Role of government in telecommunications infrastructure investment
Yifan Gu S1796933 Supervisor:
Prof. dr. Marcel Timmer Co-assessor:
dr. Bart Los
Research Methodology Supervisor:
Prof. dr. Erik Dietzenbacher 17/01/2013
Abstract
In this paper, we investigate the relationship between GDP per capita growth and telecommunications infrastructure investment for 161 countries from 1985 to 2005.
Results are used to make policy recommendations and suggest a proper government role in telecommunication provision. In detail, we find that mobile phones approximately have greater growth effect or returns than fixed- line telephone and there is no critical mass effect for either mobile phone or fixed- line telephone, but a non- linear relationship between fixed- line phone and economic growth. In our opinion, governments should encourage private firms to invest in mobile telecommunications infrastructure while invest in fixed-line telecommunications infrastructure themselves in some low fixed-line penetration countries.
Key words: telecommunication, economic growth, government intervention
1
Introduction：
Nowadays, mobile phones become a hot topic in our daily life with the invention of iphone (or similar products like Samsum Galaxy). Communications can be made everywhere every time, and new information can be shared through mobile internet immediately. However, it is not enough to have just mobile phones. Certain complements like mobile phone networks and mobile internet should be built in advance. In this paper, we want to investigate the growth effect of telecommunications infrastructure and provide some policy recommendations for governments.
In the Netherlands, most of the telephone operators are private companies, such as Vodafone, T-Mobile and KPN. Moreover, all of these firms have good performances.
For example, Vodafone has 5 million customers in the Netherlands and generates
￡32 billion revenue in the whole European market (Vodafone annual report 2011).
By contrast, in my home country China, big telecommunications companies (like China Mobile, China Unicom and China Telecom) are all government owned companies and their revenues are really high, 528 billion RMB for China Mobile, 215 billion RMB for China Unicom and 245 billion RMB for China Telecom, respectively (China Mobile annual report 2011, China Unicom annual report 2011 and China Telecom annual report 2011). It is difficult to determine which type is better, private provision or government provision. Both of the two types of firms perform well in their market. However, the puzzle is that what the role of government is in telecommunication provision. In other words, should government make the investments itself or leave it to private sector?
The role of government in economic activities is always a major point that economists emphasize on. Start from Adam Smith’s “invisible hand” to Keynesian assertion of government intervention, this debate never stops. Scholars who agree with market economy theory put emphasis on improved competition and market efficiency, while scholars who favor government intervention theory argue that market is not always efficient. A market failure usually happens when there is imperfect information, externality or monopoly in the market. Telecommunications sector, discussed in this paper, may have the potential to encounter a market failure as well due to network externalities (Datta and Agarwal, 2004). Then, the provision of telecommunications infrastructure may be below the optimal level. This phenomenon can be illustrated by Debraj’s (2008) concept of multiple-equilibrium, which states that market equilibrium is determined by expectations. When each individual firm anticipates that other firms will invest in telecommunications, the economy will end up with a higher economic growth. Conversely, if expectations are pessimistic, the economy will arrive at a lower equilibrium. Positive externality is generated because some firms may not invest or invest less, though they have an expectation that other firms will invest. Consequently, it is important for government to react and interfere with the market in order to fix the imperfections.
As mentioned, telecommunications infrastructure has network externalities: the bigger
2
the network, the higher the value is to its users. For example, when a telephone network is small, then few people have strong willingness to pay to participate because the person they want to connect with may not be in that network. However, when the network is big enough, people are more willing to join. Thus, telecommunications infrastructure investment potentially encounters the problem of market failures. That is, when the network is relatively small, there is few incentives for private companies to invest and the provision of telecommunication may not be sufficient and below the optimal level.
Telecommunication is important in our society. It not only benefits our daily life, but also plays an important role in economic development (Roller & Waverman, 2000).
Most obviously, telecommunications infrastructure investments stimulate its related industries by increasing the production of cables, mobile phones, etc. Meanwhile, it benefits other industries by accelerating information diffusion and increasing productivity (Gruber and Koutroumpis, 2010). It is therefore crucial for government to take an appropriate role when making decisions about telecommunications investments especially when there is a market failure. If a proper amount of telecommunication is invested, then it will affect the economic performance positively.
But what is an appropriate role of the government? Roller & Waverman (2000) appeal to investigate the role of government upon telecommunications infrastructure investments, but few researchers followed. This paper, therefore, aims to contribute to this topic. The main objective of this paper is to study the growth effect of telecommunications infrastructure. In detail, we want to investigate the following three research questions: (1) Does telecommunications infrastructure investment has a positive effect on economic growth and how big is it? (2) Which telecommunication service has larger growth effect, mobile phone or landline phone? And (3) Is there a non- linear relationship between telecommunications infrastructure investment and economic growth? We will then use these results to make some policy recommendations: should government invest in telecommunications infrastructure, especially in mobile phone sector and fixed-line telephone sector?
This paper is organized as follow. The next section contains a literature review of the previous researches on this topic. In section 3, the economic and econometric model is presented. The data and its sources are shown in section 4. And in section 5, results are discussed. Finally, conclusions are drawn in section 6.
