Information Economics and Policy 25 (2013) 126141

Contents lists available at SciVerse ScienceDirect

Information Economics and Policy

journal homepage: www.elsevier.com/locate/iep

Mobile communication networks and Internet technologies as drivers of technical efciency improvement
Maria del Pilar Baquero Forero
Kyoto University, Graduate School of Economics, Yoshida-honmachi, Sakyo-ku, Kyoto 606-8501, Japan

a r t i c l e

i n f o

a b s t r a c t
Empirical research on the determinants of technical efciency (TE) is essential for policy formulation, in particular in low-income countries. In this study, we estimate the variations of TE between 1980 and 2009 in 23 low-income countries and 18 high-income countries, and demonstrate that TE has increased in both country groups in view of the deployment of mobile communication networks and Internet technologies. For low-income countries, we also prove that the causal relation is from the deployment of mobile networks and Internet technologies towards the increase of TE. More specically, by estimating the stochastic production frontier for a exible transcendental logarithmic production function under the assumption of xed effects, we show that the increase in TE per additional mobile phone and Internet subscriber is the highest in Latin American and Asian countries, but the accrued TE increase in response to Internet usage is the largest in high-income countries due to an overly higher Internet diffusion. Having established that modern information and communication technologies improve the TE, we conclude discussing policies that lead to the spread of such technologies, particularly in low-income countries. 2012 Elsevier B.V. All rights reserved.

Article history: Received 2 March 2011 Received in revised form 6 November 2012 Accepted 8 November 2012 Available online 7 December 2012 Keywords: Technical efciency Mobile communications Internet technologies Stochastic production frontier Government effectiveness Political risk

1. Introduction Differences in income levels across countries are determined by the stock of production inputs such as physical capital, labor force, and human capital, as well as by the efciency of input allocation, also known as the technical efciency (TE). In this context, empirical research unveiling the determinants of income and TE most conducive to economic growth is required for policy formulation, in particular in the context of low-income countries (Easterly and Levine, 2001). This study quanties the variations of TE in 23 lowincome and 18 high-income countries between 1980 to 2009 with respect to the deployment of two specic types of infrastructure: mobile communication networks and

E-mail address: baquero.maria@ax7.ecs.kyoto-u.ac.jp 0167-6245/$ - see front matter 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.infoecopol.2012.11.004

Internet technologies. To address the impact of a more general type of infrastructure, the contribution of freight railways is also taken into account. Furthermore, two institutional performance measures are considered: government effectiveness (i.e., the quality of public services and policy implementation), and political risk (i.e., the level of economic planning, corruption, law enforcement, and bureaucracy quality). Our econometric method is based on estimating the stochastic production frontier for a exible transcendental logarithmic (translog) production function and a technical efciency equation under the assumption of xed effects (Kumbhakar and Lovell, 2000). In order to reect the particular situation of low-income countries, we classify these into the following four geographical regions: Asia, Latin America, Middle East and North Africa (MENA), and Sub-Saharan Africa. Our results are threefold. At rst, it is shown that the deployment of mobile communication networks and

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

127

Internet technologies improves TE in both low-income and high-income countries. In particular, the TE improvement per additional subscriber to mobile communication networks and Internet technologies is the highest in Latin American and Asian countries. However, the overall TE improvement in view of the deployment of Internet technologies is the highest in high-income countries owing to an overly better penetration during the analyzed period. Secondly, it is shown that the deployment of mobile communication networks and Internet technologies causes the estimated TE improvement in low-income countries, i.e. the causal relation is from the diffusion of mobile communication networks and Internet technologies towards the TE improvement in low-income countries. For highincome countries, the causality between telecommunication technologies and TE improvement could not be conrmed, since the relevant coefcients are not statistically signicant. Lastly, we demonstrate further TE improvements in response to (i) higher freight railways usage in all countries, (ii) improved government effectiveness in Europe and SubSaharan Africa, and (iii) lower political risk in high-income countries as well as, to a lesser extent, in MENA and Asia. Our ndings highlight the importance of future deployments of next-generation technologies for mobile communications and broadband Internet. In this regard, we conclude by discussing effective telecommunication policies, focusing especially on the context of low-income countries. Some of these policies include privatization, granting regulator independence, guaranteeing market access to new entrants, and regulating nal-user price. The remainder of the paper is organized as follows. Section 2 surveys related previous studies on telecommunications and economic growth, and Section 3 summarizes our main contributions. Section 4 introduces the employed stochastic production frontier method, while the panel data and empirical model strategy are explained in Section 5. Section 6 presents the summary statistics of infrastructure and institutions of countries by income and region. Section 7 presents the estimated TE coefcients and rankings for low-income and high-income countries, as well as the model robustness checks. Finally, concluding remarks are offered in Section 8. 2. Previous work on telecommunications and economic growth In general, previous works investigate the contribution of infrastructure deployment to national income, as well as the causality of the relationship between infrastructure and income. To our best knowledge, research on the effects of infrastructure on TE is still scarce, and studies on the causality relation between infrastructure and TE have not been addressed previously. 2.1. Telecommunications as determinants of national income level National income level and GDP growth have been shown to partially depend on various types of

infrastructure. In high-income countries, these include mainlines (Roller and Waverman, 2001), mainlines and transportation (Esfahani and Ramirez, 2003), mobile communication networks (Waverman et al., 2006), information technologies (Lin and Chiang, 2011; Lin, 2009; Shao and Lin, 2001), and broadband Internet (Qiang et al., 2009, Czernich et al., 2011). In low-income countries, an improvement of market efciency has been demonstrated in view of the deployment of information and communication networks (Jensen, 2007; Aker, 2008; Jensen and Oster, 2007). The studies by Roller and Waverman (2001), Esfahani and Ramirez (2003), and Czernich et al. (2011) have also addressed the problem of the causality relation between income levels and infrastructure in developed countries. In the context of the causality between telecommunications and income, it is well known that the relationship is of a two-way nature (Roller and Waverman, 2001). On one hand, infrastructure investments lead to a higher income, but on the other hand, income raises the demand for infrastructure and induces a greater supply. The mutual coupling makes it difcult to dene the dominating causality. Some country-level studies examined the causality between income and infrastructure, in particular mainlines (Roller and Waverman, 2001), mainlines and railways (Esfahani and Ramirez, 2003), and broadband Internet (Czernich et al., 2011). The rst two studies employ models that estimate simultaneous equations of demand and supply, but Roller and Waverman (2001) only addresses the case of developed countries for which there are available data on infrastructure investment. In comparison, Czernich et al. (2011) employ pre-existing networks (voice telephony and cable TV) as instruments for the supply of broadband in order to isolate the causal effect of broadband Internet on economic growth in OECD countries. However, neither the determinants of TE nor their causal relation to TE have been addressed in the above studies. The focus on telecommunications infrastructure in both developed and developing countries as well as the investigation of the role of institutions and their quality are missing entirely. 2.2. Telecommunications as determinants of technical efciency Few recent studies have attempted to identify several infrastructure/institution types as the determinants of country TE. Thompson and Garbacz (2007) used the stochastic frontier model based on a Cobb-Douglas function to show that improved telecommunications and economic freedom increase TE, especially in Africa and Latin America. However, the employed econometric model suffers from an unrealistic assumption of errors being uncorrelated to the regressors, so-called random effects. This assumption may cause the by-denition positive coefcients of human capital to become negative. Thus, the credibility of the obtained results is weak (Kumbhakar and Lovell, 2000). Using a similar model, Lin and Chiang (2011) have found that Eastern European countries gain more productive efciency than the G7 countries if the IT capital is considered as a production input. The study also analyzes the

128

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

impact on TE of national savings, external balance, rm returns on assets (ROA), and vertical integration. However, the authors do not take into account the impact of IT as a driver of TE improvement. Instead, IT is considered as a production input. IT investment as a production input is studied by using the stochastic frontier method also by Lin (2009) and Shao and Lin (2001). 2.3. General infrastructures and institutions as determinants of technical efciency The present study investigates cross-country TE variations not only with respect to telecommunications but also to more general types of infrastructure such as transportation, as well as institutional performance parameters such as government effectiveness and political risk. Our motivation stems in the fact that government effectiveness, i.e. the policy quality as dened subsequently, is known to affect economic growth in both developed and developing countries (Sa, 2011), as well as in transition economies (Beck and Laeven, 2006). Moreover, the political risk index used in our analysis (encompassing, among other factors, law enforcement, corruption, and bureaucracy quality) is argued to be a determinant of the nancial performance of multinational companies (Kesternich and Schnitzer, 2010) and stock markets (Perotti and Oijen, 2001; Bilsona et al., 2002). Foreign direct investment (FDI) has also been shown to be affected by both government effectiveness (Daude and Stein, 2007) and political risk levels (Busse and Hefeker, 2007). In this regard, none of the above-mentioned studies on telecommunications and TE considers in their analysis such more general types of infrastructure (especially, Thompson and Garbacz, 2007; Lin and Chiang, 2011; Lin, 2009; Shao and Lin, 2001), or institutions (especially, Lin and Chiang, 2011; Lin, 2009; Shao and Lin, 2001). 3. Contributions In this study, we estimate the effects of various kinds of infrastructure and institutions on TE in both low-income countries and high-income countries. To this end, we use a stochastic frontier method similarly to Thompson and Garbacz (2007), but assume xed effects and a exible translog (generalized Cobb-Douglas) production function. More specically, we conduct a joint analysis of the relationship between TE and (i) the deployment of mobile communication networks and Internet technologies, (ii) the usage of freight railways, and (iii) the degree of government effectiveness and political risk.1 For both the low-income and high-income countries, we address in our statistical model also the causality issue between TE and telecommunications by additionally controlling for lagged telecommunications variables, namely for the past values (lagged 3 years) of the diffusion of mobile communication networks and Internet technologies.
1 The political risk factor used in this study is a general measure of institutional qualities calculated based on a weighted average of thirteen indicators, including economic planning failure, corruption, law enforcement, and bureaucracy quality. For a complete list, see Table 3.

