|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 ESSAYS IN THE ECONOMICS OF NETWORKS By MIRCEA IOAN MARCU A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2008
2 2008 Mircea Ioan Marcu
3 To my family
4 ACKNOWLEDGMENTS I thank m y supervisory committee David Sappington, Chunrong Ai, Sanford Berg, and James Seale for their prompt expert advice, fo r their patience and grac iousness in guiding me. I have learned a great deal from them and my ot her professors, but also from my colleagues and friends with whom I have crossed hurdles, shar ed ideas and laughs, and who initiated me into their culture and traditions. I would not be here without the help of Lucretia and Mihai, my mother and father in law. My godparents Zoica and Mihai, and my brother in law Nelu ha ve provided encouragement and made me laugh when I needed t o. They are always on my mind. Most of all I thank my wife Mi haela, my parents Ildi and Mi rcea, and my sister Ligia who have been by my side all these years. Without their sa crifices, love, and c onstant support this would not have been possible. I never had to ask, and I never ha d a better opportunity to thank them. They are the inspiration for who one day I would like to become. While I hope to be able to repay all these special people, it is likely as one of my professors said that I will never be able to do so fully. But, as that same wise professor told me, the important thing is to remember what others have done for us, and do the same for those who follow in our footsteps.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4 LIST OF TABLES................................................................................................................. ..........7 LIST OF FIGURES.........................................................................................................................8 LIST OF ABBREVIATIONS.......................................................................................................... 9 ABSTRACT...................................................................................................................................10 CHAP TER 1 INTRODUCTION..................................................................................................................12 2 THE DIFFUSION OF MOBILE COMM UNICATIONS IN COUNTRIE S IN TRANSITION........................................................................................................................16 Introduction................................................................................................................... ..........16 Modeling Mobile Technology Diffusion................................................................................19 The Number of Potential Adopters.................................................................................. 20 The Timing (Location) Parameter................................................................................... 22 The Speed (Growth) Parameter.......................................................................................22 Empirical Specification a nd Data Description .......................................................................25 Estimation Results..................................................................................................................29 Conclusion..............................................................................................................................34 3 AN EMPIRICAL ANALYSIS OF FI XED AND MOBILE BROADBAND DIFFUSION...........................................................................................................................42 Introduction................................................................................................................... ..........42 Literature Review.............................................................................................................. .....44 Empirical Studies on Global Fixed Broadband Diffusion ...............................................45 Empirical Studies on Glob al Mobile Diffusion ............................................................... 45 The Model, Method, and Data................................................................................................ 47 Fixed Broadband.............................................................................................................47 Mobile Broadband...........................................................................................................49 Results and Analysis........................................................................................................... ....51 Fixed Broadband.............................................................................................................51 Mobile Broadband...........................................................................................................54 Concluding Remarks............................................................................................................. .55
6 4 QUALITY PROVISION IN TWO-SIDED MARKETS: T HE CASE OF MANAGED CARE........................................................................................................................... ...........72 Introduction................................................................................................................... ..........72 The Model...............................................................................................................................75 Patients Decision Making..............................................................................................77 Physicians Decision Making.......................................................................................... 79 Profit Maximization......................................................................................................... 79 Iso-Elastic Distributions of Health Risk and Cost .................................................................. 84 Proposition 1. Changes in the Marginal Utility of Quality and the Externality Exerted by Doctors on Patients ................................................................ 86 Proposition 2. Changes in the Dist ribution of Treatm ent Costs...................................... 87 Proposition 3. Changes in the MC Os Cost of Providing Quality .................................. 88 Proposition 4. Changes in the Di stribution of Health Risks ............................................ 89 Proposition 5. Magnitude of Changes Due to Increased Population Health Risk and Marginal Co st of Quality......................................................................... 89 Proposition 6. Changes in the Number of Doctors in the Market................................... 90 Proposition 7. Changes in the Maximum Cost of Treatment.......................................... 91 Conclusion..............................................................................................................................91 APPENDIX A EXPLANATORY AND INSTRUMENTAL VARIABLES USED IN CHAPTER 2, THEIR SOURCES, AND TH E ESTIMATION METHOD ................................................... 94 Definition of Explanatory Va riables and Data Sources .......................................................... 94 Instrumental Variable Definitions and Rationale for Using Them ......................................... 95 Instrumental Variables Estimation......................................................................................... 96 B DERIVATION OF THE MAIN RESULTS IN CHAPTER 4 ............................................... 97 Iso-elastic Distributions of H ealth Risk and Treatm ent Cost................................................. 97 Derivation of the Profit Maximizing Solution................................................................. 97 Sufficient Conditions for Profit Maximization................................................................ 99 Proof of Proposition 1. Changes in the Margin al Utility of Quality and the Externality Exerted by Doctors on Patients............................. 102 Proof of Proposition 2. Changes in th e Distribution of Treatm ent Costs......................104 Proof of Proposition 3. Changes in the Cost of Providing Quality ...............................104 Proof of Proposition 4. Changes in the Distribution of Health Risks ...........................105 Proof of Proposition 5. Magn itude of Changes Due to Increased Population Health Risk and Marginal Cost of Quality .......................................... 106 Proof of Proposition 6. Changes in th e Num ber of Doctors in the Market................... 108 Proof of proposition 7. Changes in the Maxim um Cost of Treatment.......................... 108 LIST OF REFERENCES.............................................................................................................111 BIOGRAPHICAL SKETCH.......................................................................................................117
7 LIST OF TABLES Table page 2-1 Number of digital mobile operators................................................................................... 36 2-2 Former socialist countri es: descriptive statistics................................................................ 37 2-3 Empirical analysis of m obile tele communications diffusi on: different fractions of potential adopters for each group....................................................................................... 38 2-4 Empirical analysis of mobile teleco mmunications diffusion: common fra ction of potential adopters for all groups........................................................................................39 2-5 Cumulative effect on cellula r penetration of a one standard deviation increase in socioeconom ic characteristics, or a 0 to 1c hange in the dummy variables of interest......40 3-1 Broadband penetration rate (top 5 OECD countries), by technology, December 2006 ....58 3-2 Mobile broadband (3G Mobile) penetr ation (top 5 OECD countries), 2005 ..................... 59 3-3 Main empirical studies of fi xed and mobile broadband diffusion .................................... 60 3-4 Variables, measurement and data sources for statistical analysis (fixed broadband) ........63 3-5 Variables, measurement and data sources for statistical analysis (m obile broadband)..... 64 3-6 Logistic regressions of broadband penetration ..................................................................65 3-7 Sample correlation matrix of variab les used in m obile broadband analysis...................... 66 3-8 Linear regressions of mo bile broadband penetration .........................................................67 4-1 The effect of parameter increases on m anaged care reimbursement, quality, number of doctors, marginal health risk, and insurance premium................................................ 110
8 LIST OF FIGURES Figure page 2-1 Evolution of cellular penetration ra tes in the Czech Republic and Estonia ....................... 41 3-1 Cable-modem broadband subscriber s per capita in the United States ............................... 68 3-2 DSL subscribers per capit a in the United States ................................................................ 69 3-3 Cable-modem broadband subscr ibers per capita in Japan ................................................. 70 3-4 DSL subscribers per capita in Japan.................................................................................. 71
9 LIST OF ABBREVIATIONS DSL Digital Subscriber Line FAO Food and Agriculture Organization FTTH Fiber To The Home GDP Gross Domestic Product ITU International Telecommunications Union MCO Managed Care Organization OECD Organization for Economic Cooperation and Development PC Personal Computer USD United States Dollar WB World Bank WDI World Development Indicators
10 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ESSAYS IN THE ECONOMICS OF NETWORKS By Mircea Ioan Marcu December 2008 Chair: David E. M. Sappington Major: Economics I conducted three studies in the economics of networks. In the first study I estimate a logistic model of mobile communications diffusion to examine the impact of competition on cellular penetration in former socialist countries in Eastern Europe and Central Asia. I find that digital technology and the introduction of competition increased the speed of diffusion of mobile communications. Countries had to choose whether to award more licenses from the beginning, or delay competition and award digi tal mobile licenses sequentially. Sequential entry had a greater impact than simultaneous entry, compensating for the delay in the introduction of competition. Larger and richer countries introduced competitio n earlier, and tended to award more than one digital mobile license initially. Failure to accoun t for this selectivity leads to upward biased estimates of the effect of co mpetition on cellular penetration. Broadband communications lie at the heart of the developing information society. In the second study I analyze the factor s that influence the diffusion of fixed and mobile broadband. Using 1999-2005 OECD data, the longe st available panel to date we find that local loop unbundling, PC penetration, population density, and Internet conten t are all associated with higher fixed broadband diffusion. Plat form competition is a signifi cant driver of cable modem broadband, but not DSL diffusion. Multiple standa rds policy, mobile application price, and
11 population density influence the diffusion of mobile broadban d. We also find that fixed broadband is neither a complement nor a substitute for mobile broadband yet. Recent developments in the economics of netw orks have shown the pot ential fallacies of using one-sided logic in two-sided markets. In the third study I develop a two-sided market model to analyze the pricing and quality decisions of a pr ofit maximizing managed care organization (MCO) in the presence of indirect network externalities between doctors and patients. The managed care organization faces trad e-offs when choosing th e quality of service, insurance premiums, and physician reimbursements. These trade-offs de pend on patient health risk and physician cost distributions, the elasticity of supply of physicians with respect to reimbursements, the marginal cost of service qual ity, and the marginal utility derived by patients from access to a broader network of physicia ns and the quality of health services. In the case of iso-elastic distri butions of patient heal th risk and physician cost of treatment, an increase in the cost of pr oviding quality decreases the qu ality provided by the MCO, which leads to fewer policyholders, lower physician reimbur sements, and fewer doctors in the preferred network. The insurance premium also decreases. An increase in the health risk of the population results in lower quality, lower reimbursements, and fewer physicians in the MCOs network. The insurance premium also decreases, but the decrease is smaller than the decrease in individuals utility due to lower quality and fewer physicians, which leads to fewer policyholders.
12 CHAPTER 1 INTRODUCTION I conduct three studies. S tudy 1 is The Diffu sion of Mobile Communi cations in Countries in Transition, study 2 is An Empirical Analys is of Fixed and Mobile Broadband Diffusion, and study 3 is Quality Provision in Two-sided Markets: the Case of Managed Care. Studies 1 and 2 are empirical studies of diffusion of new t echnologies in which netw ork externalities play an important role. They are therefore methodologi cally similar in that they are based on the nonlinear, logistic model of technology diffu sion. Study 3 is also concerned with network externalities, only this time in the cas e of networks of doctors and patients. In the first study I analyze the diffusion of mobile communications in countries in transition. These are former socialist countries in Eastern Europe and Central Asia. I estimate a logistic model of diffusion to examine the role of socio-economic characteristics and the impact of competition on cellular penetration in these countries. I use a carefully constructed set of instru mental variables specific for economies in transition to improve upon existing studies by addressing the endogenous nature of the relationship between wireline and wireless co mmunications. More importantly, I take into account the fact that countries ma y self-select when making key d ecisions such as the timing of the introduction of competition, or the nature of competition (simultaneous versus sequential).1 If, for example, countries that introduced co mpetition earlier are also those that would have benefited the most from doing so, then simple least squares techniques would overstate the effect of competition on the speed of cellular communications adoption. 1 Simultaneous competition refers to the entry of two or more digital mobile communications operators during the first calendar year of digital mobile communications pres ence in a country. Sequential competition refers to entry occurring during a subsequent calendar year, after only one digital cellular communications provider was initially present in the market.
13 I find that the introduction of competition accelerated the rates of cellular adoption significantly. Sequential entry had a more pronounced impact than simultaneous entry, compensating for the delay in the introduction of competition, perhaps through more aggressive pricing by late entrants in order to establish themselves in the market. As expected, the level of adoption was signi ficantly lower during the analog era due to capacity constraints and relatively high prices. Countries with a higher income per capita had more cellular subscribers initially, but this adva ntage declined over time as mobile services became less expensive. Countries with a higher number of fixed lines per capita experienced lower initial levels of cellula r adoption, but higher subsequent growth. The higher speed of adoption could be the result of supply increases due to incr eased rivalry between the two technologies. Alternatively, it could be that th e two technologies are not rivalrous and positive direct network effects dominate. Shortages were common in socialism. A significant fraction of the population could not get access to telecommunications servic es. Countries with a higher fraction of people waiting for a fixed line connection had a lower initial number of mob ile subscribers, but faster subsequent growth. The second study examines the diffusion of fixe d and mobile broadband. It is joint work with Sangwon Lee from the Department of Journalism and Communications. Broadband communications play an extremel y important role in the developing of the information society. Widespread broadband diffusion encourages innovation, contributes to productivity and growth, and attracts foreign investment. There is a growing body of literature on fixe d broadband diffusion, but there are still very few cross-cultural empirical studies examining the important influential factors of global
14 broadband diffusion. Results are not always consistent, and insu fficient data has prevented previous studies from capturing the nonlinear natu re of broadband diffusion. In addition, in spite of its significance and implications, there is no empirical study to our knowledge about influential factors of mobile broadband diffu sion because mobile broadband deployment began only in recent years. For the same reason, we are not aware of any study that investigates whether fixed broadband is a complement or a substitute for mobile broadband. Using 1999-2005 OECD2 data, the longest ava ilable panel to date, we estimate a logistic regression to capture the nonlinear nature and exam ine the influential fact ors of fixed broadband diffusion. We find that local loop unbundling, PC3 penetration, population density, and Internet content all influence broadband di ffusion. Platform competition is a significant driver of cable modem broadband, but not DSL diffusion. Analyzing data from 2005 for a cross-section of 51 countries we also find that multiple standards policy, mobile applica tion price, and population dens ity influence the diffusion of mobile broadband. Our results indicate that fixed broa dband is neither a complement nor a substitute for mobile broadband yet. Study 3 extends the literature on two-sided markets by developing a two-sided platform model of managed care organizations (MCOs) that for the first time analyzes the quality of service provided by the intermediary platform, not just it s pricing decisions. This is also one of the first studies to apply the theory of tw o-sided markets to health care organizations. The main feature of two-sided markets is th e presence of two-way indirect externalities between participants on different sides of an in termediary platform, usually buyers and sellers. 2 Organization for Economic Cooperation and Development. 3 Personal Computer
15 Patients care about the number of doctors in the MCOs preferre d network, while physicians care about the number of patients enrolled in the MCO and their health risk. I find that insurance premiums physician reimbursements, and the quality of managed care services reflect the magnitude of the indirect externalities. Incr eases in population health risk, such as those caused by population aging, lead to decreases in th e MCOs quality of service, physician reimbursements, and number of doctors in the preferred network. Increases in the marginal cost of quality provision also l ead to lower managed care quality, physician reimbursements, and number of doctors in th e MCOs preferred network. The magnitude of these changes depends on the patients preference for quality, the magnitude of the externality exerted by doctors on patients, the shape of the distribution of hea lth risks, and the elasticity of the supply of physicians with respect to reimbursements.
16 CHAPTER 2 THE DIFFUSION OF MOBILE COMMUNICATI ONS IN COUNTRIES IN TRANSITION Introduction Recent stud ies suggest that a priority for deve loping countries seeking to catch up with the developed world is to develop their infrastructure rapidly.4 Given their history of underinvestment in telecommunications and given the rapid pace of technological change in recent decades, former socialist countrie s in Eastern Europe and Central Asia5 have probably more to gain by investing in te lecommunications than in any ot her utility sect or. Technological innovation, in particular the development of mo bile communications, has made it possible for these countries to make up for time lost, and deve lop their communications infrastructure rapidly and at relatively low cost. Less than fifteen years after the first of thes e former socialist countries adopted mobile technology, the countries exhibit a wide array of cellular penetration leve ls, ranging from less than 10% in most former Soviet Union republics to almost 85% in the Czech Republic at the end of 2002. During this short time span, cellula r penetration has far outgrown fixed line communications in some of these countries. It seems important then to explain the success of leading countries and the inability of other countries to capitali ze on this opportunity. Socioeconomic factors, such as the leve l of economic development of each country, certainly played an important role in determining the relative success of th e countries. But part of the success may be due to government policie s affecting the adopti on of cellular technology. These countries faced several important deci sions, from the adoption of a particular technological standard to the timing and numb er of mobile licenses awarded to producers. 4 See Poot (2000) for a synthesis of empirical research on the impact of government on long-run growth, including the impact of public infrastructure investment. 5 I will hereafter refer to these countries as form er socialist countries or transition countries.
17 Among these decisions, the introduction of competition and the timing of licenses had a particularly important impact on the level and speed of mobile communication diffusion.6 There are very few studies of mobile comm unications diffusion. The most notable are the studies by Parker and Rller  on the U.S. mobile telephone industry, and those by Gruber & Verboven [2001a, b], one focusing on the Europ ean Union member countries, the other one more extensive in coverage, including over 150 de veloped and developing countries. Abstracting from technological aspects and the timing of first licenses, Parker and Rll er used a structural model to focus more deeply on competition. They found that the introduction of duopolies in U.S. states had a significant positive effect, though the effect was relatively small. Gruber and Verbovens findings are that real GDP per capita, the size of the fixed telephone network, and the size of the waiting list for fixed lines are all important determinants of cellular diffusion. They also found that competition, technological standards, and the introduction of digital technology played an important ro le in accelerating the diffusion of mobile communications. The effect of competition was considerably larger during the digital era, as the sharp capacity increase brought about by digital technol ogy made competition much more effective. A third study by Gruber  focused on 10 Ce ntral and Eastern European countries, out of the 19 transition countries that are analyzed in this paper. While this study also found a positive impact of competition on cellular penetr ation, it did not find a significant effect of digital technology and real GDP per capita on mobile communications diffusion. As Gruber himself admits, these results ar e puzzling in light of conventional wisdom and his joint work with Verboven. A possible explanation could be the small number of observations (only 64). By 6 With very few exceptions, the former socialist countries adopted GSM as their digital technological standard. Therefore, it is unlikely that technologi cal standards contributed significantly to the difference in outcomes between countries.
