.05) Swich 1 0.11* (0.02) Swich 2 0.06* (0.02) 2 R 0.9587 0.5886 0.2903 0.8624 RSS 0.0404 0.0015 0.0010 0.0000 df 26 12 7 2 Mean VIF 2.35 4.10 2.46 1.19 Shapiro-Wilk W es 0.46 0.61 0.75 0.45 (p-value=) Cook-Weisberg es 0.29 0.43 0.97 0.78 (p-value=) D-W 1.45 1.48 1.87 2.62 a Sandards errors are in parenhesis. b * Significan a he 95% level. ** Significan a he 90% level. 20
TABLE 5. ACTUAL VS. PREDICTED U.S. PENETRATION BY 4Q2006 Residenial Subscribership (in millions) Acual U.S., year-end 2006 P1: Projeced U.S., year-end 2006 (exrapolaing from he linear rend, 3Q2000-1Q2003) P2: Projeced U.S., year-end 2006 (exrapolaing from he linear rend, 1Q2003-2Q2005) Acual U.S. CM, year-end 2006 P1: Projeced U.S. CM (exrapolaing from he linear rend, 3Q2000-1Q2003) P2: Projeced U.S. CM (exrapolaing from he linear rend, 3Q2000-1Q2003) Acual Canada, 1Q2005 P1: Projeced Canada (exrapolaing from he linear rend, 3Q2000-1Q2003) Acual Canada CM, 1Q2005 P1: Projeced Canada CM (exrapolaing from he linear rend, 3Q2000-1Q2003) 25.14 (65% increase from P1, 12% increase from P2) 15.25 22.50 29.33 (11% increase from P1, 1% increase from P2) 26.41 28.97 2.92 (3% decrease from P1) 3.02 3.07 (13% decrease from P1) 3.54 Esimaed Increase in Quarerly U.S. Subscriber Growh from 1Q2003 line sharing deregulaion Model I OLS 13.29% Model I Prais Winson 8.62% Model II OLS, conrolling for 10.21% Canadian broadband rends Model V OLS, conrolling for Augus 12.00% 2005 deregulaion Esimaed Increase in Quarerly U.S. Subscriber Growh from Augus 2005 deregulaion Model V 6.33% 21
FIG. 1: RESIDENTIAL AND SMALL BUSINESS A AND COAXIAL CABLE LINES IN THE U.S., 1999-2002 12 Subscribers (in millions) 10 8 6 4 2 0 A (in millions) Cable Modem (in millions) Source: FCC, High Speed Services for Inerne Access (semi-annual repor). FIG. 2: & CABLE MODEM SUBSCRIBERS IN THE U.S., 1Q1999-4Q2006 35 30 25 20 15 10 5 0 'Line Sharing' Dereg Cable 65% above rend -5 1Q99 3Q99 1Q00 3Q00 1Q01 3Q01 1Q02 3Q02 1Q03 3Q03 1Q04 3Q04 1Q05 Number of Subscribers (in millions) 3Q05 1Q06 3Q06 Dec-99 Jun-00 Dec-00 Jun-01 Dec-01 Jun-02 Dec-02 11% above rend Sources: Legg Mason and Kagan. Noes: Projecions are linear esimaes calibraed 3Q2000-1Q2003. 22
FIG. 3: RESIDUAL U.S. PENETRATION (X1,000,000), 1Q1999-4Q2006 1Q1999 4Q1999 3Q2000 2Q2001 'Line Sharing' deregulaion 1Q2002 4Q2002 3Q2003 2Q2004 1Q2005 4Q2005 3Q2006 3 2.5 2 1.5 1 0.5 0-0.5-1 -1.5-2 -2.5 FIG. 4: & CABLE MODEM SUBSCRIBERS IN CANADA, 3Q1999-4Q2006 Number of Subscribers (in millions) 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0-0.5 13% below rend 3% below rend Cable -1.0 1Q99 3Q99 1Q00 3Q00 1Q01 3Q01 1Q02 3Q02 1Q03 3Q03 1Q04 3Q04 1Q05 3Q05 1Q06 3Q06 Sources: Legg Mason and Kagan. Noes: Projecion linearly esimaed 3Q2000-1Q2003. 23
FIG. 5: MEAN MONTHLY RATES FOR & CM SERVICE, 4Q2002-1Q2006 50 Average Revenue Per User 45 40 35 30 25 20 4Q02 1Q03 2Q03 3Q03 4Q03 1Q04 2Q04 Cable modem Source: Bernsein Research, Broadband Updae: Seasonaliy Remains a Quesion Bu Growh Coninues Unabaed; The Figh Turns o Ne Addiions (July 2006). 3Q04 4Q04 1Q05 2Q05 3Q05 4Q05 1Q06 Number of Subscribers (in millions) 30 25 20 15 10 5 0-5 1Q99 FIG. 6: & CM SUBSCRIBERS IN THE U.S., 1Q1999-4Q2006 1% above 'Line Sharing' Dereg II rend Dereg 12% above rend 'Line Sharing' Trend 3Q99 1Q00 3Q00 1Q01 3Q01 1Q02 3Q02 1Q03 3Q03 1Q04 3Q04 Cable 1Q05 3Q05 1Q06 Trend II 3Q06 Noes: Trend I calibraed 3Q2000-1Q2003; Trend II during 1Q2003-2Q2005. Sources as in Fig. 4. 24
FIG. 6: U.S. BROADBAND PENETRATION (X1,000,000), 1Q1999-4Q2006 'Line Sharing' deregulaion Cable deregulaion II 'Line Sharing' dereg. esimaes 35 30 25 20 15 10 5 0 1Q1999 4Q1999 3Q2000 2Q2001 1Q2002 4Q2002 3Q2003 2Q2004 1Q2005 4Q2005 3Q2006 FIG. 7: U.S. BROADBAND SUBSCRIBER ADDITIONS (X1,000,000), 2Q1999-4Q2006 'Line Sharing' deregulaion Cable deregulaion II 'Line Sharing' dereg. esimaes 2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 1Q1999 4Q1999 3Q2000 2Q2001 1Q2002 4Q2002 3Q2003 2Q2004 1Q2005 4Q2005 3Q2006 25
26