Master Thesis MSc. Economics
Specialization in International Economics and Globalization Faculty of Economics and Business
ICT and development: what is the impact of
telecommunications on economic growth in Sub-Saharan
Africa?
Name: Guilherme von Zuben Student number: 10825851
E-mail: guilherme.vonzuben@gmail.com Supervisor: Naomi Leefmans
Second reader: Dr. Dirk Veestraeten Date of submission: 28.04.2016
Statement of Originality
This document is written by Student Guilherme von Zuben who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
This thesis investigates the impact of ICT infrastructure on economic growth in Sub-Saharan Africa. A dynamic panel framework is applied for a pool of 44 countries between 1980 and 2014. In order to control for endogeneity, a system-GMM is employed. The findings in this thesis outline that telecommunications have a positive and statistically significant effect on economic growth, cell phones being the most important mean of communication contributing to development in the region. In addition to this, such impact is even higher for the poorest countries in the sample. These nations also display diminishing returns, meaning that countries with less ICT infrastructure tend to profit more than countries where it is already established. Given the results presented, investments in the telecom network expansion, especially the mobile telephony network, should be encouraged.
Table of Contents
1. Introduction ... 5
2. Previous empirical research on ICT and economic growth... 9
2.1 Telecommunications in a structural model ...9
2.2 Telecommunications in an endogenous growth framework...11
2.2.1 Cross-sectional Analysis ...11
2.2.2 Panel Analysis ...14
2.3 Sub-conclusion ...16
3. Methodology and Data ... 17
3.1 Model ...17
3.2 Estimation Method ...20
3.3 Data...22
4. Estimation Results... 25
4.1 Reference Equation Results ...25
4.2 Robustness...29
5. Conclusion... 34
6. References ... 36
1. Introduction
Infrastructure as a whole is seen as a powerful tool to enhance growth and equity, which may subsequently contribute to reduce income inequality and poverty. Sub-Saharan Africa (SSA) however, features a very low index of infrastructure even among developing regions. The corresponding drawbacks associated with these nations range from being landlocked to being isolated from global market centers. The implementation of infrastructure strategies, like well-developed transportation and communication frameworks can work to remedy the geographic issues faced by this region (Calderón and Servén, 2008). Indeed, the executive chairman of the World Economic Forum, Klaus Schwab, stated at the first World Summit on Information Society that investment in Information and Communications Technology (ICT) shows the best prospects to developing countries when it comes to accelerating progress (Sridhar and Sridhar, 2004).
ICTs, as defined by Grace, Kenny and Zhen-Wei Qiang (2003), “are tools that facilitate the production, transmission and processing of information. Thus a broad definition of ICTs ranges from traditional technologies such as the printed world, to the most modern communications and data delivery systems […]”. ICT infrastructure or telecommunications infrastructure, in the context of this thesis seen as landlines and mobile phones, can then contribute to economic growth in different ways. The first channel refers to the demand of products and services related to goods which are needed to satisfy the own infrastructure implementation, like cables, devices and switches. On another level, the improvement of telecommunication networks helps to increase the coordination of economic activity. The costs associated with ordering, gathering, and searching for products and services between companies are reduced, making each company’s individual level of output increase (Roeller and Waverman, 2001). Besides this, improving telecommunications infrastructure creates positive externalities. Innovations in the sector increase transparency, civil and political activity, and inclusion in the educational system (Batuo, 2015). Such developments can arise through services like e-government, e-commerce and e-learning (ibid.).
However, investments in landline networks can be very costly especially in the case of the African continent, where geographical barriers prevail. A report from the International Telecommunication Union (ITU) shows that in 2006, Africa had less than 2% of the worldwide main telephone lines, while Asia accounted for 48% of the total amount Union (ITU, 2007). Mobile phone networks however present a different story due to the system implementation being easier and cheaper. Sub-Saharan Africa features a disproportional growth rate of mobile telephony in comparison to fixed lines. As seen in Figure 1, the growth of mobile phones in SSA has been steady until the end of the 1990s and since 2000, subscriptions increased on average 37% a year, reaching the 682 million mark in 20141. However, very few studies address the
relationship between telecommunications and economic growth stressing the mobile market (Lee, Lavendis and Gutierrez, 2012). On the other hand, Figure 2 shows that the absolute number of fixed telephone subscriptions in SSA is significantly smaller and as from the year 2010 it starts decreasing, reaching 10.6 million subscriptions in 2014.
Much of these exponential growth rates are due to changes in regulation policies and liberalization processes that took place in Africa in the past decades. Once controlled by the government, the telecommunications sector saw a wave of liberalization and privatization during the 1980s, as debt crises in these countries deteriorated and the fragility of the system was revealed. These actions were taken in the context of programs initiated by the World Bank and the International Monetary Fund (IMF). In general, competition in both the mobile and fixed telephony sector was broadly introduced in the region from the year 2000. At the end of the 1990s, for instance, 70% of African countries held a monopoly scheme for the mobile segment, while in 2004 this rate amounted only 10% (Djiofack-Zebaze and Keck, 2009). The combination of liberalization policies together with technological process and decreasing costs made both private and public sectors inject a considerable amount of capital in the telecommunications market (Batuo, 2015). In addition to this, governments are creating regulating institutions to pull foreign investors into this market. The ITU points out that by the end of 2008 about 83% of African economies
had created such institutions (ibid.). This shows that African governments are aware that liberalizing and deregulating, but still creating independent monitoring institutions, lead to more investments (Smith-Hillman and Brathwaite, 2004).
The goal of this thesis is to analyze the impact of telecom (embracing mobile phones and landlines) on economic growth in SSA. Therefore, my research question is: What is the impact of the improved telecommunications infrastructure on economic growth in Sub-Saharan Africa between the years 1980 and 2014? Moreover, does this relationship display diminishing or increasing returns?
The value-added in this thesis is threefold. To begin with, this thesis aims to fill the gap in the current literature concerning the impact of telecommunication infrastructure on growth, particularly for Sub-Saharan Africa. The only paper encountered that dealt exclusively with Sub-Saharan Africa was Lee et al. (2012). Otherwise, the effect of ICT on growth has, for instance, covered OECD countries (Datta and Agarwal, 2004), China (Shiu and Lam, 2008), ASEAN countries (Ahmed, Riduzan, 2012), a pool of developing and developed countries (Hardy, 1980; Waverman, Meschi and Fuss, 2005) or the African continent as a whole (Djiofack-Zebaze and Keck, 2009; Chavula, 2013; Batuo, 2015). The telecommunication infrastructure is captured in this thesis at the aggregate level, as a sum of the number of landlines and mobile telephony. In addition to this, while for instance, Lee et al. (2012) deal with a time frame from 1975 to 2006 or Batuo (2015) from 1990 until 2013, the time period considered for this analysis starts in the year 1980 and is broadened until 2014. This time frame allows for more variation in all variables and in the case of telecommunications it enables to capture exponential growth rates in mobile penetration from the 2000s on. Furthermore, a Generalized Method of Moments (GMM) is used to account for endogeneity problems in the model.
