A model of Mobile Virtual Network Operator's entryInternational

1、foreclosing or collusive equilibria in the roaming market: a model of mobile virtual network operators entry.carlo capuano dipartimento di teoria e storia delleconomia pubblica, complesso universitario di monte s. angelo, via cintia, napoli, 80126; indirizzo e-mail: carlo.capuanounina.it. i am grate

2、ful to alfredo del monte for his continuous support not only in the development of this paper, to antonio acconcia and fabio massimo esposito that in different moments offer me their relevant suggestions. at last but not least, i thank two anonymous referees for their notations that contributed to t

3、he final version of this paper. all remaining errors are my own.universit degli studi di napoli federico iiaugust 2005the paper deals with the entry of virtual network operators into the mobile communication sector and it proves that (static) non-cooperative foreclosing or collusive equilibria in th

4、e upstream roaming market are sustainable even if there exist more incumbents. this depends on the increase in the downstream complementarity that occurs when an incumbent contracts the entrants upstream roaming access. in fact, the agreement causes a stronger attitude to implement a less aggressive

5、 price competition in the downstream market and, when the incumbents are tacitly coordinated to some foreclosing access charges, the same complementarity is the weakness of a potential upstream deviator that faces a more aggressive competition in the downstream market. in the first part of the paper

6、 we give some evidences that even if exogenous, the entries of new competitors have represented a significative increase in the diffusion speed of mobile communication service. we propose an arellano-bond dynamic panel data estimation on european countries from 1985 to 2003.jel code: d43, l13, l14,

7、l42, l51.keywords: network oligopoly, roaming, mobile, telecommunication.introduction.the mobile communication markets are ones of the more dynamic in ict sector and in the last twenty years they have been generally characterized by the highest rates of growth. as shown in figure 1, in most european

8、 countries the mobile communication services diffusion is now characterized by penetration indexes the penetration rate is generally computed as the number of activated sim cards on population. even if commonly used, it is surely a biased statistics. not lower than 66% (france), while in some countr

9、ies as italy and luxembourg, because of the sim card the acronym sim means subscriber identity mobile, and any single sim card identified any distinct activated mobile numeration. this variable is highly correlated with the number of users. duplication phenomenon, the penetration indexes are higher

10、than 110%! 020406080100120140penetration index7385757687668196115748683899185bdkdelefirlilnlapfinsukfigure 1. the penetration indexes in percentage (% of sim cards/ population) in european countries in 2003. (source of data: eurostat).until 2003, the (sold) sim cards were about 305,6 millions, when

11、just until one year before they were only 250 millions. these evidences are extraordinary when we consider that in 1998, only five years before, we computed not more than 69 millions of sim cards and the data showed a continental penetration index about 18,3 %. only in the period 2002-2003 we measur

12、ed an average growth not lower than 6%. 198019851990199520002003yearsaustriabelgium czech rep.germanydenmarkspainfinlandfrancegreecehungaryeireislanditalyluxembourghollandnorwayportugalswedenunited kingdomfigure 2. the penetration indexes in percentage (sim cards/ population in %) in european countr

13、ies from 1980 to 2003. (source of data: eurostat).nowadays the mobile network operators, mnos, are over than 50, and we compute about 120 mobile service providers, with a sector turnover in 2003 higher than 100 billions euro. in spite of the extraordinary growth and diffusion of the mobile communica

14、tion services in any countries, as shown in figure 3, the national markets are still too concentrated: in 2003 the average cr2 index we recall that the cr-x index is computed as the sum of the x higher firm market shares: for example, the cr2 is the sum of the market share of the first and second bi

15、gger firms in the market. was about 78,8 %. figure 3. concentration rate indexes, cr1 and cr2, in european countries in 2003. (source of data: eurostat).moreover, even if in many countries more than one operator have a significant market power (smp), in 14/15 of the considered sample the leader firm

16、 is the subsidiary of the incumbent fixed operator united kingdom is the only exception, where we find the lower cr2 index about 51,3 %.extracted from the 8th and the 9th european commission reports on telecommunication see. /information_society/ at the moment the paper is going t

17、o be published, data from the 10th of european commission reports on telecommunication are available. nevertheless the new series are not comparable with the others because of an important sim duplication correction that dramatically decreases the penetration rate for 2004. , the mentioned data imme

18、diately show why even if the sector is very dynamic, ex ante and ex post regulation enforcements are still necessary: in the regulatory agenda we find not only the liberalization of the markets (by beauty contests, spectrum auctions or lotteries) but also the regulation of any access prices (interco

