Commitment and Regulatory Independence in Practice in Latin American and Caribbean Countries

Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 27 CoMMItMent AnD ReGULAtoRy InDePenDenCe In PRACtICe In LAtIn AMeRICAn AnD CARIBBeAn CoUntRIes Francesc trillas* and Miguel a. Montoya**1 Abstract We present an analysis of the evolution of regulatory independence in practice for 23 Latin American and Caribbean countries in the telecommunications industry. Based on this analysis, we construct indices of regulatory independence, which improve upon the measures that have been used so far in the empirical regulation literature. Our measures are consistent with the fact that legal independence does not solve, but it relocates, the commitment problem of utility regulation. We show that legal indices may give a partially distorted picture of the commitment ability of institutions. In addition, treating independence as exogenous may underestimate its impact. The combination of de facto and de jure independence has a positive (probably modest nonetheless) impact on network penetration in telecommunications markets. keywords: independence; regulation; strategic delegation; telecommunications JeL Classification numbers: l51 1. INtROdUCtION Institutions matter in any field of economic policy. For example, the econometric treatment of the effects of good institutions on macroeconomic performance is a very active field of research, and is increasingly inspiring empirical research in microeconomic policies. Credible institutional commitment is seen as one of the few * UaB ([email protected]). ** ItesM ([email protected]). we thank Carlos Cardoso for excellent research assistance. we also thank comments received at IsNIe 2009 conference in Berkeley, especially from stefan Voigt, and feedback from Jon stern over the years. Francesc trillas and Miguel a. Montoya 28 Intersentia recommendations that survive after many studies, although the recommendation leaves broad space to fill in the details. levy and spiller (1996) stress this point for the case of the regulation of privatized utilities. The analysis of the independence of regulatory agencies is part of this increasing interest in the institutions of economic policy. as initially suggested by Rogoff (1985) in the context of monetary policy, independence is seen as a mechanism to strategically delegate into an agent (the central banker, the regulator) who is more reluctant than a representative government to engage into opportunistic behavior. Mishkin (2006) argues however that “although dealing with time-inconsistency problems by appointing a conservative policymaker has attractive theoretical properties, it is not so easy to implement in practice.” Mishkin points out difficulties with this approach to solving the time-inconsistency problem, using the example of monetary policy. It may be hard to find a central banker with the “right” preferences and it is hard to believe that politicians would naturally want to appoint central bankers with different preferences than theirs, so that a regime based on having a conservative central banker is unlikely to be stable over time. similar objections apply to the appointment of a conservative regulator. as a counter-argument, it has been suggested that “as-if Rogoff delegation” (constraining in legislation the objectives of the agent) may alleviate this problem.1 an independent regulatory agency is also seen as a mechanism to attract professional experts2 and to stabilize policies in the presence of political volatility (see evans et al., 2008). The problem of course is that independence does not solve, but it relocates, the commitment problem, which transforms itself into one of the government credibly committing not to undermine the independence of the regulator, which many jurisdictions have found very difficult. work on the measurement of telecommunications Regulatory agencies (tRa) independence has mostly analyzed legal or de jure independence but not independence in practice or de facto independence. Most, but not all, of the work on legal independence has used dichotomous dummy variables, and these, as estache et al. (2006: 12) point out, “may not capture the degree of independence”. Others use indices reporting about the legal framework: whether there is primary legislation requiring an independent regulator, how is it funded, to whom should he or she report, etc.3 1 see levine et al. (2005). 2 Bernstein (1955: 4): “In general, the Commission form has been championed by those who believe that administrative regulation requires a high degree of expertness, a mastery of technical detail, and continuity and stability of policy. These requirements, it is alleged, can only be met by a board of commissioners functioning in a neutral environment, free from partisan political considerations.” 3 stern and Cubbin (2003), Pargal (2003), edwards and wavermann (2006), Gual and trillas (2004 and 2006) and Montoya and trillas (2007) are representative of this line of research and acknowledge the need for indices of independence in practice. see also Gutierrez (2003a), Ros (1999 and 2003), Viani (2006), wallsten (2003), ai et al. (2004) and Fink et al. (2002). Commitment and Regulatory Independence in Practice in latin american and Caribbean Countries Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 29 we measure below independence in practice (for telecommunications4 regulators in latin america and the Caribbean), by summarizing in a few statistics a number of country case studies on the turnover and political vulnerability of regulators. The empirical methodology to construct indices draws from the literature on Central Bank Independence. we use these indicators to analyze the impact of independence on fixed line penetration in telecommunications, taking into account its potential endogeneity, which is aggravated by the fact that legal independence is a politically vulnerable institution. For our sample years (1990–2004) fixed lines were the main mode of telephony in our region of interest, before the boom in mobiles. Focusing on the fixed lines monopolies, we can concentrate on commitment abstracting from competition issues, which would be crucial in mobile telephony. In future research we will expand on the relationship between independence and mobile diffusion. In the rest of this paper, in section 2 we present some background on the independence debate. In section 3 we present the case studies on 23 countries and build on these to construct indices of independence in practice. In section 4 we put our measures to work, analyzing the determinants of penetration in fixed telephony, including an analysis of potential endogeneity problems. and finally we conclude in section 5. 