examination of OECD countries” “ Public sector size and economic development: a long-term longitudinal growth

1

“Public sector size and economic development: a long-term longitudinal growth

examination of OECD countries”

MSC Thesis 2020-2021 Semester 1 By Arne Mollema (S2136899)

a.p.mollema@student.rug.nl

Supervisor: Lex Hoogduin

Word count: 13136 (including references) Spelling: British English

2

Abstract

The relationship between the public sector and GDP growth is a widely contested topic in contemporary economics. Economic development and the public sector, however, are examined less often. This thesis utilised a co-integrated panel of 32 OECD countries for a period 1951-2019, to examine the relationship between economic development and public sector size. Three dependent variables were used, real GDP per capita growth, unemployment, and life expectancy. The main independent variable was government expenditures as a percentage of GDP. To investigate the relationships, fully modified OLS regressions and vector autocorrection Granger causality testing were employed. This thesis found supporting

evidence for a linearly negative and quadratically positive relationship between government expenditures and real GDP growth, and a linearly positive and quadratically negative relationship between government expenditures and unemployment/life expectancy. Additionally, this thesis revealed evidence for

3

1. Introduction

Since the resurgence of academic interest in macroeconomic growth in the 1950s, economists have studied the factors underlying the difference in macroeconomic growth between countries. Many

influential theoretical and empirical models have, for this purpose, been developed and tested rigorously. Classic examples include the distinguished Solow model (Solow, 1956) and the Ramsay-Kass-Koopmans model (1928) both of which propose macroeconomic growth to be exogenous because of population growth, and/or technological changes. More recently, increasingly sophisticated models have ascribed macroeconomic growth to physical and/or human capital accumulation (King and Levine, 1994) or research and development efforts (Jones, 1995). While the models differ substantially in their predictions, they also exhibit similarities with regards to the factors underlying macro-economic growth. More specifically, the extent to which the government can influence economic growth is a question examined by virtually every modern macro-economic growth theory. Indeed, the role and the size of government in contemporary economies remains one of the most widely contested topics in economics.

A classic divide has existed in economics, concerning the ideal size of the government. The debate has been between proponents of a large, involved government on the one hand, and a small government, that is relegated only to managing the most crucial of public tasks, on the other. Influential economists can be found on both sides of the argument. Milton Friedman, in his famous book “Capitalism and Freedom (1962)” argues that the endeavours of governments’ should be relegated only to those activities that the free market cannot efficiently carry out. Deviation from this proposition, according to Friedman, can lead to inefficiency, lack of motivation, and loss of freedom. Conversely, influential economists such as Joseph Stiglitz have criticized laissez-faire economics and point out that, many key innovations such as the internet, exist primarily because of government stimulus and contribute massively to economic growth. In recent years, theories such as “Modern Monetary Theory” have emerged that recommend a yet more active role of government through aggressive fiscal policy, and the government acting as an

employer of last resorts. Similarly, economists such as Thomas Piketty (2019) have suggested a government that guarantees its citizens a fixed income. This is referred to as “universal basic income”.

4

is true not only in America but in the rest of the developed world as well.|

While the theoretical debate is far from settled, the empirical trend is clear, as most OECD countries have seen a vast increase in the amount of government spending, both in absolute terms and as a percentage of GDP. The average government expenditures as a percentage of the GDP of countries between 1951 and 2019 have increased by 141 percent. Countries, such as Portugal and Spain have seen even sharper increases of government expenditures of 354 and 326 percent, respectively. Other countries, such as the United Kingdom and Cyprus saw milder increases in expenditures of 25 and 12 percent, respectively. Only one OECD country, namely Hungary, decreased government expenditures (by 25 percent). Thus, although the extent to which the governments of countries grew differs, the trend is clear, governments have taken an increasingly large role in modern economies.

