Effects of Mobile Number Portability on Switching Costs: Japanese Mobile Telecommunications March 2011

Abstract This paper assesses the effects of mobile number portability(MNP) introduced in the Japanese mobile telecommunications in 2006 on switching costs. Based on the two-stage nested logit model of mobile carrier choice and MNP usage choice, we show that MNP reduces the switching costs by 18% on average and increases the fraction of consumers switched the carriers by 2.6%. Keywords: Mobile Number Portability(MNP); Switching Costs; Nested Logit Models; Choice-based Sampling JEL Classification Numbers: D12; L96

1

1

Introduction Mobile Number Portability(MNP) was introduced in Japanese mobile telecommunica-

tions market in October 2006. MNP allows mobile users to retain their numbers when changing mobile from one mobile carrier to another and hence it should reduce switching costs in the choice of the carriers. From a theoretical perspective, reducing switching costs do not always make a market more competitive because firms have incentives to set monopolistic prices to their existing users, while they have incentives to offer competitive prices in order to attract new users.1 Although firms face with this trade-off in the presence of switching costs, increasing mobility among carriers might be an effective measure to induce competition in the Japanese market because most of the individuals, around 100 million out of 120 million, owned their mobile phones in 2006, and thereby there was little room to enclose new users. Therefore, the introduction of MNP aims at induce more competition in the Japanese mobile telecommunications and the Japanese government is still seeking other policies, such as the removal of Sim Lock that allows mobile carriers to restrict the use of mobile phones to their own, to reduce switching costs in the market.2 With increasing interest in the policy instruments to reduce switching costs, the demand for assessing the effects of the policies on the switching costs has emerged. Therefore, the purpose of this paper is to quantify the effects of MNP on the consumer switching in the Japanese mobile telecommunications market. To implement the analysis, we collect the individual data on the mobile phone usage and MNP adoption by web questionnaire in March 2007. Since only a fraction of individuals switched the carriers after the MNP introduc1

See Klemperer (1995) and Farrel and Klemperer (2008) for a survey on switching costs. This is not a unique story about the consequence of reducing switching costs on the market competitiveness. For example, Klemperer (1987) shows that reducing switching costs does not make a market more competitive even in the presence of a large potential market, i.e., a large population of new users that can be potentially enclosed. Klemperer (1987) considers a situation in which consumers heavily takes the future utility into account. Under the situation, they do not want to choose a firm that offer lower prices because they understand that the firm that successfully acquires market for existing users has stronger incentives to set their price higher. Indeed, recent study, Viard (2007) examines the effects of 800-number portability in US and shows that the reducing switching costs induced competition despite the presence of large potential market. Although this paper does not focus on the competitiveness of the market, whether MNP made market more competitive or not is an important area of research. 2

2

tion, we employ the choice-based sampling that intended to collect a reasonable amount of switched individual samples; therefore, we employ the weighted exogenous sampling maximum likelihood(WESML) by Mankski and Lerman (1977) that provide us with a consistent estimate in the presence of choice-based samples. In order to estimate the switching costs, we construct a model accounting for an interesting characteristic shown in our dataset: the significant fraction of switched consumers did not apply MNP despite the presence of MNP choice. For this reason, our model incorporate the endogenous choice of the MNP adoption. To be more precise, we employ a two-stage nested logit model in which consumer chooses one of mobile phone carriers at first stage, and choose whether to apply MNP at second stage if she switches the carrier. Using the dataset and the model, we analyze the impact of the past choice on the current choice in order to reveal the switching costs, while we analyze the effect of the availability of MNP choice in order to reveal the effects of MNP on switching costs. The findings of the analyses are summarized as follows. First, the switching costs in the Japanese mobile telecommunications market is around 2000-2300 Japanese yen and they are reduced by 18% by introducing MNP. Second, the benefit from MNP is around 25-35 yen on average, while there is a large variation in the benefit among individuals because MNP is beneficial only for those who switches the carriers. Third, the MNP increases a switching probabilities by around 2.5%.

1.1

Related literature

The effects of MNP on consumer switching costs in mobile telecommunications market have been studied in the literature. For example, Lee, Kim, Lee, and Park (2006) study the MNP introduction in Korean mobile phone market and Ida and Kuroda (2009) in Japanese market using the individual choice on mobile carrier switching. Although this paper also uses individual choice data on mobile phone carrier switching, it has an important difference in measurement of the effects: while both of the paper uses state preference method in the anal3

ysis, this paper employ the revealed preference method that is known to be more appropriate in studying the consumer behavior. Under the stated preference method, the respondents do not have an appropriate incentive to reveal their preference because respondents choices does not have influences on their utility. On the other hand, we can estimate the consumer behavior more precisely because their choices actually affects their utility under revealed preference method. Note that there is one exception, Kim (2006) that estimates the effects of MNP on switching costs based on the dynamic environment in Korea. The model used in Kim (2006) is based on the revealed preference, but the paper used in the estimation is market level data such as market share for each mobile carriers, while this paper uses individual data and hence the analysis in this paper allows to reveal the difference in switching behavior depending on the individual characteristics. Other than the Mobile telecommunications market, several empirical studies analyze the role of switching costs based on the revealed preference approach using individual data. For example, Chen and Hitt (2002) studies the switching costs in on-line brokerage, Shum (2004) estimates switching costs in the analysis of the effects of advertising on the consumer switching in US cereal market, and Goldfarb (2006) estimates switching costs at internet portals. The other strand of literature that quantify the switching costs applies firm behavior models to recover the switching costs (Shy (2002); Kim, Klinger, and Vale (2003)). These studies compute the switching costs based on the equilibrium pricing equation derived from a particular model in the presence of switching costs. In the telecommunications market, as in many of the countries, mobile carriers provide a large number of calling plans including many discount plans. Therefore, we avoid the firm behavior approach because of the difficulty in specifying a model that generates an equilibrium consistent with the complex tariff scheme as in actual.

