Master Thesis IE&B
“Why do some countries invest more in growth enhancing assets
than others?”
Student: Rimma Velikanova Student number: s3242900
Email: r.velikanova@student.rug.nl
Supervisor: Prof. Robert Inklaar Co-assessor: dr. Abdul Erumban
19 June 2018
Abstract
This paper estimates the model of investment in heterogeneous types of capital in order to make inferences on why some countries invest more in growth enhancing assets than others. Our research makes two contributions. First, we find that decisions of high income countries to invest in certain assets depend on that specific country’s characteristics. This show a pattern of complementarity between equipment types and other inputs whose abundance vary across countries. For example, human capital, services, government and industrial share of GDP, income per capita and financial development. We also find that complementarity between human capital and intangibles assets is subject to change.
Table of Content
1. Introduction 2
2. Literature review 4
2.1 Cross-country income differences 4
2.2 Composition of capital stock 6
2.3 Hypotheses 7
3. Model of heterogeneous types of capital 9
3.1 Theory 9
3.2 Empirical specification 12
3.3 Data preparation 12
4. Data and measures 13
4.1 Basic data 13
4.2 Capital composition measure 15
5. Regression analysis 16
5.1 Caselli and Wilson model generalisation 16
5.2 Estimation results type-by-type 17
1. Introduction
In today’s world economy, large cross-country income differences exist. Development accounting is one of the ways used to explain income differences, by looking at the sources of growth, where most emphasis is put on the total amount of investments. However, it was found that much of the cross-country income differences cannot be accounted for by total human and physical capital, which provides room for the ignorance on growth and development (Knigth, Loayza, & Villanueva, 1993; Islam, 1995; Caselli, Esquivel, & Lefort, 1996; Timmer, 1999). Looking for other reasons, it was observed that low-income countries save at similar rates as high income countries and that the marginal product of capital is similar (Hsieh & Klenow, 2007). Yet we observe that high income countries invest more in certain types of capital; partly in machinery and transport equipment, but especially in intangible assets, as shown in Figures 1-3 . The finding of intangible 1
assets is relatively new as compared to Caselli and Wilson (2004), and is related to the paper by Chen (2017). This finding is enabled by the new and comprehensive data from the Penn World Table (PWT) 9.0 and thus warrants a new look at the question why high-income countries invest
Figures report relationships between composition index and GDP per capita, composition index is explained in chapter 4.1.
1
Figure 2. Trans.eq. composition index versus income per capita
more in some assets than others and whether this is important for understanding cross-country income differences.
Previous studies have been limited by the appropriateness of estimation methods and data availability. Following the research by Caselli and Wilson (2004), we aim to cover measurement shortcomings by using data which “adds up investment to gross fixed capital formation in National Accounts” correcting biases for countries that are capital producers and for the perpetual inventory method (Feenstra, Inklaar, & Timmer, 2015). Moreover, the goal of the paper is to extend the existed analysis in terms of countries and time on factors of production by focusing on machinery, transport equipment, and intangible assets. Findings of this paper will contribute to understanding the mechanisms via which country characteristics matter for cross-country economic growth and development and contribute to the further explanation of the TFP productivity residual. Moreover, these findings contribute to the framework of Caselli (2005) who argues that the composition of capital is a key area of future research for developing accounting.
To address this question we follow the model of investment in heterogeneous types of capital where the composition of capital depends on the relative efficiency of each type of capital and capital ratios. We employ a sample of 118 economies over the period 1970-2014 combining both the PWT9.0 database and the World Bank development indicators. Implementing insights, we find that our findings do not resemble those of Caselli and Wilson (2004), primarily due to two reasons. First, due to differences in categorisation, and, second, due to the differences in capital estimates measurements. Running the main regression, we find that there is indeed complementarity between cross-country factors and assets. Services, industrial, and government share of GDP, human capital, income per capita and financial development are found to be significant factors. We found that complementarity can change over time, as human capital is exclusively complementary to machinery and intangible assets in the period 1995-2014. Further, the share of industry and services of GDP is complementary to transport equipment in both periods. Other factors such as the share of government consumption of GDP have lost its value for machinery and intangible assets, however, the services share and industrial share of GDP are found to be complementary to intangible assets in the period 1970-2014. Those results are also found to be robust by introducing additional variables.
More precisely, this implies that countries still have to invest in human capital in order to grow as it gains value over time for machinery and intangible assets, despite arguments over universal education levels. Nevertheless, these results also inline with the bigger picture on capital decomposition and income differences as taking intangible assets into the account, increases the cross-country income variation accountability.
The outline of the paper is as followed. Section 2 reviews existing theoretical and empirical evidence on income differences and capital composition. Model and empirical specification is discussed in Section 3. Section 4 specifies the data and measurement. Results and robustness checks are provided in Section 5. Section 6 concludes and discusses implications and limitations.
2. Literature review
2.1 Cross-country income differences
Large differences exist in cross country income per capita. The ratio of the 90th to the 10th percentile in the world income distribution is over 20 according to World Bank (2015) and around 28 according to Heston, Summers, and Aten (2009). One view to explain income differences, sources of growth and development is the accumulation view, which believes that economic growth is simply a mechanical process of development accounting where investments into physical capital is sufficient to account for the unprecedented growth conditional of productivity and that entrepreneurship, innovation and learning are almost automatic consequences of the investments (Krugman, 1994; Nelson & Pack, 1999). Proponents of the standard view explain growth across countries via development accounting. Development accounting answers the question outlined in Caselli (2005) — “how much of the cross-country income variance can be attributed to differences in factors of production (physical and human) and how much to differences in the efficiency with which capital is used?”. Many authors attribute differences among country incomes to differences in human capital, physical capital and productivity. For instance, Mankiw, Romer, and Weil (1992) examine the Solow growth model and Hall and Jones (1999) use the Cobb Douglas equation (1) as a benchmark for the development accounting method:
!
