d Ivem

, Kyot

Accepted 8 November 2012Available online 7 December 2012

Keywords:Technical efciency

countries and 18 high-income countries, and demonstrate that TE has increased in both

This study quanties the variations of TE in 23 low-income and 18 high-income countries between 1980 to2009 with respect to the deployment of two specic typesof infrastructure: mobile communication networks and

n estimatile transcenn and a te

efciency equation under the assumption of xed(Kumbhakar and Lovell, 2000). In order to reect thticular situation of low-income countries, we classifinto the following four geographical regions: Asia, LatinAmerica, Middle East and North Africa (MENA), andSub-Saharan Africa.

Our results are threefold. At rst, it is shown that thedeployment of mobile communication networks and

0167-6245/$ - see front matter 2012 Elsevier B.V. All rights reserved.

E-mail address: baquero.maria@ax7.ecs.kyoto-u.ac.jp

Information Economics and Policy 25 (2013) 126141

Contents lists available at SciVerse ScienceDirect

Information Econo

w.ehttp://dx.doi.org/10.1016/j.infoecopol.2012.11.004the determinants of income and TE most conducive to eco-nomic growth is required for policy formulation, in partic-ular in the context of low-income countries (Easterly andLevine, 2001).

bureaucracy quality).Our econometric method is based o

stochastic production frontier for a exiblogarithmic (translog) production functiong thedentalchnicaleffectse par-y thesecapital, labor force, and human capital, as well as by theefciency of input allocation, also known as the technicalefciency (TE). In this context, empirical research unveiling

ment effectiveness (i.e., the quality of public services andpolicy implementation), and political risk (i.e., the levelof economic planning, corruption, law enforcement, andmined by the stock of production inputs such as physical

Differences in income levels across countries are deter-

general type of infrastructure, the contribution of freightrailways is also taken into account. Furthermore, two insti-tutional performance measures are considered: govern-Mobile communicationsInternet technologiesStochastic production frontierGovernment effectivenessPolitical risk

1. Introductioncountry groups in view of the deployment of mobile communication networks and Internettechnologies. For low-income countries, we also prove that the causal relation is from thedeployment of mobile networks and Internet technologies towards the increase of TE.More specically, by estimating the stochastic production frontier for a exible transcen-

dental logarithmic production function under the assumption of xed effects, we show thatthe increase in TE per additional mobile phone and Internet subscriber is the highest inLatin American and Asian countries, but the accrued TE increase in response to Internetusage 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, partic-ularly in low-income countries.

2012 Elsevier B.V. All rights reserved.

Internet technologies. To address the impact of a moreArticle history:Received 2 March 2011Received in revised form 6 November 2012

Empirical research on the determinants of technical efciency (TE) is essential for policyformulation, in particular in low-income countries.In this study, we estimate the variations of TE between 1980 and 2009 in 23 low-incomeMobile communication networks andrivers of technical efciency impro

Maria del Pilar Baquero ForeroKyoto University, Graduate School of Economics, Yoshida-honmachi, Sakyo-ku

a r t i c l e i n f o a b s t r a c t

journal homepage: wwnternet technologies asent

o 606-8501, Japan

mics and Policy

l sevier .com/locate / iep

improvement in view of the deployment of Internet tech-

conrmed, since the relevant coefcients are not statisti-

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 127cally signicant.Lastly, we demonstrate further TE improvements in re-

sponse to (i) higher freight railways usage in all countries,(ii) improved government effectiveness in Europe and Sub-Saharan Africa, and (iii) lower political risk in high-incomecountries as well as, to a lesser extent, in MENA and Asia.

Our ndings highlight the importance of future deploy-ments of next-generation technologies for mobile commu-nications and broadband Internet. In this regard, weconclude by discussing effective telecommunication poli-cies, focusing especially on the context of low-incomecountries. Some of these policies include privatization,granting regulator independence, guaranteeing market ac-cess to new entrants, and regulating nal-user price.

The remainder of the paper is organized as follows. Sec-tion 2 surveys related previous studies on telecommunica-tions and economic growth, and Section 3 summarizes ourmain contributions. Section 4 introduces the employedstochastic production frontier method, while the paneldata and empirical model strategy are explained in Sec-tion 5. Section 6 presents the summary statistics of infra-structure and institutions of countries by income andregion. Section 7 presents the estimated TE coefcientsand rankings for low-income and high-income countries,as well as the model robustness checks. Finally, concludingremarks are offered in Section 8.

2. Previous work on telecommunications and economicgrowth

In general, previous works investigate the contributionof infrastructure deployment to national income, as wellas the causality of the relationship between infrastructureand income. To our best knowledge, research on the effectsof infrastructure on TE is still scarce, and studies on thecausality relation between infrastructure and TE have notbeen addressed previously.

2.1. Telecommunications as determinants of national incomelevel

National income level and GDP growth have beenshown to partially depend on various types ofnologies is the highest in high-income countries owing toan overly better penetration during the analyzed period.

Secondly, it is shown that the deployment of mobilecommunication networks and Internet technologies causesthe estimated TE improvement in low-income countries,i.e. the causal relation is from the diffusion of mobile com-munication networks and Internet technologies towardsthe TE improvement in low-income countries. For high-income countries, the causality between telecommunica-tion technologies and TE improvement could not beInternet technologies improves TE in both low-income andhigh-income countries. In particular, the TE improvementper additional subscriber to mobile communication net-works and Internet technologies is the highest in LatinAmerican and Asian countries. However, the overall TEinfrastructure. In high-income countries, these includemainlines (Roller and Waverman, 2001), mainlines andtransportation (Esfahani and Ramirez, 2003), mobile com-munication networks (Waverman et al., 2006), informationtechnologies (Lin and Chiang, 2011; Lin, 2009; Shao andLin, 2001), and broadband Internet (Qiang et al., 2009,Czernich et al., 2011). In low-income countries, animprovement of market efciency has been demonstratedin view of the deployment of information and communica-tion 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 relationbetween income levels and infrastructure in developedcountries.

In the context of the causality between telecommunica-tions and income, it is well known that the relationship isof a two-way nature (Roller and Waverman, 2001). On onehand, infrastructure investments lead to a higher income,but on the other hand, income raises the demand for infra-structure and induces a greater supply. The mutual cou-pling makes it difcult to dene the dominating causality.

Some country-level studies examined the causality be-tween 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 modelsthat estimate simultaneous equations of demand and sup-ply, but Roller and Waverman (2001) only addresses thecase of developed countries for which there are availabledata on infrastructure investment. In comparison, Czernichet al. (2011) employ pre-existing networks (voice tele-phony and cable TV) as instruments for the supply ofbroadband in order to isolate the causal effect of broad-band Internet on economic growth in OECD countries.

However, neither the determinants of TE nor their cau-sal relation to TE have been addressed in the above studies.The focus on telecommunications infrastructure in bothdeveloped and developing countries as well as the investi-gation of the role of institutions and their quality are miss-ing entirely.

2.2. Telecommunications as determinants of technicalefciency

Few recent studies have attempted to identify severalinfrastructure/institution types as the determinants ofcountry TE. Thompson and Garbacz (2007) used the sto-chastic frontier model based on a Cobb-Douglas functionto show that improved telecommunications and economicfreedom increase TE, especially in Africa and Latin America.However, the employed econometric model suffers froman unrealistic assumption of errors being uncorrelated tothe regressors, so-called random effects. This assumptionmay cause the by-denition positive coefcients of humancapital to become negative. Thus, the credibility of the ob-tained results is weak (Kumbhakar and Lovell, 2000).

Using a similar model, Lin and Chiang (2011) havefound that Eastern European countries gain more produc-tive efciency than the G7 countries if the IT capital is con-sidered as a production input. The study also analyzes the

impact on TE of national savings, external balance, rmreturns on assets (ROA), and vertical integration. However,the authors do not take into account the impact of IT as adriver of TE improvement. Instead, IT is considered as a

128 M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141production input. IT investment as a production input isstudied by using the stochastic frontier method also byLin (2009) and Shao and Lin (2001).

2.3. General infrastructures and institutions as determinantsof technical efciency

The present study investigates cross-country TE varia-tions not only with respect to telecommunications but alsoto more general types of infrastructure such as transporta-tion, as well as institutional performance parameters suchas government effectiveness and political risk.

Our motivation stems in the fact that government effec-tiveness, i.e. the policy quality as dened subsequently, isknown to affect economic growth in both developed anddeveloping countries (Sa, 2011), as well as in transitioneconomies (Beck and Laeven, 2006). Moreover, the politicalrisk index used in our analysis (encompassing, amongother factors, law enforcement, corruption, and bureau-cracy quality) is argued to be a determinant of the nancialperformance of multinational companies (Kesternich andSchnitzer, 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 govern-ment effectiveness (Daude and Stein, 2007) and politicalrisk levels (Busse and Hefeker, 2007).

