NET Institute* www.NETinst.org Working Paper #06-22 October 2006 Merger Simulation in Mobile Telephony in Portugal Lukasz Grzybowski, University of Alicante Pedro Pereira, Autoridade da Concorrencia, Potugal * The Networks, Electronic Commerce, and Telecommunications (?NET?) Institute, http://www.NETinst.org, is a non-profit institution devoted to research on network industries, electronic commerce, telecommunications, the Internet, ?virtual networks? comprised of computers that share the same technical standard or operating system, and on network issues in general. Merger Simulation in Mobile Telephony in Portugal? Lukasz Grzybowskiy University of Alicante Pedro Pereiraz Autoridade da Concorre^ncia November 2006 Abstract This article assesses the unilateral e ects on prices of a merger in the Portuguese mobile telephony market. We use aggregate quarterly data from 1999 to 2005 and a nested logit model to estimate the price elasticities of demand and the marginal costs of subscription of mobile telephony. Given these estimates, we simulate the e ects of the merger. We nd that the available mobile telephony subscription products are close substitutes. The merger may cause substantial price increases, even in the presence of large cost e ciencies. On average, prices increase by 7% without cost e ciencies, and by about 6% with a 10% marginal cost reduction. Key Words: mobile telephony, merger simulation, nested logit, network e ects, lock-in JEL Classi cation: L13, L43, L93 ?Financial support from the NET Institute, www.NETinst.org and the CEPR RTN Competition Policy in International Markets is gratefully acknowledged. The opinions expressed in this article re ect only the authors? views, and in no way bind the institutions to which they are a liated. yDepartamento de Fundamentos del Analisis Economico, Universidad de Alicante, Campus de San Vicente, 03080 Alicante, Spain, Phone: +49 179 6662936, E-mail: lukasz@merlin.fae.ua.es zAdC, Rua Laura Alves, 1050-188 Lisboa, Portugal, e-mail: jppereira@oninetspeed.pt 1 1 Introduction In February 2006, the rm Sonaecom proposed an operation that would involve, among other things, the merger of the Portuguese mobile telephony rms Tmn and Optimus. The Portuguese market includes a third rm, Vodafone. The revenue market shares of Tmn, Vodafone, and Optimus are, respectively: 50%, 37%, and 13%. Given that the merger reduces the number of competitors from three to two, and increases considerably the level of concentration, it raises serious anti-competitive concerns. According to the EC Merger Regulation,1 mergers must be assessed to determine whether they signi cantly lessen competition, due to the creation or reinforcement of a dominant position. In economic terms, mergers must be assessed to determine whether they enhance the market power of the rms in the market. Horizontal mergers may signi cantly lessen competition through: (i) unilateral e ects, i.e., by eliminating important competitive constraints, or, (ii) coordinated e ects, i.e., by making anti-competitive coordination between the remaining rms more likely or more e ective. This article assesses the unilateral e ects on prices of the merger between Tmn and Optimus. Our analysis uses aggregate quarterly data from 1999 to 2005. We estimate an aggregate nested logit model, based on a linear utility function for mobile telephony subscription, developed by Doganoglu and Grzybowski (2006).2 Given the estimates of the price elasticities of demand, and assuming that rms play a Bertrand game, we estimate the marginal costs. We use the estimates of the price elasticities of demand and the marginal costs to simulate the unilateral e ects on prices of the merger. Our results indicate that the merger would lead to signi cant price increases, even in the presence of substantial cost e ciencies. On average, prices increase by 7% without cost e ciencies, and by 6% with a 10% a marginal cost reduction. The price of some products could increase by as much as 13%. On average, without marginal cost e ciencies, the 1Council Regulation (EC) No 139/2004 of 20 January 2004 (OJ L 24, 29.01.2004, p.1). 2According to Crooke et al (1999), predicted post-merger price changes vary greatly with the demand spec- i cation. The price increases predicted by the logit model are lower than those predicted by the log-linear and AIDS models, but higher than those predicted by the linear demand. 