1. Literature review：
Telecommunications infrastructure includes many different industries, such as fixed- line telephone, mobile phone, radio, internet, etc. In this paper, we will focus on fixed- line telephone and mobile phone industry. These two sectors, especially mobile phone sector, became increasingly important in economic development during past few decades. On the one hand, revenues of direct related industries (e.g. mobile phone producers, network operators) increased a lot. O n the other hand, indirect related industries can also benefit from the development of the telecommunications sector,
3
because costs of doing business have been reduced dramatically due to lower search and transaction costs (Nordon, 1992) and higher productivity of managers (Roller and Waverman, 2000). But potential externalities exist when most of the benefits go to other indirect industries without compensation payment. If, for example, private firms do not take account of the indirect social benefits but only direct benefits, then telecommunications investment may be below the social optimum. Moreover, different from traditional infrastructure, telecommunications infrastr ucture has the characteristic of positive network externalities. That is, the value of the network increases with the increase of the user base (Roller & Waverman, 2000). Then, if the telecommunications infrastructure stocks do not meet a threshold (or a critical mass), incentives for private investment are low. Because of the two kinds of externalities mentioned above, it may be not profitable for private firms to invest in telecommunications infrastructure. Therefore, this study aims to investigate the potential for government intervention in telecommunications infrastructure provision.
Roller and Waverman (2009) have tested the network externalities in their paper. In addition to just testing the growth effect of telecommunication investment, three dummy variables that stand for “low, medium and high penetration rate ” are added into their equations. “low” is defined as penetration rate that is lower than 10 percent, and “high” represents the penetration rate that is over 40 percent, and “medium” is the rest. They investigate the growth effect of landline phones in 21 OECD countries over a 20-year period (1970-1990), and confirm the critical mass effect that high penetration countries have almost two times more growth effect than low and medium penetration countries. Similar results can be found in Gruber and Koutroumpis (2011) paper. Different from Roller and Waverman (2009), Gruber and Koutroumpis ’s (2011) dataset consists of more recent data from 1990 to 2007, and the country sample is larger and more general with 192 countries included. Moreover, a 40% threshold is proved. Their results show that high penetration countries have a much higher coefficient (0.102) than low (0.045) and medium (0.051) penetration countries. These results imply that only after a 40 percent penetration rate, or equivalently 40 out of 100 people own a telephone line, is reached, the growth effect will be significantly improved. We assume higher economic growth is equivalent to higher income generated by private firms in a micro sense. Then if the cost of investment is relatively too high compared to the income effect before 40% penetration rate is reached, there is few incentives for private firms to make investments.
Mobile phones, different from traditional fixed- line telephones, have the advantage that it is portable. Besides, people can use Short Message Service (SMS) instead of making phone calls and most importantly, people can browse the internet or make tele-conference using their mobile phones. All of these differences may lead to a time saving, or equivalently cost saving. For example, a business man could immediately read the latest news on internet and locate a place where he is going to have a meeting.
Lee, Levendis and Gutierrez (2009) is the only study to our knowledge that compare the return of mobile phones and fixed- line telephones. By investigating data of 44 sub-Saharan countries through the years 1975 to 2006, they find that the marginal
4
impact of mobile phones is greater than fixed- line telephones in sub-Saharan region.
Therefore, they argue that more mobile phone investments should be encouraged in sub-Saharan region. Furthermore, they test the degree of substitutability by including an interaction term between mobile phones and fixed- line phones, and conclude that the two technologies are imperfect substitutes.
When doing research about investment in mobile phone and fixed- line telephone at the same time, leapfrog effect is one of the most important issues. Leapfrog effect in telecommunications investment means that for some developing countries with low levels of traditional fixed- line telecommunications infrastructure, it is easier and cheaper to jump into the latest stage and start to provide mob ile phones (Lee, Levendis and Gutierrez 2009). Lee at al. (2009) argue that the upfront cost of fixed- line telephone networks is much higher than mobile phone, so it is better for some developing countries to take advantage of the leapfrog effect and start to provide mobile telecommunications infrastructure.
Examine causality is another way to study the role of government in telecommunications infrastructure investments. Most of the previous literatures argue that there is a two-way causal relationship between economic growth and telecommunications infrastructure investments. In other words, higher telecommunications infrastructure would lead to higher economic growth, and higher economic growth would again encourage more investment in tele communication.
Granger causality test has been applied in many papers in order to check the direction of causality. Early study like Cronin et al. (1991) apply Granger causality test based on state level data of the USA from 1958 to 1988. Their results show a two-way relationship between telecommunications and GDP. Similarly, Shui and Lam (2007) research the causal relationship for the Chinese telecommunication sector. Their findings suggest that bi-directional relationship only exist in eastern region of China, but not in central or western region. Since central and western provinces have lower income level, more telecommunications infrastructure does not imply higher economic growth. This result implies that telecommunications alone is not enough to boost economic growth. Government should invest in other complementary infrastructures (such as education, transport) as well. In addition, Chakraborty and Nandi (2003) apply Granger causality test for 12 Asian countries from the year 1975 to 2000 and they find a two-way relationship for countries with high degree of privatization. For countries with high state investment, only one-way relationship from teledensity to GDP is proved. This may suggests that by encouraging private firms to invest in telecommunications infrastructure, it will eventually end up with a virtuous circle (telecommunication and economic growth facilitate each other continuously). In their recent study, Chakraborty and Nandi (2011) extend the sample base by investigating 93 developing countries from 1985 to 2007, and they find a two-way relationship, especially for countries with relatively lower level of development.
If there is two-way causality, different test is needed. Gruber and Koutroumpis (2010)
5
apply a simultaneous equations system developed by Roller and Waverman (2000). In their model, a demand equation, a supply equation and a telecommunications infrastructure production function are formed as is shown below.