In this regard, none of the above-mentioned studies ontelecommunications and TE considers in their analysissuch more general types of infrastructure (especially,Thompson and Garbacz, 2007; Lin and Chiang, 2011; Lin,2009; Shao and Lin, 2001), or institutions (especially, Linand Chiang, 2011; Lin, 2009; Shao and Lin, 2001).

3. Contributions

In this study, we estimate the effects of various kinds ofinfrastructure and institutions on TE in both low-incomecountries and high-income countries. To this end, we usea stochastic frontier method similarly to Thompson andGarbacz (2007), but assume xed effects and a exibletranslog (generalized Cobb-Douglas) production function.

More specically, we conduct a joint analysis of therelationship between TE and (i) the deployment of mobilecommunication networks and Internet technologies, (ii)the usage of freight railways, and (iii) the degree of govern-ment effectiveness and political risk.1

For both the low-income and high-income countries,we address in our statistical model also the causality issuebetween TE and telecommunications by additionally con-trolling for lagged telecommunications variables, namelyfor the past values (lagged 3 years) of the diffusion of mo-bile communication networks and Internet technologies.

1 The political risk factor used in this study is a general measure ofinstitutional qualities calculated based on a weighted average of thirteenindicators, including economic planning failure, corruption, law enforce-ment, and bureaucracy quality. For a complete list, see Table 3.4. Modeling approach using stochastic productionfrontier

The stochastic production frontier approach employedin this study denes technical efciency (TE) as the ratioof actual observed output to the maximum feasible output,known as the stochastic production frontier. Since we al-low TE to be time variant, a TE change indicates the degreeof waste reduction achieved over time in the allocation ofinputs required for obtaining maximum outputs (Kumbha-kar and Lovell, 2000).

Roughly speaking, in order to measure TE, we estimatethe stochastic production frontier, or the maximum feasi-ble output, corresponding to the observed output of coun-tries in our sample data. The stochastic production frontieris estimated by adding a random error vit and a non-nega-tive random error uit to a standard production function.The random error uit indicates the difference between ob-served output and the maximum feasible output, i.e. thetechnical inefciency. The time-variant TE is then esti-mated as the conditional expectation of uit, given the com-posed error it = vit uit.

More specically, we employ the panel-data version ofthe xed-effects stochastic production frontier with time-variant technical inefciency effects (Stata Press, 2005)that is specied as follows:

ln yit ai RKj1bj ln xjit v it uit ai RKj1bj ln xjit it 1

where

ai is the xed effect of the ith country; b is a K 1 vector of unknown scalar parameters to beestimated;

lnxit is a K 1 vector consisting of natural logarithms ofproduction inputs of the ith country at the tth timeperiod;

lnyit is the natural logarithm of the observed productionoutput 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;r2v

;

uit is a non-negative variable, associated with the tech-nical inefciency of the output production.

The term vit encompasses the measurement error andother random factors that affect the output yit. Also, it ac-counts for the combined effects of unspecied input vari-ables in the production function. Consequently, the termRKj1bj ln xjit v it constitutes the stochastic productionfrontier representing the maximum feasible output.

The term uit represents the amount by which the ob-served countries fail to reach the frontier. For the ith coun-try, the time-variant technical inefciency effects uit areindependently truncated-normally distributed with a trun-cation point at l N l;r2u

.

The conditional mean model estimates l assuming it isa linear function of a set of covariates. The modeled condi-tional mean is given by l = c0 + czit where zit is a vector ofobservable non-random explanatory variables. These ob-servable variables are expected to explain the reduction

of technical inefciency across countries. The zit variablesinclude the diffusion of mobile communication networksand Internet technologies, freight railways usage, govern-ment effectiveness, and political risk.2

it

expectation of uit, given the value of it = vit uit. The con-

the stochastic production frontier is given by the real gross

The data employed as production inputs xit consists ofthe physical capital stock, total labor force, and humancapital. The physical capital stock in constant marketprices from 1980 to 2000 was calculated on the basis ofthe above real GDP and the physical-capital-to-output ra-tio obtained from Klenow and Rodriguez-Clare (1997).For the period between 2001 and 2009, the physical capitalstock was calculated on the basis of the methodology de-scribed in Klenow and Rodriguez-Clare (1997) by usingthe annual gross capital formation from theWord Develop-ment Indicators (WDI) database. The total labor forcewhich includes all persons engaged in some productivityactivity was extracted from the employment data of theabove mentioned CB-TED. Finally, the human capital de-ned as years of educational attainment for populationaged between 15 and 64 not enrolled in education was ta-ken from the Cohen and Soto database.

5.1.2. Technical efciency equation: infrastructure andinstitutions

The infrastructure and institutional data used asexplanatory variables zit of TE as well as their source are

Table 1List of countries by income and region.

Belgium Korea Algeria ChinaDenmark Egypt India

3 We thank the anonymous Referee for suggesting the use of theConference Board Total Economy Database.

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 129ditional expectation of uit is used since the actual value ofuit is unobservable.

It is noteworthy that even if the real value of the param-eter vector b in the stochastic frontier model was known,only the difference it = vit uit could be observed. For theparticular distributional assumptions imposed on the tech-nical inefciency effects, the expected value of TEit can becalculated by using the equation

Eit Eexpuit jit

1Ur lit=r1Ulit=r

exp lit

12r2

2

by Jondrow et al. (1982) where

lit itr2u lr2v

r2S3

r rurvrS 4

rS r2u r2v 1

2 5The stochastic production frontier is estimated by using

maximum likelihood estimation (MLE) in a one-step proce-dure 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-incomeand 18 high-income) for the period between 1980 and2009. The considered low-income countries are classiedinto four regions: Asia, Latin America, Middle East-NorthAfrica (MENA), and Sub-Saharan Africa. The list of coun-tries sorted by income and region is given in Table 1. Thedenitions and sources of the data employed to estimatethe stochastic production frontier and TE, namely produc-

2 Compared to the half-normal case, the truncated-normal distributionhas an additional parameter l to be estimated, since it allows for a non-zero mode. Therefore, it provides a more exible representation of technicalefciency.The assumption of xed effects implies that the termsvit are uncorrelated with the regressors, as well as thatthe terms uit are allowed to be correlated with the regres-sors and with vit. The main advantage of the xed-effectsmodel is that it is simple while being characterized bydesirable consistency properties (Kumbhakar and Lovell,2000). A disadvantage stems in that the model capturesthe effects of all phenomena that vary across countriesbut are time invariant for each country. It occurs whetheror not these other effects are included as regressors inthe model.

To estimate the actual technical efciency (TE) of indi-vidual countries, the best predictor for u is the conditionaldomestic product (GDP) at constant market prices as ob-tained from the Conference Board Total Economy Database(CB-TED).3tion output, inputs, infrastructure and institutions is givenin Table 2.

5.1.1. Production function: output and inputsThe data used as the production output yit to estimate

Finland NORTHAMERICA

Iran Indonesia

France (2) Jordan MalaysiaGreece Canada Syria ThailandIreland United States TunisiaItaly TurkeyNetherlandsPortugal OCEANIA SUB-SAHARAN

AFRICALATINAMERICA

Spain (1) (6) (4)Sweden Australia Cameroon ArgentinaUnited

KingdomGhana Chile

Kenya MexicoMalawi PeruSenegalZambiaHigh-income countries (18) Low-income countries (23)

EUROPE ASIA MIDDLE EASTAND

ASIA

(13) (2) NORTH AFRICA (6)Austria Japan (7) Bangladesh

Table 2Data sources 19802009.

Variable Denition

Stochastic production functionY Real Gross Domestic Product (GDP) in constant mar

sont pop

nts

ns k

apital to output ratio k/y obtained from Klenow et al. and the above real GDP.gy described in Klenow and Rodriguez-Clare (1997, p. 78) and using the Worldon.

rvice, as well as the latters degree of independence from political pressures, thehe governments commitment to such policies. (Kaufmann et al., 2008).

ints to a group of pre-set components: government stability, socioeconomicon, military in politics, religious tensions, law and order, ethnic tensions, dem-9).(PRS) group.

Table 3Political Risk Factor (PRF)a composition.

Component Points

Economic expectations vs. reality 12Economic planning failures 12Political leadership 12

130 M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141as follows. Mobile phone and Internet subscribers per 100inhabitants were obtained from the InternationalTelecommunication Union (ITU) database. Freight railwaysusage is dened as goods, in million tons, transported byrailways per kilometer and were taken from the WDI data-

K Physical capital stock in constant market priceL Total labor force economically active populatiE Human capital years of educational attainmenTechnical efciency functionM Mobile telephone subscribers per 100 inhabitaI Internet subscribers per 100 inhabitantsR Freight railways goods transported (million toGE Government effectivenesse

2.5 (highly effective) 2.5PRF Political risk factorg

100 (minimum risk) 0

a Conference Board Total Economy Database (CB-TED).b Physical capital from 1980 to 2000 was calculated from the physical c

Physical capital from 2001 to 2004 was calculated based on the methodoloBank (2007) (World Development Indicators) annual gross capital formati

c International Telecommunication Union (ITU).d World Development Indicators (WDI).e It reects perceptions of the quality of public services and of the civil se

quality of policy formulation and implementation, and the credibility of tf World Governance Indicators (WGI).g It represents the political stability per country by assigning risk po

conditions, investment prole, internal conict, external conict, corruptiocratic accountability and bureaucracy quality (Political Risk Services, 200

h International Country Risk Guide (ICRG) by The Political Risk Servicesbase. To make the freight railways usage data comparableacross countries of various sizes, the railways data were di-vided by each countries total area, also taken from the WDIdatabase.

