THE ROLE OF THE STATE IN ECONOMIC DEVELOPMENT: DO GOVERNMENT EXPENDITURES PROMOTE GROWTH IN DEVELOPING COUNTRIES?

(1)

4. Results

Government Expenditures and Economic Development 0

UNIVERSITY OF GRONINGEN,FACULTY OF ECONOMICS AND BUSINESS

THE ROLE OF THE STATE IN ECONOMIC DEVELOPMENT:

DO GOVERNMENT EXPENDITURES PROMOTE GROWTH IN

DEVELOPING COUNTRIES?

Master’s Thesis

David Schmengler (S3296164)

Institution: University of Groningen

Faculty: Faculty of Economics and Business

Academic Year: 2017-2018

Study Program: MSc International Economics and Business

Supervisor: Dr. Gaaitzen de Vries

(2)

Abstract

This paper examines the role of the state in economic development. In particular, we test the impact of aggregated and sectoral government expenditures on economic performance in developing countries. Our findings suggest that total government expenditures are negatively related to economic development in both low income and middle income developing countries. However, we find particular evidence that government expenditures prove to be more beneficial for growth when a country is characterized by a well-functional institutional design. In contrast to previous studies, we do not find positive and significant results for sectoral government expenditures. In fact, our findings suggest that transport and communication, defense and health expenditures are negatively related to economic development. At the same time, public and private investments appear to be positively related to economic development throughout our analysis. Therefore, we conclude that – instead of government expenditures – stimulating investments and institutional reforms should play a major part on the political agenda of developing countries.

(3)

Table of Contents

Table of Figures

Table 1: Descriptive Summary Statistics ... 19

Table 2: Baseline Results – Standard Fixed Effects Estimation ... 22

Table 3: Robustness Test I – Time Lag Extension ... 24

(5)

1. Introduction

The nation state and its policies played and still play an important role in the economic and socio-political development process of developing countries. In some developing countries, government activities and industrial policies were the main driver for growth, in other cases market liberalizations and a restrained government state were beneficial for economic development (Szirmai 2015). Therefore, it is not surprising that the role of the state in economic development remains one of the most controversial debates in economic literature. Whereas some researchers emphasize the developmental power of state interventions and government expenditures, others criticize that government activities may actually lead to market inefficiencies and therefore hamper economic development. For instance, Ram (1986), Romer (1989), Romer (1990) and Rubinson (1977) find that total government expenditures have a positive effect on growth and development. On the contrary, researchers such as Barro & Lee (1994), Afonso & Furceri (2010) and Dar & AmirKhalkhali (2002) emphasize that government spending has a rather deteriorative effect on economic performance.

In addition to that, other studies take the effect of disaggregated government spending into account and investigate the effectiveness of sectoral government expenditures for economic development, such as public education, defense, transport and health expenditures. In this regard, findings are equally ambiguous. For some sectoral government expenditures, researchers suggest a positive and significant relationship relative to economic development. Simultaneously, others doubt the relevance of sectoral expenditures with regard to growth and development (see for example Bose, Haque, & Osborn 2007 and Fan, Rao & Rao 2003). Complementary literature underlines the fact that government expenditures alone are not an appropriate instrument to draw meaningful conclusions about the relationship between government intervention and economic development. In fact, researchers argue that any form of government expenditure is embedded in an institutional framework. Therefore, they state that the efficiency and magnitude of government expenditures may also depend on the quality of institutions and the political framework within an economy. As a consequence, researchers like Guseh (1997) and Afonso & Jalles (2016) include proxies for institutional quality in their analyses and illustrate that the quality of institutions considerably influences the economic performance and innovative capabilities of countries.

However, we identify that the above mentioned three factors – aggregated total government expenditures, disaggregated sectoral government expenditures as well as institutional quality – have mostly been treated separately in the existing economic literature so far. Therefore, the motivation of this paper is to estimate the effect of all three determinants in one combined analysis. The combination of all three factors in one econometric framework allows us to disentangle their respective effects and analyze the impact of government expenditures on economic development for a panel dataset of 35 developing countries between 1995 and 2012.

(6)

2. Literature Review

First, this article examines the effect of both aggregated total government expenditures as well as disaggregated sectoral government expenditures on economic development. Indeed, we collect a variety of different government expenditure data. On the sectoral level, we make use of a novel sectoral government expenditure database that has not been used so far for empirical research. As a consequence, the integration of a new and consistent dataset allows use to analyze the effect of both aggregated and disaggregated government expenditures in one econometric framework. At the same time, we examine whether government expenditures prove to be more beneficial for growth when a country is characterized by a well-functioning institutional design. Briefly, we present one econometric framework that tries to capture the distinct effect of all three above mentioned factors.

Second, existing literature derives – based on its findings about the relationship between government expenditures and economic development – concrete policy recommendations for decision-makers in economic politics. On the one hand, we equally derive economic policy recommendations for policy makers in developing economies. On the other hand, we analyze our results with regard to the two main schools of thought in economic growth theory – notably the neoclassical one sector growth and the two sector structural change model. Based on our empirical findings, we try to evaluate the suitability of each of the two theories with regard to economic policies in developing countries. In particular, we critically examine the neoclassical and structural change model and provide closing remarks about the suggested role of the state in development policies.

To sum up, the aim of this paper is to examine the effectiveness of public government expenditures. We build upon recent findings in economic literature and analyze different government expenditures and their impact on economic performance in developing countries. Our motivation is being translated into the following research question: Can government expenditures promote economic development in developing countries?

The rest of this paper is organized as follows. Chapter 2 contains a literature review that presents the two schools of thought on the role of government in economic growth as well as previous empirical findings concerning the relationship between government expenditures and economic development. Chapter 3 illustrates our data and methodology. Chapter 4 presents our results and main findings. Chapter 5 contains policy recommendations and concluding remarks. Finally, Chapter 6 discusses limitations and possible extensions of this research paper.