This production function provides the means of allocating changes in a country’s observed output Yi
into the contributions of changes in factor inputs — capital Ki and labor Li — and a residual, total
factor productivity (TFP) Ai, which measures a combination of changes in efficiency in the use of
those inputs and changes in technology. The Solow model assumes that as long as it is invested in the country, it will grow up to its steady state, therefore all countries are tempted to grow. However, this model doesn’t work perfectly as a large fraction of the cross-country differences in income and productivity variance cannot be explained (Knigth et al., 1993; Islam, 1995; Caselli et al., 1996), as it acts as a residual, therefore providing room for the ignorance on growth and development (Timmer, 1999).
It is believed that standard capital accumulation would be essential only after the war or catastrophe, for example, Dunke (1990) explains the “reconstruction hypothesis” that after the WWII destructions, based on the standard neoclassical growth model, capital accumulation via investments and technology transfers in Western Europe countries led to a rapid recovery and to the pre-war economic level. Comin and Hobijn (2011) find that Western countries and Japan in the period 1950-1970 not only returned to the pre-war levels but also surpassed it through development accounting. Looking at the more recent development cases, we observe that standard capital accumulation is not imperative. For instance, Nelson and Pack (1999) found that the “Asian Miracle” originates not only because countries were able to import modern technology but also because countries were able to assimilate and absorb it, underlining the importance of learning and entrepreneurial efforts. Moreover, Chen and Inklaar (2016) argue that organization capital may promote knowledge spillovers, which will increase productivity and consequently increase income and decrease cross-country income differences. These findings lead to a conclusion that standard model for economic growth is not able to thoroughly explain existing income differences and there is more to the story.
Recent findings by Comin and Mestrieri (2014) find that technology diffusion accounts for 80 percent of the income difference between rich and poor countries. These findings indicate that the capital composition effect can better explain variation in income, compared to aggregate capital investments that believed to explain around 20 percent of change in income (Klenow and Rodríguez-Clare, 1997; Hall and Jones, 1999; Caselli, 2005). Therefore we imply that one way to account for higher income differences is to decompose physical capital and look at specific types, rather than the aggregate levels.
2.2 Composition of capital stock
One way to revisit the basic model whilst looking at the differences in income is to reduce the measure of productivity residual by looking at the composition of capital stock and investments functional forms instead of total quantity. The System of National Accounts defines capital investments as an acquisition of physical assets that are expected to enhance long-term productivity and production possibilities. That includes machinery, equipment, structures and intangible assets which cover software and other intellectual property products. Literature mostly treats physical capital as a homogeneous good. International dollar values of all types of capital goods are summed up to arrive at the aggregate capital stock of a country, with prices acting as weights. Even if the relative values of capital are captured by the prices, their relative compositions to output and contribution to capital composition are not accounted for by the prices. Subsequently, various types of capital are assumed to be perfect substitutes of one another, which is not the case. For that matter Sato (1967) estimated the elasticity of substitution between equipment and structures and found that two kinds of capital are not perfect substitutes. Data on direct measures of the quantities of equipment installed in a country by type is limited, as little is explored about the composition of physical capital across countries. Since it is challenging to disaggregate the available data, it is common to apply a proxy for the type of capital.
for capital investment. This idea is also incorporated in Alfaro and Hammel (2007). The disadvantage of this method is that it expels data on countries that produce their own equipment and do not import it, creating a biased data. Leblebicioglu and Madariaga (2015) correct for that problem by using the perpetual inventory method. Mutreja (2014) also employs the perpetual inventory method, however, focuses on mainly equipment and structures. However, it is believed that by using this method, findings will moderately diverge from actual inventory levels (Berlemann & Wesselhoft, 2016). These exist several ways to determine capital composition, however none of the listed options is optimal in terms of estimation. This implies that there is a gap in the literature, which can be filled by applying PWT 9.0 data which “adds up investments to gross fixed capital formation in National Accounts”, correcting biases for countries that are capital producers and for the perpetual inventory method (Feenstra et al., 2015).
2.3 Hypotheses
The focus of the literature on the capital composition levels as a determinant of income and productivity differences implies that it is substantial where and in what types of technology investment takes place. Subsequently, we raise a question why some countries are investing more in the technologies that generate more rapid growth than other countries, for example in intangible assets, as Figure 1 suggests. This question is related to Chen (2017) and is new to Caselli and Wilson (2004). The last focused on capital decomposition (excluding intangible assets) and found that countries are more likely to invest in some assets than in others based on the amount of various factors, such as the industry share of GDP, services share of GDP, financial development, real income and human capital.
Following Caselli and Wilson (2004) we hypothesize that differences in investment composition and capital share in the country depends on the degree to which capital is complementary with other inputs whose abundance vary across countries. We imply that there is a need for the new research in order to validate Caselli and Wilson (2004) findings, as data for the newer time period is available. We also argue that Caselli and Wilson findings should be verified for the other reasons.
focus on 40 mostly rich countries that are leaders in R&D and technology adoption, which can indicate that with the increase of the sample, their findings would deteriorate . 2
Second, Caselli and Wilson findings include period from 1970 to 1995, where in the last years significant historical moments changed the economical perspective in a slow pace over time , 3
what could have affected the societal and economical inputs composition. Moreover, it is also observed that composition of assets changed over time, and strikingly composition of intangible assets increased from 1990 (see Appendix: Figures 1-4). Intangible assets are the most R&D intensive that are complementary with human capital (OECD Science, 2017; Teece, 1989). Hence, a reason must exist for the increase in capital composition, for example, the introduction of cell phones and internet increased human capital as more people got access to education, which have increased the possibility of penetration of R&D intensive intangible assets. This would imply that level of complementarity between R&D intensive intangible assets and human capital has increased over time. In this case we would expect that the coefficient of relationships would become significant for the second estimated period. On the other hand, we could also argue that human 4
capital becomes less complementary with assets as the total level of education world-wide has become more or less equalised due to introduction of internet, which enables virtual and online education . In this case we would observe that significance of coefficient for human capital for the 5
second period would diminish. Consequently we propose that a new paper has to elaborate on the later period starting from 1995, to find whether there is a change of complementarity between human capital and intangible assets.