The government effectiveness indicator combines theperception of the public/civil service quality, the degreeof service independence from political pressures, the qual-ity of policy formulation/implementation, and the credibil-ity of the government commitment to its policies(Kaufmann et al., 2008). The indicator values range from2.5 (less effective) to 2.5 (more effective). The corre-sponding data originates from the Worldwide GovernanceIndicators (WGI) database.4

The political risk factor (PRF) (Political Risk Services,2009) takes values between 100 (lower risk) and 0 (higherrisk). It is a general measure of institutional performancethat incorporates economic risk perceptions such as eco-nomic planning failure and economic expectations vs. real-ity, as well as crucial political variables such as politicalleadership, corruption, law enforcement, quality of bureau-cracy, among other religious or racial-related indicators.The corresponding data was extracted from the Interna-tional Country Risk Guide (ICRG) of the Political Risk

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

ket prices CB-TEDa

Klenow et al. WDI and CB-TEDb

CB-TEDulation 1564 Cohen-Soto database (2000)

ITUc

ITUm/total country area) WDId

WGIf

ICRG PRShServices (PRS) group. The complete list of indicatorscomposing the political risk factor is shown in Table 3.

The two institutional measures used in this paper havebeen analyzed in previous studies on cross-country eco-nomic performance. The government effectiveness indica-tor 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 effec-tiveness indicator (Kaufmann et al., 2008) argue that, com-pared to alternative indicators used in the literature, thesedata are free of ideological and other inuences. Otherauthors considered the political risk factor, used in thisstudy, in their research on multinational companies

External conict 12Corruption in government 10Military in politics 6Organized religion in politics 6Law and order tradition 6Racial and national tensions 6Political terrorism 6Civil war risks 6Political party development 6Quality of bureaucracy 6

Total 100

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

North Africa (MENA), and Sub-Saharan Africa.

values of Internet subscribers (3 lags) in low-income re-gions and high-income countries, respectively.5

Our hypothesis is that mobile communication net-works, Internet technologies, freight railways, governmenteffectiveness indicator, and political risk factor should havenegative signs if they contribute to lower cross-countrytechnical inefciency, in other words if they help to in-crease TE. We also expect various degrees of TE improve-

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 131Models 3 and 4 incorporate mobile phone diffusion andlagged values of mobile diffusion (3 lags) in low-incomeregions and high-income countries, respectively. Finally,Models 5 and 6 include Internet subscribers and lagged(Kesternich and Schnitzer, 2010), stock markets (Perottiand Oijen, 2001; Bilsona et al., 2002), and FDI (Busse andHefeker, 2007).

5.2. Empirical model specication using a translog function

In this paper, we estimate a total of six stochastic pro-duction frontier models in order to measure the contribu-tion of infrastructure and institutions on cross-country TE,as well as to unveil the underlying causal relation betweentelecommunications and TE. Models 1 and 2 are used toestimate the contribution of telecommunications, trans-portation and institutions to a TE reduction in high-incomeand low-income country regions. Models 36 are used todetermine the causal relation between telecommunica-tions and TE in high-income countries and low-incomecountry regions.

The stochastic production frontier is estimated in atranslog functional form as this form is sufciently generaland exible, and does not impose any assumptions on theelasticities of production, nor on the elasticities of substi-tution among production inputs. The six models differ intheir efciency equation but estimate the same translogproduction function, having the real GDP as the observedproduction output yit, and the physical capital stock, laborforce and education as production inputs xit. Hence, theestimated translog production function is

lnyit bi bK lnKit bL lnLit bE lnEit bKL lnKit lnLit bKE lnKit lnEit bLE lnLit lnEit 1=2bKK ln K2it

1=2bLL

ln L2it

1=2bEE ln E2it

v it uit 6

The translog production function is characterized byallowing a non-linear relationship between the productionoutput and inputs. Specically, the function estimatesparameters of input quadratic terms (bKK, bLL and bEE) aswell as parameters of input interaction, dened as the pair-wise-products between inputs (bKL, bKE and bLE). In order todene the point of approximation to the function, the out-put and input data of the production function are normal-ized by the sample mean before applying the logarithmictransformation (Chung, 1994).

As for the data used as zit to model the mean of uit in thetechnical efciency equation, all six models consider thefreight railways usage, government effectiveness indicator,and political risk factor.

In addition to the above, Model 1 considers mobilephone subscribers, and Model 2 takes into account Internetsubscribers. To measure separately the TE change of high-income and low-income countries with respect to theabove mentioned infrastructure and institutions, bothModels 1 and 2 include four regional dummy variables oflow-income countries: Asia, Latin America, Middle East-ment by income levels and regions. The laggedtelecommunication variables should have negative signsif they indicate the causality from telecommunication dif-fusion towards technical inefciency reduction, i.e. an in-crease of TE.

6. Summary statistics

The summary statistics of real GDP, inputs (physicalcapital, labor and human capital), infrastructure (mobilephone subscribers, Internet subscribers and freight rail-ways usage), and institutional performance indicators(government effectiveness and political risk) for all consid-ered high-income and low-income countries in 2009 isshown in Table 4. The summary statistics for high-incomecountries and low-income country regions is given in Ta-bles 5 and 6, respectively.

A comparison of infrastructure and institutional qualitybetween high-income and low-income countries in 2009 isoffered in the following in conjunction with a brief analysison telecommunication policies of selected countries.

6.1. Mobile communication networks

Low-income countries in our sample have in average 71mobile phone subscribers per 100 inhabitants in 2009while high-income countries had 117 mobile phone sub-scribers per 100 inhabitants. Within low-income countries,all Latin American countries have the penetration ratesabove 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 networksshows a great disparity in our sample, specially in Asia.However, the penetration rate of mobile phones is corre-lated with the degree of market competition as indicatedby (Symeou, 2011; Fink et al., 2002; Wallsten, 2002; Liand Xu, 2002). This observation is validated in our analysisfor the case of Sub-Saharan African and MENA countries. Inparticular, the lack of competition and poor economic per-formance have left Malawi, Cameroon, and Syria laggingbehind in terms of mobile penetration (15.7, 37.8, and45.9, respectively), as well as delaying the introduction ofthird-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 inthis study. Although the estimated coefcients of both variables had theexpected signs, they were not statistically signicant because of the highcorrelation between them (0.89).

6 According to the Balancing Act Africa, a third mobile operator waslicensed in Malawi in 2008, but a delay in the network roll out causes themarket to effectively remain in a duopoly condition (http://www.balanc-ingact-africa.com/news/en/issue-no-601, accessed on August 30th, 2012).

Low-income countries High-income countriesNo. obs: 23 No. obs: 18

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

790.7 1935.9 10.1 8965.7 1152.6 2145.4 106.9 9315.2

9.56e5 2.24e6 1.16e4 1.04e 1.85e6 3.11e6 1.04e5 1.27

7.27e4 1.80e5 1960 7.77e5 2.17e4 3.35e4 1927 1.41e5

7.1071.09

20.5

0.02

0.1

63.26

132 M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141Table 4Summary statistics in 2009 by income.

Variable Units All countriesNo. obs: 41

Mean Std.dev. Min Max

GDP(CBTED)

billions1990US$

949.6 2012.6 10.1 9315.2

K (CBTED) millions1990US$

1.38e6 2.66e6 1.16e4 1.27e

L (WDI) thousands ofpeople

5.03e4 1.38e5 1927 7.77e5

H (C-S) years 9.16 2.92 2.93 13.32Mobile

subs.100 inhab 91.25 35.17 15.72 150.47

Internetsubs.

100 inhab 43.46 29.84 0.38 90.28

Railways goods km/area

0.08 0.17 0 0.79

Gov. eff.index

2.5 to 2.5 0.56 0.95 0.99 2.19

PRF 0100 71.77 11.47 53 92.5

Table 5Summary statistics in 2009 by region (high-income countries).

Variable Units EuropeNo. obs: 13On the other hand, there are countries in Latin Americaand Asia with competitive markets and mobile phone pen-etration rates higher than the average of high-income coun-tries Argentina with 128.8 subscribers per 100inhabitants, and Thailand with 122.6 subscribers per 100inhabitants. The high diffusion rate of mobile communica-tions in Argentina ismainly due to privatization andmarketderegulation that allowed the participation of new mobileoperators, while the case of Thailand can be explained bya high Telecommunications Regulatory Environment(TRE) index, indicating tariff regulation, increased competi-tion, and relative easiness of a market entry.