4

1.2

Organization of the paper

This paper is organized as follows. In the following section, we introduce the Japanese mobile telecommunications market and the data used in this study. Section 3 introduce the discrete choice model of mobile phone carrier choice and MNP application choice. Section 4 discuss the identification issues of estimating switching costs and shows the estimation results. Based on the estimates in section 4, we implement several simulations to reveal the effects of MNP on switching costs, the probability of consumers switching, and consumer welfare. Section 6 concludes the paper with caveats of the interpretation of the results presented in this study.

2

Data We here introduce the data used in this paper and do preliminary data analysis. Be-

fore turning to the introduction of the data, we first briefly introduce the Japanese mobile telecommunications market.

2.1

Mobile telecommunications market in Japan

Japanese mobile telecommunications market has three dominant carriers, NTT DoCoMo, au and Softbank: hereafter we call them as docomo, au, and softbank, respectively. Among them, docomo has been largest since the emergence of the mobile telecommunications market in Japan, and its share was 55% in October 2006. On the other hand, au was the second and softbank followed in 2006, and their shares were around 29% and 16%, respectively.3 MNP was introduced in October 2006, as a results of continuous discussion in the government since 2003. Soon after the introduction of MNP, the Japanese mobile telecommunications market experienced revolutionary change because the largest carrier, docomo,

3

The data on the share of Japanese mobile carriers are available at the web site of Telecommunications Carriers Association.

5

decreased the number of its subscribers from 52143700 to 52126200 for the first time in its history. Although docomo has increased the number of subscribers afterwards, it decreased the share from 55 to 53%, while au and softbank increased its share from 29 to 30% and from 16 to 17%, one year after the introduction.

2.2

Web questionnaire

To investigate the effect of the MNP, this paper uses the individual data on mobile phone usage and the carrier choice collected by the web questionnaire conducted in March 2007. While MNP should increase the number of consumers switched carriers, those who switched carriers were still a few. Therefore, we employ a choice-based sampling that put weight on the consumers who switched after MNP in order to reveal the individual characteristics on the decision of MNP usage. As a result, we obtain 531 out of 1578 samples on those who switched carriers between the introduction of MNP and the implementation date of web questionnaire, i.e., within 6 months. In our survey, we further put weight on the sample distribution to avoid the bias in the distribution of consumer characteristics. The choice of weight is described in Table 1. We choose the weight in order to match the distribution of individual characteristics in the samples with the actual distribution of the mobile phone users.

2.3

MNP usage

Our dataset shows that not all the individual applied the MNP even after the introduction of MNP; as shown in Table 2, the ratio of consumers who applied MNP is 64% and hence significant amount of consumers did not apply the MNP when switching carriers. We asked why the individuals did not use the MNP in the questionnaire. The results indicate that those who switched carriers without using MNP pointed out the cost regarding the process of the MNP as the reason why they did not apply MNP: 60 out of 192 respondents for the application fee (5000 yen), 30 respondents for the burden of processing. In addition to these 6

costs, 24 respondents answered that they just wanted to change their numbers. This preliminary analysis reveals the taste heterogeneity on the MNP. Therefore, we take this feature into account in the following analysis. As we will explain, the taste heterogeneity on MNP is a key in identifying the effect on the switching costs.

3

Model Our dataset reveals that not all the switched consumers applied MNP after the introduc-

tion of MNP. Therefore, we introduce a mobile carrier choice model that incorporates the consumers’ MNP usage choice. The model used has two stages: at the first stage, consumers choose the carriers from docomo, au and softbank and at the second stage, the consumers chooses MNP usage if they switch the carriers. Note that the choice set each consumer faces depends on the consumers’ current carriers; for example, as represented in Figure 1, for consumers who currently subscribe docomo, the MNP usage choice exists only when she switch to au or softbank. The choice set for consumers who subscribe au and softbank can be defined in a similar fashion. To represent the structure of the model, we apply a two-stage nested logit (NL) model. The model is helpful in this application because it enables us to quantify the benefit from MNP introduction analytically. We consider a consumer ih who currently use the carrier h. Hereafter, we suppress h of ih to simplify the notation. Consumer i’s utility from choosing mobile carrier j is ] [ Ui,j,M N Pij = (α0 + xPi α1 ) · pij + β0 + xSi β1 + (γ0 + xM i γ1 ) · M N Pij · SW IT CHij (1)

+ xij δ + ϵi,j,M N Pij (λ) = Vij (θ) + Vi,j,M N Pij (γ) + ϵi,j,M N Pij (λ),

7

where Vij (θ) ≡ (α0 + xPi α1 ) · pij + (β0 + xSi β1 ) · SW IT CHij + xij δ, [ ] Vi,j,M N Pij (γ) ≡ (γ0 + xM i γ1 ) · M N Pij · SW IT CHij .