Third, Caselli and Wilson (2004) have not found robust complementarity between assets and human capital for the period 1970-1995. We believe that it is due to the fact that their findings do not include intangible assets, which became increasingly important in the recent decade. To sum up, we believe that there is a need for an enlightened paper which will have a look at a more recent period and takes the preceding problems into the account. Based on the above-mentioned this paper hypotheses are:
H1.1: Increasing the amount of countries and extending the time period, Caselli and Wilson (2004) findings will not be generalised.
However it is important to mention that we were not able to find information that the relationships Caselli and Wilson (2004) find
2
are different across different types of countries and over time.
For example, economic liberalisation since 1980, introduction of cell phones and internet in the end of 1990s, introduction of single
3
market in 1993, Asian region growth, China’s WTO membership in 2001, rise of multinational corporations, stagnation of the world economy, economic crisis in 2009 and recovery in the following years.
It is important to note that if complementarities change over time it does not indicate causality, rather correlation.
4
In 2011 primary school completion rates around the world have reached 91 percent (World Bank, 2014).
H1.2: Difference in capital share depends on the degree to which capital is complementary with other social and economic factors.
H1.3: Complementarity of human capital and intangible assets is subject to change over time. Testing the mentioned above mentioned hypotheses, the objective of this paper is to find why some countries are investing more in technologies that generate more rapid growth than other countries. To address this we follow the model of investment in heterogeneous types of capital, where the capital type share in the total stock depends on its’ internal efficiency and degree of complementarity with other cross-country inputs. This paper will look at the sample of 118 countries in the period starting from 1970 up to 2014 and thus will avoid the narrow choice of countries and time periods used in the previous studies. This paper will use PWT 9.0 data which “adds up investments to gross fixed capital formation in National Accounts”, correcting biases for countries that are capital producers and for the perpetual inventory method, while additional covering investment in structures, what was not done before (Feenstra et al., 2015). Findings in this paper will contribute to understanding the mechanisms via which countries characteristics matter for cross-country economic growth and development and contribute to the explanation of the TFP productivity residual. Moreover, these findings contribute to the framework of Caselli (2005) who argue that composition of capital is a key area of future research for developing accounting. Previous studies have been limited by the appropriateness of estimation methods and data availability.
3. Model of heterogeneous types of capital
3.1 Theory
The starting point of our analysis will be an outline of the model by Caselli and Wilson (2004) on heterogeneous types of capital who use a CES production function as an introduction where in country i aggregate output ! is produced using intermediate inputs ! with this technology:
! ! ,
where ! is a term for a country specific total factor productivity and coefficient ! indicate the degree of (imperfect) substitutability between the inputs. Based on this equation, intermediate good ! is produced combining specific type of capital ! and labor:
! , ! ,
where ! stands for labor and capital which were used to produce the intermediate-input ! , whereas ! is the productivity of specific country ! and type ! capital. We assume that capital is heterogeneous with P different types of physical assets and each type is product specific, meaning that intermediate good p can only be produced by capital of type ! . Namely, intermediate is determined by the type of equipment used in its production, where assumption ! indicates that equipments are imperfect substitutes. For simplicity of the model, we assume that labor is 6
homogeneous within a country and heterogeneous between countries. The production function of intermediate (2) allows for capital-labor substitutability and promotes the link between our model and the theory on development-accounting. As outlined, productivity is both ! country ! and product ! specific. Prior to interpretation, it is important to explain that ! is measured as the US dollar value of the capital stock of type ! . Where equipment of type ! can be complementary with the specifics of country ! , consequently different types of capital are more applicable for different countries. As ! varies for product ! , which means that investments in different products may bring different amounts of productivity. This implies that the level of efficiency embodied in capital type ! may be higher with the R&D production intensity of that type ! .
We assume perfect labor mobility across the intermediate sectors ! , where in the
equilibrium model there will be aggregate constant returns to scale in ! and ! . The optimal choices for capital and labor for intermediate goods P require that marginal products of different capital types and labor are equalised across sectors. Assuming optimal choices of capital, labor and labor mobility, we obtain:
!
where ! is the value of capital stock. Equation (3) is derived from the model’s
equilibrium where the marginal products of different capital types are equalised based on the Caselli and Wilson (2004) method. It implies that investment will concentrate in capital types that have a xi p= Api(Lpi)1−α(Kpi)α 0 < α < 1 Ki p(Lpi) xp Ai p i p p γ < 1 Ai p i p Ki p p p i Ai p p p p Li= ∑ p Li p Ki Li Ki p Ki = (Ai p)γ/(1−γ) ∑j(Ai j)γ/(1−γ) Ki= ∑ p Ki p
Literature does not provide a specific value for elasticity of substitution between heterogeneous physical capital. Therefore, we
6
consider estimates for the elasticity of substation between varieties of inputs. We refer to Caselli and Wilson (2004) and Leblebicioglu and Madariaga (2015) and establish that elasticity of substitution between capital types is 1 with γ=0.5.