4. Modeling approach using stochastic production frontier The stochastic production frontier approach employed in this study denes technical efciency (TE) as the ratio of actual observed output to the maximum feasible output, known as the stochastic production frontier. Since we allow TE to be time variant, a TE change indicates the degree of waste reduction achieved over time in the allocation of inputs required for obtaining maximum outputs (Kumbhakar and Lovell, 2000). Roughly speaking, in order to measure TE, we estimate the stochastic production frontier, or the maximum feasible output, corresponding to the observed output of countries in our sample data. The stochastic production frontier is estimated by adding a random error vit and a non-negative random error uit to a standard production function. The random error uit indicates the difference between observed output and the maximum feasible output, i.e. the technical inefciency. The time-variant TE is then estimated as the conditional expectation of uit, given the composed error it = vit uit. More specically, we employ the panel-data version of the xed-effects stochastic production frontier with timevariant technical inefciency effects (Stata Press, 2005) that is specied as follows:

ln yit ai RK j1 bj ln xjit v it uit ai RK j1 bj ln xjit it

where  ai is the xed effect of the ith country;  b is a K 1 vector of unknown scalar parameters to be estimated;  lnxit is a K 1 vector consisting of natural logarithms of production inputs of the ith country at the tth time period;  lnyit is the natural logarithm of the observed production output for the ith country at the tth time period;  vit are independent and identically distributed (iid) ran dom errors with a normal distribution N 0; r2 v ;  uit is a non-negative variable, associated with the technical inefciency of the output production. The term vit encompasses the measurement error and other random factors that affect the output yit. Also, it accounts for the combined effects of unspecied input variables in the production function. Consequently, the term RK j1 bj ln xjit v it constitutes the stochastic production frontier representing the maximum feasible output. The term uit represents the amount by which the observed countries fail to reach the frontier. For the ith country, the time-variant technical inefciency effects uit are independently truncated-normally distributed with a trun cation point at l N l; r2 . u The conditional mean model estimates l assuming it is a linear function of a set of covariates. The modeled conditional mean is given by l = c0 + czit where zit is a vector of observable non-random explanatory variables. These observable variables are expected to explain the reduction

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

129

of technical inefciency across countries. The zit variables include the diffusion of mobile communication networks and Internet technologies, freight railways usage, government effectiveness, and political risk.2 The assumption of xed effects implies that the terms vit are uncorrelated with the regressors, as well as that the terms uit are allowed to be correlated with the regressors and with vit. The main advantage of the xed-effects model is that it is simple while being characterized by desirable consistency properties (Kumbhakar and Lovell, 2000). A disadvantage stems in that the model captures the effects of all phenomena that vary across countries but are time invariant for each country. It occurs whether or not these other effects are included as regressors in the model. To estimate the actual technical efciency (TE) of individual countries, the best predictor for uit is the conditional expectation of uit, given the value of it = vit uit. The conditional expectation of uit is used since the actual value of uit is unobservable. It is noteworthy that even if the real value of the parameter vector b in the stochastic frontier model was known, only the difference it = vit uit could be observed. For the particular distributional assumptions imposed on the technical inefciency effects, the expected value of TEit can be calculated by using the equation

Table 1 List of countries by income and region. High-income countries (18) EUROPE (13) Austria Belgium Denmark Finland France Greece Ireland Italy Netherlands Portugal Spain Sweden United Kingdom ASIA (2) Japan Korea NORTH AMERICA (2) Canada United States Low-income countries (23) MIDDLE EAST AND NORTH AFRICA (7) Algeria Egypt Iran Jordan Syria Tunisia Turkey SUB-SAHARAN AFRICA (6) Cameroon Ghana Kenya Malawi Senegal Zambia ASIA (6) Bangladesh China India Indonesia Malaysia Thailand

OCEANIA (1) Australia

LATIN AMERICA (4) Argentina Chile Mexico Peru

Eit Eexpuit jit     1 Ur lit =r 1 exp lit r2 1 Ulit =r 2

by Jondrow et al. (1982) where

tion output, inputs, infrastructure and institutions is given in Table 2.

lit r

2 it r2 u lrv

ru rv rS 2 1 2 rS ru r2 v

r2 S

3 4 5

The stochastic production frontier is estimated by using maximum likelihood estimation (MLE) in a one-step procedure using statistical software Stata (version 9.1).

5. Panel data and empirical model specication 5.1. Panel data We use a panel dataset of 41 countries (23 low-income and 18 high-income) for the period between 1980 and 2009. The considered low-income countries are classied into four regions: Asia, Latin America, Middle East-North Africa (MENA), and Sub-Saharan Africa. The list of countries sorted by income and region is given in Table 1. The denitions and sources of the data employed to estimate the stochastic production frontier and TE, namely produc2 Compared to the half-normal case, the truncated-normal distribution has an additional parameter l to be estimated, since it allows for a nonzero mode. Therefore, it provides a more exible representation of technical efciency.

5.1.1. Production function: output and inputs The data used as the production output yit to estimate the stochastic production frontier is given by the real gross domestic product (GDP) at constant market prices as obtained from the Conference Board Total Economy Database (CB-TED).3 The data employed as production inputs xit consists of the physical capital stock, total labor force, and human capital. The physical capital stock in constant market prices from 1980 to 2000 was calculated on the basis of the above real GDP and the physical-capital-to-output ratio obtained from Klenow and Rodriguez-Clare (1997). For the period between 2001 and 2009, the physical capital stock was calculated on the basis of the methodology described in Klenow and Rodriguez-Clare (1997) by using the annual gross capital formation from the Word Development Indicators (WDI) database. The total labor force which includes all persons engaged in some productivity activity was extracted from the employment data of the above mentioned CB-TED. Finally, the human capital dened as years of educational attainment for population aged between 15 and 64 not enrolled in education was taken from the Cohen and Soto database. 5.1.2. Technical efciency equation: infrastructure and institutions The infrastructure and institutional data used as explanatory variables zit of TE as well as their source are
3 We thank the anonymous Referee for suggesting the use of the Conference Board Total Economy Database.

130 Table 2 Data sources 19802009. Variable Denition

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

Source CB-TEDa Klenow et al. WDI and CB-TEDb CB-TED Cohen-Soto database (2000) ITUc ITU WDId WGIf ICRG PRSh

Stochastic production function Y Real Gross Domestic Product (GDP) in constant market prices K Physical capital stock in constant market prices L Total labor force economically active population E Human capital years of educational attainment population 1564 Technical efciency function M Mobile telephone subscribers per 100 inhabitants I Internet subscribers per 100 inhabitants R Freight railways goods transported (million tons km/total country area) GE Government effectivenesse 2.5 (highly effective) 2.5 PRF Political risk factorg 100 (minimum risk) 0
a b

Conference Board Total Economy Database (CB-TED). Physical capital from 1980 to 2000 was calculated from the physical capital to output ratio k/y obtained from Klenow et al. and the above real GDP. Physical capital from 2001 to 2004 was calculated based on the methodology described in Klenow and Rodriguez-Clare (1997, p. 78) and using the World Bank (2007) (World Development Indicators) annual gross capital formation. c International Telecommunication Union (ITU). d World Development Indicators (WDI). e It reects perceptions of the quality of public services and of the civil service, as well as the latters degree of independence from political pressures, the quality of policy formulation and implementation, and the credibility of the governments commitment to such policies. (Kaufmann et al., 2008). f World Governance Indicators (WGI). g It represents the political stability per country by assigning risk points to a group of pre-set components: government stability, socioeconomic conditions, investment prole, internal conict, external conict, corruption, military in politics, religious tensions, law and order, ethnic tensions, democratic accountability and bureaucracy quality (Political Risk Services, 2009). h International Country Risk Guide (ICRG) by The Political Risk Services (PRS) group.

