Consequences of Vertical Separation and Monopoly: Evidence From the Telecom Privatizations

Abstract

Policy variation across countries on the use of mandatory vertical separation and statutory monopoly allows the assessment of their impact on basic telephone services (local, long distance, and international service). Panel data analysis from 56 countries during the 7-year period following the privatization of the main telephone provider indicates that vertical separation and monopoly harm the consumers that were supposed to benefit: the downstream users of international telephony and the upstream users of residential local telephony. Mandatory vertical separation reduces the usage of international telephone service and the number of fixed lines in service, whereas statutory monopoly reduces the amount of fixed lines in service and increases the price of local residential telephony.

Notes

¹Few privatizing governments, such as India, Canada, and Japan, inherited a vertically separated industry. In other cases, such as Argentina and Brazil, separation was decided just before the privatization.

²Empirical evidence suggests that local residential services were priced at below marginal cost, whereas business services, such as long distance and international telephony, were priced at well above marginal cost (Crandall, 1989; Cronin, Colleran, Miller, & Raczkowski, 1997; Hausman, Tardiff, & Belinfante, 1993; Nambu, Suzuki, & Honda, 1989; Palmer, 1992).

³Although long distance and international service were also offered to residential customers, the heavy users were large businesses. This is why, especially in the more advanced economies, the impetus for privatization and liberalization came from the business sector. For an excellent analysis, see Noam (1987, 1992) and Petrazzini (1995).

⁴Although the argument for monopoly (and cross-subsidization) was mainly political, Faulhaber (1975) found an economic justification. He proved that cross-subsidization can increase social welfare as long as a multi-product monopolist exhibits economies of scope on joint production.

⁵ Bös (1993) articulated this view: “In contrast to private firms, public enterprises have often been instructed to price according to distributional objectives. This implies charging lower prices for goods which are mainly demanded by lower-income earners. In this case the public enterprises rely on internal subsidization, where the internal deficit of the low-priced goods is financed by the internal profits earned from sales to higher-income or business customers. If the privatized firm operates in a competitive market, this internal subsidization becomes impossible and distributional pricing cannot be upheld” (p. 108). Pilcher (1994) put it bluntly: “Governments, therefore, need to decide on a strategy to either introduce competition in long distance and international service or to maintain the cross-subsidy” (p. 401).

⁶This view is clearly spelled out by Bös (1993): “The government should first attempt to encourage competition, as the UK, for instance, did with its splitting of the electricity industry into electricity generation (a potentially competitive business) and distribution” (p. 108). Also, Waterschoot (1994, p. 511) underscored the influence of the U.S. antitrust case against AT&™ in shaping the post-privatization regulation of telecommunications around the world.

⁷In a related contribution to the theory of vertical integration, Arrow (1975) showed that firms will integrate vertically when there is uncertainty in the supply of the upstream (intermediate) good.

⁸See, for example, Fink, Mattoo, and Rathindran (2003) and Ros (1999).

⁹Exceptions to this are the studies focusing on the United States after the breakup of AT&T. These studies find that vertical separation did not produce the expected welfare gains from lower prices on downstream long distance and international telephone services (Crandall, 1988, 1989; Hausman et al., 1993).

¹⁰See also Alchian (1995), Crandall and Sidak (2002), and Sappington (1995).

¹¹Available at http://www.fdi.net.

¹²For such analysis, see Li and Xu (2004), Ros (1999), or Wallsten (2001).

¹³Focusing on these countries avoids the complications of controlling for state ownership, with the potential endogeneity problems associated with it. Also not included are countries that had the incumbent always under private control (such as in the United States) because these firms may behave differently than recently privatized firms. I thank an anonymous referee for pointing this out.

¹⁴Another reason for limiting the time period is to avoid endogeneity problems due to changes in telecommunications regulations. As will be explained soon, unobserved heterogeneous regulation across countries is not a problem as long as it remains time invariant over the 7 years following the privatization. The 7-year timeframe strikes a balance between the need to have enough within country variation in the explanatory variables and the need to have a short timeframe to render the policy variables exogenous.

^a Monopoly years at the time of privatization. When monopoly periods differ between international and local service, the number of years of monopoly in local service is shown in parentheses.

^b Number of years (at the time of privatization) since local and international service were vertically separated.

^c Local and international telephone providers were sold separately but allowed to merge after privatization. At the time of privatization, neither country mandated vertical separation, but the services had been vertically separated 25 years in Peru and 20 years in Trinidad and Tobago.

¹⁵See Viani (2007) for an analysis of the factors influencing the award of monopoly rights in the privatization of telephone firms.

