Efficient use of radio spectrum: the Administrative Incentive Pricing (AIP) approach

Transcription

1 Powered by TCPDF ( Effcent use of rado spectrum: the Admnstratve Incentve Prcng (AIP) approach Kansantaloustede Mastern tutknnon tutkelma Salla Emla Tasknen 2015 Talousteteen latos Aalto-ylopsto Kauppakorkeakoulu

2 Aalto Unversty School of Economcs Abstract Master s Thess Emla Tasknen Effcent use of rado spectrum: the Admnstratve Incentve Prcng (AIP) approach Purpose of the study The purpose of ths study s to offer an overvew on spectrum management methods and to provde a detaled analyss of one of them, namely the Admnstratve Incentve Prcng (AIP). Ths thess dscusses the background, purpose and determnaton of AIP by frst shortly ntroducng the alternatves for AIP and then crtcally assessng the alternatve ways to construct such payments. The man economc models dscussed nclude the opportunty cost based theory by Smth and NERA (1996) as well the subsequent, more refned model by Levne and Rckman (2007), whch extends the Smth-NERA methodology to account for market structure and nterference constrants. These models and the alternatve ways of assgnng frequences are crtcally vewed aganst ther core assumptons, regulator s objectves for spectrum management as well as from the pont of vew of economc effcency. Methodology The study s conducted as a lterature revew combnng lterature and publcatons from scentfc research regardng spectrum management and formaton of AIP payments, consultatve research commssoned by the telecommuncatons authortes as well as publcly avalable nformaton on realzed market transactons for spectrum. Results of the study Takng nto account the growng demand and utlzaton of spectrum resources and gven ts scarcty, spectrum management and effcency of use s crucal. Tradtonal admnstratve allocaton and assgnment methods need to be replaced by market based methods or complemented by market mmckng methods such as the AIP n order to fulfll effcency. AIP payments are currently constructed through three dfferent knds of methods basng the payment ether on the concept of opportunty cost, realzed market prces or treatng the formaton as an optmzaton problem where the regulator maxmzes overall welfare wth respect to the spectrum fee gven the nterference and resource constrants. The latter method (by Levne and Rckman) combnes economc modellng and nformaton theory arrvng at a group of equatons determnng optmal AIP. The key concluson s that n a settng where nterference, market structure and overall welfare ncludng consumer and producer surplus as well as the revenue mpact for the government from mposng AIP are accounted for, the optmal AIP should be hgher whenever spectrum sharng s possble. The AIP then act as Ramsey tax across sectors of the economy beng nversely related to the elastcty of demand. The method by Levne and Rckman explctly accounts for most of the crucal elements of spectrum and spectrum markets, but further studes are needed especally to account for dynamc effcency and how AIP payments may be used to promote t. Key words: Economcs, frequences, effcency, spectrum prcng, spectrum management, Admnstratve Incentve prcng, AIP

3 Table of Contents 1. Introducton Economcs of rado spectrum Demand Supply Interference, spectrum re-use and the need for regulaton Effcency and spectrum Natural resource propertes of spectrum Spectrum value Prvate value Socal value Key value drvers Spectrum management descrpton and comparson of the alternatve methods Spectrum management and ts objectves Spectrum management methods Tradtonal admnstratve methods Lotteres Frst-come-frst-serve approach Beauty contests Market-based methods Auctons Tradng secondary markets New approaches The Admnstratve Incentve Prcng (AIP) approach Spectrum as a natural resource Unlcensed spectrum Dfferent methods of defnng AIP payments Opportunty cost based AIP payments the Smth-NERA methodology Strvng for productve effcency A hypothetcal example Crtcsms of the Smth-NERA method Optmal Admnstratve Incentve Prcng of spectrum by Levne and Rckman (2007) Formulaton of the spectrum assgnment problem The core economc model Optmal spectrum prcng Optmal spectrum prcng wth lnear technology Optmal prcng wth general technology Crtque and comparsons to the Smth-NERA method AIP payments based on market transactons Conclusons Appendces

4 1. Introducton Rapd development and ncreasng demand of new types of technologcal products and servces utlzng rado spectrum 1, such as smartphones and tablets usng moble servces, combned wth the scarcty of spectrum resources 2 attest to the need of effcent frequency use. The aspraton for effcency can most clearly be seen n the ncreased nterest of telecom regulators to fnd new spectrum management methods to complement or even substtute tradtonal methods such as beauty contest and auctons n order to help ensure that frequences are used as effcently as possble. There are for example numerous studes conducted for communcatons regulators on spectrum management and prcng n partcular durng the last fve years or so, some of whch are also dscussed n ths thess. From an economc pont of vew effcent use of spectrum s by no means a trval ssue; accordng to the Economst (2004), at the begnnng of the 21 st century as much as half of the total value of frequences was stll wasted on uneconomc uses causng socal losses to the socety. Presumably the stuaton has mproved snce then as effcent market-based mechansms (especally auctons) to dvde frequences have become more popular, but the magntude of sgnfcance of the ssue n economc terms remans great. In addton to not beng n ther socally optmal use, some frequences are not used at all or are used only a part of the tme. These features have been clamed to be consequences of generally makng spectrum a de facto prvate good through lcensng as well as consequences of the lack of ncentves to use spectrum effcently. Ths n turn further attests to the mportance of effcent use of spectrum. Despte the growng sgnfcance of spectrum n the socety and the vast economc benefts t may brng when used optmally (or losses caused when used neffcently), there have been very few new approaches to spectrum management durng the last couple of decades. Academc research s stll domnated by tradtonal dvson and prcng methods such as beauty contests and auctons. Ths thess concentrates on the new approaches descrbng and dscussng ther determnaton, 1 Spectrum s an umbrella term whch descrbes a band of electro-magnetc frequences. In ths thess t specfcally refers to rado spectrum, commonly understood to cover frequences from approxmately 10kHz to 300GHz whch are usable for communcatons purposes (see e.g. Cave, Martn & Webb 2007). 2 There has been qute extensve dscusson on whether spectrum actually s a fnte and thus a scarce resource snce the development of more effcent technologes allows hgher capacty utlzaton (see e.g. Staple and Werbach 2004). However, the manstream vew s that spectrum resources currently are lmted and thus need to be regulated e.g. wth the help of lcensng. These ssues are dscussed n more detal n chapter 2 of the thess. 3

5 development and mplcatons for future spectrum management. The focus s specfcally on Admnstratve Incentve Prcng (AIP), a frequency prcng method used together wth tradtonal admnstratve methods for dvdng spectrum, whch do not attach a prce for the spectrum resources. Other new approaches, whch are sgnfcant n order to provde a holstc vew on spectrum management and to descrbe the possble role and sgnfcance of AIP, as well as tradtonal spectrum management methods n the future, nclude unlcensed spectrum and the deas of abundant spectrum (as opposed to the tradtonal vew of scarcty) and spectrum as a natural resource. The basc dea of AIP s to mprove the effcency of resource use by shftng spectrum from lower value use to hgher value use. The shft s ncentvzed by mposng a fee on spectrum users to encourage them to gve up under- or unused spectrum. In other words, AIP s a cost on hoardng and t presupposes some form of property rghts for spectrum n order for the cost to be mposed correctly. The most common property rghts regme s the one wth exclusve lcenses, whch s the prmary tool for spectrum management n most countres (Doyle 2006, p. 2). Whle AIP mposes a prce for spectrum the actual dvson of frequences (.e. frequency lcenses) s executed through admnstratve means such as beauty contests or lotteres. Settng the AIP fee naturally requres thorough consderaton as fees set too hgh can cause underutlzaton of the resource whereas fees that are substantally lower than the value of spectrum to the user(s) encourage hoardng, whch n turn may cause congeston (Cave, Doyle, Webb, 2007, p.167). The fee s usually based on the opportunty cost of spectrum use, snce t s regarded to be a clear, relatvely easly attanable measure n lne wth effcent outcomes, at least under perfect competton assumptons. The core method, called the Smth-NERA method s based on ths prncple. However, as wll be shown n chapter 4 there are also two alternatve ways proposed to calculate the optmal AIP prce; a method by Levne and Rckman (2007) whch extends the Smth- NERA methodology to account for market structure and nterference constrants, and a method whch bases AIP payments on prces realzed n market transactons. AIP should not be confused wth pure cost recovery fees set to lcense holders. The sole purpose of cost recovery fees s to cover the frequency governance costs of the regulator, such as costs of dvdng the frequences between uses and users, whereas AIP payments are ntended to reflect the value of the frequency and am to ncentvze ts effcent use. In Europe these two frequency fees (termed as admnstratve charges and fees for rghts of use) are actually requred to be kept separate 4

6 by law (EU Authorzaton Drectve, 2002, artcles 12 and 13). Also n the U.S. the responsble regulator, the Federal Communcatons Commsson (FCC) separates these charges by termng them regulatory fees and applcaton fees (Doyle 2007, p.1). Thus, spectrum lcense holders may be faced wth two knds of payments; cost recovery payments utlzed by bascally all regulators around the world (Cave et al. 2007, p.167) and ncentve prcng fees utlzed only by some regulators. Relatve to alternatve spectrum management methods, whch currently nclude auctons, frequency tradng, beauty contests, frst-come-frst-served methods and lotteres, AIP s a new method of promotng effcent use of spectrum. In ts earlest forms as an AIP lke ncentve fee amed at mproved effcency of spectrum use the method has been utlzed n the Unted Kngdom snce 1998 (Ofcom) and n the New Zealand snce 2009 (ACMA). In Ireland, Canada and Span the regulators have appled spectrum fees exceedng the cost recovery fees, but wthout an explct am at dong t accordng to clear economc prncples (ComReg, CRTC, CMT). Even though AIP s not currently n use n Fnland the Mnstry of Transport and Communcatons (MINTC), whch s the authorty responsble for spectrum management n Fnland, has nvestgated ths opportunty (see e.g. MINTC 2009). In addton, frequency assgnment and prcng are current themes n Fnland; the second auctonng of frequences n the naton s hstory (the 800MHz frequency band) was just held n 2013 and dscusson around ts successfulness has thrved n the Fnnsh meda. Due to the novelty of the subject AIP has not been studed substantally. As was prevously ponted out, academc research s centered around the tradtonal spectrum management methods and research around auctons s especally vast. Thus, there s a need for a concse representaton of current lterature and research around AIP and how t relates to spectrum management objectves and other spectrum management methods. The basc concepts and deas go far back to the prncples of mcroeconomcs whch have been appled to construct the framework for mplementng AIP. The key materals on AIP dscussed n ths thess nclude the book Essentals of modern spectrum management by Cave, Doyle and Webb (2007), the subsequent work of Doyle n The prcng of rado spectrum: usng ncentves mechansms to acheve effcency (2007) and The need for a conservatve approach to the prcng of rado spectrum and the renewal of rado spectrum lcenses (2010). In addton, reports prepared by consultants n cooperaton wth economcs professonals for regulators are dscussed; namely the Study nto the use of Spectrum Prcng by the Natonal Economc Research Assocates Economc Consultants (NERA) and Smth System Engneerng Ltd 5

7 (1996), An economc study to revew spectrum prcng by Indepen Consultng Ltd, Aegs Systems Ltd and Warwck Busness School (2004) and Admnstratve Incentve Prcng for Radofrequency Spectrum by Aegs Systems Ltd and Plum Consultng Ltd (2008). These reports present the current way of applyng the AIP, whch s utlzed n the U.K. and New Zealand. Levne and Rckman (2007) have further developed the AIP approach from Smth-NERA (1996) and Indepen (2007) by constructng a more rgorous mathematcal way of calculatng the optmal AIP n ther paper Optmal Admnstered Incentve Prcng whch s dscussed n secton 4.2. As stated above, ths thess concentrates on Admnstratve Incentve Prcng n theory and practce. The thess descrbes AIP and explans how t s expected to promote effcent use of spectrum resources. It also dscusses the dfferent stuatons n whch AIP can and should be utlzed and demonstrates the current methodologes used n determnng AIP. In addton, the thess dscusses the potental role of AIP (as well as other spectrum management methods) n the future by ntroducng the ssue of unlcensed spectrum and the deas of abundant spectrum and spectrum as a natural resource. The thess s manly executed as a lterature revew combnng theory on spectrum economcs, opportunty cost based prcng and a practcal applcaton regardng spectrum management that s the AIP. The thess presents the currently used methods of determnng AIP, but refrans from more specfc calculaton of ncentve prces for dfferent spectrum band uses (moble and fxed servces, broadcastng etc.). An mportant conceptual ssue regardng spectrum dvson s the dfference between dvdng spectrum to uses such as broadcastng or rado servces and further assgnng the frequences (.e. lcenses) between users such as broadcastng operators and rado servce provders. The former s often referred to as allocaton and the latter as assgnment of frequences (European Commsson, 2012). Thus, allocaton defnes the lcense,.e. the frequency band, the geographc area, the tme perod, and the restrctons on use whereas assgnment defnes the lcensee (Cramton 2003, p. 28). Allocaton between uses has been tradtonally determned through a process of negotatons between natonal and nternatonal regulators of spectrum (see e.g. Doyle 2007 and subsecton 3.1 of the thess) and changes n t are usually extremely slow or even unfeasble. Ths can be due to nternatonal harmonzaton agreements restrctng the use or some techncal constrants and the costs related to overcomng them. An example of the latter are the costs related to renewng or substtutng equpment whch has been bult to operate wth certan frequences. In other words, there s a dependency of nfrastructure and equpment on certan technologes utlzng partcular frequences (and beng unable to utlze others) whch leads to restrctons n frequency allocaton. 6