2. Literature Review

(7)

2.1 Theories of Economic Development

Neoclassical one sector growth models such as the Rostow (1960) model or the Harrod (1948) and Domar (1947) model focus on the importance of capital accumulation as prerequisite for economic growth. In their framework, increasing capital investment rates are the driving force for economic development (Ghatak 2003). Classical growth models build on a standard Cobb-Douglas Production Function and state that factor accumulation is considered to be a precondition for establishing higher outputs per worker. An accumulation of physical capital per worker is translated into a move along the production function and hence an increased output per worker. Such form of output growth can be characterized as “Extensive Growth”.

Although factor accumulation and investments are closely correlated with economic development, they are not the only condition for an economy to develop. Consequently, Solow (1956) extends the neoclassical growth theory and assumes that, without technological change, there will be diminishing marginal returns of capital. This corresponds to the idea that the same increase in capital stock has increasingly less impact on output growth. Therefore, Solow (1956) considers three distinct factors that influence economic development and output growth: these are increases in labor quantity and quality (represented by population growth and education), increases in capital (represented by savings and investments) as well as technological progress (Dang & Sui Pheng, 2015).

On the contrary to the neoclassical one sector growth models, economists later described the economic development process as structural change. Key aspect of these structural change models is the reallocation of labor and capital from the traditional sector (mostly represented by agriculture) to the modern sector that is characterized by a higher level of productivity (mostly represented by manufacturing).

Representatives of structural change models are the Lewis (1954) model and the two sector models by Diao, McMillan, & Rodrik, (2017) and Nelson & Pack (1999). In the latter case, Nelson & Pack (1999) assume two sectors, a craft sector and a modern sector. At the initial stage, all labor and capital are in the craft sector. Economic development and growth take place when labor and capital shift to the modern sector since the modern sector is characterized by a higher level of productivity.

(8)

neo-classical capital accumulation approach, development in the structural change model is driven by technological innovation that is being represented by a shift in the Cobb-Douglas Production Function, as more output can be generated with the same amount of inputs. In this methodology, an increase in output per worker can be characterized as “Intensive Growth”.

After having described neoclassical and structural change growth theory, it becomes evident that both approaches differ considerably in their understanding of the role of the state in economic development. Whereas neoclassical growth theorists emphasize private capital investments and a more restrained role of the state, structural change theorists emphasize a more active role of the government and identify industrial policies as an important leverage to promote entrepreneurship, innovation and structural change. To sum up, neoclassical one sector models and two sector structural change models distinguish themselves among other things in their different comprehension of the role of the state in the economic and socio-political environment. Two categories of potential roles of the state in development will be presented in the following section. We will find that each of them can be more or less closely assigned to one of the above mentioned growth theories.

2.2 The Role of the State in Economic Development

In particular for developing countries, the exact scope and optimal degree of government intervention remains a controversial debate (Cimoli, Dosi, & Stiglitz 2009; Szirmai, Naudé, & Alcorta, 2013). In a nutshell: what role can or should the government play in the economic and socio-political development of developing countries? In general, economic literature distinguishes two categories of state regarding the economic development of countries and nations: classical states and developmental states.

On the one hand, classical states are characterized by the fact market liberalization, privatization and private investments are expected to promote entrepreneurship, technological innovation and economic development (North & Thomas 1973). The role of the classical states is to guarantee a functioning institutional framework, power and law enforcement, national defense, a stable macroeconomic environment as well as an adequate transportation and communication infrastructure (Weber 1922). To be more precise, the state’s role is limited to provide important public services such as education, infrastructure and a well-functioning institutional environment. At the same time, state interventions and excessive government size are considered to lead to inefficiencies and market distortions, therefore they ought to be reduced to a limited degree. Hence, the concept of classical states is referring closer to the neo-classical growth theory, represented by Solow (1956) or Harrod (1948) and Domar (1947). Since factor input accumulation is expected to be the primary source of development, classical states can be interlinked with the “accumulationists” perspective on economic development.

(9)

economic and political relationships that can support sustained industrialization [sic]– or in short, a developmental state”. Therefore, developmental states distinguish themselves by state interventions and often aim to support domestic industries with the help of active industrial policies. These industrial policies can entail export subsidies, import substitutions as well as credit facilities. Therefore, developmental states are often characterized by active industrial policies that attempt to promote domestic industrialization and export-led economic growth. Compared to classical states, developmental states can be interlinked with the “assimilationists” perspective on economic development.1

In the last decades, both market liberalizations and privatization initiatives (classical state view) as well as state interventions and active industrial policies (developmental state view) have been used to foster and promote economic development in developing economies. In both cases, results have been manifold, both concepts have seen examples of success and failure in different countries.

For instance, in some emerging economies such as Chile and Ghana, market liberalizations and a lower degree of government intervention stimulated economic development. On the contrary, in other regions such as Africa and parts of Latin America, deregulation policies showed rather disappointing results and actually made a negative contribution to economic development (Szirmai 2015). The same conflicting results can be found for economic policies that relied on government interventions. On the one hand, South Korea in the 1960s and 1970s is a well-known example for a successful developmental state concept. In fact, the Korean government pursued active industrial policies to foster industrialization. It provided financial and administrative support for upcoming new industries (e.g. shipbuilding, automotive, electronics) and used import substitution and export subsidization to promote export-led economic growth. Governmental institutions played an important role in ‘picking winners’ and establishing the concept of infant industry promotion (Heo, Jeon, Kim, & Kim 2008; Wad 2009). This policy made a significant contribution to South Korea’s unprecedented success story and its transition from a developing country to one of the leading economic nations worldwide. On the other hand, government intervention failed in Indonesia, when active industrial policies were designated to promote the domestic aircraft and manufacturing industry in the 1970s (Chang 2010).