2 consumer surplus and the social welfare decrease both by 3%. We nd no evidence of collusion in prices. There is a growing, but still scarce, econometric literature on merger analysis. Nevo (2000) estimated a random coe cient model to study the e ects of mergers in the US ready-to-eat cereal industry. Ivaldi and Verboven (2005) analyzed the e ects of the merger between Volvo and Scania in the EU. Pinkse and Slade (2004) used a distance metric approach to study mergers in the brewing industry. Pereira and Ribeiro (2006) analyzed the e ects on broadband access to the Internet of the divestiture, the opposite of a merger, of the Portuguese telecommunications incumbent from the cable television industry. Regarding the empirical literature on mobile telephony, Rodini et al. (2002) estimated the substitutability of xed and mobile services for telecommunications access, using data on US households. Okada and Hatta (1999) analyzed jointly the demand for mobile and xed telephony services using Japanese data. Barros and Cadima (2000) estimated simultaneously di usion curves for mobile and xed telephony in Portugal. Grzybowski and Doganoglu (2006) used a nested logit model to estimate the demand for subscription of mobile telephony in Germany. Gagnepain and Pereira (forthcoming) analyzed the impact of the entry of a third rm on the Portuguese mobile telephony market, and showed that it caused the other rms to increase their cost reducing e orts. The remainder of the article is organized as follows. Section 2 gives an overview of the Portuguese mobile telephony industry. Section 3 presents the econometric model and the data. Section 4 presents the estimation results. Section 5 conducts the analysis of the impact of the merger. Section 6 concludes. 2 Mobile Telephony in Portugal In Portugal, the rm associated with the telecommunications incumbent, Tmn, started operations in 1989 with the analogue technology C-450. In 1991, the sectorial regulator, ICP- ANACOM, assigned two licenses to operate the digital technology GSM 900. One of the licenses 3 was assigned to Tmn. The other license was assigned to the entrant Vodafone. In 1997, the sectorial regulator assigned three licenses to operate the digital technology GSM 1800. Two licenses were assigned to Tmn and Vodafone. A third license was assigned to the entrant Optimus, which was also granted a license to operate GSM 900. In 2001, the sectorial regulator assigned licences to operate the 3G technology IMT2000 /UMTS. Three licenses were assigned to Tmn, Vodafone, and Optimus. A fourth license was assigned to the entrant Oniway, which was not granted a license to operate GSM, and never operated before. Service began in 2001. After its inception in 1989, the Portuguese mobile telephony industry had a fast di usion, analyzed in Pereira and Pernias (2006) and Pereira et al. (2006). In 2005, the penetration rate of mobile telephony in Portugal was 108%.3 After entering the market in 1992, Vodafone gained revenue market share rapidly. During the duopoly period, i.e., from 1992 to 1997, Tmn and Vodafone shared the market. The entry of Optimus led to an asymmetric split of the market, which suggests that this event had a substantial impact in the industry, analyzed in Gagnepain and Pereira (2006). In February 2006, the holding company Sonaecom, that owns Optimus, made an hostile take bid for the holding company Portugal Telecom, PT, the telecommunications incumbent. The merger of Tmn and Optimus was one of the conditions for the operation to go through. 3 Econometric Model 3.1 Demand We assume that all consumers have access to a xed telephone. Consumers face a two stage decision. In the rst stage, they decide whether to continue to use only xed telephony, or to use also mobile telephony. In the second stage, they decide to which mobile telephony product they subscribe, if they decided to use mobile telephony in the rst stage. This is a standard 3The penetration rate is the numbers of subscribers per 100 inhabitants. 