¹⁶Note that, for each country, the earliest observation of y_it corresponds to the year of the privatization sale. At that time, contemporaneous or future values of w_it were predetermined and, thus, no endogeneity problem exists.

¹⁷In Peru and in Trinidad and Tobago, the local and international service providers merged after the privatization sale, but the rules were not changed. When the buyers purchased these firms, they knew in advance that they would be allowed to merge. Again, the rules were predetermined and not changed, confirming the exogeneity of vertical separation.

¹⁸Of course, a government can choose to renege on a contract, but this could be costly. In fact, many countries sought to signal their commitment not to change the rules by signing the Convention on the Settlement of Investment Disputes with the International Centre of Settlement of Investment Disputes. This intends to circumvent poor systems of rule of law prevailing in many less-developed countries (see http://www.worldbank.org/icsid/about/main.htm).

¹⁹See also Wooldridge (2002, p. 306).

²⁰For that matter, any (unobserved) country-specific characteristic will be adequately captured by the fixed effects as long as it is constant over the 7 years following the privatization sale.

²¹For a good illustration of this and related problems, see Berg and Tschirhart (1988, pp. 505–511). See also Alchian and Kessel (1962), Comanor and Leibenstein (1969), Leibenstein (1966), and Williamson (1963).

²²Other studies that use a continuous monopoly variable to capture the effect of the length of monopoly on output and prices include Duso (2005) and Wallsten (2004).

²³There is, however, one way to turn a dummy variable into almost time invariant. In two countries (Peru and Trinidad & Tobago), the government had two separate firms for the provision of local and international services, and sold these separately. Because vertical integration was allowed, soon after privatization, these firms integrated vertically. Defining a dummy variable to be equal to one if the provider of local service was separated from that of international service would mean that, for Peru, this variable would take the value of one in the first year (t ₀) and zero in the following 7 years. The case for Trinidad and Tobago would be similar, except that it took 2 years for these firms to merge. The bigger concern here, however, would be that the test of whether vertical separation reduces phone services would be determined by only three observations. This would clearly be inadequate.

²⁴For a theoretical exposition of optimal access fees, see Laffont and Tirole (2000, pp. 80–83, 97–105).

²⁵See, for example, Barro (1991) and Brunetti (1997) on the effect of political instability on investment and economic growth.

²⁶For example, in 1991, the share of international service on total revenue for the Jamaican telephone monopoly was 77% (Wint, 1996, p. 59).

²⁷A test of weak instruments is performed on these variables. The null hypothesis of both instrument coefficients being zero is rejected with a p value of .134. To further test the suitability of these instruments, an F test is performed on the overall significance of all the instruments used. The null hypothesis that all instrument coefficients are zero is rejected with more than a 99% confidence level. The p value of this test is reported in the first-stage regression results in the Appendix.

²⁸I thank an anonymous referee for pointing this out.

^a Sources include the Economist Intelligence Unit Viewswire and Country Information Databases, each firm's annual reports, each regulator's Web sites, and the Commission of the European Communities.

²⁹See Froot (1989) and Williams (2000).

³⁰ Table 3 shows (at the bottom) the p values of rejecting incorrectly the null hypothesis (H₀) in Hausman's (1978, 1983) test, where H₀ is the hypothesis that fixed lines per person (in logs) is exogenous.

³¹See also Wallsten (2004) for similar findings with a smaller sample and using ordinary least squares estimation.

^a Hausman test of endogeneity, H₀: Instrumented variable is exogeneous. P value is the probability of rejecting H₀ incorrectly.

^b F test of overall significance of a regression, H₀: All coefficients are zero. The p value is the probability of rejecting H₀ incorrectly.

* = 99% confidence.

** = 95% confidence.

*** = 90% confidence.

³²Commission of the European Communities, Seventh Report on the Implementation of the Telecommunications Regulatory Package (November 26, 2001).

³³The dummy variable takes the value of one if managerial control was transferred, and zero otherwise. A problem is that not all privatization sales specify whether managerial control was transferred. In those sales, I assume managerial control was transferred if either (a) the government announced the sale of equity to a private “strategic partner”; or (b) as a result of the sale, the government's equity share fell below 50%. The main sources for these data were the Privatisation International Yearbook and monthly issues, and the Economist Intelligence Unit Viewswire and Country Information online databases.

³⁴The data source is Wallsten et al. (2004).

³⁵The exact way to compute this is 100 × [exp(β) – 1], where β is the coefficient of vertical separation.