(2)
higher efficiency. Adopting equation (3) we can demonstrate that the optimal capital ratio of different types of capital will be induced by their relative efficiencies:
!
To show how aggregate output is determined by composition of capital we use Equation 3 in the aggregate output equation:
!
where ! is the index for capital investments, defined by:
!
where composition of capital depends on the relative efficiency of each type of capital. Therefore higher ! indicates a higher productivity, subsequently resulting in a higher level of the aggregate output. However, as productivity by capital type is not detectable we construct a productivity proxy by substituting the efficiency of capital types p=1,…, P from Equation (4) to Equation (6) and derive the following results:
!
The Equation (7) expresses capital composition in terms of capital 1 productivity and capital ratios (relatively to capital type 1). We choose structures equipment as a numerate since it is the least R&D intensive product according to the OECD calculations of the R&D intensity by industry. On average, structures expenditure as a percent of gross value added in log is less than 1%, compared to machinery and transport equipment, which are considerably larger (OECD Science, 2017). This argumentation follows the Caselli and Wilson (2004) argumentation. Therefore, ! can be assumed to be constant across countries and therefore will not lead to significant variation in the capital composition index in Equation (7).
This paper uses capital stocks values for the capital rations ! in the capital composition Equation (7), while Caselli and Wilson (2004) could not accurately measure the value of the stocks by type and country, therefore converting import flows into stocks.
3.2 Empirical specification
Further, the unified empirical framework for the composition of capital takes the following panel regression specification:
!
where ! , where ! equals composition index ! , ! is a set of explanatory variables whose values can vary across time, ! stands for independent variables which are time-invariant, ! is an intercept term which varies over time, ηi is an unobserved and time-invariant country effect, τt is time-effects that capture global shocks. ! and ! are the error terms, where ! is time-variant and ! varies across countries but not across time. ! also represents the effects of all stable variables that are not included in the model. In order to avert the loss of zero observations and in order to contract values for the convenience of regression interpretation, we take logs on both sides and estimate log-log model (Hill, Griffiths, & Lim, 2012). As our data have number of countries in a specific period of time, we refer to data as a panel data or cross-sectional time series data.
To estimate a model, we choose for both fixed effects and random effects estimation techniques. Using a fixed effects model enables us to control for the time-invariant variables that have not been measured but that affect dependent variables. A fixed effects model allows to capture individual heterogeneity by the intercept and assume that slope coefficients are constant for all types of equipment in the model. On the other hand, we use random effects model if we believe that there is no omitted variables or ! is not correlated with the ! , because omitted variables are not correlated with the variables in the model. We restrain to apply OLS estimates as it ignores heterogeneity across time or countries, where standard errors will be too small overstating the reliability of the last squares estimates. We also restrain to use a pseudo-maximum likelihood approach that is applied by Caselli and Wilson (2004), as in the large sample with several missing estimates (which is the case for our sample) the convergence of the calculations can be false because poisson command is sensitive to numerical computation (Silva & Tenreyro, 2010).
3.3 Data preparation
By running a fixed effects model we are able to reject null hypothesis and conclude that the intercept parameter is not identical for at least one country in the sample. Therefore, heterogeneity exists across countries and a pooled model is not suitable. We also use the Breusch and Pagan
Lagrangian multiplier test. As the p-value is lower than the significance interval, we rejected the null hypothesis and conclude that estimates of the models are different, therefore a pooled model should not be used and random and fixed effects models are the most applicable. Using the Wald heteroskedasticity test for fixed effect models, we reject the null hypothesis and conclude that heteroskedasticity exists in our dataset. Moreover as our panel data has a long time series (more than 30 years) we have to test for serial correlation. We use Wooldridge test for serial correlation where we reject the null hypothesis and conclude that autocorrelation exists. In order to correct for both heteroskedasticity and autocorrelation we will use robust standard errors. In order to choose between random and fixed effects we apply the Hausman test where we reject the null hypothesis. Therefore we are obliged to use the fixed effects model, because the random effects estimator is not consistent. The above mentioned tests, definitions, results and interpretation can be found in the Appendix, Table 4.
4. Data and measures
4.1 Basic data
This paper is based on the balanced panel data set that includes 118 countries over a period 7
of forty-four years, from 1970 to 2014. This sample includes a mix of advanced and developing countries in significant numbers. This diversity overcomes the sample selection bias that may arise when focusing on a sample of only developed countries (DeLong, 1988). This large data set will enable to estimate growth experiences across two time periods: 1970-1995 and 1995-2014. We restrict our panel to the two time periods, because country characteristics can change relatively slow over time. Moreover, by applying these periods we eliminate spikes that yearly data can generate and avoid any yearly gaps that can be present. The time periods are based on the availability of data, and historical and political occasions. For instance, in 1970 the Great Divergence period started, and starting from 1990 trade liberalisation and digital revolutions took place. Commending these events we expect to observe that the determinants of investments in capital embodied technology will change. The choice of the sample is based on the availability of data for capital composition measure.
Please refer to Table 6 in the Appendix for the list of countries.
To test for the first hypothesis we reproduce Caselli and Wilson findings, including all mentioned variables, however, for the second and third hypotheses we drop FDI variables, as those are scarcely informative and deficient in terms of data availability.
Data on country characteristics is obtained from several sources. Human capital is measured as an index and obtained via PWT database 9.0. Financial development measure, measured as a private credit by deposit money banks and other financial institution as a % of GDP, openness to trade, which is measured as a % of GDP, and real GDP per capita in millions US$ are all obtained from World Bank Development Indicators. For the measures of FDI and sectoral and governmental shares, the World Bank development indicators database was used. Table 1 in the Appendix provides an overview of all data measurements and data sources. Please refer to Table 2 and Table 3 in the Appendix for summary statistics of independent and dependent variables respectively.
