This section describes the practical approach chosen to analyze the problem of this thesis, where we put ICT capital as explanatory variable to GDP growth. We begin by describing the data selected for this study followed by a classification of countries. Finally, a presentation of the production function used to analyze our problem and some descriptive statistics, are examined.
For us to analyze the impact of investments in information and communication technologies to output growth, we need a comprehensive dataset consisting of variables measuring economic inputs of several countries globally. National statistical offices most often estimate domestic values in their own national currencies and are therefore not directly internationally comparable. Fortunately, the Conference Board3 has conducted a great database for this type of study by collecting economic data for 123 countries worldwide, during the period of 1950 – 2016. The dataset we are ought to use is the Total Economy Database (TED) (2016a) and includes variables for two sources of capital, output growth, labor and human capital. Due to limited data accessibility of ICT capital for developing countries, we end up with a total number of 101 countries. As mentioned in the theoretical part, ICTs have only been considered to affect productivity for several years due to the rapid growth during the 1990’s, after the so-called IT boom (Venturini, 2006; Edquist & Henrekson, 2004). The great increase in these kinds of investments during the past three decades, therefore makes up an interesting period for us to investigate. In accordance with Niebel (2014), we start in 1995 but add another twenty years to gather the impact up until recently. Thus, we end up with a total number of 21 years during the period 1995 – 2015. Additional data which is not covered by the TED, is accessed from the World Bank (2017c).
Since previous empirics have found support for the relationship between ICT and GDP, we assume this relationship to flow from ICT investments to output growth (Röller & Wavermann, 2001; Wolde-Rafuel, 2007) and have, just as Niebel (2014), Yosefi (2011) and Hanclova et al. (2015), put ICT as an independent variable explaining GDP. To distinguish for growth in ICT investments, TED is dividing capital into two categories; growth of ICT capital services and non-ICT capital services. ICT capital services are provided by assets as telecommunication materials, computer hardware and software, while the non-ICT capital services are referred to the growth in the services provided by transports, machinery, buildings, construction and other types of non-ICT assets. Hence, the empirical part is based on capital services instead of capital stocks. According to OECD (2001), Inklaar and Timmer (2013) and Niebel (2014), this measure is more appropriate since it reflects upon the assumption that short-lived assets (computers, phones, etc.) are having greater impact in production, as indicated by user cost of capital. ICT as well as non-ICT capital services, are thus calculated as the growth of stocks in single ICT assets weighted by the share in total ICT capital compensation. As theory suggests, we expect both types of capital services to contribute to the growth of GDP. ICT services are however expected to have relatively greater marginal impacts in the lower income groups, due to the law of diminishing returns on investments (Solow, 1956; Romer, 1990; Gottfries, 2013). Since developed countries are assumed to hold better prerequisites, they should in general also be regarded as more efficient users and eager to invest in new technologies. We have therefore reason to believe that the impact of ICT could be high in the high-income group as well (Avgerou, 2003)
We include labor quality as one of the control variables since we assume human capital to influence the level of production (Glaeser et al., 2004). Labor quality consists of three levels of skills; low, medium and high, and is a composition of data sources from different origins4 (The Conference Board, 2016b). Since these sources are set up differently in terms of definitions, time and countries, TED has estimated the relationship between them using a seemingly unrelated regression (SUR) model. The employment data covers all employees, including self-employed, apprentices and the military, engaged in productive activities within the country’s border. The data is gathered from different sources by the Conference Board5. The level of employment and human capital are vital factors for raising output and this study will thus expect positive effects on GDP growth in most of the income groups (Solow, 1956; Romer, 1990; Gottfries, 2013).
From World Development Indicators, we add export as a measure for a country’s openness to the rest of the world (World Bank, 2017c). In addition to labor quality, we follow Niebel and assume that the level of trade and export are important factors for GDP growth. Theory suggests that the more globalized a country is, the easier it is to accumulate capital and access new technology through the exchange of knowledge across borders (Romer, 1990; Dollar and Kraay, 2002). The expectations about exports having positive effects on GDP growth are therefore present in our empirical analysis.
To grasp potential disparities between countries, especially different income levels, we separate them in accordance to their level of development. We classify these as different income groups rather than dividing them into developing-, emerging- and developed economies. We follow the country classification table drawn by the World Economic Situation and Prospects, WESP (United Nations, 2017), where each country is categorized into the low-, lower middle, upper middle or the high-income country group. The classification is dependent on each country’s gross national income (GNI) per capita. Countries having a GNI less than $1.025 per capita are drawn under the category of low-income countries, countries between $1.026 and $4.035 are classified as lower middle income, upper middle income countries between $4.036 and $12.475 and high-income countries for those with more than $12.475 per capita. All countries can be found in table A1.1 in Appendix 1.