18 1997, the last year in Grubers dataset, transition countries had relatively limited mobile communications experience, especially with digi tal technology. A considerable number of years have passed since then and a ne w empirical investigation of ce llular communications diffusion in transition countries is warranted. I analyze the di ffusion of mobile communications in 19 out of the 27 former socialist countries in Easter n Europe and Central Asia from 1990 to 2002. The present study employs a careful ly constructed set of instrumental variables specific for economies in transition to improve upon existing studies by addressing th e endogenous nature of the relationship between wireline and wireless communications. Mo re importantly, it takes into account the fact that countries ma y self-select when making key d ecisions such as the timing of the introduction of competition, or the nature of competition (simultaneous versus sequential).7 If, for example, countries that introduced co mpetition earlier are also those that would have benefited the most from doing so, then simple least squares techniques would overstate the effect of competition on the speed of cellular communications adoption. I find that the introduction of competiti on accelerated rates of cellular adoption significantly. Sequential competition had a mo re pronounced impact than simultaneous competition, perhaps due to more aggressive pricing by late entrants in order to establish themselves in the market. As expected, the leve l of adoption was signifi cantly lower during the analog era due to capacity constraints and relativ ely high prices. Countries with a higher income per capita had more cellular subscribers initially, bu t this advantage declined over time as mobile services became less expensive. Countries with a higher number of fixed lines per capita 7 For the purposes of this study, simultaneous compe tition means the entry of two or more digital mobile communications operators during the first calendar year of digital mobile communications presence in a country. Sequential competition refers to entry occurring during a subsequent calendar year, after only one digital cellular communications provider was initially present in the market. Due to capacity limitations, none of the former socialist countries introduced competition during the analog era, with the possible exception of Russia, which was excluded from the analysis due to its relative size compared to all other countries in the region.
19 experienced lower initial levels of cellular adoption, but higher subsequent growth. The higher speed of adoption could be the result of supply increases due to increased rivalry between the two technologies. Alternatively, it could be th at the two technologies are not rivalrous and positive direct network eff ects dominate. Countries with a high er fraction of people waiting for a fixed line connection had a lower initial number of subscribers, but faster subsequent growth. The rest of the paper proceeds as follows. Sec tion 2 presents the logistic model of cellular technology diffusion employed. Secti on 3 presents the data and the empirical specification. Section 4 presents and discusses th e results. Section 5 concludes. Modeling Mobile Technology Diffusion When analyzing the ado ption of new technologie s, it is essential to account explicitly for the prevailing stages of adoption (Dekimpe et al. ). For a variety of cultural and socioeconomic reasons, some countries adopt new technologies earl ier than others. If leading countries are also the ones to first adopt certain policy decision s, such as the introduction of competition, then a simple cross section analysis would inappropriately attribute the higher level of cellular penetration in early adopting countries to the implem entation of the respective policy decision. Following Gruber and Verboven [2001b], I use a l ogistic function to model the diffusion of mobile technology in order to account for its dynamic nature. The logistic specification is appealing because it provides a means of capturi ng the existence of ne twork externalities in mobile telephony. The flow of new adopters is related to the stoc k of existing subscribers in a way similar to that of epidemic models. Initially, the flow of adopters is small. As the stock of cellular subscribers increases, more and more people adopt th e technology, leading to an exponential increase in the number of adopters. However, the flow of subscribers declines as the stock approaches the total number of potential ad opters in the market. Over time, the diffusion of
20 the new technology thus follows th e classic S shaped pattern. The logistic specification appears to approximate well the actual e volution of cellular pene tration rates, as can be seen from Figure 2-1 depicting the evolution of cellular penetratio n rates in the Czech Re public and Estonia, two of the most advanced transition countries in terms of cellular penetration (see appendix). Letting ity denote the total number of individuals in country i that adopted cellular technology by time t and letting ity denote the total number of poten tial adopters, the standard logistic diffusion equation is ) exp(1*tba y yitit it it (2-1) where itaand itbare parameters, as discussed below. The shape of this function is determined by three elements: the number of potential adopters the timing of initial adoption, and the speed of adoption. The Number of Potential Adopters Not all individuals in a country adopt a new technology, such as m obile telephony. Regardless of how inexpensive mobile telephony might become, some individuals never choose to consume the service. The to tal number of potential adopters *ity in a country may be specified as a fraction i of total population (POPit) itiitPOPy*. (2-2) This fraction varies with income, prices, the fraction of urban population, and other country characteristics.8 In practice, the estimation of the number of potential adopters poses significant problems. In principle, this potential could be estimated as a fixed effect for each 8 Unfortunately, as in other studies, reliable data on the price of mobile services is not available for my sample.
21 country. However, most former socialist countries were in their early stag es of cellular adoption throughout the 1990s. Therefore, there are insufficien t observations to esti mate consistently the potential number of adopters in each country.9 A solution to this problem is to pool the da ta and estimate a common fraction of potential adopters for a group of countries, exploiting the fact that differe nt countries are in different stages of adoption. I estimate th e common fraction of potential a dopters for all countries from Equation 2-1 using nonlinear least squares, after adding a mean zero disturbance term, without including country characteristics in the estimation.10, 11 Once estimated, I treat the fraction of potential adopters as a known parameter, as in Dekimpe et al.12 There is considerable heterogeneity in socio economic characteristics even between former socialist countries in my sample. To account for this heterogeneity, I further divided the countries into four groups, and allowed each group to have a different fraction of potential adopters. Countries were divide d into groups based on geographic and historical considerations in order to reduce as much as possible th e heterogeneity due to both observable and unobservable factors.13 The four groups are: the Baltic Stat es, the other former members of the 9 The difficulty in estimating the fraction of potential adop ters is not a problem only in this study. By the time researchers have enough observations to estimate the potential of a market, that potential is typically well known, and estimations are of limited importance. 10 This is also the approach in Grub er and Verboven (2001a), justified by the relative homogeneity of the European Union member countries they analyzed. 11 Regressions in which the fraction of potential subscribers varies linearly with the real GDP per capita and the fraction of people in each country living in urban areas produced unreasonable results for some countries, such as negative fractions of potential adopters. This result might refl ect the absence of data on th e price of mobile services. 12 Dekimpe et al.  suggest that parameters of adop tion equations can be estimated sequentially instead of simultaneously. They treat the total number of potential adop ters as a known parameter. Their decision is based upon industry interviews. They define the total number of potential adopters in each country as the percentage of literate people living in urban areas and having sufficient income to afford basic telephone service. 13 Kocenda (2001), and Kutan and Yigit (2004) use a similar division of these countries based on geographical and institutional consederations and find evidence of conver gence in macroeconomic fundamentals within groups.
22 Commonwealth of Independent Sates (CIS), the Ba lkans, and countries in Central Europe (CE). A list of countries in each group is presented in Table 2-1. Interesti ngly, this geographicalhistorical division captures most of the hete rogeneity in the levels of adoption reached by countries at the end of 2002. The Timing (Location) Parameter The param eteritain Equation 2-1 indicates the lag (lea d) in adopting the new technology. It determines the initial level of adoption.14 A positive value shifts th e S shaped function upwards while a negative one shifts it downwar ds, without modifying the S-shape. The Speed (Growth) Parameter The param eter itb in Equation 2-1 measures the speed of diffusion. Differentiating Equation 2-1 with respect to time provides *1it itit it it ity yy b ydt dy (2-3) Equation 2-3 shows that itb is equal to the growth rate in the number of adopters, relative to the fraction of potential s ubscribers who have not yet adopted the technology. The second derivative is positive for 2 1*it ity y, and negative for 2 1*it ity y.15 The number of adopters thus follows a symmetric S-shaped pattern, the maximum speed being reached when half of the total number of potential subscriber s has adopted the new technology. 14 Note that 0 1* tas e y yita it it. 15 The second derivative is * 2 2 22it it it it itit itit ity yy y yy yb dt yd
23 The location and growth parameters can be al lowed to vary with country characteristics and policy variables, as in Equations 2-4 and 2-5 it J j j it j iitXD a 1 0 (2-4) it J j j it j iitXD b 1 0. (2-5) The variables j itDare dummy variables capturing the effect of certain events j that take place at time j iT in country i (the introduction of comp etition for example). Thus j itDequals zero forj iTt, and equals one forj iTt. The vector itXincludes continuous variables affecting both the timing and the speed of adoption. The parameters 0 iand 0 iare country-specific timing and speed fixed effects. They capture differences in the timing and speed of adoption due to country characteristics that are not included in the model. All parameters with th e exception of the speed and timing country fixed effects, 0 iand0 i, are assumed to be the same for all countries. Following Gruber and Verboven [2001b], I allow every event to impact both the location and growth parameters. However, I te st the restriction that an event j itD does not lead to a discontinuous jump or fall in the number of subscribers, and impos e the restriction for the events for which I could not reject it.16 For the restriction to hold, the number of subscribers immediately before the occurrence of the ev ent should equal the num ber of subscribers immediately after the event occurred. That is 16 I could not reject the hypothesis that no discontinuity was caused by the introduction of competition. Imposing restriction (6) makes the effect at j iTtyears from the introduction of competit ion to be the same across countries. However, the existence of discontinuities could not be rejected for the introduction of digital mobile communications. This reflects the sharp increase in capacity associated wi th digital technology.
24 ,) ( ) (0 0 0 0 j iit jk k it k j ii it jk k it k j i j iit jk k it k i it jk k it k iTXD XD TXD XD (2-6) which simplifies to j i j jT (2-7) The events for which constraint 2-7 holds im ply only a change in the speed of adoption. Let J denote the set of these events, and K denote the set of events th at lead not only to changes in the speed of adoption, but also to discontinuities in the level of adoption. When the potential number of adopters is known, the following transformation which permits a simple linear form of Equation 2-1 is possible tbaz yy yititit itit it *ln. (2-8) The dependent variable, itz is the natural logarithm of the fraction of potential adopters that did not adopt by time t but that adopt at time t Note that itz can be rewritten in terms of cellular per capita penetration rates as tba m m zitit iti it it ln, (2-9) where it it itPop y m is the per capita mobile penetration rate. Substituting (4) into (7), and account ing for restriction (6) provides Kk k it k Kk k it k j i Jj j it j iti itiittD D TtDtX Xz )( )(0 0 (2-10) Equation 2-10 can be estimated using sta ndard least squares dummy variables and instrumental variables technique s by adding a mean zero disturbance term. Restriction 2-7 can be
25 tested by including j in Equation 2-10, and applying standard F and t tests on the significance of parametersj. One estimation concern is the relatively la rge number of parameters that must be estimated, since there is both a location and a growth parameter for each country. Because of the linearity of (8), the location parameters 0i behave as standard fixed effects in least squares dummy variables regression. The growth parameters 0ihowever will be inconsistent unless the number of observations for each country is suffici ently large. To address this issue, I use an approach similar to the one I used to estimate th e number of potential adopt ers. In particular, I am assuming that all countries within a group have the same location and speed fixed effects.17 The number of fixed effects to be estimated then is only six, instead of thirty-six, two less than twice the number of countries.18 Empirical Specification and Data Description The focus solely on former socialist countries implies some limitations, but also provides some advantages. In particular, the relative homog eneity of these countries is well-suited for the customary assumption of common parameters of in terest across all countri es. I analyze 19 of the 27 former socialist countries in Eastern Eu rope and Central Asia between 1990 and 2002. Several countries were excluded from the an alysis because of missing data (Bosnia and Herzegovina, Belarus, Macedonia, Tajikistan, Turkmenistan, and Yugoslavia).19 Slovenia was excluded because it had a much higher level of GDP per capita than all other former socialist 17 Gruber and Verboven [2001b] addressed the same problem by imposing partial or complete convergence in the level of adoption between countries, thus restricting the speed parameters. 18 Since we are dealing with dummy variables, the locatio n and growth parameters of one group are excluded to avoid multicolinearity. The remaining three pairs of estimated location and growth parameters represent differences in the initial level of adoption and the subsequent speed of adoption relative to the excluded group, that of Central European countries. 19 Some of these countries experienced armed conflicts during the period of study.
26 countries, and my empirical methodology presumes homogeneity between groups of countries.20 Russia was excluded because it is an outlier am ong former socialist coun tries both in terms of population and land area. Russia can be consid ered as having seve ral telecommunication markets. It would be inappropriate to assess the competitiveness of these individual markets based on aggregate data for all of Russia. The level of adoption and speed of a doption vary with country socioeconomic characteristics (see Equation 2-10). The country characteristics included in itX are real GDP per capita expressed in constant 1995 USD, the num ber of mainlines per capita, and the ratio between registered applications for a fixed line and the number of mainli nes in service (waiting list).21 Mainlines are fixed telephone lines connecting the subscribers terminal equipment to the public switched network. Throughout the paper I use the terms mainlines, fixed lines, and wirelines interchangeably. Descriptive statistics are presented in Table 2-2.22 The average cellular penetration rate at the end of 2002 was 25.4%. In the short time span since its introduction, mobile telephony has surpassed fixed telephony in many of these countries. For comparison, the average fixed lin e penetration rate at the end of 2002 was only 21.9%. There is however a wide discrepancy in ce llular penetration levels. Most former Soviet Union Republics had cellular penetration levels sm aller than 10%, compared to almost 85% in the leading country, the Czech Republic. 20 Slovenia was the only former socialist country to be cl assified in the high-income category by the International Telecommunications Union during this period. 21 The length of the waiting list is affected positively by unmet demand, and negatively by the frustration of joining a waitlist without hope of receiving service. Ideally, one w ould also like to account for the effect of the average waiting time for the installation of a fixed line on cellular penetration. However, data on average waiting time is scarce. 22 An additional measure of countries political instability pr oved insignificant in the estimations. The measure is the political constraint variable POLCON V constructed by Heni sz (2000) on the basis of a spatial model of political interaction.
27 On the assumption that mobile communications is a normal good, countries with higher per capita income should have a high er initial level of cellular penetr ation, other things equal. In theory, fixed line penetration may have a pos itive or a negative effect on fixed network expansion, depending whether a dopters view mobile communications as a complement or a substitute for fixed communications. The telecommunications market in socialist countries has long been characterized by significant excess demand for fixed lines. The situation persisted even after the fall of communism, as building communications infrastruct ure required time and a considerable amount of investment. Low and unbalanced tariffs that favored residen tial customers aggravated the problem of excess demand for wireline services One would expect c ountries with a higher fraction of people waiting for the installation of a fixed line to adopt cellular technology faster. Several countries started adop ting cellular technology during th e analog era, before the advent of digital mobile communications. Digital technology brought a sharp increase in capacity and in the number of cellular subscribers. I use a dummy variable (ANALOG) to account for the possible differences in the le vel and speed of adoption between the two technologies. An additional dummy variable (SINGLE_D) accounts for the possibility that countries which only adopted the digital techno logy may have a different level and speed of adoption than countries which started with analog technology. Te sts of restriction 2-7 on these variables rejected the hypothesis of no discontinuity. All former socialist countries in Eastern Europe and Central Asia introduced competition in digital mobile communications by the end of 2002, with the exception of Armenia, Macedonia
28 and Turkmenistan. Table 2-1 shows the number of digital cellular operators in each country, and the years when these operators entered the market.23 Monopolists typically find it prof itable to raise prices and restrict output. The introduction of competition may lead to lower prices, increased demand for mobile services, and faster adoption of cellular technology. Bu t competing firms might also collude and keep prices high. There is ample theoretical literature that st resses the role interconnection charges play in facilitating collusion between competing netw orks (see Armstrong , Laffont, Rey, and Tirole [1998a, b] for example). In the extreme case the price fi rms charge could be the monopoly price. More firms may also have more resources av ailable for investment. But more resources do not necessary mean more investment. Because of demand uncertainty, competing firms might free ride on each others information gathering, and thus might wait a relatively long time to invest (Armstrong and Vickers ). Whether competition leads to faster cellular technology adoption or not is therefore an empirical question. Parker and Rller  find that duopolies in US mobile communications markets behaved much less competitively than expected. Gruber and Verboven [2001a, b] and Gruber  find that the introduction of competition had a positive effect on the speed of adop tion of cellular technology. I use three variables to capture the effect of competition on mobile communications diffusion: a dummy variable for the introduction of digital communications competition (COMPD), and two other dummy vari ables that make it possible to compare the relative effect of introducing competition simultaneously (SIMULTANEOUS) versus sequentially 23 The dates of entry reflect actual launch dates for the prov ision of services, rather than dates when licenses were awarded. Entry dates, as well as the number of subscrib ers of each company, were collected by the author from various issues of the Global Mobile and Global Wireless publications.
29 (SEQUENTIAL). Tests of restrict ion 2-7 could not rej ect the hypothesis of no discontinuity for any of the competition variables. An analysis of the patterns of entry in form er socialist countries reveals that richer countries and more populous countries introduced competition earlier than smaller, poorer countries. Low initial demand for mobile comm unications in poorer and less populous countries did not allow more than one profitable firm in the market. Richer and more populous countries also tended to introduce competition simultaneously. It is important, therefore, to account for the endogeneity of such decisions, and see if the e ffect of competition variables remains positive and significant when accounting for self-selec tion on behalf of these countries. I estimate the model using instrumental variab les, after adding a mean zero error term to Equation 2-10.24 A significant contribution of this paper is the use of a carefu lly constructed set of instruments specific for transition countries, including va riables such as government expenditure as a fraction of GDP, and the fract ion of bank assets held by state-owned banks. Details about the estimating procedure, data sour ces, the instruments used, and the rationale for using them are given in the appendix. Estimation Results The results from estimations in which the four groups of countries were allowed to have different fractions of potential adopters are presented in Table 2-3. The results were very similar when all groups were assumed to have the sa me fraction of potential adopters. They are presented in Table 2-4. The robustness of the results with respect to different assumptions 24 Gruber and Verboven [2001a, b], and Gruber  estimated their models using nonlinear least squares. However, the number of main lines per capita and the fr action of people on a waiting list are endogenous to cellular penetration. Furthermore, the introduction and type of co mpetition also appear to have been endogenous decisions. A Hausman test rejected the null hypothesis of exogeneity of these variables at 1% level of significance.
30 regarding the total number of pot ential adopters is comforting and helps validate the treatment of the fractions of potential adopt ers as known parameters. The first two columns in Table 2-3 present results in which mainlines per capita, the waiting list, as well as all competition variables, were instrumented. As expected, countries with a higher level of income per capita had higher in itial adoption levels, al l other things equal. However, their advantage seems to diminish over time, probably because of decreasing prices of mobile services. These findings ar e in contrast with those of Gruber  who found no effect of real GDP per capita on cellu lar penetration rates in 10 tr ansition countries, but are in accordance with the results of Gruber and Verboven [2001a, b]. Countries with higher wireline penetrati on had fewer mobile s ubscribers per capita initially, but had higher subsequent growth.25 This finding may reflect the fact that former socialist countries were relatively poor in the early 1990s, and most people who owned a fixed line did not acquire mobile servi ces. As prices declined, perhaps driven by supply increases due to lower costs and increased rivalry between the two technologies, mo re people started buying mobile services. Alternatively, the higher rate of growth could be the result of positive direct network effects between technol ogies. Many customers in thes e countries bought prepaid phone cards and used mobile phones only as a means of being contacted, rather than to make calls. Most people who adopted cellular technology did not give up their fixed lines due to high tariffs for mobile-to-fixed and fixed-to-mobile calls. The relative poverty of these countries could also explain the lower ini tial level of adoption in countries with a higher fraction of the popul ation waiting for fixed lines. Generally, people 25 This finding is contrary to that of Gruber and Verboven (2001b), who find that countries with higher fixed line penetration also had a higher initial fraction of cellular subs cribers. However, the differen ce in results is not due to the difference in estimating techniques between the two studie s. In the present study, l east squares dummy variables estimation produced the same qualitative results as instrumental variables.