2. tHeORetICal FRaMewORK aNd Related lIteRatURe In infrastructure industries, the importance of institutions is mainly driven by the sunk nature of the investments needed, which is the source of a time inconsistency problem, highlighted by levy and spiller (1996), shirley et al. (2000), Noll (2000), Noll and shirley (2002), Gutiérrez (2003a), levine et al. (2005) and Newbery (2000), among others. Many countries face major difficulties in inducing sufficient investment to meet demand at an acceptable cost. The reason is that given the long-lived nature of sunk assets, unless firms have expectations that prices will be sufficiently high over time, they will be reluctant to invest. 4 Compared with electricity or water, the telecommunications sector is less prone to regulatory opportunism, given that assets are not as long lived. Then if we find that regulatory opportunism is an issue in telecommunications, we would predict that a similar and more serious phenomenon arises in these other sectors. Underinvestment in the face of regulatory time inconsistency problems, however, has resurfaced in telecommunications as countries debate how to create a framework to promote next generation fiber networks, in spite of the increasing role played by competition and technological change. Francesc trillas and Miguel a. Montoya 30 Intersentia The following picture of an extensive form game summarizes with the simplest of models the time inconsistency problem in regulation that gives rise to the “independence” solution: F R I = 1 I = 0 P = 1+ε P = 0 (ε,1–ε) (–1,2) (0,0) In the game described in the picture, first a firm (F) makes a decision on whether to undertake a specific investment (for example, a fiber optic network) or not, and next the regulator (R), if the firm has invested, decides whether to fix a price that remunerates the investment, or to expropriate this investment (zero price). The payoff of the firm is P-I, whereas the payoff of the (consumer welfare maximizing) regulator is 2I–P.5 By backwards induction, if the firm has invested (I = 1), the regulator will rationally fix P=0, and, anticipating this, the firm will not invest (I = 0). Hence, in a sub-game perfect equilibrium, there is no investment. More realistic settings would include many other real world details, but if there is no commitment and assets are sunk, under-investment would remain a serious concern, perhaps in the form of bad maintenance or use of inefficient technologies. The expropriation of the quasi-rents derived from specific investments may not necessarily take the form of too low prices, but it can take other forms, such as unexpected investment requirements, costly unanticipated quality improvements, or requirements to hire inefficient staff. Policy makers may follow this path and still benefit from the (already in place) investments. ex ante, however, investors will anticipate this, and investment levels will be sub-optimal. The opportunity cost of reneging will depend on country characteristics, such as the institutional endowment, the degree of income inequality, or the nature of fiscal systems. In countries with skewed income distributions, governments pay a political price in terms of not satisfying the median (relatively poor) voter if they do not renege on promises made to remunerate specific investments. The problem may be alleviated by long term contracts, repeated interactions, reputational mechanisms or institutions that make credible that the P=0 path will not be taken. Historically, public ownership (the state internalizing the firm’s problem) has been a way to alleviate time inconsistency, but in the recent decades policy makers in many countries recognized that the costs of public ownership in terms of public funds and 5 any objective function of the regulator equal to γI–P with γ>1 would yield the same results. we use γ=2 for concreteness. Commitment and Regulatory Independence in Practice in latin american and Caribbean Countries Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 31 inefficient practices outweighed its benefits. Thus, the solution of privatizing and strategically delegating into an independent regulator who cares to a certain extent about the firm’s rents, in a similar way that governments delegate into an inflationaverse central banker (see levine et al., 2005). The need for experts in technologically complex sectors and the wish to give certainty in politically volatile regions reinforce the argument of independence. However, it is not axiomatic that independence will automatically yield relatively better investment results than the existing alternatives. There are certainly other mechanisms to achieve commitment, to attract experts, and to avoid political volatility. For example, it is widely recognized that Chile has achieved a high degree of commitment in privatizing utilities through a very detailed and difficult to change legislation. It is an open empirical question whether Chile is an exception that can be explained by its unique history and a very specific political and institutional system that makes policy reversal very difficult, or whether it is an example that can be generalized. One problem of course is that independence does not solve, but it relocates, the commitment problem, which transforms itself into one of the government credibly committing not to undermine the independence of the regulator, which many countries have found very difficult, as we show below in