4 nested logit structure, where one branch is degenerated and involves no further choices.4 We index the nests with subscript g = 0; 1, where the nest g = 0 represents the choice of only xed telephony, and the nest g = 1 represents the choice of xed telephony and mobile telephony. We index consumers with subscript i, products with subscript j, and time period with subscript t. Denote by Ui0t, the utility of the outside option of consumer i at time t. Value Ui0t may vary over time, e.g., due to its dependence on the prices of xed telephony. There are three mobile telephony products, each associated to one of the three rms currently in the market. The utility derived by consumer i from using xed telephony together with the mobile telephony product j in period t is given by: Uijt = Ui0t + rj ? fipjt + Vt + ?jt + ?gt + (1 ? )?ijt; (1) where rj is the stand alone value of product j, pjt is the price of product j in period t, Vt is the expected network bene t function in period t, ?jt is the unobserved utility of product j in period t, ?gt is the common value of all products in nest g in period t, on [0; 1) is a parameter that measures correlation of the consumers? preferences within the nest xed telephony and mobile telephony, and ?ijt is an idiosyncratic taste variable. By normalizing with respect to the utility of the outside option, Ui0t, the choice between alternatives becomes independent of the determinants of the utility of xed telephony. The consumer?s tastes for products within a nest are correlated. When the consumer?s tastes for products within a nest are independent, = 0, the nested logit model reduces to the logit model. When ?! 1, the alternatives within the nest are perfect substitutes and the post- merger prices do not change. Variable ?jt is interpreted as the mean value of the consumers? valuations for unobserved product characteristics, such as product quality, and serves as the econometric error term.5 Variable ?gt has a distribution dependent on , and variable ?ijt has 4A discrete choice model ts well mobile telephony, because consumers usually have one mobile telephone and the outside option, xed telephony, is clearly de ned. However, nested logit model has the property of independence of irrelevant alternatives within the nests. This property may be tested following Hausman and McFadden (1984). 5Prices, stand alone values and unobserved qualities are the only product characteristics in the model. 5 an extreme value distribution. The probability that consumer i subscribes to product j in period t is: Pijt = Pijt(g = 1)Pijt ?Uijt ? maxk2f1;::::;Ng;k 6=jUikt ?; (2) Denote by Dgt, the value of the nest g in period t, de ned by: Dgt = X j2Gg exp ?jt 1 ? ? ; (3) where G is the set of products in nest g, and ?jt = rj ? fipjt + Vt + ?jt is the mean utility level of product j in period t. Expression (2) may be written in a closed form as: Pijt = exp ? ?jt 1? ? Dgt D1? gt 1 +PGg=1 D1? gt : (4) Denote by sjt, the share of product j in period t, i.e., the number of consumers that choose product j in period t divided by the number of consumers that make subscription decisions in period t. Probability (2) is equivalent to this share, i.e., sjt = Pijt. The share of mobile telephony is given by st = PN j=1 sjt. Denote by sjtjg=1, the share of product j of mobile telephony services in period t, i.e., the number of consumers that choose product j in period t divided by the number of consumers that choose to consume mobile telephony in period t. The share of the outside good, i.e., the number of consumers that choose not to subscribe a mobile telephony product divided by the number of consumers that make subscription decisions in period t is given by s0t = 1 ? st = 1 ? PN j=1 sjt. Following Berry (1994), we invert the observed market shares to compute the mean utility levels of each product, and treat them as observed utility levels. Using the observed utility level and the speci cation in (1), we arrive to the following equation: log(sjt) ? log(1 ? st) = rj ? fipjt + Vt + log(sjtjg=1) + ?jt: (5) The utility function (1) represents only consumers whose current choice does not depend on the past, i.e., of consumers that just entered the market, or consumers that where previously in the market but have negligible or no switching costs. 6 There are three types of consumers: (i) consumers that just entered the market, (ii) con- sumers that where previously in the market but are not locked-in to a product, and (iii) con- sumers that where previously in the market and are locked-in to a product. The two rst types of consumers choose which product they subscribe. The third type of consumers continue to use the product that they subscribed previously. We lack data on the number of consumers that switch of providers over time, and the choices they make. Thus, regarding the third type of consumers we consider two cases: (i) only consumers with contracts are locked-in, and (ii) consumers with contracts and 50% of consumers with pre-paid cards are locked-in.6 Since about 80% of the Portuguese consumers use pre-paid cards, in the rst case, the majority of consumers are not locked-in, and may react to price