Demand equation: Pen = g(GDPC,TelePr) Supply equation: TeleRev=γ(GDPC,TelePr)
Telecommunications infrastructure production function: ∆Pen = φ(TeleRev)
In the demand equation, “Pen” stands for fixed- line penetration rate (number of fixed- lines per 100 people), which is a proxy for quantity demand. GDPC and TelePr stand for GDP per capita and telecommunication service price respectively.
Telecommunications infrastructure demand is thus a function of income per capita and telephone service price. “TeleRev”, in the supply side equation, is aggregate telecommunication revenue in a country. This equation shows that telecommunications infrastructure is a function of income per capita and telephone service price. Finally, telecommunications infrastructure production function shows the relationship between revenue and investment. That is, the change of telecommunications infrastructure stock is the function of revenue. Together with a macro production function (similar with the production function used in this paper), these set of equations constitute an equation system that endogenizes telecommunications infrastructure into a micro supply-demand model.
Figure 1: Simultaneous equation model
Put it simple, the value of penetration rate “pen” is firstly determined by the supply-demand equations that is circled in the above picture. Then, since penetration
6
is endogenized, economic growth can be estimated by penetration rate without worrying about the problem of reverse causality.
Roller and Waverman (2000) conclude with a significant positive relationship especially when a critical mass is reached. Thus, for low penetration countries, it is theoretically supported by Roller and Waverman (2000) that a “big push” can greatly accelerate economic growth. And in my point of view, if the costs of such a “big push”
are extremely high compared to the returns, it is the government’s responsibility to make this “big push” due to the existence of network externality.
One of the shortages of using simultaneous equation system is that the data requirements are high, though this methodology is highly advanced and accurate.
Among these, telephone service price data are especially difficult to get. As far as we know, the only institution that provides these data is the International Telecommunication Union, which does not provide the data for free. However, other models, which do not rely on these data, are available as well. They correct for reverse causality in other ways. Transaction cost theory model and macro production function model are the most popular models used in previous literature. Datta and Agarwal (2004) paper is taken as a representative using a macroeconomic growth framework that was initially developed by Barro (1991). Economic growth is put on the left hand side while telecommunications variables are put on the right hand side, and lagged value of GDP per capita is controlled for checking the convergence hypothesis in their model as well. Using data from 22 OECD countries through 1980 to 1992 they show that the relationship is positive and significant. In order to confirm that the result is not simply caused by reverse causality, it is further tested using lagged values of telecommunications. They conclude that the results are robust.
Furthermore, squared term of teledensity is included in their model, and the significant and negative coefficient of it indicates a concave relationship between teledensity and economic growth. They conclude that the marginal effect of telecommunications on GDP growth is larger for countries with lower teledensity and the marginal effect is gradually reduced to zero or even negative with the increase of teledensity.
Apart from the growth model, models of transaction costs are frequently used as well, such as Norton (1992) and Madden and Savage (2000). These authors explain the growth effect through the change of transaction costs due to telecommunications investment. They believe that more telecommunications infrastructure make transactions cheaper and then lower transaction costs facilitate economic growth as a consequence. However, their empirical model has not much difference with macro model discussed previously, or a Cobb-Douglas macroeconomic production function is applied. And the positive results they consequently obtained are used to indicate that lower transaction costs facilitate economic growth.
Unfortunately, there are few studies in previous literature that account for the role of government. Most studies focus on the causal relationship and macroeconomic
7
performance of telecommunications infrastructure, but the function of government is usually ignored. Thus, this paper aims to fill this gap and provide more policy implications.
2. Conceptual model Economic model
Three kinds of models have been discussed in the previous section: simultaneous equation model, transaction costs theory model and macroeconomic growth model.
Among them, simultaneous equations system is the most advanced and accurate model but it requires expensive data which we cannot obtain. The transaction costs theory model encounters the disadvantage that it limits the contribution of telecommunications to just lower transaction costs. We therefore decide to use macroeconomic growth model in this study. In order to determine a proper government role, we basically want to test the growth effect of telecommunications infrastructure investment in a Cobb-Douglas production equation developed by Canning and Pedroni (2004). The advantage of applying a macroeconomic production function is that we can obtain a social rate of return, while using microeconomic method (such as cost-benefit analysis), only the rate of returns that directly relate to firm themselves are usually calculated. Social return is important for policy recommending because government should consider the overall social welfare, not only just firms’. Therefore, we are going to use the following production function to estimate the effects of telecommunications infrastructure on output:
GDP = A𝐾𝛼𝐿1−𝛼 (1) Where Gross Domestic Product (GDP) represents total output of a country, A is total factor productivity; K is capital stock; and L stands for total labor force. So GDP is a function of capital stock and labor force and technology. Theoretically, capital (K) and labor (L) are assumed to be positively related to output. And to keep the model simple, capital and labor are assumed to have constant returns to scale.
𝐺𝐷𝑃𝐶 = 𝐺𝐷𝑃 𝐿⁄ = 𝐴(𝐾 𝐿⁄ )𝛼 (2) After rearranging, equation (1) can be changed into a per capita form, where GDPC represents GDP per capita and 𝐾 𝐿⁄ is capital per capita. Since we are going to test the growth effect of telecommunications investment, equation (2) is then transformed into a growth form.
g = a + αk (3) where g and k are the growth of GDP per capita and capital stock per capita respectively. As a further step, telecommunications variable is added to the model.