as follows. Mobile phone and Internet subscribers per 100 inhabitants were obtained from the International Telecommunication Union (ITU) database. Freight railways usage is dened as goods, in million tons, transported by railways per kilometer and were taken from the WDI database. To make the freight railways usage data comparable across countries of various sizes, the railways data were divided by each countries total area, also taken from the WDI database. The government effectiveness indicator combines the perception of the public/civil service quality, the degree of service independence from political pressures, the quality of policy formulation/implementation, and the credibility of the government commitment to its policies (Kaufmann et al., 2008). The indicator values range from 2.5 (less effective) to 2.5 (more effective). The corresponding data originates from the Worldwide Governance Indicators (WGI) database.4 The political risk factor (PRF) (Political Risk Services, 2009) takes values between 100 (lower risk) and 0 (higher risk). It is a general measure of institutional performance that incorporates economic risk perceptions such as economic planning failure and economic expectations vs. reality, as well as crucial political variables such as political leadership, corruption, law enforcement, quality of bureaucracy, among other religious or racial-related indicators. The corresponding data was extracted from the International Country Risk Guide (ICRG) of the Political Risk

Table 3 Political Risk Factor (PRF)a composition. Component Economic expectations vs. reality Economic planning failures Political leadership External conict Corruption in government Military in politics Organized religion in politics Law and order tradition Racial and national tensions Political terrorism Civil war risks Political party development Quality of bureaucracy Total
a

Points 12 12 12 12 10 6 6 6 6 6 6 6 6 100

Source: International Country Risk Guide (ICRG) by the Political Risk Services (PRS) group.

4 Kaufmann et al. (2008) construct the government effectiveness indicator by applying factor analysis to 35 primary indicators from 32 independent data sources.

Services (PRS) group. The complete list of indicators composing the political risk factor is shown in Table 3. The two institutional measures used in this paper have been analyzed in previous studies on cross-country economic performance. The government effectiveness indicator has been utilized in studies on economic growth (Sa, 2011; Beck and Laeven, 2006), and FDI (Daude and Stein, 2007). In addition, The authors of the government effectiveness indicator (Kaufmann et al., 2008) argue that, compared to alternative indicators used in the literature, these data are free of ideological and other inuences. Other authors considered the political risk factor, used in this study, in their research on multinational companies

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

131

(Kesternich and Schnitzer, 2010), stock markets (Perotti and Oijen, 2001; Bilsona et al., 2002), and FDI (Busse and Hefeker, 2007). 5.2. Empirical model specication using a translog function In this paper, we estimate a total of six stochastic production frontier models in order to measure the contribution of infrastructure and institutions on cross-country TE, as well as to unveil the underlying causal relation between telecommunications and TE. Models 1 and 2 are used to estimate the contribution of telecommunications, transportation and institutions to a TE reduction in high-income and low-income country regions. Models 36 are used to determine the causal relation between telecommunications and TE in high-income countries and low-income country regions. The stochastic production frontier is estimated in a translog functional form as this form is sufciently general and exible, and does not impose any assumptions on the elasticities of production, nor on the elasticities of substitution among production inputs. The six models differ in their efciency equation but estimate the same translog production function, having the real GDP as the observed production output yit, and the physical capital stock, labor force and education as production inputs xit. Hence, the estimated translog production function is

values of Internet subscribers (3 lags) in low-income regions and high-income countries, respectively.5 Our hypothesis is that mobile communication networks, Internet technologies, freight railways, government effectiveness indicator, and political risk factor should have negative signs if they contribute to lower cross-country technical inefciency, in other words if they help to increase TE. We also expect various degrees of TE improvement by income levels and regions. The lagged telecommunication variables should have negative signs if they indicate the causality from telecommunication diffusion towards technical inefciency reduction, i.e. an increase of TE. 6. Summary statistics The summary statistics of real GDP, inputs (physical capital, labor and human capital), infrastructure (mobile phone subscribers, Internet subscribers and freight railways usage), and institutional performance indicators (government effectiveness and political risk) for all considered high-income and low-income countries in 2009 is shown in Table 4. The summary statistics for high-income countries and low-income country regions is given in Tables 5 and 6, respectively. A comparison of infrastructure and institutional quality between high-income and low-income countries in 2009 is offered in the following in conjunction with a brief analysis on telecommunication policies of selected countries. 6.1. Mobile communication networks Low-income countries in our sample have in average 71 mobile phone subscribers per 100 inhabitants in 2009 while high-income countries had 117 mobile phone subscribers per 100 inhabitants. Within low-income countries, all Latin American countries have the penetration rates above the average of all low-income countries. In contrary, the Sub-Saharan countries have the lowest average rate 49 mobile phone subscribers per 100 inhabitants. The diffusion of mobile communication networks shows a great disparity in our sample, specially in Asia. However, the penetration rate of mobile phones is correlated with the degree of market competition as indicated by (Symeou, 2011; Fink et al., 2002; Wallsten, 2002; Li and Xu, 2002). This observation is validated in our analysis for the case of Sub-Saharan African and MENA countries. In particular, the lack of competition and poor economic performance have left Malawi, Cameroon, and Syria lagging behind in terms of mobile penetration (15.7, 37.8, and 45.9, respectively), as well as delaying the introduction of third-generation mobile services.6
5 We also estimated a model based on two telecommunications variables - mobile phone and Internet subscribers, but the results are not reported in this study. Although the estimated coefcients of both variables had the expected signs, they were not statistically signicant because of the high correlation between them (0.89). 6 According to the Balancing Act Africa, a third mobile operator was licensed in Malawi in 2008, but a delay in the network roll out causes the market to effectively remain in a duopoly condition (http://www.balancingact-africa.com/news/en/issue-no-601, accessed on August 30th, 2012).

lnyit bi bK lnK it bL lnLit bE lnEit bKL lnK it lnLit bKE lnK it lnEit bLE   lnLit lnEit 1=2bKK ln K 2 it 1=2bLL     2 2 ln Lit 1=2bEE ln Eit v it uit

The translog production function is characterized by allowing a non-linear relationship between the production output and inputs. Specically, the function estimates parameters of input quadratic terms (bKK, bLL and bEE) as well as parameters of input interaction, dened as the pairwise-products between inputs (bKL, bKE and bLE). In order to dene the point of approximation to the function, the output and input data of the production function are normalized by the sample mean before applying the logarithmic transformation (Chung, 1994). As for the data used as zit to model the mean of uit in the technical efciency equation, all six models consider the freight railways usage, government effectiveness indicator, and political risk factor. In addition to the above, Model 1 considers mobile phone subscribers, and Model 2 takes into account Internet subscribers. To measure separately the TE change of highincome and low-income countries with respect to the above mentioned infrastructure and institutions, both Models 1 and 2 include four regional dummy variables of low-income countries: Asia, Latin America, Middle EastNorth Africa (MENA), and Sub-Saharan Africa. Models 3 and 4 incorporate mobile phone diffusion and lagged values of mobile diffusion (3 lags) in low-income regions and high-income countries, respectively. Finally, Models 5 and 6 include Internet subscribers and lagged

132 Table 4 Summary statistics in 2009 by income. Variable Units

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

All countries No. obs: 41 Mean Std.dev. 2012.6 2.66e6 1.38e5 2.92 35.17 29.84 0.17 0.95 11.47 Min 10.1 1.16e4 1927 2.93 15.72 0.38 0 0.99 53 Max 9315.2 1.27e 7.77e5 13.32 150.47 90.28 0.79 2.19 92.5

Low-income countries No. obs: 23 Mean 790.7 9.56e5 7.27e4 7.10 71.09 20.5 0.02 0.14 63.26 Std.dev. 1935.9 2.24e6 1.80e5 1.95 30.17 14.6 0.04 0.56 6.73 Min 10.1 1.16e4 1960 2.93 15.72 0.38 0 0.99 53 Max 8965.7 1.04e 7.77e5 10.72 128.84 57.61 0.19 1.21 79.5

High-income countries No. obs: 18 Mean 1152.6 1.85e6 2.17e4 11.78 117.02 72.81 0.15 1.44 82.64 Std.dev. 2145.4 3.11e6 3.35e4 1.43 21.89 13.95 0.24 0.49 5.26 Min 106.9 1.04e5 1927 7.85 68.41 44.05 0 0.52 73 Max 9315.2 1.27 1.41e5 13.32 150.47 90.28 0.79 2.19 92.5