In both country groups, mobile phone users exceedmainlines subscribers. However, mobile phone diffusion isspecially important in low-income countries, where xedtelephony penetration rates are remarkably low. The big-gest differences between mobile and xed telephony diffu-sion rates are in countries of Sub-Saharan Africa such asGhanawith 63.4 mobile phone subscribers per 100 inhabit-ants and only 1.11 xed line subscribers per 100 inhabit-ants, and Senegal with 55.0 subscribers per 100inhabitants and just 2.32 xed line subscribers per 100inhabitants, as well as MENA countries such as Algeriaand Jordan 93.7 and 101.1 mobile phone subscribers per100 inhabitants versus merely 7.53 and 7.99 land line

Mean Std.dev. Min

GDP (CBTED) billions 1990US$ 489261.4 492515.9 106,920K (CBTED) millions 1990US$ 784590.6 822259.1 104,020L (WDI) thousands of people 10411.15 9837.78 1927H (C-S) years 11.23 1.31 7.85Mobile subs. 100 inhab 126.27 16.23 95.09Internet subs. 100 inhab 71.07 16.14 44.05Railways goods km/area 0.059 0.073 0.001Gov. eff. index 2.5 to 2.5 1.43 0.55 0.51PRF 0100 82.92 5.71 73Other high-incomeNo. obs: 51.95 2.93 10.72 11.78 1.43 7.85 13.3230.17 15.72 128.84 117.02 21.89 68.41 150.47

14.6 0.38 57.61 72.81 13.95 44.05 90.28

0.04 0 0.19 0.15 0.24 0 0.79

4 0.56 0.99 1.21 1.44 0.49 0.52 2.19

6.73 53 79.5 82.64 5.26 73 92.5subscribers per 100 inhabitants, respectively. It isnoteworthy that the mentioned exemplary countries arecharacterized by mobile market liberalization, good tele-communications infrastructure, and innovative services.7

6.2. Internet technologies

High-income countries in our sample in 2009 have anaverage of 72.8 Internet users per 100 inhabitants whilelow-income countries are characterized only by 20.5 usersper 100 inhabitants. As noted in the context of the mobilephone diffusion, the highest average Internet penetrationrate within low-income countries can be found in Latin

Max Mean Std.dev. Min Max

1,384,657 2,877,466 3,699,330 538,735 9,315,2192,487,902 4,608,526 5,074,374 792653.6 12,668,51227,123 51178.8 54458.94 10,845 141,46613.32 13.20 0.08 13.07 13.29150.46 92.95 15.60 68.40 110.7290.27 77.30 3.23 72.03 80.900.241 0.37 0.38 0.02 0.792.19 1.45 0.29 1.11 1.7792.5 81.9 4.32 77.5 87

7 As reported by the BuddeComm consultancy, Ghana is the regionalpioneer having launched the rst cellular mobile network in Sub-SaharanAfrica in 1992, and privatized the telecommunications incumbent as earlyas in 1996. Similarly, the incumbent operator in Senegal has been offeringthe most efcient network and the lowest mobile prices in West Africa, aswell 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 awell developed infrastructure and price competition between three mobileoperators. Despite its low GDP per capita, Jordan has a relatively welldeveloped telecoms sector and the most liberalized market in the regionowing to an intense competition in the mobile market that has led to highsubscriber numbers and reduced prices.

America (29 users per 100 inhabitants) while the lowestaverage penetration rate is in Sub-Saharan African coun-tries (6 users per 100 inhabitants).

We comment on two cases of successful Asian countries Malaysia and Thailand. Malaysia has widely deployed ad-vanced Internet infrastructures such as ber optics net-works thanks to a government effort since 1991.Remarkably, the Internet penetration rate of 57.6 usersper 100 inhabitants in Malaysia is higher than in thehigh-income countries of Greece (44), Portugal (48.6),and Italy (48.5). As for Thailand, it has had a strong Internetgrowth in recent years due to the liberalization of theInternet gateway market.

Except for the above cases, our sample data indicatethat Asian and Sub-Saharan African countries are generallylagging behind in the deployment of Internet technologies,and have the lowest Internet penetration rates (e.g., 0.4users per 100 inhabitants in Bangladesh, 3.8 in Cameroon,4.7 in Malawi, 5.44 in Ghana, and 5.31 in India). The Inter-net penetration rates are very low in these countries owing

Table6

Summarystatistics

in20

09by

region

(low

-incom

ecoun

tries).

Variable

Units

MiddleEast

North

Africa

Sub-Saharan

Africa

Asia

LatinAmerica

No.

obs:

7No.

obs:

6No.

obs:

6No.

obs:

4

Mean

Std.de

v.Min

Max

Mean

Std.de

v.Min

Max

Mean

Std.de

v.Min

Max

Mean

Std.de

v.Min

Max

GDP(CBTE

D)

billions19

90US$

2.50

e52.09

e53.44

e45.87

e52.43

e41.44

e41.01

e44.38

e42.45

e63.44

e61.86

e58.97

e64.02

e53.02

e51.54

e58.21

e5K(CBTE

D)

million

s19

90US$

3.36

e52.90

e53.66

e47.40

e54.25

e45.21

e41.16

e41.46

e42.90

e63.95

e61.64

e51.04

e75.02

e54.35

e51.93

e51.14

e6L(W

DI)

thou

sands

ofpe

ople

1.50

e41.10

e419

602.84

e48.97

e35.22

e34.82

1.86

e42.40

e53.11

e51.09

e47.77

e51.92

e41.64

e472

774.34

e4H(C-S)

years

7.35

1.53

5.15

10.18

5.26

1.30

2.93

6.49

7.24

1.99

4.97

10.15

9.20

1.03

8.39

10.71

Mob

ilesubs.

100inhab

79.57

19.26

45.97

101.06

42.46

16.97

15.72

63.38

72.50

36.56

31.06

122.56

97.05

22.62

77.75

128.83

Internet

subs.

100inhab

26.93

9.53

13.46

38.28

6.28

2.21

3.84

10.03

21.10

21.20

0.38

57.60

29.64

3.32

26.47

33.98

Railw

ays

good

sk

m/area

0.01

0.00

90.00

30.02

0.00

30.00

20.00

020.00

70.04

0.07

0.00

050.18

0.03

0.02

0.00

30.06

Gov.eff.inde

x2

.5to

2.5

0.17

0.50

0.74

0.41

0.49

0.31

0.80

0.06

40.00

70.63

0.98

0.98

0.23

0.69

00

.42

1.20

PRF

010

062

.07

6.63

55.5

7261

.25

4.63

56.5

6762

.25

6.77

5372

69.87

7.76

6279

.5

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 133to an underdeveloped infrastructure as well as lack of priv-atization and regulation.8 Nevertheless, the recent WiMAXand Wi-Fi technologies for mobile Internet access could mit-igate the limitations of country infrastructure since they cancost-efciently provision last-mile broadband Internetaccess in remote or rural areas in low-income countries(Galperin, 2005; Mishra et al., 2005; Gunasekaran andHarmantzis, 2007). According to the Buddecom consultancy,Bangladesh, Cameroon, and Malawi have already grantedseveral licenses to WiMAX operators.

In the MENA region, some countries such as Turkey,Tunisia, and Jordan have relatively high Internet penetra-tion rates compared to other low-income countries (35.3,33.5, 29.3 users per 100 inhabitants, respectively), mainlythanks to regulatory reforms, market competition, andinfrastructure investments,9 but other countries such asAlgeria (13.5 users per 100 inhabitants) and Syria (18.6 usersper 100 inhabitants) have the lowest penetration rates, pri-marily due to regulatory barriers limiting the existing infra-structure in achieving its potential.10

8 According to Pyramid Research, Bangladesh still has to develop its ICTinfrastructure in order to promote Internet diffusion. Cameroon has notcompleted the privatization of its incumbent operator and has not allowedthe entry of new operators by allocating additional licenses. Despite theexistence of 15 licensed Internet Service Providers, Malawi is characterizedby a low Internet penetration rate due to limited service availability andhigh cost of international bandwidth, as well as insufcient regulatoryregime.

9 In Turkey, regulatory reform has helped improve competition anddevelop the Internet market. Tunisia has one of the most developedInternet markets and some of the lowest broadband prices in North Africadue to competition between eleven ISPs, access to a nationwide ber opticbackbone network and international access via submarine and terrestrialber. Finally, the Internet broadband market in Jordan is relativelycompetitive due to infrastructure-based competition and the introductionof WiMAX services. In addition, the quality of broadband services in Jordanis expected to improve through the deployment of two new internationalterrestrial cables.

10 Despite its well-developed infrastructure that includes a national berbackbone and one of Africas rst ber-to-the-home deployments, the lowInternet penetration rate in Algeria stem from regulatory barriers hinderingcompetition. In Syria, the Internet penetration rate is very low becausebroadband services are expensive and difcult to subscribe to.