To sum up, and based on the above mentioned discussion, we raise the question whether government interventions or rather market liberalization policies are the more appropriate policy instrument to promote economic development. So far, a lot of empirical research has been done to examine the relationship between government intervention and economic development. Results are twofold. Whereas some economists emphasize the developmental power and advantages of active industrial policies, others warn of excessive government size that may lead to market distortions, inefficiencies and a misallocation of resources. In the following section, we will summarize the key findings of previous economic literature about

1_{For a broader and more extensive discussion about accumulationists and assimilationists, see}

(10)

government spending and its impact on economic development. Building upon these key findings, we will develop three hypotheses that will be examined in the main part of this paper.

2.3 Government Expenditures, Institutional Quality and Growth

To begin with, many researchers have examined the impact of overall government spending on economic development in a variety of countries. Results are double-folded. Barro & Lee (1994) find that economic development, measured as the growth rate of real GDP per capita, is negatively correlated to government size and the ratio of government spending. In addition to that, Afonso & Furceri (2010) examine the effect of government size and fiscal volatility on growth for a set of OECD and EU countries. Their overall results equally suggest that both indicators tend to hamper economic development in both country samples. Concretely, government’s total revenue as well as government’s total expenditures seem to impediment the real growth of GDP per capita for both OECD and EU countries. Moreover, Dar & AmirKhalkhali (2002) examine the role of government spending by explaining the differences in economic growth rates of 19 OECD countries over the 1971–1999 period using a random coefficients model. Their results show that “[…] on average, total factor productivity growth, as well as the productivity of capital, are weaker in countries where government size is larger”. Therefore, they conclude that greater government expenditures lead to policy-induced market distortions that may result in inefficiencies and crowding-out effects.2 According to them, these negative effects are actually expected to hamper private capital investments and technological innovation.

On the contrary, other studies find that government activities may have a positive impact on economic development. Ram (1986) analyzes a cross-sectional dataset of 115 countries from 1960 to 1980 and finds strong evidence that government spending has a positive effect on economic performance and growth. Furthermore, he concludes that the positive effect of government size on growth could be higher in low-income and developing countries. Ram’s findings find support in Romer (1989), Rubinson (1977) and Romer (1990) who equally emphasize the developmental power of government activities with regard to economic performance.

To sum up, it becomes obvious that empirical findings regarding government spending and its impact on development are contradictory. However, we acknowledge that previous studies indeed find evidence for the fact that the state can actually exert developmental power and foster growth. Our first hypothesis focuses on the above mentioned controversy between small and large government spending and its impact on economic development. Relying on Ram (1986) and others, we hypothesize that government expenditures are beneficial for development in developing countries.

2 _{Crowding out is defined as a situation where personal consumption of goods and services and}

(11)

Hypothesis 1: Government expenditures have a beneficial impact on economic development in developing countries

In addition to that, researchers find evidence that the efficiency of government expenditures and its impact on output growth is reliant on other components as well. In fact, the magnitude and significance of government spending may also depend on external factors, such as the governmental form or the institutional environment of an economy. For example, Guseh (1997) examines the relationship between government size and economic growth by taking the form of the political regime and the quality of economic and political institutions into consideration. He uses time-series data for 59 middle-income developing countries between 1960 and 1985. Guseh (1997) finds that government spending has adverse effects on growth in developing countries, but he emphasizes that these negative effects are even more substantial in non-democratic socialist regimes than in democratic and market-oriented societies. Apart from the economic and political regime, other studies also find that a country’s institutional quality impacts the effectiveness of government activities. Afonso & Jalles (2016) find a negative effect of government expenditures on development, but at the same time they show that institutional quality has a positive impact on the level of real GDP per capita. Thus, the effectiveness of government policy and the innovative capabilities of an economy are considerably interlinked with its institutional design (Acemoglu, Johnson, & Robinson 2001; Hall & Jones 1999).

Therefore, our second hypothesis focuses on the controversy between high quality and low quality institutional states and their potential impact on economic development. Considering the above mentioned empirical findings, we hypothesize that the potential positive effect of government expenditures on economic performance unfolds more considerably in developing economies with a well-functioning institutional regime.

Hypothesis 2: The beneficial impact of government expenditures on economic development unfolds more considerably in developing countries with higher institutional quality

In addition to the above mentioned discussion, some parts of the economic literature report an inverted u-shaped curve relationship between government spending and economic performance. This phenomenon describes the fact that government intervention is beneficial for economic development up to a certain threshold. As soon as government spending exceeds this threshold, it actually hampers growth (Barro 1990; Armey 1995; Rahn & Fox 1996). The reasoning behind this argumentation is that in countries with a larger share of government, the incentives for private sector productivity growth are smaller than in other economies with a respectively lower share of government spending (Folster & Henrekson 2001).

2.4 Sectoral Government Expenditures and Growth

(12)

decomposing them in its structural components. Hence, government spending and its impact on development is being analyzed on the sectoral level by separately investigating e.g. agriculture, defense, education, health, social security and transport and communication expenditures. Empirical findings for selected domains will be presented in the following section.

AGRICULTURE EXPENDITURES

The agricultural sector often accounts for the largest share of the national economy in developing countries. Therefore, agriculture still plays a considerably important role in economic and social policies of these countries. Elias (1985) finds a positive relationship between governmental agriculture expenditures and output growth in Latin America. Simultaneously, Fan, Rao & Rao (2003) find that public agriculture expenditures were a vital aspect for the promotion of economic growth in Africa and Asia. In addition to that, Diakosavvas (1995) shows that agriculture expenditures directly influence the performance of the agricultural sector and therefore the imminent food supply of large parts of the population. Therefore, one can assume that agriculture expenditures and agricultural policies in general can directly influence the daily life of large parts of the population in developing economies.