8 Ths n turn leads to neffcences n spectrum use when ntal allocatons are not optmal. Alternatvely one can conclude that the system of allocatng and assgnng frequences (through lcensng) has lead to the current technologes beng developed as opposed to what mght have been f dfferent knds of allocaton and assgnment mechansms, such as freely tradable spectrum rghts or more unlcensed spectrum, had been used. As opposed to allocaton between uses, the assgnment of spectrum between dfferent users s a process conducted and managed by the natonal authortes. It s executed wth the help of varous alternatve spectrum management methods dscussed n chapter 3. AIP payments can nfluence both assgnment and allocaton, snce the lcense renounced as a result of the payment ends up wth a new user, whch may operate n a dfferent market than the prevous owner,.e. have a dfferent use for the resource. However, n practce there s lkely to be much greater dscreton over the assgnment of rghts n the short to medum term, as nternatonal agreements and techncal restrctons restrct the changes n allocatons. The evaluaton of both allocatve and assgnment mpacts s however of mportance, snce concentratng only on assgnment may neglect sgnfcant opportuntes to enhance effcent use of spectrum. Ths s easy to understand knowng that effcent use of spectrum resources requres allocatng them to the most socally proftable uses; f ths cannot be done due to restrctons mposed then some of the potental socal benefts are lost. As a matter of fact, t has even been suggested that mprovements n allocaton mpose sgnfcantly larger effcency gans than mprovements n assgnment (Cramton 2003, p.28). The descrbed defntons of allocaton and assgnment are followed throughout ths thess. The structure of the thess s as follows. Frst the economcs of rado spectrum such as demand, supply and effcency determnaton as well as the concept of spectrum value are ntroduced n chapter 2 to buld a bass for further dscusson. Then alternatve frequency assgnment methods are descrbed and compared n chapter 3, especally wth respect to ther fulfllment of economc effcency, as t s the man objectve of AIP utlzaton and the spectrum management n general. The current and suggested methods for calculatng AIP are ntroduced and compared n chapter 4, agan manly n relaton to the concept of economc effcency. Chapter 5 summarzes the thess fndngs and makes suggestons for further studes. 7

9 2. Economcs of rado spectrum Ths chapter goes through the fundamental economc characterstcs of rado spectrum,.e. the part of the electro-magnetc frequences whch covers frequences from approxmately 10kHz to 300GHz and can be used manly for communcatons purposes. These characterstcs are essental n understandng the ways n whch rado spectrum s managed; ncludng the motvaton behnd Admnstratve Incentve Prcng payments. The characterstcs dscussed nclude supply and demand, effcency n the context of spectrum, as well as the concept of nterference as the man motvaton behnd spectrum management. In addton, t s qute natural (and yet very rare) to draw parallels between spectrum and tradtonal natural resources (such as land, fsheres or fossl fuels) snce the resemblance n many characterstcs s evdent. Thus, the vewpont of spectrum as a natural resource s dscussed n order to better understand the possble alternatve ways of vewng spectrum resources. The chapter s concluded by a dscusson on one of the key concepts of ths thess, namely spectrum value and ts determnants. It s noteworthy that even these basc characterstcs are not all unambguous; they may not be mutually agreed upon or have not been studed extensvely. A good example of a dsagreement s vewng spectrum supply as lmted, an approach changng due to technologcal development. Another example of a new approach to economcs of spectrum yet to be extensvely studed s the prevously descrbed vewpont of spectrum as a natural resource. Snce both examples can have sgnfcant effects on the way spectrum can and should be managed (see chapter 3 next for ths dscusson) they are also brought up n ths chapter dscussng the fundamental economc propertes of spectrum. 2.1 Demand Frequences can be regarded as nputs of producton; they have no ntrnsc value, but ther value s constructed through utlzng them to produce dfferent knds of products and servces. No other resources are requred to produce spectrum, but some factors of producton such as labor and captal are needed to make use of t. Thus, the total demand for spectrum can be derved from the demand of the end products and servces produced by usng spectrum as an nput (Indepen, Aegs and Warwck, 2004, p.17). These products and servces are multfold and nclude both commercal and publc servces. Examples of the former nclude TV and rado broadcastng as well as moble phone 8

10 servces, GPS devces and wreless consumer electroncs such as mcrowaves and automatc garage door openers. The latter nclude servces of the mltary and other publc safety promoters, the use of emergency frequences beng one example. In addton, rado spectrum s used to provde products and servces utlzed by both prvate end consumers (ctzens, frms) as well as publc enttes: examples such as navgaton and avaton applcatons are the most common ones. Frequences dffer n terms of ther physcal propertes, whch the technologes utlzng them have to account for. 3 Smply put the range and penetraton power are hgher wth lower frequences and thus less nfrastructure (cell stes ncludng masts, towers and related equpment such as transmtters) s needed to cover a larger area. In contrast hgher frequences have a larger bandwdth capacty,.e. they allow the sgnal to carry more data, whch s why they are often utlzed for example n the urban areas wth many users. Due to the dfferent characterstcs dfferent frequency bands are not equally useful for all purposes. Ths clearly mples that the demand for spectrum s not homogenous ether; some frequency bands have sgnfcant excess demand whle others reman relatvely unused or specalzed to certan applcatons. Good examples of the former are the 3G and 4G frequences 4 whch face a huge excess demand due to ther commercal value especally for the telecommuncatons operators. The strong demand s reflected n the realzed aucton prces for these frequences, whch are throughout sgnfcantly hgher than prces pad for other frequences wth dfferent usages (see e.g. FCC and spectrum auctons). Examples of frequences wth a lower demand are frequences of specalzed usage such as the ones used for rado astronomy, whch can utlze even the extremely hgh frequences (EHF s) at about GHz. The demand s low smply because the utlzaton of these frequences requres specal equpment and technology. The regulator can affect spectrum demand through ncentve prcng, e.g. by settng AIP payments. The sze of the effect of a prce change depends naturally on the elastcty of demand of spectrum (marked by ϵ n the analyss n secton 4.2). As spectrum s an nput to producton the relevant term to dscuss s the elastcty of derved demand for spectrum. Ths n turn s dependent on (see Marshall 1920, Hcks 1932 and the Hcks-Marshall Law of Derved Demand) 5 : 3 For a detaled descrpton of spectrum s physcal propertes see e.g. Electromagnetc waves (InTech, 2011), avalable at: In addton, the key propertes affectng spectrum value are dscussed n detal n subsecton G and 4G technologes operate manly at 700MHz-2,6GHz frequency bands, for more nformaton see e.g. Internatonal Telecommuncaton Unon (ITU) at 5 Elastcty of demand of an nput (.e. elastcty of derved demand) and varables affectng t were frst dscussed by Marshall (1920) and Hcks (1963) resultng n the so called Hcks-Marshall Law of Derved Demand. The law was orgnally constructed n the context of labor demand. Here t s appled n the context of spectrum demand 9

11 The prce elastcty of demand for all the servces and end products produced usng spectrum as an nput The avalablty and prce elastcty of supply of alternatve nputs and thus the producton technologes avalable Spectrum's share of total cost of producton The prce elastcty of demand for the servces produced usng spectrum as an nput dffers greatly between the servces produced. The prce elastcty for many commercal applcatons s very hgh (e.g. consder customers tendency to compare and swtch ther moble operators), whereas the opposte s often true for non-commercal/publc servces (such as emergency servces). As for alternatve nputs for spectrum, substtutes do exst. Frstly, lack of spectrum can to some extent be substtuted by ncreasng captal (.e. buldng nfrastructure). As an example one can thnk of a stuaton where a moble communcatons frm lacks lower frequences whch requre less nfrastructure snce ther range s longer, but has a suffcent amount of hgher frequences, whch may be as sutable for the technology utlzed / servce provded as the lower ones but due to ther shorter range requre more nfrastructure. Thus, the lack of lower frequences can to some extent be made up for by utlzng the hgher frequences by ncreasng captal K. In addton, as ths example demonstrates a lack of one type of spectrum nput (e.g. lower frequency bands) can be substtuted by other frequences (e.g. hgher ones). Thus, snce dfferent frequences are not dentcal, spectrum tself s a substtute for spectrum. The substtutablty naturally depends on the technologes avalable and ther sutablty for dfferent spectrum bands. Tradtonally the cost of spectrum relatve to total costs of producton has been nonexstent, snce spectrum has been assgned practcally for free, except for the relatvely small cost recovery fees mposed by the regulators to cover ther spectrum management costs (see the next chapter 3 dscussng alternatve spectrum management methods for more detaled nformaton). The demand for spectrum s also affected by nnovaton: development of more effcent technologes may result n demand decreases as fewer spectrum resources are needed to produce the same amount of output. 10

12 2.2 Supply By nature spectrum s a common access resource snce t s avalable to anyone. It s an ntangble resource whch, as prevously mentoned, s not produced or refned from anythng; t just exsts as a part of the electromagnetc spectrum. However, n most countres practcng spectrum management most of the spectrum s owned by the state and leased (usng lcenses) under varous terms of use. Thus, spectrum frequences are effectvely made de facto prvate goods through regulaton. An excepton s the so-called unlcensed spectrum, for whch exclusve usage rghts are not mposed. Unlcensed spectrum s dscussed n more detal n subsecton Snce spectrum cannot be used up (although t can n theory be n full utlzaton), t s also a nonexhaustble resource. Despte ts non-exhaustble nature the supply of spectrum s fxed n a sense that there exsts a certan, fnte amount of spectrum the relevant spectrum bands under dscusson here beng rado spectrum bands, whch cover frequences from approxmately 10kHz to 300GHz and are usable for communcatons purposes (Cave, Doyle & Webb 2007, p.4). However, as was prevously dscussed n relaton to spectrum demand the heterogenety (of spectrum use and thus demand) mples that even though the overall spectrum resources are regarded as scarce, there exst also spectrum bands wth excess supply. In addton, the development of more effcent technologes, such as cogntve rados and ultrawdeband 6, enables a hgher capacty utlzaton rate of frequences whch lessens the scarcty of frequences. As Staple and Werbach (2004, p.50) state: the extent to whch there appears to be a spectrum shortage largely depends not on how many frequences are avalable but on the technologes that can be deployed. Ths n turn mples that f more effcent technologes could develop rapdly and wth relatvely low costs, spectrum as a resource would evolve from scarce to abundant. Ths would have profound effects on the way spectrum could and should be managed. The current manstream vew however s, that spectrum resources are lmted snce at least for now technologes effcent enough to challenge ths do not exst, they are too costly or complex to be employed commercally or can only be utlzed wth respect to some of the frequences (Cave et al. 2007, p ). Ths fact s sgnfcant snce the dea of scarcty s key n justfcaton of spectrum regulaton n general and thus the use of AIP, as wll be dscussed next. 6 Cogntve rados refer to a technology whch allows rados to move across the frequency band seekng for free spectrum capacty whch can then be utlzed. Ultra-wdeband n turn refers to a technology, whch can be used at a low energy level for short-range, hgh-bandwdth communcatons usng a large porton of the rado spectrum. 11

13 2.3 Interference, spectrum re-use and the need for regulaton Scarcty of spectrum resources combned wth hgh demand causes congeston whch n turn often results n nterference between dfferent spectrum users. In other words, when users of lmted spectrum resources transmt at the same tme, on the same frequency and suffcently close to each other they wll typcally cause nterference whch mght render both of ther system unusable. Even f users transmt on neghborng frequences, they can stll nterfere snce wth practcal transmtters sgnals transmtted on one channel leak nto adjacent channels, and wth practcal recevers sgnals n adjacent channels cannot be completely removed from the wanted sgnal (Cave et al. 2007, p.3). The frst type of nterference s called co-channel nterference and the leakages nto adjacent channels adjacent-channel nterference. Thus, nterference s a negatve externalty mposed by one user of spectrum on other users and t s drven by congeston. From the early days of spectrum management nterference has been the man reason justfyng the need for spectrum regulaton by the governments and ther agences. As Melody (1980, p.393) summarzes: Cooperaton among all users s essental f the spectrum s to be used effectvely by anyone. Another mportant reason behnd regulatory needs s that suffcent access to spectrum has to be ensured for applcatons of socal or publc value. Examples of these knds of applcatons are the emergency servces. A natonal regulator s a natural entty to execute the requred supervson n coordnaton wth other natonal regulators as well as nternatonal organs. Even though one cannot rely solely upon the market mechansm n achevng effcency regulators can utlze market based mechansms such as auctons or market outcome mmckng ncentves such as AIP to acheve superor outcomes relatve to alternatves such as pure admnstratve assgnment of spectrum (Cave et. al 2007, p.171). The form of regulaton and the regulatve authortes are dscussed n more detal n chapter 3. Snce frequences dffer n ther demand the constrants mposed by nterference are not equal for all parts of the rado spectrum. For frequences wth excess supply nterference poses no problem whereas for hgh demand (the most valuable) frequences careful management s needed to ensure that as many users as possble are able to utlze the resource wthout unnecessary nterference. One way of accountng for the dfferng nterference constrants when defnng the AIP payments s by 12