8
That is,
g = a + αk + βteledensity (4) In equation (4), teledensity stands for the sum of fixed- line penetration rate and mobile phone penetration rate (number of fixed- lines and mobile phones subscribers per 100 people). Since telecommunications investment data is relatively difficult to find, penetration rates are used as proxies for capital stock data.
Econometric model
Based on the economic model developed above, a corresponding econometric model has been formulated:
𝑔𝑖,𝑡 = 𝛽0,𝑖 + 𝛽1𝑘𝑖,𝑡 + 𝛽2𝑡𝑒𝑙𝑒𝑑𝑒𝑛𝑠𝑖𝑡𝑦𝑖,𝑡+ 𝛽3𝐺𝐷𝑃𝐶𝑖,𝑡−1+ 𝜀𝑖,𝑡 (5) where subscript i and t represent different countries and time periods respectively.
Lagged value of GDP per capita is included to check the convergence hypothesis which says economic growth tends to be lower in countries with higher GDP per capita. In addition, an error term is included to cover the effects of other variables on economic growth that are not part of the model. Notice that fixed effect is an important aspect of this model. Roller and Waverman (2000) shows that coefficients are overestimated without country fixed effects. Therefore, the constant coefficient 𝛽0 is allowed to have different values in different country i in this model.
In general, capital stock should have a positive sign and lagged value of GDP per capita should have a negative one according to classical economics. Most importantly, we expect telecommunications infrastructure stock are positively related to GDP growth (Gruber and Koutroumpis, 2010; Roller and Waverman, 2000, etc).
𝑔𝑖,𝑡 = 𝛽0,𝑖 + 𝛽1𝑘𝑖,𝑡 + 𝛽2𝑝𝑒𝑛𝑖,𝑡 + 𝛽3𝑚𝑜𝑏𝑝𝑒𝑛𝑖,𝑡 + 𝛽4𝐺𝐷𝑃𝐶𝑖,𝑡−1+ 𝜀𝑖,𝑡 (6) As a second version of our model, we distinguish mobile phone sector (represented by mobpen) and fixed-line telephone sector (represented by pen) within telecommunications infrastructure. By doing this, we can compare the returns of mobile network and the returns of fixed- line telecommunication infrastructure. As mentioned, mobile phones are expected to have greater growth effect than traditional fixed- line telephones, not only because they have more functions but also because this expectation is supported by Lee, Levendis and Gutierrez (2009) ’s paper. In my opinion, if β3 is greater than β2, then government should encourage private firms to invest in mobile telecommunications infrastructure and if there is a market failure due to externalities, government should make the investment itself until it is profitable for private firms to invest.
In the third version of our model, equation (7) is extended to include dummy variables which distinguish different level of penetration rates. By doing this, we want to test a
9
40% threshold that significantly improves the growth effect.
𝑔𝑖,𝑡 = 𝛽0,𝑖 + 𝛽1𝑘𝑖,𝑡 + (𝛽2𝑙𝑜𝑤 + 𝛽3ℎ𝑖𝑔ℎ)𝑝𝑒𝑛𝑖,𝑡 + (𝛽4𝑙𝑜𝑤 + 𝛽5ℎ𝑖𝑔ℎ)𝑚𝑜𝑏𝑝𝑒𝑛𝑖,𝑡+ 𝛽6𝐺𝐷𝑃𝐶𝑖,𝑡−1+ 𝜀𝑖,𝑡 (7) where dummy variables low is defined as all the countries that have penetration rates that are lower than 40% follow Roller and Waverman (2000). Similarly, high is a dummy variable stands for countries which have penetration rate higher or equal to 40%. In previous research, Roller and Waverman (2000) find that above 40%
penetration rate contributions to economic growth significantly increase. This suggests that the relationship between telecommunications infrastructure and output may not be linear. At low levels of penetration rate, the growth effect of telecommunications maybe small, while influential effects will be found when a critical mass is reached. Thus, dummy variables are included to check this effect.
The final version of our model is again established to test the non-linear relationship between teledensity and economic growth. Following the way suggested by Datta and Agarwal (2004), quadratic terms of teledensity are included.
𝑔𝑖,𝑡 = 𝛽0,𝑖 + 𝛽1𝑘𝑖,𝑡 + 𝛽2𝑝𝑒𝑛𝑖,𝑡 + 𝛽3𝑝𝑒𝑛𝑖,𝑡2 + 𝛽4𝑚𝑜𝑏𝑝𝑒𝑛𝑖,𝑡 + 𝛽5𝑚𝑜𝑏𝑝𝑒𝑛𝑖,𝑡2 +
𝛽6𝐺𝐷𝑃𝐶𝑖,𝑡−1+ 𝜀𝑖,𝑡 (8) As is shown in equation (8), positive coefficients of square terms (e.g.𝛽3 > 0) indicate a convex relationship between teledensity and economic growth while negative coefficients imply a concave relationship. We will calculate the approximate value of penetration rates that maximizing/minimizing economic growth, using the condition 𝑑𝑔
𝑑(𝑝𝑒𝑛)
⁄ = 0 and 𝑑𝑔 𝑑(𝑚𝑜𝑏𝑝𝑒𝑛)⁄ = 0.
After building the econometric model and combine it with theory, several hypothesis tests can be made. The aim of this paper is to make policy suggestions through comparing the growth effect of different investments, fixed capital and infrastructure investment. There are three hypotheses.