EDUCATION EXPENDITURES

When it comes to human capital and education, many researchers including Bassanini & Scarpetta (2001), Cohen & Soto (2007) and de la Fuente & Doménech (2006) argue that there is a significant and positive relationship between human capital accumulation and economic development. For example, Hansson and Henrekson (1994) use a disaggregated analysis and find that government spending and transfers have a negative effect on development, whereas education expenditures induce a positive effect. Bose, Haque, & Osborn (2007) examine the growth effects of government expenditures for a panel of 30 developing countries over the 1970s and 1980s with a focus on disaggregated expenditures. The results of their analysis are twofold. On the one hand, they find that the share of government capital expenditures in GDP is positively and significantly correlated with economic growth, but current expenditures are insignificant. On the other hand, government investments in education and total expenditures in education are the only forms of disaggregated government expenditures that have a positive and significant impact on economic performance. For other forms of government spending, such as transport, communication and defense expenditures, this is not the case. These findings are supported by Blankenau (2005) who also finds that education spending is likely to increase growth.

TRANSPORT AND COMMUNICATION EXPENDITURES

(13)

Although – as stated above – Bose, Haque, & Osborn (2007) do not find a positive relationship between public transport and communication investments and growth, other authors do observe one. Easterly & Rebelo (1993) use a cross-section dataset of 100 countries for the period 1970-1988. They find that “[…] the share of public investment in transport and communication is robustly correlated with [economic] growth […].” Their results are in line with Aschauer (1989) who equally states that there is a positive correlation between public investments in infrastructure capital and the level of output. Munnell (1990) emphasizes that government spending for capital and infrastructure is necessary to improve US infrastructure facilities. According to his findings, public infrastructure investments will raise the growth rate in capital per worker as well as overall labor productivity and GDP growth.

HEALTH EXPENDITURES

For a long time, health has been considered as the result of increasing welfare and economic growth. But recently, attention has shifted and researchers also acknowledge that the relationship between health and economic development is not only demand driven, but that health can actually be a decisive determinant for positive income effects (van Zon & Muysken 2003). For example, Piabuo & Tieguhong (2017) use a comparative analysis to analyze the impact of health expenditures on growth in the ‘Economic Community For Central African States (CEMAC)’ and five other African countries. Their results show that health expenditures have a positive and significant effect on economic performance in both samples. Moreover, Fan, Rao & Rao (2003) state that government spending on healthcare was particularly effective in promoting economic growth in Africa. At the same time, all forms of government spending except health were insignificant for growth in Latin America. In addition to that, McDonald & Roberts (2002) find that health capital has a significant impact on economic growth rates, especially in developing countries. This is equally in line with the findings of Islam (1995).

However, these results are put into perspective by Webber (2002). He argues that reducing undernutrition would only make a modest contribution to economic development while increasing education and enrollment ratios have a positive and more significant effect. Therefore, according to Webber (2002), policies to increase economic development should favor investments in education over health.

(14)

3. Data and Methodology

Hypothesis 3: Government expenditures for education and infrastructure have a more beneficial impact on economic development than other sectoral government expenditures

To sum up, this paper’s objective is to investigate the relationship between government expenditures and economic performance on three different layers. Our first hypothesis focuses on the overall impact of government spending on economic development at the aggregated level. Our second hypothesis takes institutional quality into consideration and raises the question if government expenditures unfold more effectively in developing countries with higher institutional quality. Our third and last hypothesis puts emphasize on the sectoral level and investigates the impact of disaggregated and sectoral government expenditures on economic performance. This integrated approach is a main contribution of this paper, since these three layers have generally been treated separately in previous literature.

3. Data and Methodology

To answer our research question and examine our above mentioned hypotheses, this paper uses the neoclassical Solow growth model as starting point for its economic framework. A simplified translation of the Solow model is illustrated in Equation (1),

y = A f(K, L)

(1)

where the level of output y is defined as a function of capital K and labor L. The variable A represents the level of technology, indicating that technology and innovation directly influence both the productivity of the factor inputs K and L as well as the overall output level of an economy. Dividing both sides of the equation by L, Equation (1) can be rewritten as

= ∝ + β + ε

(2)

where y / L denotes the level of output per unit of labor, α and β represent unknown regression parameters that need to be estimated, K / L denotes a factor input coefficient (capital K divided by labor L) and ε is an error term that captures and incorporates the level of technology.

3.1 Economic Model

As stated in Durlauf, Johnson, & Temple (2005), most influential previous studies on economic growth and development made use of an econometric framework that builds upon the above illustrated Solow model as baseline for their growth regressions. To examine the relationship between government expenditures and economic development, we make use of an economic framework that has commonly been used to estimate the impact of different explanatory variables on economic development. According to Durlauf, Johnson, & Temple (2005), most cross-country growth regressions are denoted as a version of Equation (3)

(15)

where log yi,0 and Xi represent those growth determinants that are suggested by the Solow

growth model. At the same time, Zi stands for other growth determinants that are outside the

β

₆

(G

_it

x INST

_it

) + ή

_/0 (6)

In their framework, β represent unknown regression parameters that need to be estimated, INSTit and Git denote proxies for institutional quality and government size; εit and ήit are

_i

+ µ

_t

+ ϵ

_it (7)

where yit constitutes the level of output, represented by the level of real GDP per capita. Git denotes total government expenditures, INVit represents public and private investments, Hit stands for human capital and education and INSTit constitutes a proxy for institutional quality. β0 is our intercept term, λi represents a country effect and µt accounts for time effects. We decide to incorporate country specific effects since they allow for permanent differences in the level of income between countries. At the same time, the inclusion of time-specific effects is a common practice in growth regression estimations. We include time effects in our analysis, since output means are expected to increase over time, given the worldwide increase in productivity levels (Durlauf, Johnson, & Temple 2005). Simultaneously, the incorporation of time-specific effects equally allows us to capture global shocks and economic crises (Islam 1995). As in Afonso & Jalles (2016), the variable εit is an error term satisfying the

assumptions of zero mean and constant variance.