In Latin America, low Internet penetration rates in Peru(27.7 users per 100 inhabitants) andMexico (26.5 users per100 inhabitants) are the result of market restrictions, low

Australia and Austria. Slightly over the level of Finland

frontier models are given in Tables 8, 12, and 13. We ob-

instead of less exible functional forms such as the Cobb-Douglas function.

The output elasticity of inputs calculated based on theestimation results of Models 1 and 2 are given in Table 7.The highest output elasticity among the three productioninputs is that of human capital, ranging between 0.80

Table 7Input elasticities of output.

Input elasticities ofoutput

Model 1 mobilephones

Model 2Internet

Physical capital 0.313 0.311Labor 0.306 0.243Human capital 0.804 0.871

Fig. 1. Estimated technical efciency improvement from 1980 to 2009 byregion.

134 M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141tained positive and statistically signicant coefcients ofphysical capital, labor, and education in all the six esti-mated models. The majority of coefcients of input inter-action (pair-wise products) is also statistically signicantwhich validates the choice of translog production function

11 According to the ITU Report (2008), Peru is a world leader in publicInternet access with 56% inhabitants accessing the Internet from publicplaces.and France, China is the low-income country with thehighest use of railways used for cargo transports. High-in-come countries such as Portugal and Greece as well as themajority of low-income countries are characterized by aminimum usage of railways for the transportation ofgoods.

The average of the government effectiveness indicatorranging from 2.5 (the lowest effectiveness) to 2.5 (thehighest effectiveness) is 0.13 for low-income countriesand 1.4 for high-income countries in 2009. The most ef-cient countries within the high-income group are Denmark(2.2), Finland (2.1), and Sweden (2) while the least efcientgovernments in the same group are Spain (0.9), Greece(0.6), and Italy (0.5). As for the low-income countries, theworst average government effectiveness was obtained forSub-Saharan Africa (0.49) while Latin America is topranking (0.23).

The average political risk factor of high-income coun-tries in 2009 is 82.6. In this group, the countries havingthe lowest political risk are Finland (92.5), Austria (90),and Sweden (88.5). In contrary, the highest political riskcharacterizes Spain (77), and Greece (73). In comparison,low-income countries have the average political risk factorlower by 25% (63.2). Within low-income countries, LatinAmerica has a slightly lower average political risk (70)while MENA, Sub-saharan Africa, and Asia have the sameaverage of 62 points.

7. Estimation results

7.1. Production function estimates and input elasticities ofoutput

The estimation results of the six stochastic productioncomputer ownership, and low competition. In this context,it is noteworthy that subsidies to the poorest householdsincrease computer ownership while the provisioning ofpublic Internet access enhances Internet penetration (ITUReport, 2008).11

6.3. Freight railways usage, government effectiveness, andpolitical risk

In terms of the freight railways usage by region, theUnited States and Canada are the countries that trans-ported the most goods by railways in 2009, followed by and 0.87. Physical capital has a higher output elasticity

(0.31) compared to the elasticity of labor (between 0.24and 0.30). Thompson and Garbacz (2007) also obtained ahigher elasticity of capital (0.801) compared to the elastic-ity of labor (0.167) for high-income countries but also ob-tained a higher elasticity of labor (0.517) compared to theelasticity of capital (0.458) for low-income countries.

The results by Thompson and Garbacz (2007) also indi-cate a negative output elasticity for human capital for bothlow-income countries (0.0067) and OECD countries(0.095). The difference between the output elasticity ofinputs of our study and that of Thompson and Garbacz(2007) is partly due to different functional forms (translogvs. Cobb-Douglas). Moreover, Thompson and Garbacz(2007) consider a time trend which is not statistically sig-nicant, except for low-income countries and Africa.

7.2. Technical efciency estimates

The estimated TE changes per country/region during theanalyzed period 19802009 are shown in Fig. 1. For high-income countries, the average TE improved by 6% duringthe rst decade, from 72.1% in 1980 to 76.5% in 1990.The most remarkable increase was obtained for the follow-ing decade when TE rose by 11% and reached 84.9% in2000. In the last decade, an increase of 8% followed andTE attained 91.6%. In case of low-income countries, theaverage TE grew only by 7.2% in the rst two decades, from48.7% in 1980 to 52% in 2000. In contrast, the estimated TEincrease was the highest (16.8%) from 2000 to 2009 withTE reaching 60.1%.

Within the low-income country regions, Latin Americanand Asian countries are characterized by the highest TElevels and the most signicant change. During the last dec-ade of the studied period, TE levels rose by 24.3% rise from58.7% to 73% in Latin America. In Asia, they rose by 21.7%from 55% to 67.6%. In contrary, the average TE of Sub-Sah-aran African and MENA countries grew at a slower rate of13% and 10%, respectively, during the same decade.

In both country types, a notable increase in TE tookplace when mobile communication networks and Internettechnologies became the most widespread (late 1980s inhigh-income countries, and after the year 2000 in low-income countries).

The following subsections discuss in more detail thespecic contribution of telecommunication technologies,transportation, and institutions to the changes in TE inthe studied countries.

7.3. Technical efciency improvement by telecommunicationsand their relation of causality

According to our estimation results summarized inTable 8, the reduction in technical inefciency for high-income countries is 0.002 for an additional percentagepoint in mobile phone users, and 0.003 for an additionalpercentage point in Internet users. For Asian and Latin

Table 8Technical efciency reduction by mobile phones and Internet subscribers.

Stochastic frontier truncated-normal model

Model 1 mobile phones Model 2 Internet

Primary index equation parametersb0 0.89 (0.072)a b0 0.719 (0.072)bCapital 0.386 (0.018) bCapital 0.393 (0.018)bLabor 0.237 (0.064) bLabor 0.125 (0.057)bEducation 1.109 (0.074) bEducation 1.163 (0.083)bCapital labor 0.021 (0.015) bCapital labor 0.013 (0.015)bCapital education 0.032 (0.020) bCapital education 0.015 (0.021)bLabor education 0.179 (0.032) bLabor education 0.216 (0.031)bCapital2 0.035

(0.006) bCapital2 0.036 (0.006)

bLabor2 0.021 (0.012) bLabor2 0.045 (0.012)bEducation2 0.132

(0.043) bEducation2 0.161 (0.044)

Offset [mean = lit] parameters in one-sided error uitc0 0.923 (0.083) c0 0.788 (0.077)

cMobile phones 0.002 (0.0002) cInternet 0.003 (0.0002)cMobMENA 0.0006 (0.0003) cIntMENA 0.0001 (0.001)cMobSUBSAH 0.006 (0.001) cIntSUBSAH 0.038 (0.008)cMobLATIN 0.002 (0.0005) cIntLATIN 0.006 (0.001)cMobLowASIA 0.001 (0.0003) cIntLowASIA 0.003 (0.0008)cRailways 0.127 (0.074) cRailways 0.435 (0.094)cRailMENA 0.108 (0.056) cRailMENA 0.180 (0.054)cRailSUBSAH 0.262 (0.072) cRailSUBSAH 0.356 (0.071)cRailLATIN 0.213 (0.063) cRailLATIN 0.140 (0.037)cRailLowASIA 3.691 (0.740) cRailLowASIA 3.034 (0.743)cGov.Eff. 0.030 (0.034) cGov.Eff. 0.090 (0.031)cGEEUR 0.688 (0.144) cGEEUR 0.234 (0.050)cGEMENA 0.102 (0.132) cGEMENA 0.065 (0.144)cGESUBSAH 74.174 (6.222) cGESUBSAH 69.794 (6.163)cGELATIN 0.233 (0.113) cGELATIN 0.075 (0.108)cGELowASIA 0.029 (0.068) cGELowASIA 0.084 (0.069)cPol.Risk 0.005 (0.0009) cPol.Risk 0.006 (0.0009)cPRFMENA 0.002 (0.001) cPRFMENA 0.002 (0.001)cPRFSUBSAH 0.008 (0.001) cPRFSUBSAH 0.009 (0.002)c 0.002 (0.001)

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 135PRFLATINcPRFLowASIA 0.003 (0.001)

Variance parameters for compound errorln (r2) 4.851 (0.045)ilgtc 0.087 (0.270)r2u 0.003 (0.0006)r2v 0.004 (0.0004)c 0.478 (0.067)

OthersNo. of obs. 1230Log likelihood 1299.576

a 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.cPRFLATIN 0.001 (0.001)cPRFLowASIA 0.005 (0.001)

ln (r2) 4.810 (0.044)ilgtc 0.461 (0.270)r2u 0.005 (0.0006)r2v 0.003

(0.0005)c 0.613 (0.064)