The empirical model in this study is based on the neoclassical Solow model, presented in the theoretical framework. The analytical framework follows the methodologies outlined by Yousefi (2011) and Niebel (2014) in their studies. We put GDP growth as dependent variable by having capital, labor and technology on the right-hand side of the equation, without the assumption of constant returns to scale. The contribution of capital to GDP is measured by having two capital variables; ICT and non-ICT capital services. In addition, we augment the production function by including two control variables, export and labor quality, since those are considered to affect the level of production as well. We end up with the following augmented production function:
where gYi,t is the annual growth rate of GDP, gKICTi,t and gKNICTi,t are the growth rates of ICT and non-ICT capital services respectively, and gEi,t is the growth rate of employment. Finally, Xi,t stands for the additional variables for export and growth of labor quality. All variables apart from exports, are presented in annual growth rates as the log change of previous year’s value and are measured in terms of 2015 US Dollars purchasing power parities (PPP).
We run panel estimations for the four different country groups by using two types of models; pooled ordinary least square (OLS) and fixed effect models (FEM). We cut of the 5% highest and lowest observations and replace them with the 5th and 95th percentile value, to avoid problems of extreme outliers in the data. The process is called Winsorising and is a more robust method to correct for outliers, in comparison to trimming (IHS EViews, 2010). To avoid problems of non-stationarity, a Fisher Augmented Dickey-Fuller test for panel data analysis is conducted (Appendix 3). The ADF results do not find any problems of non-stationarity in the aggregated sample, nor in the high-income group. Problems of unit roots are solved among the other three groups by differencing the affected time-series. The drawback is that we end up with less number of years. Consequently, for all groups except the most developed we can only estimate the period 1996 – 2015. It should be noted that the Jarque-Bera normality tests indicate non-normal distributions among residuals in the highest income and lower middle income countries, as well as for the total sample (Appendix 4).
The OLS regression is used as a benchmark for the analysis, where the assumption about no individuality among countries is present. They are all having the same intercept and the same slope. The assumption that all countries are holding the same characteristics is somewhat unrealistic (Gujarati & Porter, 2009) and not the purpose of this study. Countries should not hold similar prerequisites for using factors efficiently in production, due to differences in investments behavior (Avgerou, 2003). We will therefore adopt a FE model to control for unobserved heterogeneity across countries, by letting all nations to have individual intercepts and control for time-invariant qualities to not affect GDP growth. The difference between the two models will show if some constant variables omitted from the model, affect the contribution of ICT to GDP growth. Both models are estimated with robust standard errors, using Whites period coefficient covariance estimator to remove potential problems with heteroscedasticity (Eviews, 2016).
Below, there are four tables reporting summary statistics for each income group. Every table gives an overview of the statistical nature in each subgroup by presenting the values of the mean, median, maximum, minimum and standard deviation for each variable. The summary statistics for the full sample can be found in table A2.1 in Appendix 2.
As can be seen in table 4.1 – 4.4, the mean value for the annual change in ICT capital investments are presented for each income group. As expected, the highest growth rate is reached within the high-income group. This observation is consistent with previous literature and implies that the richer an economy is, the more it tends to invest in ICT due to the knowledge and experience in how to use and implement these assets (Avgerou, 2003; Beil, Ford & Jackson, 2005; Venturini, 2006). The highest maximum value of 65.2% is presented in the low-income group, indicating that the largest investments are not placed in any of the top-three highest income groups. The change in non-ICT investment is still relatively high in the most developed countries, yet not as large as the investments for ICT in any of the four income groups. This implies that some of the less developed countries have invested heavily in ICTs during the past two decades. This is in accordance with the findings of Niebel (2014) and Dedrick et al. (2013) and might display the realization of the so called leapfrogging effect discussed by Steinmuller (2001).
Another main variable in focus for the empirical part is the annual growth rates for GDP. In accordance with the neoclassical theory, the highest rate is reached in the low-income group followed by the lower middle income group. The same relationship can be noticed for the mean value of employment growth, as the highest value comes from below and ends with the lowest mean in the high-income economies. These recognitions display the observations by Dedrick et al. (2013) and Niebel (2014), about the rise in income among less developed countries during the past two decades. Furthermore, the rise might represent the success from increased investments in ICT and non-ICT capital as well as increased employment rate which should have a plausible effect on output growth (Solow, 1956; Romer, 1990; Gottfries, 2013). Although the highest average values for GDP are reached for the low-income group, the highest minimum values are reached for the upper middle and in the lower middle income group for the employment rate.