changes. Denote by Zjt, the number of subscribers of product j in period t, denote by Z ljt, the number of subscribers of product j in period t that are locked-in, and denoted by yjt, the number of consumers that choose product j in period t. We approximate the number of consumers that choose product j in period t by the di erence yjt = Zjt ? Z ljt?1. Denote by Mt, the market size in period t, i.e., the number of consumers that can potentially subscribe to mobile telephony in period t. We assume that Mt = 1:08Pt, where 1:08 is the penetration rate of mobile telephony in 2005:4, and Pt is the population of Portugal in period t. Denote by mt, the number of consumers that e ectively make subscription decisions in period t. This value is given by mt = Mt? PN j=1 Z ljt?1. The share of product j is then given by sjt = yjtmt . Denote the by zt = 1Mt PN j=1 Zjt, the penetration rate of mobile telephony in period t. We assume that due to lags in information transmission, each period consumers only observe the number of subscribers of the previous period. Thus, the expected network bene ts are a function of the penetration rate of the previous period: Vt = V (zt?1).7 In addition, we assume that the expected network bene ts are a linear function: V (zt?1) = flzt?1.8 Network e ects 6Alternatively, one could assume that all consumers of mobile telephony of the previous period are locked-in. 7Of course this begs the question of why consumers do not forecast the contemporary penetration rate. Using zt in the utility function, which depends on current prices, pjt, would greatly complicate the derivation of the rst-order conditions. In addition, values zt and zt?1 are highly correlated. 8Consumers also derive network bene ts from a xed line network. Given the linearity of the expected network 7 are an important aspect of mobile telephony. Ignoring them could lead to overestimating the price elasticities of demand. By making network e ects depend on the penetration rate, i.e., the total number of subscribers, we are assuming perfect compatibility between the various mobile telephony products, and the lack of price mediated network e ects, due to di erence on-net and o -net prices. Using demand equation (5), the substitution matrix for the nested logit model is given by: St = 2 66664 @s1t @p1t @s2t @p1t @s3t @p1t @s1t @p2t @s2t @p2t @s3t @p2t @s1t @p3t @s2t @p3t @s3t @p3t 3 77775 ; where @sjt @pkt = 8 < : ? fi1? sjt ?1 ? sjtjg=1 ? (1 ? )sjt ? if k = j; fi 1? sjt ? sktjg=1 + (1 ? )skt ? if k 6= j: (6) and @s0t @pkt = fiskts0t: (7) The elements of matrix St are calculated using the estimates of and fi. Using (6), for the products within the mobile telephony nest, the elasticity of demand of product j with respect to the price of product k is given by: Esjtpkt = @sjt @pkt pkt sjt : Using (7), the elasticity of the market demand for mobile telephony subscription with respect to the price of product j is given by: Estpkt = @(1 ? s0t) @pkt pkt st = ?fiskt(1 ? st) pkt st : The elasticity of the market demand for mobile telephony subscription with respect to the lagged penetration rate is given by: Estzt?1 = flzt?1(1 ? st): bene ts function and the linearity of the utility function (1), any network bene ts from xed telephony are cancelled out when we normalize with respect to the outside option. 8 Following Anderson et al. (1992) and Ivaldi and Verboven (2004), the net consumer surplus for the nested logit model, CSt, equals: CSt = 1fi 1 + GX g=1 D1? gt ? : (8) 3.2 Supply We index rms with subscript f . Denote the price vector by p, and the demand for product j by Qj(p). The pro t function for rm f is: f = X j2 f ?(pj ? cj)Qj(p) ? where f is the set of products owned by rm f , and cj is the marginal cost of rm j. Denoted by , the 3 ? 3 ownership matrix, formed according to the rule: fj = 8 < : 1 if rm f sells product j; 0 otherwise. We assume that rms choose prices and play a static non-cooperative game, i.e., a Bertrand game. Additionally, we assume that rms only take into account in the pro t maximization problem consumers that are not locked-in. The Nash equilibrium of the game is given by the rst-order conditions:9 @ f @pk = Qk + JX j=1 fj @Qj@pk (pj ? cj) = 0; (9) Denote by ? St, the Hadamard, or element by element product of matrices and St. System (9) can be written in matrix notation as: Q + ( ? St) (p ? c) = 0: (10) Initially, there are three mobile telephony rms: Tmn, Vodafone, and Optimus. Each rm owns one of the mobile telephony products. Thus: =I. 9We assume that a Nash equilibrium exists. See Anderson et al. (1992) for a proof of existence for the nested logit model with multiproduct rms, assuming symmetry. 