31 waiting for main lines had modest or little income.26 These people could not afford a mobile phone initially. However, as prices decreased, thes e people started to buy cellular phones, and countries with more people on waiting lists expe rienced relatively rapid growth in mobile adoption. In contrast with Gruber , but in accord ance with Gruber and Verboven [2001a, b], I find that technological factors had an important impact on mobile technology diffusion. The level of adoption was significantly lower during th e analog era due to capacity constraints and relatively high prices. Countries that started adopting cellular communications only during the digital era had lower initial levels of adoption re lative to countries that started the adoption with analog technology. However, they also experien ced higher levels of gr owth, as people who valued mobile communications highly, and w ould have bought the services had they been available during the analog era, bought the servic es when digital mobile communications were introduced. The introduction of competiti on increased the speed of adoption of ce llular technology. The effect is significant even after accounting for the endogeneity of competition variables.27 The effect was particularly pronounced when competition was introduced sequentially.28 The greater impact of sequential competition could re flect the fact that countries which introduced competition sequentially often introduced competition later than countries which introduced 26 Bribes were not uncommon in these countries. Rich or influent people did not have to wait in line for a connection. 27 As mentioned previously, cellular prices are absent from the estimations, as data on cellular prices in these countries is very scarce. Nevertheless, prices of mobile phones and mobile calls have declined throughout the 1990s. To the extent that price decreases are due to competition, the absence of prices from estimations does not lead to biased coefficients. However, if price decreases are also due to other omitted factors, the coefficients of competition variables are overestimated. The use of fixed effects he lps reduce the possible bias due to omitted variables. 28 The coefficients on sequential and simultaneous competiti on in Table I are statistically different at 5% level of significance.
32 competition simultaneously. To the extent there is a catching up effect caused by technological progress, decreases in the cost of mobile services or international learning, it would be reflected in the difference between the eff ect of sequential entry and that of simultaneous entry. Another possibility is that with sequential entry the ne w competitor has to price more aggressively to establish itself in the market. A t test based on the Herfindahl index of market concentration revealed that sequential entry led indeed to higher industry concentration at the end of 2002.29, 30 The third and fourth columns in Table 2-3 present results of estimations in which competition variables were treated as if th ey were exogenous. The comparison with the corresponding results in the first two columns shows the importance of accounting for selfselection on behalf of countries The coefficients from the in strumented specifications are smaller. The difference between the coefficients on simultaneous competition is significantly different from zero at 5% level of significance, a nd so is the difference between the coefficients on sequential competition obtained from the two estimations. The difference between the coefficients on competition from columns 1 and 3 is significant at 10%. These differences translate into sizeable differences in predicted ad option levels, as is evident from the results in Table 2-5. Simulation results in Table 25 were computed using the specifications in Table 2-4 columns 2 and 4, and are based on average and st andard deviation values of variables for the 29 Collusion is easier achieved when the competing networks are of similar size (Laffont, Rey and Tirole [1998a, b]), that is, when the Herfindahl Index is smaller for a given number of competitors. The average value of the Herfindahl index at the end of 2002 was 4150 for countries that introduced competition simultaneously, and 5600 for countries that introduced competition sequentially. The difference is statistically significant at 1% level of significance. 30 The Herfindahl index was employed in alternative specifications of the diffusion model instead of the competition variables described above. The coefficient on the Herfindahl in dex variable turned out to be marginally insignificant. This is due to the reduction in sample size caused by missing values in the number of subscribers of each telecommunications operator.
33 year 2002. All countries are assumed to have the same mark et potential of 59%.31 The average country was considered to be a member of th e third group of countries the Balkans, for the purposes of these simulations. The last two columns present the e ffect of separate changes in each variable.32 The predicted effect of sequential compe tition is the total effect on the cellular penetration level at the end of 2002, assuming th at competition was introduced sequentially in 1999. The predicted effect of simultaneous competition is the total effect on the cellular penetration level at the end of 2002 resulting fr om the introduction of two or more digital operators in 1996. The results il lustrate the tradeo ff between introducing digital competition simultaneously versus sequentially. Sequential competition naturally occurred later than simultaneous competition, but the catching-up effect due to the higher coefficient of sequential competition meant that by 2002 the effects of th e two types of entry were very similar. Table 2-5 also demonstrates the importance of accounting for the endogeneity of policy decisions. When the endogeneity of competition po licy is not accounted for, the model predicts there would be 23 to 24 more ce llular subscribers per 100 individua ls as a result of sequential competition, and 20 to 21 more cellular subscribers as a result of simultaneous competition. In contrast, the corresponding increases when accounting for the endogeneity of policy decisions are only 19 and 18 subscribers per 1 00 individuals, respectively. 31 This value, obtained by constraining all countries to ha ve the same market potential, is close to the value of 62% obtained by Gruber and Verboven [2001a] for the European Union member countries. The value obtained by Gruber  for 10 out of these 19 countries was 17%. 32 The total effect on cellular penetration from changing more variables at the same time does not equal the sum of the effects of separate changes in each variable due to the nonlinear functional form. The negative effect of a one standard error increase in GDP per capita does not necessarily mean that richer countries will have lower penetration levels than poorer ones. Richer countries generally also have higher levels of fixed line penetration, which is associated with reduced mobile adoption. A higher GDP per capita has a positive effect on cellular penetration rates roughly through 2000, after which the effect becomes negative.
34 Conclusion I have analyzed the determin ants of mobile communicati ons adoption in the former socialist countries from Eastern Europe and Central Asia. The so cioeconomic characteristics of these countries played an important role in determining the level and speed of mobile communications diffusion. As expected, countries w ith a higher level of real per capita income had higher initial levels of ce llular penetration, a lthough their advantage diminished over time. Countries with a higher number of fixed lines pe r capita and countries with higher waiting lists for fixed lines experienced lower initial levels of cellular adoption, but higher subsequent growth. Technological factors had an important imp act on mobile technology diffusion. The level of adoption was significantly lower during the analog era due to capacity constraints and relatively high prices. The introduction of competition led to a substa ntial acceleration in the speed of adoption of cellular communications. Furthermore, and perhaps more surprisingly, sequential competition led to higher growth in the number of subscriber s than simultaneous competition. Thus, there appears to be a catching up effect of sequentia l competition, compensating for the fact that sequential entry naturally occurs later than simultaneous entry. This research also demonstrates the importa nce of accounting for the endogeneity of key policy decisions, such as the timing of the introduction of competition. Failure to account for this endogeneity leads to upward biased estimates of the effect of competition on cellular adoption rates. The grouping of countries reflects both historic al and geographical cons iderations in order to minimize the heterogeneity between group member s. A first division is that between former Soviet Union republics and the other former socialist countries This division reflects the
35 common history of the former Soviet Union Republic s, and the fact that most of the republics are still in the sphere of influence of the Russian Federation. Among former Soviet Union Republics, the Baltic States have slightly different cultural and historical background. Th ey are closer in this respect to the former socialist countries that were not part of the Soviet Union. The Baltic States are also the only former Soviet Union countries that have joined the European Union. Countries that were not member s of the Soviet Union were fu rther divided into those in Central Europe, and those in the Balkan Peninsul a. This is not only a ge ographical division. It also reflects the fact that count ries in the Balkan Peninsula we re for a long period of time under the occupation of the Ottoman Empire, while the other countries were (a t least partially) under Austro-Hungarian occupation. The two empires left their mark not only on the culture, but also on the economic institutions of these countries.
36 Table 2-1. Number of digital mobile operators Analog since Country name Group 1994 1995 1996 1997 1998 1999 2000 2001 2002 Albania Balkan 1* 2 Armenia CIS 1* 1994 Azerbaijan CIS 1* 2 1993* Bulgaria Balkan 1 2 Bosnia and Herzegovinaa CE 1 2 3* 1993* Belarusa CIS 1* 2 1991 Czech Republic CE 2* 3 1991* Estonia Baltic 1 2 3* Georgia CIS 2 1990* Croatia CE 1* 2 1990 Hungary CE 2* 3 1994 Kazakhstan CIS 2* 3 Kyrgyzstan CIS 1* 2 1992* Lithuania Baltic 2* 3 1992* Latvia Baltic 1* 2 1995* Moldova CIS 1* 2 3 Macedoniaa Balkan 1* 1992* Poland CE 2 3* 1993* Romania CE 2 3* 4 1991 Russiaa CIS 1991* Slovakia CE 2* 3 1996* Tajikistana CIS 1 2 Turkmenistana CIS 1 1993* Ukraine CIS 4* 5 1993 Uzbekistan CIS 2* 5 Yugoslaviaa Balkan indicates entry by the incumbent fixed line operator into the mobile market. a indicates that the country is not part of estimations.
37 Table 2-2. Former socialist countries: descriptive statistics Country name Group Real GDP per capita in 1995 CST. USD Main lines per 100 inhabitants Cell phones per 100 inhabitants Average 1990-2002 Std. dev.1990 2002 Average2002 Albania Balkan 8121621.2%7.1%2.6% 27.6% Bulgaria Balkan 153111724.2%36.8%31.1% 33.3% Macedoniaa Balkan 244216814.8%27.1%20.0% 17.7% Yugoslaviaa Balkan 16.6%23.3%20.0% 25.7% Estonia Baltic 397061620.4%35.1%29.1% 65.0% Latvia Baltic 252458023.4%30.1%28.0% 39.4% Lithuania Baltic 226139721.2%27.0%26.5% 47.5% Bosnia and Herzegovinaa CE 123947814.0%*23.7%10.8% 19.6% Czech Republic CE 511236215.8%36.2%27.5% 84.9% Croatia CE 459469217.2%41.7%29.6% 53.5% Hungary CE 47635439.6%36.1%25.0% 67.6% Poland CE 30745078.6%29.5%18.5% 36.3% Romania CE 15149510.2%19.4%14.3% 23.6% Slovakia CE 389544713.5%26.8%22.7% 54.4% Armenia CIS 109427115.7%14.3%15.3% 1.9% Azerbaijan CIS 410908.6%11.4%9.2% 10.7% Belarusa CIS 134219115.3%29.9%21.7% 4.7% Georgia CIS 5462609.9%13.1%11.5% 10.2% Kazakhstan CIS 15202568.0%13.0%10.8% 6.4% Kyrgyzstan CIS 4301037.2%7.8%7.7% 1.0% Moldova CIS 88236410.6%16.1%13.2% 7.7% Russiaa CIS 262448814.0%24.2%18.6% 12.0% Tajikistana CIS 5462814.5%3.7%4.2% 0.2% Turkmenistana CIS 15924956.0%7.7%7.3% 0.2% Ukraine CIS 116239713.6%21.6%17.5% 8.4% Uzbekistan CIS 501526.9%6.7%6.8% 0.7% Min 410521.2%3.7%2.6% 0.2% Max 511269224.2%41.7%31.1% 84.9% Average 201533612.7%21.9%17.3% 25.4% data available only since 1994. a indicates that the country is not part of estimations.
38 Table 2-3. Empirical analysis of mo bile telecommunica tions diffusion: different fractions of potential adopters for each group Mobile diffusion parameters Variables z IVa 1 z IVa 2 z Partial IVb 3 z Partial IVb 4 Location parameters for country characteristics Mainlines per capita -29.273*** (-3.73) -30.011*** (-3.80) -28.011*** (-3.62) -27.639*** (-3.69) Waiting list -8.727*** (-3.14) -9.346*** (-3.49) -8.617*** (-3.11) -9.378*** (-3.56) Real GDP per capita 0.0018*** (4.65) 0.0018*** (4.60) 0.0017*** (4.30) 0.0018*** (4.55) Growth parameters for country characteristics Mainlines per capita 3.594*** (3.90) 3.717*** (4.17) 3.458*** (3.79) 3.454*** (3.97) Waiting list 1.008** (2.41) 1.100*** (2.69) 0.976** (2.35) 1.079*** (2.73) Real GDP per capita -0.00016*** (-3.02) -0.00017*** (-3.36) -0.00016*** (-3.00) -0.00017*** (-3.34) Growth parameters for competition variables Competition 0.211** (2.20) 0.243*** (2.58) Sequential entry 0.497*** (2.75) 0.635*** (3.99) Simultaneous entry 0.201* (1.91) 0.257*** (2.80) Location parameters for technology variables Digital only -2.940** (-2.20) -3.204** (-2.44) -2.823** (-2.13) -3.050** (-2.35) Analog -1.925* (-1.85) -2.162** (-2.03) -1.996* (-1.92) -2.329** (-2.26) Growth parameters for technology variables Digital only 0.239 (1.55) 0.281* (1.84) 0.229 (1.49) 0.270* (1.78) Analog 0.161 (0.88) 0.190 (1.02) 0.175 (0.96) 0.220 (1.22) Adjusted R2 N observations N countries Min # obs. per country 0.875 157 19 4 0.881 157 19 4 0.876 157 19 4 0.883 157 19 4 Asymptotic normal statistics in parentheses. Significant at 10%; **significant at 5%; ***significant at 1%. a Mainlines per capita, waiting list, as well as all competition variables were instrumented. b Only mainlines per capita and waiting list were instrumented. The dependent variable is z = ln (mit/( i-mit)), where mit is the number of cellular subscribers per capita, and country groups have different estimated market potentials: i=0.91 if country belongs to the Central Europe or Baltic groups, I=0.63 if the country belongs to the Balkan group, and I=0.15 if the country belongs to the CIS group. All estimations include timing and speed country group fixed effects.
39 Table 2-4. Empirical analysis of mobile te lecommunications diffusion: common fraction of potential adopters for all groups Mobile diffusion parameters z IVa 1 z IVa 2 z Partial IVb 3 z Partial IVb 4 Location parameters for country characteristics Mainlines per capita -30.492*** (-3.83) -29.832*** (-3.73) -29.433*** (-3.74) -28.193*** (-3.69) Waiting list -8.172*** (-2.99) -8.596*** (-3.22) -8.074*** (-2.97) -8.632*** (-3.29) Real GDP per capita 0.0017*** (4.18) 0.0017*** (4.28) 0.0017*** (4.15) 0.0017*** (4.24) Growth parameters for country characteristics Mainlines per capita 3.818*** (4.04) 3.749*** (4.13) 3.707*** (3.95) 3.565*** (4.00) Waiting list 0.912** (2.22) 0.979** (2.41) 0.883** (2.16) 0.972** (2.47) Real GDP per capita 0.00015*** (-2.80) 0.00016*** (-2.95) 0.00015*** (-2.79) 0.00016*** (-2.92) Growth parameters for competition variables Competition 0.214** (2.22) 0.241*** (2.55) Sequential entry 0.468** (2.55) 0.577*** (3.54) Simultaneous entry 0.220** (2.08) 0.252*** (2.73) Location parameters for technology variables Digital only -3.165** (-2.37) -3.286** (-2.49) -3.069** (-2.31) -3.202** (-2.45) Analog -1.967* (-1.87) -2.121** (-1.98) -2.034* (-1.95) -2.226** (-2.15) Growth parameters for technology variables Digital only 0.286* (1.84) 0.307** (2.00) 0.278* (1.800) 0.302** (1.98) Analog 0.174 (0.94) 0.192 (1.03) 0.187 (1.02) 0.209 (1.16) Adjusted R2 N observations N countries Min # obs. per country 0.897 155 19 4 0.901 155 19 4 0.898 155 19 4 0.902 155 19 4 Asymptotic normal statistics in parentheses. *significant at 10%; **significant at 5%; ***significant at 1%. a Mainlines per capita, waiting list, as well as all competition variables were instrumented. b Only mainlines per capita and waiting list were instrumented. The dependent variable is z = ln (mit/(0.59-mit), where mit is the number of cellular subscribers per capita, and all countries are assumed to have the same ma rket potential. The estim ated common fraction of potential subscribers is 0.59. All estimations include timing and speed country group fixed effects.
40 Table 2-5. Cumulative effect on cellular penetrat ion of a one standard deviation increase in socioeconomic characteristics, or a 0 to 1c hange in the dummy variables of interest Mobile diffusion based on estimations with common fraction of potential adopters for all groups of 59% Simulation Setup Cumulative predicted effect on cellular penetration by the end of 2002 from Table 2-4, col. 2 (competition variables instrumented) Cumulative predicted effect on cellular penetration by the end of 2002 from table 2-4, col. 4 (competition variables not instrumented) Variables 2002 Mean 2002 Std. dev. One std. dev, or 0 to 1 increase One std. dev., or 0 to 1 increase Mainlines per capita 0.219 0.101 28.1% 27.0% Waiting list 0.062 0.054 2.5% 2.4% Real GDP per capita 2331 1760 -5.1% -5.1% Sequential entry 0 18.9% 23.5% Simultaneous entry 0 17.7% 20.4% Digital only 0 Analog 0
41 Figure 2-1. Evolution of cellular penetrati on rates in the Czech Republic and Estonia
42 CHAPTER 3 AN EMPIRICAL ANALYSIS OF FIXED AND MOBI LE BROADBAND DIFFUSION Introduction Broadband communications lie at the heart of the devel oping information society. Widespread broadband diffusion encourages innova tion, contributes to productivity and growth, and attracts foreign investme nt. The International Telecommunication Union (ITU) defines broadband as a network offering a combined speed equal to, or greater than, 256 Kbit/s in one or both directions (ITU, 2005; ITU, 2006). As of December 2005, more than 166 countries had launched fixed broadband services and 68 econom ies had launched mobile broadband services (ITU, 2006). Fixed broadband may be defined as transmission capacity with sufficient bandwidth to permit combined provision of voice, data, a nd video through a fixed line such as DSL and cable modem (ITU, 2003). Mobile broadband pr ovided by 3G mobile systems support data transport rates of at least 256 kbit/s (Kilobit per second) for all radio environments, which exceed the rates under second generation wireless networks (ITU, 2006; ITU, 2003; Shelanski, 2003). The 3G mobile systems enable many advan ced video applications such as mobile videoconferencing, video phone/mail, mobile TV/V ideo player, and digital audio/video delivery (Xavier, 2001; ITU, 2001). In spite of the overall rapid gr owth in broadband diffusion, ma ny countries are still in the early stages of fixed and mob ile broadband deployment and are assessing policy strategies to promote faster adoption. Ma ny countries have considered local loop unbundling (LLU)33 and facilities based competition as important policy initiatives to promote rapid fixed broadband diffusion. Platform competition (facilities-ba sed competition among several different broadband 33 Local loop unbundling refers to the process of requiring incumbent operators to open, wholly or in part, the last mile of their telecommunications networks to competitors (ITU, 2003; OECD, 2003).