No. of obs. 1230Log likelihood 1289.145

in 20

Inte

78.183.183.971.277.726 0.765 1.779 8572.031 0.283 1.743 8789.990.273.475.277.768.385.944.048.661.180.948.5

a.ell asents

t proreauc

136 M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141Table 9Technical efciency ranking of high-income countries.

High-income countries technical efciency TE determinants

Country Ranking by TEa Mobile phonesb

2009 2000 1990 1980

United States 0.998 0.997 0.995 0.991 97.197United Kingdom 0.997 0.996 0.995 0.993 129.976Finland 0.996 0.993 0.988 0.845 144.238France 0.995 0.995 0.994 0.991 95.091Canada 0.994 0.983 0.923 0.857 68.408Australia 0.990 0.956 0.886 0.712 110.720Netherlands 0.988 0.968 0.826 0.770 128.132Sweden 0.978 0.908 0.738 0.701 122.832Austria 0.962 0.843 0.729 0.692 140.756Belgium 0.931 0.819 0.743 0.701 115.110Japan 0.929 0.809 0.672 0.630 90.088Ireland 0.906 0.776 0.584 0.530 109.451Denmark 0.877 0.805 0.677 0.633 133.941Greece 0.855 0.721 0.622 0.636 117.829Portugal 0.822 0.753 0.660 0.632 142.756Spain 0.766 0.683 0.611 0.610 110.952Korea 0.765 0.649 0.563 0.544 98.353Italy 0.729 0.629 0.556 0.508 150.466

a The efciency coefcient per country is dened as TEi = exp(ui)b Mobile phone subscribers per 100 inhabitants in 2009, ITU data.c Internet subscribers per 100 inhabitants in 2009, ITU data.d Goods transported (million tons km/country area) in 2009, WDI date Perceptions of the quality of public services and of the civil service, as w

policy formulation and implementation, and the credibility of the governmf Components: government stability, socio-econ. conditions, investmen

tensions, law and order, ethnic tensions, democratic accountability and buAmerican countries, the reductions are 0.003 and 0.004for an additional percentage point in mobile phone users,respectively, and 0.006 and 0.009 for an additional per-centage point in Internet users, respectively.

These results conrm that the contribution to TE peradditional mobile phone and Internet subscriber is higherin Latin America and Asia compared to the high-incomecountries. However, the accrued increase of TE is higherin high-income countries due to overall higher penetrationrates of both mobile telecommunication networks andInternet.

Tables 9 and 10 summarize, respectively, the estimatedTE levels, as well as the penetration levels of mobile com-munication networks and Internet technologies in high-in-come and low-income countries in 2009. Table 11summarizes the correlation coefcients between the pene-tration rates of mobile communication networks and Inter-net technologies, as well as the technical efciencyrankings in 2009.

We observe that the correlation between efciencyrankings and mobile phone penetration levels in 2009 ispositive for low-income countries (0.45), and negative(0.28) for high-income countries. This implies that themost technically efcient high-income countries do nothave the highest mobile penetration rates. In contrary,the most (least) efcient countries in the low-incomecountries have the highest (lowest) mobile penetration.As for the Internet technologies, the most technically ef-cient countries are also the ones characterized by a higherInternet penetration. The correlation is specially strong in64 0.026 2 84.579 0.124 1.985 88.552 0.241 1.634 9000 0.060 1.475 82.523 0.017 1.256 77.567 0.001 1.298 87.515 0.037 2.191 8453 0.004 0.608 7315 0.008 1 8381 0.015 0.935 7707 0.019 1.112 77.547 0.022 0.516 79.5

the latters degree of independence from political pressures, the quality ofcommitment to such policies in 2009, WGI data.le, internal and external conict, corruption, military in politics, religiousracy quality in 2009, PRS data.09

rnetc Railways usaged Government effectivenesse Political riskf

39 0.791 1.387 82.588 0.020 1.475 77.538 0.166 2.131 92.567 0.041 1.442 78.5Latin America (0.98), Asia (0.64), and in the high-incomecountries (0.616).

Tables 12 and 13 summarize the estimation results ofthe impact of lagged variables of mobile phone and Inter-net diffusion on technical inefciency reduction, i.e. a TEincrease in low-income and high-income countries.

We observe that for low-income countries, the causalrelation is from the mobile communication networks andInternet technologies towards technical inefciency reduc-tion (i.e., TE increase). This claim is justied by the statis-tically signicant coefcient of the lagged variables (3lags) of mobile phone subscribers: 0.001 for MENA,0.003 for Sub-Saharan Africa and Asia, and 0.002 for La-tin America. Similarly, statistically signicant parametersof the lagged variables (3 lags) of Internet subscribers wereobtained for MENA (0.002), Sub-Saharan Africa (0.02),Latin America (0.008), and Asia (0.01). For high-incomecountries, a signicant causal relation between TE andtelecommunications infrastructure was not observed.

7.4. Technical efciency improvement by transportation andinstitutions

Higher usage of freight railways, increased governmenteffectiveness, and improved political risk have a varyingimpact on TE in low-income and high-income countriesduring the analyzed period.

According to the estimation results summarized inTable 8, there is a decrease in technical inefciency for

Table 10Technical efciency ranking of low-income countries.

Low-income countries technical efciency TE determinants in 2009

Country Ranking by TEa Mobile phonesb Intern

2009 2000 1990 1980

28.8457.6025.805.300.388.69

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 137AsiaChina 0.978 0.683 0.610 0.540 56.103Malaysia 0.754 0.632 0.500 0.446 110.598Thailand 0.726 0.646 0.604 0.617 122.567India 0.568 0.500 0.502 0.462 45.448Bangladesh 0.520 0.454 0.424 0.394 31.068Indonesia 0.512 0.418 0.442 0.505 69.248

Latin Americaevery additional percentage point in railways usage for allthe studied countries, especially in Asia.12

Government effectiveness has an impact on technicalinefciency reduction only for Europe and Sub-SaharanAfrica.13 No negative impact of government effectivenesswas observed in other low-income regions.

Chile 0.958 0.779 0.672 0.698 96.935 33.98Argentina 0.730 0.569 0.471 0.529 128.837 30.39Peru 0.670 0.509 0.428 0.537 84.692 27.72Mexico 0.561 0.490 0.491 0.545 77.750 26.47

Middle East and North AfricaTunisia 0.632 0.543 0.476 0.437 93.495 33.54Jordan 0.584 0.463 0.442 0.501 101.064 29.27Turkey 0.573 0.487 0.481 0.446 83.912 35.30Egypt 0.521 0.480 0.446 0.461 66.689 20.04Iran 0.465 0.460 0.439 0.487 72.088 38.28Algeria 0.462 0.401 0.455 0.483 93.793 13.46Syria 0.399 0.469 0.422 0.461 45.974 18.65

Sub-Saharan AfricaKenya 0.650 0.601 0.617 0.586 48.652 10.03Ghana 0.641 0.503 0.455 0.429 63.383 5.44Cameroon 0.594 0.467 0.462 0.456 37.892 3.84Senegal 0.561 0.642 0.548 0.455 55.061 7.36Zambia 0.503 0.390 0.436 0.399 34.066 6.31Malawi 0.425 0.386 0.340 0.313 15.723 4.69

a The efciency coefcient per country is dened as TEi = exp(ui)b Mobile phone subscribers per 100 inhabitants in 2009, ITU data.c 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

policy formulation and implementation, and the credibility of the governmentsf Components: government stability, socio-econ. conditions, investment pro

tensions, law and order, ethnic tensions, democratic accountability and bureauc

Table 11Correlation between technical efciency ranking and technical efciency determin

TE determinants High-income countries Asia

Mobile phones 0.236 0.316Internet 0.616 0.648Railways usage 0.457 0.775Government effectiveness 0.695 0.539Political risk factor 0.482 0.628

a Middle East and North Africa.

12 The high coefcient for Asia may be explained by the freight railwaysusage level in China, which is the highest among other low-incomecountries as well as the majority of high-income countries.13 In the case of Sub-Saharan African countries, Cameroon, Ghana, andMalawi are characterized by signicant improvements of the indicator fromthe second half of 1990.0 0.189 0.116 67.58 0.005 0.989 724 0.004 0.152 585 0.047 0.014 62.50 0.001 0.985 536 0.001 0.211 60.5etc Railways usaged Government effectivenesse Political riskfFinally, political risk factor reduces technical inef-ciency, especially in high-income countries, and to a lesserextent, in MENA and Asian countries. For Sub-Saharan Afri-ca, there is no improvement in TE due to political risk fac-tor changes, and for South American countries, thecoefcients are not statistically signicant.

7.5. Robustness check

As a robustness check, in estimation results not shownin this paper, we included two additional variables sepa-rately in Models 1 and 2 to check whether the parameters

3 0.023 1.209 79.59 0.029 0.420 65.51 0.003 0 622 0.066 0.17 72.5

9 0.019 0.413 720 0.005 0.280 710 0.012 0.351 59.53 0.004 0.300 58.59 0.028 0.742 55.58 0.003 0.590 60.56 0.011 0.608 57.5

8 0.003 0.658 571 0.001 0.064 671 0.005 0.806 65.54 0.003 0.400 583 0.007 0.674 63.53 0.0002 0.520 56.5

the latters degree of independence from political pressures, the quality ofcommitment to such policies in 2009, WGI data.le, internal and external conict, corruption, military in politics, religiousracy quality in 2009, PRS data.

ants in 2009.