9 In the course of the analysis, we will assume two alternative forms for the matrix : (i) one associated with the merger of Tmn and Optimus, and (ii) one associated with joint pro t maximization. 3.3 The Data We use quarterly data from Portugal for the period 1999:1 to 2005:4, obtained from the rms. For each rm and period, we have: the number of subscribers, Zjt, the number subscribers with pre-paid cards, the number of subscribers with contracts, the revenues from voice tra c, the voice tra c, the operating costs, the cost of materials, the cost of labor, and the number of employees. We also have the interest rate on ten year treasure bonds. We construct average prices, pjt, by dividing the total rm revenues from tra c by the total number of minutes of tra c. 4 Econometric Implementation 4.1 Demand Estimates We estimate two models.10 In Model I, only consumers with contracts are locked-in. In Model II, consumers with contracts and 50% of consumers with pre-paid cards are locked- in. Otherwise the models are identical. In particular, in both models the demand for mobile telephony subscription is a function of: (i) the average prices, (ii) the lagged penetration rate of mobile telephony, and (iii) rm speci c dummy variables. The price and network coe cients are common across the three demand functions. [Table 1] Table 1 presents the estimation results for Model I. The model is estimated rst using ordinary least squares, OLS. Reported R-squared is about 0.60 for each equation. The Breusch- 10The model was estimated using the SAS procedure PROC MODEL. The merger simulation was conducted in Matlab. 10 Godfrey test rejects the null hypothesis of no autocorrelation and the White test rejects the null hypothesis of no heteroscedasticity in all three demand equations. Besides, the estimate of is larger than 1, but not statistically signi cant. To account for the problems of autocorrelation and heteroscedasticity of the error term, and the problem of endogeneity of prices and the within group shares, we estimate the model through the general method of moments, GMM, using the Newey and West (1987) consistent covariance matrix estimator. We use as instruments the weighted average across rms of the cost of materials per subscriber, the cost of labor per employee, and the number of workers per subscriber, as well as the interest rate of ten year treasure bonds, a time trend, and a dummy variable for the fourth quarter.11 The GMM estimates of all parameters are statistically signi cant. The estimates of fi and decrease considerably. In particular, is estimated to be 0:865, but statistically signi cantly di erent from 1. This implies that mobile telephony products are very close substitutes. Table 1 presents also the estimation results for Model II. The OLS estimates are presented merely as a reference. The GMM estimates are again all statistically signi cant. Compared to Model I, the estimate of fi increases, and the estimates of and fl decrease. The estimate of is 0:744. However, it is not statistically di erent from the value 0:865, the estimate of for Model I. The estimates of fi and fl are also not statistically signi cantly di erent from the estimates of Model I. 4.2 Price Elasticities of Demand Table 2 presents the average price elasticities of demand for the GMM estimates. The second, third, and fourth columns give the own and cross price elasticities of demand for the mobile telephony products. The last column gives the elasticities of the market demand of mobile telephony with respect to the prices of the mobile telephony products and the lagged 11For instruments we need variables that are correlated with prices and within group shares, but uncorre- lated with the unobservable demand shocks. Standard candidates for instruments are input factors (Evans and Heckman, 1983). 