4.2 Capital composition measure
The capital composition index is constructed using Equation (7). To calculate the index we indicate that ! , therefore there is a perfect elasticity of substitution ! . For the robustness check we set up a higher elasticity of substitution where ! . The PWT database collects data for four categories of capital goods ! , see Table 1 for the details. The capital stock at a constant national price for type p asset is determined through equation:
!
where ! is an investment in country i for asset type p during year t and ! is an asset type specific depreciation rate.
To construct capital stock at the constant national prices for each equipment type we turn to the equation on current-cost net capital stock:
!
where current-cost net capital stock ! and capital stock deflator ! are given in PWT data. Therefore we are able to estimate capital stock at constant national prices for type of asset:
The final step is to convert national prices for type of asset by using market estimated exchange rates from PWT data: . 8
5. Regression Analysis
5.1 Caselli and Wilson model generalisation
To test our first hypotheses, we are interested to simulate Caselli and Wilson (2004) findings. Findings are reported in Table 7 in the Appendix, using an estimation of a fixed effects model. It is important to note that Caselli and Wilson data on the capital composition is more defined. Authors use a commodity-flow method to collect data and categorise nine capital types based on the ISIC categorisation, while we estimate four capital types based on OECD National Accounts, country Nationals Accounts, EU KLEMS, ECLAC National Accounts, and a
commodity-Kpt/xr Table 1
Categories of Capital Goods
Capital type Lable Description Depreciation rate
Structures K_Struc Capital stock of residential and non-residential structures
Residential structures 1.1% Non-Residential structures 3.1%
Machinery K_Mach Capital stock of machinery including computers, communication equipment and other machinery
Computers 31.5% Communication equipment 11.5% Other machinery 12.6% Transport equipment
K_TraEq Capital stock of transport equipment Transport equipment 18.9% Intangible
assets
K_Intang Capital stock of other assets including software and other intellectual property products
Software 31.5%
Other intellectual property products 15%
Cultivated assets 12.6%
Note: This table contains information over that capital types used in the regression, labelling composition and corresponding depreciation rate used in data collection.
Applying this capital measure, we are able to obtain capital composition through the time (Please refer to the Figures 1-4 in the
8
flow method (Feenstra et al., 2015). Table 5 in the Appendix presents ISIC, EU KLEMS and PWT database corresponding codes. Caselli and Wilson (2004) report separately computers, electricity equipment, non-electrical machinery and communication equipment, whereas we report converged machinery assets. Moreover, Caselli and Wilson estimate separately other transport, motor vehicles and aircraft, whereas we estimate a data for overall transport. The authors estimate professional goods physical capital, which we do not cover due to data availability. On the other hand, we estimate structures and intellectual property rights, which are not reported by Caselli and Wilson.
Caselli and Wilson found that cross-country differences in human capital, institutions, and composition of GDP have explanatory power for the differences in the composition of capital. More specifically, outward FDI imply higher imports of computers and aircraft. The industry share of GDP predicts more non-electrical assets. Services share of GDP is associated with more investments in motor vehicles. Income per capita and human capital have positive relationships with computers and electrical equipment, however these relationships are not robust. Comparing Table 7 to Caselli and Wilson findings, we find that our model of fit is higher than that of the authors for all three columns. We find that the estimators have the same sign and significance for the income per capita in the case of machinery and for the time trend and services share in the case of transport equipment. Therefore we can conclude that our estimations cannot resemble the model by Caselli and Wilson, therefore we are not able to validate the first hypothesis. We believe that the difference in estimators exist due a number of reasons. First, PWT 9.0 data set of capital is not detailed enough comparing to that of Caselli and Wilson. Second, data collection methods differ, as the Caselli and Wilson paper focuses on imports, whilst the PWT data set focuses on both initial stocks and further investments, which corrects for the fact that some countries have not imported, as they already had high levels of technology or produced it domestically. Third, we use structures as numeraire while authors use fabricated metal products. Moreover, we include intangible assets, which Caselli and Wilson do not review. These points suggest that the method used in this paper applies different data composition measures, which implies divergent outcomes.
5.2 Estimation results type-by-type
relative results. Panel B and C contains regression results for transport and intangible assets respectively. All panels incorporate the same number of variables, such as time trend, the government share of GDP, industry share of GDP and services share of GDP, where agriculture share of GDP act as an omitted variable. Panels also include control variables such as GDP per capita and human capital. We drop out the constant term as it does not provide compelling information in the fixed effects model estimations. It is observable that R-square values are lower than 60%, which is explained by the fact that cross sectional data implies cross sections heterogeneity, comparing to time-series data. Overall, the model reports high significance levels for indicators alone which is an evidence that the model has a good fit.
Intangible assets composition indexes show positive time trends for both periods, displaying that their intrinsic efficiencies have increased, or identically, that the worldwide quantities have risen of those characteristics complementary to intangible capital types. This is contrary to both machinery and transport equipment in the two periods.
Looking at the independent variables, the government consumption share of GDP is positively related to capital investments in every type of capital and for both periods relatively to structures. The industry share of GDP holds similar results, as it predicts relatively more machinery and transport equipment capital. The exception is the period from 1970 to 1995 for the intangible assets, where industrial share is insignificant. Services’ share of GDP is associated with more investment in machinery in the first period, more investment in intangible assets in the second period and more investments in transport equipment for both periods. According to the baseline specification, human capital is complementary with intangible assets in the second period and remarkable with machinery. Human capital is not complementary for transport equipment. Real income per capita has a positive relationships for most of the observations, except for intangible assets in the first period.