On the right hand side of the equation we include total government expenditures (Git) as percentage of GDP to capture the aggregated effect of government spending on economic development. Furthermore, we include public and private investments (INVit) in our regression equation. According to the neoclassical growth theory, investments and capital accumulation are key determinants for economic development. We consider both public and private investments in order to disentangle the distinct effect of both investment sources and their potential effect on economic performance. As a consequence, we hope to clarify whether public investments can have a positive and significant effect on growth or whether private investments only are positively associated with development.

Additionally, we include a proxy for human capital and education (Hit). In fact, we expect to see higher educated societies to grow faster than countries with poorer education standards. Finally, we equally include a proxy for institutional quality (INSTit), since researchers found that countries with an advanced institutional design are more likely to perform better than their less developed counterparts.

Apart from that, it is important to note that we include a lagged dependent variable in our framework in order to avoid reverse causality. We feel that reverse causality is of particular interest for our research question. As a reminder, our model is designed to estimate the impact of government expenditures on economic development. But at the same time, we are aware of the fact that the level of output (in our case represented by the level of real GDP per capita) does not necessarily rise as a consequence of increasing government expenditures and investments. On the contrary, it is also possible that government expenditures increase as a result of income growth and higher GDP per capita.3

3_{The phenomenon of increasing government expenditures as a result of income growth is known as}

β

₅

(G

_it

x INST

_it

)+ λ

_i

+ µ

_t

+ ϵ

_it (8)

_i

+ µ

_t

+ ϵ

_it (9)

With regard to our econometric approach, it is worth mentioning that the majority of previous panel data growth studies use a fixed effects (within-group) estimator rather than a random effects estimator. Fixed effects estimators allow for a set of different country-specific intercepts. However, the slope parameters are meant to be the same and identical across countries, by relying on over time variation within each country. The between variation, namely the variation across countries, is not captured in a fixed effects estimation.

(18)

motivation for the use of fixed effects appears in the Mankiw, Romer & Weil (1992) implementation of the Solow model (Durlauf, Johnson, & Temple 2005).

However, the disadvantage of fixed effects estimators is that they do not estimate time-invariant explanatory variable components across countries, such as geographical location or resource abundancy. However, fixed effects models only control for these time-invariant variables (Williams 2015). For our research question, it ought to be sufficient to control for these time-invariant effects, since we are not particularly interested in differences in geographical location or resource abundancy and their potential impact on economic performance. Therefore, a fixed effects estimation seems to be the preferred option for our economic model. This approach is confirmed by estimating both a Breusch-Pagan and a hausman test for our panel dataset. As we have to reject the hausman null hypothesis that the coefficients for fixed and random effects are the same (p < 0.000), we are supposed to use a fixed effects estimation.

With regard to our econometric framework, we follow the approach of Afonso & Jalles (2016). First, we estimate our growth regressions by running a fixed effects estimation, including time and country-specific effects. These regression estimates denote our baseline results. After that, we slightly change the assumptions of our model and run two robustness tests. With the help of these robustness tests, we investigate whether our baseline results still hold under different assumptions. In our first robustness tests, we change the assumption of a one year time lag and extend it to two and three years respectively. In our second robustness test, we divide our sample into two sub samples, notably one that only includes low income developing countries, and the other one including middle income developing countries. This allows us to examine whether our baseline results hold for both country groups. A more detailed explanation of our robustness tests will be given in the corresponding section 4.2. 3.2 Data Sources

(19)

Firstly, the dependent variable on the left hand side of our regression equation denotes each country’s level of economic development and is represented by the level of real GDP per capita. Annual data on the level of real GDP per capita (GDPCAPit) have been retrieved from the IMF World Economic Outlook Database, October 2017 (IMF 2017c) and are expressed in 2011 international US dollars (PPP).

Secondly, we include different variables of government expenditures in our dataset. At the aggregated level, total government expenditures (TOTGOVEXPit) as percentage of GDP are extracted from the IMF World Economic Outlook Database, October 2017 (IMF 2017c). Government expenditures on the sectoral level (SECTEXPit) have been retrieved from the

SPEED Database of the International Food Policy Research Institute (IFPRI 2015; Bingxin, Magalhaes, & Benin 2015). The 2015 version of SPEED contains information about 10 public expenditure sectors in 147 countries from 1980 to 2012. Sectors include agriculture, communication, education, defense, health, mining, social protection, fuel and energy, transport, and transport and communication (as a group). The SPEED database is focused primarily on the areas that are most important to developing countries and places a large emphasis on productive sectors (such as transport and communication infrastructure), as well as sectors that contribute to the well-being of citizens (such as health, education and social protection) (Yu, Fan, & Magalhães 2015). The database relies on multiple sources, including the International Monetary Fund, the World Bank and national governments. To our best knowledge, the SPEED Database is one of few databases that includes comprehensive data on sectoral government expenditures for a large number of countries. Nevertheless, we could not identify worth mentioning economic publications that made use of this data collection. In this regard, this paper contributes to the existing literature by using a fairly different and sophisticated dataset on sectoral government expenditures. For our analysis, we include government expenditures as percentage of GDP for the following sectors and industries: agriculture (AGRICit), education (EDUCit), health (HEALTHit), defense (DEFit), transport and communication (TRANSCOMit) and social protection (SOCIALit).

Thirdly, we calculate public and private investment rates as percentage of GDP for all countries in our dataset. We distinguish between general government investments (gross fixed capital formation) as percentage of GDP (GOVINVit) and private investments (gross fixed capital formation) as percentage of GDP (PRIVINVit). The ratios as percentage of GDP are obtained by dividing the total amount of investments by total GDP, both values in constant 2011 international US Dollars. The corresponding data has been retrieved from the IMF Investment and Capital Stock Dataset (IMF 2017a, 2017b).