(2)

SW IT CHij takes 1 if j ̸= h and takes zero otherwise, and M N Pij takes 1 if consumer i apply MNP for j ̸= h, and zero otherwise. Note that the consumer i never faces the MNP usage choice and hence M N Pij takes always zero if she continues to use the current carrier or j = h; then, we denote the utility simply as Uih . xPi , xSi and xM are the vectors of i consumer i’s characteristics variables that affect the price sensitivity, switching behavior, and MNP usage, respectively. The variables includes Music, Game, and Wallet that are dummy variables for the consumer i’s music, game, and electronic money function usage, respectively, Student that takes one if consumer i is student4 , Allowance that is consumer i’s monthly allowance, and Address that is the number of registered address in her mobile phone. xij is the cross term between dummy variables for mobile carriers and consumer characteristics variables. In this paper, we treat docomo as reference and thereby we have the dummy variables for au and softbank choices, denoted as au and softbank, respectively. pij is the charge when consumer i chooses carrier j. Note that the charge depends on a billing plan that determines basic charges, dialing and data communication charges, etc. Moreover, each carriers offer several discount plans such as family discount plan that discounts basic monthly charges for every members of the family who subscribe the same carrier, in-network discount plan that allows one to make a phone call to consumers who subscribe the same carrier at a discount rate, and so on. Therefore, although we have information on the billing information of the carrier that consumer i currently subscribe, it is difficult to reveal how much she will be charged if she subscribe different carriers. One possible way to resolve this problem is to apply a discrete-continuous model proposed by Hanemann (1984). The model allows us to incorporate not only the discrete choice 4

The student here includes pupils and the junior high school, high school and college students.

8

but also the continuous choice such as the minutes of calls and the volume of packet use, from the same utility maximization decision. However, the method is difficult to apply to the telecommunications market because of the presence of too many complex billing plans. Therefore, we compute the charges for each mobile carrier in the following way. First, we assume the minutes of calls and the volume of data communication for each consumer are the sames across the carriers. Second, using the information on the minutes of calls and the volume of data communications for each consumer, we compute the charges for all available plans for each carriers. Third, we choose the lowest charges for each carrier and set them as pij . ϵi,j,M N Pij follows a Generalized extreme value (GEV) depending on a parameter λ, which give rise to the choice probability introduced in the following section. λ represents the correlation across nests and, as McFadden (1978) shows, to be consistent with the utility maximization problem, λ should takes the value between zero and one. We will confirm whether the condition is met in this application. Finally, θ = (α0 , α1 , β0 , β1 , δ0 ) and γ = (γ0 , γ1 ) are the parameters to be estimated. Note that θ and γ are related to the mobile carrier choice at the first stage and the MNP usage at the second stage.

4

Estimation We here turn to an estimation procedure. From the property of NL model, we can write

the consumer ih ’s choice probability of mobile carrier j ̸= h as the product of the probability on the MNP usage choice conditional on the choice j, and the marginal choice probability of the mobile carrier j:

Pi,j.M N Pij (θ, γ, λ) = Pij (θ, γ, λ)Pi,M N Pij |j (γ, λ),

9

(3)

where Pij (θ, γ, λ) = and

exp (Vij (θ) + λIij (γ, λ)) ∑ exp (Vih (θ)) + l̸=h exp (Vil (θ) + λIil (γ, λ))

(4)

( ) exp Vi,j,M N Pij (γ)/λ ( ). Pi,M N Pij |j (γ, λ) = ∑ M N Pij ∈{0,1} exp Vi,j,M N Pij (γ)/λ

(5)

Iij represents the expected value of MNP usage that is expressed as  Iij (γ, λ) = ln 

∑

(

)



exp Vi,j,M N Pij (γ)/λ  (6)

M N Pij ∈{0,1}

[ ( )] = ln 1 + exp (γ0 + xM i γ1 ) · SW IT CHij /λ If the consumer ih chooses the same carrier h, the choice probability is expressed as:

Pih (θ, γ, λ) =

exp (Vih (θ)) ∑ exp (Vih (θ)) + l̸=h exp (Vil (θ) + λIil (γ, λ))

(7)

Since the data used in this paper is collected by the choice-based sampling, the ordinary maximum likelihood generates biased estimates. In this study, we employ the weighted exogenous sampling maximum likelihood (WESML) proposed by Mankski and Lerman (1977), which allows us to obtain consistent estimators in a simple way. To implement WESML, we first define the following weighting function:     1−QS wij =

1−HS

   QS

if j = h, (8) otherwise

HS

where HS and QS are the fraction of consumers who switched their carriers in population and sample, respectively. Then, we define the following weighted log-likelihood:

W LL(θ, γ, λ) =

∑∑ h

i∈Nh

 wih yih ln Pih +

∑

∑

j̸=h M N Pij ∈{0,1}

10

 wij yi,j,M N Pij ln Pi,j,M N Pij 

(9)

where Nh is the set of consumers whose current carrier is h, yih and yi,j,M N Pij are dummy variables for consumer ih ’s choice. The parameters in the model can be estimated by maximizing this weighted log-likelihood. Note that in order to implement the WESML, we need the figure on the fraction of consumer switched the carriers, HS , that is not directly observed to us. In this paper, we compute it based on the information on the number of consumers who used MNP until March 2007 and the ratio of consumers who applied MNP to total switched consumers that can be computed from our dataset. When we implement the questionnaire, the number of MNP users is about 2 million that is amount to the 2% of total mobile phone subscribers. Note that since this is based on the figure only a half year after the introduction of MNP, those who switched a carrier should increased afterwards. We here consider that each consumer make a carrier choice when she updates her mobile phone handset because the carrier choice always incorporate the handset choice. In Japanese telecommunications market, consumers usually update their mobile handsets every two years; therefore, we assume that each consumer make a carrier choice every two years. Therefore, if the number of MNP users increased proportionately, the fraction of MNP users is amount to 8% of total mobile phone subscribers. Combining the ratio of MNP users in our dataset, about 36%, HS can be computed as 8% multiplied by the inverse of the ratio. However, those who uses MNP might be concentrated on soon after the introduction of MNP and thereby those who applies MNP should not increase proportionately. To take this issue into account, we consider the case of 6%, in addition to 8%.