Does the expansion of government contribute to economic development or hamper it? Economic development leads to innovation, less crime, and healthier populations and allows societies to grow socially and culturally. Conversely, economic recession leads to unemployment, higher crime rates, and increased prevalence of mental and physical health issues. Therefore, the question is crucial for public policymakers to understand in their efforts to increase the wellbeing of the population. Perhaps the relationship between government and development has never been more relevant. The COVID-19

induced global economic recession had led leaders to request huge stimulus packages and demand a more involved role of government post lockdown. Although such reactions are understandable given the devastating effects of the pandemic, government overreach may exacerbate, rather than reduce, these effects.

While the relation between government size and economic growth has been studied extensively, contradicting results have been found and a consensus has so far not been reached. Moreover, the

literature focuses almost exclusively on economic growth rather than development. While growth is a part of economic development it is not the only relevant measure. Furthermore, studies tend to focus on single, or several countries. Analysing single countries is worthwhile to uncover the country-specific link

5

most studies analyse relatively short periods. The economic impact of long-term government projects and spending may be observed considerably later than its implementation which is why long-term analysis is more appropriate. To my knowledge, Lamartina and Zaghini (2011) conduced only study that examined government size and long-term economic growth in developed economies. However, that study too focused on growth rather than economic development. In conclusion, the relationship between government size and economic development in developed economies is understudied and not fully understood.

This thesis utilises a panel of 32 Organisation for Economic Co-operation and Development (OECD) countries to examine the relationships between government and development. The relations are measured by using fully modified ordinary least squares regression (FMOLS) and vector error correction (VECM) Granger causality testing. Economic development is measured as GDP growth, unemployment, and life expectancy while government size is measured as government expenditures as a percentage of GDP. To answer the question “what is the effect of government size on long-term economic development in developed economies, this thesis proceeds as follows. Section 2 provides an overview of the existing literature and the theoretical framework. Section 3 details the data sources and the econometric model. Section 4 presents the regression and Granger causality results. Section 5 provides a summary and conclusion.

2.

Literature review

2.1 Government and Growth

6

describe conditions of human wellbeing”; examples include; life expectancy, inequality, poverty, and suicide rates.

The causal relationship between government size and growth has been studied extensively and many different, often contradicting results, have been obtained. The ambiguity in the literature is illustrated by Nyasha and Odhiambo (2019), who distinguish four different views with regards to the government-growth relationship, all of which have sound theoretical foundations and are supported by empirical evidence. These four “schools of thought”; can be classified following their participants' adherence to the directional causality between government size and growth. The most dominant of these schools are the Keynesians, who assert government size driven economic growth, and the Neoclassicals, who propose economic growth driven government size. Two alternative schools of thought are the bidirectional causality view and the reconciliation view. Adherents of the bidirectional causality view regard

government size and economic growth to be mutually causal. Those belonging to the reconciliation view propose that the relationship between government size and economic growth is both Keynesian and Neoclassical depending on the relative size of government.

2.2 The Neoclassical View

2.2.1 Government inefficiency

7

They found that modern public firms more often employ hiring practices based on political connection rather than merit. Moreover, public firms tend to pursue political objectives such as low unemployment, resulting in excessive hiring practices, higher labour expenditures, and reduced efficiency. It should be observed that, while some studies (Boardman and Vining, 1989) found that public firms outperform private firms, these studies are often based on static analysis. Ehrlich et al. argue that productivity differences due to ownership structure changes may be inconclusive in the short run, which is why long-term dynamic analysis is more appropriate.