11 penetration rate. [Table 2] For Model I, the own price elasticities of demand are high and range from ?2:61 for Tmn to ?6:32 for Optimus. A 1% increase in the price of Optimus decreases its sales by 6:32% and increases the sales of Tmn and Vodafone by 1.33% each. The values in the last column of Table 2 are interpreted as follows. A 1% price decrease in the price of Optimus increases the number of subscribers of mobile telephony by 0:06%. This is the outcome of two opposing e ects: a decrease in the number of subscribers of Optimus, and an increase in the number of subscribers of Tmn and Vodafone. Similarly for Tmn and Vodafone. If all rms increase simultaneously their prices by 1%, the number of subscribers of mobile telephony decreases by the sum of these values, which is 0:341%. The small values of the elasticity of the market demand of mobile telephony subscriptions with respect to the prices of the mobile telephony products suggests that mobile and xed telephony are weak substitutes. If the lagged penetration rate increases by 1%, the number of subscribers increases by 0:99%. The estimates of the own and cross price elasticities of demand for the mobile telephony products are larger for Model I than for Model II. However, the elasticities of the market demand of mobile telephony subscription with respect to the prices of the mobile telephony products and the lagged penetration rate are larger for Model II than for Model I.12 5 Merger Analysis The merger of Tmn and Optimus would result in a market with two mobile telephony rms: (i) one controlling the two products previously owned by Tmn and Optimus, to which we shall refer as Tmn?Optimus, and (ii) Vodafone, which would maintain its product. The merger 12The lower the number of locked-in consumers, the larger the share of consumers choosing mobiles, sjt, and the smaller the share of the outside option, s0t. Thus, using equation (6) we get higher own and cross-price price elasticities for the mobile telephony products, but lower elasticities of the market demand for mobiles. 12 consists of change from matrix to matrix m, given by: m = 2 66664 1 1 0 1 1 0 0 0 1 3 77775 : We use the demand estimates in Table 2 to compute the marginal costs given by equation (9), which are presented in Table 3. Then, given these estimates, and replacing with m, we solve system (10) with respect to prices, to estimate the price of each product after the merger. Table 3 presents the estimated prices after the merger for Model I and II under three scenarios: (i) there are no cost e ciencies, (ii) there is 5% reduction in the costs of Tmn and Optimus, and (iii) there is 10% reduction in the costs of Tmn and Optimus. [Table 3] For Model I, after the merger, the prices of mobile telephony increase on average by 7%. The largest increase, 13%, occurs for Optimus, and the lowest, 3%, occurs for Vodafone.13 If the merger generates a 10% reduction in the costs of the merging rms, on average, prices increase by 6%.14 The estimates of the price elasticities are smaller for Model II than for Model I. Therefore, the price increases are larger for Model II than for Model I. For Model II, after the merger, the prices of mobile telephony increase on average by 8%. [Table 4] 13These simulations ignore adjustments in the consumption levels, i.e., in the number of calls and minutes, resulting from price changes. However, the price elasticities of the demands for calls and duration are very small. 14One could also consider the case in which after the merger, the less e cient Optimus is able to provide services with the marginal cost level of Tmn. This case is, however, unlikely, since the merger is hostile and is conducted by Optimus. Besides, it generates implausible results. The price of Optimus decreases by 10% and the price of Tmn increases by 40%. The price of Vodafone remain almost unchanged. This occurs because the post-merger margins are equal for both rms, but the pre-merger margins di er signi cantly. 13 Table 4 reports the impact of the merger on pro ts, consumer surplus, and social welfare, with 95% con dence intervals.15 For Model I, on average, after the merger the consumer surplus decreases by 4%. The pro ts of all rms increase, and on average pro ts increase by 26%. Social welfare decreases by 4%. However, the con dence intervals for changes in pro ts, consumer surplus, welfare are large and include both positive and negative values. For Model II, the decrease in consumer surplus is larger. [Table 5] We use the information on the operating costs of the rms to asses the plausibility of the assumption that rms play a Nash equilibrium. More speci cally, we compare the estimated price-cost margins, pjt?bc n jt pjt , with the observed price-cost margins, revjt?opcjt revjt , where revjt is the revenue and opcjt is the operating cost, respectively, of rm j in period t. Table 5 presents simple statistics on estimated and observed margins. Unfortunately, there are large di erences between these values. This could be due to the de nition of operating costs, which are average accounting costs, rather than economic marginal costs. In particular, the observed margins of Optimus are negative. The observed margins of Tmn are much higher than those of Vodafone. We also estimated the marginal costs under the assumption that rms maximized jointly their prices. In this case matrix consists only of ones. The estimates of all the marginal costs are negative. Recall that the mobile telephony products are very close substitutes, while the outside product is not a close substitute. We take this as evidence that rms do not collude on prices. 