GDP (CBTED) K (CBTED) L (WDI) H (C-S) Mobile subs. Internet subs. Railways Gov. eff. index PRF

billions 1990US$ millions 1990US$ thousands of people years 100 inhab 100 inhab goods km/ area 2.5 to 2.5 0100

949.6 1.38e6 5.03e4 9.16 91.25 43.46 0.08 0.56 71.77

Table 5 Summary statistics in 2009 by region (high-income countries). Variable Units Europe No. obs: 13 Mean GDP (CBTED) K (CBTED) L (WDI) H (C-S) Mobile subs. Internet subs. Railways Gov. eff. index PRF billions 1990US$ millions 1990US$ thousands of people years 100 inhab 100 inhab goods km/area 2.5 to 2.5 0100 489261.4 784590.6 10411.15 11.23 126.27 71.07 0.059 1.43 82.92 Std.dev. 492515.9 822259.1 9837.78 1.31 16.23 16.14 0.073 0.55 5.71 Min 106,920 104,020 1927 7.85 95.09 44.05 0.001 0.51 73 Max 1,384,657 2,487,902 27,123 13.32 150.46 90.27 0.241 2.19 92.5 Other high-income No. obs: 5 Mean 2,877,466 4,608,526 51178.8 13.20 92.95 77.30 0.37 1.45 81.9 Std.dev. 3,699,330 5,074,374 54458.94 0.08 15.60 3.23 0.38 0.29 4.32 Min 538,735 792653.6 10,845 13.07 68.40 72.03 0.02 1.11 77.5 Max 9,315,219 12,668,512 141,466 13.29 110.72 80.90 0.79 1.77 87

On the other hand, there are countries in Latin America and Asia with competitive markets and mobile phone penetration rates higher than the average of high-income countries Argentina with 128.8 subscribers per 100 inhabitants, and Thailand with 122.6 subscribers per 100 inhabitants. The high diffusion rate of mobile communications in Argentina is mainly due to privatization and market deregulation that allowed the participation of new mobile operators, while the case of Thailand can be explained by a high Telecommunications Regulatory Environment (TRE) index, indicating tariff regulation, increased competition, and relative easiness of a market entry. In both country groups, mobile phone users exceed mainlines subscribers. However, mobile phone diffusion is specially important in low-income countries, where xed telephony penetration rates are remarkably low. The biggest differences between mobile and xed telephony diffusion rates are in countries of Sub-Saharan Africa such as Ghana with 63.4 mobile phone subscribers per 100 inhabitants and only 1.11 xed line subscribers per 100 inhabitants, and Senegal with 55.0 subscribers per 100 inhabitants and just 2.32 xed line subscribers per 100 inhabitants, as well as MENA countries such as Algeria and Jordan 93.7 and 101.1 mobile phone subscribers per 100 inhabitants versus merely 7.53 and 7.99 land line

subscribers per 100 inhabitants, respectively. It is noteworthy that the mentioned exemplary countries are characterized by mobile market liberalization, good telecommunications infrastructure, and innovative services.7

6.2. Internet technologies High-income countries in our sample in 2009 have an average of 72.8 Internet users per 100 inhabitants while low-income countries are characterized only by 20.5 users per 100 inhabitants. As noted in the context of the mobile phone diffusion, the highest average Internet penetration rate within low-income countries can be found in Latin
7 As reported by the BuddeComm consultancy, Ghana is the regional pioneer having launched the rst cellular mobile network in Sub-Saharan Africa in 1992, and privatized the telecommunications incumbent as early as in 1996. Similarly, the incumbent operator in Senegal has been offering the most efcient network and the lowest mobile prices in West Africa, as well as a wide range of value-added services such as mobile Internet access. Algeria has one of the highest teledensities in the continent, as a result of a well developed infrastructure and price competition between three mobile operators. Despite its low GDP per capita, Jordan has a relatively well developed telecoms sector and the most liberalized market in the region owing to an intense competition in the mobile market that has led to high subscriber numbers and reduced prices.

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

133

Middle East North Africa No. obs: 7

3.44e4 3.66e4 1960 5.15 45.97 13.46 0.003 0.74 55.5

America (29 users per 100 inhabitants) while the lowest average penetration rate is in Sub-Saharan African countries (6 users per 100 inhabitants). We comment on two cases of successful Asian countries Malaysia and Thailand. Malaysia has widely deployed advanced Internet infrastructures such as ber optics networks thanks to a government effort since 1991. Remarkably, the Internet penetration rate of 57.6 users per 100 inhabitants in Malaysia is higher than in the high-income countries of Greece (44), Portugal (48.6), and Italy (48.5). As for Thailand, it has had a strong Internet growth in recent years due to the liberalization of the Internet gateway market. Except for the above cases, our sample data indicate that Asian and Sub-Saharan African countries are generally lagging behind in the deployment of Internet technologies, and have the lowest Internet penetration rates (e.g., 0.4 users per 100 inhabitants in Bangladesh, 3.8 in Cameroon, 4.7 in Malawi, 5.44 in Ghana, and 5.31 in India). The Internet penetration rates are very low in these countries owing to an underdeveloped infrastructure as well as lack of privatization and regulation.8 Nevertheless, the recent WiMAX and Wi-Fi technologies for mobile Internet access could mitigate the limitations of country infrastructure since they can cost-efciently provision last-mile broadband Internet access in remote or rural areas in low-income countries (Galperin, 2005; Mishra et al., 2005; Gunasekaran and Harmantzis, 2007). According to the Buddecom consultancy, Bangladesh, Cameroon, and Malawi have already granted several licenses to WiMAX operators. In the MENA region, some countries such as Turkey, Tunisia, and Jordan have relatively high Internet penetration rates compared to other low-income countries (35.3, 33.5, 29.3 users per 100 inhabitants, respectively), mainly thanks to regulatory reforms, market competition, and infrastructure investments,9 but other countries such as Algeria (13.5 users per 100 inhabitants) and Syria (18.6 users per 100 inhabitants) have the lowest penetration rates, primarily due to regulatory barriers limiting the existing infrastructure in achieving its potential.10

Std.dev. Mean Max Std.dev. Mean Max Std.dev. Mean Min Max Min Min

Latin America No. obs: 4

Asia No. obs: 6

Sub-Saharan Africa No. obs: 6

5.87e5 7.40e5 2.84e4 10.18 101.06 38.28 0.02 0.41 72

2.43e4 4.25e4 8.97e3 5.26 42.46 6.28 0.003 0.49 61.25

1.44e4 5.21e4 5.22e3 1.30 16.97 2.21 0.002 0.31 4.63

1.01e4 1.16e4 4.82 2.93 15.72 3.84 0.0002 0.80 56.5

4.38e4 1.46e4 1.86e4 6.49 63.38 10.03 0.007 0.064 67

2.45e6 2.90e6 2.40e5 7.24 72.50 21.10 0.04 0.007 62.25

3.44e6 3.95e6 3.11e5 1.99 36.56 21.20 0.07 0.63 6.77

1.86e5 1.64e5 1.09e4 4.97 31.06 0.38 0.0005 0.98 53

8.97e6 1.04e7 7.77e5 10.15 122.56 57.60 0.18 0.98 72

4.02e5 5.02e5 1.92e4 9.20 97.05 29.64 0.03 0.23 69.87

3.02e5 4.35e5 1.64e4 1.03 22.62 3.32 0.02 0.690 7.76

1.54e5 1.93e5 7277 8.39 77.75 26.47 0.003 0.42 62

Min

8.21e5 1.14e6 4.34e4 10.71 128.83 33.98 0.06 1.20 79.5

Max

8 According to Pyramid Research, Bangladesh still has to develop its ICT infrastructure in order to promote Internet diffusion. Cameroon has not completed the privatization of its incumbent operator and has not allowed the entry of new operators by allocating additional licenses. Despite the existence of 15 licensed Internet Service Providers, Malawi is characterized by a low Internet penetration rate due to limited service availability and high cost of international bandwidth, as well as insufcient regulatory regime. 9 In Turkey, regulatory reform has helped improve competition and develop the Internet market. Tunisia has one of the most developed Internet markets and some of the lowest broadband prices in North Africa due to competition between eleven ISPs, access to a nationwide ber optic backbone network and international access via submarine and terrestrial ber. Finally, the Internet broadband market in Jordan is relatively competitive due to infrastructure-based competition and the introduction of WiMAX services. In addition, the quality of broadband services in Jordan is expected to improve through the deployment of two new international terrestrial cables. 10 Despite its well-developed infrastructure that includes a national ber backbone and one of Africas rst ber-to-the-home deployments, the low Internet penetration rate in Algeria stem from regulatory barriers hindering competition. In Syria, the Internet penetration rate is very low because broadband services are expensive and difcult to subscribe to.

Table 6 Summary statistics in 2009 by region (low-income countries).