section 3. Then a potential, albeit partial and imperfect, measure of independence in practice is for how long do politicians respect the period in office of appointed regulators. and this is the research strategy we follow below. an independent (or at least separate from government) regulator is generally associated with attempts to reform regulation. wallsten (2001, 8) argues that having a separate regulator is a sign of how willing a country is to reform regulation. There are at least ten studies about telecommunications regulation which measure legal (de jure) independence of the regulatory agency; five of them use dichotomous variables and the others use indices, but none of them measures independence in practice. In most (but not all) cases the measures that have been used so far have a positive impact on some performance measures. although in theory other mechanisms could alleviate the time inconsistency problem faced by regulators, the literature points out that it is difficult,6 at least for developing countries, to find credible (and consistent over time) alternatives to an independent regulatory agency. an important issue is then what do we mean exactly by independence. edwards and waverman (2005: 25) argue that independence is more than a group of formal institutional rules; it also has important informal aspects which usually depend on centuries of legal and political traditions, cultural norms and individuals. 6 see Gutierrez (2003a) and levine et al. (2005). Francesc trillas and Miguel a. Montoya 32 Intersentia we therefore think that more emphasis should be put on de facto regulatory independence. edwards and waverman (2005) suggest that there seems to be a negative correlation between formal (de jure) regulatory independence and independence in practice (de facto), because in those countries with weak informal mechanisms to ensure regulatory independence, these are compensated with strong formal arrangements in order to persuade potential investors that there is no regulatory bias.7 The drawback of legal indices or dichotomous variables which measure regulatory independence is that they only reflect the state of legislation, and events and politicians may leave the law aside.8 an implication of this is that inferences derived from such data sets for telecommunications, electricity and other industries can potentially give distorted pictures about the real effect of regulatory governance, as stern and Cubbin (2003: 22) point out. Thus there is the need to enlarge the data bases to develop empirical work which tests the effects of a well functioning regulatory regime using data about the process or practice of regulation, for example, as stated by levine, et al. (2005: 469), by measuring “the percentage of commissioners or directors of regulatory agencies who end their term prematurely.” Pargal (2003) also points out, when explaining the limitations of his research, that lack of data made it impossible to assess the importance of aspects of independence such as security and duration of the contract of the regulator.9 In contrast, the measurement of independence in practice for Central Banks is quite developed in the literature.10 according to eijffinger and de Haan (1996) a Central Bank is independent if its monetary policy is not influenced by political cycles or by the preferences of politicians. Cukierman (1992: 383) clarifies that there is no obvious measure of actual (in practice) independence as opposed to legal Central Bank independence. He points out that this is not because it is not important, but “because it is difficult to find a group of systematic measures of actual independence when this diverges from legal independence”. as a proxy for independence in practice he uses the average time (or turnover ratio) in the position of Central Bank governor or chairman. He finds that the measure of legal independence (a 16-variable index) and the turnover rate of the 7 However, they do not measure empirically the practice of independence in their work. 8 an illustrative example of the problems of appointment, continuity and independence of commissioners in a regulatory agency is the regulation of electricity in India, where laws establish that commissioners’ appointments must be for 5 years with a compulsory retirement age of 62. Most commissioners are appointed around 60 years old, so that they can only stay in their jobs for 2 or 3 years, according to stern and Cubbin (2003: 18). 9 see also wallsten (2003) or estache et al. (2006). 10 see eijffinger and de Haan (1996), de Haan and Kooi (2000) and arnone, laurens and segalotto (2006). Commitment and Regulatory Independence in Practice in latin american and Caribbean Countries Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 33 Central Bank governor differ by a larger amount in developing rather than in developed countries. Cukierman and webb (1995) calculate yet another variable of independence in practice which they call (political) vulnerability index. This reflects the political influence in the central bank by measuring the probability that the Central Bank governor will be replaced immediately after a political change in government. They use a sample of 67 countries for the 1950 to 1989 period. They find that in a six months period the governor changes in one quarter of the cases after a non-radical (i.e., not related to a military rebellion, restoration of democracy or constitutional change) political change. There exists high variation among countries, political vulnerability being more than three times larger in developing economies than in developed ones. They also find that their measure is positively correlated with the level and variance of inflation, with real growth and with real interest rates. The authors show that the frequency of governor changes is higher the closer the date of the political transition. They stress that the political turnover measure by itself is an imperfect measure of independence in practice. a low ratio does not necessarily mean high independence (in some cases such as denmark, United Kingdom or Iceland it is; but not in others such as countries with stable authoritarian regimes). Instead, they argue that a high ratio does reflect low Central Bank independence, because high turnover ratios mean that the governor’s period is shorter than the executive’s and this makes the governor more vulnerable to the influence of the President or the political majorities. and hence will be less prone to try to implement long run policies. de Haan and Kooi (2000) analyze Central Bank independence in practice for 82 developing countries. They conclude that (lack of) independence is only correlated with inflation if we take into account countries with large price increases. They report that independence works better in developed rather than in developing countries. 