15The con dence intervals were computed using a standard bootstrapping method. The covariance matrix of the parameter estimates was used to generate 1000 draws from a joint-normal distribution. For each draw, we computed the percentage change in post-merger pro ts, consumer surplus, and welfare, relative to the pre-merger situation. In this way we got a distribution of potential pro t changes. 14 6 Conclusion In this article we simulated the unilateral e ects of a merger in the Portuguese mobile telephony market. We estimated the price elasticities of demand using a nested logit model and quarterly data from 1999:1 to 2005:4. The mobile telephony subscription products seem to be close substitutes. Using the estimates of the price elasticities of demand and the assumption that rms play a Bertrand game we estimated the marginal costs. We used the estimates of price elasticities of demand and the marginal costs to simulate the unilateral e ects of a merger between Tmn and Optimus. We found that the merger can lead to substantial price increases even with cost e ciencies. In mobile telephony, the complexity of the consumers? behavior and the rms? strategies is very challenging for economic modelling. However, we had only limited aggregate data, which constrained the sophistication of our analysis. In addition, competition authorities or sectorial regulators usually have limited resources, such as time and personnel, to collect data and evaluate the merger e ects. This is a compelling argument for using simple methods. There is scarce evidence of the accuracy of merger simulations, due to the small number of cases in which simulations were conducted. The merger in mobile telephony in Portugal is a natural experiment that, in case of acceptance, can be used to assess whether standard analysis of this type may provide us with accurate predictions about unilateral e ects. 15 Bibliography Anderson, S., De Palma, A. and J-F. Thisse. 1992. "Discrete Choice Theory of Product Di erentiation." MIT Press. Barros, P. and N. Cadima. 2000. "The impact of mobile phone di usion on the xed - link network."CEPR Discussion Paper, no.2598. Berry, S.T. 1994. "Estimating Discrete-Choice Models of Product Di erentiation." RAND Journal of Economics, 25(2), pp.242-262. Crooke, P., Froeb, L., Tschantz, S. and G. Werden. 1999. "The E ects of Assumed Demand Form on Simulated Post-Merger Equilibria." Review of Industrial Organization, 15(3), pp.205-217. Doganoglu, T. and L. Grzybowski. 2006. "Estimating Network E ects in Mobile Telephony in Germany." Information Economics and Policy, (forthcoming). Evans, D.S. and J.J. Heckman. 1983. "Multiproduct Cost Function Estimates and Natural Monopoly Tests for the Bell System." in: Breaking up the Bell, North-Holland, New York, pp.253-287. Gagnepain, P. and P. Pereira. 2006. "Entry, costs reduction, and competition in the Portuguese mobile telephony industry." International Journal of Industrial Organization, (forthcom- ing). Hausman, J. and D. McFadden. 1984. "Speci cation Tests for the Multinomial Logit." Econo- metrica, Vol.52(5), pp.1219-1240. Ivaldi, M. and F. Verboven. 2004. "Quantifying the e ects from horizontal mergers in European competition policy." International Journal of Industrial Organization, 23, pp.669-691. 16 Nevo, A. 2000. "Mergers with di erentiated products: the case of the ready-to-eat cereal industry." Rand Journal of Economics, pp.395{421. Okada, Y. and K. Hatta. 1999. "The Interdependent Telecommunications Demand and E - cient Price Structure." Journal of the Japanese and International Economies, Vol.13(4), pp.311-335. Pinkse, J. and M. Slade, 2004. "Mergers, brand competition, and the price of a pint." European Economic Review, 48(3), pp.617{643. Pereira, P. and F. Pernias, 2006. "The Di usion of Mobile Telephony in Portugal before UMTS: A Time Series Approach," Univesidad de Valencia, unpublished. Pereira, P., P ernias, J. and T. Ribeiro, 2006. "The Determinats of the Di usion of Mobile Telephony." Portuguese Competition Authority. Pereira, P. and T. Ribeiro, 2006. "The Impact on Broadband Access to the Internet of the of the Dual Ownership of Telephone and Cable Networks." Portuguese Competition Authority. Rodini, M., Ward M. and G. Woroch, 2003. "Going Mobile: Substitution Between Fixed and Mobile Access." Telecommunications Policy, 27, pp.457-476. 