14 combnng basc graph theory wth economc theory as wll be shown n subsecton 4.2 along wth the optmal AIP calculatons by Levn and Rckman (2007). Interference requres the use of dfferent frequences for some communcatons, but the possblty of sharng frequences exsts for others. Frequency reuse means usng the same rado frequences on rado transmtter stes wthn a geographc area, provded that they are separated by suffcent dstance n order to mnmze nterference (Althos - GSM tutoral). In other words, to avod harmful nterference adjacent areas (sometmes called cells) use dfferent frequences, but geographc areas whch are suffcently far away from each other may use the same spectrum resources. In practce ths s seen for example n the fact that n dfferent countres the same frequences are (re-)used; e.g. operators n Fnland and Sweden may both use the same frequences to provde moble servces. 2.4 Effcency and spectrum The key purpose of spectrum management s to maxmze the value of frequency use to socety by encouragng effcent use of spectrum and thus allowng as many effcent users as possble whle also ensurng manageable nterference between users (Cave et al. 2007, p.3). In order to create spectrum management tools to try to fulfll ths objectve one must have a clear understandng on what s meant by effcency n spectrum polcy. Ths n turn s not as unambguous as one mght thnk; varous practces n determnng effcency n spectrum polcy exst (concepts such as allocatve, productve, techncal, dynamc and functonal effcency are often mentoned) and they are not always correctly understood by regulators desgnng the polcy tools. In some cases, these effcency measures can even be conflcted and achevng one does not guarantee that other effcency ndcators are successfully fulflled. An example of ths knd of a stuaton s an aucton whch manages to allocate spectrum to the bdders who value them the most and can use them most effcently n the long term. Stll, n the short term these acqured resources may be left fallow, whch ndcates that even though the so called allocatve effcency (see the defnton below) s fulflled, techncal effcency requrng constant full utlzaton of the resource s left unfulflled (for a more detaled dscusson see e.g. Freyens & Yerokhn, 2011). In general effcency s concerned wth the socety utlzng scarce resources such as spectrum n order to satsfy dfferng needs of varous agents, whch nclude for example consumers, frms and the regulator as a socal planner. To be more specfc, ths thess follows the approach gven by 13

15 Indepen et al. (2004) and followed by Cave et al. (2007) as well as Doyle (2007), whch suggests that effcency n the context of spectrum management should be understood as economc effcency. Economc effcency accounts for both statc (effcency exstng at a pont n tme) and dynamc effcency (effcency over tme) by havng three dmensons: Allocatve, productve and dynamc effcency. Allocatve effcency has to do wth the economy producng the most desred types of goods and servces n a way that Pareto optmalty s satsfed (Indepen et al. p.20). In other words, spectrum should be allocated across dfferent uses n a way that the mx of goods and servces produced s optmal; no other mx can ncrease the well-beng of one economc agent wthout harmng the wellbeng of another agent 7. Allocatve effcency can thus be mproved by encouragng the utlzaton of spectrum as an nput n the producton of products and servces most valued by the consumers. Productve effcency refers to producng the goods and servces at the lowest possble cost where cost s measured n terms of nputs such as captal, labor and spectrum (Doyle 2007, p.2). Thus, beng productvely effcent mples producng on the producton possblty fronter. As productve effcency requres that no nputs are wasted t s closely related to techncal effcency, a concept frequently emphaszed n spectrum management. Techncal effcency s concerned about the utlzaton rate of spectrum,.e. that the maxmum output s produced wth a mnmal amount of nputs, and t s often seen as an ntegral part of productve effcency (see e.g. Doyle 2007). It s mportant to note the addtve nature of the dfferent effcency measures. For example techncal effcency s a part of the overall economc effcency, but t can as well occur whle the overall economc effcency s left unfulflled: the spectrum resources may be fully utlzed, but by nonoptmal users. Dynamc effcency n turn refers to frequences beng allocated and used n a way that encourages (an optmal amount of) nnovaton and R&D (Cave et al. p.170). It can also be nterpreted as allocatng the nputs to producton over tme n a way whch mantans productve and allocatve effcency n response to changes n technology and consumer preferences (Ofcom, 2006, p.54). Thus, through dynamc effcency t s possble to further mprove effcency over tme; for example by nvestng n the development of new technologes a spectrum (lcense) owner can ncrease the 7 It s noteworthy that a Pareto effcent soluton s not necessarly socally optmal; there may be a way to ncrease one economc agent's welfare more than another agent's welfare decreases whch ncreases socal welfare. 14

16 utlzaton rate of frequency and n economc terms shft hs/her producton possbltes fronter outwards producng more wth the same resources. When all the condtons necessary for economc effcency are acheved, the economy satsfes the requrements of perfect competton thus enablng the attanment of socally optmal solutons as well. Markets utlzng spectrum naturally do not satsfy these strct requrements: there exst externaltes, the most mportant of whch s the nterference dscussed above. The consequent need for regulaton as well as varous transactonal, admnstratve as well as poltcal constrants restrct the flexblty to allocate and assgn spectrum to the most socally valuable users and uses. Thus, the regulators must compromse between dfferent objectves whle accountng for nterference. These ssues are dscussed n more detal n chapter 3 n connecton wth spectrum management and ts methods. 2.5 Natural resource propertes of spectrum Spectrum as a resource wth sgnfcant value to the socety and establshed need for regulaton s n many fundamental ways smlar to the tradtonal natural resources such as land, forests or fossl fuels. Actually, many governmental actors as well as researchers seem to be agreeng on the ssue that spectrum s, at least to some extent, a natural resource, but have yet to fulfll ther mplct promse to treat t as such (Ryan 2005). Ths subchapter shortly dscusses the connecton between natural and spectrum resources. The ssue s of relevance, snce t may have mpacts on AIP s (as well as other current prcng and assgnment methods ) use and valdty n the future. On one hand, relatng spectrum to natural resources mght mean that current economc models for spectrum regulaton must be adapted to accommodate the specal features of natural resources. On the other hand, ths mght open up new possbltes of regulatng and prcng spectrum. Spectrum possesses a varety of features that have tradtonally been regarded as propertes of natural resources. It s smlar to ar or sunshne n a sense that they are all ubqutous,.e. can be found and exst everywhere. In prncple spectrum s also non-excludable and non-exhaustble (even though t can be fully utlzed). The scarcty and possbltes for externaltes have however resulted n a stuaton where spectrum has become a common property resource, collectvely managed by governments (natonally and through nternatonal cooperaton) and leased under varous terms; 15

17 much n the way n whch for example land tenures functon. Another way n whch spectrum resources resemble land resources s the heterogeneous nature of both. However, spectrum resources also dffer from natural resources n some elementary ways. In addton to beng non-exhaustble, spectrum resources are also nstantly renewable; whenever a certan applcaton stops usng a frequency the same frequency becomes usable for any other applcaton. In any case an nterestng queston arses: what could we learn from all the exstng research and polces appled to natural resources? 2.6 Spectrum value Consderatons of spectrum value are essental to admnstratve prcng decsons, snce the basc dea of AIP s to encourage effcent use of spectrum resources through a fee whch reflects the value of spectrum n ts optmal, feasble use. Ths fee then gves spectrum users ncentves to reconsder ther current use and need of spectrum. As a result spectrum resources can be re-allocated or re-assgned from lower value use to hgher value use mplyng better overall effcency. However, the problem a regulator faces whle reflectng on spectrum prcng s the complex nature of spectrum value. Ths s well explaned by ITU (2012, p.1) whch states that the buldng blocks of spectrum value are as much poltcal and socoeconomc as they are purely fnancal. Fnancal value refers to the value derved through market sales of spectrum (e.g. auctons), whch actually reflects the prvate value of spectrum to ts users,.e. what they are wllng to pay for the resource. Due to market mperfectons ths s usually not consstent wth the value of spectrum to the socety (or socal value), whch also accounts for an array of objectves of the regulator (the socal planner), such as market structure (see secton 3.1 for a detaled descrpton of spectrum management objectves). These objectves may not always be commercally vable, but are seen as socally preferable; such as the rollout of servces nto rural areas. Valung spectrum also rases the queston of whether frequences used to produce publc servces such as defense and emergency servces should be prced and thus be subject to AIP n the frst place. Ths secton 2.6 dscusses the possble nconsstency between prvate and socal value of spectrum along wth key spectrum drvers. Snce spectrum prces reflect (prvate) spectrum value and are n a 16

18 market envronment determned by demand and supply ths secton s strongly lnked to those of demand (2.1) and supply (2.2.). Thus all prevous dscusson naturally apples here Prvate value It seems plausble that a ratonal frm values access to spectrum, or n the presence of a lcense regme the lcense, based on expected (dscounted) future returns provded by the access. Aegs and Plum (2008) offer a useful framework to examne prvate spectrum value for the spectrum utlzng frms. Accordng to them the total spectrum value (TV) for the frms conssts of two elements: the expected net present value of future returns (NPV) and the opton value (OV). The expected net present value of returns nclude returns from spectrum use.e. from enhancement of exstng servces or creaton of new ones (termed as the project based value, PV) and defensve or strategc value (DV) from ganng addtonal profts by utlzng some level of market power. The defensve value s thus concerned wth acqurng spectrum resources to protect ones market share e.g. by restrctng entry of new players or rasng compettors costs. Defensve value s assumed to be nonnegatve based on the presence of mperfectons n the market. The concept of opton value becomes relevant when there exsts sgnfcant uncertanty over future applcatons and ther value and there are sunk or rreversble costs assocated wth nvestments (Aegs & Plum 2008, p.9). It refers to the value of flexblty spectrum offers even f left partly or totally unutlzed; t offers the spectrum (lcense) holder better abltes to respond to changng crcumstances by keepng the spectrum on hold. Postve opton values would mply clearly postve prces even for spectrum resources for whch there exsts excess supply. The determnaton of spectrum value as defned by Aegs and Plum (2008) can be summarzed as: TV = PV + DV + OV, where DV, OV 0 and PV + DV = NPV of returns (1) However, the applcaton of regulatory control as well as the competton law mples that at least n theory the sgnfcance of defensve value should be small and t can be largely gnored when assessng spectrum values. In addton, spectrum lcenses often carry wth them dfferent knds of coverage and rollout condtons, or even straghtforward use t or lose t type of terms. Ths n turn mples that the spectrum lcense owner rarely gets to keep the acqured spectrum unutlzed. Thus, the role of opton value s also assumed to be qute nsgnfcant. So, the man component of 17

19 spectrum value from the users pont of vew s thus the spectrum s ablty to generate revenues/proft for the frm. The current methods of determnng AIP payments are manly based on the dea of the opportunty cost of spectrum use (see chapter 4 for more detals) and thus reflect the prvate value of spectrum. The same s true for spectrum resources that are auctoned (or traded n secondary markets), snce the bdders n the aucton naturally bd accordng to ther own prvate valuatons of the spectrum resource. Ths emphass of prvate value may however dffer from the value of spectrum to socety, especally when the markets are not perfect. The socal value aspect s dscussed next Socal value Accordng to basc economc theory, n the absence of externaltes the prvate optmum level of producton equals the socal optmum. Thus, n such a market the socal value of spectrum equals prvate value,.e. the valuaton of the most effcent frm. Ths valuaton n the absence of market dstortons was depcted n the prevous subsecton by the gans from enhancement of exstng servces or creaton of new ones,.e. the project based value (PV). As was already acknowledged however, n the market for spectrum the externalty of nterference exsts mplyng that socal and prvate valuatons of spectrum dffer. Furthermore, f the frm acqures sgnfcant market power upon obtanng the lcense, socal and prvate values dverge (McMllan 1995, p.193). Obtanng market power would be ndcated by a postve defensve value parameter (DV) n the prevous subsecton Ths also seems to often be the case for spectrum resources snce the current lcense assgnment methods have n many countres led to hghly concentrated market structures (Mlgrom, Levn & Elat 2011, p.12). Therefore, the optmalty of usng prvate valuaton nformaton as a bass of prcng spectrum s further challenged. However, prvate spectrum valuatons may be the only feasble valuatons avalable or at least most easly attanable, snce e.g. nformaton on realzed aucton prces s avalable. In addton, they may be close enough to the optmum gven that the dstortons (such as externaltes and market power mpacts) are relatvely small. Aegs and Indepen (2005, p.5) 8 also acknowledge that opportunty cost estmates used as a bass for AIP fees may not need to be adjusted to account for the socal value, because the opportunty cost estmates are calculated n the presence of polces, such as coverage requrements, that are desgned to promote the socal aspects. 8 Ths artcle focuses specfcally on prcng frequences that are used to provde broadcastng servces, but the same concluson can be drawn for all spectrum resources 18