43 platforms) is often thought to be crucial for reducing prices, improving th e quality of service, increasing the number of customers and prom oting investment and innovation (ITU, 2003; Maldoom et al., 2003). Experts differ on whether single or multiple standards promote faster diffusion of mobile communications. There is a growing body of literature on fixed broadband diffusion, but there are still very few cross-cultural empirical studies examining the important influential factors of global broadband diffusion. The results ar e not always consistent, and in sufficient data has prevented previous studies from capturing the nonlinear nature of broadband diffusion. In addition, in spite of its significance and implications, there is no empirica l study to our knowledge about influential factors of mobile broadband diffu sion because mobile broadband deployment began only in recent years. For the same reason, we are not aware of any study that investigates whether fixed broadband is a complement or a substitute for mobile broadband. Using 1999-2005 OECD (Organization for Econo mic Cooperation and Development) data, the longest available panel to date, we estimate a logistic regression to capture the nonlinear nature and examine the influential factors of fixed broadband diffusion. We find that local loop unbundling, information and communications technology (ICT) infrastructure, population density, and Internet content all influence broadband diffusion. Platform competition is a significant driver of cable modem broadband, but not DSL diffusion. Analyzing data from 2005 for a cross-section of 51 countries we also find that multiple standards policy, mobile applic ation price, and population dens ity influence the diffusion of
44 mobile broadband. 34 Our results indicate that fixed broa dband is neither a complement nor a substitute for mobile broadband yet. This chapter is organized as follows: Section 2 summarizes the existing literature on fixed and mobile communications diffusion, Section 3 presents the methodology and data, Section 4 presents the results, and Section 5 concludes. Literature Review There has been a steady growth of fixed and mobile broadband adoption in the world. There were some 216 million fixed broadband subscribers a nd just over 60 million mobile broadband subscribers at the end of 2005 (ITU, 2006). The adoption of fixed broadband service during the first 10 years of its availability has proceeded more rapidly than the corresponding adoption of previous services such as cellular telephone and di al-up Internet services across OECD countries (OECD, 2006). The dominant fixed broadband access platforms are DSL (Digital Subscriber Line) (with 61.85 % of the market) and cable modem (32.29 %). Other platforms such as fiber-to-the-home (FTTH) a nd wireless broadband access serve approximately 6 % of the market (ITU, 2005). As of December 2006, Denmark, Netherlands, Iceland, Korea, and Switzerland had the highest broadband pene tration rates among OECD countries (OECD, 2007; see Table 3-1). The extent of mobile broadband diffusion varies widely across countries. As of November 2005, Korea, Israel, Canada, Japan, and the Unite d States were the top five mobile broadband economies in terms of the per centage of 3G subscribers (IT U, 2005; see Table 3-2). WCDMA and CDMA 2000 are the two main standards for 3G wireless tec hnologies (Gandal et al., 2003). Most of the European Community adopted WC DMA for 3G wireless services. By November 34 We are constrained to use linear regression to analyze mobile broadband diffusion by the small number of years of data available. Previous studies of first and second generation mobile communications diffusion have evidenced the nonlinear nature of the adoption process (see Gruber and Verboven (2001), and Marcu (2004) for example).
45 2005, almost 92 percent of the European 3G mobile customers subscribed to WCDMA technology-based services (ITU, 2005). On the other hand, many countries in the Americas, Asia, and Africa adopted CDMA 2000 or both CDMA 2000 and WC DMA in their 3G markets. Empirical Studies on Global Fixed Broadband Diffusion There is a growing body of em pirical research about fixe d broadband diffusion. Some empirical studies find that inter-modal compe tition, local loop unbundling (LLU), income, low broadband prices, and demogra phic variables such as populat ion density increase fixed broadband diffusion (Distaso et al., 2006; Garc ia-Murillo, 2005; Grosso, 2006; Lee, 2006b). In their study of 30 OECD countri es, Cava-Ferreruela and Alabau-Mu oz (2006) find that technological competition, low cost of deploying infrastructures, and predilection to use new technologies are key factors for broadband s upply and demand. Analyzing data from 14 European countries, Distaso et al. (2006) argue that inter-platform competition drives broadband diffusion, but that competition in the DSL market does not play a significant role. Using logit regression, Garcia-Murillo ( 2005) finds that unbundling an in cumbents infrastructure only results in a substantial increas e in broadband deployment for mi ddle-income countries, but not for their high-income counterpart s. Kim and others (2003) suggest the preparedness of a nation and the cost conditions of de ploying advanced networks are the most consistent factors explaining broadband uptake in OECD countries. Using generalized least squares, Grosso (2006) finds that competition, income, and unbundling increase broadband diffusion (see Table 3-3). Empirical Studies on Global Mobile Diffusion Previous empirical studies on global mobile diffusion find that standardization policy, competition, and low user cost are influential factors of global mobile diffusion (Gruber, 2001; Gruber and Verboven, 2001b; Liikanen et. al., 2 001; Kioski and Kretsc hmer 2002; Rouvinen, 2006; see Table 3-3). Studies in the economics of standards have focused on the private and
46 social incentives for standardiz ation (Gandal, 2002; David and Greenstein, 1990). In essence, there are both advantages and disadvantages to market-mediated multiple standards relative to a government-mandated single standard. Although ma rket mediated standards might lead to limited network externalities and economies of scale, multiple wireless standards and different types of services across technol ogies enable the existence of diverse competing systems which may lead to more and better mobile services. Gruber and Verboven (2001 b) find that the early diffusion of digital technologies in mobile markets was faster in Europe where most countries had adopted a single standard. Kios ki and Kretschmer (2002) concl ude that standardization has a positive but insignificant effect on the timing of initial entry of 2G services but can also lead to higher prices as it dampens competition. It a ppears that while a government mandated standard was useful in stimulating mobile diffusion in th e initial stage (e.g., first generation mobile), standardization policies become less relevant and even limiting as mobile technology becomes more advanced (Cabral and Kretschmer, 2004). As long as mobile networks are interconnected and coverage is effective, there is little need for compatibility (Ganda l et al., 2003). Cabral and Kretschmer (2004) examined the effectiveness of public policy in the context of competing standards with network externalities and conclude d that current mobile diffusion levels are quite similar between the United States (multiple standards) and Europe (mostly single standard). More recently, Rouvinen (2006) finds that standards competition hinders and market competition promotes diffusion in both developed and de veloping countries. Tabl e 3-3 summarizes the variables and findings of empirical st udies on global broadband diffusion. In spite of a growing body of literature that addresses the factors contributing to fixed broadband diffusion at the national level, the re sults of empirical st udies are not always consistent and insufficient data has prevente d previous studies from capturing the nonlinear
47 nature of broadband diffusion. In particular, the results concerning the e ffects of income, fixed broadband price, competition, and urbanization on fixed broadband diffusion are mixed (see Table 3-3 and Ridder, 2007). In addition, few empirical studies have focu sed on the factors that affect mobile broadband diffusion globally. We are not aware of any st udy that investigates whether fixed broadband is a complement or a substitute for mobile broadband. The Model, Method, and Data Fixed Broadband We estimate a logistic model of fixed broadba nd penetration. The logi stic specification is appealing because it provides a means of captu ring the existence of network externalities (Gruber and Verboven (2001b)). In the beginning few people have broadband Internet access. Because people value the possibility to inter act with and to access content provided by other people, more and more people adopt the technology as the stock of subscribers increases, leading to an exponential increase in the number of broadb and users. However, th e flow of subscribers declines as the stock approaches the total num ber of potential adopters in the market, perhaps due to congestion or low valuation for broadba nd services among the rema ining non-subscribers. In many OECD countries the S-shaped time profile of the logistic curve appears to approximate well the diffusion of fixed broadba nd (see Figures 3-1 to 3-4). Letting ity denote the percentage of country i s population that has broadband access to the Internet by time t the standard logistic diffusion equation is ) exp(1*tba y yitit it it (3-1) where ita and itb are parameters, as discussed below, and*ity is the penetration ceiling or percentage of potential adopters.
48 Not all individuals in a country adopt a new technology, such as broadband, regardless of how inexpensive the technology may become. This is captured in the model by the ceiling parameter*ity The parameterita in Equation 3-1 is a constant of integration that gives the initial value of broadband penetration.35 A positive value shifts the S-shaped function upwards while a negative one shifts it downw ards, without modifying the S-shape. The parameter itb in Equation 3-1 captures the speed of diffusion. This can be seen by differentiating Equation 3-1 with respect to time *1it itit it it ity yy b ydt dy (3-2) Equation 3-2 shows that itb is equal to the growth rate in the number of adopters, relative to the fraction of potential subscribers who have not yet adopted the technology. We allow the speed of diffusion to vary with policy variablesj itD and country socioeconomic characteristicsitX in linear fashion36 it J j j it j itXD b 1 0. (3-3) The country characteristics included initX are variables that are li kely to influence the supply of and the demand for broadband. We exp ect the demand for broadband to increase with average disposable income (measured by real GDP per capita expressed in constant 2000 US Dollars), the number of comput ers per 100 inhabitants, and Internet content specific to the 35 Note that 0 1* tas e y yita it it. 36 Two broad classes of logistic diffusion models have been proposed: the variable-ceiling logistic and the variablespeed logistic (Fernandez-Cornejo and McBride (2002)). Letting the ceiling vary with country characteristics poses significant estimation problems. There is no guarantee that the parameter will stay at theoretically justifiable levels, or that the model will converge. The variable-speed logis tic model is easier to estimate and the speed of adoption can be positive or negative, depending on the movement of exogenous factors.
49 country in question (we use the number of Internet hosts under country-related code names per 10000 inhabitants as a proxy for Internet content).37 Higher population density and percentage of urban population decrease deployment cost increasing the supply of broadband. We are mainly interested in the impact of policy variables on broadband penetration. The policy variables included in our study are dummy variables capturing th e unbundling of the local loop and the existence of platform competition. Our measure of platform competition equals one for years in which both cable and DSL subscr iber existed in the country. The local loop unbundling dummy equals one for years when unbundling was in effect and zero otherwise. The dummy variables thus change over time, depending on the ti ming of the introduction of competition and the year unbundling began. Some previous studies have found that inter-modal competition and local loop unbundling are important determinants of broadband penetration (Cava-Ferreruela and Alabau-Mu oz, 2006). A third policy variable, the share of state ownership in the incumbent telecommunications operator turned out to be insignific ant and did not affect the qualitative results of our estimations, so we excluded it from the reported results and estimations. 34 records the variables in our re gression analysis, their measures, and the data sources. Mobile Broadband We employ a secondary dataset and linear regression to examine the in fluential factors of global mobile broadband diffusion. As mobile broadband is relatively young, data is only available for a cross-section of 51 countries. Tabl e 3-5 records the variables used, their measures, and the data sources. 37 A host is a domain name that has an IP (Internet Protoc ol) address associated with it. There were approximately 64 million hosts connected under OECD country-related count ry code domains in January 2004. The largest country code domain at that time was jp (Japan), though the number of hosts under various US-related domains (.us, .edu, .mil, .gov) was larger.
50 The dependent variable, the 3G mobile diffusi on rate, is measured by the number of 3G mobile subscribers per 100 inhabi tants in a country. Previous studi es (Madden et al. 2004; Koski and Kretschmer, 2002; Gruber 2001; Ahn and Lee, 1999) also used mobile subscribers per 100 inhabitants to measure mobile diffusion. To exam ine the effect of sta ndardization policy, we employ a dummy variable equal to one if a count ry had multiple standards, and zero otherwise. We use the price per minute of a local m obile call during peak hours in US Dollars collected by the ITU to measure th e price of mobile services in each country. This measurement refers to the average price of one -minute mobile calls within the same network, off-net, and to a fixed line during peak hours (ITU, 2005). Regard ing non-voice mobile applications, we adopted the price of short message services (SMS) as a proxy for the price of mob ile broadband relevant applications. SMS is a feature available in many new digital phone s that lets users receive and send short text messages. For the measurement of fixed broadband price, we employ USD per 100Kbit/s. Education could be a driver of mobile broa dband adoption, as literate people and those with higher education/degree leve ls probably have more to benefit from broadband Internet access (Garcia-Murillo, 2005; Clements and Abramowitz; 2006). For the measurement of education we use the UNDP education index38, which is based both on the adult literacy rate and the combined gross enrollment ratio for primary, secondary and tertiary schools (UNDP, 2005). We use population density measured in inhabitants per km2 and the percentage of urban population to capture demographic aspects of diffusion. Previous st udies have found that population density is a key driver of broadband deployment, while the share of urban population 38 The United Nations Development Programme (UNDP ) education index measures a countrys relative achievement in both adult literacy and combined primary, secondary and tertiary gross enrolment. Initially an index for adult literacy and one for combined gross enrolment ar e calculated and then these two indices are combined to create the education index, with two-thirds weight given to adult literacy and one-third weight to combined gross enrolment (UNDP, 2005).
51 was not significant (Gru ber, 2001; Liikanen et. al, 2004; Ko ski and Kretschmer, 2002; Rouvinen, 2006). We use GDP per capita expressed in US dolla rs as a measure of disposable income. Finally, to assess the importanc e of information and communicat ion technology infrastructure, we adopted the usual ICT measures of estim ated PCs per 100 inhabitants (ITU, 2006) and teledensity (i.e., main telephone lines per 100 inhabitants). Data comes fr om the International Telecommunication Union and UNDP. In summary, the linear regre ssion composite model we employ to examine the influence of quantifiable variables on the diffu sion patterns of 3G mobile in individual countries is Yt (3G Mobile Penetration Rate) = 0 + 1(Dummy-Standardization Policy) + 2(ICT Infrastructure) + 3(Mobile Price) + 4(Population Density) + 5(Education) + 6(Teledensity) + 7(Income) + 8(Fixed Broadband Price) + 9(Price of Mobile Application) + t (3-4) Results and Analysis Fixed Broadband Our data covers all 30 OECD countries. To our knowledge, no other study employs data as recent as 2005 when examining broadband diffusion. We estimate the variable-speed logistic model described in Equations 3-1 and 3-3 by nonlinear least squares, after adding disturbances to Equation 3-1. The results are presented in Table 36. In addition to the re sults of fixed broadband penetration estimations, Table 3-6 also presents results of separate estimations for DSL and cable penetration.39 39 Other high speed technologies such as WiFi and fiber to the home (FTTH) played a relatively insignificant role in Broadband Internet penetration for th e countries and years in our sample.
52 The use of personal computers was associated with higher broadband and cable penetration levels, though not with DSL penetration. Income was significant only in the cable equation, and had an unexpected negative sign. It is possible that we are unable to adequately disentangle the effects of these variables. Per capita disposab le income is highly correlated with our ICT infrastructure indicator, the number of personal computers per 100 inhabitants (the correlation is close to 0.8). While this may lead to multicollinea rity and imprecise estimates of the coefficients, theory suggests they are both ve ry important determinants of broadband penetration. For this reason we opted to report results with both variables included in the estimations. Alternative estimations that included only one of the variables provided qualitatively similar results. Our proxy for available Inte rnet content (the number of Internet hosts per 10000 inhabitants) is positively associated with the dependent variables in all our models. However, this result should be interpreted with care, as hi gher broadband pe netration may in turn lead to a larger number of Internet hosts. That is, the variables may be endogenous. The same can be said about the number of PCs per 100 inhabitants. To address this potential endogeneity problem we also ran estimations using one y ear lags of the two variables. The Internet content proxy became insignificant in the cable and DSL estimations (t he variable was only si gnificant at 10% level before using the lags). All other results we re qualitatively and numerically similar. We find that higher population density increa ses the speed of broadband penetration. However, the percentage of urban population is insignificant in all thr ee equations. We include both the percentage of urban population and populat ion density in the esti mations, as there is little correlation between them in our sample. Th eoretically they are both drivers of cost, and hence the supply of broadband. The relatively low correlation is probably due to the existence of two types of countries in our sample: large countri es like the US, Canada, and Australia, with a
53 high percentage of urban populati on yet relatively low population de nsity, and smaller countries, like most European countries, that have bot h a relatively high populat ion density and a high percentage of urban population. LLU encourages competition by reducing economi c barriers to entry, thereby allowing new entrants to build some com ponents of their networks and obtain other components from the incumbent DSL operator (OECD, 2001). Competiti on may bring real choice for customers and reduce prices in the broadband access market (Lee, 2006b; Maldoom et al., 2003). However, LLU may also result in lower broadband penetr ation by reducing the ince ntives of incumbent suppliers to invest in broadband infrastructure. There is great va riation in the ways local loop unbundling has been implemented in different count ries. Unfortunately, we were unable to find measures such as leased line prices or the shar e of leased lines in to tal lines for a large enough number of observations in our sample. Ne vertheless, the dummy variable capturing the implementation of local loop unbundling is significant in all our estimations. Local loop unbundling appears to have significantly in creased the speed of broadband diffusion. According to an ITU report, the existen ce of strong platform competition among DSL, cable modem, fiber, and wireless broadband in a ma rket may ensure that prices remain low (ITU, 2003). In this context, in the broadband access ma rket, regulation across platforms should be as competitively neutral as possible to sustain strong platform competition. An interesting result is that of the effect of platform competition on broadband penetration. While platform competition is insignificant in our estimations of overall br oadband penetration, a clos er look at the equations we estimated separately for cable and DSL penetr ation reveals the fact that platform competition is significantly associated with a higher speed of cable modem broadband diffusion.