For labor quality and exports, higher rates are explored in the high-income group. The unrealistically low median as well as the relatively lower mean value for the upper middle income group, might indicate that there could be some biasedness or measurement errors in the data for export. However, the tables give an overview about the relatively lower rates in the low-income groups in comparison the developed.
In this section, a presentation and analysis of the main findings from the regressions will be presented. We investigate whether ICT capital investments and their impact on GDP growth differ between countries at four levels of income.
Table 5.1 examines the results from the regression for the aggregated sample in period 1995 – 2015. The growth in ICT capital services is insignificant for the two models with only a few percentage points (0.012% and 0.014%) differing the output elasticities. For the non-ICT capital services, the elasticities range from 0.479% to 0.437%, indicating a higher importance in relation to ICT. The growth of employment seems to have somewhat higher importance for economic growth when accounting for country-specific effects (0.521%) compared to the OLS (0.458%). When controlling for labor quality and export in the regression, we only find significant results for the labor variable. When we exclude the control variables, we find that ICT capital has a significant effect in both models, as can be seen in table A5.1.
Since the purpose of this study is to compare the contribution of ICT growth across different countries, the output in table 5.2 presents the results for each subsample for the same period. When including exports and labor quality, the low-income group do not find ICT contributing to GDP growth at all. In this group, we also find the only significant export coefficient. On the other hand, in the FE model for high-income countries the labor quality seems to find its only significant impact on GDP. Although insignificant results, the control variables seem to affect the impact of ICT on GDP positively for the lower middle income group, where the coefficient increase from 0.0405% and 0.014% to 0.044% and 0.027% in OLS and FEM respectively. For the high-income group, the impact from the control variables are mixed, with a slight increase in OLS for ICT and a decrease in FEM. In the other two groups, when including exports and labor quality, we get lower elasticities of ICT capital.
When we exclude labor quality and export, ICT capital is significant in the three highest income groups with an exception for FEM in the lower middle income group. The elasticities reveal somewhat higher values in all subgroups in comparison to the aggregated sample. The magnitude of the non-ICT capital services is considerably higher for all subsamples, in comparison to the ICT services. It is surprising to see is the elasticity in the upper middle income countries is around 1.3. This is a remarkably high value in relation to the other groups. The variable suggests a 1% increase in growth of non-ICT capital services would affect growth of GDP with 1.3%.
Analysis & Discussion
The output from the empirical part presents interesting results. As expected, the result for the full sample in table A5.1 shows a positive relationship between ICT capital services and GDP growth rate, supporting previous work made by both Niebel (2014) and Yousefi (2011). The values of capital services provided by non-ICT assets are considerable higher, implying that classical assets like buildings, cars and machinery should still be more important for the global growth in relation to new technologies within the ICT sector. Increased coefficients are observed when including export and labor quality as control variables. However, there are no longer any evidence for ICT capital services affecting growth. Consistent with Gottfries (2013) and Glaeser et. al (2004), labor quality seems to have an important role for growth in the full sample. Since traditional theory is predicting exports to have a positive effect on growth (Romer, 1990; Dollar and Kraay, 2002), it is surprising to see the opposite in our empirical findings. The insignificant results might however indicate that there could be some measurement errors. Since these results are aggregations of countries with different income levels, we should not assume that these values are true for all countries globally (Avgerou, 2003; Dimelis & Papaioannou, 2009). Again, all economies do not share the same prerequisites for investing and using new technologies, one should therefore not rely on the results based on an aggregated sample of different countries.
The impact of ICT on GDP growth affects the four subgroups in the output differently. Regarding ICT capital services in the high-income group, there are no differences in the significance level between the two models, nor any remarkable changes in the contributions to GDP growth. This could suggest that there are not many unobserved heterogeneity effects, constant over time, that affect the impact on the contribution of ICT to GDP growth rate. The significance levels are in line with both Yousefi and Niebel as well as theory; high-income countries should have a high stock of ICT capital and historical experience in how to efficiently use these assets. When allowing every country to have its own intercept we notice that labor quality becomes significant to GDP growth. This might partially support the statement by Avgerou about context-specific characteristics and that these could affect the educational contribution to GDP. In addition, it is expected that these countries have developed appropriate infrastructures supporting these types of assets. Our results somewhat contradict the statements made by Edquist and Henrekson as well as Venturini, stressing that growth in high income countries are today driven by investments and usage of ICT capital. The contributions of non-ICT related capital shown in our output, seem to still be relatively more important in relation to ICT capital services.