The empirical growth framework is based on a dynamic panel data analysis in line with Islam (1995) and Lee et al. (2012), while the determinants of the growth equation are mainly based on the latter, Barro (1991), Sala-i-Martin (1997) Datta and Agarwal (2004) and Batuo (2015). Thus, the annual growth rate of GDP per capita is regressed on the lagged value of the dependent variable, the level of GDP per capita, the telecommunication variable as a sum of mobile phones and landlines and a set of control variables accounting for investments, trade openness and government
consumption. To solve for endogeneity problems that might arise, the GMM approach will be applied. The data for the thesis stems from the World Bank.
In section 2 an overview of selected papers about the impact of telecommunications on growth in Sub-Saharan Africa will be reviewed, categorized by framework and methodology. Thereafter, the underlying model of this thesis together with the estimation method and the data used will be described in section 3. Section 4 presents the estimation results followed by robustness tests to confirm the accuracy of the results. After these are interpreted, section 5 presents the conclusion and recommendations to promote economic progress.
2. Previous empirical research on ICT and economic growth
The impact of telecommunications on economic development has been an area of interest in the literature considering different regions of the world through distinct periods of time. Up until this point, such papers have been conducted using different empirical approaches. Chapter 2.1 outlines studies using a structural model, whereas chapter 2.2 deals with papers that use an endogenous growth framework to scrutinize the telecom and economic growth relationship. Here, it is of interest to differentiate between cross-country and panel analyses, as displayed in Chapter 2.2.1 and 2.2.2. Subsequently, in section 2.3 a conclusion will be drawn.2.1 Telecommunications in a structural model
In order to investigate the contribution telecommunications can have on economic growth, Roeller and Waverman (2001) make use of a structural model including a system of four equations. They consist of an aggregate production function, a demand function for telecommunications infrastructure, a supply function of telecommunications investment, and a telecommunications infrastructure production function. While the first equation accounts for the one-way relationship going from telecommunications to GDP, the other three equations endogenize the supply and demand for infrastructure and its investments. Such equations’ system is meant to control for a possible simultaneity bias between economic growth and telecom
investments. After applying a nonlinear GMM estimator, the authors conclude that for the sample average of 21 OECD countries between the year 1970 and 1990, a 1% increase in the penetration rate – number of mainlines per capita – would increase the average country’s output by 0,05%. In addition to this, Roeller and Waverman (2001) also differentiate the countries pooled using dummy variables in the output equation that indicate if the country has a low, middle or high rate of mainlines per capita2.
The main goal here is to investigate if a non-linearity in the impact of telecommunications on output is present. They conclude that although the coefficient for countries with a low penetration rate is positive and significant, it turns to be insignificant for medium penetration rates. However, when it comes to the countries with a high level, the effect is twice as large as for the other two sub-divisions. These results support the idea of a critical mass in telecommunications.
Sridhar and Sridhar (2004) employ the same framework proposed by Roeller and Waverman (2001), looking at both landlines and mobile phones as the telecommunications variable. Moreover, they consider a sample of 63 developing countries between 1990 and 2001. The estimations results show that a 1% increase of penetration rate – here the total number of mobile and main lines per 100 inhabitants – makes the economy grow by 0,14%. However, when only mobile phones are regressed, the results are even larger. A 1% increase in the number of cell phones per 100 inhabitants would increase output by 6,75%. Nonetheless, the authors attribute this result, for instance, to the lack of data for various countries or to the skyrocketing number of mobile telephony subscriptions that the countries in question had during the short period of time analyzed. Except from the caveats regarding missing data and assumptions made by the authors, they recognize the compelling impact that the mobile telephony has on developing countries to accelerate economic growth.
Waverman et al. (2005) extend the methodology proposed by Roeller and Waverman (2001) focusing on the impact that cellular phones can have on GDP considering 92 countries from which 38 were developing between the years 1996 and 2003. However, when transposing the value of the coefficient for the penetration rate
2 When the penetration rate is less or equal to 20%, the country is considered to have a low penetration rate. When the number of landlines per 100 inhabitants ranges between 20 and 40, it has a medium rate of penetration. Equal to or above the 41% level, the country features a high penetration rate (Roelller and Waverman, 2001).
to the original level of penetration for low and lower-middle income countries, Waverman et al. (2005) observe that mobile penetration would explain the whole output growth for the underlying period. The authors tried to control for other factors that may influence economic growth by changing the specifications, but the results turned out to be insignificant.
2.2 Telecommunications in an endogenous growth framework
In this section empirical studies based on endogenous growth models investigating the impact of telecommunications on economic growth will be considered. While authors in Chapter 2.2.1 opt for the use of a cross-country analysis, the literature presented in Chapter 2.2.2 applies a dynamic panel data model.
2.2.1 Cross-sectional Analysis
All authors reviewed in this section (Waverman et al., 2005; Djiofack-Zebaze and Keck, 2009 and Chavula, 2013) base their analysis on the endogenous growth approach proposed by Barro (1991). The latter empirically tests the ‘convergence’ hypothesis as the neoclassical growth theory3 presumes (Solow, 1956; Cass, 1965;
Koopmans, 1965) and scrutinizes the role of human capital for economic growth, following several endogenous growth models4. Barro (1991) utilizes a cross-section
analysis for a pool of 98 OECD countries, where the average GDP per capita over the period from 1960 to 1985 is regressed on the 1960-level value of GDP per capita – which tests the convergence hypothesis –, the 1960-level stock of human capital and the 1960-1985 average of usual growth control variables, like government consumption, investment ratio to GDP and political stability. The findings of this paper are that, holding the other variables constant, human capital (measured as primary and secondary school enrollment) has a significantly positive effect on the
3 That means for countries with similar technological degree and preferences, the per capita growth is opposite to the initial income level, in other words, poor countries would tend to grow faster than rich countries (Barro, 1991).
4 Romer (1990), for instance, argues that the largest the stock of human capital, the fastest an economy grows. Therefore, economic underdevelopment was also partially caused by a low stock of human capital.
growth rate per capita, whereas the initial level of GDP per capita is significantly negative. Investments have a positive impact on growth, while the measurements for political instability5 and government consumption affect the growth negatively.