3. tHe CONstRUCtION OF MeasURes OF INdePeNdeNt ReGUlatION IN PRaCtICe In this section, we first explain the data collection process about the beginning and end of the period in office of the head of tRa (telecommunications Regulatory authority). Next we explain the construction of a vulnerability index and a turnover rate which use the changes in the head of the agency as a measure of independence in practice. Next we explain the construction of an index of independence in practice that reports about and weighs for the reason why the head of agency leaves his or her position. From the combination of these two indices of independence in practice with an index of legal independence we create two indices that combine Francesc trillas and Miguel a. Montoya 34 Intersentia legal issues and the practice of regulation for 23 latin american and Caribbean countries. 3.1. data COlleCtION to create useful measures of independence in practice for latin american tRa’s it is key to have information about the exact date of beginning and end of the period in office of the head of the agency, as well as the reason for which the head of the agency leaves his or her position (end of term in office, dismissal or other).11 we obtained complete data on the beginning and end of the director or chair person of the agency for 23 countries in the latin america and Caribbean region for the period between 1990 and 2004.12 to obtain the necessary data we resorted mainly to three groups of sources: i) The tRa’s themselves. In some cases, in their web pages there are press bulletins, decrees, contracts, speeches or notes about the beginning and/or end of the time in office of the head.13 ii) data bases of news about the region, such as ISI Emerging Markets, ProQuest and Lexis-Nexis. we also used the press of every country as well as contact with some analysts or local telecommunications research centers.14 iii) Bulletins and web pages of Asociación Hispanoamericana de Centros de Investigación y Empresas de Telecomunicaciones (aHCIet), International telecomunication Union (ItU) and Foro Latinoamericano de Entes Reguladores de Telecomunicaciones (ReGUlatel).15 11 we think of the head of the agency as the most important member of the institution. we are aware of examples, such as Colombia, where government of the agency is in the hands of a council where each member has the same weight and the presidency rotates among the members (16 months each). 12 For Haiti and Guyana it was impossible to obtain complete data for the period. we also exclude Cuba and Puerto Rico for their specific political or market characteristics. 13 For an analysis comparing the web sites of tRa in latin america and the Caribbean, see Mahan (2005). 14 In the case of Mexico, for example, we contacted the Instituto del Derecho de las Telecomunicaciones (Idet), the Programa de Investigación de las Telecomunicaciones from Centro de Investigación y Docencia Económica (CIde) and a specialised editorialist in newspaper Reforma. 15 Their web pages are: www.ahciet.net (aHCIet), www.itu.int (ItU), www.regulatel.org (ReGUlatel). Commitment and Regulatory Independence in Practice in latin american and Caribbean Countries Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 35 table 1. telecom regulation agency and their presidents, 1990–2004 Country name of Agency year of creation President/ Director name start end Months argentina Comisión Nacional de telecomunicaciones 1990 Ceferino Namuncura Jun-04 — — Falvio M. Madaro Jun-03 Jun-04 13 adolfo luis Italiano Mar-02 Jun-03 15 Carlos Forno dec-99 Mar-02 27 Roberto Catalan Feb-97 dec-99 35 alberto Gabrielli Mar-96 Feb-97 11 Raúl agüero May-95 Mar-96 10 Oscar Gonzalez Jun-94 May-95 11 Rinaldo Colomé Oct-93 Jun-94 8 José l. Palazzo Nov-91 Oct-93 23 Raúl Otero Jan-90 Nov-91a22 Barbados Fair trading Commission 2001bNeville Nicholls dec-03 — — Frank King Jan-01 dec-03 35 Belice Public Utilities Commission 1999 Gilbert Canton Jan-99 — — Bolivia superintendencia de telecomunicaciones 1995 René Bustillo P. Jan-03 — — Guido loayza M. dec-97 Jan-03 61 Carlos saravia d. Nov-95 dec-97 25 Brazil agência Nacional de telecomunicações 1997 Pedro J. ziller de a. Jan-04 — — luiz G. schymura de O. May-02 Jan-04 20 antônio C. Valente da s. Mar-02 May-02 1 Renato Navarro G. Nov-97 Mar-02 53 Chile subsecretaria de telecomunicaciones 1977 Chistian Nicolai O. Mar-00 — — Juanita Gana aug-97 Mar-00 31 Gregorio san Martín Mar-94 aug-97 41 Roberto Pliscoff Mar-90cMar-94d48 Francesc trillas and Miguel a. Montoya 42 Intersentia where the period in office is longer than the region’s average, especially in countries with high stability in their cabinets and presidential periods.23 3.3. tURNOVeR Rate The turnover rate measures the average ratio of the actual duration relative to the legally prescribed duration, in months, of the time in office of the director or chair of the agency. It is a simple and interesting figure. In many cases, the regulator does not complete the prescribed period in office.24 The average duration of a regulator in latin american and Caribbean tRa’s in the period 1990 to 2004 is approximately two years and a half, being those of lowest duration those in Colombia, Nicaragua and argentina and those of highest duration those in Belize, Peru and Jamaica. Notice that the most stable countries are former British colonies and Chile. The turnover rate does not distinguish amongst the reasons behind the change of the regulator. It only captures the effect of any type of changes, from the end of the legal period to resignations or removals for different reasons which diminish the effective time in the direction of tRa’s relative to what is prescribed by law.25 Note that the prescribed time in office can usually be evaded without violating the law, but clearly affecting the degree of independence.26 If we focus on the 17 countries which have a period established by law, only in four of them the regulator completed the full period: Belize, Jamaica, Uruguay and Peru (which actually goes beyond the prescribed period), and the remainder (with the exception of Bolivia, Colombia and surinam), stay in approximately half of the prescribed period (see table 2). 