Hypothesis 1:
Telecommunications infrastructure as a whole (mobile penetration + fixed-line penetration) has a positive relationship with GDP per capita growth. In terms of equation (5), 𝛽2> 0.
Hypothesis 2：
Growth effect of mobile phone is greater than that of fixed-line telephone. In the language of equation (6), 𝛽3 > 𝛽2.
Hypothesis 3:
10
The relationship between teledensity and economic growth is non- linear. There exists a threshold of 40% penetration rate that significantly stimulate economic growth, or in the language of equation (7),
𝛽2 < 𝛽3 𝑎𝑛𝑑 𝛽4 < 𝛽5.
Or there is a non- linear relationship between teledensity and economic growth, in terms of equation (8),
β3 ≠ 0 𝑎𝑛𝑑 𝛽5≠ 0.
3. The data:
As is described in our econometric model, factors such as Gross domestic product (GDP) per capita, population, overall capital level, fixed- line penetration rate and mobile phone penetration rate are included in this study. In the following section, these data will be discussed in detail.
The first mobile phone, Motorola Dyna TAC 8000X, was introduced into the market in 1984 (The Associated Press, 2005), so our dataset starts from the year 1985, and ends at 2005. The reason why 2005 is chosen is that we want to exclude the influence of worldwide financial crisis happened around 2008. Further, the dataset consists of 161 countries in total. Country list can be found in appendix. Among them, there are 114 developing countries and 47 developed countries according to World Bank definitions. Due to “leapfrog” effect, developing countries might benefit most from rapid innovations (Chakraborty and Nandi, 2011). For some less developed countries, they can skip early stage of telecommunication development and jump into latest level of technology. Therefore, telecommunications as a fast developing sector may lead to more rapid growth in developing countries. Thus, it is important to include developing countries as well and not like previous studies (like Roller and Waverman, 2000 or Datta and Agarwal, 2004) only focus on OECD countries.
GDP per capita measures gross domestic product per person. It is the sum of gross value added by all resident producers in the economy plus any product taxes and minus any subsidies not included in the value of the products. It is obtained from World Bank database and it is measured in constant 2005 U.S. dollars converted by Purchasing Power Parity (PPP).
Explanatory variable K stands for total capital stock. This data is constructed using the method mentioned by Calderon et al. (2011). Initially, a zero capital stock is assumed in the year 1955. Because “the capital stock in 1955 has only a very minor effect on the capital stock that results for 1985, and hence it is immaterial whether we set the level of capital stock at zero or some other arbitrary level in 1955 (p. 5)”. And then the data of investment from 1985 to 2005 obtained from the Penn World Table 7.1 is used to calculate the average growth rate of investment. The average growth rate is then used to calculate the investment level in 1955, and consequently adding all
11
the investments from 1955 to 1985 we arrive at the initial capital stock in 1985. For the years after 1985 and up to 2005, the capital stock is accumulated by adding the investment to the capital stock of last year and depreciated, assuming an average depreciation rate of 8%. These data used to construct total capital are all in constant 2005 U.S. dollars.
Population data is also obtained from the Penn World Table 7.1, which is measured in thousands of people. It is used to convert total capital data constructed earlier into per capita terms.
MobPen and Pen are similar variables for mobile phones and fixed- line phones penetration rates. MobPen measures mobile phone subscriptions per 100 people using mobile telephone service (such as prepaid subscriptions), while Pen measures fixed telephone lines per 100 people. Dummy variables “high and low” will be added on to distinguish different level of penetration rates. High is defined as penetration rate which is higher or equal to 40% while low represents penetration rate lower than 40%.
Table 1: Descriptive statistics of the dataset
Variable Description Recourse Mean Min. Max. Obs.
GDPC GDP per
capita (constant 2000 US$)
World Bank database
6215.96 82.67 62143.34 3186
K Total capital stock
Penn World Table 7.1
2302917 2245.12 1.64e+07 3381
Pop Total
population (in thousands of people)
Penn World Table 7.1
33985.22 41.41 1297765 3381
MobPen Mobile subscriptions
per 100
people
World Bank database
10.59 0.00 127.45 3292
Pen Fixed- lines
per 100
people
World Bank database
16.54 0.02 89.24 3330
The descriptive statistics are shown in table 1. All the means, maximums, minimums and number of observations of the variables used in the model are presented in the table in an attempt to give an overall impression. As can be seen, GDP per capita has the most missing values (5.77% of the total observations), while data on total capital stock and population are well collected without any missing values. Mobile penetration rates and fixed- line penetration rates have less missing values, 2.63% and
12
1.51% respectively. When running regressions, data will be adjusted until there is no missing value. In our case, for example, if the mobile penetration rate for Albania in the year 1990 is missing, then all the data for Albania in 1990 will be abandoned and they are not counted as an observation in our regression analysis. Consequently, we arrive at a common sample of 2959 observations in total which is effective.
4. Results and analysis:
We are going to test hypothesis one first, or is there a positive relationship between economic growth and overall telecommunication infrastructure investment? Mobile penetration rate and fixed-line penetration rate are combined in order to create a new proxy for overall teledensity. The meaning of the new variable teledensity is then regarded as: number of fixed telephone lines and mobile subscribers per 100 people.