5.3. Robustness check
6. Conclusions
Why do some countries invest in more growth enhancing physical assets than others? We answer this question by accounting for the role of country specific factors with regards to physical capital composition. Looking at the 118 countries for the two time periods 1970-1995 and 1995-2014, we used the capital composition index to test its complementarity with social and economic factors, such as the government share of GDP, industry share of GDP, services share of GDP, human capital and income per capita.
6.1 Implications
These findings imply that there are a number of country-specific characteristics, which have a statistically significant effect on the composition of capital. Particularly, these country-specific economic and social factors affects how much and in which kind of assets country invests. We also find that complementarity of those factors and capital varies over time, therefore countries have to adapt to the changing environment. Specifically, this implies that countries still have to invest in the human capital in order to grow as it gains value through the time for machinery and intangible assets, despite arguments over universal education levels. Looking at the broader picture, the important question is whether these findings matter for explaining the income differences across countries. Caselli and Wilson (2004) find that capital decomposition could account for 10 percent of the overall cross-country difference variance. Chen (2017) find that after taking intangible assets into the account, cross-country income variation accounted for 40 percent. Even though, we do not estimate relationship between capital quality measures and income, due to the time limitation, we could still assume that including intangible assets in regression analysis would increase the overall contribution of capital to cross-country income differences. However, more research is needed to account for this statement.
6.2 Limitations
Reference List
Alfaro, L., & Hammel, E. 2007. Capital flows and capital goods. Journal of International Economics, 72 (1): 128–50.
Berlemann, M., & Wesselhoft, J. 2016. Estimating aggregate capital stocks using the perpetual inventory method. Review of Economics, 65(1): 1-34.
Caselli, F. 2005. Accounting for cross-country income differences. In P. Aghion & S. Durlauf (Eds.), Handbook of Economic Growth, Elsevier: 679–741.
Caselli, F., & Coleman, W. J. 2001. Cross-country technology diffusion: The case of computers. American Economic Review – AEA Papers and Proceedings, 91: 328-335.
Caselli, F., Esquivel, G., & Lefort, F. 1996. Reopening the Convergence Debate: A New Look at Cross-Country Growth Empirics. Journal of Economic Growth, 1(3): 363-89.
Caselli, F., & Wilson, D. J. 2004. Importing technology. Journal of Monetary Economics, 51: 1–32. Chen, W. 2017. Cross-country income differences revisited: Accounting for the role of intangible
capital. Review of Income and Wealth.
Chen, W., & Inklaar, R. C. 2016. Productivity spillovers of organization capital. Journal of Productivity Analysis, 45: 229–45.
Comin, D. A., & Hobijn, B. 2010. An exploration of technology diffusion. American Economic Review, 100: 2031-2059.
Comin, D., & Hobijn, B. 2004. Cross-Country technology adoption: Making the theories face the facts. Journal of Monetary Economics, 51: 38-83.
Comin, D., & Hobijn, B. 2011. Technology diffusion and postwar growth. NBER Macroeconomics Annual, 25(1): 209-246.
Comin, D., Hobijn, B., & Rovito, E. 2006. Five facts you need to know about technology diffusion. Working paper series. no. 11928.
Comin, D., & Mestrieri, M. 2014. Technology diffusion: Measurement, causes, and consequences. Handbook of Economic Growth, 2: 565-622.
DeLong, B. 1988. Productivity growth, convergence, and welfare: Comment. American Economic Review, 78(5): 1138-1154.
Dougherty, C., & Jorgenson, D. W. 1996. International comparisons of the sources of economic growth. American Economic Association, 86 (2): 25-29.
Dunke, R. H. 1990. Reassessing the Wirtshaftswunder: Reconstruction and postwar growth in West Germany in an international context. Oxford Bulletin of Economics and Statistics 52:
451-92.
Eaton, J., & Kortum, S. 2001. Trade in capital goods. European Economic Review, 45(7): 1195-235. Erunbam, A. A., & de Jong, S. B. 2006. Cross-country differences in ICT adoption: A consequence
of Culture? Journal of World Business, 41(4): 302-314.
Feenstra, R., Inklaar, R., & Timmer, M. 2015. The next generation of the Penn World Table. American Economic Review, 105(10): 3150-3182.
Gregorio, D. D., Kassicieh, S. K., & Neto, R. D. 2005. Drivers of e-business activity in developed and emerging markets. IEEE Transactions on Engineering Management, 52(2): 155-166. Hall, R. E., & Jones C. I. 1999. Why do some countries produce so much more output per worker
than others? Quarterly Journal of Economics, 114(1): 83–116.
Hall, B. H., & Khan, B, 2003. Adoption of new technology. In D. C. Jones (Eds.), New Economy Handbook. Academic Press.
Heston, A., Summers, R., & Aten, B. 2009. Penn World Table version 6.3. Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania.
Hill, R. C., Griffiths, W. E., & Lim, C. G. 2012. Principles of Econometrics. John Wiley And Sons Ltd.
Hsieh, C., & Klenow, P. J. 2007. Relative prices and relative prosperity. American Economic Review, 97(3): 562-85.
Islam, N. 1995. Growth empirics: A panel data approach. The Quarterly Journal of Economics, 110(4): 1127-1170.
Jorgenson, D. W. 2001. Information technology and the U.S. economy. Americal Economic Review, 91(1): 1-32.
Kiiski, S., & Pohjola, M. 2002. Cross-country diffusion of the Internet. Information Economics and Policy, 14(2): 297-310.