23 Chile is the clearest of these cases. It has a low inverse vulnerability index, but the period in office of the telecommunications deputy secretary (the non-independent regulator, actually a cabinet minister) is longer than the region’s average. For our period, in this country there has been only one regulator that has not finished his period with the country’s president. 24 There are some striking examples in the region. argentina changed the regulator almost a dozen times since the creation of the agency in 1990, including two interventions (2002 and 2004) where the government removed the board and assigned a controller who took charge of the application of government’s policy in the sector. In ecuador, between 1995 and 2004 there were six presidents in the country and the same number of regulators. 25 The local press mentioned colourful scandals around some turnover episodes, such as one of telephone spying related to France telecom in el salvador, or due to alleged nepotism in Panama, where the spouse of the private secretary of the country’s president was appointed as regulator. There are also cases of removals in Mexico or Peru due to bad practices to benefit the incumbent firm. 26 For example, in Costa Rica the general regulator is appointed by the executive by law and for the same period as the president, allowing for re-election, but there is no re-election for the country’s president. In practice, the regulator has never been re-elected. another potentially interesting issue is that the retirement age can be a way to appoint an official for a short period of time, as the Pravas (2003) report suggests for the case of India. For latin america and the Caribbean, only Belize prescribes 55 years old as the retirement age, but it has not affected the period in office of the regulator. There can also be changes in the regulator by changing the rules of the agency. For example, in the case of Mexico, a new president in the country can change the rules (something the executive can do) without being necessary to change the law and after changing the commissioners.. Commitment and Regulatory Independence in Practice in latin american and Caribbean Countries Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 43 table 2. Duration of tRA’s heads by law, turnover Rate and Political Vulnerability Index, Period 1990–2004 Country Legally prescribed (years) Legally prescribed (months) Actual Period (months) turnover Rate Political Vulnerability Index Latin America 4 54 32 60% 0.59 1argentina 560 16 27% 0.33 2Barbados* 448 24 50% 1.00 3Belice 672 72 100% 1.00 4Bolivia 560 40 67% 1.00 5Brasil 560 24 40% 1.00 6Chile Na Na 36 Na 0.00 7Colombia 1.3 16 13 81% 0.67 8 Costa Rica 4 48 22 46% 0.00 9dominican R. 4 48 21 44% 0.00 10 ecuador 448 20 42% 0.20 11 el salvador 7 84 22 26% 0.00 12 Guatemala Na Na 22 Na 0.00 13 Honduras 448 18 38% 1.00 14 Jamaica 5 60 60 100% 1.00 15 Mexico Na Na 27 Na 1.00 16 Nicaragua Na Na 15 Na 0.00 17 Panama Na Na 27 Na 0.50 18 Paraguay 560 22 37% 0.67 19 Peru 560 66 110% 1.00 20 surinam 560 42 70% 0.00 21 trinidad and tobago 336 18 50% 1.00 22 Uruguay* 448 48 100% 1.00 23 Venezuela Na Na 24 Na 0.67 Na: Not applicable, Countries with no duration prescribed by law. * The years by law in Barbados are 5 and in Uruguay 6, but we only count 4, from its creation in 2001 to the final year of our sample, 2004. source: Computed by the authors Francesc trillas and Miguel a. Montoya 44 Intersentia 3.4. PRaCtICal INdePeNdeNCe INdeX together with the information on dates of beginning and end of periods in office, we also obtained information on the reason publicly given for which the head of the agency left his or her position (the full list of reasons is available upon request). The index of independence in practice (PI) uses this information. It is computed as follows: we give a value of 0 when the regulator is dismissed or fired; we give a value of 1 to any other reason for leaving the position, except for those cases where the established period ends (change of government or disappearance of the agency); and we give a value of 2 when the regulator stays in his or her position to the end of the established period. This index does not directly interact with political transitions, but makes it possible to have a time-varying index that includes information about all changes at the top of the agency. 3.5. COMBINed INdICatORs The vulnerability index has a clear limitation when it comes to its use in statistical analysis: it is an average figure for all the studied period, so that it only takes one value for the fifteen years of the period. In order to overcome this limitation, we have created an Index of Independence in law and in Practice (lPI1), which combines the timevarying (legal) Index of Regulatory Independence (IRI) shown in Montoya and trillas (2007)27 and the index of inverse vulnerability, through the multiplication of both indices. as a result, the lPI1 is an index that varies over time and that captures the independence in law and in practice, being the first index of this type in the literature. It takes into account the same information as IRI (or indices that are shown in Montoya and trillas, 2007, to be highly correlated with this, such as the one presented in Gutiérrez, 2003a), and it adds for the first time information about the degree of independence in practice relative to the political majorities. we also have created an Index of Independence in law and in Practice (lPI2), which combines the time-varying Index of Regulatory Independence (IRI) and the practical independence index (PI, which is also time-varying), through the multiplication of both indices. as a result, the lPI2 is an index that captures the independence in law and in practice, and which takes into account the reason why the head of tRa leaves the agency. 