4.1

Identification

Before reporting the estimation results, we here discuss what variation of our dataset allows us to identify the coefficients of interest. In the discussion, we refer to summary statistics reported in Table 4. The table lists the statistics for all samples, samples without switch, samples with switch and samples with MNP usage. 11

Switching costs The switching costs depends on the coefficients on SWITCH and the coefficients on Charge, consumers’ sensitivity on charges, i.e. the coefficients on Charge. The coefficient on SWITCHij should takes the negative value if the consumer tends to choose the same carrier even though its charge is higher than the other carriers. Therefore, the coefficients are identified from the choice between current and past carriers, and the consumer level variations in the charges among carriers. On the other hand, the coefficient on Charge should takes the negative value if consumers tend to choose the carriers that provides lower charge; therefore, it is identified from the variation in charges and the consumers’ current choices. Since both of the coefficients are identified from the similar variation, it is worth discussing what variation in our dataset allows us to separately identify SWITCH and Charge. In our dataset, we find that a significant portion of the consumers chooses the carriers that did not offer the lowest charges. The observation is consistent with the presence of switching costs, but it might be due to the consumer insensitiveness on charges. To confirm which factors lead to this consumer choice tendency, we look at the difference between the charges offered by past carrier and the lowest charges among carriers. The difference is about 382 yen on average for the consumers who do not switch the carriers, while it is about 759 yen for the consumers who switched; therefore, the consumers switched the carriers if they can save large money by switching, while they choose to stay if they can save only a little. The casual observation implies that the consumers are not insensitive to charges, but the presence of switching costs that restrict the mobility of consumers among carriers. State dependence v.s. heterogeneity The problem in identifying the switch cost is, as Heckman (1981) discussed, the presence of consumer heterogeneity in taste for each alternative. The presence of heterogeneity bias the estimated switching costs upwardly because choosing same carrier may be attributable to the unobservable consumer taste on the carrier rather than the presence of switching costs. Note that the switching costs are identified from the effect of past choice, i.e. so called state 12

dependence for each consumer; therefore, this identification problem is sometimes called spurious state dependence. The usual way to distinguish the switching costs from the heterogeneity is to use the repeated consumer choices. However, it is difficult to obtain this kind of data in mobile telecommunications market because consumers do not face with a carrier choice less frequently.5 Unfortunately, we do not have an adequate way to deal with this problem, and hence we do not take account of the effect of heterogeneity in our estimation. As a result, the switching costs estimated in this paper should be biased upwardly. However, this paper aims at the introduction of MNP whose effects on switching costs are revealed conditional on the carrier switching; therefore, we consider that our simulation results on the reduction rate of the switching costs are free from the spurious state dependence problem. Endogeneity in charge The other problem in identifying switching costs is an endogeneity problem in Charge: as is well-known problem in the demand estimation, the assumption may be problematic because there should be the characteristics and demand shocks observable to consumers but unobservable to researchers. One of the way to remove the effect of the unobserved factor is to include the fixed effect for each alternative in the estimation. Our dataset is individual consumer choice and hence exploiting this variation we can include the carrier specific dummy variables.6 Note that there may be unobserved characteristics for specific consumer groups, which can not be controlled though the carrier fixed effects. In reality, the carriers had the strategy to provide services for specific group of consumers. However, such services are usually discount services such as student discount, that are taken into 5

See Keane (1997), for the analysis of state dependence and heterogeneity in frequently-purchasedconsumer-goods market . 6 For the demand estimation using market level data such as Berry, Levinsohn, and Pakes (1995), each observation is choice level, i.e. sales, price and characteristics for each automobiles. Therefore, the choice specific dummy variables can not be included in the estimation. Rather, Berry, Levinsohn, and Pakes (1995) employ a instrumental variable methods to deal with the endogeneity problem. Note that it is difficult to implement the estimation using a similar instrument used in Berry, Levinsohn, and Pakes (1995) because the instrument vary only across carriers but not across individuals. Our identification strategy is similar to Hausman (1997) and Nevo (2001) that propose ways to identify the price coefficients in the presence of the choice specific fixed effect.

13

account in the construction of the Charge variable. Therefore, we consider the unobserved characteristics and demand shocks are adequately controlled by including the carrier fixed effects and thereby estimate the model assuming the error term ϵi,j,M N Pij to be uncorrelated with the charges in the estimation. Effects of MNP The effects of MNP is identified from the MNP usage choice of those who switched carriers. As we discussed in the previous section, the consumers face with the transaction costs when they apply the MNP, while it allows them to switch carriers without changing their number has been used. As shown in the table 2, the fraction of consumers who used the MNP is about 64%, more than half of switched consumers, and thereby the coefficient γ0 should takes the positive value. Note that, we have the data on individual switching behavior only after the introduction of MNP; therefore, the change in switching probability before and after the introduction do not contribute to identifying the effects of MNP. Ideally, we should investigate the data on switching behavior not only after but also before the introduction that should generates more precise estimates on MNP. In addition, we take account of the difference in taste for MNP among individuals. In particular, we include the student dummy variable and the variable of the number of address registered in the mobile phone in the second stage. The choices of the variables is because: the students have little business relationship and hence the cost of changing the number should be smaller; the cost of changing the number should larger for those who have a larger number of addresses because they have to inform their new phone number of a larger people. As shown in table 4, the fraction of students used MNP is actually merely 10% within the sample with switching, while it is about 15% in all sample. Similarly, Address is larger within MNP users.