Based on Barro’s (1991) underlying model, Waverman et al. (2005) employ a second approach besides the structural model described in section 2.1 to investigate the implications of telecommunications infrastructure on economic growth. The sample was extended over 16 years more in comparison to the structural model, spanning from the year 1980 to 20036. The average GDP per capita for this period is
regressed on the 1980-level number of fixed-lines per 100 inhabitants and the average of mobile phone subscribers per 100 inhabitants for this period. As control variables, the authors use the average ratio of investment to GDP for the period and as a measure for human capital, the proportion of the population above 15 years old that held at least a primary school diploma in the year 1980. After testing for the endogeneity of the mobile penetration variable using a Hausman test, which afterwards confirmed the consistency of the OLS estimation, the sample was divided in income levels. As a result, Waverman et al. (2005) find out that mobile telephony significantly adds to economic growth for all levels of income. This applies particularly for developing countries, where the impact can be two times higher than for industrialized countries. The estimation shows that for low-income countries, a 10% increase in the number of mobile phones generates a 0,59% growth in GDP per capita. The authors point out, however, that some possible issues were not covered by this study. For instance, endogeneity with respect to other right hand side variables was not tested. Moreover, Waverman et al. (2005) stress that there might be some other regressors not taken into consideration, like the institutional quality, that can both influence output growth and the level of mobile penetration, causing a spurious relationship between them.
Djiofack-Zebaze and Keck (2009) investigate the effect that liberalization measures in the telecommunications sector have on economic growth in SSA. The
5 The political stability is measured in two variables: the number of revolutions and coups per year and the number per million-population of political assassinations per year (Barro, 1991).
6 For the 1980s, mobile phones and human capital are considered exogenous since there were no mobile phones back then. This assumption is pulled off from 1990s on. Therefore, endogeneity is controlled from this period on (Waverman et al., 2005).
dataset comprises a pool of 45 Sub-Saharan African countries for the period between 1997 and 2003. They control for endogeneity and country-fixed effects using a GMM estimator by Arellano and Bond (1991) and use a 3SLS estimation to compare the results. The underlying regression comprises the common variables of Barro’s endogenous growth model plus the inclusion of further control variables for human development (life expectancy), macroeconomic policies (inflation), openness (exports as a share of GDP) and governance (political stability)7. The variable representing
liberalization is the number of mobile network providers in the market. With respect to the results obtained, the GMM estimation shows that the variable accounting for liberalization is statistically insignificant. However, when liberalization is seen as “actual performance” and proxied as the number of mobile subscribers8 one would
reach a positive and statistically significant coefficient9. In the GMM estimation, a
1%-growth in the number of subscribers means a 0,6% increase in real GDP per capita. Similarly, in the 3SLS methodology the impact in the economy amounts 0,7% considering the new measure for liberalization (Djiofack-Zebraze and Keck, 2009).
Chavula (2013) also makes use of Barro’s (1991) cross-sectional endogenous growth model to evaluate the impact of telecommunications on economic growth. 49 African countries are analyzed in this paper. The average GDP per capita between the year 1990 until 2007 is regressed on the 1990-level GDP, the 1990-level human capital as primary school enrollment and the average gross domestic investment to GDP ratio for the given period. As parameters for the telecom variable the averages of both landlines and cellular phones per 100 people and Internet user per 1000 people are considered10. Chavula (2013) also divides the countries in sub-samples of
upper-middle-, low-middle- and low-income nations according to the World Bank. As a result, he reasons that for the sample without the sub-division in income levels both fixed-lines and cell phones add to economic growth positively and significantly,
7 Djiofack-Zebaze and Keck (2009) follow Greenaway, Morgan and Wright (2002) for the selection of these variables. Fixed capital formation and the ratio of secondary education are chose to be the proxy for investment and human capital, respectively.
8 Percentage of population subscribed and having mobile telephone service activated within the last 9 months (Djiofack-Zebaze and Keck, 2009).
9 Nevertheless, the authors’ choice of taking the number of subscribers as an instrument for liberalization should be put into question. The variable might be capturing positive expenditures of mobile phone usage rather than liberalization.
whereas the Internet does not. Numerically, this means a 0,15% and 0,22% increase in per capita GDP growth, if respectively landlines and mobile penetration increases 1% per 100 people. However, when they are divided into the three income sub-samples, mobile telephony is the only ICT that has a significant impact on economic growth in low-middle and low-income nations. In this case, a 1% increase in mobile penetration per 100 people would add 0,26% to the economic growth in low-middle countries and 0,15% in low-income nations. The other two indicators remain insignificant. On the other hand, upper-middle countries’ growth is significantly affected by the three means of communication presented. A 1% increase in the number of Internet users per 1000 people would then increase economic growth by 0,21%. Likewise, a 1% increase in the number of mobile and fixed lines per 100 people would boost economic growth by 0,39% and 0,24% respectively.
2.2.2 Panel Analysis
Datta and Agarwal (2004), Lee et al. (2012) and Batuo (2015) adhere to a methodological improvement proposed by Islam (1995) to scrutinize the impact of telecommunications on economic growth. In his paper, the latter author sees a drawback present on cross-sectional analyses. Such empirical studies do not allow for variations in the production functions, which means, they assume identical preferences and technology states for all countries. Since in this framework single cross-country regressions are estimated, such variations are econometrically hard to measure. In order to overcome this problem, Islam (1995) proposed a dynamic fixed effects panel data model that takes into account individual ‘country-effects’, correcting for omitted variable bias. In this model, a lagged value of the dependent variable is added to the right-hand side of the equation in order to take capture the short-run autoregressive behavior. Thus, the correlation between the dependent variable and its previous values is taken into consideration. The author concludes that using panel data, one obtains a higher rate of convergence than using cross-sectional data.11
11 In order to obtain a feasible comparison with previous papers, Islam (1995) makes use of the same database (Summers-Heston data set), time period and country sample as in Mankiw, Romer and Weil (1992), which is similar to Barro (1991).
Datta and Agarwal (2004) follow Islam (1995) and utilize a dynamic panel data to investigate the role of telecommunications infrastructure on economic growth in 22 OECD countries from the year 1980 to 2002. The growth framework from Barro (1991) is adopted to determine the variables accounting for economic growth and extended to examine the impact of telecommunications on economic development. Real growth of GDP per capita is regressed on the lagged value of the dependent variable and the level real GDP per capita to test for convergence. The extension for telecommunications is quantified as a variable representing landlines per 100 inhabitants. Its lagged values (one and two periods) check for the causality between economic growth and telecommunications, while a squared of the telecom variable verifies the nature of returns to scale. As control variables, the author uses the rate growth of population, the share of fixed investment in GDP, the share of government consumption in GDP and the degree of openness of the respective country, measured as the total of imports plus exports. The results of the regression show that population growth and government consumption have a significantly negative effect on economic growth, whereas investments and openness have a significant positive impact on the dependent variable. Moreover, a positive and statistically significant relationship is found between telecommunications and real growth of GDP per capita, after controlling for the other explanatory variables. Numerically, one new landline per 100 inhabitants generates a 0,005% growth in the economy (Datta and Agarwal, 2004).