54 One concern is potential self -selection on behalf of c ountries implementing policy decisions like unbundling the local loop. To address this issue and to furthe r check our results we ran separate estimations for the sub-samples of EU (European Union) and non EU member countries. EU member countries had to implemen t local loop unbundling to comply with an EU directive, so there is less con cern about self-selection in the EU sub-sample. The results appear to be robust. As before, LLU was positive and si gnificant in all the estimations, while platform competition was significant only in the cable eq uation. Unbundling speeded up the upgrading of switches and ensured the wide av ailability of DSL (ITU, 2005). Mobile Broadband Our examination of mobile broadband di ffusion employed 51 observations using linear regression analysis. Two models were identified from the multiple regression analysis. Initially, all nine independe nt variables were employed.40 As multicollinearity is usually a concern with a relatively low number of observa tions, Table 3-7 shows the correlation matrix among independent variables. Multic ollinearity does not appear to be an issue in our sample. The left panel in Table 3-8 provides the results of this first extende d regression. The regression model is significant at 1% level. The multiple standards dummy variable, populati on density, and the proxy for the cost of mobile applications (cost of SMS service) are all statistically signi ficant at the 1% level. Mobile service price, income, teledensity, ICT infrastructure, and education are not statistically significant. Also insignificant is the price of fixed broadband, so fi xed and mobile broadband appear to be neither substitutes, nor complements yet. 40 As in previous studies, the share of urban population was insignificant in our preliminary analysis. We therefore report the results of estimations that include only population density.
55 To check the stability of the results, the mo st insignificant variables were removed and a second, reduced model was estimated.41 Teledensity, fixed broadband price, ICT infrastructure, and mobile service price were removed. The ri ght panel in 3-8 provide s the results of the resulting reduced model. The regression is still significant at the 1% level. The multiple standards dummy variable, populatio n density, and the proxy for the co st of mobile applications (cost of SMS) were again statistically significan t at the 1% level. Furthermore, the estimates appear to be very stable when comparing th e two regressions. We th erefore conclude that multiple standards, higher population density, an d lower mobile application prices were all significant drivers of mo bile broadband diffusion. Our findings support the assertion of Cabral and Kretschmer (2004) that, as mobile technology becomes more mature, st andardization and its scale and efficiency benefits seem to become less relevant. This illustrates the impor tance of market mediated multiple standards when a new technology evolves in to a different stage of deve lopment characterized by more advanced, differentiable features. In this contex t, technological diversit y is likely to foster innovative applications and better consumer choices. Concluding Remarks In this study we analyze the factors that influence the diffusion of fixed and mobile broadband. The results of logist ic regression indicate that ICT infrastructure, population density, and broadband content are associated with faster fixed broadba nd diffusion. We also find that LLU policies have been successful in promo ting fixed broadband di ffusion in many OECD countries. The effects of platform competition on fixed broadband diffusion are very interesting. We found that platform competiti on contributed mainly to cable modem diffusion. More refined 41 The removed variables had p-values above 25%, and with the exception of fixed broadband price they all had pvalues above 40%.
56 research about the impacts of platform competiti on is necessary in the future. A high level of ICT infrastructure and broadband content are also important drivers of fixed broadband diffusion. Higher population density contribut es significantly to fixed br oadband deployment. In light of the recent debates and allegations that some countries have fallen behind while others are ahead in the broadband adoption race, our results s uggest that larger, less densely populated countries are indeed at a disadvantage when it comes to the deployment of broadband communications. One should theref ore exercise caution and conduc t a careful analysis before concluding that certain countries are ahead or are falling behind.42 Our linear regression results suggest that a market-based multiple standards policy significantly contributes to the diffusion of mobile broadband se rvices. The price of mobile broadband applications is also an influential factor of mobile br oadband diffusion. It seems that a lower price for popular mobile applications is critical in attracting the consumers to the more advanced non-voice mobile services (Lee et al., 2007). In other words, a competitive pricing system for applications that can leverage the unique characteristic s of mobile broadband networks encourages rapid and wi despread diffusion of mobile br oadband. Just like in the case of fixed broadband, population density contributes significantly to mobile broadband diffusion. This study is limited by the relatively small number of observations in our sample, particularly on mobile broadband. As more data becomes available it may be possible to capture the potentially nonlinear na ture of mobile broadband diffusion, and to construct better measures of broadband policies such as LLU and platform competition. Based upon studies that suggest mobile telephony serves as a s ubstitute for fixed phone services (Feldmann, 2003), one might 42 In a policy paper written at the same time as this paper, Ford et al. (2007) propose a policy-relevant method for comparing broadband adoption among countries.
57 expect a similar relati onship between fixed and mobile broa dband. Our preliminary findings did not support the hypothesis that fixed and mobile broadband are substitutes, nor did they support the hypothesis that they are complements. It ma y be too early to indicate a permanent result, considering that mobile broadba nd is relatively new in the mark etplace. Further research is necessary to better assess the nature of the re lation between fixed and mobile broadband, as well as the effects of broadband policies on broadband diffusion.
58 Table 3-1. Broadband penetration rate (top 5 OECD countries), by technology, December 2006 DSL Cable Fibre/LANOther Total Rank Total subscribers Denmark 19.6 9.4 2.6 0.4 31.9 1 1,728,359 Netherlands 19.5 12.0 0.4 0 31.8 2 5,192,200 Iceland 28.8 0 0.2 0.6 29.7 3 87,738 Korea 11.4 10.7 7.0 0 29.1 4 14,042,728 Switzerland 18.8 8.8 0 0.9 28.5 5 2,140,309 Data were derived from Organization for Economic Cooperation and Development broadband statistics (2007).
59 Table 3-2. Mobile broadband (3G Mobile) pe netration (top 5 OECD countries), 2005 Data were derived from the International Te lecommunication Union In ternet reports (2005). 3G Mobile penetration 3G as % of all mobile subscribers Rank Total 3G subscribers Korea 57.37 75.2 1 27,509,000 Israel 27.78 25.4 2 1,823, 000 Canada 23.31 49.4 3 7,400,000 Japan 20.10 28.1 4 25,700,000 United States 16.68 27.4 5 49,550,000
60 Table 3-3. Main empirical studies of fixed and mobile broadband diffusion Study Fixed broadband Main independent variables Significant findings Study Mobile broadband Main independent variables Significant findings Kim et. al. (2003) Broadband price Preparedness of a nation Gruber (2001) Income Late mobile adoption Dial-up service price Population density Urban population Multiple operators 30 countries Income 140 countries Fixed penetration High fixed penetration Preparedness of a nation Wait time Wait time Competition Digital mobile competition Population density Number of mobile operators Policy (unbundling, cross ownership, government funding) Market transition index Garcia-Murillo (2005) Broadband price Broadband price Liikanen et al. (2001) Income Digital mobile introduction Income Income Urban population hinders analog mobile diffusion 92 countries Education Population dens ity 80 countries Mobile telephony operation Generation-specific results differ from Competition Competition Analog/digital penetration generic results Population density Internet access Number of analog/digital standards Policy (unbundling, cross ownership) Unbundling Years since introduction Content Standard (dummy) Personal computers Fixed penetration Internet access Age-dependency ratio
61 Table 3-3. Continued Study Fixed broadband Main independent variables Significant findings Study Mobile broadband Main independent variables Significant findings Distaso et. al. (2006) Intra-modal competition Inter-modal competition Gruber and Income Competition Inter-modal competition LLU price Verboven (2001b) Fixed penetration Single standard 14 countries Rights of way Digital mobile Incumbent pre-empt LLU price 140 countries Standard sequential entry Price of leased line Competition Price of ten minutes call Cava-Ferreruela and Broadband price Technological competition Koski and Income Between and within standards Alabau-Mu oz (2006) Competition Cost of Kretschmer (2002) Urban population competition Infrastructure investment deploying infrastructures Competition Lower user cost 30 countries Telecom services penetration Economic indicators 32 countries Analog mobile penetration Internet indicators Demographic indicators Dominant digital mobile standard Economic indicators Mobile operators (dummy) Demographic indicators Education indicators Social indicators
62 Table 3-3. Continued Study Fixed broadband Main independent variables Significant findings Study Mobile broadband Main independent variables Significant findings Grosso (2006) Competition Competition R ouvinen (2006) Income Standards Income Income Population Competition 30 countries Unbundling Unbundling 165 coun tries Standards Network effects Fixed Internet penetration Fixed penetration / user cost D e v e l o p m e n t Technology Democracy
63 Table 3-4. Variables, measurement and data sour ces for statistical analysis (fixed broadband) Variables Measurement Data sources Fixed broadband diffusion Fixed broadband subscribers per 100 inhabitants OECD (1999-2005) Cable modem diffusion Cabl e modem subscribers per 100 inhabitants OECD (1999-2005) DSL diffusion DSL subscribers per 100 inhabitants OECD (1999-2005) Income GDP per capita ITU (1999-2005) ICT infrastructure Estimated PCs per 100 inhabitants ITU (1999-2005) LLU Dummy (1 for with LLU, 0 for no LLU) OECD (1999-2005) Population density Popul ation density (per km2) ITU (1999-2005) Urban population Percentage of ur ban population Euromonitor (1999-2005) Broadband content Internet hosts per 10000 inhabitants ITU (1999-2005) Platform competition Dummy (1 for with DSL and cable modem,0 for only DSL or cable modem) OECD (1999-2005)
64 Table 3-5. Variables, measurement and data sour ces for statistical analysis (mobile broadband) Variables Measurement Data Sources 3G mobile diffusion 3G mobile subscribers per 100 inhabitants ITU (2005), ITU (2006) Standardization policy Dummy variable (1 for multiple standards or 0 for mandated standard) www.3gtoday.com, ITU (2005), ITU (2006) ICT infrastructure Estimated PCs pe r 100 inhabitants ITU (2005), ITU (2006) Mobile service price Per minute local call (USD) peak ITU (2005), ITU (2006) Population density Popul ation density (per km2) ITU (2005), ITU (2006) Price of mobile application Price of SMS service ITU (2005), ITU (2006) Income GDP per capita ITU (2005), ITU (2006) Education UNDP education index UNDP (2004), UNDP (2005) Teledensity Main telephone lines per 100 inhabitants ITU (2005), ITU (2006) Fixed broadband price USD pe r 100kbit/s ITU (2005), ITU (2006)
65 Table 3-6. Logistic regressi ons of broadband penetration Broadband penetration Cable penetration DSL penetration Variables Coefficientst-stat Coeffi cientst-stat Coefficients t-stat Ceiling 21.3516210.35***10.222676.8***13.36406 6.07*** Initial level parameter -3.52229-6.66***-1.96148-4.72***-4.63713 -5.28*** Natural speed -0.18011-0.86-2.33559-5.34***0.052397 0.21 Income -3.14E-06-0.58-1E-05-2.49**1.83E-06 0.28 ICT Infrastructure 0.6091491.98*0.8611862.92***0.451714 1.14 LLU 0.2411453.89***0.3453262.6**0.184515 2.94*** Population density 0.000911.97*0.0008341.94*0.000875 1.74* Urban population 0.001070.360.0003430.090.002213 0.58 Internet hosts 0.00012.46**5.23E-051.75*0.000116 1.83* Platform competition0.0359290.671.7744714.12***-0.04641 -0.52 R-squared 0.88710.75130.8512 Number of observations 205 205 205 Nonlinear regression with robust standard errors; statistically significant at the 10% level; ** statistically significant at the 5% level; ***s tatistically signifi cant at the 1% level
66 Table 3-7. Sample correlation matrix of vari ables used in mobile broadband analysis ICT Infrastructure Mobile price Population density Cost of mobile application Income Education Teledensity Fixed broadband price ICT infrastructure 1 Mobile service price .273 1 Population density .044 .041 1 Price of mobile application -.094 .263 -.037 1 Income .135 .301 -.080 .335 1 Education .435 .022 -.059 .110 -.016 1 Teledensity .049 -.076 .014 .293 .147 .208 1 Fixed broadband price -.135 -.033 -.082 -.149 .007 -.001 .348 1
67 Table 3-8. Linear regressions of mobile broadb and penetration Extended model Reduced model Variable Coefficients t-stat Coefficients t-stat Constant -70.648 -1.487 -54.936 -1.340 Multiple standards policy 9.580*** 3.582 9.328*** 3.766 Mobile service price 3.837 .579 Income .001 1.511 .001 1.527 Population density .030*** 3.662 .031*** 3.908 Fixed broadband price -.073 -1.165 ICT infrastructure -.053 -.851 Price of mobile application -51.276*** -3.051 -41.587*** -2.871 Education 77.324 1.528 60.867 1.435 Teledensity .040 .611 R-Squared .553 .534 Number of observations 51 51 ***Statistically significant at the 1% level
68 Figure 3-1. Cable-modem br oadband subscribers per capita in the United States 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 1998 1999 2000 2001 20022003200420052006 Year
69 Figure 3-2. DSL subscribers per capita in the United States 0 0.02 0.04 0.06 0.08 0.1 0.12 1998 1999 2000 2001 20022003200420052006 Yea r
70 Figure 3-3. Cable-mode m broadband subscribers per capita in Japan 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 1998 1999 2000 2001 20022003200420052006 Yea r
71 Figure 3-4. DSL subscrib ers per capita in Japan The nonlinear nature of broadband diffusion is readily apparent. In Japan, one of the leading countries in terms of broadband penetration, fast er technologies like FTTH have begun replacing DSL and cable-modem. The diffusion pattern alread y exhibits a slowdown in the diffusion of DSL and cable-modem broadband, a characteris tic of the logistic S-shaped curve. 0 0.02 0.04 0.06 0.08 0.1 0.12 1998 1999 2000 2001 20022003200420052006 Yea r
72 CHAPTER 4 QUALITY PROVISION IN TWO-SIDED MARKETS: THE CASE OF MANAGED CARE Introduction Direct externalities, such as those exerted by insured individuals on other members of the same insurance pool through moral hazard and/or a dverse selection, have long been analyzed in health care markets.43 This study abstracts from issues of moral hazard and adverse selection to focus on indirect externalities between doctors and patients and two-sided market aspects of managed care. Two-sided markets are markets in which there are two-way indirect externalities between the participants in a market that inte ract through an inte rmediary platform.44 The number (and/or quality, or some other characte ristic) of members on one side of an intermediarys network affects the utility of enrollees on the other side. A case in point is a managed care organization (MCO) that needs to attract both doctors and patients. Patients care about the availability of doctors. At the same time, doctors care about th e number of patients en rolled in the managed care plan. Yet neither patients nor doctors take into acc ount the indirect externality they exert on members of the other side when they decide whether to participate in managed care. The presence of indirect externalities leads to important departures from fundamental results in traditional microeconomics. For example, departures from marginal cost pricing may not necessarily reflect market power. Such departures could also reflect the relative magnitude of the indirect externalities exerted by each side (A rmstrong ). The pricing structure in such markets is often skewed in favor of one side, wh ose consumption is subsidized from the revenues collected on the other side. Most software platforms and internet portals operate in this manner. 43 See Arrow [1963, 1968], Zeckhauser , Ellis and McGuire , Ma and McGuire , Ma and Riordan . 44 See Rochet and Tirole  for a formal definition of two-sided markets.
73 Recognizing the importance and consequences of th e indirect externalities is essential to avoid the fallacies of applying th e one-sided logic to two-si ded markets (Wright ). Managed care differs from other two-sided markets through the fact that the MCOs expected cost is linked to the expected benefit derived by enrollees through their health risk. The cost of producing a video game on the other hand is the same regardless how much gamers enjoy the game. Therefore, of particular importance in analyzing managed care is the health risk of policyholders, which in turn depends on the quality of health care provided by the MCO, the size of its physician network, and th e insurance premium it charges. To date, two-sided market models have focused almost exclusively on the pricing decisions of intermediaries.45 The study by Pezzino and Pignatarro  written at the same time as the present study is an exception. They study quality provision by competing hospitals in a horizontal differentiation model with regulate d insurance premiums. Bardey and Rochets  analysis of competition between managed care organizations is th e only other two-sided approach to health care mark ets the author is aware of. I extend the literature by analyzing the prof it maximizing choice of quality of service provided by managed care organizations In this paper, the quality variable captures the degree to which the MCO assists physicians in providing health care servi ces. It can be thought of as a quality-increasing input supplied by the MCO, such as diagnostic tests or sophisticated health information processing, storage, and retrieval. I find that the higher the marginal cost of pr oviding quality, the lower the elasticity of the number of doctors with respect to reimbursements, the more enrollees care about quality, and the lower the indirect externality exerted by docto rs on patients, the more the MCO can reduce 45 Evans and Schmalensee , and Rochet and Tirole  provide an overview of the two-sided markets literature.
74 physician compensation when increasing quality a nd maintain the same premium and number of enrollees, particularly at low levels of quality relative to reimbursement. For the case of iso-elas tic distributions of pa tient health risks a nd physician costs, an increase in the marginal cost of providing quality decreases the amount of the quality-increasing input provided by the MCO. This in turn decreases the effectiveness of physicians, which leads to lower physician reimbursements and fewer docto rs in the preferred ne twork. Individuals are unwilling to pay the same insurance premium as before, and the demand for managed care shifts downwards. The insurance premium decreases, bu t the decrease does not fully compensate for the decreases in quality and number of physicia ns. Thus, the MCO reduces the risk of its patients mix to avoid providing the more costly quality. For populations more at risk of becoming ill, as is the case of the ageing population of the United States, the MCO provides services of lower quality, pays lower physician reimbursements, and includes fewer providers in its preferred ne twork. The insurance premium decreases, but the MCO finds it profitable to enroll some higher risk individuals. The percentage decreases in managed care quality, access to doctors, and reimbursements to physicians due to increases in population health risks and in th e cost of quality provision are larger in markets where the externality exerted by doctors on patients is greater, individuals care more about quality, and the elasti city of the supply of doctors w ith respect to reimbursements is greater. Thus, although MCOs provide services of higher quality and access to more doctors in markets where individuals care more about quality and access to doctors, these markets are also the ones most affected when the cost of quality provision and population health risks increase. The results illustrate the role played by indi rect externalities in shaping the pricing and quality decisions of managed care organiza tions and can explain the growing public
75 dissatisfaction with managed care. They unders core the importance of implementing and monitoring the achievement of quality standards in the face of ongoing increases in health risk due to population ageing, and th at of adjusting the rates paid by government programs like Medicare and Medicaid according to the health risk of managed care enrollees. The rest of the study is organized as follows. Section 2 presents the general model. Section 3 presents the results for iso-elas tic distributions of patients hea lth risks and physicians cost of treatment. Section 4 concludes. The Model There is a continuum of individuals of mass Np that differ according to their probability of becoming ill, )1,0( The probability of becoming ill is distributed according to the cumulative distribution function)( F, with twice continuously differentiable probability density function )( f that is positive everywhere in (0,1). The expected utility of type individuals from joining the MCO depends on the number of physicians enrolled in the managed care network (n), the insurance premium (P) they have to pay, and the quality of service provided by the MCO (q) 46 )1,0(,, )( PnqU. (4-1) For simplicity, I assume patients pay no out of pocket costs in the form of coinsurance, copayments, or deductibles. The model thus abst racts from moral hazard and adverse selection issues that have been extensively analyzed el sewhere in order to focus on network externality aspects of managed care. 46 As it is apparent from the utility specification, the model abstracts from income effects. This assumption is not incompatible with global risk aversion and health insurance demand from individuals if the probability of getting ill is small, the insurance premium is small, and the cost of treatment is large (Bardey and Rochet ).