Std.dev. Mean

GDP (CBTED) K (CBTED) L (WDI) H (C-S) Mobile subs. Internet subs. Railways Gov. eff. index PRF

Variable

billions 1990US$ millions 1990US$ thousands of people years 100 inhab 100 inhab goods km/area 2.5 to 2.5 0100

Units

2.50e5 3.36e5 1.50e4 7.35 79.57 26.93 0.01 0.17 62.07

2.09e5 2.90e5 1.10e4 1.53 19.26 9.53 0.009 0.50 6.63

134

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 Table 7 Input elasticities of output. Input elasticities of output Physical capital Labor Human capital Model 1 mobile phones 0.313 0.306 0.804 Model 2 Internet 0.311 0.243 0.871

In Latin America, low Internet penetration rates in Peru (27.7 users per 100 inhabitants) and Mexico (26.5 users per 100 inhabitants) are the result of market restrictions, low computer ownership, and low competition. In this context, it is noteworthy that subsidies to the poorest households increase computer ownership while the provisioning of public Internet access enhances Internet penetration (ITU Report, 2008).11 6.3. Freight railways usage, government effectiveness, and political risk In terms of the freight railways usage by region, the United States and Canada are the countries that transported the most goods by railways in 2009, followed by Australia and Austria. Slightly over the level of Finland and France, China is the low-income country with the highest use of railways used for cargo transports. High-income countries such as Portugal and Greece as well as the majority of low-income countries are characterized by a minimum usage of railways for the transportation of goods. The average of the government effectiveness indicator ranging from 2.5 (the lowest effectiveness) to 2.5 (the highest effectiveness) is 0.13 for low-income countries and 1.4 for high-income countries in 2009. The most efcient countries within the high-income group are Denmark (2.2), Finland (2.1), and Sweden (2) while the least efcient governments in the same group are Spain (0.9), Greece (0.6), and Italy (0.5). As for the low-income countries, the worst average government effectiveness was obtained for Sub-Saharan Africa (0.49) while Latin America is top ranking (0.23). The average political risk factor of high-income countries in 2009 is 82.6. In this group, the countries having the lowest political risk are Finland (92.5), Austria (90), and Sweden (88.5). In contrary, the highest political risk characterizes Spain (77), and Greece (73). In comparison, low-income countries have the average political risk factor lower by 25% (63.2). Within low-income countries, Latin America has a slightly lower average political risk (70) while MENA, Sub-saharan Africa, and Asia have the same average of 62 points. 7. Estimation results 7.1. Production function estimates and input elasticities of output The estimation results of the six stochastic production frontier models are given in Tables 8, 12, and 13. We obtained positive and statistically signicant coefcients of physical capital, labor, and education in all the six estimated models. The majority of coefcients of input interaction (pair-wise products) is also statistically signicant which validates the choice of translog production function
11 According to the ITU Report (2008), Peru is a world leader in public Internet access with 56% inhabitants accessing the Internet from public places.

instead of less exible functional forms such as the CobbDouglas function. The output elasticity of inputs calculated based on the estimation results of Models 1 and 2 are given in Table 7. The highest output elasticity among the three production inputs is that of human capital, ranging between 0.80 and 0.87. Physical capital has a higher output elasticity (0.31) compared to the elasticity of labor (between 0.24 and 0.30). Thompson and Garbacz (2007) also obtained a higher elasticity of capital (0.801) compared to the elasticity of labor (0.167) for high-income countries but also obtained a higher elasticity of labor (0.517) compared to the elasticity of capital (0.458) for low-income countries. The results by Thompson and Garbacz (2007) also indicate a negative output elasticity for human capital for both low-income countries (0.0067) and OECD countries (0.095). The difference between the output elasticity of inputs of our study and that of Thompson and Garbacz (2007) is partly due to different functional forms (translog vs. Cobb-Douglas). Moreover, Thompson and Garbacz (2007) consider a time trend which is not statistically signicant, except for low-income countries and Africa. 7.2. Technical efciency estimates The estimated TE changes per country/region during the analyzed period 19802009 are shown in Fig. 1. For highincome countries, the average TE improved by 6% during the rst decade, from 72.1% in 1980 to 76.5% in 1990. The most remarkable increase was obtained for the following decade when TE rose by 11% and reached 84.9% in 2000. In the last decade, an increase of 8% followed and TE attained 91.6%. In case of low-income countries, the average TE grew only by 7.2% in the rst two decades, from 48.7% in 1980 to 52% in 2000. In contrast, the estimated TE increase was the highest (16.8%) from 2000 to 2009 with TE reaching 60.1%.

Fig. 1. Estimated technical efciency improvement from 1980 to 2009 by region.

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

135

Within the low-income country regions, Latin American and Asian countries are characterized by the highest TE levels and the most signicant change. During the last decade of the studied period, TE levels rose by 24.3% rise from 58.7% to 73% in Latin America. In Asia, they rose by 21.7% from 55% to 67.6%. In contrary, the average TE of Sub-Saharan African and MENA countries grew at a slower rate of 13% and 10%, respectively, during the same decade. In both country types, a notable increase in TE took place when mobile communication networks and Internet technologies became the most widespread (late 1980s in high-income countries, and after the year 2000 in lowincome countries).
Table 8 Technical efciency reduction by mobile phones and Internet subscribers. Stochastic frontier truncated-normal model Model 1 mobile phones Primary index equation parameters b0 0.89 bCapital 0.386 bLabor 0.237 bEducation 1.109 bCapital labor 0.021 bCapital education 0.032 bLabor education 0.179 bCapital2 0.035 bLabor2 bEducation2 0.021 0.132 (0.072)a (0.018) (0.064) (0.074) (0.015) (0.020) (0.032) (0.006) (0.012) (0.043) (0.083) (0.0002) (0.0003) (0.001) (0.0005) (0.0003) (0.074) (0.056) (0.072) (0.063) (0.740) (0.034) (0.144) (0.132) (6.222) (0.113) (0.068) (0.0009) (0.001) (0.001) (0.001) (0.001) (0.045) (0.270) (0.0006) (0.0004) (0.067)

The following subsections discuss in more detail the specic contribution of telecommunication technologies, transportation, and institutions to the changes in TE in the studied countries. 7.3. Technical efciency improvement by telecommunications and their relation of causality According to our estimation results summarized in Table 8, the reduction in technical inefciency for highincome countries is 0.002 for an additional percentage point in mobile phone users, and 0.003 for an additional percentage point in Internet users. For Asian and Latin

Model 2 Internet b0 bCapital bLabor bEducation bCapital labor bCapital education bLabor education bCapital2 bLabor2 bEducation2 0.719 0.393 0.125 1.163 0.013 0.015 0.216 0.036 0.045 0.161 0.788 0.003 0.0001 0.038 0.006 0.003 0.435 0.180 0.356 0.140 3.034 0.090 0.234 0.065 69.794 0.075 0.084 0.006 0.002 0.009 0.001 0.005 4.810 0.461 0.005 0.003 0.613 1230 1289.145 (0.072) (0.018) (0.057) (0.083) (0.015) (0.021) (0.031) (0.006) (0.012) (0.044) (0.077) (0.0002) (0.001) (0.008) (0.001) (0.0008) (0.094) (0.054) (0.071) (0.037) (0.743) (0.031) (0.050) (0.144) (6.163) (0.108) (0.069) (0.0009) (0.001) (0.002) (0.001) (0.001) (0.044) (0.270) (0.0006) (0.0005) (0.064)

c0

Offset [mean = lit] parameters in one-sided error uit 0.923 0.002 0.0006 0.006 0.002 0.001 0.127 0.108 0.262 0.213 3.691 0.030 0.688 0.102 74.174 0.233 0.029 0.005 0.002 0.008 0.002 0.003

c0 cInternet cIntMENA cIntSUBSAH cIntLATIN cIntLowASIA cRailways cRailMENA cRailSUBSAH cRailLATIN cRailLowASIA cGov.Eff. cGEEUR cGEMENA cGESUBSAH cGELATIN cGELowASIA cPol.Risk cPRFMENA cPRFSUBSAH cPRFLATIN cPRFLowASIA
ln (r2) ilgtc

cMobile phones cMobMENA cMobSUBSAH cMobLATIN cMobLowASIA cRailways cRailMENA cRailSUBSAH cRailLATIN cRailLowASIA cGov.Eff. cGEEUR cGEMENA cGESUBSAH cGELATIN cGELowASIA cPol.Risk cPRFMENA cPRFSUBSAH cPRFLATIN cPRFLowASIA

Variance parameters for compound error ln (r2) 4.851 ilgtc 0.087 0.003 r2 u

r2 v c
Others No. of obs. Log likelihood
a

0.004 0.478 1230 1299.576

r2 u r2 v c
No. of obs. Log likelihood

Values within parenthesis are standard errors. Parameter is statistically signicant at 1% level. Parameter is statistically signicant at 5% level. Parameter is statistically signicant at 1% level.