(20)

As soon as we encounter missing data values surrounded by two nonmissing data points, we make use of interpolations. In this scenario, we calculate the linear trend between the two nonmissing values in order to obtain meaningful observation points. At the same time, we encounter missing values at the end of the time series, namely for the years 2011 and 2012. In this situation, an extrapolation in required. Our extrapolation uses information from nonmissing and calculated observations in previous years. Here, a fifteen-year average growth rate method is used, where the missing value is imputed based on the fifteen-year average annual growth rate of previous data points. Consequently, we obtain meaningful and coherent observations points for each year within the time period between 1995 and 2012. Lastly, we incorporate data about institutional quality (INSTit) in our dataset. Data about institutional quality and design have been retrieved from Kaufmann, Kraay, & Mastruzzi (2010). In their paper, Kaufmann, Kraay, & Mastruzzi (2010) present a governance index that evaluates a country’s governance and institutional performance with regard to six different criteria: Voice and Accountability (VOICEit), Political Stability (POLSTABit), Government Effectiveness (GOVEFFit), Regulatory Quality (REGQLTit), Rule of Law (RULELAWit) and Control of Corruption (CORRit). For each of the aforementioned criteria, their index ranges from -2.5 (weak performance) to +2.5 (strong performance). In this regard, it is worth to mention that their governance index did not contain data for the years 1995, 1997, 1999 and 2001. To obtain observation points for these years, the arithmetic mean of the previous and subsequent year has been calculated. To conclude, our variable INSTit denotes the calculated average values of all six governance criteria for each country and year. We report descriptive summary statistics for our panel dataset in Table 1. A detailed and concise overview of our variables and their data sources are illustrated in the appendix.

Table 1: Descriptive Summary Statistics

Unit Obs. Mean S.D. Min Max

GDPCAP 2011 Int. US Dollars (PPP) 627 8.459,64 5.936,96 739,76 26.108,92

TOTGOVEXP Percentage of GDP 593 25,29 9,59 0,00 63,32 AGRIC Percentage of GDP 598 0,94 0,80 0,04 6,67 EDUC Percentage of GDP 597 3,76 2,67 0,16 16,22 HEALTH Percentage of GDP 597 1,67 1,29 0,04 9,78 DEF Percentage of GDP 543 2,07 1,97 0,00 17,52 TRANSCOM Percentage of GDP 590 1,37 1,18 0,00 7,69 SOCIAL Percentage of GDP 588 2,32 2,57 0,01 11,89 GOVINV Percentage of GDP 630 4,90 4,10 0,40 25,20 PRIVINV Percentage of GDP 630 14,40 6,50 0,40 47,60

H Average Yrs of Schooling 558 6,61 2,04 2,65 12,53

(21)

4. Results

With reference to our panel dataset presented in Table 1, we expect heteroskedasticity to play a role in our analysis. We assume that heteroscedasticity might be an issue because larger economies with a higher GDP per capita are likely to be more diverse and flexible with respect to their aggregated and sectoral government expenditures. With regard to our regression analysis, this implies that as the size of an economy becomes larger, the more uncertainty is associated with the predicted value of our dependent variable y. Therefore, the variance of the error term εit increases the wealthier the economies in our dataset get. This

phenomenon is often encountered when using cross-sectional or panel data and especially relevant for analyses regarding any form of private or public expenditures (Hill, Griffiths, & Lim 2012).

To test our dataset for heteroscedasticity, we plot the least squares residuals of each explanatory variable against our dependent variable real GDP per capita. The residuals indicate that there is indeed evidence of heteroscedasticity in our dataset. This impression is confirmed when computing a White Test. Therefore, we will use heteroscedasticity robust standard errors in our empirical analysis. By doing so, we avoid underestimating the variance and standard errors of our estimated regression coefficients (Hill, Griffiths, & Lim 2012). Furthermore, the reported summary statistics in Table 1 give evidence that our dataset is rather diversified. Some variables such as real GDP per capita, total government expenditures and education expenditures as percentage of GDP show large standard deviations since their observation points are largely spread across our panel dataset. For instance, the least developed country in our sample has a real GDP per capita of 739$, whereas the most developed country’s real GDP per capita accounts for 26.100$.

We illustrate the distribution of our variables in different histograms. Hence, we can observe that the distribution of our explanatory variables is skewed since individual countries in our dataset have above average education or defense expenditures that represent outliers in our sample. As indicated in our model equations, we therefore decide to transform our dataset by taking the natural logarithm ln(x) of all variables with entirely positive values. This is therefore the case for all variables reported in Table 1, excluding the index variable INSTit

that contains both positive and negative values. The ln(x) transformation of spread and skewed data is common practice in the empirical literature to linearize the relationship between dependent and explanatory variables.

4. Results

(22)

4. Results

4.1 Baseline Results

First and foremost, we find strong evidence that aggregated total government expenditures are negatively associated with economic performance in developing countries. For our fixed effects estimation in Table 2, we find a negative and significant coefficient for total government expenditures on real GDP per capita.

(23)

4. Results

Table 2: Baseline Results – Standard Fixed Effects Estimation

All models are estimated with a Fixed Effects Estimation, using heteroskedastic-consistent standard errors as well as time and country dummies. Robust p values are reported in parentheses. Level of significance: *** p<0.01, ** p<0.05, * p<0.1; T&C = Transport and Communication; TOTGOVEXP = Total Government Expenditures; INST = Institutional Quality

Moreover, after integrating sectoral government expenditures in our empirical framework, we do not find particular evidence for our third hypothesis. We find a positive coefficient for agriculture expenditures, indicating that the latter might be beneficial for economic development. All other sectoral government expenditures show a negative coefficient, suggesting that they are not positively associated with the level of real GDP per capita. However, all sectoral government expenditure coefficients show an insignificant sign.