Then, regression analysis is conducted according to equation (5). The results are shown below.
g= 0.015+ 0.158*k+ 0.0002*teledensity- 4.30E-07*gdpc(-1)
(0.001) (0.014) (5.05E-05) (1.89E-07)
It is clear that teledensity has a positive effect on economic growth, and it is significant at 1% level. However, hypothesis one still needs to be statistically tested.
Therefore, we are going to test H0: 𝛽2≤ 0 against H1: 𝛽2> 0. The t-statistic equals to 3.96 (=0.0002/0.0000505), suggesting a rejection of null hypothesis and we conclude that overall teledensity is significantly and positively related to economic growth.
Stationarity is one of the most criticized points of research on infrastructures, since both GDP per capita level and infrastructure stocks might be non-stationary series. If these data are used, this will lead to a problem of spurious regression, because there is a danger of obtaining significant regression results from unrelated data. So a Dickey-Fuller test is used here to check for unit roots.
Table 2: Panel unit root test with constant (ADF test)
Statistics Prob.
g 863.150 0.000
Mobpen 594.341 0.000
Pen 225.760 1.000
Results are shown in table 2. The p- value of per capita GDP growth and mobile penetration rate are estimated to be 0.000, thereby rejecting the null hypothesis of nonstationarity. Hence, economic growth and mobile penetration are stationary series.
On the other hand, the p-value of fixed-line penetration rates indicates that they are nonstationary series.
To correct for non-stationarity of fixed-line penetration data, we further test its
13
stationarity with a constant and a time trend. The result is shown in table 3.
Table 3: Panel unit root test with constant and trend (ADF test)
Statistics Prob.
Pen 436.285 0.000
As is shown in the table above, fixed- line penetration rate is stationary with a constant and a time trend. Since economic growth is stationary and fixed-line penetration rate is stationary with a time trend, fixed-line penetration rate need to be de-trended before running regression tests. Therefore, we have changed our econometric model slightly into the following form including a time trend.
𝑔𝑖,𝑡 = 𝛽𝑜,𝑖+ 𝜏𝑡 + 𝛽1𝑘𝑖,𝑡+ 𝛽2𝑝𝑒𝑛𝑖,𝑡+ 𝛽3𝑚𝑜𝑏𝑝𝑒𝑛𝑖,𝑡+ 𝛽4𝐺𝐷𝑃𝐶𝑖,𝑡−1+ 𝜀𝑖,𝑡 (9) Where t represents the time fixed effects that is added to de-trend the fixed-line penetration rates. Equation (9) is an example of the new version of equation (6) after correction. Equations (7) and (8) can be easily transformed into a new version by adding the time trend t and we are not going to list them separately here.
Including mobile phone penetration and fixed- line penetration at the same time, there might be a problem of collinearity. That is, mobile penetration and fixed-line penetration rate data may move together in systematic ways, which means a high correlation between these two variables. A high correlation then leads to a large variance and consequently the OLS estimates may not be significantly different from zero. Therefore, it is important to check for collinearity and correct for this problem if it is in presence. There are several ways to detect a collinearity problem. The easiest way is to calculate the correlation between pen and mobpen. The result shows a correlation of 0.54 with a p-value of 0.000. However, it is quite subjective to conclude that there is no collinearity since the correlation is 0.54. To be cautious, variance inflation factor (VIF) is used to check for collinearity as well.
𝑝𝑒𝑛𝑖,𝑡 = 𝛽𝑜,𝑖 + 𝜏𝑡 + 𝛽1𝑘𝑖,𝑡+ 𝛽2𝑚𝑜𝑏𝑝𝑒𝑛𝑖,𝑡+ 𝛽3𝐺𝐷𝑃𝐶𝑖,𝑡−1+ 𝜀𝑖,𝑡 (10) In detail, an OLS regression is firstly conduct according to equation (10). And then use the R-square obtained from regression analysis to calculate the VIF factor (VIF=1/(1-R2)), which is equal to 4.629 (=1/(1-0.784)) in this analysis. Since the value of VIF is smaller than 5, we conclude that there is no collinearity problem between mobile penetration and fixed-line penetration.
Least squares estimator fails if there is correlation between an explanatory variable and the error term and this is one of the usual problems when investigating telecommunications infrastructure as mentioned previously. Therefore, the Huasman test is used to confirm the existence of endogeneity. The null hypothesis for Huasman test is H0: correlation between teledensity and the error term is zero while the alternative is H1: correlation between teledensity and the error term is not zero. If the null hypothesis is rejected, then teledensity is correlated with the error term and
14
instrumental variable need to be used to correct for endogeneity. The implementation of Hausman test to assess the endogeneity of a teledensity requires mainly three steps.
Firstly, estimate equation (9) and save the estimated residuals as e.g. ehat. Secondly, use ehat as an explanatory variable in the following equation.
𝑔𝑖,𝑡 = 𝛽𝑜,𝑖+ 𝜏𝑡 + 𝛽1𝑘𝑖,𝑡+ 𝛽2𝑝𝑒𝑛𝑖,𝑡−1+ 𝛽3𝑚𝑜𝑏𝑝𝑒𝑛𝑖,𝑡−1+ 𝛽4𝐺𝐷𝑃𝐶𝑖,𝑡−1+ 𝛽5𝑒ℎ𝑎𝑡 + 𝜀𝑖,𝑡 (11) And finally, check the t-statistic for the coefficient associated with ehat 𝛽5 and if the corresponding p- value is lower than 0.01, then the null hypothesis of no correlation between penetration rate and the error term is rejected. The results are shown in table 4.