136

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

Table 9 Technical efciency ranking of high-income countries. High-income countries technical efciency Country Ranking by TEa 2009 United States United Kingdom Finland France Canada Australia Netherlands Sweden Austria Belgium Japan Ireland Denmark Greece Portugal Spain Korea Italy
a b

TE determinants in 2009 Mobile phonesb Internetc Railways usaged Government effectivenesse Political riskf

2000 0.997 0.996 0.993 0.995 0.983 0.956 0.968 0.908 0.843 0.819 0.809 0.776 0.805 0.721 0.753 0.683 0.649 0.629

1990 0.995 0.995 0.988 0.994 0.923 0.886 0.826 0.738 0.729 0.743 0.672 0.584 0.677 0.622 0.660 0.611 0.563 0.556

1980 0.991 0.993 0.845 0.991 0.857 0.712 0.770 0.701 0.692 0.701 0.630 0.530 0.633 0.636 0.632 0.610 0.544 0.508 97.197 129.976 144.238 95.091 68.408 110.720 128.132 122.832 140.756 115.110 90.088 109.451 133.941 117.829 142.756 110.952 98.353 150.466 78.139 83.188 83.938 71.267 77.726 72.031 89.964 90.279 73.452 75.200 77.723 68.367 85.915 44.053 48.615 61.181 80.907 48.547 0.791 0.020 0.166 0.041 0.765 0.283 0.026 0.124 0.241 0.060 0.017 0.001 0.037 0.004 0.008 0.015 0.019 0.022 1.387 1.475 2.131 1.442 1.779 1.743 2 1.985 1.634 1.475 1.256 1.298 2.191 0.608 1 0.935 1.112 0.516 82.5 77.5 92.5 78.5 85 87 84.5 88.5 90 82.5 77.5 87.5 84 73 83 77 77.5 79.5

0.998 0.997 0.996 0.995 0.994 0.990 0.988 0.978 0.962 0.931 0.929 0.906 0.877 0.855 0.822 0.766 0.765 0.729

The efciency coefcient per country is dened as TEi = exp (ui) Mobile phone subscribers per 100 inhabitants in 2009, ITU data. Internet subscribers per 100 inhabitants in 2009, ITU data. d Goods transported (million tons km/country area) in 2009, WDI data. e Perceptions of the quality of public services and of the civil service, as well as the latters degree of independence from political pressures, the quality of policy formulation and implementation, and the credibility of the governments commitment to such policies in 2009, WGI data. f Components: government stability, socio-econ. conditions, investment prole, internal and external conict, corruption, military in politics, religious tensions, law and order, ethnic tensions, democratic accountability and bureaucracy quality in 2009, PRS data.
c

American countries, the reductions are 0.003 and 0.004 for an additional percentage point in mobile phone users, respectively, and 0.006 and 0.009 for an additional percentage point in Internet users, respectively. These results conrm that the contribution to TE per additional mobile phone and Internet subscriber is higher in Latin America and Asia compared to the high-income countries. However, the accrued increase of TE is higher in high-income countries due to overall higher penetration rates of both mobile telecommunication networks and Internet. Tables 9 and 10 summarize, respectively, the estimated TE levels, as well as the penetration levels of mobile communication networks and Internet technologies in high-income and low-income countries in 2009. Table 11 summarizes the correlation coefcients between the penetration rates of mobile communication networks and Internet technologies, as well as the technical efciency rankings in 2009. We observe that the correlation between efciency rankings and mobile phone penetration levels in 2009 is positive for low-income countries (0.45), and negative (0.28) for high-income countries. This implies that the most technically efcient high-income countries do not have the highest mobile penetration rates. In contrary, the most (least) efcient countries in the low-income countries have the highest (lowest) mobile penetration. As for the Internet technologies, the most technically efcient countries are also the ones characterized by a higher Internet penetration. The correlation is specially strong in

Latin America (0.98), Asia (0.64), and in the high-income countries (0.616). Tables 12 and 13 summarize the estimation results of the impact of lagged variables of mobile phone and Internet diffusion on technical inefciency reduction, i.e. a TE increase in low-income and high-income countries. We observe that for low-income countries, the causal relation is from the mobile communication networks and Internet technologies towards technical inefciency reduction (i.e., TE increase). This claim is justied by the statistically signicant coefcient of the lagged variables (3 lags) of mobile phone subscribers: 0.001 for MENA, 0.003 for Sub-Saharan Africa and Asia, and 0.002 for Latin America. Similarly, statistically signicant parameters of the lagged variables (3 lags) of Internet subscribers were obtained for MENA (0.002), Sub-Saharan Africa (0.02), Latin America (0.008), and Asia (0.01). For high-income countries, a signicant causal relation between TE and telecommunications infrastructure was not observed.

7.4. Technical efciency improvement by transportation and institutions Higher usage of freight railways, increased government effectiveness, and improved political risk have a varying impact on TE in low-income and high-income countries during the analyzed period. According to the estimation results summarized in Table 8, there is a decrease in technical inefciency for

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 Table 10 Technical efciency ranking of low-income countries. Low-income countries technical efciency Country Ranking by TEa 2009 Asia China Malaysia Thailand India Bangladesh Indonesia Latin America Chile Argentina Peru Mexico 0.978 0.754 0.726 0.568 0.520 0.512 0.958 0.730 0.670 0.561 2000 0.683 0.632 0.646 0.500 0.454 0.418 0.779 0.569 0.509 0.490 1990 0.610 0.500 0.604 0.502 0.424 0.442 0.672 0.471 0.428 0.491 0.476 0.442 0.481 0.446 0.439 0.455 0.422 0.617 0.455 0.462 0.548 0.436 0.340 1980 0.540 0.446 0.617 0.462 0.394 0.505 0.698 0.529 0.537 0.545 0.437 0.501 0.446 0.461 0.487 0.483 0.461 0.586 0.429 0.456 0.455 0.399 0.313 56.103 110.598 122.567 45.448 31.068 69.248 96.935 128.837 84.692 77.750 93.495 101.064 83.912 66.689 72.088 93.793 45.974 48.652 63.383 37.892 55.061 34.066 15.723 28.840 57.608 25.804 5.305 0.380 8.696 33.983 30.399 27.721 26.472 33.549 29.270 35.300 20.043 38.289 13.468 18.656 10.038 5.441 3.841 7.364 6.313 4.693 0.189 0.005 0.004 0.047 0.001 0.001 0.023 0.029 0.003 0.066 0.019 0.005 0.012 0.004 0.028 0.003 0.011 0.003 0.001 0.005 0.003 0.007 0.0002 0.116 0.989 0.152 0.014 0.985 0.211 1.209 0.420 0 0.17 0.413 0.280 0.351 0.300 0.742 0.590 0.608 0.658 0.064 0.806 0.400 0.674 0.520 67.5 72 58 62.5 53 60.5 79.5 65.5 62 72.5 72 71 59.5 58.5 55.5 60.5 57.5 57 67 65.5 58 63.5 56.5 TE determinants in 2009 Mobile phonesb Internetc Railways usaged Government effectivenesse

137

Political riskf

Middle East and North Africa Tunisia 0.632 0.543 Jordan 0.584 0.463 Turkey 0.573 0.487 Egypt 0.521 0.480 Iran 0.465 0.460 Algeria 0.462 0.401 Syria 0.399 0.469 Sub-Saharan Africa Kenya 0.650 Ghana 0.641 Cameroon 0.594 Senegal 0.561 Zambia 0.503 Malawi 0.425
a b

0.601 0.503 0.467 0.642 0.390 0.386

The efciency coefcient per country is dened as TEi = exp (ui) Mobile phone subscribers per 100 inhabitants in 2009, ITU data. Internet subscribers per 100 inhabitants in 2009, ITU data. d Goods transported (million tons km/country area) in 2009, WDI data. e Perceptions of the quality of public services and of the civil service, as well as the latters degree of independence from political pressures, the quality of policy formulation and implementation, and the credibility of the governments commitment to such policies in 2009, WGI data. f Components: government stability, socio-econ. conditions, investment prole, internal and external conict, corruption, military in politics, religious tensions, law and order, ethnic tensions, democratic accountability and bureaucracy quality in 2009, PRS data.
c

Table 11 Correlation between technical efciency ranking and technical efciency determinants in 2009. TE determinants Mobile phones Internet Railways usage Government effectiveness Political risk factor
a

High-income countries 0.236 0.616 0.457 0.695 0.482

Asia 0.316 0.648 0.775 0.539 0.628

Latin America 0.350 0.982 0.453 0.709 0.569

MENAa 0.716 0.530 0.044 0.927 0.778

Sub-Saharan Africa 0.845 0.420 0.079 0.251 0.367

Middle East and North Africa.

every additional percentage point in railways usage for all the studied countries, especially in Asia.12 Government effectiveness has an impact on technical inefciency reduction only for Europe and Sub-Saharan Africa.13 No negative impact of government effectiveness was observed in other low-income regions.
12 The high coefcient for Asia may be explained by the freight railways usage level in China, which is the highest among other low-income countries as well as the majority of high-income countries. 13 In the case of Sub-Saharan African countries, Cameroon, Ghana, and Malawi are characterized by signicant improvements of the indicator from the second half of 1990.