Dependent Variable: Real GDP per capita (t+1)

(1) (2) (3)

VARIABLES GDPCAP t+1 GDPCAP t+1 GDPCAP t+1

Total Government Expenditures -0.268** -0.119*

(0.031) (0.073) Agriculture Expenditures 0.012 (0.471) Education Expenditures -0.032 (0.483) Health Expenditures -0.039 (0.145) Defense Expenditures -0.018 (0.470) T&C Expenditures -0.018 (0.248)

Social Protection Expenditures -0.015

(24)

4. Results

Therefore, our baseline results do not allow us to deviate substantial conclusions about the role of sectoral government expenditures and their impact on the level of real GDP per capita. At the same time, it is worth mentioning that we obtain considerably high R-squares for our fixed effects estimation, indicating that our model explains a large part of the relationship between our dependent and explanatory variables.

4.2 Robustness Tests

In the following, we conduct two robustness tests to examine how our core regression coefficient estimates behave as soon as we modify the specification of our econometric model. Robustness checks are a common instrument in empirical literature to test the validity and robustness of regression results. In general, there are several ways to test for robustness, for instance by adding or removing explanatory variables, introducing different time lags or running the same regressions for a specific sub sample (group of countries with a specific characteristic) compared to the original panel estimation.

We decide to apply our first robustness test by introducing different time lags in our model equations. As described in section 3.1, we assume a one year time lag between the actual spending of government expenditures and its potential impact on economic performance. This simplified assumption certainly does not hold for all possible circumstances. In fact, government expenditures as well as public and private investments might take a longer period of time until they unfold their economic impact. In particular, this might be the case for all sorts of investments that are related to manufacturing and capital-intensive industries such as defense and infrastructure. Therefore, we acknowledge that investments in property, plant and equipment (PP&E) require the consideration of a larger time lag. In general, it takes a longer period of time before such capital-intensive assets are ready to operate and they can actually contribute to economic development. Regression results with extended time lags for our dependent variable real GDP per capita (t+2 and t+3) can be found in Table 3.

(25)

4. Results

Table 3: Robustness Test I – Time Lag Extension

All models are estimated with a Fixed Effects Estimation, using heteroskedastic-consistent standard errors as well as time and country dummies. Robust p values are reported in parentheses. Level of significance: *** p<0.01, ** p<0.05, * p<0.1; T&C = Transport and Communication; TOTGOVEXP = Total Government Expenditures; INST = Institutional Quality

Concerning our second hypothesis, our baseline results seem to hold after computing our first robustness test. We still find evidence that the level of institutional quality affects the effectiveness of total government expenditures. The interaction term between total government expenditures and institutional quality remains positive and statistically significant, suggesting that government expenditures are more positively related to economic development when institutional quality is high.

After exposing sectoral government expenditures to our modified time lag assumption, most of our baseline results remain unchanged. However, we now find a statistically significant negative coefficient for social protection expenditures with regard to the level of real GDP per capita. At the same time, we still do not find significant results for other forms of sectoral government expenditures. In fact, the coefficient for agriculture expenditures remains

(1) (2) (3) (4) (5) (6)

VARIABLES GDPCAP t+2 GDPCAP t+2 GDPCAP t+2 GDPCAP t+3 GDPCAP t+3 GDPCAP t+3

Total Government Expenditures -0.239** -0.099 -0.188 -0.052

(0.046) (0.102) (0.106) (0.401) Agriculture Expenditures 0.004 -0.001 (0.809) (0.955) Education Expenditures -0.043 -0.051 (0.298) (0.247) Health Expenditures -0.036 -0.033 (0.154) (0.205) Defense Expenditures -0.024 -0.028 (0.457) (0.503) T&C Expenditures -0.016 -0.015 (0.358) (0.407)

Social Protection Expenditures -0.015 -0.018*

(0.128) (0.086) Public Investment 0.085* 0.069* 0.051 0.075* 0.057 0.060 (0.065) (0.077) (0.255) (0.090) (0.128) (0.207) Private Investment 0.038 0.037 0.057** 0.010 0.011 0.041 (0.170) (0.168) (0.032) (0.747) (0.715) (0.249) Human Capital 0.194 0.172 0.042 0.201 0.171 0.071 (0.211) (0.256) (0.758) (0.165) (0.235) (0.585) Institutional Quality -0.061 -0.779* -0.019 -0.079 -0.770** -0.048 (0.252) (0.053) (0.720) (0.134) (0.046) (0.382) TOTGOVEXP # INST 0.238* 0.229* (0.062) (0.061) Constant 9.321*** 8.866*** 8.764*** 9.066*** 8.635*** 8.701*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Observations 464 464 422 433 433 404 R-squared 0.853 0.863 0.823 0.853 0.864 0.828 Number of id 31 31 28 31 31 28

(26)

4. Results

positive when we consider the level of real GDP per capita in t+2, but the coefficient loses its positive sign in t+3.

Therefore, the slight evidence that agriculture expenditures have a more positive role for development than other sectoral expenditures only survives when applying a two year time lag. All other expenditures still show a negative coefficient, but without statistical significance. As a consequence, there is continuously no particular evidence for our third hypothesis that assumes that education and transport and communication expenditures are more beneficial for development than other sectoral expenditures.

Our second robustness test takes account of the fact that we have a relatively large dispersion between countries within our dataset. As already stated in section 3.2, our panel dataset is characterized by a variety of different countries that stand at diverging stages of their development process. It is therefore legitimate to assume that our findings so far are not necessarily valid for the entire sample of developing countries. It is rather more accurate to assume that government expenditures and state interventions might unfold differently depending on the stage of development a country is currently situated in.

For this reason, we split our panel dataset into two sub samples and run our regression equations for both of these sub samples. This approach help us to examine whether our results are equivalent for both sub samples and independent from the current status of development of a developing country.