14

4.2

Estimation results

We here report the estimation results. As is mentioned, we estimate two cases of weight, 8% and 6% in addition to the case with no weight. The estimation results is summarized in Table 4. First, we report the results of the first stage. As shown in the table, the coefficient on Charge is negative and significant: the consumers tend to choose the carriers that offers lower charges. Although the cross term of charge and allowance and its square is not significant except for the case of no weight, the results are economically significant because the results indicate that consumers become less price sensitive as an increase in their allowance. The coefficient on SWITCH is negative and significant for all cases but it depends on the weight chosen. In particular, it takes larger value for small weight and hence the coefficient in case (ii) amounts to 1.5 times in case (iii). We include the cross term of SWITCH and music, game and wallet function dummy variables because those who use of these function have to incur additional costs in switching carriers. The coefficients takes the positive value for all specifications and hence reasonably estimated though only some of them are significant. The au and softbank dummy variables and their cross term captures the differences in consumers’ perception on the quality of carriers in comparison with docomo. The results indicate that the coefficient on softbank is significantly negative, which indicate that consumers consider softbank as low quality carrier. This result may be attributable to the narrow coverage area of softbank. The coefficient of the product of au and Music is positive and significant for two of them, while the coefficient on au×Game and softbank ×Game are negative for all cases and significant for most of them: these results are consistent with the discussion in Ishikawa (2006) that indicates the excellence of the music distribution service “LISMO” offered by au and the game distribution service “i-appli” offered by docomo.7

7

Note that we treat the usages of music, game and wallet service as consumer characteristics; however, it should be a choice for consumers and hence we should construct the model that incorporate the choice of these service usage. However, it requires us to include too many choices, which makes it difficult to implement the estimation using our dataset.

15

Next, we explain the estimation results of the second stage. All the coefficients reported in the table are not significant except for the case of no weight. These results may be because WESML does not provide efficient estimates; in particular, the standard error gets larger when the weight is small as in our application.8 Therefore, to take this problem into account, we report the standard error of the simulation results in the following section. Though the estimates are not significant, the value of the estimates lies in a economically significant range. In particular, as discussed earlier, the estimate of λ should take the value between zero and one to be consistent with utility maximization problem. (McFadden (1978)) Note that the coefficient on constant is positive and significant, which means that consumers value the MNP on average. The effect of the number of registered addresses are positive, which implies that those who has larger networks obtains larger benefit from the MNP usage. Finally, the results also indicate that the students do not care their number change so much.

5

Simulation Using the estimates obtained in the previous section, we now implement simulations to

measure the switching cost for each consumer and the effect of MNP on the switching costs. We further quantify the effects of MNP introduction on welfare and the switching probability.

5.1

Switching costs

We first estimate the switching costs using the estimates shown in the previous section. Before turning to the analysis, we explain the meaning of switching costs obtained here. According to the classification in Klemperer (1995)9 , the switching costs includes the discount 8

As Greene (2008) notes, “WESML is not a free lunch”. To overcome this problem, Cosslett (1981) and Imbens (1992) provide ways to obtain consistent estimates in the choice-based samples. These method require more complex procedure to obtain the estimates and hence we will try them in the future research. 9 Klemperer (1995) classifies the switching costs into five categories: (1) Need for compatibility with existing equipment, (2) Transaction costs of switching suppliers, (3) Costs of learning to use new brands, (4) Uncertainty about the quality of untested brands, (5) Discount coupons and similar device

16

service such as continuous subscription and family discount. The switching costs estimated in this section, however, do not include what relates to the discount because the effect of discount is captured through the construction of charge variables. Therefore, the switching costs presented here relate to psychological costs of changing the number and new e-mail address, and the processing the MNP. We now explain the procedure to simulate the switching costs before and after the MNP introduction. First of all, we define the choice probability of carrier j when MNP is unavailable: exp (Vij (θ)) P¯ij (θ) = ∑ l exp (Vil (θ))

(10)

We define the switching cost sij as the value of charges that allows consumer i to recover the utility when switching carriers. Then, the switching cost sij is represented as the following formula: NP P˜ (pij − sM , SW IT CH = 1) ≡ P¯ (pij , SW IT CHij = 0) ij

(11)

Note that since pij represents the monthly charge, the measurement of switching costs is in monthly charges. From equation (11), the switching costs can be expressed as follows:

sij =

1 (β0 + xSi β), αi

(12)

where αi is a price sensitivity for consumer i, αi = α0 + xPi α1 .

(13)

As shown in equation (12), our model assumes that the switching cost is independent of the carrier chosen. This assumption might seems to be unrealistic because it is known that the carriers compete in enlarging the switching costs after the MNP introduction; for example, softbank introduced the discount services that allowed docomo and au subscribers to carry forward their continuous subscription discount to the softbank in order to reduce 17

the switching costs when moving to softbank. However, as is mentioned, we control for such effects in the construction of the Charge variable. We next turn to the analysis of the effects of MNP on the switching costs. The role MNP in the choice probability is represented by the term λIij in equation (4), which corresponds to the expected benefit from the MNP usage. In this paper, we define the switching cost NP under MNP, sM , to be satisfies: ij

NP P˜ (pij − sM , SW IT CH = 1) = P¯ij (pij , SW IT CHij = 0) ij

(14)

where,

NP P˜ (pij − sM , SW IT CH = 1) = ij

exp (Vij (θ, SW IT CHij = 1) + λIij ) ∑ . exp (Vij (θ, SW IT CHij = 1) + λIij ) + l̸=j exp(Vil (θ)) (15)

NP The equation (14) implies that sM is defined as the value that allows consumer i to recover ij

the utility cost of switching to carrier j. Then, the effects of MNP on the switching costs can be computed as:

NP sM − sij = ij

] 1 [ λ ln(1 + exp(γo + xM i γ1 )/λ) . αi

(16)

Note that the difference is also the same for all carriers. Table 5 reports the simulation results of the switching costs and the effects of MNP on the switching costs for the cases (i) - (iii) in Table 4. As shown in the table, the estimated switching costs are significantly different among the cases, 1145 yen for the case (iii) an 2328 yen for the case (i). The difference is due to the fact that the larger weight on the number of switched consumers results in the smaller switching costs. As we mentioned, though we do not know the exact weight, it should lies between 6 to 8%. Therefore, in this paper, we consider between 2057 and 2328 yen as a reasonable range of the switching costs for average consumers. Intuitively, the results can be took it that the consumers do not switch the

18

current carriers even though the competing carrier offers the charges less than the current carriers by 2057 to 2328 yen. Note, however, that since our estimation does not control for the heterogeneity in taste, these estimated switching costs should be overestimated. However, as is mentioned, since the effect of MNP is conditional on switching, the reduction rate should be correctly estimated. Note also that the standard errors of the parameter estimates in table 4 used to compute the simulation are not small. In the table, we report the standard error of the switching costs and the effects of MNP using delta method. As shown in the table, the standard errors are large in the estimates of the benefit from MNP; indeed the cases (i) and (ii) are statistically insignificant. While the table reports the average switching costs and the effect of MNP, Figure 2 and 3 reports the distribution of the switching costs and the effects of MNP. These figures show that there is a large difference in switching costs and the effect of MNP among consumers. The difference is due to the difference in the characteristics of consumers as shown in the estimation results. For example, those who use music, game, wallet function has a higher switching costs, and the students obtain a small benefit from MNP, while those who has a larger number of registered address obtain a larger benefit. In particular, those who has high allowance has a higher switching costs because they do not care the charges so much and hence require a large compensation in changing the carriers. For the same reason, these consumers should have large benefit from the MNP introduction.

5.2

Consumer welfare

Next, we simulate the effects of MNP on consumer welfare. Before turning to the analysis, we first explain the relationship between the effects on the switching costs and consumer welfare. Obviously the effects of MNP on the consumer welfare appear through the reduction of the switching costs: those who does not switch the carrier do not benefit from the MNP. Therefore, most of the consumers do not gain any benefits from the MNP because only 19

fraction of the consumers switched the carriers. Compared to the effects of MNP on the switching costs in the previous section, since they can be interpreted as the gains of consumer welfare conditional on all the consumers switched the carriers, the benefit from the MNP should be much smaller than them. To be precise, the gain of consumer welfare takes the value around the switching probability times the reduction of switching costs. As shown in Small and Rosen (1981), the gains of consumer welfare for consumer i can be derived as follows. [ ( ) ( )] ∑ ∑ 1 ∆E(CSi ) = ln exp(Vih (θ)) + exp(Vil (θ) + λIil (γ, λ)) − ln exp(Vij (θ)) αi j l̸=h (17) Using this equation, we compute the welfare gains based on the weight of 8%, 6% and no weight. The simulation results is summarized in table 6. Since the welfare gain depends on the switching probability, it takes the larger value as the weight gets larger. For the weight from 6% to 8%, the welfare gains are between 25 and 35 yen for average consumers. Note that similar to the discussion made in the previous section, the standard error of the welfare gains are also non-negligible. The standard errors reported in the table are computed by the delta method. The results indicate that though the z-value of the estimates exceeds unity, they are still statistically not significant at 10% level for the cases (i) and (ii). Therefore, we have to notice that the results present here have some errors. Again, the gains of consumer welfare also vary from consumer to consumer. Figure 4 and 5 show the distribution of the welfare gains under the weight of 8% and 6%, respectively. The figures reveal, as predicted, that the most of consumers do not benefit from the MNP. This is because most of consumers do not switch the carriers even after the introduction of the MNP. However, there are some consumers earned large benefit, 390-420 yen for 8% weight and 360-390 yen for 6%weight, from the MNP.

20

5.3

Switching probability

To understand more the role of MNP in consumer welfare, we here classify the consumers into three types: those who never switch carriers despite the availability of MNP; those who switch carriers despite the availability of MNP; those who switch carriers only when the MNP is available. Note that the first type of consumers never benefit from MNP, while the second type of consumers do benefit from MNP to the amount that the reduction of the switching costs shown in the previous section. The third type of consumers obtain the highest utility from the past carrier in the absence of MNP, while they can increase the utility by switching in the presence of MNP; in other words, their utility obtained from the different carrier minus the reduction of switching cost does not reach the level of the utility obtained from the past carrier. Therefore, the welfare gains for the third type of consumers lies between zero and the reduction of the switching costs. Table 7 reports the effects of MNP on switching probability. For the cases of 8% weight and 6% weight, the switching probabilities increase by 2.77% and 2.42%, respectively. In terms of the classification introduced before, these value corresponds to the share of the third type consumers. Similarly, the switching probability without MNP implies the share of the second type consumers and the rest are the share of the first type consumers. As we mentioned, for the case (i), 86.98% of consumers gains no benefit from MNP, 10.25% of consumers gains from the MNP to the values correspond to the reduction of switching costs, and 2.77% of consumers gains from the MNP to the value between zero and the reduction of switching costs.