Table 4: The Hausman Test for endogeneity
Coefficient Std. Error t-Statistic Prob.
ehat 1.000 2.96E-05 33750.36 0.000
As is shown in the above table, a p-value of 0.000 clearly rejects the null hypothesis of no correlation between teledensity and the error term. We find that both current mobile penetration rate and fixed- line penetration rate are correlated with the error term which proves the existence of endogeneity. To control for endogeneity, instrumental variables can be used. In this paper, past penetration rate is applied as an instrument for penetration rate. Moreover, this approach can reduce the influence of causality as well. The reason is simple and intuitive, that is, past penetration rate facilitates economic growth now but current economic growth will not influence the penetration rate in the past. Although, causality is not completely eliminated and OLS method may lead to biased estimates, they still bring us some information about the growth effect of telecommunications. In addition, total capital stock may have a causal relationship with economic growth as well, but this is not the main object of our study. Therefore, we do not correct for the causality problem for total capital stock.
Then, a first regression is processed and the results are shown in table 5 regression (1).
As shown in the table below, most explanatory variables are not significant except for the time trend and growth of per capita total capital stock. Moreover, the value of R-square is quite low (0.062). These insignificant results might be caused by not including the country fixed effects as discussed previously and the quality of results are improved a lot when fixed effects are added in regression (2).
15
Table 5: Regression results: dependent variable g
(1) (2) (3) (4)
Constant 0.001 0.051*** 0.077*** 0.049***
(0.531) (8.212) (9.810) (7.604)
t 0.001*** 0.002*** 0.003*** 0.002***
(6.071) (9.178) (10.041) (8.600)
k 0.157*** 0.130*** 0.130*** 0.130***
(11.366) (7.622) (7.640) (7.594)
Mobpen(-1) -3.01E-05 0.0003*** 0.0004*
(-0.391) (3.016) (1.654)
High 0.00029***
(3.106)
Low 0.00035*
(1.728)
(Mobpen(-1))^2 -1.37E-06
(-0.538)
Pen(-1) 1.42E-05 -0.0013*** -0.0038***
(0.114) (-4.404) (-6.919)
High -0.0014***
(-4.383)
Low -0.0013***
(-3.575)
(Pen(-1))^2 3.98E-05***
(5.296)
GDPC(-1) 4.25E-08 -6.60E-06*** -9.21E-06*** -6.37E-06***
(0.170) (-6.399) (-7.818) (-8.600)
Country fixed effect
No Yes Yes Yes
Time trend Yes Yes Yes Yes
R-square 0.062 0.203 0.212 0.204
Obs. 2959 2959 2959 2959
Prob. 0.000 0.000 0.000 0.000
*Significant at 10% level.
**Significant at 5% level.
***Significant at 1% level.
Note: t-statistics are shown in the brackets.
Country fixed effect is added to the model in regression (2) of table 5. It is essential to include fixed effect into the model and intercept β0 is allowed to vary from country to country, and a “Redundant Fixed Effects” test is carried out to detect whether country fixed effect should be included. The test result (p- value 0.000) clearly rejects
16
the null hypothesis that there is no cross section fixed effect in the regression.
Compare with regression (1), all the coefficients are significant at 1% level now, which is a great improvement. The constant term is equal to 0.051, which indicates the value of economic growth when all the explanatory variables are zero. The growth of per capita total capital stock is significant and meets the expected sign. That is, if the growth of per capita total capital increases by 1 unit, economic growth will increase by 0.13. Lagged value of GDP per capita is negative and significant at 1%
level, which proves the convergence hypothesis: countries with higher GDP per capita tend to develop at a lower rate than countries with lower GDP per capita. Turn to teledensity variables. Mobile penetration is significant and positive, which says that with more mobile phone subscribers, economic growth tends to be higher. However, coefficient associated with fixed- line penetration rate is negative, which is a surprise for us. This puzzle is left here for the moment, and we will discuss it in the upcoming regressions.
Similar to Datta and Agarwal (2004), square term of penetration rates are added in regression (3) in order to test the non- linear relationship between teledensity variables and economic growth. As can be seen, parameters for constant term, total capital stock and lagged value GDP per capita change slightly, so we do not discuss them again.
The coefficient of the fixed- line penetration rate is still negative (-0.0038) and significant at 1% level. A negative coefficient in this case does not imply that more fixed- line penetration decreases economic growth. Together with a positive and significant coefficient of square term fixed- line penetration rate, it implies a convex relationship between fixed- line penetration and economic growth, and the minimum point is reached when fixed-line penetration rate is positive. This minimum point of fixed- line telecommunications investment is estimated at 47.74%
(=0.0038/(2*0.0000398)) penetration rate. This result implies that if a country with a fixed- line penetration rate lower than 47.74% may be better not to invest anymore.
Due to leapfrog effect discussed previously, it is not difficult to understand. Countries with a low fixed- line penetration rate could just jump into the latest technology and invest in mobile phones. Conversely, the relationship between mobile penetration and economic growth is concave and there is a maximum point for mobile phone investment since the coefficient associated with square term of mobile penetration is negative (-1.37E-06). Different from Datta and Agarwal (2004) who estimate a maximum point at around 80% penetration rates, we conclude that the maximum point is reached when mobile penetration is equal to 145.99% (=0.0004/(2*1.37E-06)).
However, this result is not significant even at 10% level and the coefficient of square mobile penetration rate is not significantly different from zero.