Narrowing the geographical coverage to the Sub-Saharan region, Lee et al. (2012) focus on the analysis of the impact of mobile phones on economic development. The authors use the same framework as Datta and Agarwal (2004), and add an interaction term between landlines and cellular phones, in order to analyze the marginal impact that mobile telephony might have, given the existing fixed-line network. They examine a sample of 44 Sub-Saharan countries between the years 1975 and 2006 and consider all explanatory variables not to be strictly exogenous. Thus, they use a two-step difference GMM to control for fixed-effects and endogeneity. When considering the whole time period, the results for the control variables go in accordance with Datta and Agarwal (2004) apart from the growth population variable, which turns out to be insignificant. Landlines significantly affect growth in a positive way, meaning that for a new telephone per hundred people, the economy
grows by 0,05%. Cell phones on the other hand turn out to be insignificant. After splitting the sample in two to account for the mobile telephone expansion around the turn of the new millennium, they came to the conclusion that after 2000 a 1%-increase in the number of mobile phones per 100 inhabitants leads to a 0,019% growth in the economy, whereas the effect of landlines wasn’t significant at all. Furthermore, Lee et al. (2012) come to the conclusion that the impact of a single cell phone as well as the marginal impact of mobile phones is greater where landlines are scarce.
Batuo (2015) collects data for 44 African countries12 from 1990 until 2010 and
follows a similar methodology as in Datta and Agarwal (2004), with the difference that the telecom variable is expressed as the sum of fixed and mobile lines per 100 inhabitants. Furthermore, he employs a one-step system GMM estimator in order to overcome possible endogeneity bias between economic growth and the telecom variable, according to Blundell and Bond (1998). For the control variables, he concludes that investments, openness and FDI have a significantly positive impact on growth, while population growth does not. The sum of mobile and fixed lines is positive and significantly relevant to economic growth, making GDP per capita rise between 0,5% (GMM model) to 0,8% (OLS model) if telecom density increases by 10%. Furthermore, investments in telecommunications feature increasing returns to scale. It means that countries with an ongoing high rate of telecom infrastructure tend to grow more (Batuo, 2015).
2.3 Sub-conclusion
This section has shown that the previous literature finds that telecommunications have a positive impact on economic growth. However, it can be of interest not only to narrow down the research to the Sub-Saharan African region, but also investigate if its more underprivileged nations can profit from the diffusion of ICT in a higher degree.
12 This pool of countries also includes non-Sub-Saharan African countries like Morocco, Tunisia and Egypt.
The Generalized Method of Moments appears to be a powerful method to account for endogeneity and fixed-effects. However, none of the papers using this method apply a two-step system GMM, which is preferred due to its efficiency13.
Therefore, it will be chosen over the two-step difference and one-step system GMM. Furthermore, the time range should be further expanded to account for recent developments in telecommunications diffusion.
3. Methodology and Data
In this section the empirical model proposed will be discussed. Chapter 3.1 outlines the model and the methodology used in this paper. From this, the data applied for this study will be described in chapter 3.2.
3.1 Model
The model presented in this paper is based on the studies portrayed in section 2.2.2. Thus, a dynamic panel framework is applied to investigate the impact of telecommunications on economic growth. As argued briefly, this framework controls for omitted variable bias by taking into account individual country-effects and by including a lagged dependent variable to the regression (Islam, 1995). This is done to avoid omitted dynamics, which lead to misspecification of the model, as argued by Bond (2002).
The variables used14 were chosen taking into consideration the papers by Barro
(1991), Sala-i-Martin (1997), Islam (1995), Datta and Agarwal (2004), Lee et al. (2012) and Batuo (2015). Consequently, the regression model used has the following form:
€
gdpgri,t =αgdpgri,t −1+βgdpi,t −1+δteli,t +γXi,t+µi+υi,t (1)
€
Xi,t = inv
[
i,t,govi,t,openi,t]
13 Section 3.2 will discuss this matter in greater detail. 14 The variables are also summarized in Appendix 1.
where t and i index time and country respectively, while µi represents the unobserved
country fixed effects andυi,t the error term. The dependent variable gdpgri,t represents
the growth rate in GDP per capita measured annually in percent. Its lagged value
gdpgri,t-1 is included as a right-hand side variable to account for the dynamic process.
The lagged value of the GDP per capita, gdpi,t-1, tests for the convergence hypothesis as
the neoclassical growth theory suggests. Therefore, we expect for this variable a negative sign or diminishing returns. This means that a higher initial level of GDP per capita results in a lower rate of growth. The control variables are represented by the matrix Xi,t, which includes: invi,t, govi,t and openi,t. The investments variable invi,t is
represented by the gross capital formation of a country as a percentage of total GDP and is expected to be positive. The share of general government consumption expenditure as a share of the GDP is denoted as govi,t. According to previous literature,
government consumption should have a negative impact on economic growth. The variable openi,t represents the degree of economic openness of a country, here proxied
as the sum of exports and imports as a percent of its GDP. Trade is commonly assumed to affect economic growth positively. The teli,t variable describes the
telecommunications infrastructure. The sum of both landlines and mobile telephones per 100 inhabitants is chosen to represent the telecom penetration in Sub-Saharan Africa, because the focus is on the aggregate impact of telecom on economic growth rather than on the quantification of their single impact.15 A positive correlation is
expected between telecommunications and economic development, as the previous literature suggests.
Variables accounting for human capital, institutional quality and population growth are commonly seen in growth regressions, however they are dismissed here given the estimation results seen in the reviewed papers. First, human capital is hard to capture in growth regressions performed with panel data, as Islam (1995) reminds. Insignificant and/or negative values may arise due to a divergence between theory and practice. Besides this, data availability is a common obstacle, especially in the case of Sub-Saharan Africa. Records for school enrollment in the region are either scarce or not available, given the long period considered in this thesis, which can lead
to very unbalanced panels, distorting and/or biasing the results. In addition to this, the literature16 exploring the relationship between institutional quality and economic
growth relies in its majority on the World Bank’s Worldwide Governance Indicators (WGI), which contains dimensions of governance like voice and accountability, regulatory quality and rule-of-law. This database also comprises Sub-Saharan economies, but it is only available from the year 1996 onwards on a two-year basis. The Ease of Doing Business and the Fraser Institute’s Economic Freedom of the World Index are also used by Bruinshoofd (2016) and Góes (2015) respectively as a proxy for institutional quality. However, they also have data availability problems with respect to the time frame and countries needed for the analysis in this paper. Regarding the population growth variable, only Datta and Agarwal (2004) find significant values for the population growth variable among the papers analyzed. The others either do not use it as an explanatory variable or the values are insignificant. Also Sala-i-Martin’s (1997) ‘I just ran two million regressions’ does not give support to either a significantly positive or negative impact of population growth on economic growth17.