6

Conclusion This paper examines Japanese mobile telecommunications market to measure the switch-

ing costs and the effects of MNP on the reduction of the switching costs. Using the individual

21

data on the mobile phone usage and switching behavior, we construct and estimate the twostage nested logit model that incorporates a mobile carrier choice at the first stage and an MNP usage choice at the second stage. The estimation results reveal that the switching cost in the Japanese mobile telecommunications market is about 2000-2300 yen in monthly charge unit before the introduction of the MNP, and the MNP decreases the switching costs by 18% on average. In addition, we show that the MNP increase the consumer welfare by 25-35 yen and the switching probabilities by about 2.6%. Japanese telecommunications market has already matured in the sense that most of the consumer has their mobile phones. Therefore, the reduction of switching costs should contribute to enhance the competition among mobile carriers. Finally, we note the caveat on the robustness of our results. First of all, this paper do not take account of the role of heterogeneity that is another reason why consumers do not switch. Ignoring heterogeneity makes the switching costs in this paper to be overestimated. Next, our estimates indicate a large standard error and thereby the simulation results quantified in this paper have significant error. This may be because the data used in this study is single cross section after the MNP introduction and hence does not capture the changes in switching probability before and after the introduction in identifying the switching costs.

22

References Berry, Steven, James Levinsohn, and Ariel Pakes. 1995. “Automobile prices in market equilibrium.” Econometrica 63 (4):841–890. Chen, Pei-Yu and Lorin M. Hitt. 2002. “Measuring switching costs and the determinants of customer retention in internet-enabled businesses: a study of the online brokerage industry.” Information Systems Research 13 (4):255–274. Cosslett, Stephen R. 1981. “Efficient estimation of discrete choice models.” In Structural Analysis of Discrete Data with Applications, edited by Charles Manski and Daniel McFadden. Cambridge: MIT Press. Farrel, Joseph and Paul Klemperer. 2008. “Coordination and lock-in: competition with switching costs and network effects.” In Handbook of Industrial Organization vol. III. Amsterdam: North-Holland. Goldfarb, Avi. 2006. “State dependence at internet portals.” Journal of Economics & Management Strategy 15 (2):317–352. Greene, William H. 2008. Econometric Analysis. Prentice Hall, 6th ed. Hanemann, W. Michael. 1984. “Discrete/continuous choice models of consumer demand.” Econometrica 52 (3):541–563. Hausman, Jerry. 1997. “Valuation of new goods under perfect and imperfect competition.” In The Economics of New Goods, edited by Timothy F. Bresnahan and Robert J. Gordon. Chicago: National Bureau of Economic Research. Heckman, James. 1981. “Heterogeneity and state dependence.” In Studies in Labor Markets, edited by Rosen. Chicago: University of Chicago Press. Ida, Takanori and Toshiumi Kuroda. 2009. “Discrete choice analysis of demand for mobile telecommunication service in Japan.” Empirical Economics 36 (1):65–80. 23

Imbens, Guido W. 1992. “An efficient method of moments estimator for discrete choice models with choice-based sampling.” Econometrica 60 (5):1187–1214. Ishikawa, Tsutsumu. 2006. Web 2.0 Jidai no Keitai Senso. Kadokawa-Shoten. In Japanese (English translation: Competition in the Japanese mobile telecommunications market in the era of web 2.0). Keane, Michael P. 1997. “Modeling heterogeneity and state dependence in consumer choice behavior.” Journal of Business & Economic Statistics 15 (3):310–327. Kim, Jinyoung. 2006. “Consumer’s dynamic switching decisions in the cellular service industry.” Mimeo. Kim, Moshe, Dron Klinger, and Bent Vale. 2003. “Estimating switching costs: the case of banking.” Journal of Financial Intermediation 12:25–56. Klemperer, Paul. 1987. “The competitiveness of markets with switching costs.” RAND Journal of Economics 18 (1):138–150. ———. 1995. “Competition when consumers have switching costs: an overview with applications to industrial organization, macroeconomics, and international trade.” Review of Economic Studies 62 (4):515–539. Lee, Jongsu, Yeonbae Kim, Jeong-Dong Lee, and Yuri Park. 2006. “Estimating the extent of potential competition in the Korean mobile telecommunications market.” International Journal of Industrial Organization 24:107–124. Mankski, Charles and Steven Lerman. 1977. “The estimation of choice probabilities from choice based samples.” Econometrica 45 (8):1977–1988. McFadden, Daniel. 1978. “Modelling the choice of residential location.” In Spatial Interaction Theory and Planning Models, edited by Anders Karlqvist, Lars Lundqvist, Folke Snickars, and Jorgen W. Weibull. Amsterdam: North-Holland, 75–96. 24

Nevo, Aviv. 2001. “Measuring market power in the ready-to-eat cereal industry.” Econometrica 69 (2):307–342. Shum, Matthew. 2004. “Does advertising overcome brand royalty? Evidence from the breakfast-cereal market.” Journal of Economics & Management Strategy 13 (2):241–272. Shy, Oz. 2002. “A quick-and-easy method for estimating switching costs.” International Journal of Industrial Organization 20 (1):71–87. Small, Kenneth A. and Harvey S. Rosen. 1981. “Applied welfare economics with discrete choice models.” Econometrica 49 (1):105–130. Viard, V. Brian. 2007. “Do switching costs make markets more or less competitive? The case of 800-nuber portability.” RAND Journal of Economics 38 (1):146–163.

25

Table 1: Choice-based sampling: allocation of the sample (i) Stayed sample (Planned number of observations：1000) Mobile carriers Ratio(%) Region Ratio(%) Age Ratio(%) NTT docomo 50 Kanto 50 10’s 10 au 30 Tokai 15 20-50’s 85 softbank 20 Kansai 35 60’s5 (ii) Switched sample (Planned number of observations：500) Mobile carriers Ratio(%) Region Ratio(%) Age Ratio(%) NTT docomo 20 Kanto 50 10’s 10 au 60 Tokai 15 20-50’s 85 softbank 20 Kansai 35 60’s5 Note: The distribution of the sample characteristics matches with the ratios of mobile carriers, region and age in this table. Kanto, Tokai and Kansai regions indicate the eastern, central and western part of Japan.