Finally, political risk factor reduces technical inefciency, especially in high-income countries, and to a lesser extent, in MENA and Asian countries. For Sub-Saharan Africa, there is no improvement in TE due to political risk factor changes, and for South American countries, the coefcients are not statistically signicant. 7.5. Robustness check As a robustness check, in estimation results not shown in this paper, we included two additional variables separately in Models 1 and 2 to check whether the parameters

138

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

Table 12 Causality between technical efciency and mobile phones subscribers. Stochastic frontier truncated-normal model with lagged mobile phone subscribers Model 3 low-income countries Primary index equation parameters b0 0.151 bCapital 0.408 bLabor 0.348 bEducation 0.877 bCapital labor 0.035 bCapital education 0.009 bLabor education 0.095 bCapital2 0.023 bLabor2 bEducation2 0.076 0.126 (0.105)a (0.019) (0.059) (0.084) (0.010) (0.022) (0.034) (0.006) (0.015) (0.046) (0.061) (0.0002) (0.0003) (0.0009) (0.001) (0.0004) (0.0005) (0.039) (0.020) (0.0005) (0.197) (0.288) (0.001) (0.00009) (0.037) b0 bCapital bLabor bEducation bCapital labor bCapital education bLabor education bCapital2 bLabor2 bEducation2 Model 4 high-income countries 0.257 0.406 0.418 0.845 0.041 0.003 0.092 0.021 0.076 0.164 0.798 0.001 0.001 0.000005 (0.103) (0.019) (0.057) (0.083) (0.014) (0.023) (0.034) (0.006) (0.035) (0.052) (0.061) (0.0003) (0.0003) (0.00002)

c0

Offset [mean = lit] parameters in one-sided error uit 0.780 0.001 0.001 0.0005 0.002 0.001 0.002 0.403 0.058 0.003

c0 cMobile phones cMobLAGGED cMobLAGGEDEUR

cMobile phones cMobLAGGED cMobLAGGEDMENA cMobLAGGEDSUBSAH cMobLAGGEDLATIN cMobLAGGEDASIA cRailways cGov.Eff. cPol.Risk

cRailways cGov.Eff. cPol.Risk

ln (r2) ilgtc

0.428 0.071 0.003 4.913 1.867 0.006 0.0009 0.866 1106 1147.537

(0.041) (0.019) (0.0005) (0.212) (0.296) (0.001) (0.00009) (0.034)

Variance parameters for compound error 5.028 ln (r2) ilgtc 1.703 0.005 r2 u

r2 v c
Others No. of obs. Log likelihood
a

0.001 0.845 1106 1159.331

r2 u r2 v c
No. of obs. Log likelihood

Values within parenthesis are standard errors. Parameter is statistically signicant at 1% level. Parameter is statistically signicant at 5% level. Parameter is statistically signicant at 1% level.

of interest in the technical efciency equation changed signicantly. The rst variable is the share of households with a television set (TV), obtained from the World Development Indicators (WDI). Television is believed to contribute to the reduction of technical inefciency by conveying information and knowledge, albeit less efcient than Internet technologies. The second variable added to the models is the index of economic freedom (IEF), taken from the Heritage Foundation.14 This institutional variable can reduce technical inefciency because it permits to minimize distortions that affect not only market transactions but also the output and welfare of both consumers and producers. After controlling for mobile telecommunication networks, Internet, freight railways, and institutions, it was observed that television sets do not contribute to reduce technical inefciency (exact results are omitted due to
14 The index of economic freedom (IEF) is an average of the following ten component scores: business freedom, trade freedom, scal freedom, government size, monetary freedom, investment freedom, nancial freedom, property rights, freedom from corruption and labor freedom.

space limitations). Similarly, in the case of economic freedom, we obtained an indication of reduced technical inefciency but the coefcients are not statistically signicant. Most importantly, the inclusion of both variables does not change the main parameter estimates in the technical efciency equation, nor the estimates of the inputs in the stochastic production frontier. Therefore, our estimated results are robust with respect to an inclusion of additional variables in the proposed models. We have also checked the technical efciency rankings obtained from Models 1 and 2 and the robustness check models by including separately the two additional variables (TV and IEF). None of the technical efciency rankings of both high-income and low-income countries has changed in a signicant way. Therefore, the country technical efciency rankings are also robust with respect to the inclusion of additional variables to the main models.

8. Conclusions By estimating a stochastic production frontier for a exible transcendental logarithmic production function under

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 Table 13 Causality between technical efciency and Internet subscribers. Stochastic frontier truncated-normal model with lagged internet subscribers Model 5 low-income countries Primary index equation parameters b0 0.699 bCapital 0.388 bLabor 0.286 bEducation 0.990 bCapital labor 0.072 bCapital education 0.009 bLabor education 0.109 bCapital2 0.010 bLabor2 bEducation2 0.099 0.189 (.069)a (0.020) (0.061) (0.084) (0.015) (0.025) (0.035) (0.007) (0.013) (0.053) (0.058) (0.0006) (0.0008) (0.003) (0.012) (0.002) (0.003) (0.054) (0.018) (0.0005) (0.050) (0.845) (0.0005) (0.0003) (0.044) b0 bCapital bLabor bEducation bCapital labor bCapital education bLabor education bCapital2 bLabor2 bEducation2 Model 6 high-income countries 0.127 0.402 0.371 0.912 0.043 0.036 0.127 0.021 0.059 0.205 0.689 0.003 0.0005 0.0004 (0.104) (0.018) (0.058) (0.080) (0.015) (0.022) (0.034) (0.006) (0.015) (0.047) (0.082) (0.001) (0.001) (0.002)

139

c0

Offset [mean = lit] parameters in one-sided error uit 0.581 0.003 0.001 0.001 0.020 0.007 0.010 0.091 0.038 0.002

c0 cInternet cInternetLAGGED cIntLAGGEDEUR

cInternet cIntLAGGED cIntLAGGEDMENA cIntLAGGEDSUBSAH cIntLAGGEDLATIN cIntLAGGEDLowASIA cRailways cGov.eff. cPol.risk

cRailways cGov.eff. cPol.risk

ln (r2) ilgtc

0.269 0.061 0.003 4.694 1.614 0.007 0.001 0.833 1188 1156.035

(0.104) (0.022) (0.0006) (0.048) (0.663) (0.001) (0.0008) (0.091)

Variance parameters for compound error 4.805 ln (r2) ilgtc 2.823 0.007 r2 u

r2 v c
Others No. of obs. Log likelihood
a

0.0004 0.943 1106 1162.366

r2 u r2 v c
No. of obs. Log likelihood

Values within parenthesis are standard errors. Parameter is statistically signicant at 1% level. Parameter is statistically signicant at 5% level. Parameter is statistically signicant at 1% level.

the assumption of xed effects, we provide empirical evidence of infrastructure and institutions being determinants of technical efciency (TE) improvement in both low-income and high-income countries from 1980 to 2009. In particular, we demonstrate various levels of TE increase due to the deployment of mobile communication networks and Internet technologies, usage of freight railways, improved government effectiveness, and lower political risk. Importantly, the causal relation is shown to be from the deployed mobile communications networks and Internet technologies towards the TE in low-income countries in Latin America, Asia and Sub-Saharan Africa. For high-income countries, the particular causality could not be determined since the results were not statistically signicant. Consequently, to benet from further TE improvements, policy makers, especially in low-income countries, must implement telecommunication policies that increase the penetration rates of mobile communication networks and Internet technologies. Both current and next-generation systems are of interest. In general, the spread of telecommunication technologies can be achieved by privatizatizing networks and

enabling market competition, promoting market access to new entrants, controlling the end-user price, granting subsidies to increase computer ownership, and providing public access to telecommunication services such as the Internet. To be more specic, previous studies concerning the effectiveness of various telecommunications policies have shown that the privatization of incumbent operators of mobile communication networks, as well as policies ensuring competition increase the TE of the telecommunications sector and trigger technology diffusion (Symeou, 2011). However, for better market outcomes, the competition and privatization must be implemented simultaneously (Fink et al., 2002; Li and Xu, 2002). In addition to privatization and competition policies, the regulator authority must be granted independence (Fink et al., 2002; Howard and Mazaheri, 2009). Competition in the mobile telecommunication sector can be also promoted by guaranteeing market access to new entrants by awarding new spectrum licenses, or regulating the share of telecommunications infrastructure among providers. However, in this context, it has been shown that the possibility of awarding spectrum licenses on the basis of either an auction or a so-called beauty