17 Appendix Table 1: Demand Estimates Model I OLS GMM Estimates std (t) Estimates std (t) ropt 1.141 2.769 ( 0.41) -1.886 0.304 ( -6.20) rtmn -0.656 1.564 (-0.42) -2.153 0.123 (-17.43) rvod 0.984 2.432 ( 0.40) -1.804 0.285 ( -6.32) fi -9.644 4.464 (-2.16) -3.258 1.302 ( -2.50) 1.676 1.081 ( 1.55) 0.864 0.118 ( 7.31) fl 5.645 0.647 ( 8.72) 4.753 0.311 ( 15.27) mse/R-sq Opt. 0.9249 0.59 1.0496 0.51 mse/R-sq Tmn 1.0511 0.57 1.0800 0.52 mse/R-sq Vod. 0.9363 0.63 1.0863 0.53 N*Obj. 72.8092 7.2117 Model II OLS GMM Estimates std (t) Estimates std (t) ropt -0.158 2.395 (-0.07) -2.470 0.186 (-13.25) rtmn -1.539 1.408 (-1.09) -2.671 0.075 (-35.54) rvod -0.141 2.157 (-0.07) -2.323 0.214 (-10.83) fi -9.176 4.409 (-2.08) -3.767 1.489 ( -2.53) 1.322 0.970 ( 1.36) 0.743 0.139 ( 5.34) fl 5.515 0.657 ( 8.39) 4.628 0.321 ( 14.40) mse/R-sq Opt. 0.9604 0.57 1.0453 0.49 mse/R-sq Tmn 1.0249 0.56 1.0807 0.51 mse/R-sq Vod. 1.0103 0.60 1.1062 0.51 N*Obj. 74.8365 8.9926 18 Table 2: Demand Elastictites: Price and Lagged Subscribers Model I Optimus TMN Vodafone Mobiles st Optimus -6.32 1.33 1.33 -0.063 TMN 2.44 -2.61 2.44 -0.170 Vodafone 2.47 2.47 -5.86 -0.108 Net.E ect 0.985 Model II Optimus TMN Vodafone Mobiles st Optimus -3.93 0.75 0.75 -0.088 TMN 1.37 -1.71 1.37 -0.234 Vodafone 1.39 1.39 -3.71 -0.144 Net.E ect 1.365 19 Table 3: Simulation of Post-Merger Equilibrium Prices Model I p.pre mc ef.0% markup% p.post price % Optimus 0.2924 0.2257 0.2283 0.3353 0.1288 Tmn 0.1881 0.1044 0.4450 0.2140 0.1200 Vodafone 0.2939 0.2137 0.2729 0.3056 0.0349 mean 0.2380 0.2566 0.0713 p.pre mc ef.5% markup% p.post price % Optimus 0.2935 0.2144 0.2695 0.3291 0.0985 Tmn 0.1896 0.0992 0.4769 0.2139 0.1042 Vodafone 0.2948 0.2137 0.2753 0.3045 0.0261 mean 0.2389 0.2565 0.0641 p.pre mc ef.10% markup% p.post price % Optimus 0.2923 0.2031 0.3052 0.3195 0.0721 Tmn 0.1888 0.0940 0.5022 0.2103 0.0946 Vodafone 0.2943 0.2137 0.2740 0.3019 0.0196 mean 0.2378 0.2531 0.0565 Model II p.pre mc ef.0% markup% p.post price % Optimus 0.3131 0.2257 0.2794 0.3686 0.1684 Tmn 0.2164 0.1044 0.5174 0.2474 0.1378 Vodafone 0.3178 0.2137 0.3277 0.3317 0.0416 mean 0.2307 0.2490 0.0772 p.pre mc ef.5% markup% p.post price % Optimus 0.3144 0.2144 0.3182 0.3622 0.1448 Tmn 0.2177 0.0992 0.5444 0.2470 0.1310 Vodafone 0.3192 0.2137 0.3306 0.3316 0.0371 mean 0.2314 0.2489 0.0736 p.pre mc ef.10% markup% p.post price % Optimus 0.3115 0.2031 0.3480 0.3493 0.1137 Tmn 0.2139 0.0940 0.5606 0.2402 0.1186 Vodafone 0.3158 0.2137 0.3233 0.3260 0.0304 mean 0.2299 0.2455 0.0664 20 Table 4: Simulation of Post-Merger Equilibrium Pro ts, Consumer Surplus and Welfare Model I Mod.II pre x 1 ? fi = 0:95 % pre x 1 ? fi = 0:95 % Optimus 0.0148 0.0014 [ 0.0000, 0.0051] 9,5% 0.0166 0.0013 [ 0.0002, 0.0033] 7,8% Tmn 0.0303 0.0072 [-0.0000, 0.0254] 23,8% 0.0379 0.0079 [-0.0000, 0.0161] 20,8% Vodafone 0.0272 0.0105 [ 0.0000, 0.0345] 38,6% 0.0312 0.0113 [ 0.0003, 0.0219] 36,2% CS 4.5937 -0.2009 [-0.4994,-0.0000] -4,4% 2.0070 -0.1289 [-0.1983,-0.0038] -6,4% Welfare 4.6661 -0.1819 [-0.4431,-0.0000] -3,9% 2.0927 -0.1084 [-0.1652,-0.0031] -5,2% Optimus 0.0149 0.0039 [ 0.0007, 0.0093] 26,2% 0.0170 0.0033 [ 0.0023, 0.0067] 19,4% Tmn 0.0309 0.0077 [-0.0001, 0.0283] 24,9% 0.0384 0.0085 [-0.0000, 0.0153] 22,1% Vodafone 0.0273 0.0085 [-0.0069, 0.0359] 31,1% 0.0317 0.0102 [-0.0004, 0.0201] 32,2% CS 4.4662 -0.1566 [-0.5112, 0.1425] -3,5% 1.9921 -0.1170 [-0.1849, 0.0080] -5,9% Welfare 4.5392 -0.1364 [-0.4438, 0.1394] -3,0% 2.0792 -0.0949 [-0.1462, 0.0122] -4,6% Optimus 0.0143 0.0068 [ 0.0035, 0.0179] 47,6% 0.0165 0.0058 [ 0.0040, 0.0147] 35,2% Tmn 0.0306 0.0078 [-0.0003, 0.0276] 25,5% 0.0371 0.0087 [-0.0001, 0.0150] 23,5% Vodafone 0.0273 0.0067 [-0.0095, 0.0311] 24,5% 0.0308 0.0084 [-0.0021, 0.0168] 27,3% CS 4.3214 -0.1187 [-0.4569, 0.2056] -2,7% 2.0642 -0.0986 [-0.1626, 0.0384] -4,8% Welfare 4.3937 -0.0973 [-0.3979, 0.2016] -2,2% 2.1487 -0.0757 [-0.1267, 0.0473] -3,5% Table 5: Observed and Estimated Price-Cost Margins mean std min max obsopt -0.137 0.160 -0.71426 0.08068 obstmn 0.223 0.068 0.03140 0.34624 obsvod 0.071 0.076 -0.16373 0.15950 estopt 0.160 0.022 0.13518 0.24609 esttmn 0.390 0.053 0.30918 0.46351 estvod 0.173 0.023 0.14230 0.23966 21