However, the estimation of equation (1) presents a challenge. As discussed through the papers reviewed in section 2, the causality between telecommunications and economic growth might transit in both directions, meaning that not only a higher level of telecom infrastructure would affect growth positively, but also a higher economic growth rate would result in higher ICT levels. This reverse causation also applies to the other explanatory variables and to the lagged dependent variable. Therefore, one needs to control for this joint endogeneity of the regressors towards the dependent variable (Calderón and Servén, 2004).
16 Please refer to Easterly and Levine ﴾2003﴿, International Monetary Fund ﴾2003), Kuncic ﴾2013﴿ and Fabro and Aixalá ﴾2013).
17 Barlow (1994) provides a discussion on the insignificant correlation between population growth and economic growth, suggesting that in two-variable models there might be a partial negative correlation between economic growth and population growth.
3.2 Estimation Method
Applying the Generalized Method of Moments (GMM) can solve the above-mentioned endogeneity problem. However, differently from Lee et al. (2012) and Batuo (2014), who respectively chose a two-step difference and a one-step system GMM to overcome such issues, a two-step system GMM is adopted. The system-GMM provides more efficient results than the difference-GMM, according to Blundell and Bond (1998).
In order to understand how the system-GMM works, it is helpful to understand the difference-GMM framework first. Arellano and Bond (1991) proposed the latter method, where primarily the equation is first-differenced in order to eliminate the fixed effects. Equation (1) would then become:
€
Δgdpgri,t =αΔgdpgri,t −1+βΔgdpi,t −1+δΔteli,t+γΔXi.t+ Δυi,t (2)
The fixed-effects are purged away, but the right-hand side variables remain endogenous. The term Δgdpgri,t-1, for instance, is correlated with the error term Δυi,t
(gdpgri,t-1 in Δgdpgri,t-1=gdpgri,t-1-gdpgri,t-2 correlates with υi,t- in Δυi,t=υi,t-υi,t-1). The same
also applies to the other regressors considered to be endogenous. In order to overcome this situation, one must find valid instruments for such variables (Roodman, 2006). Arellano and Bond (1991) suggest the use of internal instruments of the respective variables, being gdpgri,t-2 the first possible candidate since it is related with
Δgdpgri,t-1, but not with the error term ΔυI,t and at the same time it does not reduce
sample size, as when instrumenting with the difference Δgdpgri,t-2 (Roodman, 2006).
Assuming that the error terms υi,t are not serially correlated and that the regressors
are weakly exogenous, instrumenting a variable with its second or further lag onwards is acceptable as long as they are not correlated with the error term. Notwithstanding, the difference-GMM can lead to biased results when the right-hand variables are persistent over time (Blundell and Blond, 1998). Also, when the data is partially available, it can augment the gaps when taking the first-differences (Roodman, 2006).
In order to overcome these shortcomings, Arellano and Bover (1995) and Blundell and Bond (1998) present a different approach. The authors combined two equations into a system where, beside the difference equation used in the
difference-GMM and its lagged level instruments, a level equation is considered using as instruments lagged differences of the endogenous variables. We have then a system with equations (1) and (2) combined. In the level equation another assumption is required for the validity of the instruments: the lagged differences are not correlated with the individual fixed effects in the level equation (Blundell and Bond, 1998). Moreover, system-GMM can be differentiated in one-step and two-step. According to Windmeijer (2005):
“One-step GMM estimators use weight matrices that are independent of estimated parameters, whereas the efficient two-step GMM estimator weighs the moment conditions by a consistent estimate of their covariance matrix. This weight matrix is constructed using an initial consistent estimate of the parameters in the model.”
Two-step system GMM is preferred to one-step GMM due to its efficiency. However, one drawback of the two-step system GMM is that its standard errors are mostly downward biased. Nevertheless, the command xtabond2 in Stata adjusts those according to the Windmeijer’s (2005) finite-sample correction making then more reliable. As Roodman (2006) states, the use of a system-GMM allows us to explore more moment conditions of the data in levels, but one should ponder the number of instruments. Too many instruments can on the one hand weaken the Sargan/Hansen test and on the other hand ‘overfit endogenous variables’. Therefore, if not stated the contrary, the minimum number of instruments is considered for each variable. For the lagged dependent variable (gdpgri,t-1) and the lagged level GDP per capita level(gdpi,t-1)
the lag one level value for the first-difference equations and the first-differenced lag one value for the equations in level is used. For the telecommunications variable teli,t
and the Xi,t control variables matrix the same apply, but using the second lagged
values instead of the first, respectively.
Moreover, in order to verify the validity of the GMM estimation, two tests are applied: the Arellano-Bond test checks for the autocorrelation in the residuals while the Hansen test checks if the specification used is not over identified. The null hypothesis of the first test is that the error term υi,t is serially uncorrelated. Therefore,
expected in first difference transformations, since Δυi,t=υi,t-υi,t-1 and Δυi,t-1=υi,t-1-υi,t-2
share a common term, υi,t-1. Consequently, autocorrelation in level equations is
detected observing second order correlation (and upwards) in differences. Furthermore, the null hypothesis of the Hansen test is that all instruments are valid and the model is correctly specified, meaning the instruments are together exogenous and uncorrelated with the residuals. As a result, the p-value should be as big as possible in order not to reject the null hypothesis (Roodman, 2006).
3.3 Data
The underlying sample of this empirical analysis consists of 44 Sub-Saharan African countries, which are listed on Appendix 2. Liberia, São Tomé and Principe, Somalia and South Sudan are not included due to data unavailability. All data used for this study stem from the World Development Indicators (World Bank, 2016). The time span considered ranges from the year 1980 until 2014. This 35-year period was chosen because it allows for considerable variation in all the variables throughout these years. In the case of telecommunications, landlines are already present in some nations from the 1980s, while mobile penetration is first seen in the mid-1990s following exponential growth rates from the 2000s on as seen on Figures 3 and 4. Furthermore, for many African countries data availability prior to 1980 is scarce.
Although not taken into consideration by the researched telecommunications literature utilizing the GMM method – Lee et al. (2012) and Batuo (2015) –, I follow Islam (1995) and the majority of the empirical growth studies that suggest to average the data in 5-year periods. This transformation aims to mitigate business cycle effects that may emerge since growth is highly persistent. Moreover, a small time frame combined with a large number of entities is an underlying assumption of the GMM estimation. As Roodman (2006) points out, the number of instruments on both difference and system approaches tend to explode the larger the time frame is. On the other hand, for a small number of entities the Arellano-Bond autocorrelation test may be biased. Consequently, for this investigation, seven sub-periods are defined: 1980-1984, 1985-1989, 1990-1994, 1995-1999, 2000-2004, 2005-2009 and 2010-2014.