20 Thus, n addton to takng nto account the gans from enhancement of exstng servces or creaton of new ones socal value ncludes consderatons about factors such as market structure and nvestment regulaton for example ensurng nvestments and thus the exstence of servces n rural areas. The mechansms used to calculate AIP payments, ntroduced n chapter 4, dffer wth respect to ther consderatons on prvate versus socal value Key value drvers Valuaton dfferences between spectrum bands result from varous spectrum propertes, physcal as well as other, a part of whch were already dscussed n subchapter 2.1. Tryng to provde a complete descrpton of the propertes would be an onerous task provdng very lttle beneft for the further analyss. However, t s useful to dentfy the key value drvers n order to be able to dscuss the AIP payment formaton and justfcaton n the forthcomng chapters. For ths purpose a classfcaton of the key value drvers s formed based on Smth & NERA (1996) and Aegs and Plum (2008). Whle consderng the value of a certan spectrum asset, the followng aspects should be taken nto account: A. Frequency amount o Bandwdth o Area sterlzed B. Frequency propertes o Propagaton characterstcs o Locaton of use e.g. urban vs. rural o Possble nternatonal harmonzaton C. Exstence of alternatves o Utlzaton possbltes for dfferent applcatons o Re-use opportuntes (frequency sharng) o Congeston level 19

21 D. Other qualtatve factors o Convenence of use o Ease of equpment avalablty o Mantenance or qualty of transmssons From the pont of vew of AIP, whch s most often determned based on opportunty cost, the especally nterestng aspects relate to the exstence of alternatves (pont C above). Naturally the more applcatons can use the band the more demand there s and thus the value of the spectrum band s ncreased. The same value ncrease through demand explans why congested bands are more valuable than uncongested ones. Possbltes for frequency sharng naturally reduce congeston and thus can be expected to have an opposte effect on the value of the frequency. Chapter 2 dscussed the fundamental economc characterstcs of spectrum thus buldng a bass for further dscusson of spectrum allocaton, assgnment and prcng as well as the motvaton behnd applyng AIP. As spectrum s an nput of producton ts demand was shown to be derved from the demand of the end products and servces produced usng frequences. The heterogenety of demand was also justfed and man reasons behnd varyng demands across dfferent spectrum bands were explaned to be caused by the heterogeneous nature and the vast amount of dfferent applcatons for spectrum resources. The scarcty of spectrum supply was shown to be a controversal ssue; however the current understandng beng that the supply s lmted. The concept of nterference was establshed as the man motvaton for spectrum management and effcency n the spectrum context was defned as economc effcency, whch was shown to nclude the concepts of allocatve, productve and dynamc effcency. As a curosty spectrum was paralleled wth natural resources, snce they are smlar n many ways and ths approach may arouse new ways of managng and prcng spectrum. Fnally, one key element of AIP, spectrum value was dscussed from the prvate and socal pont of vews. Prvate spectrum value was broken down nto project and defensve values (formng the total returns from spectrum usage) and the opton value. It was also shown that ths prvate value s lkely to dvert from the socal value of spectrum whenever market dstortons are present. Fnally, the key value drvers were dentfed as the amount of frequency (bandwdth), ts key propertes (propagaton characterstcs, locaton of use and harmonzaton), the exstence of alternatve uses and possbltes for sharng. Next the dfferent spectrum management practces,.e. alternatve ways of assgnng and prcng frequences, are ntroduced and compared. The comparsons are made especally wth respect to the 20

22 fulfllment of economc effcency. The man objectve of chapter 3 s to gve the reader a thorough understandng on alternatves for the AIP method and advantages as well as downsdes of AIP relatve to other spectrum management methods. 21

23 3. Spectrum management descrpton and comparson of the alternatve methods Ths chapter offers a descrpton of spectrum management, ts objectves as well as key methods. The dfferent methods are also compared wth respect to the attanment of the objectves of the regulator ( socal planner ) and especally relatve to economc effcency (.e. allocatve, productve and dynamc effcency), snce that s the man goal to be fulflled by the use of AIP. The man focus s naturally on AIP and the comparsons are made n orders to understand when and why AIP s preferred as a prcng method relatve to other alternatves and how t s used to complement alternatve spectrum management methods. 3.1 Spectrum management and ts objectves The man objectve of spectrum management s naturally to ensure that the value to the socety from scarce spectrum resources s maxmzed. Ths s done by allowng as many effcent users as possble whle keepng nterference at an acceptable level. To fulfll ths task spectrum s allocated to dfferent uses and further assgned to users. As was dscussed n 2.3 the need for regulaton s justfed by the exstence of nterference. Due to the fact that nterference can extend beyond natonal geographcal boundares and snce there exsts also nherently nternatonal uses of spectrum, such as avaton, spectrum management needs to operate at nternatonal as well as natonal level (Cave et al. 2007, p.5). The fgure 1 below, constructed based on Cave et al. (2007) llustrates the nternatonal spectrum management framework. The examples gven for each level of regulaton (lght blue boxes) consst of the ones dscussed n ths thess and are not meant to be exhaustve. 22

24 Fgure 1. The spectrum management framework Source: Cave et al (modfed) The Internatonal Telecommuncaton Unon (ITU) s an agency of the Unted Natons responsble for nformaton and communcaton technologes (ITU 2012). It accounts for the hghest level of spectrum management allocatng the global rado spectrum to dfferent uses, whch vary from prescrptve, such as satellte, to uses allowng sgnfcant nterpretaton and varaton, such as fxed or moble (Cave et al. 2007, p.5). Under ITU there are the mult-natonal bodes further coordnatng and harmonzng spectrum management across regons, such as the European Unon (EU) and Confederaton of European Post and Telecommuncaton Agences (CEPT). Natonal regulators, such as the Mnstry of Transport and Communcatons n Fnland (MINTC) then operate wthn the gudelnes provded by the nternatonal regulatng bodes. It s mportant to note that the gudelnes provded are non-bndng, but devatons by ndvdual countres are expected not to cause nterference on other countres (Cave et al. 2007, p.6). Snce the use of a certan frequency band s often defned through nternatonal coordnaton and harmonzaton n the way descrbed above (although the allocaton mght not be bndng as such) the natonal regulator s ablty to allocate spectrum to dfferent uses s restrcted. Thus, spectrum management at the natonal level s usually concerned wth assgnment decsons wthn predefned 23

25 uses rather than allocaton between uses. However, the assgnment methods ntroduced can be appled also to allocaton of spectrum, where the natonal regulator assesses the possble benefts of dfferent uses and allocates spectrum accordngly possbly devatng from the nternatonal gudelnes. Consequently the spectrum management methods presented n ths chapter are often referred to as allocaton methods as opposed to assgnment methods. As was prevously stated, n most countres the prmary tool for spectrum management s a lcensng system, whch s a form of property rghts. A spectrum lcense gves ts holder an exclusve rght to transmt at a gven frequency. Lcense condtons, whch defne the contents of the property rght,.e. what the lcense enttles the lcense holder to do and on the other hand what the holder s requred to do, are multfold. They can for example be defned to restrct the partcular technology that can be used such as GSM, or a partcular use, such as moble (Cave et al. 2007, p.105). However, many regulators nowadays express wllngness to grant more servce and technology neutral lcenses mposng fewer restrctons on the use of spectrum (see e.g. Mnstry of Transport and Communcatons n Fnland 2012). Naturally ths knd of deregulaton can be clamed to ncrease economc effcency through allowng spectrum to be used to produce the servces most valued n the socety (ncreasng allocatve effcency) wth a technology that s regarded the most effectve (ncreasng techncal and thus productve effcency). It also allows for expermentng wth new technologes whch may ncrease the dynamc effcency through nnovaton. Thus, n theory the lcense condtons should be as unrestrcted as possble mplyng that the natonal regulators would also be n charge of allocaton of spectrum n addton to ts assgnment between users. Yet n practce the problem whch arses s agan the scarcty mposed nterference; the more nonharmonzed the usage terms the more lkely nterference s to occur between users. In other words, the exstence of externaltes (.e. nterference) requres devatons from socally optmal solutons. As the socal planner the regulator has a set of objectves to fulfll n order to maxmze the spectrum s value to the socety. The man goals of the regulator usually nclude the followng (ITU 2012): Effcent usage and assgnment of the spectrum resources Rapd and effectve ntroducton of a new wreless technology (.e. broadband wreless access or BWA) Reducton of the dgtal dvde, through the development of wreless servce n remote, rural or generally low populaton densty areas 24

26 Protecton or promoton of socal welfare and/or publc servce Mnmzaton of potental nterference and coexstence ssues Government revenue generaton Some of the objectves, such as effcent usage of spectrum resources and promoton of socal welfare are clearly complementary goals both ncreasng overall welfare. Instead goals such as ncreasng rural rollout do not necessarly ncrease the overall socal welfare but are however mposed and desrable due to equty reasons. Ths s true for the rollout objectve snce such nvestments are costly wth lttle revenue ganed, and thus not economcally vable for frms to make. The socal losses made by heavy nvestments are unlkely to be recompensed by the ncrease n consumer surplus snce very few customers are located n these sparsely populated rural areas. Ths example llustrates the contradcted nature of some of the objectves of the regulator and emphaszes the need for prortzaton. 3.2 Spectrum management methods As for the assgnment and/or prcng methods by whch key spectrum management objectves can be reached, regulators have three sets of methods n ther use. Frstly there are tradtonal admnstratve methods, whch do not nclude any marked-based processes but are, as ther name suggests, purely admnstratve gvng the regulator a lot of power over the assgnment or prcng of the spectrum resources. These methods nclude lotteres, frst-come-frst-serve methods and beauty contests. Secondly, regulators may use market-based methods, whch nclude auctons and secondary markets for spectrum,.e. spectrum tradng. These methods mpose fewer nformaton requrements for the regulators, snce the regulators do not have to admnstratvely set the prce, but t s set by a market process. However, even the use of these methods requres careful plannng of the framework to ensure optmal outcomes: aspects to consder nclude the terms wth whch partcpants are allowed to partcpate (e.g. fnancal credblty), possble bddng lmtatons (e.g. bddng caps) as well as the techncal executon of the process (e.g. programs used). Thrdly, the regulators have n ther use a set of newer prcng and assgnment methods, whch are here termed as new methods. There have not been many new approaches ntroduced nto spectrum management durng the last few decades, but the three most sgnfcant ones, AIP (combned wth admnstratve methods), the vewpont of frequences as natural resources and the specal case of unlcensed spectrum are dscussed under the ttle new methods. Fgure 2 summarzes the avalable spectrum management methods. 25

27 Fgure 2. Spectrum management methods Source: Modfed from varous spectrum regulator sources (ITU, Ofcom, MINTC, FCC) When dfferent assgnment/prcng methods are compared, the essental settng whch the regulator faces s that of the prncpal-agent settng famlar from economc theory. The task of spectrum assgnment and prcng can be thus thought of as a prncpal-agent type of game between the regulator ( prncpal ) and the users of spectrum ( agents ). The problem the regulator faces whle decdng the optmal mechansm for dvdng frequences as well as governng ther use s that of asymmetrc nformaton. The task would be easy f the regulator possessed suffcent nformaton about the ndvdual valuaton (and thus the cost structures as well as technologcal solutons avalable) that spectrum users have for spectrum. In that case the choce of the assgnment method would be nsgnfcant, snce the regulator would always be able to offer Pareto effcent solutons,.e. optmal prces for the dfferent spectrum lcenses. In the real world however, the problem of adverse selecton exsts as the regulator s mperfectly nformed about the characterstcs of the spectrum users. The exstence of ths problem makes the dfferent allocaton methods unequal from the economc effcency pont of vew or at least mposes many requrements on the nformaton the regulator should have n order to prce spectrum admnstratvely n a way that promotes effcency. Market-based mechansms whch are desgned to reveal the valuaton wthout the regulator needng to determne t based on an enlghtened guess are preferred n ths respect. However, there may be other reasons related to e.g. poltcal pressure or some socal value such as equty whch favor the use of admnstratve methods. The AIP n turn s a method n the mddle n a sense that t s combned wth admnstratve methods to ncorporate market-based ncentves to them. The alternatve methods as well as benefts and weaknesses of each alternatve are dscussed next so that each of the alternatve approaches consttutes ts own subchapter. The man focus s on the attanment of economc effcency,.e. how well the alternatve methods satsfy allocatve, 26

28 productve and dynamc effcency goals. Ths s due to the reason that achevng economc effcency s the man goal of AIP utlzaton (see e.g. Ofcom 2010). 3.3 Tradtonal admnstratve methods Frst we concentrate on the tradtonal assgnment methods, whch as such mpose no prce on frequences (although t s common for the regulators to mpose small cost recovery fees), but can be combned wth prcng schemes such as the AIP. These nclude lotteres, frst-come-frst-serve methods and beauty contests, sometmes also called hearngs. Admnstratve lcensng methods, especally beauty contests, are stll wdely used even though market-based mechansms have ncreased ther popularty durng the last couple of decades Lotteres Lotteres are random selecton processes whereby the lcensees are selected by chance. Due to ther obvous lack of any knd of systematc aspraton towards economc effcency, or for any of the other key objectves of spectrum management depcted n 3.1, they are nowadays rarely used n spectrum management (FCC 2012). The key motvaton behnd ther use, manly n the 1980 s, was that they succeeded n assgnng the lcenses quckly. However, as demand started to grow, ths became evermore challengng. An llustratve example of ths s the applance of lotteres n the U.S. n 1982, when beauty contest awards lacked severely behnd causng costs to the applcants, the government and ultmately to the publc as forgone servces (McMllan 1994, p.4). Consequently the government tred replacng the beauty contests wth lotteres, but the prospect of wndfall gans attracted nearly applcatons, some of whch were submtted by users not beng techncally competent to develop and operate the subsequent spectrum utlzng servces Frst-come-frst-serve approach The frst-come-frst-serve (FCFS) approach, accordng to ts name, s an admnstratve decson assgnng the lcense(s) to the frst credble applcant(s). It s typcally used for low-valued frequences wth weak demand, snce for those frequences there wll be no necessty to resolve mutually exclusve or competng requests (ITU 2010, p.17). The FCFS method was especally popular before the 21 st century as there was enough spectrum n almost every band to accommodate most or all users and permt adequate separaton among potentally ncompatble uses (Neto & 27