19、nnection and roaming). by the way, in each country antitrust authorities are continuously involved in abuse of single or collective dominance investigations.the first part of the paper gives some evidences that even if exogenous, the entries of new competitors have represented a significant increase

20、 in the diffusion speed of mobile communication service. the proposed arellano-bond dynamic panel data estimation underlines the implication that the risk of foreclosing the market to new mobile virtual network is not out of interest from regulatory point of view. and this evidence introduces our th

21、eoretical model about static non cooperative foreclosing or collusive equilibria in the upstream roaming market.the second part of the paper explores the mobile virtual network operators (mvnos) entry mechanism into vertical integrated oligopolies, i.e. network oligopolies. the mvnos are not owners

22、of any physical infrastructures but they typically contract the roaming access with one or more incumbents to use their essential facilities. these access agreements are more and more diffused and represent what we call the “l(fā)iberalization frontier”. in fact, when the presence of more vertical integ

23、rated incumbents induces an upstream price competition, then entry is always accommodated and no structural regulation is needed. otherwise, in order to avoid foreclosing or anticompetitive equilibria a specific national regulatory authoritys enforcement should be required. unfortunately, there not

24、exist robust theoretical results about downstream entry into vertical oligopolies and, maybe as a consequence, we observe very different approaches by the national regulatory authorities, nras, with respect to the roaming access mechanismin spain, norway, sweden, finland and ireland the nras have co

25、mpelled all mnos to guarantee the access by roaming to mvno and service providers. in holland and denmark the nras compelled only the smp operators to guarantee any roaming agreements. differently, without any particular enforcement by oftel, roaming accesses have been contracted in uk. similarly, w

26、ithout any nras enforcement in other countries, as in france (3) or in german (1), some mvno entries occurred. in italy no enforcement happens. but recently the antitrust authority opens an investigation against the three incumbents that refused to contract with any potential mvnos. . the main resul

27、t of this paper is proving that when not regulated, as in italy, incumbents may have a unilateral, i.e. non-cooperative, incentive to foreclose the market to new competitors. anticipating our results, in this paper we stress the increase in downstream complementarity that occurs when upstream an inc

28、umbent signs a roaming agreement with a new competitor: the higher the downstream traffic served by the new competitor, the higher the upstream roaming revenue. indeed, there exists a stronger attitude to implement a less aggressive price competition in the downstream market. but, the crucial point

29、of our analysis is that when the incumbents are tacitly coordinated to some foreclosing upstream pricing, the deviator incumbent becomes the one much more affected by downstream complementarity with the entrant. indeed, if deviation occurs the other incumbents have an immediate instrument of punishm

30、ent by setting lower downstream prices: the incentives to more aggressive strategies are just due to the fall in downstream complementarity. as a last point, we check the role played by network effects in mature market and we find out that there not exist any univocal relationships between network e

31、ffects and competitive or non-competitive equilibria sustainability. note that in oligopoly, any non-competitive equilibrium, i.e. foreclosing or collusive one, represents a clear abuse of collective dominance an evidence of this kind of abuse of collective dominance has recently been the object of

32、an italian antitrust investigation on the three mobile network incumbents that refused to contract access with potential mvnos. see case n. a357-tele2/tim-vodafone-wind. 1 the econometric model of diffusion.we estimate the impact of entry of new competitors on the penetration rate of the mobile comm

33、unication services , measured as the number of sim cards on the population. a similar analysis has been developed in gruber and verboven (2001) estimating a logistic model using panel data up to 1997. their study has underlined the importance of ex ante regulation (liberalization of the market) and

34、technological progress (the transition from analogue to digital signal). our target is testing the same results using a longer time series, up to the 2003, considering 16 european countries (austria, belgium, czech republic, denmark, finland, france, german, italy, island, ireland, luxembourg, neder

35、land, norway, spanish, sweden, united kingdom). indeed, we estimate the following logistic specification the logistic specification is almost standard. see for example greene (2000, pp. 215-216) or, for an application, gruber and verboven (2001). from which where the subscripts i,t, represent countr

36、y and time period, respectively; is the unobserved individual heterogeneity and is such that note that we assume lack of serial correlation but not necessarily independence over time.when and we obtain the following specification that we estimate. the variable t is a time counter, gsmit , mnoit and

37、popit are exogenous explanatory covariates we recall that a variable xit is said to be strictly exogenous if . in our case all the regressors are defined by a regulatory enforcement, then we are confident that the error term at time t has not any feedback on the subsequent realization of xit.: gsm i