76 The parameter determines the magnitude of th e externality exerted by doctors on patients. Doctors do not factor in to their decision-making the effect their participation has on the utility of policyholders. The size of the provider network matters because the MCO does not pay for any treatment that its enrollees might receive outside its network. The model thus captures the case of a health maintenance organization (H MO). A larger network of physicians may lower patients transportation costs, or increase the likelihood that patients will find physicians matching their preferred style of treatment. Ther e is reason to believe i ndividuals feel strongly about the choice of providers MCOs offer. Con cern about limited choice of providers has even prompted some states to enact laws ensuring th at willing physicians are not excluded from managed care networks (the so-called Any Willing Provider laws), and patients have access to providers of their choice (Freedom of Choice laws).47 The utility of policyholders also depends on the quality of care provided by the MCO (q). The quality variable in this model captures the degree to whic h the MCO assists physicians in providing health care. It can be thought of as a quality-incre asing input supp lied by the MCO, such as diagnostic tests or sophi sticated health information pro cessing, storage, and retrieval. Managed care organizations use various means to control costs, from pre-authorization of procedures to utilization review. While these ar e meant to increase efficiency and decrease the cost of health care, they may also restrict th e provision of health care by physicians and thus potentially affect the quality of treatment. Quality issues have not played a prominent role in twosided market models to date, as studies have focused mainly on the credit card industry and 47 Some restrictions apply even under Any Willing Provider laws. Providers have to have a certain quality certification and have to be willing to accept the same c ontractual conditions (reimbur sements, pre-authorization etc.) as enrolled physicians of the same quality.
77 electronic platforms.48 However, quality of service is a critic al issue in health care, perhaps more important than in most other industries, and hence merits careful attention. There is a continuum of physicians of mass Nd >1 that differ in their cost of treating patients Cc ) (, where C >1 is the highest cost incurred by any doctor for treating a single patient. The fraction ] 1,0[ of this maximum cost incurred by type physicians is distributed according to the cumulative distribution function G( ) with corresponding density function, g( ) which is assumed to be twice continuous ly differentiable and positive everywhere.49 To focus on the quality provided by the MCO, I assume that doctors do not differ in terms of quality. To this end, I also assume that doctors act in the best interest of their patients and do not engage in skimping (lowering the quality of service they provide) or dumping (the practice of refusing to tr eat high cost patients). The utility that a physician de rives from treating a patient (v( )) is the difference between the reimbursement paid by the MCO in the form of a fee for service ( r ), and the cost of treatment ( c( ) )50 )()( crv (4-2) The MCO does not observe patients type or physicians cost of treatment. However, the distributions of patient and doctor types are common knowledge. Patients Decision Making If they choose to do so, individuals can secure health care services from an alternative provider, called plan A Plan A provides services of quality qA, and access to a network of 48 An exception is the study by Pezzino and Pignatarro . 49 The cost c( ) captures both monetary and non-monetary costs of treatment. For simplicity, I assume doctors have constant marginal costs of treating patients. 50 The results of the model would not change if the MCOs paid physicians a fixed amount per year per enrollee regardless whether individuals seek treatment or not (capitation), rather than a fee for service.
78 physicians of size nA, in exchange for a premium PA. Plan A might be viewed as a traditional insurance plan (which gives enro llees access to all the doctors in the market) or Medicare fee for service (FFS), whose premium, quality, and number of physicians the MCO takes as given. I assume premiums are sufficiently high that patient s choose not to join bo th plans, and the MCO finds it profitable to offer a lower quality of se rvice and number of doctors in order to attract individuals with lower probability of getting ill. To avoid confusion, in what follows I will refer to the MCO as plan B or MCO B and index its characteristics accordingly. Patients join MCO B if they derive a higher level of expected utility from joining it than from joining plan A AAAABBB BPnqUPnqU (4-3) Given the alternative offered by plan A the MCO chooses to provi de a lower quality of service network size combination such that it at tracts individuals with lower probability of getting sick BBAAnqnq (4-4) MCO B charges a lower insurance premium than plan A otherwise no patients will join MCO B Therefore, there exists )1,0( for which individuals are i ndifferent between joining MCO B and joining plan A BBAA BA BABABAnqnq PP nnqqPP ),,,,, (. (4-5) Patients of lower health risk ( ) join MCO B while patients of higher health risk ( ) join plan A. Empirical evidence suggests that managed care plans ha ve been successful in attracting a mix of patients with lo wer health risk than traditional health insurers (Miller and Luft , Glied , Deb and Triv edi , and Liu and Zimmer ) and their insurance premiums are often substantially lower (Altman, Cutler, and Zeckhauser ).
79 Physicians Decision Making I assume physicians face no capacity constraint s. Consequently, they enroll in the network of MCO B if the utility they derive from treating patients is nonnegative 0)()( crvB B. (4-6) Only physicians for which the (monetary and non-monetary) cost of treating patients is lower than the reimbursement per patient ( C rB) will join the MCOs preferred network. The number of physicians that join MCO B is thus d B BN C r Gn (4-7) I assume MCO B does not observe the cost of tr eatment of each physician and cannot preferentially direct patients towards certain physicians (e.g., those that have the lowest cost of treatment). Moreover, patients are equally likely to go to any physician because treatment is free once the premium has been paid, and physicians are assumed to be of the same quality. Therefore, physicians enrolling in the preferred network of MCO B expect to equally share cases in the MCO with the other enrolled physicia ns. The total utility a physician of type expects to get from joining MCO B is the product of the expected utility per patient (Bv) and the expected case load (BD) 0)( )( )()(df rG N CrDvVB p BB B B. (4-8) Physicians utility thus depends on the number and type of enrollees. Individuals do not take this indirect externality into account when deciding wh ether to enroll in the MCO. Profit Maximization I assume MCO B is a for-profit MCO, and thus maximizes its profits given by
80 0)()(dfqkrPNBBB BpB, (4-9) where )(BBqkis MCO Bs per patient cost of providing quality qB, which is increasing and convex in qB. The first order conditions characterizing MCO Bs profit maximizing choice of premium, reimbursement, and quality are given by 0)()( )( )( B BBB B B p B BdP d fqkr dP dF PFN P (4-10) 0)()( )( )(0 B BBB B Bp B Bdr d fqkrdf dr dF PN r (4-11) 0)()( )()( )(0 B BBB BB B Bp B Bdq d fqkrdfqk dq dF PN q (4-12) where BA BBAAB BPPnqnqPdP d 1 (4-13) B B BB BAB B BBAA BBBA B B B Br n nq PPr n nqnq nqPP r n ndr d 1 2 2 1 (4-14) BB BA BBAA BBBA B Bnq PP nqnq nqPP qdq d1 2 2 1 (4-15) are the total derivatives of with respect to PB, rB and qB. Sufficient conditions for profit maximization are provided in Appendix B. Equations 4-10, 4-11, and 4-12 reflect the familiar microeconomic paradigm that the profit maximizing firm chooses the insurance premium, physician reimbursement, and quality such that the marginal benefit of each instrument equals its marginal cost. For example, increasing quality allows MCO B to attract more individuals B pdq dF N )( from which it collects premiumBP, a
81 fraction of which will need treatment which costs )(BBBqkr in addition to the increased costs of providing quality to all policyholders 0 ')()(dfNqkpBB (Equation 4-12). While these results are intuitive, they diffe r from the familiar microeconomic theory of one-sided markets through the pr esence of indirect externality effects. The profit maximizing premiums, reimbursements, and quality depend on the indirect externality physicians exert on policyholders, and the externality exerted by policyholders on doctors. When deciding whether to join MCO B, individuals do not take into account the effect their decision has on the utility of doctors in the MCOs network. Similarly, physicia ns do not take into a ccount the added benefit policyholders would enjoy from being able to choose among a larger network of providers. The first order conditions can be rewr itten in the familiar Ramsey form B B BBB BP qkrP 1 )( (4-16) where )( )( F P P FB B B (4-17) is the elasticity of the numbers of enrollees wi th respect to the insurance premium. The more elastic is the number of enrollees with respect to changes in the insurance premium the lower will be the markup charged by MCO B. Physician reimbursement has taken the place that usually belongs to the marginal cost of treatment in one-sided market models. Th e utility of policyholders depends on the reimbursement paid to doctors by MCO B through its effect on the number of doctors in the preferred network C N C r g r nd B B B (4-18)
82 Similarly, the utility of doctors depends on the insurance premium provided by MCO B through its effect on the number a nd type of policyholders B pdP dF N )(. Two-sided market models so far have focused on the optimal pricing decisions of intermediaries. The present model also analyzes the optimal choice of quality of the intermediary MCO. The utility of physicians from joining MCO B depends indirectly on the quality of service provided by MCO B to patients through its effect on the number and type of policyholders Bdq dF )( (Equation 4-12). Combining the first order conditions for profit maximization yields 0 0)()()( )( dfqk dq d df dr d F dP dBB B B B. (4-19) This in turn can be rewritten to analyze the tradeoffs faced by MCO B when choosing the quality of service and physician reimbursement cstP cst B B B B B B BBBq r r n n q qk )('. (4-20) Increasing the quality of service provided to patients allows the MCO to reduce the reimbursement per patient it pays doctors, charge the same insurance premium and attract the same number of patients. Similarly, increasing the reimbursement attracts more doctors which provide policyholders additional utility, allowing the MCO to reduce the quality of service it offers and maintain the same number of patients and insurance premium. Using Equations 4-14 and 4-15, Equation 4-20 can be further rewritten as
83 cstP cst B B B B B B B B BBBq r n r r n q r qk )('. (4-21) The higher the marginal cost of providing quality ) (' BBqk and the lower the elasticity of the number of doctors with respect to reimbursements B B B Bn r r n the more the MCO can reduce payments to its doctors when it increases quality and maintain the same premium and number of enrollees. Also, the more enrollees care about quality (higher ) and the less they care about the number of doctors (lower ), the more the MCO can reduce reimbursements when increasing quality and keep the same premium and number of enrollees, particularly at low levels of quality relative to reimbursement. Similarly, the first order conditions can be re written to analyze the tradeoffs faced by MCO B when choosing the quality of se rvice and the insurance premium cstr cst B B B B BBBq P P q F df qk )( )( )(0 '. (4-22) The higher the marginal cost of providing qua lity for the average policyholder, the more the MCO has to increase the insurance premium to be able to pay the same reimbursement, and maintain the same number of doctors and enrollees. Using Equations 4-13 and 4-15 Equa tion 4-22 can be rewritten as cstr cst B B BB BBBq P nq F df qk 1 0 ')( )( )(. (4-23) By increasing quality, MCO B can raise the insurance premium more if the marginal individual has a higher probability of becomi ng ill and if patients care more about the quality of service
84 (higher ), and still maintain the same number of policyholders, pay the same reimbursement, and keep its network of physicians. Finally, the first order conditi ons can also be used to analyze the relation between physician reimbursement and the insurance premium cstq cst B B B BBr P P r F df )( )(0. (4-24) The higher the average health risk of policyhol ders, the more the MCO has to increase the insurance premium if it wants to attract more physicians by increasing reimbursements, and maintain the same quality of service. Using Equations 4-13 and 3-14, Equation 4-24 can be further rewritten as cstq cst B B B B BBBr P r n nq F df 1 0)( )( (4-25) An increase in reimbursements allows the MCO to raise the insurance premium more when patients care more about the number of physicians in the preferred network, and when the supply of physicians is more responsive to increases in reimbursements. Iso-Elastic Distributions of Health Risk and Cost The model is solved in Appendix B assuming iso-elastic distributions of patient and physician types and constant marg inal costs of providing quality )( F )( G BBBqk )('. (4-26) These power distributions are a convenient way to summarize the concen tration of high-risk patients and high-cost physicians in the market (Bardey and Roch et ). Distributions with higher dominate those with lower in the sense of first order st ochastic dominance. The higher
85 the elasticity of the dist ribution of patient types the higher the number of illness cases. Similarly, the higher the elasticity of the distri bution of physician types the higher the number of doctors with high cost of treating patients. Solving the model yields the following solution (see Appendix B for derivations) 1 11 d B BN C r (4-27) 1 1 11 C N nd B B (4-28) 1 11 d B B BN C q (4-29) 1 1 1 11 )1()1( 1 ]1)1)([( )1( 1 d B AA d B AA A BN C nq N C nq P P (4-30)
86 1 11 )1()1( d B AA AN C nq P (4-31) The following propositions describe the impact of changes in parameters of interest on MCO B s reimbursement, quality, number of docto rs, marginal health risk, and insurance premium. They are summarized in Table 4-1. Th e proofs of these propositions are sketched in the appendix. The results were derived using Ma thematica 5. A separate file detailing the derivations is available from the author upon request. Proposition 1. Changes in the Marginal Utilit y of Quality and the Externality Exerted by Doctors on Patients The MCO provides more of the quality increasing input, pays higher reimbursements, and attracts more doctors when individu als care more about quality (higher ) and/or the number of doctors in the network (when is larger and thus the externality exerted by doctors on patients is greater). The marginal h ealth risk insured by the MCO decrea ses as more individuals choose the alternative plan A that provides higher quality and access to more physicians. Proof: see Appendix B. When individuals care more about quality the MCO provides more of the quality increasing input. This increases the ma rginal utility of doctors, so MCO B pays higher reimbursements and attracts more doctors. Similarl y, the indirect extern ality exerted by doctors on patients does not affect solely the number of doctors in MCO B s network. When policyholders care more about th e number of doctors the MCO finds it profitable to provide doctors with more of the inputs they need to better treat patients. However, if individuals care
87 more about quality or access to doctors more of them will choose the alternative plan A that provides higher quality of service and access to more physicians. Proposition 2. Changes in the Distribution of Treatment Costs The MCO reduces the risk of its mix of pati ents when the elasticity of the supply of physicians with respect to reimbursements increases ( is higher). Proof: see Appendix B. In the model, captures both the concentr ation of doctors with hi gh treatment costs, and the elasticity of the supply of physicians with respect to reimbur sement. Just as firms charge lower prices when demand for their products is more elastic, so does the MCO have to pay higher reimbursements to attract doctors when the s upply of doctors is more elastic. Distributions of doctor types with higher values of dominate those with lower values in the sense of first order stochastic dominance. There is more mass c oncentrated at higher leve ls of treatment cost. It is more costly for the MCO to attract physicians. Consequently, the MCO reduces the health risk of its patient mix and insures fewer individuals. The effect of an increase in on reimbursement, the number of doctors, and the quality provided by the MCO is ambiguous. Decreasing reimbursements or keeping them constant may be profit maximizing if the number of doctors do es not decrease by much and the willingness to pay of policyholders doesnt decrease too much. If a decrease in reimbursement would lead to a large decrease in the number of doctors and in th e policyholders utility and willingness to pay, then the MCO may find it profitable to increase re imbursement, even if the net effect is still a reduction in the marginal hea lth risk insured by the MCO.
88 Proposition 3. Changes in the MC Os Cost of Providing Quality MCO B provides less of the quality increasing i nput, pays lower reimbursements, attracts fewer doctors, charges a lower insurance premium, and insures individuals with lower health risk when its marginal cost of providing quality (B ) increases. Proof: see Appendix B. The MCO decreases reimbursements and attrac ts fewer doctors because doctors are less efficient when they have less of the quality increasing input at their dis position, and the marginal utility derived by policyholders from access to doctors decreases. The demand for managed care shifts downwards, as individuals willingness to pay for manage d care decreases. The insurance premium decreases as a result, but the decrease does not fully compensate for the decrease in policyholders utility due to lower quality and fewer physicians. Thus, the MCO reduces the risk of its patients mix and insures fewer people in order to avoid providing the more costly quality. Of particular importance is the way in which the distribution of health risks in the population affects the insurance premium, reim bursement, quality provided, as well as the number and mix of patients served by the MCO. More people are at risk of becoming ill as a result of the ageing of the US population. In the model this corresponds to a shift in the distribution of health risks towards higher he alth risks, captured by a higher value of The distribution of health risks for higher statistically dominates those with lower in the sense of first order stochastic dominance. Comparative statics with respect to can illustrate the effects of population ageing, or allow for comparisons betw een markets with differe nt distributions of health risks.
89 Proposition 4. Changes in the Distribution of Health Risks When the health risk of the population increases the MCO provi des lower quality, pays lower reimbursements, attracts fewer physicians, charges a lower insurance premium, and finds it profitable to attract some individuals with higher health risk. Proof: see Appendix B. Faced with a higher number of illness cases the MCO tries to control costs by reducing quality, paying lower reimbursements, and provi ding access to fewer doctors. Thus, the model can explain the growing public dissatisfaction with the quality of health care provided by managed care organizations and their attempts to control costs through a variety of mechanisms such as mandatory referrals and pre-authorization procedures. The MCO insures some individuals with higher health risk because fe wer low-risk individuals remain. Because the policyholders willingness to pay decreases when quality and acce ss to physicians are reduced the MCO charges a lower insurance premium. Proposition 3 describes the role played by the indirect externality exerted by doc tors on policyholders, as well as other factors, in determining the magnitude of these changes. Proposition 5. Magnitude of Changes Due to Increased Population Health Risk and Marginal Cost of Quality The percentage decreases in MCO B s quality, reimbursement, and number of doctors due to increases in population health risk or in the ma rginal cost of quality provision are greater when people care more about quality (higher ), the externality exerte d by doctors on patients is greater (higher ), and the concentration of doctors with high treatment costs is higher (higher ).