Latin America MENAa Sub-Saharan Africa

0.350 0.716 0.8450.982 0.530 0.420

0.453 0.044 0.0790.709 0.927 0.2510.569 0.778 0.367

Table 12Causality between technical efciency and mobile phones subscribers.

Stochastic frontier truncated-normal model with lagged mobile phone subsc

Model 3 low-income countries

138 M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141Primary index equation parametersb0 0.151 (0.105)a

bCapital 0.408 (0.019)bLabor 0.348 (0.059)bEducation 0.877 (0.084)bCapital labor 0.035 (0.010)bCapital education 0.009 (0.022)bLabor education 0.095 (0.034)bCapital2 0.023

(0.006)

bLabor2 0.076 (0.015)bEducation2 0.126

(0.046)

Offset [mean = lit] parameters in one-sided error uitc0 0.780 (0.061)

cMobile phones 0.001 (0.0002)cMobLAGGED 0.001 (0.0003)cMobLAGGEDMENA 0.0005 (0.0009)cMobLAGGEDSUBSAH 0.002 (0.001)cMobLAGGEDLATIN 0.001 (0.0004)cMobLAGGEDASIA 0.002 (0.0005)cRailways 0.403 (0.039)

of interest in the technical efciency equation changedsignicantly.

The rst variable is the share of households with a tele-vision set (TV), obtained from the World DevelopmentIndicators (WDI). Television is believed to contribute tothe reduction of technical inefciency by conveying infor-mation and knowledge, albeit less efcient than Internettechnologies.

The second variable added to the models is the index ofeconomic freedom (IEF), taken from the HeritageFoundation.14 This institutional variable can reducetechnical inefciency because it permits to minimize distor-tions that affect not only market transactions but also theoutput and welfare of both consumers and producers.

After controlling for mobile telecommunication net-works, Internet, freight railways, and institutions, it wasobserved that television sets do not contribute to reducetechnical inefciency (exact results are omitted due to

cGov.Eff. 0.058 (0.020)cPol.Risk 0.003 (0.0005)Variance parameters for compound errorln (r2) 5.028 (0.197)ilgtc 1.703 (0.288)r2u 0.005 (0.001)r2v 0.001 (0.00009)c 0.845 (0.037)

OthersNo. of obs. 1106Log likelihood 1159.331

a 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.

14 The index of economic freedom (IEF) is an average of the following tencomponent scores: business freedom, trade freedom, scal freedom,government size, monetary freedom, investment freedom, nancial free-dom, property rights, freedom from corruption and labor freedom.b0 0.257 (0.103)bCapital 0.406 (0.019)bLabor 0.418 (0.057)bEducation 0.845 (0.083)bCapital labor 0.041 (0.014)bCapital education 0.003 (0.023)bLabor education 0.092 (0.034)bCapital2 0.021

(0.006)

bLabor2 0.076 (0.035)

bEducation2 0.164 (0.052)

c0 0.798 (0.061)

cMobile phones 0.001 (0.0003)cMobLAGGED 0.001 (0.0003)cMobLAGGEDEUR 0.000005 (0.00002)

cRailways 0.428 (0.041)ribers

Model 4 high-income countriesspace limitations). Similarly, in the case of economic free-dom, we obtained an indication of reduced technical inef-ciency but the coefcients are not statistically signicant.

Most importantly, the inclusion of both variables doesnot change the main parameter estimates in the technicalefciency equation, nor the estimates of the inputs in thestochastic production frontier. Therefore, our estimated re-sults are robust with respect to an inclusion of additionalvariables in the proposed models.

We have also checked the technical efciency rankingsobtained from Models 1 and 2 and the robustness checkmodels by including separately the two additional vari-ables (TV and IEF). None of the technical efciency rank-ings of both high-income and low-income countries haschanged in a signicant way. Therefore, the country tech-nical efciency rankings are also robust with respect tothe inclusion of additional variables to the main models.

8. Conclusions

By estimating a stochastic production frontier for a ex-ible transcendental logarithmic production function under

cGov.Eff. 0.071 (0.019)cPol.Risk 0.003 (0.0005)

ln (r2) 4.913 (0.212)ilgtc 1.867 (0.296)r2u 0.006 (0.001)r2v 0.0009

(0.00009)c 0.866 (0.034)

No. of obs. 1106Log likelihood 1147.537

ribers

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 139Table 13Causality between technical efciency and Internet subscribers.

Stochastic frontier truncated-normal model with lagged internet subsc

Model 5 low-income countries

Primary index equation parametersb0 0.699 (.069)a

bCapital 0.388 (0.020)bLabor 0.286 (0.061)bEducation 0.990 (0.084)bCapital labor 0.072 (0.015)bCapital education 0.009 (0.025)bLabor education 0.109 (0.035)bCapital2 0.010 (0.007)

bLabor2 0.099 (0.013)bEducation2 0.189

(0.053)

Offset [mean = lit] parameters in one-sided error uitc0 0.581 (0.058)

cInternet 0.003 (0.0006)cIntLAGGED 0.001 (0.0008)cIntLAGGEDMENA 0.001 (0.003)cIntLAGGEDSUBSAH 0.020 (0.012)cIntLAGGEDLATIN 0.007 (0.002)cIntLAGGEDLowASIA 0.010 (0.003)cRailways 0.091 (0.054)cGov.eff. 0.038 (0.018)the assumption of xed effects, we provide empirical evi-dence of infrastructure and institutions being determi-nants of technical efciency (TE) improvement in bothlow-income and high-income countries from 1980 to2009. In particular, we demonstrate various levels of TE in-crease due to the deployment of mobile communicationnetworks and Internet technologies, usage of freight rail-ways, improved government effectiveness, and lowerpolitical risk.

Importantly, the causal relation is shown to be from thedeployed mobile communications networks and Internettechnologies towards the TE in low-income countries inLatin America, Asia and Sub-Saharan Africa. For high-in-come countries, the particular causality could not be deter-mined since the results were not statistically signicant.Consequently, to benet from further TE improvements,policy makers, especially in low-income countries, mustimplement telecommunication policies that increase thepenetration rates of mobile communication networks andInternet technologies. Both current and next-generationsystems are of interest.

In general, the spread of telecommunication technolo-gies can be achieved by privatizatizing networks and

cPol.risk 0.002 (0.0005)Variance parameters for compound errorln (r2) 4.805 (0.050)ilgtc 2.823 (0.845)r2u 0.007 (0.0005)r2v 0.0004 (0.0003)c 0.943 (0.044)

OthersNo. of obs. 1106Log likelihood 1162.366

a 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.Model 6 high-income countries

b0 0.127 (0.104)bCapital 0.402 (0.018)bLabor 0.371 (0.058)bEducation 0.912 (0.080)bCapital labor 0.043 (0.015)bCapital education 0.036 (0.022)bLabor education 0.127 (0.034)bCapital2 0.021

(0.006)

bLabor2 0.059 (0.015)bEducation2 0.205

(0.047)

c0 0.689 (0.082)

cInternet 0.003 (0.001)cInternetLAGGED 0.0005 (0.001)cIntLAGGEDEUR 0.0004 (0.002)

cRailways 0.269 (0.104)cGov.eff. 0.061 (0.022)enabling market competition, promoting market access tonewentrants, controlling the end-user price, granting subsi-dies to increase computer ownership, and providing publicaccess to telecommunication services such as the Internet.

To be more specic, previous studies concerning theeffectiveness of various telecommunications policies haveshown that the privatization of incumbent operators ofmobile communication networks, as well as policies ensur-ing competition increase the TE of the telecommunicationssector and trigger technology diffusion (Symeou, 2011).However, for better market outcomes, the competitionand privatization must be implemented simultaneously(Fink et al., 2002; Li and Xu, 2002). In addition to privatiza-tion and competition policies, the regulator authority mustbe granted independence (Fink et al., 2002; Howard andMazaheri, 2009).

Competition in the mobile telecommunication sectorcan be also promoted by guaranteeing market access tonew entrants by awarding new spectrum licenses, or regu-lating the share of telecommunications infrastructureamong providers. However, in this context, it has beenshown that the possibility of awarding spectrum licenseson the basis of either an auction or a so-called beauty

cPol.risk 0.003 (0.0006)

ln (r2) 4.694 (0.048)ilgtc 1.614 (0.663)r2u 0.007 (0.001)r2v 0.001 (0.0008)c 0.833 (0.091)

No. of obs. 1188Log likelihood 1156.035

International Economics 82, 208218.

140 M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141contest creates a trade-off between government revenuesand the diffusion of third generation of mobile phone tech-nologies (Baquero and Kuroda, 2011).