38、s a dummy variable that assumes the unit value after the introduction of the digital signal technology, mno is the number of mobile network operators, pop is the population. all data are collected from oecd, eurostat and eu reports.our depend variable is a measure of the penetration rate. because of

39、 network effects we assume that its level at time t is strongly correlated with its lagged value. this is the reason we use yit-1 as an endogenous variable we recall that a variable xit is endogenous when but . our idea is that entries of new competitors and the technological shock can better explai

40、n the deviation from an autoregressive process. so, we estimate our specification by random-effects gls regression as benchmark and we compare it with a proper gmm estimation. number of obs = 288 group variable (i): country number of groups = 16 r-sq: within = 0.9808 obs per group: min = 18 between

41、= 0.9826 avg = 18 overall = 0.9804 max = 18 random effects u_i gaussian wald chi2(4) = 14154.63 corr(u_i, x) = 0 (assumed) prob chi2 = 0.0000 y | coef. std. err. z p|z| 95% conf. interval ly | .923213 .0152037 60.72 0.000 .8934143 .9530117 tmno | .0046482 .0010116 4.60 0.000 .0026655 .0066308 gsm |

42、.0640511 .0342051 1.87 0.061 -.0029896 .1310918 tpop | -4.59e-11 4.15e-11 -1.11 0.268 -1.27e-10 3.53e-11 _cons | .01779 .0455913 0.39 0.696 -.0715674 .1071473 sigma_u | 0 sigma_e | .18547202 rho | 0 (fraction of variance due to u_i) table 1: random-effects gls regression.1.1 dynamic panel data metho

43、dology.we work with a pooled data set of cross-country and time-series observations (data details are given above). we use an estimation method that is suited to dynamic as the generalized method of moments (gmm) for dynamic models of panel data developed by arellano and bond (1991) and arellano and

44、 bover (1995).the general regression equation to be estimated is the where the subscripts i,t, represent country and time period, respectively. y is the dependent variable of interest, that is, the logistic transformation of the penetration index. x is a set of time- and country-varying explanatory

45、variables while is the vector of coefficients to be estimated. finally, is an unobserved country specific effect, and is the error term.parameter identification is achieved by assuming that future realizations of the error term do not affect current values of the explanatory variables, that the erro

46、r term is serially uncorrelated, and that changes in the explanatory variables are uncorrelated with the unobserved country-specific effect. as arellano and bond (1991) and arellano and bover (1995) show, this set of assumptions generates moment conditions that allow estimation of the parameters of

47、interest. the instruments corresponding to these moment conditions are appropriately lagged values of both levels and differences of the explanatory and dependent variables. the latter is because the model is dynamic. since typically the moment conditions over-identify the regression model, they als

48、o allow for specification testing through a sargan-type test. analytically, when we assume that (scalar) and t 3, we estimate a first different transformation .table (1) and table (2) respectively report the random-effects gls regression and the arellano-bond dynamic panel-data estimation we use the

49、 stata statistical software: releas 8.0. college station, tx: stata corporation. for estimation. the latter is derived by imposing a two step estimation using the exogenous variables as instrumental ones. there, all the estimated coefficient are significative with a p-value lower than 5%. even if th

50、e lagged value yit-1 explains most of the panel dynamicity, the number of operators mno has a positive impact on the diffusion speed in both the estimations. this is not true for gsm dummy variable that only in the arellano-bond estimation is significative with a p-value lower than 5%. this variable

51、 explains not negligible shifts of the s-shaped diffusion function forward note that the regressors in are location variables: these variables shift the s-shaped diffusion function forwards or backwards, without affecting the s-shape otherwise. differently the regressors in affect the diffusion spee

52、d. by the way, their final impacts are higher because of the multiplicative effects due to the autoregressive specification. moreover, even if quite negligible because of the different scales, the population variable pop negative affects the penetration growth of the service. with respect the gmm es

53、timation, the sargan test for the over-identification of the restrictions is satisfied the null hypothesis of the sargan test is that the covariance restrictions used in the gmm estimation are computed with valid instrumental variables. even if some studies have found that the two step errors tend t

54、o be biased downward in small samples, it is important to note that the two step sargan test may be better for inference on model specification (see arellano bond(1991) for details). and, more important, the arellano-bond tests for autocovariance in residuals of order 1 and 2 avoid to reject the null hypothesis note that the presence of first order autocorrelation in