Klenow, P. J., & Rodríguez-Clare, A. 1997. The Neoclassical revival in growth economics: Has it gone too far? In B. Bernanke & J. Rotemberg (Eds.), NBER Macroeconomics Annual, vol. 12: 73-103. Cambridge, MA: MIT Press.
Krugman, P. 1994. The myth of Asia’s miracle. Foreign Affairs, 73(6): 62-78.
Leblebicioglu, A., & Madariaga, J. 2015. Financial flows, composition of capital, and growth. IMF Economic Review, 63 (2): 325-352.
Mankiw, N. G., Romer, D., & Weil, D. N. 1992. A contribution to the empirics of economic growth. Quarterly Journal of Economics, 107(2): 407–37.
Mutreja, P. 2014. Equipment and structures capital: Accounting for income differences. Economic Inquiry, 52(2): 713-731.
Nelson, R. R., & Pack, H. 1999. The Asian miracle and modern growth theory. The Economic Journal, 109(457): 416-436.
OECD Science. Technology and Industry Scoreboard. R&D intensity by industry, 2015. Accessed online https://www.oecd-ilibrary.org/science-and-technology/oecd-science-technology-and- industry-scoreboard-2017/r-amp-d-intensity-by-industry-2015_sti_scoreboard-2017- graph122-en. Viewed April 5, 2018.
Sato, K. 1967. A two-level constant-elasticity-of-substitution production function. The Review of Economic Studies, 34(2): 201–218.
Savvides, A., & Stengos, T. 2009. Human Capital and Economic Growth. Stanford: Stanford University Press.
Silva, S., & Tenreyro, S. 2010. On the existence of the maximum likelihood estimates in Poisson regression. Economics Letters 107: 310–312.
Teece, D. J. 1989. Capturing value from knowledge assets: The new economy, markets for knowhow, and intangible assets. California management review, 40, (3).
Timmer, M. P. 1999. Climbing the technology ladder too fast? An international comparison of productivity in South and East-Asian manufacturing, 1963-1993. CCSO Working Papers 199919, University of Groningen, CCSO Centre for Economic Research.
World Bank., World Development Indicators, Washington D.C.: World Bank Publications, 2015. World Bank. 2014. World Bank support to education: A system to achieve learning for all. Accessed
Appendix: Tables and Figures
Table 1
Description of variables
Variable Measurement Lable Definition Source
Structures Constant national prices
K_Struc Capital stock of residential and non-residential structures
PWT 9.0
Machinery Constant national prices
K_Mach Capital stock of machinery and non-transport equipment PWT 9.0 Transport equipment Constant national prices
K_TraEq Capital stock of transport equipment
PWT 9.0 Intangible assets Constant national
prices
K_Intang Capital stock of other assets including software and other intellectual property products
PWT 9.0
Human capital Index hc Human capital index is
based on average years of schooling PWT 9.0 Real GDP per capita at chained PPPs in millions 2011 US$
rgdpo Real GDP per capita World Bank
Development Indicators Openness to trade % of GDP opent Trade is the sum of
exports and imports of goods and services measured as a share of gross domestic product
World Bank Development Indicators
Financial development
% of GDP findev Private credit by deposit money banks and other financial institutions to GDP (%)
World Bank Global Financial Development
Inward FDI % of GDP ifdi Foreign direct
investment, net inflows (% of GDP)
World Bank Development Indicators
Outward FDI % of GDP ofdi Foreign direct
investment, net outflows (% of GDP)
Industrial sector share
% of GDP indshare Industry, value added (% of GDP) World Bank Development Indicators Service sector share
% of GDP seshare Services, etc., value added (% of GDP)
World Bank Development Indicators Government share % of GDP govshare General government final
consumption expenditure (% of GDP)
World Bank Development Indicators
Note: Table presents description of independent variables, its measurement value, label used in the regression, definition and the source.
Table 2
Summary statistics for independent variables
1970 2014 N of countries Mean Std. deviation # of countries Mean Std. deviation Inward FDI 70 -0.2929 1.3118 118 0.9675 1.0425 Outward FDI 70 -1.3706 1.3141 118 -0.3995 1.6785 Government’s share of GDP 118 2.6020 0.3202 118 2.7381 0.3607 Industrial sector’s share of GDP 118 3.3262 0.4307 118 3.2731 0.4455 Service sector’s share of GDP 118 3.7133 0.3309 118 4.0571 0.2378 Human capital 118 0.4519 0.3205 118 0.9140 0.2921 Real GDP per capita 118 8.4626 1.1927 118 9.2959 1.1748 Openness to trade 118 3.6645 0.6778 118 4.3702 0.4710 Financial development 118 2.8460 0.7909 118 3.6890 0.8370
Table 3
Summary statistics for capital stocks Capital type
Structures Machinery Transport equipment Intangible assets 1970
Available N of countries 118 118 118 118
Capital stock share mean 14.28325 11.75422 11.07953 6.620938
Std. deviation 6.60128 6.652971 6.622939 6.698785
min 6.468924 4.505408 3.470598 -7.045984
max 42.77551 41.47597 40.18269 32.44485
2014
Available N of countries 118 118 118 118
Capital stock share mean 11.9676 10.04787 8.83976 8.150076
Std. deviation 2.098234 2.00833 2.006292 2.595517
min 7.618549 6.199897 3.466194 2.555268
max 16.98331 15.30783 14.57271 13.43577
Note: This table list summary statistics for the four capital types for two years. Column 1 lists number of countries, mean, standard deviation, minimum and maximum values for structures, whereas column 2, 3 and 4 lists details for machinery, transport equipment and intangible assets respectively.