A common criticism levelled against publicly versus privately owned firms’ comparisons is that

differences in efficiency are attributed to ownership factors as opposed to other, external factors. Several studies have analysed the efficiency differences between public and private firms with regards to factors other than ownership structure. Pelzman (1971), was one of the first modern economists to empirically consider behavioural differences between privately and publicly owned firms. In his paper, he analysed PSE’s that were not constrained in any way by the government, for example through the subsidisation of customers, and presents evidence for differences in pricing behaviour with regards to certain customer groups; however, the results are far from conclusive. Vickers and Yarrow (1991) propose that efficiency differences can be partly attributed to the competitive environment in which the firm operates. They consider that public firms operate primarily in less competitive industries and have access to readily available soft government loans to bail them out. Thus, those firms have less of a necessity to innovate, and/or increase efficiency, to stay in business. Vickers and Yarrow subsequently argue that the reason for efficiency differences between public and private firms is not solely due to ownership structure, but discrepancies in the competitive environment and government help as well. Furthermore, Kole and Mulherin (1997), analyzed the differences between public and private firms in the same industries which, they argue, minimizes the external differences, and allows for the isolated study of ownership structure effects. They found that although public firms had higher corporate turnover rates, the performance of public firms was not significantly different from their private counterparts.

8

common criticism of external versus ownership factors is valid, it should be observed that multiple influential studies such as the ones mentioned above, controlled for multiple of these external factors. Ehrlich et al. (1994) controlled for the competitiveness and the regulatory burden of the environment while Karpoff controlled for; access to funding, objectives, available technology, and country of origin. If private entities are indeed more efficient than public ones, it should follow logically that privatization contributes significantly to economic growth. Plane (1997), for example, found a significant positive effect of privatisation on GDP growth in a study on 35 developing economies in the period 1988--1992. Similarly, Barnett’s (2000) study revealed a significant correlation between privatisation and GDP growth, however; he asserts that the results are not sufficient to establish causality.

2.2.2 Crowding out and public debt

In Neoclassical theory, an additional explanation as to why government spending is unable to increase economic growth is; crowding-out effects. John Maynard Keynes (1929) was one of the first economists to describe “crowding-out” although he originally named the term “diversion”. Keynes defined diversion (crowding-out hereafter) as the displacement of private economic activity by public economic activity, although he asserted it would only occur when an economy was nearing full capacity. More precisely, crowding out transpires when increases in public spending results in decreases in private spending. It should be noted that Neoclassicals concede that public spending can stimulate the economy in the short run due to wage/price stickiness (Buiter, 1977). In the long-run, however, crowding-out effects dominate.

Economists have identified several channels of transmission through which crowding-out takes place, depending on the mode of finance. The government has three main ways of financing expenditures; taxes, debt, and money printing.

9

attractive to investors, domestic and foreign, because they pay higher returns. This increased demand for domestic assets results in increased demand for the domestic currency, causing the currency to appreciate. The appreciated currency increases/decreases the relative prices of domestic/foreign goods and leads to a reduction in net exports (Ball and Mankiw, 1995).

Similarly to tax financing, assuming increased private saving does not fully compensate, funding government expenditures through issuing debt also increases the public deficit and decreases national saving. As stated, this leads to higher interest rates, lower investment and lower net exports. Additionally, higher public debt increases public policy uncertainty and expectations of future confiscation (Cochrane, 2011a). Investors may interpret increased public debt as an expanded probability of future financial distress and choose to invest in short-term high-risk assets, or not at all (Panizza and Presbitero, 2013). Similarly, households may expect current deficits to be corrected by future increases in the tax rates, prompting them to reduce consumption and increase savings.

When the government finances expenditures through the creation of new money, several issues may occur. Firstly, an increased money supply causes prices of goods to rise, this referred to as monetary inflation (Cecchetti and Groshen, 2000). Inflation reduces the purchasing power of households and increases unemployment (Jones and Manuelli, 1995). Secondly, as a result of higher inflation, the cost of borrowing increases. This happens because the central bank is likely to attempt to control inflation by increasing the interest rate to disincentivize investment spending (Svenson, 1997).