Table 1 displays the summary statistics for the variables used in the regression considering the seven 5-year periods for all countries. The numbers confirm the substantial cross-country variation for all variables. The growth rate of real GDP per capita for instance has a mean of 1.3% per year, but ranges from -12% to 52% per year. The former rate is attributed to the Democratic Republic of Congo for the period from 1990 to 1994. At this time, a combination of political instability, ethnical conflicts, the dissociation with allies in the developed world and the cease of international aid led to a economic and political collapse in the country (Putzel, Lindemann and Schouten, 2008). On the other extreme is Equatorial Guinea, whose discovery of offshore oil along its coast made the GDP skyrocket between 1995 and 1999 (McSherry, 2006). Not surprisingly, the country also presents the highest levels of investment (180% of the GDP) and trade (440% of GDP) for the same time span. Telecommunications density is also quite heterogeneous. The highest penetration rates are seen for the last period of observation, namely from the year 2010 until 2014 in Gabon. There, 171 mobile phones for 100 people were found while the number of landlines in Mauritius was around 30 per 100 inhabitants, a timid number compared to cell phones. When observing the telecom density per 100 people, that means, when both means of communications are added, Gabon possesses the highest
Table 1: Summary Statistics, 1980-2014 in 5-year periods
Variables Mean Standard Deviation Minimum Maximum
Growth rate of real GDP per capita (%) 1.295 4.711 -11.97 51.47 Real GDP per capita (current USD) 1 290 2,339 119.1 20 816
INV/GDP (%) 21.64 15.98 0 179.9
GOV/GDP (%) 16.27 7.721 0 50.94
OPEN/GDP (%) 74.69 47.24 12.88 440.7
Fixed lines (per 100 people) 2.091 4.561 0.0184 29.79
Mobile telephones (per 100 people) 14.94 29.20 0 171.3
Telecom density per 100 people 17.13 31.35 0.0206 172.7
rate with an increase of one unit in comparison to the number of cellular phones, amounting to 179%. According to The Report: Gabon 2014 (Oxford Business Group, 2014), the country – among other nations in the region – might be reaching a saturation point for mobile telephony. However, since the majority of cell phones are prepaid accounts, one user can hold several different SIM cards, not necessarily utilizing all of them at the same time. Thus, such mobile telephony saturation numbers might be exaggerated. On the other hand, sharing phones is a common practice in Africa, as Aker and Mbiti (2010) point out. In Kenya for instance, one third of the population states that they share their mobile phones with either relatives or friends. A possible reason for this behavior entails the cost motive, especially in poorer countries. Despite the fact that the mobile penetration rate per se is hard to measure, it doubtlessly lays above the 100% mark in the case of Gabon, according to the Oxford Business Group (2014), indicating the high reach ICT has over the population. It is in the interest of this thesis to analyze if such diffusion triggers economic growth in the Sub-Saharan region.
4. Estimation Results
Under this section, chapter 4.1 presents the results for the benchmark equation (1) considering the whole sample and data averaged in 5-year periods. Moreover, the presence of network externalities and a sample variation is tested. Afterwards, two robustness tests are displayed in section 4.2.
4.1 Reference Equation Results
Table 2 summarizes the results obtained over the impact of telecommunications in Sub-Saharan Africa between the years 1980 and 2014 divided in seven 5-year-averaged observations, as discussed in section 3.3. The methodology used was a two-step system GMM proposed by Arellano and Bover (1995)/Blundell and Blond (1998). Regression 1.1 in the left column represents the estimation of equation (1) in Chapter 3.1. The outcome shows that the lagged value of the GDP per capita gdpt-1 is
significant at the 1%-level and has a negative sign, supporting the convergence hypothesis of the neoclassical growth theory. It can be thus confirmed that the GDP per capita in the Sub-Saharan African region tends to grow in a faster pace in comparison to richer countries. The lagged value of the growth GDP per capita and the investments variable are both significant at the 95% confidence interval. For the latter, this result is expected implying that the more it is invested in a country, the more the economy grows. For this baseline regression, the coefficients for government consumption and openness (captured by trade as a share of GDP) are respectively negative and positive, but statistically insignificant. This condition will be further scrutinized in the next section.
Table 2: Telecommunications and Economic Growth in SSA (full sample, 5-year periods)
Variables (1.1) (1.2) (1.3) (1.4) (1.5) (1.6) gdpgrt-1 0.226** 0.275 0.265*** 0.210** 0.237 0.231* (0.0918) (0.185) (0.0881) (0.0983) (0.161) (0.127) gdpt-1 -0.000700*** -0.000429 -0.000675*** -0.000510*** -0.00124 0.000989 (0.000165) (0.000463) (0.000162) (0.00018) (0.00131) (0.00122) inv 0.141** 0.155 0.147** 0.128** -0.0277 -0.0231 (0.0621) (0.136) (0.0623) (0.0523) (0.0491) (0.0411) gov -0.128 -0.216 -0.129 -0.1074 -0.134 -0.0533 (0.103) (0.135) (0.105) (0.101) (0.089) (0.112) open 0.0176 -0.0129 0.00984 0.0080 0.00946 -0.0308 (0.0325) (0.074) (0.031) (0.0171) (0.0241) (0.0248) fix 0.164 (0.165) mob 0.0223*** (0.007) tel 0.0212*** -0.0055 0.0365*** 0.0990*** (0.00687) (0.01394) (0.0135) (0.0217) tel2 0.000193* -0.000886*** (0.000103) (0.000249) constant -0.861 2.068 -0.468 -0.245 2.789 2.743 (2.434) (3.815) (2.442) (2.094) (1.851) (1.936) Observations 248 248 248 248 204 204 Number of countries 44 44 44 44 36 36 Instruments 13 13 13 19 19 22 AR(2) 0.177 -0.0259 0.217 0.0821 1.382 0.945 p-value of AR(2) 0.859 0.979 0.828 0.935 0.167 0.345 Hansen J statistic 2.125 4.706 2.329 14.31 18.02 17.76 p-value of Hansen 0.908 0.582 0.887 0.216 0.115 0.218 Standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1)
Since the main goal of this study is to identify the relationship between ICT and economic growth, I now concentrate on the telecommunications variable. In equation 1.1 the regressor representing the sum of landlines and mobile phones per 100 people is positive and statistically significant at the 1%-level. A positive relationship between the two variables goes in line with the findings obtained by all the literature reviewed in chapter 2, regardless of methodology applied. To analyze the impact expressed in numbers the following is considered: since tel is measured per 100 people, an increase in one unit represents a 1% increase in telecom density. Further, growth of GDP per capita is represented in percentages. Therefore, the coefficient of 0.0212 indicates that an increase of 10% in telecom density per capita increases GDP by 0.212%. This number is half of the one found by Batuo (2015) using the one-step system GMM and bellow the ones found by cross-sectional analyses, although none of those considered just the Sub-Saharan African region in their sample. In addition to it, Lee et al. (2012) did not find any significant result for this variable. Such an increase means that – following a calculation done by Lee and Levendis (2013) – for an economy to grow 1%, the telecom density would have to increase on average by 47,5%. Although seemingly high, this rate appears to be plausible for a sample where only 8 countries reached a “saturation” rate of 100%, in the latest 5-year period and the mean for telecom density amounts to 17.13% on average.