90 The decreases due to increased population health risk are lower in population that had high health risk to start with (higher initial ).51 Proof: see Appendix B. Proposition 1 showed that the MCO provides hi gher quality and access to a wider network of physicians in markets in which individuals ca re more about quality and access to physicians. Proposition 5 shows that these are also the mark ets in which the MCO reduces quality and its network of physicians more to contain costs when the population health risk increases for reasons such as population ageing. Proposition 6. Changes in the Nu mber of Doctors in the Market The MCO pays higher reimbursements, attracts more doctors, provides more of the quality increasing input, insures more policyholders, and charges a highe r insurance premium when the number of doctors in the market Nd increases. Proof: see Appendix B. The MCO pays its doctors higher reimburseme nts when the number of doctors in the market increases because the same increase in reimbursements attracts more doctors in MCO B s network. This in turn increases the utility of policyholders, more of whom enroll in managed care. The insurance premium increases as a result of the increase in the policyholders willingness to pay. We have assumed that doctors have constant marginal cost of treating patients and they join the MCO if the reimbursement they receive covers their cost of treatment. Having more doctors in the network decreases the number of patients treated by each doctor. If the doctors 51 The externality exerted by the enrollment of an additional doctor in MCO Bs network, the marginal utility of policyholders with respect to the number of doctors, increases when increases if the number of doctors in the network is greater than one (Assumption 4-63 in the Appendix). Similarly, Assumption 4-64 in the Appendix insures that the marginal utility of quality increases when increases.
91 per patient cost of treatment depe nds on the number of patients they have to treat, this direct externality may negatively affect doctors by increasing their cost of treatment. Consequently, the change in reimbursement and number of doctors in MCO B s network resulting from an increase in the number of doctors in the market will depend on the relative magnitude of the direct and indirect externality. Proposition 7. Changes in the Maximum Cost of Treatment The MCO pays lower reimbursements, attracts fewer doctors, provides less of the quality increasing input, insures fewer policyholders, and charges a lower insurance premium when the cost of treatment incurred by doctors increases in the sense of an increase in the maximum cost of treatment C Proof: see Appendix B. While the decrease in the insurance premium charged by the MCO is somewhat surprising, it is due to the absence of income effects in the model, and to the fact that the alternative plan A is non-strategic and the MCO ta kes its quality, network of docto rs, and insurance premium as given. This limits the MCOs ability to increase its insurance premium. The MCO then resorts to quality reductions and reduced access to physicians to curtail costs. This framework is appropriate for analyzing the behavior of managed care organizations given the choices made by the Center for Medicare and Medicaid Service regarding traditional Medicare FFS. Future research should model the strategic interaction between health insurers to study also the alternative plans choice of quality, number of doctors, and insurance premium. Conclusion Recent developments in the economics of networ ks have shown the potential fallacies of using one-sided logic in two-sided markets. This paper has deve loped a two-sided market model
92 of managed care to analyze the pricing and quality decisions of a profit maximizing MCO in the presence of indirect network extern alities between doctors and patients. Managed care differs from other two-sided mark ets through the fact that the intermediarys expected cost of providing service to enrollees is linked to the expected benefit they derive from joining through the probability of getting ill. The MCO faces tradeoffs when choosing quality, premiums, and reimbursements, tradeoffs that depe nd on the distribution of patients health risk and physicians cost, the elasticity of supply of physicians wi th respect to reimbursements, the marginal cost of service quality, the utility derived by patients from the quality of health care services, and the magnitude of the indirect externality exerted by doctors on patients. In the case of iso-elastic distributions of patient health risk and physician cost of treatment, an increase in the cost of providing quality d ecreases the quality provided by the MCO, which reduces the effectiveness of phys icians, the marginal utility de rived by policyholders, and thus the reimbursement and the number of doctors in the preferred network. The demand for managed care services shifts downward, and the insurance premium also decreases. However, the decrease in premium is smaller than the decrease in pa tient utility due to lower quality and number of physicians in the network. Thus, the MCO reduces th e health risk of its patient mix, and avoids providing, at least in part the more costly quality. For populations more at risk of becoming ill, as is the case of the ageing population of the United States, the MCO provides services of lower quality, pays lower physician reimbursements, includes fewer provi ders in its preferred network, and yet finds it profitable to insure some higher risk individuals. Although MCOs provide services of higher qual ity and access to more doctors in markets where individuals care more about quality and acc ess to doctors, these markets are also the ones
93 in which the MCO reduces quality and access to p hysicians more aggressively to curtail costs when the cost of quality provision or population health risks increase. These results can explain the increasing public dissatisfaction with managed care quality and underscore the importance of implementi ng and monitoring the achievement of quality standards, as well as adjusting the capitation rates paid by government programs like Medicare and Medicaid to account for health risk. While most of the research on two-sided market s has so far been theoretical, there is a growing empirical two-sided ma rket literature (Rysman  Rysman ). Yet to my knowledge there is no empirical st udy analyzing the two-sided mark et aspects of managed care. Given the potential important implications of indirect externalities, empi rical studies would be useful to gauge the relative magnitude and the role played by indirect externalities in health care markets.
94 APPENDIX A EXPLANATORY AND INSTRUMENTAL VARIA BLES USED IN CHAPTER 2, THEIR SOUR CES, AND THE ESTIMATION METHOD Definition of Explanatory Variables and Data Sources REAL GDP PER CAPITA Real GDP per capita expressed in 1995 constant USD (WDI)52 MAIN LINES PER CAPITA Number of main lines per capita (ITU)53 WAITING LIST Ratio between registered applications for a main line and the number of main lines in operation (ITU) COMPETITION Dummy variable = 1 every y ear after the intr oduction of two or more digital mobile communi cations operators (GM and GW)54 SEQUENTIAL ENTRY Dummy variable = 1 every year after the introduction of competition, if a single digital mobile operator was present in the market from the beginning of digital communications adoption (GM and GW) SIMULTANEOUS ENTRY Dummy variable = 1 every year after the introduction of competition, if two or more digital mobile operators were present in the market from the beginning of digital communications adoption (GM and GW) ANALOG Dummy variable = 1 if onl y analog technology was introduced in the country (GM and GW) DIGITAL ONLY Dummy variable = 1 every year after the introduction of digital mobile communications if the country did not start adopting mobile communicati ons with analog technology (GM and GW) 52 WDI: World Bank, World Deve lopment Indicators Database 53 ITU: International Telecommunication Union, World Telecommunication Indicators Database 54GM and GW: Informa UK Ltd., Global Mobile; Crain Communications Inc., Global Wireless
95 Instrumental Variable Definition s and Rationale for Using Them TIME Number of years since 1989. In general, competition was more likely to be introduced during the latter years. This instrument also captures the trend in the number of fixed lines per capita and in the fraction of people waiting for a fixed line. POPULATION Country population (WDI). More populous countries were more likely to introduce competition earlier, and tended to introduce competition simultaneously. TVs PER CAPITA Number of television receivers per capita (ITU). This variable is a good instrument for the number of fixed lines per capita. Former socialist countries had re latively low numbers of television receivers per capita when communism fell. The number of television receivers increased considerably af terwards, much like the number of fixed telephone lines. TDSGNI Total debt service as a fraction of GNI (WDI). This variable can serve as an indicat or of how attractiv e a country is. For richer countries, total debt service represents a lower fraction of GNI, and richer countries tended to in troduce competition earlier, and simultaneously. But this variable may also capture the degree to which governments were cash constrained. Ca sh constrained governments had to promote economic reforms, especially privatization, and had greater incentives to award (cellular) licenses. GVTEXP Government expenditure as a fraction of GDP (WDI). A higher fraction of government spending in GDP is characteristic for countries in which reforms and the privatization process did not work well, and the public sector played an important role. These countries were generally less attractive to foreign investors. OPEN Ratio of exports + imports over GDP (WDI). There is evidence that foreign direct investment responded positively to the trade policy of former socialist countries (Deichman et. al ). STBKS Fraction of bank assets held by st ate-owned banks (Sherif et. al ).55 A higher fraction of state-owned ba nk assets generally means higher government control on economic activity. It is the variable that is most 55 Missing values of STBKS and other in strumental variables were replaced by lags, where available. This does not lead to biased estimates.
96 correlated with the waiting list for fixed lines. One of the reasons why telecommunications in former social ist countries were underdeveloped, compared to other utility industries, was the desire of the ruling party to control the entire society. Phone ta pping was common practice in these countries, and a higher number of fixed lines meant less government control. Countries with a lower frac tion of state-owned bank assets were more successful in introducing reforms, and in attracting foreign investors. DISTBRUX Distance from Bruxelles to the capital of each country (Rand McNally New Millenn ium World Atlas ). Countries closer to the European Union were more developed, more successful in introducing reforms, and more attractive markets. CALORIE Number of calories per capita per day (FAO).56 Captures the level of development of each country, as well as cultural factors. Instrumental Variables Estimation All instruments (and other exogenou s variables) have been used to estimate firs t stage logit models for the competition variables. The pseudo R square measures of goodness of fit are 0.63 for the introduction of competition, 0.47 for th e introduction of simultaneous competition, and 0.50 for the introduction of sequential competition.57 The fitted probabilities were then used in the second stage as instruments (together with initial instruments and the exogenous variables), in order to obtain correct estimates of the coefficients standard errors. The use of fitted probabilities in the second stage together with the initial instruments provides more efficient estimates, because both linear and nonlinear co mbinations of the initial instruments are exploited. 56 Food and Agriculture Organization 57 The first stage results are omitted from the paper as th ey would require too much space. They are available from the author upon request. All the first stage regressions, including those for the number of mainlines per capita and the length of the waiting list, were statistically significant The smallest F statistic was significant at 1% level of confidence.
97 APPENDIX B DERIVATION OF THE MAIN RESULTS IN CHAPTER 4 Iso-elastic Distributions of He alth Risk and Treatment Cost Derivation of the Profit Maximizing Solution To better understand the results, the model is solved assuming iso-elastic distributions of patient and physician types )( F andccG ) (, and constant marginal costs of providing qualityBBBqk )('. It is easiest to express the insurance prem ium charged by the MCO in terms of quality, reimbursement, and the cutoff probability from Equation 4-5 and solv e the model with respect to quality, reimbursement, and the cuto ff probability of illness. We have ) (BBAA ABnqnqPP (4-32) The MCO B s profit is 0)()( ) ( dfqkrnqnqPNBBB BBAA ApB. (4-33) Replacing 1)(f in Equation 4-33 and integrating yields 1 11 )( ) ( BBB BBAA ApBqkrnqnqPN (4-34) The first order conditions with respect to Bq, Br are 0)( )1)( (1 BBB BBAA Ap Bqkr nqnqPN. (4-35) 0 11 1 BB B p B Bnq r N r. (4-36) 0 1 )('1 1 1 B BBp B BqknqN q. (4-37) Simplifying, we get 0)()1() ( BBB BBAA Aqkr nqnqP. (4-38)
98 0 1 BB Bnq r. (4-39) 0 1 )('1 B BBqknq. (4-40) Combining Equations 4-39 and 4-40 we get )('BB B Bqk q r (4-41) Taking into account thatBBBqk)(', we can express quality in terms of reimbursement B B Br q (4-42) Replacing Equations 4-42 and 4-7 into E quation 4-39 and solving for reimbursement yields 1 11 d B BN C r. (4-43) Because d B BN C r n we can compute the number of doctors in the preferred network d d B BN N C C n 1 11 1 (4-44) Replacing the expression for the reimbursement from Equa tion 4-43 into Equation 4-42 yields the following expression for the quality provided by MCO B
99 1 11 d B B BN C q. (4-45) Substituting Equations 4-43, 4-44, and 4-45 into Equation 4-38 and solving for the marginal health risk of MCO Bs patients yields 1 11 )1()1( d B AA AN C nq P (4-46) Finally, substituting in Equation 4-32 yields the solution for the insurance premium 1 1 1 11 )1()1( 1 ]1)1)([( )1( 1 d B AA d B AA A BN C nq N C nq P P (4-47) Sufficient Conditions for Profit Maximization Note that the denominator in Equati ons 4-46 and 4-47 can be written as 0)1()1()1()1( BB AA B AAnq nq r nq D (4-48) where the second equality comes from Equation 4-39. The denominator is positive since MCO B offers lower quality and number of doctors than the alternative plan A and so are and PB as a result. We also need
100 1 11 1 1 B AA AnqP (4-49) to guarantee that 1 The second order derivatives of MCO B s profit function with respect to Bq, and Br are 2 2 2)( )1)( ()1( BBB BBAA Ap Bqkr rqnq PN (4-50) 1 2 2 2)1( BB B p B Bnq r N r (4-51) 1 1 2 21 )('' )1( B BB p B Bqknq N q (4-52) 0)1( 12 BB B p B Bnq r N r (4-53) 0)1( 1 )('1 2 B BBp B BqknqN q (4-54) 1 1 2 BB B p BB Bnq r N qr (4-55) Taking Equation 4-38 into account, Equation 4-50 can be written 02 2 2 Ap BPN (4-56) The Hessian matrix of the MCOs profit function is 1 2 1 1 1 1 1 2 21 )('')1( 0 )1( 0 0 0 B BB p BB B p BB B p BB B p Apqknq N nq r N nq r N nq r N PN H. (4-57)
101 The diagonal elements of the He ssian matrix are negative if 0)('' Bqk, 10 10 and10 (4-58) that is, if the cost of providing quality is convex, and the utility of enrollees exhibits diminishing returns to quality and reimbursement. Furthermor e, all the 22 determinant minors are positive under these assumptions if the utility of policyholders also exhibits decrea sing returns to scale, that is if 0 1 (4-59) The fact that the 2 determinant minors are po sitive is obvious with the exception of the one corresponding to the second derivatives with respect to quality a nd reimbursement, which can be written as 0)1( 1 )('' 1 1 )('')1( )1( )(2 222 2 222 1 2 1 1 1 1 1 2 2 BB B B BB B p B BB p BB B p BB B p BB B pnq r qk nq r N qknq N nq r N nq r N nq r N HDet (4-60) Under the conditions in Inequali ties 4-58 and 4-59 the Hessian matrix in Equation 4-57 is negative definite. These conditions together with the first order conditions in Equations 4-35, 436, and 4-37 are sufficient conditions for profit maximization. The following assumptions on the parameters are made in addition to the sufficient conditions for maximization C > 1, Nd > 1 (4-61) d BN C 1 11 (4-62) 1 1 11d BN C (4-63)
102 d BN C 11 (4-64) Assumption 4-62 is needed to guarantee that 1 0 C r C r GB B. Assumption 4-63 ensures that nB > 1 and the utility of policyholders is higher for higher values of This allows us to interpret an increase in as an increase in the external ity exerted by doctors on patients. Similarly, Assumption 4-64 guarantees that qB > 1 and policyholders deri ve a higher utility from quality when increases. Inequalities 4-62, 4-63, and 4-64 can be simultaneously satisfied if 0 1 and the marginal cost of the qual ity increasing input is not too large d BN 1 (4-65) Under all these assumptions, the impact of changes in parameters of interest on the MCOs reimbursement, quality, number of doctors, marg inal health risk, and insurance premium is summarized in Table 4-1. The following section sketches the proofs of the main propositions. The results were derived using Mathematica 5. For brevity, only the main results of the derivations are presented. A separa te file detailing the derivations is available from the author upon request. Proof of Proposition 1. Changes in the Marg inal Utility of Quality and the Externality Exerted by Doctors on Patients Differentiating Equations 4-43 th rough 4-47 with respect to it can be shown that 0 1 ln1 1 B B Bq r r (4-66) 0 1 ln1 1 1 B B B B Bq r r n n (4-67) 0 1 ln 1 1 B B Bq q q (4-68)
103 0 )1( ln ln 1 BB AA BBBAAAnq nq qnqqnq (4-69) 2 2)1()1()1( )1( ln)1(ln)(1B AA BB AAA B BABr nq nq nqq q rPP (4-70) One would expect the MCOs insurance prem ium to increase when policyholders care more about quality (higher ) because the MCOs quality and num ber of doctors increase and the marginal health risk decreases. The change in MCO Bs insurance premium is ambiguous because the utility of the alternat ive plan also increases when pa tients care more about quality. A small enough decrease in reimbursement coupled w ith small increases in quality and number of doctors can lead to a decrease in the marginal hea lth risk when the alternative plans quality and number of doctors are large enough. Differentiating Equations 4-43 th rough 4-48 with respect to to study the effects of an increase in the externality exer ted by doctors on patients yields 0 )1( ln1 1 B B Bn r r (4-71) 0 1 ln 1 B B Bn q q (4-72) 0 )1( ln1 1 1 B B B B Bn r r n n (4-73) 0 )1( ln ln 1 BB AA BBBAAAnq nq nnqnnq (4-74) 2 2 2 2)1()1()1( )1( )1()1()1( ln)1(ln)(B AA BB BA B AA AAA B BABr nq nq rP r nq nqn n rPP (4-75) The change in MCO Bs insurance premium when policyho lders care more about access to doctors (higher ) is ambiguous for reasons similar to those for which the change in MCO Bs
104 insurance premium is ambiguous when policyh olders care more about quality (higher ). The utility from joining the alternative plan A also increases. Proof of Proposition 2. Changes in the Distribution of Treatment Costs Differentiating Equations 4-43 th rough 4-47 with respect to yields )1( lnln 1 1 Cr r rB B B (4-76) 1 )lnln1)(1( 1 ln 1Cr r r C r n nB B B B B B (4-77) 1 lnln1 1Cr q qB B B (4-78) 0 )1()1( lnln 1 BB AA B Bnq nq rCr (4-79) 2 22)1()1()1( )1( lnln)(1B AA B AAB BABr nq r nqrC rPP (4-80) The sign of the derivatives de pends on the magnitude of (BrC lnln ), with the exception of the cutoff health risk. The MCO reduces th e marginal health risk it insures when the concentration of high treatm ent cost doctors increases. Proof of Proposition 3. Changes in the Cost of Providing Quality Differentiating Equation 4-45 with respect toB we see that the MCO provides less of the quality increasing input when the marginal cost of quality provision increases 0 )1( 11 B BB Bq q. (4-81) The reimbursement paid to physicians also decreases 0 )1( 1 B BB Br r. (4-82)
105 The reason is that the reductio n in the amount of the quality increasing input provided by the MCO also diminishes the marginal utility of physicians. Therefore, the number of physicians in the MCOs network also decreases 0 )1( 1 B BB Bn n. (4-83) With lower quality and access to fewer doctors the marginal willingness to pay of policyholders decreases and so does the insurance premium charged by the MCO 0 )1()1()1( )(2 2 B AA B AA BA B Br nq nq rP P (4-84) However, the decrease is such that the MCO insures a mix of patients with lower health risk, as can be seen by differentiating Equation 4-46 w ith respect toB and simplifying 0 )1( 12 BB AAB BB Bnq nq nq. (4-85) Finally, the MCO attracts a smaller number of individuals 0 )(1 B p B p B pN N FN (4-86) Proof of Proposition 4. Changes in the Distribution of Health Risks Differentiating Equations 4-43, 4-44 and 4-45 with respect to it can be shown that 0 )1( 1 )(1 1 1 B B B B B Bn n q q r r. (4-87) When the concentration of high-risk patients increases the MCO reduces quality and includes fewer physicians in its network by paying lower reimbursements to control costs. To see how the mix of patients insured by the MCO changes when the distribution of heath risks changes, differentiate Equation 4-46 with respect to Taking also the expression for reimbursement from Equati on 4-43 into account we get
106 2) 1 ( )1()1( )1(B AA BAA Ar nq rnq P (4-88) From Equation 4-39 we have BB Brq r)1(. (4-89) Substituting Equation 4-89 in Equation 4-88 yields 0 ) 1 ( )1(2 22 B AA BBAA Ar nq rqnqP (4-90) The derivative in Equation 4-90 is positive because the quality and number of doctors provided by MCO B is lower than that provided by the alternative plan A. The MCO finds it profitable to attract some higher risk individuals because fewer individuals with low risk remain. To determine how the insurance premium ch arged by the MCO changes, rewrite Equation 4-5 as ) (BBAA BAnqnqPP (4-91) Totally differentiating Equation 4-91 with respect to we get B BB BBAAPnq nqnq)( ) ( (4-92) Because 0 0 Bq and 0 Bn it must be that 0 BP Hence, the insurance premium also decreases. Proof of Proposition 5. Magnitude of Changes Due to Increased Population Health Risk and Marginal Cost of Quality To show that the percentage decreases in qua lity, reimbursement, and number of doctors in MCO B s network due to increases in population health risk are greater when the externality is greater, differentiate the absolute value of the decreases in Equation 4-87 with respect to
107 0 )1( 1 )(1 111 12 B B B B B Bn n q q r r (4-93) Similarly, one can show that the pe rcentage decreases are greater when individuals care more about the quality of health care service 0 )1( 1 )(1 1 111 12 B B B B B Bn n q q r r (4-94) the elasticity of the supply of doctors with respect to reimbursements is higher 0 )1( 1 )(1 1 12 B B B Bq q r r (4-95) 0 )1( 1 )(1 1 12 B Bn n (4-96) the concentration of high h ealth risks in the population is smaller to start with 0 )1( 1 )(1 1 111 1 B B B B B Bn n q q r r (4-97) To show that the percentage decreases in qua lity, reimbursement, and number of doctors in MCO B s network due to increases in the marginal cost of providing quality are greater when the externality is greater, differentiate the absolute value of the decreases in Equations 4-81, 4-82, and 4-83 with respect to 0 )(1 111 12 B BB B BB B BB Bn n q q r r. (4-98) Similarly, one can show that the percentage decreases are greater when individuals care more about the quality of health care service
108 0 )(1 1 111 12 B BB B BB B BB Bn n q q r r (4-99) the elasticity of the supply of doctors with respect to reimbursements is higher 0 )(1 1 12 B BB B BB Bq q r r (4-100) 0 )(1 )1( 12 B BB Bn n. (4-101) Proof of Proposition 6. Changes in the Number of Doctors in the Market Differentiating Equations 4-43 th rough 4-47 with respect to Nd yields 0 )1( 11 dBd B Bd BNqN q rN r (4-102) 0 )1( 11 dBd BNnN n (4-103) 0 )1( 1 BB AAd BB dnq nqN nq N (4-104) 0 )1()1()1( )(2 22 B AA d AA BA d Br nq N nq rP N P (4-105) Proof of proposition 7. Changes in the Maximum Cost of Treatment Differentiating Equations 4-43 th rough 4-47 with respect to C yields 0 )1( 11 CqC q rC rB B B B (4-106) 0 )1( )1(1 CnC nB B (4-107) 0 )1( 1 BB AA BBnq nqC nq C (4-108)
109 0 )1()1()1( )(2 222 B AA AA BA Br nq C nq rP C P (4-109) The MCO prefers to reduce quality and the numb er of doctors in its preferred network to restrain costs, even though this mean s it has to charge a lower premium.