27 It must be emphasized that this index includes a measure of privatization of the incumbent. IRI has 10 components: (i) years of effective operation of the agency since its legal creation. (ii) Percentage of private ownership of the incumbent. (iii) degree to which the regulatory agency has powers in the allocation of fixed telephony licences. Powers to (iv) set fixed line tariffs, (v) to allocate spectrum, (vi) in administering universal service. (vii) Budget independence. (viii) term in office for regulators, (ix) appointment rules; and (x) dismissing powers. IRI is similar to the indices of edwards and waverman (2006) and Gual and trillas (2006). The latter reports that an equally weighted index does not yield significantly different results than those derived from principal components indices. Commitment and Regulatory Independence in Practice in latin american and Caribbean Countries Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 45 Finally, LPI3 is an equally weighted sum of the original components of IRI plus the vulnerability index. 3.6. RaNKINGs Concerning the relative position of the 23 countries, some countries which occupy the first positions in the legal index (IRI) are also on top of the legal and practical indices (lPI1 and lPI2), such as Peru and Bolivia. In general, the countries occupying the first places at lPI1 and lPI2 show a correct legal framework, and the longevity of some of their directors, together with the fact that turnover takes place long after political change and/or the head is not dismissed.28 The ten lowest positions are occupied by countries characterized by agencies that are part of a Ministry, so with little legal independence, and hence where independence in practice is meaningless (Chile and surinam); or by rules prescribing that new governments must appoint regulators (dominican Republic and Guatemala); or by a high politicization in the appointment of the head of the agency so that turnover is frequent (Nicaragua). Concerning examples of countries that experience clear changes in their relative positions, we note that el salvador for instance has a correct legal framework but a low level of independence de facto combined with the legal framework. Conversely, Jamaica has a low position in terms of legal framework, but a high position when this legal framework is combined with independence in practice. 3.7. dIsCUssION It can be argued that our measure of independence in practice is a partial one, since we are missing many elements by which politicians may influence regulators. However, many of the ways politicians have to influence regulators are informal and difficult to measure in a consistent or comparable way. we certainly focus on that part of the practice of independence that is related to the job stability of regulators, following the lead of the literature on Central Bank independence. It could also be argued that, in some cases, a regulator that stays in his position could precisely do so because he or she “plays by the rules” and does what politicians want him or her to do; conversely, those who abandon their position could do so as a way to demonstrate their true independence. However, our rankings (for example, improving the relative standing of the Peruvian regulator or relatively downgrading the argentinean regulator) are 28 For example, Bolivia had one of the regulators that stayed in his position for longer, “surviving” to three different country presidents. Peru had one of the most stable agencies in the sub-continent, with only two regulators in 10 years. Francesc trillas and Miguel a. Montoya 46 Intersentia according to the intuition of what are generally understood by experts as well functioning independent institutions or otherwise. 4. tHe IMPaCt OF ReGUlatORy INdePeNdeNCe ON NetwORK PeNetRatION 4.1. sPeCIFICatION, data aNd CROss seCtION we are interested in measuring the impact of regulatory independence on network penetration in fixed telephony. although in developed countries access to a telephone line is usually taken for granted, this is not the case for many developing countries. In our region of study for our sample years (1990–2004), many countries had very low levels of telephone penetration to start with, and the cross country and time variation is high. Graph 1 shows for example the evolution of telephone penetration for Brazil, el salvador, Panama and Venezuela. Of all these, the country that reached the highest growth was Brazil, still falling below the 25% level by the end of the sample period. Graph 1. Main telephone lines per 100 inhabitants. Four LA countries 25.0 20.0 15.0 10.0 5.0 0.0 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 Linies per 100 inhab. Brasil El Salvador Panama Venezuela Commitment and Regulatory Independence in Practice in latin american and Caribbean Countries Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 47 For our sample years (1990–2004) fixed lines were, where available, the main mode of telephony in our region of interest, before the boom in mobiles. Focusing on the fixed lines monopolies, we can concentrate on commitment abstracting from competition issues, which would be crucial in mobile telephony, but would complicate our analysis. In future research we will expand on the relationship between independence and mobile diffusion, interacting with technology and rivalry. to analyze the impact of independence on telecommunications performance, the general model we use can be expressed as: Yit = Β1it + Β2X2it + Β3X3it + … Xkit + μit (1) where Xit = (X2it, X2it, … XKit) are the explanatory variables, including an independence index (IRI, lPI1 or lPI2) and control variables. Β = (Β1, Β2, … ΒK) are their respective parameters and μit is an error term. we use individual fixed effects for the 23 countries. The error term is modeled as: μit = μi + νit (2) where μi denotes non-observable individual effects and νit denotes the remainder of the residual. The performance dependent variable Yit in equation (1) are fixed telephone lines for every 100 inhabitants, obtained from the International telecommunications Union data base. The indices IRI,29 lPI1 and lPI2 are our main explanatory variable of interest. along the lines of other work in this field we expect that network expansion can be potentially affected by the economic structure and industry institutions. For this reason we control for two economic structure variables: GdP measured in purchasing power parity per capita (GdPppp) and population density. we obtained those variables from the data base of the world Bank. we expect that an increase in income per capita and density are associated with higher demand for communications and telephone services. Our starting point is a cross section regression of average telephone penetration over the sample period for each country, to average GdP per capita and average population density. This has an adjusted R square of 0.75, both coefficients are positive and significant despite a very low number of observations (23), and the p-value associated to the F test is 0. If we add as regressors (table 3) either LPI1 or IRI to this regression, the coefficients on gdp and density keep their sign and significance. In the regression where LPI1 is added, the coefficient on the independence variable is negative and non-significant, whereas the adjusted R square value barely changes. However, when the IRI index is added (see table 3), this has a negative impact which is significant at the 10% level and the adjusted R square improves. This negative and weakly significant result is very similar to that found by Gual and trillas (2004) for a different group of 37 countries in a cross section exercise. The negative relationship between IRI and lines penetration 29 Gual and trillas (2004 and 2006) find that the correlation between the equally weighed index and the one derived from principal components analysis is above 0.9. Based on this, IR1 is an equally weighted index. Francesc trillas and Miguel a. Montoya 48 Intersentia may be due to reverse causation: when line penetration is very low, the marginal productivity of independence increases and this may encourage policy makers to increase independence. This coefficient ceases to be significant when IRI is instrumented by polconiii (an index30 of political and institutional constraints developed31 by Henisz and zelner),32 but the Hausman test indicates that the difference between Ols and IV coefficients is not significant (although the IV coefficients are lower in magnitude, though still negative). However, with such small sample, the usefulness of IV is very doubtful, since it has good properties only asymptotically. Besides, a cross section analysis may fail to account for time invariant non-observed country characteristics. For these reasons, we turn to panel data results in the next section. table 3. Parameter estimates for Main Lines per 100 Inhabitants 23 Countries, average values 1990–2004. Cross-section Regressors oLs oLs IV 123 IR1 –11.903* –7.9605 t–1.80 –0.57 LPI1 –3.826 t–0.61 GDPpc 0.003*** 0.003*** 0.003*** t6.58 6.01 6.30 Density 0.014* 0.015* 0.015* t2.05 1.97 2.03 Instrument polconiii adjusted R-sqr 0.7811 0.7489 0.777 N-obs. 23 23 23 Notes. *** statistically significant at 1%. ** statistically significant at 5%. *statistically significant at 10%. 30 The data base on “policy constraints” due to Henisz (2000) is based on four variables that help estimate the probability of policy reversal for the period between 1900 and 2004, for 234 countries. 31 For the study of the privatization, liberalization and regulation of telecommunications, it has been used by Henisz and zeller (2001), Gual and trillas (2004 and 2006) and mentioned by Gutierrez (2003b) and Jamison et al. (2005). Henisz, zelner and Guillen (2005) create and use a data base with three variables to measure (formal) regulator independence, competition and privatization for 205 countries in the period 1960–1999. Both data bases can be obtained in www.management.wharton. upenn.edu/henisz/. 32 Polconiii has a 0.5 correlation coefficient with IR1 and a -0.1 correlation coefficient with lines penetration. The first coefficient is surprising, since one would expect that legal independence is a substitute for other constraints (for example, Chile does not need independence because it has other constraints), although other constraints help to keep in practice whatever level of legal independence is prescribed by law. The results rather indicate that legal independence is on average a complement of other constraints. Commitment and Regulatory Independence in Practice in latin american and Caribbean Countries Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 49 4.2. PaNel data aNalysIs we report static panel data estimates with country fixed effects, instrumenting for independence and not instrumenting for it (table 4). we use five indices of independence: – IRI, a legal independence index similar to the one used in Gual and trillas (2004). – PI, the index of vulnerability of the regulators that varies with time presented in the previous section. – LPI1, which multiplies IRI and the vulnerability index at one month reported in the previous section. – LPI2, which multiplies IRI and PI. – LPI3, an index that adds an equally weighted sum of the original components of IRI plus the vulnerability index. In all regressions, we control for GdP per capita in purchasing power parity and for density. Both have a positive and significant impact in all regressions.33 as it can be seen from the results, adding issues of “independence in practice” to a legal independence