140

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141

contest creates a trade-off between government revenues and the diffusion of third generation of mobile phone technologies (Baquero and Kuroda, 2011). As for the policies related to Internet technologies, previous research has shown better Internet diffusion in developing countries in response to regulation aiming to (i) promote market access of additional Internet Service Providers, and (ii) control the end-user prices (Wallsten, 2002). Furthermore, subsidies targeting higher rates of computer ownership by the poorest households as well as the provisioning of public Internet represents additional measures that enhance the usage of Internet in low-income countries (ITU Report, 2008). The ongoing standardization and initiated commercialization of the next-generation networking technologies such as Worldwide Interoperability for Microwave Access (WiMAX), Wireless Fidelity (Wi-Fi), and Long Term Evolution Advanced (LTE-A) may possibly become the main driver of Internet penetration in both low-income and highincome countries. The reason for the above statement is that in developing countries, the low deployment cost of WiMAX or Wi-Fi enable economically viable solutions for last-mile broadband Internet access on both the local level (e.g., in small villages and remote locations in Latin America Galperin, 2005, India Mishra et al., 2005) and several developing countries (Gunasekaran and Harmantzis, 2007). In high-income countries, the claim justication follows from the technological superiority of the LTE-A network over current solutions as well as the rapid spread of increasingly capable portable tablets (expected to reach 500 million by 2015). In particular, since LTE will allow transmission speeds high enough to rival xed broadband connections, auctions of spectrum for fourth generation mobile telephony using LTE are already being run in many countries (Hoernig and Valletti, 2012). In this context, further research should take into account the impact of the deployment of innovative mobile communication services and increased broadband Internet capacity. This study focused only on the diffusion of basic mobile communications and Internet service. Further differentiation can be done on the level of service content - the next-generation mobile phone services in developed countries are primarily used for entertainment purposes (music downloads and mobile games) whereas in developing countries, data services such as agricultural advices, health care and money transfer are of higher importance. Another issue consists in analyzing whether the productivity gains stemming from a universal Internet access in developing countries can be obtained also by deploying the next-generation technologies such as WiMAX, Wi-Fi, and LTE. The usage of capable/inexpensive computing devices such as the tablets and netbooks should be considered as a related factor as well. Acknowledgements The comments and suggestions by the Journal Editor Tobias Kretschmer and two anonymous Referees have greatly contributed to this study. The author would like

to express a deep gratitude to Professors Takanori Ida and Kazuhiro Yuki from Kyoto University, as well as Assistant Professor Toshifumi Kuroda from Tokyo Keizai University for their helpful advices and careful review support. This study was supported in part by the Monbukagakusho scholarship from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of the Government of Japan. References
Aker, J.C., 2008. Does digital divide or provide? the impact of cell phones on grain markets in Niger. BREAD Working Paper, pp. 161. Baquero, M., Kuroda, T., 2011. Treatment effects analysis of the third generation mobile phone spectrum auction. Journal of Information and Communication Research (in Japanese) 100 (29-3), 4959. Beck, T., Laeven, L., 2006. Institution building and growth in transition economies. Journal of Economic Growth 11, 157186. Bilsona, C.M., Brailsforda, T.J., Hooper, V.C., 2002. The explanatory power of political risk in emerging markets. International Review of Financial Analysis 11, 127. Busse, M., Hefeker, C., 2007. Political risk, institutions and foreign direct investment. European Journal of Political Economy 23, 397415. Chung, J.W., 1994. Utility and Production Functions. Blackwell, Oxford, England. Czernich, N., Falck, O., Kretschmer, T., Woessmann, L., 2011. Broadband infrastructure and economic growth. Economic Journal, Royal Economic Society 121552, 505532. Daude, C., Stein, E., 2007. The quality of institutions and foreign direct investment. Economics & Politics 19 (3), 317344. Easterly, W., Levine, R., 2001. Its not factor accumulation: Stylized facts and growth models. NBER Working Paper Series, pp. 159. Esfahani, H., Ramirez, M., 2003. Institutions, infrastructure and economic growth. Journal of Development Economics 70, 443477. Fink, C., Mattoo, A., Rathindran, R., 2002. An Assessment of Telecommunications Reform in Developing Countries. World Bank Policy Research Working Paper. 2909, pp. 137. Galperin, H., 2005. Wireless networks and rural development: opportunities for Latin America. Information Technologies and International Development 2 (3), 4756. Gunasekaran, V., Harmantzis, F.C., 2007. Emerging wireless technologies for developing countries. Technology in Society 29 (1), 2342. Hoernig, S., Valletti, T., 2012. Mobile telephony. In: Peitz, M., Waldfogel, J. (Eds.), The Oxford Handbook of the Digital Economy. Oxford University Press, New York, pp. 136160. Howard, P.N., Mazaheri, N., 2009. Telecommunications reform, Internet use and mobile phone adoption in the developing world. World Development 37 (7), 11591169. ITU, 2008. Measuring information and communication technology availability in villages and rural areas. International Telecommunication Union (ITU), Geneva, Switzerland. Jensen, R., 2007. The digital provide: Information (technology), market performance and welfare in the south Indian sheries sector. The Quarterly Journal of Economics 122, 583597. Jensen, R., Oster, E., 2007. The power of tv: Cable television and womens status in India. NBER Working Paper Series, pp. 10571094. Jondrow, J., Lovell, C.K., Materov, I.S., Schmidt, P., 1982. On the estimation of technical inefciency in the stochastic frontier production function model. Journal of Econometrics 19 (23), 233238. Kaufmann, D., Kraay, A., Mastruzzi, M., 2008. Governance matters vii: Aggregate and individual governance indicators 19962007. Policy Research Working Paper. World Bank, pp. 1105. Kesternich, I., Schnitzer, M., 2010. Who is afraid of political risk? Multinational rms and their choice of capital structure. Journal of International Economics 82, 208218. Klenow, P.J., Rodriguez-Clare, A., 1997. The neoclassical revival in growth economics: has it gone too far? NBER Macroeconomics Annual 12, 73103. Kumbhakar, S., Lovell, C., 2000. Stochastic Frontier Analysis. Cambridge University Press, Cambridge, England. Li, W., Xu, L.C., 2002. The Impact of Privatization and Competition in the Telecommunications Sector Around the World. Darden Business School Working Paper No. 02-13, pp. 143. Lin, W.T., 2009. The business value of information technology as measured by technical efciency: evidence from country-level data. Decision Support Systems 46, 865874.

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 Lin, W.T., Chiang, C.Y., 2011. The impacts of country characteristics upon the value of information technology as measured by productive efciency. International Journal of Production Economics 132, 1333. Mishra, S.M., Hwang, J., Filippini, D., Moazzami, R., Subramanian, L., Du, T., 2005. Economic analysis of networking technologies for rural developing regions. Internet and Network Economics 3828, 184194. Perotti, E.C., Oijen, P.V., 2001. Privatization, political risk and stock market development in emerging economies. Journal of International Money and Finance 20, 4369. Political Risk Services, 2009. International Country Risk Guide. Published online. Qiang, Z., Rossotto, C., Kimura, K., 2009. Economic Impacts of Broadband. In: World Bank, Information and Communications for Development 2009: Extending Reach and Increasing Impact. Washington, USA. Roller, L., Waverman, L., 2001. Telecommunications infrastructure and economic development: a simultaneous approach. The American Economic Review 91, 909923. Sa, Y., 2011. Government size, economic growth and unemployment: evidence from advanced and developing economy countries (a time

141

series analysis, 19962006). International Review of Public Administration 16 (295), 95116. Shao, B.B.M., Lin, W.T., 2001. Measuring the value of information technology in technical efciency with stochastic production frontiers. Information and Software Technology 43, 447456. Stata Press, 2005. Stata Base Reference Manual, Release 9. Texas, United States of America. Symeou, P.C., 2011. Economy size and performance: an efciency analysis in the telecommunications sector. Telecommunications Policy 35 (5), 426440. Thompson, H., Garbacz, C., 2007. Mobile, xed line and Internet service effects on global productive efciency. Information Economics and Policy 19, 189214. Wallsten, S., 2002. Regulation and Internet Use in Developing Countries. World Bank Policy Research Working Paper No. 2979, pp. 124. Waverman, L., Meschi, M., Fuss, M., 2006. The impact of telecoms on economic growth in developing countries. Vodafone Policy Paper Series 1, 123. World Bank, 2007. World Development Indicators. C.D. World Bank Development Data Group.