To split our sample into two sub samples, we first calculate the arithmetic mean of our logged dependent variable lnGDPCAPit across all observations in our dataset. We denote this

(27)

4. Results

Table 4: Robustness Test II – Sub Sampling

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

VARIABLES GDPCAP t+1 GDPCAP t+1 GDPCAP t+1 GDPCAP t+1 GDPCAP t+1 GDPCAP t+1 GDPCAP t+2 GDPCAP t+2 GDPCAP t+2 GDPCAP t+2 GDPCAP t+2 GDPCAP t+2

Total Government Expenditures -0.327 0.173 -0.243*** -0.240*** -0.313 0.168 -0.201*** -0.203***

(0.115) (0.307) (0.000) (0.000) (0.108) (0.280) (0.003) (0.002) Agriculture Expenditures -0.012 0.006 -0.021 0.000 (0.678) (0.628) (0.457) (0.996) Education Expenditures -0.073 0.007 -0.051 -0.009 (0.435) (0.883) (0.552) (0.837) Health Expenditures 0.006 -0.061** 0.011 -0.051** (0.911) (0.024) (0.827) (0.042) Defense Expenditures -0.046 0.008 -0.092* 0.021 (0.183) (0.758) (0.084) (0.419) T&C Expenditures -0.045* 0.001 -0.047* 0.007 (0.086) (0.925) (0.070) (0.617)

Social Protection Expenditures -0.022 0.004 -0.023 0.002

(0.175) (0.449) (0.146) (0.710) Public Investment 0.069 0.019 0.044 0.082** 0.082** 0.039 0.061 0.017 0.061 0.068* 0.068* 0.030 (0.233) (0.730) (0.457) (0.014) (0.014) (0.255) (0.250) (0.733) (0.292) (0.067) (0.066) (0.452) Private Investment 0.085 0.060 0.041 0.032 0.032 0.090*** 0.093 0.046 0.085 0.004 0.004 0.042 (0.245) (0.384) (0.410) (0.104) (0.106) (0.001) (0.259) (0.550) (0.161) (0.850) (0.849) (0.128) Human Capital 0.331 0.310 0.259 0.021 0.022 -0.169 0.292 0.251 0.194 0.046 0.045 -0.096 (0.522) (0.426) (0.338) (0.870) (0.868) (0.167) (0.552) (0.507) (0.432) (0.696) (0.696) (0.391) Institutional Quality 0.002 -1.704** 0.109 0.017 -0.015 0.023 0.002 -1.576** 0.087 -0.020 0.003 -0.014 (0.990) (0.011) (0.217) (0.805) (0.956) (0.796) (0.990) (0.010) (0.292) (0.728) (0.993) (0.851) TOTGOVEXP # INST 0.595*** 0.010 0.553*** -0.007 (0.003) (0.895) (0.002) (0.933) Constant 8.576*** 6.897*** 7.508*** 10.130*** 10.117*** 9.699*** 8.602*** 6.980*** 7.761*** 9.884*** 9.893*** 9.479*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Observations 185 185 172 310 310 267 173 173 166 291 291 256 R-squared 0.856 0.888 0.830 0.901 0.901 0.894 0.867 0.896 0.847 0.883 0.883 0.882 Number of id 12 12 11 19 19 17 12 12 11 19 19 17

Annotations: see Table 3

Low income developing countries Middle income developing countries Low income developing countries Middle income developing countries

(28)

4. Results

First and foremost, our second robustness test equally suggests that total government expenditures are negatively related to economic development. We find a negative and statistically significant coefficient for government expenditures in middle income developing countries. At the same time, columns (1) and (7) also suggest a negative association in low income developing countries, but here without statistical significance. Briefly, state intervention and aggregated government expenditures seem to be negatively associated with development in developing countries. This finding is especially controversial with regard to the one of Ram (1986) who states that government expenditures are more beneficial for growth in low income developing countries. In our analysis, we do not find evidence for this statement.

However, our robustness tests reveals that the negative impact of total government expenditures has to be put into perspective. Again, we find a positive and significant relationship for both types of investments, notably public and private investments in both sub samples. Most of them show a significant statistical sign. Therefore, the rejection of our first hypothesis is not entirely evident. Although total government expenditures are likely to hamper economic development, public authorities’ and government investments seem to be positively related to economic performance.

(29)

5. Policy Implications and Concluding Remarks

To conclude, our results suggest that – on the aggregated level – total government expenditures are negatively related to the economic development of developing countries. After we split our sample into two sub samples, notably low and middle income developing countries, this result is confirmed for both sub samples. We find that total government expenditures seem to have a negative impact economic performance in low income as well as in middle income developing countries. This differentiation proves to be a new finding with regard to existing development economics literature that tends to pool low and middle income developing countries together in one analysis.

However, this finding is put into perspective when looking at the effect of public and private investments on economic performance. Throughout our analysis, we find positive and significant coefficients for both forms of investment. Therefore, we conclude that public investments can actually be beneficial for the economic development of developing countries. Consequently, we are careful with entirely rejecting our first hypothesis, since there are potential forms of government intervention that might have positive effects on economic development.

With reference to institutional quality, we find strong evidence that better institutional quality induces an increase in the effectiveness of government expenditures. In most parts of our regression analysis, the interaction term between total government expenditures and institutional quality shows a positive and significant coefficient. This paper therefore confirms previous studies that emphasized the important role of institutional quality in economic development policies and highlighted its role in the development process.

Concerning sectoral government expenditures, we find fairly unexpected results with regard to our third hypothesis. In low income developing countries, T&C and defense expenditures are actually negatively associated with development. The same negative correlation can be observed for health expenditures in middle income developing countries. On the contrary, agriculture expenditures are the only form of sectoral government expenditures that show a positive but insignificant coefficient in our baseline results. However, this positive correlation does not hold when we impose our two robustness test. Consequently, we do not find positive results with regard to the role of sectoral government expenditures and their potential impact on development – at least for our dataset and sample of countries. This layer is therefore of particular interest for further research that might make use of an alternative methodological approach.

However, which policy recommendations can we derive from our empirical results?