In equations 1.2 and 1.3, the telecommunications variables are considered individually as landlines and cell phones. Equation 1.2 dealing only with fixed phones is in its whole insignificant, whereas the one where just mobile phones are taken into consideration exhibits very similar results to the ones found in equation 1.1. The significance levels are the same as in 1.1 and the coefficients vary little. The value of the variable representing mobile phones is positively significant at the 1%-level adding up to 0.0223. Lee et al. (2012) find a coefficient of 0.0191, a little smaller than the findings of this paper, while Waverman et al. (2005) find a coefficient of 0.039 but for a mixed sample of developed and developing countries and a shorter time frame. The differences to the results obtained in this paper are probably due to the augmented period and the narrowed region analyzed here. Moreover, since the telecom density variable has a value of 0.0212, one can say that it almost entirely
captures the impact of mobile telephony. Given the non-significance of the landlines coefficient and the skyrocketing expansion in Sub-Saharan Africa of mobiles phones per 100 people in comparison to fixed lines (see Figure 3 and 4), one can further infer that cell phones have a greater importance as a mean of communication and as a propeller for growth in the region than fixed lines.
Following Roeller and Waverman (2001), Datta and Agarwal (2004), Lee and Levendis (2013) and Batuo (2015), equation 1.4 checks for network externalities in telecommunications. The squared value of the telecom density variable is added to the equation as in 1.1. to check for non-linearity in telecommunications. This means that for pen2>0 we find increasing returns, indicating that the impact of additional
telecommunications is bigger where it is already present. The opposite, pen2<0,
suggests diminishing returns, that is, countries with less ICT infrastructure tend to profit more than countries where it is already well established. A coefficient equal to zero (pen2=0) indicates constant returns. The outcome points out in the direction of
increasing returns, since pen2 is equal to 0.000193 and statistically significant in the
10%-level18, in accordance with Batuo (2015). Though insignificant, the negative
coefficient of the telecom density variable pen combined with the positive value of
pen2 might however support the critical mass theory as in Roller and Waverman
(2001), meaning that positive network externalities of telecom on growth are effective after a certain threshold value is achieved. However, due the insignificancy of pen, no conclusion can be inferred.
Regressions 1.5 and 1.619 repeat the procedure of equations 1.1 and 1.4, but
this time examining a smaller sample of thirty-six countries considered to be low- and lower-middle-income nations according to the World Bank for 2016. This implies that these countries have a GNI per capita of less than USD 4 126 a year (see Appendix 3). This sub-sample was chosen in order to analyze if the impact of telecommunication can be higher the lower the stage of development of a country in the sample is. Regression 1.5 shows that the only significant coefficient is the telecommunications one and its impact on economic growth is higher than when considering the whole
18 For the equation 1.4 19 instruments are used, because for variables tel and tel2 the lags from the
second until the fourth order are adopted as instruments.
19 For the equations 1.5 and 1.6, 19 and 22 instruments are respectively used, since the second and the third lags are used as instruments.
sample. The telecom density is positive and significant at the 1%-level amounting 0.0365. This means that an increase of 10% in telecom density pushes up the economy by 0.365%. All other coefficients become insignificant, but follow a same trend for their values as in the main equation 1.1 except for the investments, which turn out to be negative. Additionally, equation 1.6 tests for the returns to scale of telecommunications using its square as additional variable. Both pen and pen2 are
significant at the 1%-level, but differently from the equation considering the whole sample, the coefficients for them are respectively positive and negative indicating diminishing returns. This means that for such countries, investments in telecommunications infrastructure can lead to a greater impact on economic growth.
Test-wise, the p-values for all equations lie way above the 10%-mark rejecting the null of the AR (2) test and confirming the validity of the instruments in the Hansen-test. Thus, these results support the efficacy of the estimations.
4.2 Robustness
In this section two tests will be performed in order to check the robustness of the results obtained in the previous segment. The first involves excluding possible outliers from the sample. Next, a check dividing the sample in three-year instead of five-year periods will be conducted. For this analysis, the equations 1.1 and 1.5 are considered as benchmark for the robustness tests.
One question that arises after obtaining the results in section 4.1 is if the outcome really draws a correct picture for the whole sample. In order to verify this, a q-q plot of the residuals is graphed to check for eventual outliers. If the dataset follows a normal distribution, the dots representing the observations should be close to the reference line. The points away from it might indicate the existence of outliers within the observations. I opt to take the four points outside of the ±10 residual range out of the sample to check if the results in the last section hold. As seen in Figure 5, two points upwards and two downwards are chosen and they correspond to Equatorial Guinea for the periods between 1985-1989 and 1995-1999, Democratic Republic of Congo for 1990-1994 and Rwanda between 1995-1999. As already discussed in chapter 3.3, Equatorial Guinea has the highest GDP growth rate for the
period from 1995 until 1999 due to the discovery of oil while the Democratic Republic of Congo found itself in an economic collapse during this same period. As a result, the nation featured the smallest growth in the whole sample. Rwanda experienced the aftermath of a genocide in the period 1995-1999 while between the years 1985 and 1989 Equatorial Guinea experienced a growth collapse, as commodity prices went down and the CFA exchange rate overvalued (Iman and Salinas, 2008). After they are removed the distribution becomes smoother, as Figure 6 displays.
In Table 3, regression 2.1 represents the main equation 1.1 but without the four outliers described above. Although the autoregressive variable gdpgrt-1becomes
insignificant, the outcome shows that the values of the coefficients for the control variables do not diverge that much, with the only difference that now they are all significant. The government consumption gov, for instance, is significant at the 10%-level and amounts -0.143 instead of -0.128. Furthermore, the openness coefficient
open becomes significant at the 5%-level adding to a 0.375% increase in economic
growth, if the trade volume expands by 1%. The telecom density tel continues to be positive and significant, but at the 5%-level instead. Its value fluctuates marginally, indicating that a 1% point increase in the telecom infrastructure boosts economic growth by 0.0185% instead of the previous 0.0212%. The same procedure is applied for regression 1.5 considering the lower-income countries of the sample.