29 Wellenus, 2005, p.2). An example of frequences assgned usng the FCFS method are lnk frequences n Fnland. The credblty of applcants requres that n order to be granted the lcense the applcant must adhere to certan techncal standards and regulatons. Ths aspraton towards selectng a credble applcant s what essentally separates the FCFS method from lotteres mplyng slghtly more effcent assgnments of the resource. However, savngs n admnstratve costs of allocaton/assgnment relatve to more complex management methods seems to be the only sgnfcant beneft the FCFS method offers. In addton, ths beneft s lkely to be revoked by the probablty of not assgnng spectrum to the user valung t the most,.e. the probablty that the most effcent user of the resource s not the quckest to respond to the offer Beauty contests In a beauty contest the awardng authorty (regulator) releases an nvtaton to bd for the spectrum lcenses n queston. The nvtaton contans a set of crtera, such as populaton to be served (.e. coverage), speed of deployment, project vablty, spectrum effcency and ablty to stmulate competton, based on whch responses are evaluated. The selecton crtera can be weghted dependng on the objectves of the regulator. After the responses to the nvtaton have been submtted the awardng authorty judges the qualty of applcants responses aganst the crtera and assgns spectrum lcenses accordngly. (Ofcom 2012, FCC 2012) Snce a beauty contest ncludes actually assessng the benefts the applcants would brng to socety nstead of assgnment based on draws or FCFS prncples t s the most nterestng admnstratve spectrum management method as far as economc effcency s concerned. Beauty contests are stll wdely used, processes well establshed and understood by the regulators (tenderers) as well as the applcants. For example n Fnland, lcenses even to the most valuable rado frequences have tradtonally been assgned usng beauty contests. Market based mechansms were frst tred as the 2,6GHz lcenses were auctoned to moble communcatons operators n The second frequency aucton, for one of the most valued spectrum bands n the 800MHz area s scheduled to be held durng 2013 (MINTC 2012). The man ssues wth beauty contests relate to the low beneft-cost rato: such comparatve processes can be very tme-consumng and resource-ntensve (especally related to more 28

30 straghtforward tradtonal methods) and yet not able to assgn spectrum to the agents valung them the most, thus leadng to non-optmal solutons from the effcency pont of vew. Admnstratve decsons and ther mpartalty are easy to call nto queston and whereas the regulator may be able promote natonal/socetal goals more easly through tenderng terms, comparatve processes are n practce often decded on the bass of mnor dfferences among applcants (ITU 2010, p.18). The award usually favors establshed companes (e.g. ncumbents), snce they are able to cte a track record to support ther case (Ofcom 2012), whch may mpar dynamc effcency by restrctng the market entry of nnovatve, new companes. In general, as the tradtonal admnstratve methods do not mpose any prce on frequences, the ssue of economc rents or wndfall profts arses. In addton, users of spectrum have few ncentves to gve up underused spectrum or on the other hand nvest n spectrally effcent technologes or servces. Next we wll move from dscussng pure admnstratve methods to market-based methods, whch address and correct some of these ssues. 3.4 Market-based methods Instead of relyng on a regulator to perform spectrum allocaton or assgnment, market-based methods are based on the assumpton that market mechansms, whle properly montored and supported, are the most effectve way of complementng the task. Market-based assgnment methods nclude auctons and spectrum tradng or a secondary market for spectrum lcenses Auctons Auctons represent a market based prcng and assgnment 9 mechansm whereby the prce and the lcensee of the frequency are determned n a bddng process. Asde from pre-determned requrements and condtons for the tenderng process, only prce matters. The dea of auctonng arwaves rather than assgnng them through admnstratve lcensng methods was frst proposed by Leo Herzel n hs 1951 artcle 'Publc Interest' and the Market n Color Televson Regulaton. followed by Ronald Coase n hs 1959 artcle The Federal Communcatons Commsson. However, t took approxmately forty years before the admnstratve lcensng methods descrbed prevously started to be replaced by market based prcng of frequences. New Zealand was the frst 9 As was dscussed prevously, the methods dscussed may n theory also be used to allocate spectrum between uses, but n practce ths s very rare 29

31 country n the world to legslate spectrum auctons n 1989 and t also executed the frst aucton only a year later n 1990 (McMllan, 1994, p ). Another poneer has been the Unted States whch held ts frst spectrum aucton n 1994 and has snce executed nearly a hundred auctons (FCC, 2012). In Europe the frst auctons were held at the begnnng of the 21 st century as 3G lcenses were assgned. Currently auctons are the domnant assgnment method for spectrum and they are carred out especally when there exsts strong competton for scarce spectrum wth a hgh commercal value (Mlgrom et al., 2011, p. 22). Spectrum auctons are generally consdered to be the most effcent spectrum management tools n achevng allocatve effcency. Ths s due to the fact that n a compettve bddng stuaton where only the prce pad matters (n determnng the wnner(s)), the lcenses are obtaned by those who value them the most and are thus best equpped to utlze them effectvely. Ths naturally presupposes that the frms are unbased n ther estmaton of the future profts generated by the use of the resource,.e. n determnng the total value (TV) as descrbed prevously n The framework of value creaton s also useful for dentfyng the key ssues and possble shortcomngs of auctons as a spectrum management method. As prvate frms evaluate the value of spectrum resources not only based on returns on spectrum use, but also based on the possble defensve or strategc value (termed as DV n 2.6.1) the spectrum resource entals, an aucton could end up n a stuaton where a frm would be wllng to bd on spectrum just to nterfere wth compettors. Ths also llustrates the problematc relatonshp between prvate and socal value and the fact that these two often do not meet. Regulators have however n ther dsposal measures to prevent such a stuaton from arsng. Aucton condtons such as use t or lose t are examples of ths. In addton, ant-compettve outcomes such as large operators acqurng an undue concentraton of the avalable spectrum can be restrcted by lmtng the amount of spectrum one applcant may bd on (.e. establshng bddng caps). Another key ptfall related to auctons s the phenomenon of wnner s curse. When there s ncomplete nformaton, the wnner of the tender process tends to overvalue the resource tendered and thus overpay (for emprcal proof see e.g. Bajar and Hortacsu, 2003). Ths n turn may lead to lower nvestment level and thus hnder development and effcency especally from the pont of vew of dynamc effcency. 30

32 Even though auctons are consdered the preferred assgnment method to ensure ntal effcent dstrbuton of the spectrum, the queston related to dynamc effcency remans; how to ensure that spectrum contnues to be used n an economcally effcent manner n the future? As wth other resources, economsts recommend that spectrum users be allowed to transfer ther spectrum rghts (ITU 2010, p.18). The emergence of secondary markets for spectrum wll be dscussed next n subchapter Tradng secondary markets In order to ensure that spectrum resources contnue to be used effcently, secondary markets may be needed. The core dea s very smple and equvalent to that of the basc prncple of AIP payments: movng spectrum from lower-value to hgher-value uses and users untl the value of any margnal unt of bandwdth s equal for all, or untl the cost of spectrum to any buyer equals ts value to some next-best user. The man constrant for creatng any form of free tradng n spectrum s the externalty of nterference. Most adversares of free trade state that preventng nterference among techncally dfferent servces would requre extremely complex engneerng analyss and could lead to ltgaton among spectrum users. Other counterarguments nclude for example not satsfyng socally desrable requrements and not beng able to restrct ant-compettve outcomes. (ITU 2010, p. 17). A totally free market spectrum approach has not been mplemented by any country (ITU 2012, p. 32). However, spectrum tradng s to some extent allowed n Australa, Guatemala, New Zealand and the Unted Kngdom (Doyle 2006, p.1). Admnstratve Incentve Payments can be thought of as a gradual step towards tradng frequences as they enable and encourage users to gve up frequences that are ether not utlzed or are of more value to someone else. AIP wll be dscussed n short n the next subchapter (3.5.1) under new approaches for spectrum management as well as more extensvely durng the rest of ths thess (chapters 4-5). All n all, market based mechansms are preferred over pure admnstratve ones whenever there exsts a suffcent amount of actors n the market (.e. demand exceeds supply) and there s genune competton between the players. 31

33 3.5 New approaches As was stated at the begnnng of ths thess there have been very few new ways of thnkng about spectrum resources, ther management and prcng durng the last couple of decades. The most sgnfcant ones, dscussed n ths subchapter, nclude the Admnstratve Incentve Prcng (AIP), spectrum as a natural resource and the concept of unlcensed spectrum. All of these methods pursue to ether fnd fundamental ways of treatng and regulatng spectrum dfferently than before (spectrum as a natural resource and unlcensed spectrum) or to complement more tradtonal methods n order to enhance effcency of the resource use (AIP) The Admnstratve Incentve Prcng (AIP) approach As was prevously explaned, Admnstratve Incentve Prcng s a frequency prcng method, utlzaton of whch presupposes some form of property rghts n order to mpose a fee on spectrum usage. Most commonly the property rght regme conssts of lcenses offerng exclusve usage rghts to spectrum for a specfed tme and possbly wth other usage requrements (e.g. regardng coverage). AIP s combned wth the tradtonal admnstratve assgnment methods, whch themselves mpose no explct prce for spectrum, to better mmc the market based outcomes. Thus, the AIP approach s not an assgnment method, but a prcng method combned wth admnstratve assgnment methods. AIP ams at mposng a market prce for frequences, whch encourage spectrum users to gve up spectrum that s ether unused or otherwse valued less than the charged AIP payment. If AIP truly reflects the hghest market valuaton or hghest value of alternatve use, ths encourages the transference of spectrum resources to agents valung them the most. The use of ncentve prcng of spectrum as opposed to or as a complement for the pure admnstratve allocaton and assgnment methods was frst proposed by Levn as early as 1970 n hs paper Spectrum allocaton wthout market. Levn s approach to spectrum management envsaged an ncremental path towards effcent prcng, wth revealed and stated preference methods beng used to reveal opportunty costs. Another economst promotng the ssue was Melody who (1980, p.396) dentfed the substantal possbltes for economc rents or wndfall profts to be ganed by the frms utlzng spectrum wth the contemporary spectrum management mechansms. 32

34 AIP or an AIP lke prcng (.e. prces clearly above pure admnstratve fees) for frequences s currently utlzed n the Unted Kngdom, New Zealand, Ireland, Canada and Span. In addton, ts applcaton has been consdered n Fnland. Many of the current applcatons are however stll mplemented wthout an explct am at dong t accordng to clear economc prncples (ComReg, CRTC). The telecom regulator n the UK, Ofcom, defnes the role of AIP as follows: AIP should contnue to be used n combnaton wth other spectrum management tools, n both the commercal and the publc sectors, wth the objectve of securng optmal use of the rado spectrum n the long term. AIP s role n securng optmal use s n provdng long-term sgnals of the value of spectrum (Ofcom 2010). Instead of lump sum payments (thnk of aucton payments), AIP payments are charged as an annual fee from the spectrum lcensees. Theoretcally the sum of the net present values of AIP payments should be equal to the aucton prce pad for the same frequency. Due to ts annual nature, t could be clamed that AIP lessens the need for the spectrum lcense holders to predct ther revenue and profts streams far nto the future, thus mprovng flexblty of the players to operate. At any pont n tme, when a spectrum user regards the value of spectrum to t less than the AIP payment, t s ncentvzed to gve t up resultng n new assgnment (or n some cases allocaton) of the resource. In auctons on the other hand, the pad prce as a whole s regarded as sunk cost. AIP may also be used n connecton wth admnstratve methods such as beauty contests n order to account for both monetary and non-monetary objectves. In addton, t can also be used on frequences dedcated to publc use, whch often are favored n spectrum assgnments sue to ther socetal purpose. There are some frequences for whch nether AIP, combned wth an admnstratve assgnment method, nor auctonng can be used. Examples nclude the so-called unlcensed frequences, whch are dscussed n subsecton On the other hand, AIP payments assume that a regulator s able to set a payment that reflects the real value of spectrum to ts users. In practce ths s a challengng task due to asymmetrc nformaton, the many dfferent uses that exst for specfc spectrum resources (whch one s the one brngng hghest value?) and conflcts between prvate and socal value already dscussed n secton 2.6 of ths thess. Three dfferent ways of determnng the sze of AIP payments have been proposed by economsts and regulators. The startng pont s usually the opportunty cost of spectrum use. The core method, 33