³⁶The direct effect of additional years of vertical separation on international service is 0.8% as stated before. The indirect effect of vertical separation on international telephony is found by first computing its effect on fixed lines per person. From the coefficient of vertical separation in the first column in Table 6, we find this to be 0.9% (see the next section). Second, we need the elasticity of international telephony with respect to fixed lines. From Table 4 we find it to be approximately 0.5; therefore, the indirect effect of vertical separation on international service is 0.45 (0.9 × 0.5), and the direct plus indirect effect of additional years of vertical separation on international telephony is 1.3%.

³⁷ Wallsten (2004) found that if the number of monopoly years increase by 1%, international service usage declines by 0.087%. As an illustration, the difference between a 1-year monopoly and a 5-year monopoly translates into a decline of 17.4% on international minutes. This is not too far from the 19.2% decline on minutes per person estimates in Table 4 (direct effect only).

³⁸Again, the indirect effect of additional years of monopoly on international telephony is found by first finding its effect on fixed lines per person. From the first column in Table 6 (see the next section), we find this to be about 4.5%. Second, we need to use the elasticity of international telephony with respect to fixed lines, which we found to be approximately 0.5 from the fourth column in Table 4. Therefore, the indirect effect of monopoly on international telephony is approximately 2.2% (0.5 × 4.5), and the direct plus indirect effect of additional years of monopoly on international telephony is 7%.

³⁹The null hypothesis of this instrument coefficient being zero is rejected with a p value of .131. To further test the suitability of the instruments used, an F test is performed on the overall significance of the instruments. The null hypothesis that all instruments' coefficients are zero is rejected with more than a 99% confidence level. The p value of this test is reported in the first-stage regression results in the Appendix.

⁴⁰See Drukker (2003) and Wooldridge (2002, pp. 282–283) for the test used.

⁴¹See Arrow (1975) for the effect of uncertainty on the incentive on firms to integrate vertically.

^a Hausman test of endogeneity, H₀: Instrumented variable is exogeneous. P value is the probability of rejecting H₀ incorrectly.

^b F test of overall significance of a regression, H₀: All coefficients are zero. The p value is the probability of rejecting H₀ incorrectly.

* = 99% confidence.

** = 95% confidence.

*** = 90% confidence.

⁴²As in the previous section, this variable takes the value of one if managerial control was transferred, and zero otherwise.

⁴³The data source is Wallsten et al. (2004).

⁴⁴ Wallsten (2004) found that if the number of monopoly years increase by 1%, the number of fixed lines in service declines by 0.07%. As an illustration, the difference between a 1-year monopoly and a 5-year monopoly translates into a decline of 14% on fixed lines in service. This is not too distant from the 18% decline estimated in Table 6.

⁴⁵A test of weak instruments is performed on these variables. For each instrumented endogenous variable, the null hypothesis of the four instrument coefficients being zero is tested. This hypothesis is rejected with more than a 99% confidence level, except in the first-stage regression of the natural logarithm of the number of faults per line (column 4 in the Appendix). In this case, the null hypothesis cannot be rejected. To further test the suitability of the instruments used, an F test is performed on the overall significance of the instruments. The null hypothesis that all instrument coefficients are zero is rejected with more than a 99% confidence level. The p value of this test is reported in the first-stage regression results in the Appendix.

⁴⁶Even AT&T's chairman, Charles Brown, agreed with the conventional view: “[W]ith competition, this subsidization of local rates by AT&T's long distance service is no longer possible and will be gradually phased out. Long distance rates will come down, and local rates will rise” (as cited in Temin & Galambos, 1987, p. 307).

⁴⁷This is computed as the one-time connection charge plus the monthly rate for perpetuity. A discount rate of 5% is assumed in the perpetuity formula. Therefore, Price = Connection Charge + Monthly Rate/0.05.

⁴⁸This dummy variable takes the value of one if managerial control was transferred, and zero otherwise.

^a Hausman test of endogeneity, H₀: All three instrumented variables are exogeneous. P value is the probability of rejecting H₀ incorrectly.

^b F test of overall significance of a regression, H₀: All coefficients are zero. The p value is the probability of rejecting H₀ incorrectly.

* = 99% confidence.

** = 95% confidence.

*** = 90% confidence.

⁴⁹Again, the exact estimation of this is computed by 100 × [exp(β) − 1], where β is the coefficient of the monopoly variable.

^a First stage of the fourth column in Table 4.

^b First stage of the fourth column in Table 6.

^c First stage of the fourth column in Table 7.

^d F test of overall significance of a regression, H₀: All coefficients are zero. The p value is the probability of rejecting H₀ incorrectly.

* = 99% confidence.

** = 95% confidence.

*** = 90% confidence.