Table 4
Tests and diagnostics
Test name Definition Outcome Result Solution
Breusch-Pagan Lagrange multiplier (LM)
The null hypothesis in the LM test is that variances across entities is zero. This is, no significant
difference across units (i.e. no panel effect)
chibar2(01) = 15115.51
Reject the null hypothesis and conclude that we can run random effects model and simple OLS
regression is not appropate. Prob > chibar2
Modified Wald test
A test for
heteroskedasticiy for the fixed- effects model. The null hypothesis is homoskedasticity. chi2 (133) = 4.0e+06 Prob>chi2 = 0.0000
Reject the null hypothesis and conclude heteroskedasticity. robust standard errors/ cluster
Wooldridge test A test for serial correlation. The null hypothesis is no serial correlation. F(1, 124) = 19.173 Prob > F = 0.0000
Reject the null hypothesis and conclude serial correlation.
robust standard errors/ cluster
Hausman test To decide between fixed or random
effects. Null hypothesis is that the preferred model is random effects vs. the alternative the fixed effects. It tests whether the errors are
correlated with the regressors. The null hypothesis means there is not correlation. chi2(7) = (b- B)'[(V_b- V_B)^(-1)](b-B)=87.12 Prob>chi2 = 0.0000
Reject the null hypothesis and conclude that fixed effects model is the one appropriate.
Note: This table represents a number of tests, which were done prior the regression estimation. Column 1 lists the names of the tests, column 2 lists the definition of the tests, column 3 shows outcome produced in STATA, column 4 derives conclusions based on the results and column 5 lists the appropriate solution for the specific problem.
Table 5
Capital type codification
ISIC EU KLEMS PWT 9.0
Code Asset Name Code Asset Name Corresonding
Asset Name
382-3825 Non electrical
machinery
N11OG Other machinery
3830-3832 Electrical equipment excl communication equipment N11321G N11OG Computer equipment Other machinery equipment and weapons Machinery 3832 Radio tv comm equipment N11322G Telecommunicati ons equipment Machinery
3842+3844+3849 Other transport N1131G Transport
equipment
Transport equipment
3843 Motor vehicles N1131G Transport
equipment Transport equipment 3845 Aircraft N1131G Transport equipment Transport equipment 385 Professional goods - - -N111G, N112 Structures and buildings Structures N117G Intellectual property products Intangible assets Note: This table represents ISIC codes for the assets used in the Caselli and Wilson (2004) paper and corresponding EU KLEMS codes, which were compressed lated into the four assets, used in PWT database. Colums 1 represent ISIC codes, column 3 represents EU KLEMS codes. Column 2 and 4 describes asset type based on the code in the precending column. Where the last column shows the corresponding asset name used in the PWT 9.0 database.
Table 6
List of countries
Angola Albania Argentina Armenia Australia Austria Burundi Belgium Benin Burkina Faso Bangladesh Bulgaria Belize Bolivia Brazil Barbados Brunei Botswana Central African Republic Switzerland Chile China Cameroon Congo Colombia Costa Rica Cyprus Czech Republic Germany Denmark Dominican Republic Algeria Ecuador Egypt Spain Estonia Finland Fiji France United Kingdom Ghana Greece Honduras Croatia Hungary India Ireland Iran Iraq Israel Italy Jamaica Jordan Japan Kazakhstan Kenya Kyrgyzstan Cambodia Republic of Korea Lao People's DR Liberia Sri Lanka Lesotho Lithuania Luxembourg Latvia Morocco Republic of Moldova Madagascar Mexico Mali Malta Mongolia Mozambique Mauritius Malawi Malaysia Namibia Nigeria Nicaragua Netherlands Norway Nepal New Zealand Pakistan Panama Peru Philippines Poland Portugal Paraguay Romania Russian Federation Saudi Arabia Sudan Senegal Singapore Sierra Leone El Salvador Serbia Slovakia Slovenia Sweden Swaziland Syrian Arab Republic Togo Thailand Tajikistan Tunisia Turkey Tanzania Uganda Ukraine Uruguay Venezuela Yemen South Africa Zimbabwe
Table 7
Type by type relative physical capital regression based on Caselli and Wilson (2004) sample of 40 countries in a time period from 1970 to 1995.
Variable Panel A: Machinery
Panel B: Transport
equipment Panel C: Intangible assets
(1) (2) (3) Time Trend 0.0003 -0.0026*** -0.0031*** (0.0002) (0.0004) (0.0006) Inward FDI 0.0002 0.0019 0.0034 (0.0008) (0.0011) (0.0018) Outward FDI -0.0010* 0.0002 -0.0042*** (0.0005) (0.0008) (0.0012) Gov’t share 0.0002 -0.0084 0.0036 (0.0049) (0.0073) (0.0113) Industrial share -0.0365*** -0.0448*** -0.0267 (0.0081) (0.0122) (0.0188) Services share 0.0347*** 0.0650*** 0.033 (0.0094) (0.0141) (0.0217) Human capital -0.0915*** -0.1116** 0.2996*** (0.0241) (0.0362) (0.0557)
Income per capita 0.0819*** 0.0782*** 0.1493***
(0.0065) (0.0098) (0.0151)
# countries 40 40 40
N 482 482 457
R2 0.425 0.514 0.485
Note: Dependent variable is the capital composition index based on the Equation (7). Robust standard errors in parentheses. Panel A contains results for machinery. Panel B contains results for transport equipment. Panel C contains results for intangible assets. Column 1 report capital
composition index for machinery. Column 2 report capital composition index for transport equipment. Column 3 report capital composition index for other assets. ***, **, * denote
Figure 1. Structures composition versus years Figure 2. Transport equipment composition versus years