110 Table 4-1. The effect of parameter incr eases on managed care reimbursement, quality, number of doctors, marginal he alth risk, and insurance premium Variables of Interest Parameters Reimbursement rB Quality qB Number of doctors nB Cutoff probability Insurance premium PB Concentration of health risks in the population + Importance of quality for policyholders + + + +/Indirect externality parameter (Importance of number of doctors for policyholders) + + + +/Elasticity of supply of doctors with respect to reimbursement +/+/+/+/Marginal cost of quality B Doctors maximum treatment cost per patient C Number of doctors in the market Nd + + + + +
111 LIST OF REFERENCES Ahn, H., Le e, M., 1999. An Econometric Anal ysis of the Demand for Access to Mobile Telephone Networks. Information Economics and Policy 11, 297-305. Altman, D., Cutler, D. M., and R. J. Zeckhauser, 2003. Enrollee Mix, Treatment Intensity, and Cost in Competing Indemnity and HMO Plan s. Journal of Health Economics 22, 23-45. Armstrong, M., 1998. Network Interconnection in Telecommunications. The Economic Journal 108, 545-564. Armstrong, M., 2006. Competition in Two-Sided Ma rkets. Rand Journal of Economics 37(3), 668-691. Armstrong, M. and J. Vickers, 1996. Regulatory Reform in Teleco mmunications in Central and Eastern Europe. Economics of Transition 4, 295-314. Arrow, K. J., 1963. The Welfare Economics of Medical Care. American Economic Review 53, 941-973. Arrow, K. J., 1963. The Economics of Moral H azard: Further Comment. American Economic Review 58, 537-539. Bardey, D. and Rochet, J-C., 2006. Competiti on Among Health Plans: a Two-Sided Market Approach. Universita del Rosario, unpublis hed manuscript. Electronic document at http://economia.uniandes.edu.co/es/c ontent/download/9220/45219/file/Bardey.pdf Cabral, L. M. B., Kretschmer, T., 2004. Standards Battle and Public Policy. London School of Economics, unpublished manuscript. Electronic document at http://www.chicagofed.org/news_and_conferen ces/conferences_and_events/files/2004_sta ndards_cabral.pdf Cava-Ferreruela, I., AlabauMu oz, A., 2006. Broadband Policy Assessm ent: a Cross-National Empirical Analysis. Telecommunications Policy 30, 445-463. Clements, M., Abramowitz, A., 2006. The Development and Adoption of Broadband Service: a Household Level Analysis. Paper presen ted at the 34th Research Conference on Communication, Information and Inte rnet Policy. Arlington, Virginia. Crain Communications Inc., 2001. Cellular and PC S Carriers of the World, Part 5. Global Wireless 4(4). David, P., and Greenstein, S., 1990. The Economics of Compatibility Standards: an Introduction of Recent Research. Economics of Innovation and New Technology 1, 3-41. Deb, P., and P. K. Trivedi, 2006. Specificati on and Simulated Likelihood Estimation of a NonNormal Treatment-Outcome Model with Selection: Application to Health Care Utilization. Econometrics Journal 9, 307-331.
112 Deichman, J., A. Eshghi, D. Haughton, S. Sayek, and N. Teebagy, 2003. Foreign Direct Investment in the Eurasian Transition St ates. Eastern European Economics 41, 5-34. Dekimpe, M., P. Parker, and M. Sarvary, 1998. Staged Estimation of International Diffusion Models: An Application to Global Cellu lar Telephone Adoption. Technological Forecasting and Social Change 57, 105-132. Distaso, W., Lupi, P., Maneti, F.M., 2006. Plat form Competition and Broadband Uptake: Theory and Empirical Evidence from the European Union. Information Economics and Policy 18, 87-106. Ellis, R. P. and T. G. McGuire, 1993. Supply-Si de and Demand-Side Cost Sharing in Health Care. Journal of Economic Perspectives 7, 135-152. Evans, D.S. and R. Schmalensee, 2005. The Industrial Organization of Markets with Two-Sided Platforms. NBER Working Paper No. 11603. Feldmann, V., 2003. Mobile Overtakes Fixed: Im plications for Policy and Regulation. ITU, Geneva. Electronic document at http://www.itu.int/osg/spu/ni/mobileoverta kes/Resources/Mobileovertakes_Paper.pdf Fernandez-Cornejo, J, McBride, W. D., 2002. A doption of Bioengineered Crops. US Departm ent of Agriculture Agricultura l Economics Report, 810. Food and Agriculture Organizati on, 2003. FAOSTAT Statistical Da tabase. Available online at http://faostat.fao.org/ Ford, G. S., Koutsky, T. M., Spiwak, L. J., 2007. The Broadband Performance Index : a PolicyRelevant Method of Comparing Broadband A doption Among Countries Phoenix Center Policy Paper, 29. Gandal, N., 2002. Compatibility, Standardiza tion, and Network Effects: Some Policy Implications. Oxford Review of Economic Policy 18, 80-91. Gandal, N., Salant, D., Waverman, L., 2003. Standards in Wireless Telephone Networks. Telecommunications Policy 27, 325. Garcia-Murillo, M., 2005. International Broa dband Deployment: the Impact of Unbundling. Communications and Strategies 57, 83-108. Glied, S., 2000. Managed Care. Handbook of Health Economics, Ed. by A. J. Culyer and J. P. Newhouse, Vol. 1A. Grosso, M., 2006. Determinants of Broadband Penetr ation in OECD Nations. Paper presented to the Australian Communications Policy and Research Forum. Gruber, H., 2001. Competition and Innovation. Th e Diffusion of Mobile Telecommunications in Central and Eastern Europe. Informa tion Economics and Policy 13, 19-34.
113 Gruber, H. and F. Verboven, 2001a. The Diffusion of Mobile Telecommunications Services in the European Union. European Economic Review 25, 577-588. Gruber, H. and F. Verboven, 2001b. The Evol ution of Markets under Entry and Standards Regulation The Case of Global Mobile Te lecommunications. Intern ational Journal of Industrial Organization 19, 1189-1212. Henisz, W., 2000. The Institutional Environmen t for Economic Growth. Economics and Politics 12(1), 1-31. Informa UK Ltd., 2003. Eastern Europe: Cellu lar Telephone Subscribers by Network. Global Mobile 10(5), 7-14. International Telecommunication Union, 2002. World Telecommunication Indicators Database, 5th edition. ITU, Geneva. International Telecommunication Union, 2001. 3G Mobile Licensing Policy: from GSM to IMT2000a Comparative Analysis. ITU Workshop on Licensing 3G Mobile, September 19-21. ITU, Geneva. International Telecommunication Union, 2003. ITU Internet Report 2003: Birth of Broadband. ITU, Geneva. Electronic document at http://www.itu.int/osg/spu/publicatio ns/sales/birthofbroadband/index.htm l International Telecommunication Union, 2004. ITU Internet Report 2004: The Portable Internet. ITU, Geneva. Electronic document at http://www.itu.int/osg/spu/publ ications/portableinternet/ International Telecomm unication Union, 2005. ITU Internet Report 2005 : The Internet of Things. ITU, Geneva. Electronic document at http://www.itu.int/osg/spu/publ ications/internetofthings/ International Telecomm unication Union, 2006. ITU Internet Report 2006: Digital Life. ITU, Geneva. Electronic document at http://www.itu.int/osg/spu/publications /digitalife/ Ki m, J. H., Bauer, J.M., Wildman, S.S., 2003. Broadband Uptake in OECD Countries: Policy Lessons from Comparative Statistical Analys is. Paper presented at the 31st Research Conference on Communication, Information a nd Internet policy. Arlington, Virginia. Kocenda, E., 2001. Macroeconomic Convergence in Transition Countries. Journal of Comparative Economics 29(1), 1-23. Koski, H., Kretschmer, T., 2002. Entry, Standards and Competition: Firm Strategies and the Diffusion of Mobile Telephony. ETLA Discussion Papers, 827. Kutan, A.M. and T.M. Yigit, 2004. Nominal and Real Stochastic Convergence of Transition Economies. Journal of Compar ative Economics 32(1), 23-36.
114 Laffont, J., P. Rey, and J. Tirole, 1998a Network Competition. I. Overview and Nondiscriminatory Pricing. RAND Journal of Economics 29 (1), 1-38. Laffont, J., P. Rey, and J. Tirole, 1998b. Netw ork Competition. II. Price Discrimination. RAND Journal of Economics 29 (1), 38-57. Lee, S., 2006a. The Impacts of Market-Based St andardization Policy on Mobile Deployment in OECD Countries: an Empirical Analysis. Paper presented at the Association for Education in Journalism and Mass Communicati on Annual Convention. San Francisc o, California. Lee, S., 2006b. Broadband Deployment in the United States: Examining the Impacts of Platform Competition. The International Journal on Media Management 8, 173-181. Lee, S., Chan-Olmsted, S.M, Kim, H. 2007. The Deployment of Third-Generation Mobile Services: A Multinational Analysis of Contri buting Factors. Paper presented to the 2007 Association for Education in Journalism and Mass Comm unication Annual Convention. Washington, D.C. Liikanen, J., Stoneman, P., Toivanen, O., 2004. Intergenerational Effects in the Diffusion of New Technology: the Case of Mobile Phones International Journal of Industrial Organization 22, 1137-1154. Liu, F. and D. M. Zimmer, 2006. The Effect of Switching Private Insu rance Plans on Health Care Utilization. Contributions to Economic Analysis and Policy Vol. 5(1), Article 22. Ma, C. A. and T. G. McGuire, 1997. Optimal He alth Insurance and Provider Payment. American Economic Review 87, 685-704. Ma, C. A. and M. H. Riordan, 2002. Health Insurance, Moral Hazard, and Managed Care. Journal of Economics and Management Strategy 11(1), 81-107. Madden, G., Coble-Neal, G., Dalzell, B., 2004. A Dynamic Model of Mobile Telephony Subscription Incorporating a Network Eff ect. Telecommunications Policy 28, 133-144. Maldoom, D., Marsden, R., Sidak, J. G. and H. J. Singer, 2003. Competition in Broadband Provision and its Implications for Regulatory Policy. Report for the Brussels Round Table. Brussels, Belgium. Electronic document at http://papers.ssrn.com /sol3/papers.cfm?abstract_id=463041 Marcu, I. M., 2004. Competition and Mobile Communications in Former Socialist Countries. Paper presented at the 32nd Research C onference on Communication, Information and Internet Policy. Arlington, Virginia. Miller, R. H. and L. S. Luft, 1994. Managed Care Plan Performance Since 1980. Journal of the American Medical Association 271, 1512-1519. Organization for Economic Cooperation and Development, 2001. The Development of Broadband Access in OECD Countries. OECD, Paris.
115 Organization for Economic Cooperation and Deve lopment, 2003. Development in Local Loop Unbundling. OECD, Paris. Organization for Economic Cooperation and Development, 2005. Communications Outlook 2005. OECD, Paris. Organization for Economic Cooperation and Deve lopment, 2007. OECD Broadband Statistics. OECD, Paris. Electronic document at www.oecd.org/sti/ict/broadband. Parker, P. and L. Rller, 1997. Collusive Conduc t in Duopolies: Multimar ket Contact and CrossOwnership in the Mobile Telephone Industry. RAND Journal of Economics 28, 304-322. Pezzino, M. and G. Pignatarro, 2007. Competition in the Health Care market: a Two-Sided Approach. Annals of the 19th Conference of the Italian Society for Public Economics (SIEP). Electronic document at http://www-1.unipv.it/websiep/wp/200771.pdf Poot, J., 2000. A Synthesis of Em pirical Research on the Impact of Government on Long-Run Growth. Growth and Change 31, 516-546. Rand McNally & Company, 2001. New Millennium World Atlas Deluxe CD-ROM. Chicago, Illinois. Ridder, J., 2007. Catching-up in Broadband: What Will it Take? OECD, Paris. Rochet, J-C. and J. Tirole, 2003. Platform Comp etition in Two-Sided Markets. IDEI Working Papers 152, Institut dcono mie Industrielle, Toulouse. Rochet, J-C. and J. Tirole, 2005. Two-Sided Mark ets: A Progress Report. IDEI Working Papers 275, Institut dconomie Industrielle, Toulouse. Electronic document at http://idei.fr/doc/wp/2005/2sided_markets.pdf Rouvinen, P., 2006. Diffusion of Digital Mobile Telephony: Are Developing Countries Different? Telecommunications Policy 30, 46-63. Rysman, M., 2004. Competition between Networks: A Study of the Market for Yellow Pages. Review of Economic Studies 71(2), 483-512. Rysman, M., 2007. The Empirics of Antitrust in Two-Sided Markets. Competition Policy International 3(1), 197-209. Shelanski, H., 2003. Competition Policy for 3G Wireless Services. In Noam E.M., Steinbock D. (Eds.), Competition for the Mobile Internet. Kluwer Academic Publishers, Norwell, MA, pp. 39-54. Sherif, K., M. Borish, and A. Gross, 2002. State-Owned Banks in Transition: Origin, Evolution, and Policy Responses. World Bank Publication, 022803.
116 United Nations Development Programme 2004. Human Development Report 2005 UNDP, New York. United Nations Development Programme 2005. Human Development Report 2006 UNDP, New York. Wolfram Research, Inc., 2003. Mathematica 5. Champaign, IL. World Bank, 2003. World Development Indicators Online. Wright, J., 2004. One-Sided Logic in Two-Sided Markets. Review of Network Economics 3(1), 44-64. Xavier, P., 2001. Licensing of Third Generation (3G) Mobile: Briefing Paper. ITU Workshop on Licensing 3G Mobile. ITU, Geneva. Zeckhauser, R., 1970. Medical Insurance: A Case Study of the Tradeoff between Risk Spreading and Appropriate Incentives. Journa l of Economic Theory 2, 10-26.
117 BIOGRAPHICAL SKETCH Mircea Ioan Marcu graduated in 1998 with a Bachelor of Science degree in econom ics from Babe-Bolyai University in Cluj-Napoca Romania, majoring in finance and banking. He went on to get a Master of Arts degree in quan titative economics from the same university the next year. In fall 1999 Mircea moved to Budapest, Hungary to study Economics at Central European University. After graduating with a Master of Arts degree in 2001, he moved to the United States to pursue doctoral studies in Economic s at the University of Florida. At the University of Florida, Mircea specialized in industrial organization, microeconomics, and econometrics. He was empl oyed by the Department of Economics both as a teaching and a research assistant. He was an instructor of managerial economics in the summers of 2004, 2005, and 2007. He also taught economics of transition and European Union accession, an upper-level elective, in Fall 2005 and Spring 2006. After successful completion of all the requirements for the degree, Mircea ea rned the Doctor of Philosophy degree in December 2008. He serves as a researcher at the University of Florida Institute fo r Child Health Policy and as a consultant to ANACOM, the Portugue se National Communications Authority.