According to the third version of our model (equation (7)), we have run regression (4) to distinguish the growth effect for countries with different mobile phone penetration levels. As is shown, there is still not much change for constant term, capital stock and lagged GDP per capita. For mobile phone penetration rates, they are displayed separately by low (below 40%) and high penetration rates (over or equal to 40%). In detail, high mobile penetration countries have an average growth effect of 0.00029, or
17
when there is a 1% increase in penetration rate, economic growth will increase by 0.00029. For low mobile penetration countries, the growth effect is slightly larger but less significant (significant at 10% level). For fixed-line penetration rate, they are displayed by low and high as well with the same range. Both of the high and low fixed- line penetration rates have negative coefficients, and again this may be due to the presence of non-linear relationship but not a real negative growth effect.
To sum up, no critical mass effect has been found for either fixed-line telecommunications infrastructure or mobile telecommunication infrastructure.
According to our results, fixed- line telecommunications infrastructure has a convex relationship with economic growth. For mobile telecommunications infrastructure and economic growth, the relationship is linear. Therefore, hypothesis 3 is rejected for mobile telecommunications but accepted for fixed-line telecommunications.
In order to compare the growth effect or returns of mobile telecommunication and fixed- line telecommunication infrastructure investment, we conduct a new regression combining the significant results in regression (3) and (4) of table 5. That is, the first power of mobile penetration, the first power of fixed- line penetration and square term of fixed-line penetration are included in the new regression. In addition, the interaction term of mobile and fixed- line penetration rate is included to test for the degree of substitutability. The results are shown in the following table.
18
Table 6: Regression results: dependent variable g
(5) (6) (7)
Developed countries
Less developed countries
Constant 0.076*** 0.233*** 0.080***
(9.755) (7.962) (10.480)
t 0.003*** 0.003*** 0.003***
(10.236) (4.918) (9.192)
k 0.130*** 0.586*** 0.119***
(7.643) (8.546) (6.537)
Mobpen(-1) 0.0004* 0.0005** 0.0007**
(1.676) (2.004) (2.058)
Pen(-1) -0.0039*** -0.0049*** -0.0038***
(-6.600) (-5.207) (-4.309)
Pen(-1)^2 4.20E-05*** 4.09E-05*** 7.86E-05***
(5.018) (4.975) (4.355)
GDPC(-1) -9.18E-06*** -5.29E-06*** -3.28E-05***
(-7.642) (-8.277) (-10.039)
Mobpen(-1)*Pen(-1) -2.45E-06 -8.22E-06** -5.82E-06
(-0.498) (-2.092) (-0.580)
Country fixed effect Yes Yes Yes
Time trend Yes Yes Yes
R-square 0.212 0.347 0.231
Obs. 2959 506 2446
Prob. 0.000 0.000 0.000
*Significant at 10% level.
**Significant at 5% level.
***Significant at 1% level.
Note: t-statistics are shown in the brackets.
Results of regression (5) in table 6 basically meet all the expectations we have made until now. One unit increase in the growth of per capita total capital increases economic growth by 0.13. And a minus sign of lagged value GDP per capita proves the convergence hypothesis. Importantly, mobile telecommunications infrastructure investment has a linear relationship with economic growth, that is, 1% increase in mobile penetration rate increases economic growth by 0.0004. However, this estimate is less significant (10% level). Moreover, a convex relationship between fixed-line telecommunications infrastructure investment and economic growth is found as well.
The minimum point is reached when fixed-line penetration rate equals to 46.43%
(=0.0039/(2*0.000042)). Further, refer to the method applied by Lee, Levendis and Gutierrez (2009), an interaction term between mobile phones and fixed-line phones is included. Although the coefficient associated with this interaction term is not significant, the negative sign still weakly suggest that mobile phones and fixed-line phones are imperfect substitutes. And if we exclude the interaction term from
19
regression (5), estimates of other variables become more significant especially for mobile phone penetration rate (increase from 10% level to 1% level). This again implies that the substitution effect is not that strong.
According to regression (5), mobile telecommunications infrastructure and economic growth have a linear relationship, while fixed- line telecommunications infrastructure and economic growth have a quadratic relationship. The growth effect can be approximately compared. Assume all the explanatory variables, except mobile penetration and fixed- line penetration, are zero. Further assume mobile penetration and fixed- line penetration do not influence each other, that is, when estimating the curve of mobile phones, we assume that fixed-line phones take the value zero, and vice versa. Then we can compare the growth effect through the following figure.
Figure 2: Relationship for penetration rates and economic growth
The lower curve stands for the function of fixed- line telecommunications infrastructure and the higher curve is the function of mobile telecommunications infrastructure. As is shown in figure 2, along the range of fixed- line penetration rate (from 0 to 90%), the growth effect of fixed- line telephones is always lower than mobile phones. However, this comparison is rather an approximate. It is difficult to compare the growth effect when both of mobile and fixed- line penetrations take non-zero values.
In the final two regressions, we make a distinction between more developed countries (GDP per capita great than 13000 in constant 2000US$) and less developed countries (GDP per capita less than 13000 in constant 2000US$). Compare the results in regression (6) and (7), we find that overall capital stock has larger growth effect in developed countries than in less developed countries, which is reasonable. The relationships between the two teledensity variables and economic growth are the same as we discussed before, indicating that previous results we got are robust. Moreover, we find a substitution effect in both developed and less developed countries. The