Table 3: Robustness tests
Variables (2.1) (2.2) (2.3) (2.4) (2.5) gdpgrt-1 0.0954 0.292** 0.215** 0.189*** 0.124* (0.119) (0.136) (0.0991) (0.0593) (0.0679) gdpt-1 -0.000621* -0.000982 -0.000555** -0.00181* -0.000647*** (0.000346) (0.00113) (0.000218) (0.00102) (0.000136) inv 0.0921* -0.0461 0.118* 0.140 0.144** (0.052) (0.0419) (0.06) (0.0884) (0.0629) gov -0.143* -0.151* -0.178 -0.284** -0.219* (0.0763) (0.0797) (0.123) (0.127) (0.123) open 0.0375** 0.0103 0.0289 0.0119 0.0382** (0.018) (0.022) (0.02) (0.0183) (0.0151) tel 0.0185** 0.0342*** 0.0174*** 0.0265** 0.0200*** (0.00949) (0.0113) (0.00507) (0.0107) (0.00752) constant -0.977 3.100* -0.498 1.851 -0.934 (1.914) (1.766) (2.156) (1.863) (2.272) Observations 240 200 445 364 439 Number of countries 44 36 44 36 44 Instruments 13 19 13 13 13 AR(2) 0.91 1.066 0.897 -0.452 -0.187 p-value of AR(2) 0.363 0.287 0.37 0.652 0.851 Hansen J statistic 3.15 14.27 9.54 4.633 8.992 p-value of Hansen 0.790 0.284 0.145 0.592 0.174 Standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1)
Since Equatorial Guinea falls into the high-income economies category, just the Democratic Republic of Congo (1990-1994) and Rwanda (1995-1999) are dropped out. The results are shown in equation 2.220, which shows similar results for the tel as
in 1.5 now with a value of 0.0342. It indicates that a 1% increase in the telecom density leads to a 0.0342% growth in the economy. Besides this, the coefficient for the lagged-dependent variable gdpgrt-1and the one for government expenditure gov are
significant, while the others not. The coefficients, however, do not fluctuate much. A second robustness check is done in regression 2.3 taking into consideration equation 1.1. Here the data is averaged in 3-year periods21 instead of 5-year ones,
resulting in twelve observations. As in regression 1.1, gdpgrt-1, gdpt-1 and inv are
significant with similar values (0.215, -0.000555 and 0.118 respectively). gov and
open remain insignificant, while tel continues to be positively significant, but now at
the 1%-level. Its slightly lower coefficient points out that augmenting telecom density by 10% would increase growth by 0.174%. The possibility of having outliers is also considered for the regression considering data averaged in 3-year periods as in regression 2.5, leaving three extreme points are taken as possible candidates: two on the lower part of the graph and one on the upper side, as Figure 7 shows.
20 For the equation 2.2, 19 instruments are used, since the second and the third lags are used as instruments.
21 The year 1979 is added to the sample in order to have exactly twelve periods with the same amount of years for the sample.
They all represent Equatorial Guinea in the periods 1982-1984, 1988-1990 and 1997-1999 respectively, quite similar to the spans in the 5-year period approach. The 3-year division blends the discrepant residuals for Rwanda 1995-1999 and Democratic Republic of Congo out. According to Figure 8, the residuals follow a more normalized distribution after the three above mentioned periods are disregarded. Nonetheless, the outcome of equation 2.5 reveals that the coefficient for the telecommunications density is significantly positive at the 1%-level and very similar to the one in equation 1.1 accounting for 0.0200. This means that an expansion of 10% in telecom density would represent a 0.2%-growth in the economy. As in 2.1, excluding the possible outliers for the sample made the variables gov and open significant, while these and the rest of the variables indicate no much variation in their values.
The 3-year period robustness check is also performed for the equation considering the poorer countries in the sample 1.5. As regression 2.4 shows, the value of the telecom coefficient is significant at the 5%-level and diminishes by 0.01 percentage point when compared to regression 1.5. This means that a 10% increase in telecom density leads to 0.265% economic growth. Furthermore, the gov coefficient is like in regression 2.2 negative but now significant at the 5-% level. Moreover the lagged GDP per capita coefficient gdpt-1 is like in the other equations negative,
although now significant at the 10%-level. The coefficient for investments inv is positive and shows values similar to the regression considering the whole sample,
despite being insignificant. A regression considering possible outliers is not performed for this subgroup because when data is averaged in 3-year periods, the only three possible outliers represent the Equatorial Guinea, which belongs to the high-income nations group. In this section as well the p-values for all equations lie way above the 10%-mark rejecting the null of the AR (2) test and confirming the validity of the instruments in the Hansen-test. Thus, these results support the efficacy of the estimations.
Overall, the results show that the coefficient for telecommunications is positive and significant for all the equations considered, confirming that access to mobile phones and landlines has an impact of economic growth for SSA. Furthermore, a sub-sample containing the low-income countries in the sub-sample reveal that telecom density affects growth in a greater extent for such nations.
5. Conclusion
This thesis investigates the impact of telecommunications on economic growth in SSA. It adds value to the existing literature by not only focusing on this specific region, but also considering a larger time frame than previous studies for the analysis. Besides this, a more up-to-date GMM method compared to the reviewed literature is applied to account for endogeneity problems between the regressors and the dependent variable. A sample of 44 Sub-Saharan countries is evaluated for the period between 1980 and 2014.
After dividing the sample in seven 5-year sub-periods and employing a dynamic panel in combination with a two-step system GMM, the results outline that telecommunications indeed have a positive and statistically significant effect on economic growth. An increase of 10% in the telecom density boosts per capita GDP growth by 0.212%. The outcomes analyzing fixed lines and mobile phones separately show that while the whole regression considering fixed lines was insignificant, the results obtained in the regression taking into account mobile phones were not only comparable to the ones for the equation including both, but also the telecom coefficients were very similar. The mobile phone variable was thus in its whole
captured by the telecom density variable (a value of 0.223, compared to the aggregate value of 0.212). This indicates that in Sub-Saharan Africa cell phones play a more crucial role as a mean of telecommunication and as a propeller for growth than fixed lines. In addition to it, the results support the convergence theory, implying that poorer countries tend to grow at a higher pace than relatively richer nations. Furthermore, one observes that the impact of telecommunication on growth is even higher when considering only the poorest economies in the region. Numerically, a 10% increase in the telecom density leads to a 0.365% growth in per capita GDP. These low-income nations also display diminishing returns, that is, these countries tend to profit more than countries where ICT infrastructure is already well established. In order to check for the validity of the outcomes, the data was also averaged in 3-year instead of 5-year periods and also possible outliers were excluded from the sample. The results remain similar, showing not much variation in the values.
Therefore, investments in telecommunications infrastructure are strongly recommended for the economic development in SSA and should be encouraged by its governments. ICT diffusion can have further spillovers on financial integration, education and health care. Thus, it can be for the interest of future research to focus on the correlation between these other sectors, economic growth and telecommunications. On a country level, the mobile telephony saturation problem can be investigated; although at the moment it does not affect that many countries and as discussed in chapter 3.3, the actual numbers are hard to measure.