26

Table 2: Ratio of samples who used MNP

MNP users No MNP users Total

Number of observations Ratio(%) 339 64 192 36 531 100

27

Table 3: Summary Statistics (i) All sample Variables Mean S.E. Charge 4907 3086 Allowance 29930 34679 Address 112 111 Music 0.352 0.478 Game 0.342 0.474 Wallet 0.098 0.297 Student 0.150 0.357 Number of observations 1279

(ii) Without switch Mean S.E. 4612 2678 30433 34282 112 111 0.310 0.463 0.333 0.472 0.082 0.275 0.148 0.355 912

(iii) With switch Variables Mean S.E. Charge 5641 3828 Allowance 28678 35663 Address 112 113 Music 0.455 0.499 Game 0.362 0.481 Wallet 0.136 0.344 Student 0.155 0.363 Number of observations 439

(iv) With MNP Mean S.E. 5853 4162 31791 38529 121 113 0.459 0.499 0.328 0.470 0.160 0.368 0.097 0.297 314

Note: We drop a sample if there is an inconsistent answer or a gap between the observed and computed charges has more than 5000 yen or 50%. According to this adjustment, the dataset decreased from 1537 to 1279 observations.

28

Table 4: Estimation results: Nested Logit model

Variables Charge Charge × Allowance Charge × (Allowance)2 SWITCH SWITCH × Music SWITCH × Game SWITCH × Wallet au au × Music au × Game au × Wallet softbank softbank × Music softbank × Game softbank × Wallet

Variables Constant Address Student λ Log likelihood

First (i) WESML(8%) Coef. S.E. -17.780 2.206 *** 0.935 0.731 -0.024 0.038 -3.394 0.572 *** 0.519 0.374 0.084 0.362 0.524 0.508 0.505 0.307 0.836 0.469 * -0.935 0.461 ** -0.042 0.684 -1.916 0.322 *** 0.532 0.469 -0.839 0.477 * 0.077 0.573

Coef. 0.422 0.001 -0.617 0.409

S.E. 0.416 0.001 0.507 0.368

stage: Mobile carrier choice (i) WESML(6%) (i) ML(no-weight) Coef. S.E. Coef. S.E. -17.599 2.637 *** -18.215 1.322 *** 0.918 0.887 0.923 0.424 ** -0.024 0.048 -0.023 0.020 -3.778 0.758 *** -2.062 0.261 *** 0.537 0.466 0.500 0.201 ** 0.078 0.447 0.086 0.200 0.492 0.627 0.635 0.294 ** 0.526 0.388 0.440 0.149 *** 0.818 0.584 0.881 0.240 *** -0.949 0.573 * -0.831 0.236 *** 0.003 0.867 -0.223 0.341 -1.920 0.402 *** -1.899 0.166 *** 0.547 0.578 0.443 0.257 * -0.841 0.590 -0.825 0.261 *** 0.155 0.685 -0.216 0.353

Second stage: Coef. 0.479 0.001 -0.681 0.450

454.424

MNP usage choice S.E. Coef. 0.572 0.292 0.001 0.001 0.661 -0.457 0.495 0.310

367.410

S.E. 0.160 0.001 0.234 0.162

886.609

Note: ***, ** and * represent significance at 1, 5, and 10% level, respectively. For the representation purposes, we multiply the coefficient on Charge × Allowance by 10000 and on Charge × (Allowance)2 by 100000000.

29

* * * *

Table 5: Switching cost and the effects of MNP on switching cost (i) WESML(8%) Average S.E. Switching cost (yen) 2057 487 Effect of MNP (yen) 311 298 Rate of change (%) 17.8 -

(ii) WESML(6%) Average S.E. 2328 681 350 409 17.7 -

30

(iii) ML(no Average 1145 218 23.5

weight) S.E. 206 119 -

Table 6: Effects of MNP on consumer welfare (i) WESML(8%) Average S.E. 35 25

(ii) WESML(6%) Average S.E. 25 22

31

(iii) ML(no weight) Average S.E. 67 27

Table 7: Effects of MNP on switching probability (i) WESML(8%) With MNP Without MNP Difference 13.02 10.25 2.77

With MNP 8.56

(ii) WESML(6%) Without MNP Difference 6.14 2.42

(iii) ML(no weight) With MNP Without MNP Difference 32.09 29.59 2.50

32

Figure 1: Mobile carrier and MNP usage choice

Consumer ih (h = docomo)

au

docomo

softbank

Not use MNP

Use MNP

33

Use MNP

Not use MNP

Figure 2: Changes in switching costs (8%)

700 %HIRUH013\HQ 600 $IWHU013\HQ 500

400

300

200

100

0 0

300 600 900 1200 1500 1800 2100 2400 2700 3000 3300 3600 3900 4200 4500

34

Figure 3: Changes in switching costs (6%)

700 %HIRUH013\HQ

600

$IWHU013\HQ 500

400

300

200

100

0 0

300

600

900 1200 1500 1800 2100 2400 2700 3000 3300 3600 3900 4200 4500

35

Figure 4: Effects on consumer welfare (8%, yen)

1000 900 800 700 600 500 400 300 200 100 0 0

30

60

90

120 150 180 210 240 270 300 330 360 390 420 450

36

Figure 5: Effects on consumer welfare (6%, yen)

1000 900 800 700 600 500 400 300 200 100 0 0

30

60

90

120 150 180 210 240 270 300 330 360 390 420 450

37