In the upper middle income group, the results show reduced importance of ICT capital services in relation to the high-income group. This is not in favor of previous research and theory studying upper middle economies, which shows that these countries should have higher elasticities of ICT compared to higher income economies (Yousefi, 2011; Hanclova et al., 2015). On the other hand, the results about relatively lower contributions in this income group are in line with the findings of Niebel, if one consider the upper middle income group as ‘emerging’ economies. Interestingly, when including both control variables, the significant contribution of ICT investments to GDP growth in FEM disappears. The result estimated by the fixed effects model now suggests that ICT capital services no longer is contributing to the growth of GDP. Non-ICT capital services however shows increased importance when controlling for omitted variables. Our results also show that labor qualities are insignificant in this group. This could be another explanation why ICTs do not affect GDP growth in these countries. Researchers argue that the labor force in low-income countries do not have the required education, nor skills, to handle these technologies efficiently and are thus not able to benefit from high-tech investments of ICT (North, 1991; Acemoglu, Johnson & Robinson, 2001). This could also be a consequence of low institutional qualities, since King et al. (1994) and Glaeser et al. (2004) state that institutions have a major role in the development of human capital and vice versa. However, these results measure the growth in labor quality and the effects on growth in GDP from one year to another, implying that the change in human capital does not contribute to the change in GDP output. This could mean that the change in labor quality are time-invariant over the years and is therefore not contributing to the change in GDP growth rate.
The contribution of ICT capital services in the lower middle income countries is like the results found in the upper middle income group. When controlling for fixed effects the variable is no longer contributing to growth of GDP. As in the upper middle income economies, a potential reason for this could be the impact of low quality institutions. The results for the non-ICT capital however, show much less importance in explaining the GDP growth in relation previous group. This is not consistent with the theory presented by the neoclassical model about the law of diminishing marginal return on capital, which states that countries with relatively lower stocks of capital should have higher elasticities. On the other hand, the contribution seems to be higher in this group in comparison to the high-income group.
For low-income countries, we do not find any significant contribution from the growth in ICT capital to GDP growth. Here, the growth of non-ICT capital services seems to be of higher importance. This might be explained by the suggestion of Wavermann, Meschi and Fuss (2005) about the fact that lower income countries seem to hold less incentives to invest in infrastructure for new technologies. The explanation to this is that country-specific characteristics and established infrastructure, do not support these kinds of investments. This could imply that lower income countries in general suffer from lower contributions of ICT in relation to other types of capital that meet more primary needs. Wavermann et al. also argue that another explanation for why less developed economies exhibit insignificant contributions from ICT capital could be the consequence of the so-called telecom income trap. This implies that the cost of ICTs exceeds the gains in productivity. The lower levels of income might also affect the support for building efficient infrastructures for telecommunications, as stressed by Röller and Waverman (2001). In comparison to the lower middle income group, the growth in employment seem to be of much higher importance in these types of countries. This could partially be explained by the lower stocks of capital making labor more important in the production. In addition, one might also assume higher unemployment rates in low-income countries, making the growth of employment a much more important variable to GDP growth in relation to previous group.
Overall, the results show that ICT capital still has a more important role among the World’s richest countries. In contrast to Niebel, we do not find any significant impact of ICT in the three lower income groups when accounting for country-specific effects. Interestingly, both middle-income groups exhibit significant impacts of ICT in the OLS model. It can however, be noticed that the fixed effect model fails to support the effects of ICT capital on growth in most countries. One could therefore assume there is some time-invariant heterogeneity aspect that rendering ICT negligible when accounting for it. Many less developed countries are today affected by current or historical consequences of war, famine and poverty and might therefore find it hard to establish stable economies and increased growth. This is in line with the research of North (1991) and Acemoglu et al. (2001), which arguing that historical aspects still influence the configuration of institutions today.
Table of Contents
2. Background & Related Literature
2.1 Definition of ICT
2.2 Related Literature
3. Theoretical Framework
4. Empirical Framework
4.2 Country Classification
4.3 Empirical Model
4.4 Descriptive Statistics
5. Empirical Analysis
5.2 Analysis & Discussion
GET THE COMPLETE PROJECT