35 called the Smth-NERA method s based on ths prncple. In addton to ths method, there exst two alternatve ways to calculate the optmal prce; a method by Levne and Rckman (2007), whch extends the Smth-NERA methodology to account for market structure and nterference constrants, and a method whch bases AIP payments on prces realzed n market transactons. All of these methods and especally the most refned method of Levne and Rckman are dscussed n detal n chapters 4 and Spectrum as a natural resource As prevously dscussed, spectrum resources resemble natural resources n many respects. Thus, t would be only logcal to explore the possbltes that exstng natural resource regulaton, prcng and tradng schemes could brng to the dscusson of managng scarce spectrum resources. The dea of spectrum as a natural resource s dscussed by many scentsts, recently for example by Ryan (2004 and 2005) n hs artcles Applcaton of the Publc-Trust Doctrne and Prncples of Natural Resource Management to Electromagnetc Spectrum and Treatng the Wreless Spectrum as a Natural Resource. Ryan concludes that there seems to exst a consensus on electromagnetc spectrum beng, to some extent, a natural resource. However, t s not treated as such even though the current ways of regulaton and spectrum management may be unsutable gven ths fact. As the focus s on the AIP model ths thess refrans from more detaled dscusson around the topc, but brngs the ssue up as a potental and nterestng topc for further studes related to spectrum management Unlcensed spectrum As was prevously dscussed, the use of AIP presupposes a lcensng regme wth exclusve usage rghts for the lcense holders. Thus, one cannot thoroughly cover the concept of effcent use of spectrum or meanngfully talk about the future applcablty of AIP wthout also dscussng the other alternatve to exclusve lcenses.e. the so-called unlcensed (or lcense-exempt) spectrum. Ths subchapter provdes a concse representaton of unlcensed spectrum, ts connecton wth economc effcency and mplcatons on the use of AIP. 34

36 Unlcensed spectrum smply refers to those frequency bands n whch users can operate wthout a lcense. In other words, dfferent users share the same spectrum resources (aptly often called spectrum sharng). Thus, these parts of spectrum are treated as non-excludable, but stll rvalrous (due to nterference) common-pool resources as opposed to the prvate goods approach of lcensed spectrum. As wth any commons the problem of overuse, congeston and thus nterference s lkely to occur, when spectrum users do not account for the externalty of nterference they mpose on others (also called the tragedy of the commons n economc lterature). In order to avod excess nterference the users must therefore use certfed rado equpment and must comply wth the techncal requrements (e.g. power lmts) set by the regulator. The regulator s, whether t s a natonal regulator or an nternatonal governng body, challenge to allocate spectrum resources to specfc uses as descrbed prevously, ncludes the decson of how much spectrum to allocate to unlcensed uses. The dea of unlcensed spectrum as such s not novel; there have always been unlcensed spectrum bands and before the dscovery of the value of rado spectrum (due to technologcal development) and thus the strct regulaton of spectrum, spectrum resources were nherently unlcensed. The reason why unlcensed spectrum deserves to be ntroduced under the headng of new approaches s ts ncreased sgnfcance snce the late 1990s and the current heated dscusson around t. The cause of dscusson has mostly to do wth dynamc effcency,.e. the ablty of unlcensed spectrum to encourage nnovaton. Ths s due to the fact that many valuable nnovatons ncludng spectrum, such as the development of W-F n the 2,4GHz band, have taken place on spectrum bands that are unlcensed (Mlgrom, Levn & Elat 2011, p. 1). Ths seems natural, snce nnovaton s often best encouraged n an open envronment; just thnk about all the techncal applcatons developed n open-source envronments wthout restrctng property rghts aspects, an example beng the operatng system Lnux. Consequently, ths has aroused the queston of whether more spectrum should be allocated to unlcensed uses to encourage nnovaton. The development of nterference restrctng technologes has further ntensfed the debate snce t has the possblty to overcome nterference-related problems of unlcensed spectrum. Snce unlcensed spectrum lays asde any barrers of entry t s especally effcent n encouragng thrd-party nnovaton,.e. nnovaton by partes who do not necessarly own any lcensed spectrum. Ths s manly because the nnovators no longer have to seek and pay for the approval of current lcense holders to let them develop and test ther deas n the spectrum bands that they have no 35

37 usage rght to. The subsequent nnovatons can be substtutes for technologes utlzed by lcensed spectrum, thus ncreasng competton and effcency of spectrum use; n whch case the permsson to develop them n the current lcensees spectrum s lkely to be revoked. Alternatvely they can be complementary technologes, whch ncrease the total demand for spectrum related servces thus ncreasng the value of lcensed spectrum. An example of the latter s W-F, a technology allowng an electronc devce such as a smartphone or a tablet to exchange data wrelessly (usng unlcensed frequences) over a computer network. The avalablty of W-F ncreases the demand for electronc devces that are able to utlze t, at the same tme ncreasng demand for e.g. 3G moble servces (usng lcensed frequences), snce the devces are forced to use 3G outsde W-F hotspots. The value ncreasng effects that the unlcensed spectrum has on lcensed spectrum may even revoke the effect of revenues lost by the socety, when nstead of lcensng spectrum and sellng the lcenses, spectrum s allocated to unlcensed free use. (Mlgrom et al. 2011) However, as any form of property rghts, lcenses protect the usage of the resource and ncrease predctablty over future events - or alternatvely decrease the rsk of dsturbances by e.g. compettors. Thus, lcensng encourages the lcensees to make related nvestments, such as buldng the nfrastructure. As many uses of lcensed spectrum, such as 3G and 4G wreless moble technologes and rado as well as TV broadcastng, requre large nfrastructure nvestments lcensng s a preferred method to ensure that these nvestment are made. Another concern wth lcensed spectrum has been the techncal and thus productve effcency of spectrum use (Mlgrom et al. 2011, p. 11). Ths s because exclusve lcenses provde the lcensees wth a rght to use the spectrum n queston at all tmes and possbly all over the naton (.e. natonal lcenses) even f the resources are only needed at certan tmes a day or n certan geographcal areas. For example, just as people tend to consume more electrcty durng the day than at nght many spectrum utlzng servces (thnk about e.g. moble phone usage wthn Fnland) are consumed durng the day rather than durng nghts. Ths ndcates that at certan tmes (or n certan areas) the spectrum resources are severely underused. Several studes confrm ths underutlzaton of part of the lcensed spectrum (see e.g. Santvanez et al. 2006, Cave et al. 2007, Calabrese 2009). As a soluton to the underutlzaton problem the ntroducton of unlcensed spectrum or spectrum sharng has been suggested, but the nterference constrants have dscouraged rapd and substantal changes so far. One example of the frst steps towards more sharng can however be seen n the U.S. where FCC n cooperaton wth the Natonal Telecommuncatons and Informaton Admnstraton (NTIA), responsble for managng the spectrum used for federal purposes, 36

38 announced that t plans to take spectrum resources n the 3.5GHz area (specfcally frequences between MHz), whch are currently used for radar and share t wth wreless carrers (Arstechnca, 2012). All n all, havng both unlcensed and lcensed spectrum s attractve due to the dverse advantages that they offer. In other words, the exstence and benefts from unlcensed spectrum do not mtgate the applcablty or necessty of other spectrum management mechansms, such as AIP, but complement them. In addton, one mght envson possble hybrd solutons combnng elements from unlcensed and lcensed spectrum. An example would be a mechansm where spectrum s unlcensed, but users pay an access fee dependng on the level of congeston of the band (Cave et al. 2007, p.203). In ths case AIP payments (at least n some form) mght be appled also to unlcensed spectrum contradctng the prerequste of a lcense regme. In ths chapter spectrum management and ts key objectves as well as the alternatve spectrum management methods were ntroduced. The alternatve methods were also compared to each other prmarly aganst ther fulfllment of economc effcency consstng of productve, allocatve and dynamc effcency. Spectrum management methods were shown to nclude the admnstratve methods,.e. lotteres, frst-come-frst-serve methods and beauty contests, market-based methods,.e. auctons and secondary markets for spectrum, as well as the newer approaches of AIP, the vewpont of frequences as natural resources and the specal case of unlcensed spectrum. Wth respect to economc effcency the market-based methods were shown to be preferred. However, there was also shown to be stuatons where these cannot be utlzed or where other objectves than economc effcency necesstate the use of other methods; the most mportant case beng unlcensed spectrum, whch exstence s clamed to encourage nnovaton ncreasng dynamc effcency. Of the methods that are ether market based (auctons, spectrum tradng) or try to mmc market outcomes (AIP), auctonng was shown to be preferred for hgh demand and value frequences under compettve settngs whereas AIP can also be utlzed n connecton wth admnstratve assgnment methods and for prcng publc servce frequences. Due to ncreasng demand and multfold applcatons for spectrum resources, new ways of thnkng about spectrum management, assgnment and prcng where shown to be crucal. Regardng spectrum as a natural resource and applyng management methods and prcng used n connecton wth tradtonal natural resources was dentfed as a key opportunty for further studes. 37

39 Spectrum management was shown to nclude the nternatonal level of coordnaton through organzatons such as ITU, whch oversees spectrum management and the allocaton of spectrum to uses, multnatonal level coordnaton through organs such as the EU and a natonal level governed by the natonal communcatons regulators. The decson-makng power of natonal regulators n practce determnng the AIP payments was shown to be restrcted by nternatonal coordnaton n a way that t mostly ncludes the assgnment, but not the allocaton decsons. The multplcty and sometmes even contradctory nature of dfferent objectves of regulators was also ponted out. In the next chapter the exstng alternatve ways to calculate AIP payments are ntroduced, compared and analyzed wth respect to the fulfllment of the requrement of economc effcency. Chapter 5 then concludes the thess. 38

40 4. Dfferent methods of defnng AIP payments After the need for admnstratve prcng of spectrum s establshed the queston of how to determne the fee arses. So far there have been three alternatve ways suggested for calculatng AIP payments. Frstly, there s the method based on opportunty cost of spectrum use, whch s currently utlzed (wth mnor dfferences n calculaton practces) n the Unted Kngdom and n New Zealand. Ths method s based on the fundamental economc understandng that a prce based on opportunty cost guarantees that spectrum users cost spectrum resources as any other nputs n ther producton (Doyle 2007, p.1). Ths n turn mples productve effcency. An observaton of opportunty cost based prcng wth spectrum resources orgnally dates back to Levn (1970) who stated that although a system of freely-transferable rghts that works would be by far the best from a strctly economc vewpont, t may be mpossble to conduct (at least for all spectrum bands). Thus, we should be able to determne shadow prces that are derved from maxmum sums that current spectrum users and systems desgners would be wllng to pay rather than do wthout some small amount of spectrum (these sums naturally referrng to opportunty costs) to ensure effcent use of resources that cannot be prced by the market. Ths methodology was later on proposed n touch wth spectrum prcng by Smth and NERA (1996) and further elaborated by Indepen, Aegs and Warwck Busness School (2004). The elaboraton manly conssted of takng nto account the possblty of re-allocaton of spectrum between uses, whle Smth and NERA ntally consdered only changes n assgnments between users as a result of mposng AIP payments. Accordng to the frst developers ths method wll be referred to as the Smth-NERA method. Secondly, Levne and Rckman (2007) have proposed a more rgorous method whch bulds on the Smth-NERA methodology. More specfcally, Levne and Rckman have developed an optmal prcng scheme that allows for consumer surplus, nterference constrants and ther mplcatons for productve effcency, revenue mplcatons and market structure. Ths method shall be referred to as the Levne-Rckman method from here on. Thrdly, t has been also suggested that AIP payments could be drectly derved form observed market prces of spectrum generated n auctons or on the secondary markets for spectrum. 39

41 Ths chapter presents these three alternatve ways developed and utlzed up to date to estmate AIP payments. The focus s on the methodologes, not on the absolute values gven per spectrum band, snce the values strongly depend on the characterstcs of technologes and uses of spectrum as was dscussed n subsecton (the key value drvers). In other words, usng these methodologes the AIP payments for dfferent spectrum bands can be calculated, but some adjustments have to be made n order to take nto account the ndvdual characterstcs of dfferent uses and technologes utlzng the spectrum bands 10. Whle dscussng the dfferent methods for constructng AIP payments I provde comments and crtque related to the key assumptons. Separate sectons 4.1.3, and 4.3 summarze these dscussons per model. 4.1 Opportunty cost based AIP payments the Smth-NERA methodology In ther paper Study nto the use of Spectrum Prcng prepared for the Radocommuncatons Agency Smth and NERA (1996) construct a smple framework to examne AIP payments based on opportunty costs. Ther prmary focus s on assgnment decsons whereas Indepen et. al (2004) extend ths framework to cover also allocaton changes caused by mposed AIP payments. Thus, the Smth-NERA methodology presented next accounts for both effcency gans from shftng spectrum resources from neffcent users to effcent ones as well as effcency gans from alterng the allocaton,.e. the use of spectrum. However, the same constrants and hndrances for allocaton changes that were dscussed n the prevous chapters apply Strvng for productve effcency Introducton to the Smth-NERA approach s ntated by ntroducng the Frst Welfare Theorem statng that a compettve equlbrum (.e. a Walrasan equlbrum) s a Pareto optmum. Thus, when perfectly compettve markets preval n equlbrum the prce mechansm establshes relatve prces such that the cost to socety of producng X n terms of Y reflects consumers wllngness to pay for such a transformaton,.e. the opportunty cost (Indepen et al. 2004, p.21). Ths theorem attests to the desrablty of compettve markets. The Second Fundamental Theorem of Welfare Economcs then states that out of all possble Pareto-effcent outcomes, one can acheve any partcular one by enactng a lump-sum wealth redstrbuton and then lettng the market take over. 10 For a more detaled descrpton of the adjustments needed n the AIP payment by use/technology see e.g. Ofocm