index changes the magnitude of the impact but not the sign or the statistical significance. In a static panel, the impact of regulatory independence on network penetration is positive and significant. when other institutional variables are added as regressors (such as reg, polconiii or checks: these are respectively a measure of overall regulatory quality obtained by the world Bank, the constraints index built by Henisz and zelner, and an index of checks and balances also developed by the world Bank) the results of independence do not change (not reported in the table). when both an index of independence in practice (namely, the one that varies over time, PI), and an index of legal independence are included as regressors (column 5), both have a positive and significant impact, and the impact of the legal index is higher in magnitude. However, this result changes when independence is treated as endogenous, although instruments then are weak (see column 10 and the explanations about exogeneity tests below). The first 5 regressions assume that regulatory independence is exogenous. But regulatory independence may be correlated with the error term for the following reasons: – Measurement error, both because in international comparisons we may be comparing slightly differing dimensions and because in aggregating through an index we may assign arbitrary weights. 33 In the regressions where independence is considered exogenous, the R square is the within R square. In the regressions where independence is endogenous the R square is the centered R square, as reported by stata. Francesc trillas and Miguel a. Montoya 50 Intersentia table 4. Parameter estimates for main lines per 100 inhabitants. 23 countries. 1990–2004. Panel Data, country fixed effects Regressors 12345678910 LPI1 6.561*** 12.307*** z7.78 7.34 LPI2 2.165 *** 7.67 *** z4.47 5.93 LPI3 5.525*** 8.783*** z7.07 7.39 PI .6962** 4.489* z2.28 1.65 IRI 5.268*** 4.152*** 8.997*** –.771 z6.55 4.43 7.23 –0.13 Density .1809*** .113*** .1867*** .1143*** .124*** .163*** .051 .1502*** .0632** .1614* z7.42 4.02 7.23 4.16 4.36 6.20 1.55 4.77 2.02 2.25 GDPpcpppr .0023*** .0025*** .0024*** .0025*** .0023*** .0018*** .002*** .0013*** .0022*** .0014* z8.51 9.14 8.21 9.01 8.32 5.88 7.82 3.12 7.75 2.10 R-sqr 0.5443 0.5222 0.4897 0.5312 0.5299 0.4779 0.4899 0.2823 0.5057 0.2975 N-obs. 344 344 344 344 344 344 344 344 344 344 Instruments —————efi, staff efi, staff efi, staff efi, staff efi, staff, polconiii (weak) Notes. * statistically significant at 1%. ** statistically significant at 5%. *** statistically significant at 10%. Hausman test of Ho: † Fixed effects vs Random effects. ‡ The explanatory variable is not endogenous. Commitment and Regulatory Independence in Practice in latin american and Caribbean Countries Competition and Regulation in Network Industries, Volume 12 (2011), No. 1 51 – Reverse causality, because as mentioned in the cross section exercise, a low penetration level may trigger measures to increase penetration, such as regulatory independence. – Omitted regressors, because institutions or social preference may at the same time be causing a high penetration level and a high level of independence. Hence, since we were suspicious of potential endogeneity problems in our independence measures, in columns 6–10 we endogenize the independence indices by two politicoinstitutional variables, efi (economic freedom index, an index developed by the Fraser Institute) and staff (a relative measure of the size of the incumbent operator). In column 3 we need an additional instrument to run the overidentification test, and we use polconiii. except in column 10 where the instruments turn out to be weak, in all the other tests the instrumental variable estimates are higher in magnitude than the estimates computed without instrumenting. Using the LPI1 index, for example, the regression results treating independence as exogenous imply that, if the model is correct, when a country has the independence level in 2004 of the highest country in the ranking (Peru), it has 4 more lines per 100 inhabitants than when it has the independence level of the lowest country in the ranking, Costa Rica, everything else constant. However, if independence is (correctly) treated as endogenous, the impact of the difference is to have 7.6 more lines. we conclude that treating regulatory independence as exogenous may substantially underestimate the impact of independence on network penetration. However, finding good instruments is difficult (Kennedy, 2008), since only in the regression in column 6 the instruments were fully convincing, at least in the narrow sense of passing the overidentification sargan test and a weak exogeneity test, although in all cases the null hypothesis of exogeneity was rejected using the Hausman test.34 In column 6, the Crag-donald wald F statistic of weak exogeneity of the instruments is 63.846, which is above the stock-yogo critical value at 10% of 19.93, meaning that the hypothesis of weak exogeneity is rejected. The sargan statistic of overidentification is 1.04 with a P-value of 0.3060, which does not reject that the instruments are uncorrelated with the error. Finally, the chi-square value of the Hausman test is 15.70, with an associated p-value of 0.0013, which rejects the hypothesis that there is no difference between the IV estimates and the conventional panel data estimates. In column 7, the Crag-donald wald F statistic of weak exogeneity of the instruments is 125.081, which is above the stock-yogo critical value at 10% of 19.93, meaning that the hypothesis of weak exogeneity is rejected. The sargan statistic of overidentification is 3.920 with a P-value of 0.047, which does not reject that the instruments are uncorrelated with the error, but only at the 5% level. Finally, the chi-square value of the Hausman test is 15.3, with an associated p-value of 0.0015, which rejects the hypothesis that there is no difference between the IV estimates and the conventional panel data estimates. 34 Other instruments such as reg and checks were tried, but the test results were even weaker.