As for the policies related to Internet technologies, pre-vious research has shown better Internet diffusion indeveloping countries in response to regulation aiming to(i) promote market access of additional Internet ServiceProviders, and (ii) control the end-user prices (Wallsten,2002). Furthermore, subsidies targeting higher rates ofcomputer ownership by the poorest households as wellas the provisioning of public Internet represents additionalmeasures that enhance the usage of Internet in low-in-come countries (ITU Report, 2008).

The ongoing standardization and initiated commercial-ization of the next-generation networking technologiessuch as Worldwide Interoperability for Microwave Access(WiMAX), Wireless Fidelity (Wi-Fi), and Long Term Evolu-tion Advanced (LTE-A) may possibly become the main dri-ver of Internet penetration in both low-income and high-income countries.

The reason for the above statement is that in developingcountries, the low deployment cost of WiMAX or Wi-Fi en-able economically viable solutions for last-mile broadbandInternet access on both the local level (e.g., in small villagesand remote locations in Latin America Galperin, 2005, In-dia Mishra et al., 2005) and several developing countries(Gunasekaran and Harmantzis, 2007). In high-incomecountries, the claim justication follows from the techno-logical superiority of the LTE-A network over current solu-tions as well as the rapid spread of increasingly capableportable tablets (expected to reach 500 million by 2015).In particular, since LTE will allow transmission speeds highenough to rival xed broadband connections, auctions ofspectrum for fourth generation mobile telephony usingLTE are already being run in many countries (Hoernigand Valletti, 2012).

In this context, further research should take into ac-count the impact of the deployment of innovative mobilecommunication services and increased broadband Internetcapacity. This study focused only on the diffusion of basicmobile communications and Internet service.

Further differentiation can be done on the level of ser-vice content - the next-generation mobile phone servicesin developed countries are primarily used for entertain-ment purposes (music downloads and mobile games)whereas in developing countries, data services such asagricultural advices, health care and money transfer areof higher importance.

Another issue consists in analyzing whether the pro-ductivity gains stemming from a universal Internet accessin developing countries can be obtained also by deployingthe next-generation technologies such as WiMAX, Wi-Fi,and LTE. The usage of capable/inexpensive computing de-vices such as the tablets and netbooks should be consid-ered as a related factor as well.

Acknowledgements

The comments and suggestions by the Journal EditorTobias Kretschmer and two anonymous Referees havegreatly contributed to this study. The author would likeKlenow, P.J., Rodriguez-Clare, A., 1997. The neoclassical revival in growtheconomics: has it gone too far? NBER Macroeconomics Annual 12,73103.

Kumbhakar, S., Lovell, C., 2000. Stochastic Frontier Analysis. CambridgeUniversity Press, Cambridge, England.

Li, W., Xu, L.C., 2002. The Impact of Privatization and Competition in theTelecommunications Sector Around the World. Darden BusinessSchool Working Paper No. 02-13, pp. 143.

Lin, W.T., 2009. The business value of information technology asmeasured by technical efciency: evidence from country-level data.Decision Support Systems 46, 865874.to express a deep gratitude to Professors Takanori Idaand Kazuhiro Yuki from Kyoto University, as well as Assis-tant Professor Toshifumi Kuroda from Tokyo Keizai Univer-sity for their helpful advices and careful review support.This study was supported in part by the Monbukagakushoscholarship from the Ministry of Education, Culture, Sports,Science and Technology (MEXT) of the Government ofJapan.

References

Aker, J.C., 2008. Does digital divide or provide? the impact of cell phoneson grain markets in Niger. BREAD Working Paper, pp. 161.

Baquero, M., Kuroda, T., 2011. Treatment effects analysis of the thirdgeneration mobile phone spectrum auction. Journal of Informationand Communication Research (in Japanese) 100 (29-3), 4959.

Beck, T., Laeven, L., 2006. Institution building and growth in transitioneconomies. Journal of Economic Growth 11, 157186.

Bilsona, C.M., Brailsforda, T.J., Hooper, V.C., 2002. The explanatory powerof political risk in emerging markets. International Review ofFinancial Analysis 11, 127.

Busse, M., Hefeker, C., 2007. Political risk, institutions and foreign directinvestment. 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. Broadbandinfrastructure and economic growth. Economic Journal, RoyalEconomic Society 121552, 505532.

Daude, C., Stein, E., 2007. The quality of institutions and foreign directinvestment. Economics & Politics 19 (3), 317344.

Easterly, W., Levine, R., 2001. Its not factor accumulation: Stylized factsand growth models. NBER Working Paper Series, pp. 159.

Esfahani, H., Ramirez, M., 2003. Institutions, infrastructure and economicgrowth. Journal of Development Economics 70, 443477.

Fink, C., Mattoo, A., Rathindran, R., 2002. An Assessment ofTelecommunications Reform in Developing Countries. World BankPolicy Research Working Paper. 2909, pp. 137.

Galperin, H., 2005. Wireless networks and rural development:opportunities for Latin America. Information Technologies andInternational Development 2 (3), 4756.

Gunasekaran, V., Harmantzis, F.C., 2007. Emerging wireless technologiesfor 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. OxfordUniversity Press, New York, pp. 136160.

Howard, P.N., Mazaheri, N., 2009. Telecommunications reform, Internetuse and mobile phone adoption in the developing world. WorldDevelopment 37 (7), 11591169.

ITU, 2008. Measuring information and communication technologyavailability in villages and rural areas. InternationalTelecommunication Union (ITU), Geneva, Switzerland.

Jensen, R., 2007. The digital provide: Information (technology), marketperformance and welfare in the south Indian sheries sector. TheQuarterly Journal of Economics 122, 583597.

Jensen, R., Oster, E., 2007. The power of tv: Cable television and womensstatus in India. NBER Working Paper Series, pp. 10571094.

Jondrow, J., Lovell, C.K., Materov, I.S., Schmidt, P., 1982. On the estimationof technical inefciency in the stochastic frontier production functionmodel. Journal of Econometrics 19 (23), 233238.

Kaufmann, D., Kraay, A., Mastruzzi, M., 2008. Governance matters vii:Aggregate and individual governance indicators 19962007. PolicyResearch 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

Lin, W.T., Chiang, C.Y., 2011. The impacts of country characteristics uponthe value of information technology as measured by productiveefciency. 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 ruraldeveloping regions. Internet and Network Economics 3828, 184194.

Perotti, E.C., Oijen, P.V., 2001. Privatization, political risk and stock marketdevelopment in emerging economies. Journal of International Moneyand Finance 20, 4369.

Political Risk Services, 2009. International Country Risk Guide. Publishedonline.

Qiang, Z., Rossotto, C., Kimura, K., 2009. Economic Impacts of Broadband.In: World Bank, Information and Communications for Development2009: Extending Reach and Increasing Impact. Washington, USA.

Roller, L., Waverman, L., 2001. Telecommunications infrastructure andeconomic development: a simultaneous approach. The AmericanEconomic Review 91, 909923.

Sa, Y., 2011. Government size, economic growth and unemployment:evidence from advanced and developing economy countries (a time

series analysis, 19962006). International Review of PublicAdministration 16 (295), 95116.

Shao, B.B.M., Lin, W.T., 2001. Measuring the value of informationtechnology in technical efciency with stochastic productionfrontiers. Information and Software Technology 43, 447456.

Stata Press, 2005. Stata Base Reference Manual, Release 9. Texas, UnitedStates of America.

Symeou, P.C., 2011. Economy size and performance: an efciency analysisin the telecommunications sector. Telecommunications Policy 35 (5),426440.

Thompson, H., Garbacz, C., 2007. Mobile, xed line and Internet serviceeffects on global productive efciency. Information Economics andPolicy 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 oneconomic growth in developing countries. Vodafone Policy PaperSeries 1, 123.

World Bank, 2007. World Development Indicators. C.D. World BankDevelopment Data Group.

M.d.P. Baquero Forero / Information Economics and Policy 25 (2013) 126141 141

Mobile communication networks and Internet technologies as drivers of technical efficiency improvement1 Introduction2 Previous work on telecommunications and economic growth2.1 Telecommunications as determinants of national income level2.2 Telecommunications as determinants of technical efficiency2.3 General infrastructures and institutions as determinants of technical efficiency

3 Contributions4 Modeling approach using stochastic production frontier5 Panel data and empirical model specification5.1 Panel data5.1.1 Production function: output and inputs5.1.2 Technical efficiency equation: infrastructure and institutions

5.2 Empirical model specification using a translog function

6 Summary statistics6.1 Mobile communication networks6.2 Internet technologies6.3 Freight railways usage, government effectiveness, and political risk

7 Estimation results7.1 Production function estimates and input elasticities of output7.2 Technical efficiency estimates7.3 Technical efficiency improvement by telecommunications and their relation of causality7.4 Technical efficiency improvement by transportation and institutions7.5 Robustness check

8 ConclusionsAcknowledgementsReferences