42 In other words, n a perfectly compettve economy where polcy nstruments are non-dstortng, the frst-best welfare maxmzng outcome can be acheved. As far as spectrum management s concerned, the fact that equlbrum prces n a perfectly compettve market are n accordance wth effcent outcomes brngs us to the subsequent concluson that prces equatng supply and demand for spectrum are lkely to promote effcency. As economes n practce are not perfectly compettve, and applance of lum-sum wealth redstrbutons s bascally mpossble, only second best outcomes are possble. However, accordng to the general theory of the second best 11 t s not desrable to set prces at frst-best levels when dstortons persst elsewhere n the economy. In addton, accordng theory on optmal taxaton 12, t s not recommended to tax the use of nputs when pursung welfare maxmzng outcomes n a second-best settng. Ths would suggest that the use of nputs n a compettve economy should satsfy condtons necessary for productve effcency. (Indepen et al. 2004, p.21) Indepen et al. (2004, p.22) conclude that when compettve markets exst government polcy should be drected towards the promoton of competton where possble and desrable, and tax nstruments should be used manly on fnal goods and servces to acheve second-best welfare maxmzng outcomes. Gven ths, the use of spectrum should satsfy condtons needed for productve effcency. If ths holds, polcy as a whole ought to be consstent wth a second-best welfare maxmum. As a corollary, settng spectrum prces that promote productve effcency s desrable for effcency A hypothetcal example After t has been argued that the use of nputs should satsfy productve effcency, the relatonshp between spectrum usage, prcng and productve effcency can be studed. Indepen et al. (2004) as well as Doyle (2007) provde smple, complementary examples utlzng basc mcroeconomc theory to dscuss the lnk between effcency and spectrum prcng. These examples allow for dentfyng the necessary condtons for productvely effcent spectrum use and thus act as a 11 Smth NERA refer to Lpsey and Lancaster (1956) dscussng the theory of the second best; R.G. Lpsey and K. Lancaster (1956) The general theory of the second best, Revew of Economc Studes, vol. 24, pp Smth NERA refer todamond and Mrrlees (1971) dscussng optmal taxaton; Peter Damond and James Mrrlees (1971) Optmal taxaton and publc producton 1: Producton effcency and 2: Tax rules, Amercan Economc Revew, vol. 61, pp and

43 cornerstone for developng ncentve spectrum prcng. These examples are presented and dscussed next. 13 Assume that the avalable spectrum resources le on a unt nterval [0,1] and they are used by two sectors, 1 and 2. In other words, there are two dfferng uses for the spectrum resources n the economy, for example broadcastng and telephony. Alternatvely, one mght thnk of a certan frequency band beng assgned between two users; for example a band of frequences allocated to TV broadcastng use assgned between two TV broadcastng companes. The two sectors utlze two types of nputs, spectrum and labor n order to produce the respectve fnal outputs (broadcastng and telephony servces). Thus, labor and spectrum are regarded as substtute goods. Note that the other nput n the producton n addton to spectrum mght as well be any other nput, e.g. base statons, so that lack of spectrum resources could be replaced by nvestng n nfrastructure. In other words, substtutes for spectrum exst. There are many other sectors besde these two n the economy, but they do not utlze spectrum. However, they do make use of other nputs such as captal and they also demand labor. Thus, any amount of labor unused n the two sectors utlzng spectrum s valued n the other sectors of the economy. The total amount of labor n the economy equals L and the wage rate w>0 s determned on a compettve market. The labor resources used by sector 1 equal l 1 and by sector 2 l 2, the amount of labor utlzed n sectors 1 and 2 equalng l 1 + l 2 L. In addton, the prces of all fnal outputs produced n the economy are determned n a compettve market and frms take the prces as gven. Prces for the outputs produced by sectors 1 and 2 are denoted wth p 1 and p 2 respectvely. However, there exsts a market mperfecton, whch s the lack of a market for spectrum. In other words, spectrum s allocated to the sectors usng admnstratve proceedngs such as lotteres, beauty contests or frst-come-frst serve methods nstead of wth the help of market mechansms. Note that ths stll s the case n many countres. For smplcty assume that frequences as such (excludng cost recovery) carry a zero prce, as was prevously shown to usually be the case n admnstratve allocaton. The costs of the regulator from spectrum management are covered through general taxaton. 13 I have ntertwned these two examples nto one unform example. In order to do ths I have made some mnor modfcatons to the notaton of the example presented by Indepen et. al (2004) n order to allow for a suffcently theoretcal presentaton - note that the orgnal paper was ntended for regulatory use and was thus ntentonally expressed n layman s terms rather than usng a notaton n lne wth the economc practce. The basc story and results naturally stay unaffected by the notaton modfcatons. 42

44 From the avalable amount of spectrum (=1) let 0 < s < 1 be the amount of spectrum orgnally allocated to sector 1 and 1-s allocated to sector 2. The frms n each sector are proft maxmzers 14 choosng nputs and thus the output n order to acheve ths goal. The proft s maxmzed wth respect to possble spectrum resource constrants reflectng the scarcty of spectrum; the usage of spectrum cannot exceed supply. Other types of scarcty constrants, the nterference constrants dscussed prevously, are mposed by the fact that spectrum can be re-used n some sectors, but not n others due to nterference. A more mathematcal presentaton of the optmzaton problem s gven n subsecton XX, but here the basc prncple of the opportunty cost based prcng can be shown wth ths more general example. The total output produced n sector 1 s denoted by Q 1 (s,l 1 ) and n sector 2 by Q 2 (1-s,l 2 ). In ths example the producton functon s assumed to be concave wth dmnshng returns.e. Q > 0 and Q < 0, wth =1,2, but the logc also apples to producton functons of other forms 15. From the spectrum demand pont of vew there are naturally three scenaros whch may occur: demand for spectrum n equal to supply n each sector, demand s below supply n each sector or demand for one or both sectors exceeds the supply. In the absence of excess demand (the frst and second scenaros) t can be argued that the economy s at an effcent pont, snce no further proft can be ganed by substtutng costly labor or other nputs wth free spectrum resources. If t were possble the proft maxmzng and thus cost mnmzng frms would have done t ultmately causng excess demand for spectrum. From the regulators (and effcency s) pont of vew the nterestng scenaro s the last one, where excess demand for spectrum exsts. Ths s the case wth many spectrum resources such as the ones used for moble servces (especally the 3G and 4G frequences mentoned prevously). The exstence of excess demand rases the queston of whether the current resources could be reallocated to acheve effcency gans. The effcency gans could be acheved f a re-allocaton of spectrum freed up some of the labor (the other nput) resources wthout necesstatng a reducton n the overall output produced n each sector. The freed labor resources could be used n other sectors 14 It s mportant to note one mportant shortcomng of the llustratve example: there are also frms utllzng spectrum that are not proft-maxmzers such as many publc sector users of spectrum. The example can be however be extended to cover them by allowng for cost mnmzaton rather than proft maxmzaton and the mplcatons on spectrum prcng preval. The necessty of mposng AIP on publc spectrum users s also dscussed n more detal n chapter Dfferent forms of producton functons for spectrum utlzng servces and ther plausblty, as well as effects on optmal prcng of spectrum, are dscussed next n secton

45 to ncrease the overall output of the economy. Thus, the ntal spectrum allocaton would be productvely neffcent. In order to examne whether a re-allocaton of spectrum resources could nduce effcency gans the effect of a margnal change n spectrum allocaton s consdered. The re-allocaton s assumed to be such that outputs Q 1 and Q 2 reman constant,.e. the use of the other nput (labor) s adjusted. Keepng the output constant allows the focus to be on productve effcency. The re-allocaton from an neffcent pont to an effcent one can be llustrated n the Edgeworth box n fgure 3, where pont b llustrates the orgnal neffcent pont. Fgure 3. Effcency llustrated n the Edgeworth box Source: Indepen et al In fgure 3 the spectrum nputs are n full utlzaton (s 1+ s 2 = s),.e. techncal effcency s acheved. The soquants, whch portray the dfferent combnatons of nputs wth whch the output remans constant, correspond to the prevously determned concave producton functons wth dmnshng returns to scale. Isoquants for sector 1 are depcted as convex to the orgn and soquants for sector 2 n the opposte corner wth ncreasng output towards the orgn. As can be seen from the fgure current market outcome at pont b s neffcent as a re-allocaton n spectrum (and consequently n labor resources) brngs forth an mprovement n overall quantty of output wthout mparng the quantty produced by the other sector. When a re-allocaton cannot beneft the other sector wthout harmng the other, the allocaton satsfes productve effcency and the soluton 44

46 les on the contract curve of Pareto effcent outcomes.e. on the curve n whch the soquants of the two sectors are tangental. Ths s the bolded curve n fgure 3. For example, the regulator may strve to the effcent pont c n the Edgeworth box. Ths happens by allocatng more spectrum resources to sector 1 and consequently decreasng ts use of labor. We can assess the extent of neffcency n the ntal allocaton (pont b) n terms of the other nput used by the spectrum utlzng sectors,.e. labor. Suppose that after a margnal change n the spectrum resources, for example an ncrease (decrease) n s, denoted by Δs, the same output n sector 1 Q 1 can be produced by usng Δl 1 unts less (more) labor. It s now possble to nfer value of spectrum Δs n terms of labor,.e. wδl 1 ; the value of nput resources that would be saved by allocatng Δs to sector 1 nstead of sector 2. The same holds naturally for sector 2 where the value of Δs s wδl 2. These values are the margnal benefts (MB) of spectrum and represent estmates of the margnal opportunty cost of spectrum snce by defnton opportunty cost s the cost of an alternatve that must be forgone n order to pursue a certan acton 16. These values allow AIP payments to be calculated correctly, but requre an understandng of close substtutes for spectrum and ther relatonshp wth spectrum resources. Thus, the scarcty necesstates trade-offs and tradeoffs result n opportunty costs. When prces are set equal to opportunty cost, the frms treat spectrum as any other nput n producton, choose the nputs to mnmze these costs thus achevng productve effcency. The margnal beneft curves for the two sectors of the example are llustrated n fgure 4. The decreasng returns to scale can be seen n the downward slopng shape of the margnal beneft curves. The ntal allocaton of spectrum s (amount s for sector 1 and 1-s for sector 2) wth the correspondng margnal benefts of the sectors beng MB 1 and MB 2 can be seen to be neffcent, snce re-allocatng spectrum from sector 2 to sector 1 wth a hgher margnal beneft would mprove effcency. 16 In fact snce the Smth-NERA method looks at opportunty costs calculated at the margn by vewng how small ncremental changes n spectrum affect nput substtutablty, the approach actually studes margnal rate of techncal substtuton and values attaned can be vewed as margnal techncal opportunty costs derved from the producton functons (Indepen &Aeds 2007, p.13). 45

47 Fgure 4. Margnal beneft functons of spectrum Source: Indepen et al Effcency s n turn satsfed at s* where the margnal benefts across sectors are equal and thus no further mprovements can be made. In practce, the regulator however does not have the possblty of gettng accurate nformaton about the shape of the margnal beneft functons. Nevertheless, ths s not necessary snce the regulator may use estmated margnal benefts at current assgnments and allocatons,.e. the opportunty costs of spectrum at the current stuaton. Doyle (2007, p.7-9) contnues by llustratng ths wth a followng hypothetcal example. Frst assume that spectrum resources consst of three non-overlappng spectrum bands a,b and c n the nterval [0,1]. These frequences have been allocated to three dfferent and competng uses I, II and III. The current allocatons are depcted n table 1 below wth the hghlghted cells and the numbers stand for margnal benefts across dfferent uses. In addton, a substtute nput s depcted n the rght-most column. 46

48 Table 1. Margnal benefts of spectrum Source: Doyle 2007 Thus, use I utlzes frequency band a, use II frequency band b and use III frequency band c. The hghlghted cells,.e. the current allocatons depct the estmated opportunty cost values that a regulator can estmate most easly. The way of calculatng them n practce s explaned n detal n the next subsecton Agan, the neffcency of the ntal allocaton can be seen, snce the margnal benefts for spectrum bands across uses (the colums n the table) are not equalzed. The frequences n band a allocated to use I are n ther most effcent use, snce the margnal beneft of the two alternatve uses II and III are lower than the margnal beneft at current use. However, frquences n band b have a hgher margnal beneft at use I (MB=75) than n ther current use II (MB=60). Thus, by allocatng more spectrum from use II to use I ncreases effcency. Ths re-allocaton naturally affects the margnal benefts of the uses that are affected by the change. The changes are depcted n table 2 below. Specfcally, when more spectrum b s allocated to use I the margnal beneft n use I of frequency band b wll fall below 75 due to the decreasng returns to scale. Smlarly the margnal beneft n use I of frequency band a falls below 100, snce there are now more spectrum resources allocated to use I. In addton, an opposte effect s seen n use II of frequency bands a, b and c. Table 2. Margnal benefts of spectrum after the frst re-allocaton Source: Doyle

49 Thus, now the margnal benefts of frequency bands a and b n uses I and II have changed as well as the MB of frequency band c n use II. Now the MB s of frequency band b between uses I and II are equalzed, whch mples that for frequency band b the allocaton s effcent. Agan the hghlghted cells are the ones that the regulator can more easly calculate n practce, snce they correspond to the current allocatons and assgnments. We can see that there stll s scope for further effcency gans by re-allocatng spectrum n band c form use III (MB=15) to use II (MB=32). Followng the same logc as above ths re-allocaton wll decrease the MBs of frequency bands a, b and c n use II as the total amount of spectrum n that use ncreases. In addton, the change wll ncrease the MBs of frequency bands a, b and c n use III, whch now has overall less spectrum. As a result the MBs of spectrum band b between uses I and II wll no longer be equal and yet a further re-allocaton of frequency band b s needed. After the requste re-allocatons an effcent soluton llustrated by table 3 s acheved. Table 3. Margnal benefts of spectrum at the effcent soluton Source: Doyle 2007 In table 3 above there s equalty of margnal benefts across uses n the two hghest values. No further re-allocatons would yeld better outcomes and thus the soluton s effcent. Thus, arrvng at an effcent outcome,.e. achevng allocatve effcency, s an teratve process where one (or possbly several) re-allocatons or assgnments are made at a tme and the changed margnal benefts then calculated agan to show the possble the need for further re-allocatons or re-assgnments wthn uses. It can also be seen that n order to acheve the effcent outcome the regulator needs to know about the MBs of frequency bands n neghborng uses. These values are proposed to be evaluated based on the costs of alternatves. 48

50 In other words, ths method bascally suggests that the regulator needs to be able to frst dentfy all frequency bands and assocated uses and then determne the margnal benefts for each of them utlzng a least cost alternatve,.e. determnng, what s the spectrum user s alternatve (substtute) way of offerng the servce f a margnal amount of spectrum s taken from hm. Usng these estmatons for MB s the regulator dentfes the drecton of needed change n spectrum reallocaton and sets the prces for spectrum accordngly to encourage these changes. The actual prce set depends on whether the maxmum margnal beneft s offered by the current use or by a use that currently does not utlze the spectrum resource. In the former case the prce s set at the value of the current use whereas n the latter case the prce should le on the nterval between MB* and the current use margnal beneft (for a more detaled dscusson, see Doyle 2007, p 9-11). Ths procedure starts an teratve process towards the effcent allocaton or assgnment, whch may take up to fve years or so (Doyle 2007, p. 10) Crtcsms of the Smth-NERA method Whle the Smth-NERA methodology s smple and thus can be easly communcated to regulators settng the polces, ts straghtforwardness comes at a prce. In partcular, the methodology assumes perfect markets and thus refrans from dscussng any ssues related to market structure. Ths s a key weakness of the approach, snce many of the spectrum-utlzng markets are hghly concentrated, whch s also reflected as hgh end prces of spectrum utlzng products and servces for the customers. For example, n Germany, the Unted States, Span and Greece, where there exst no challenger operators, the moble data rates per ggabyte are tmes hgher than n the hghly competed markets n Fnland, Denmark and the UK (Taloussanomat, ). Another key lmtaton of the model s that t does not explctly account for nterference or nterference constrants, whch were prevously shown to be the major drver behnd the need for spectrum management n the frst place. Ths s manly due to the fact that the Smth-NERA model s based on smplstc economc assumptons of a non-exstng state of the market (perfect competton). As wll be shown n the next subchapter 4.2 ths s a key ssue corrected n the approach by Levne and Rckman, acheved by supplementng the basc economc theory wth graph theory, whch allows for constructng nterference constrants to the model. In addton, the approach does take nto account productve effcency, but t gnores other effects such as the consumers' wllngness to pay and revenue rased by the government (Levne & 49

51 Rckman 2007, p.2). The dynamc effcency as such s also left unexplaned and depends on the dea that a frm that values the spectrum resources most today s also the best nnovator n the long run. If we thnk about the prvate value buld-up dscussed n we see that ths often may not be the case: for example a possblty to stall or damage compettors, ndcated by a nonnegatve defensve value (DV) parameter, mght ncrease the valuaton a spectrum user has for the spectrum resources wthout havng anythng to do wth the actual wllngness and capabltes of the user to guarantee suffcent nvestments to acheve dynamc effcency. Fnally, the fact that the smallest decrements n spectrum used for evaluatng the opportunty cost may actually be very large for some servces mples that the assumpton of the output remanng constant may be unrealstc (Aegs & Plum 2008, p.19). Many of these ssues are addressed by the extended model for AIP determnaton by Levne and Rckman, whch s dscussed next. 4.2 Optmal Admnstratve Incentve Prcng of spectrum by Levne and Rckman (2007) As was dscussed n the prevous chapter the AIP calculaton approach developed by Smth and NERA and further enhanced by Indepen et al. has been crtczed for ts smplstc assumptons as well as omttng certan structures of the actual economes n whch the frms operate. To correct some of these flaws Levne and Rckman (2007) have constructed a mathematcal framework combnng models from nformaton technology and economcs to explan the structure and attanment of optmal AIP payments. In partcular they take nto account nterference and market structures (other than perfect competton proposed by Smth-NERA) and allow for revenue mplcatons to the regulator or government from the collecton of AIP payments. AIP determnaton s regarded as an optmzaton problem where the regulator effectvely maxmzes overall welfare wth respect to the spectrum fee gven resource and nterference constrants. Ther work combnes both approaches of productve and allocatve effcency and s ntroduced and dscussed n ths subchapter Formulaton of the spectrum assgnment problem The assgnment of spectrum to users can also be called channel assgnment snce a channel s smply a specfed frequency range. 17 The spectrum assgnment problem s n ths secton termed as 17 For example n Fnland the natonal (tv) channel 5 operates on frequences from 174MHz to 181MHz (Fcora 2012). 50

52 the channel assgnment problem n order to follow the well-known termnology used throughout the lterature dscussng ths ssue. The channel assgnment problem can be seen as a mathematcal problem of dvdng scarce spectrum resources (.e. channels) between competng, though predetermned (snce allocaton has been conducted) set of demands whle takng nto account the constrants mposed by nterference (Levne & Rckman 2007, p. 3). In other words, the problem specfcaton conssts of nformaton on requrements (demand) for spectrum across the system, the constrants mposed to lmt nterference and the specfcaton of the objectve to be fulflled whle satsfyng the spectrum requrements and the nterference constrants. The spectrum requrements are ntroduced by specfyng the amount of dstnct channels each transmtter ste regures. For n dfferent transmtter stes T 1, T 2, T 3 Tn there exst correspondng demands of m 1, m 2, m 3 m n channels, where ste T requres m dstnct channels. We have a set of constrants each relatng to a sngle transmtter ste T, known as co-ste constrants or to a par of transmtter stes (T, T j ), known as nter-ste constrants. For smplcty, the dfferent channels are labeled wth nteger numbers, whch correspond to the locaton of the channels n the spectrum band. When f requres that, ( ) 1 and f ( ) 2 are channels both assgned to a transmtter ste T the co-ste constrant f ( ) ( ) 1 f 2 (1) where s a specfed mnmum channel separaton,.e. dstance n the spectrum between two dstnct channels, whch ensures that nterference between the channels s kept tolerable. Then naturally for channels f () and f ( j) assgned to dfferent transmtter stes T and T j the nter-ste constrant requres that, f ( ) f ( j) j (2) where smlarly to (1) j s a specfed mnmum channel separaton. The nterference lmtng coste and nter-ste constrants are thus specfed by and j constructng the constrant matrx, where form the dagonal entres and j the non-dagonal entres. These constrants reflect the use of protecton rato (.e. the sgnal-to-nterference rato) n the rado communty. 51

53 The objectve of the channel assgnment problem can be specfed as ether mnmzng the span requred (.e. the dfference between the hghest and lowest channel used) subject to the constrants, or as a fxed spectrum problem where gven the maxmum span (.e. the amount of spectrum avalable) the channels are assgned to as many spectrum requrements as possble. The latter approach s used by Levne and Rckman. Ths approach mples that the transmtter network and power are fxed and thus effectvely taken nto account by the constrant matrx. An alternatve would be to have these as extra varables n the model to be optmzed wth the channel assgnment, but such theoretcal work s scarce and thus the more smplstc approach s taken. (Levne & Rckman 2007, p.5) The channel assgnment problem constructed above has been studed wth the help of basc graph theory and specfcally the graph-colorng problem. 18 A graph s a collecton of abstract nodes, of whch some are joned by edges. The colorng problem attaches a color to each of the nodes n a way that no adjacent nodes share the same color and the overall amount/number of colors s mnmzed. Ths mnmum number of colors s called the chromatc number of the graph. Ths problem relates drectly to channel assgnment when the nodes are thought of as transmtter stes and the colors as channels. For example, f we determne that m equals 1.e. each transmtter requres only one channel, and j equals 1 f the nodes T and T j are joned and 0 otherwse we end up wth the mnmum span channel assgnment problem dscussed above (snce each ste requres one channel the values for co-ste constrants are mmateral). In physcal terms, we model cochannel (nstead of adjacent channel) nterference and the edges represent the rough locaton of potental coverage blackspots. (p.5) The next step s to relate the channel assgnment problem to an economc model explanng spectrum demand n terms of market condtons and costs. Ths s done by assumng that each node or transmtter ste ncorporates a local market wth an olgopolstc market structure. Ths s n accordance wth many of the actual product and servce markets utlzng spectrum n real lfe (as was prevously dscussed n 4.1.3) and brngs a clear correcton to the Smth-NERA model whch assumes perfect competton. There should be no restrctons n nterpretng the local markets as natonal markets (e.g. the Fnnsh moble communcatons market) or alternatvely as local markets wthn natonal markets (e.g. moble communcatons market wthn the Eastern Fnland). The frms 18 For a more thorough explanaton of graph theory and the graph colorng problem see e.g. Graph Theory by Destel Renhard (Sprnger 2006). 52

54 operatng n these local markets are assumed to provde homogeneous products. Note that products across markets can stll dffer. A spatal nterpretaton of the transmtter stes s to regard them as cells.e. specfc regons of servce. Thus, a transmtter ste conssts of all the transmtters used by the frms n the local market and they may share some of the transmtters, perhaps aganst a fee. In each cell a local olgopoly offers a local servce, produced wth spectrum resources (channels) as nputs. The frms purchase a lcense (whch s equvalent to chargng an AIP from ther use) from the regulator to use a certan amount of channels dependng on the level of output. Wthn a cell the frms are so close to each other that no spectrum re-use or sharng can occur between them, and we assume that there exsts only co-channel nterference,.e. =1 n (1) and j =0 or 1 n (2). Ths assumpton by Levne and Rckman s n accordance wth realty, as a certan geographcal dstance s requred n order to be able to share frequences (see chapter 2.3 for more detals). The demand of spectrum s defned by the sum of demands of the ndvdual frms, to be modeled n Each cell s gven a color and a shared color ndcates that spectrum sharng s possble between the regons. Fgures 5 and 6 llustrate ths n a world wth four local markets (.e. four nodes). Fgure 5. A four node graph of four markets Source: Levne and Rckman 2007 In fgure 5 the nodes A-D represent the transmtter stes around whch the local markets are formed. Fgure 6. A coloured map of four markets 53

55 Source: Levne and Rckman 2007 In Fgure 6 the graph has been colored wth two colors n the prevously determned way where no adjacent nodes receve the same color. The four local markets are formed around the four nodes. The numbers nsde crcles represent the demand for frequences (or channels) n each market and shared colors across dagonal markets ndcate the possblty for these markets to share frequences. Due to ths possblty of sharng the total demand for frequences, whch equals 60 channels ( ) can be satsfed wth a mnmum of 40 dstnct channels. An exemplary dvson of frequences s as follows. Market A s assgned wth channels 1-10, of whch market D can re-use all ten channels and n addton requres ten channels more, say Market B s assgned wth channels 11-30, of whch market C can re-use channels The core economc model Ths subsecton constructs the economc model, whch s then combned wth the spectrum assgnment problem depcted n the prevous secton. Frst a sngle local market wth N competng frms and a homogenous servce offered at prce P s consdered (sectors and sector-crossng sales are ncluded later on). The total output Q n the local market equals the ndvdual frms outputs Q N k 1 q k, where k=1,2, N. The output can be thought of as mnutes of a servce requrng frequences as nputs. The demand curve s gven by Q D(P) wth the usual property D '( P) 0 statng that demand dmnshes as prce ncreases. It s also assumed that lm PD( P) 0. The nverse demand s denoted by P D 1 ( Q) P( Q). The output of each frm s produced usng labor (L), captal (K) and frequences (Z) as nputs accordng to a general CES producton functon. Later on Levne and Rckman specfy ths, but frst the basc form s assumed (wthout the frm subscrpt): P 54