2.2.3 Growth-led government expenditure

In summary; Neoclassical thought asserts that a larger government hampers economic growth. This happens mainly as a consequence of public inefficiency and crowding-out effects. However, as a result of unrelated economic growth the size of the government increases as well. This happens because of

increased demand for public institutions such as policing and infrastructure, increased welfare and cultural spending, market regulation, and public substitution of private activity, for example in education and transportation. Thus, in Neoclassical theory, economic growth causes public sector growth and not vice versa, this assertion is referred to as “Wagner’s Law” (Wagner, 1958). The validity of Wagner's law has been studied extensively. Bohl (1995) analysed the causal relationship between government

10

unidirectional Granger causality from economic growth to public expenditures. More recently, Lamartina and Zaghini (2011) employed Granger causality tests examine 23 OECD countries, and found that the causality ran from growth to expenditures. Similarly, Kumar, Webber and Fargher (2012) studied the relationship in New Zealand for the period 1960--2007 and found evidence of unidirectional Granger causality from economic growth to public sector growth.

2.3 The Keynesian View

2.3.1 Market inefficiency

Under Keynesian theory, government spending and increased government size can increase economic growth, in the short and long-term, due to market inefficiencies. Such expansionary fiscal policy, they argue, is especially potent during recessions when the self-regulatory mechanisms of the free market fail to drive the economy back to equilibrium. Romer (1986) proposes an infinite horizon model of long-term economic growth with utility maximising households and profit maximising firms. The equilibrium in absence of government intervention in this model is not Pareto-optimal because rational firms ignore positive externalities. More specifically, firms only recognize private returns to physical and human capital accumulation but neglect the positive external effect due to a change in the aggregate human and physical capital stock. Consequently, consumption in equilibrium is too high, capital investment and research are too low, and the social welfare optimum is not achieved. The government can attain Pareto improvements not available to private agents through taxation, subsidisation, or other schemes.

Furthermore, as a result of downward wage rigidity, the market absent government intervention is unable to return to full employment equilibrium. In times of economic downturn, wages don’t adjust downward fast enough to clear the labour market. When this happens as a result of cost minimisation by firms or implicit contracts it is referred to as involuntary unemployment (McDonald and Solow, 1981). In any case, government intervention in the form of expansionary demand-side fiscal policy is required to drive the economy back to equilibrium (Dutt, 1987)

2.3.2 Innovation, education, and crowding-in

11

grown substantially and as a result, growth has increased and progress has been made towards the societal goal of less reliance on fossil fuels (Bezemer, 2019, p. 56). Renewable energy is not the only sector in which government spending can stimulate innovation. As Romer (1986) demonstrated, private firms in varying sectors underinvest in absence of government, thus by stimulating research and development (R&D) and innovation, the government can increase growth. Indeed, Grossman and Helpman (1994) show that technological progress is a crucial driver of economic growth and that the government can stimulate such growth through various means. Similarly, Mazzucato (2011) proposes that state involvement in the economy is beneficial to long term growth through innovation support which, she argues, has become stagnant in the private sector. Furthermore, Landau (1986) argues that government expenditures can boost economic growth when it is used for educational purposes as education stimulates the accumulation of human capital, which is an important mechanism of economic growth. Thus, public spending on education and innovation crowds-in rather than crowds-out, private spending.

2.3.3 Government expenditure-led growth

In summation, in Keynesian theory, a larger government causes economic growth by correcting for market inefficiencies and stimulating innovation and education. Several studies have found evidence supporting the Keynesian position. Ghali (1998), used a vector error-correction model (VECM) and found Granger causality running from government expenditures to economic growth in several OECD countries. Furthermore, Loizides and Vamvoukas (2005) utilised a Granger causality framework and found

unidirectional causality running from government expenditures to economic growth in Greece, the UK and Ireland. Moreover Dogan and Tang (2006), examined the relationship in South East Asia and found unidirectional causality running from government size to economic growth.

2.4 Alternative Views

2.4.2 Reconciliation

12

infrastructure (Cherchye, Ooghe, and Van Puyknbroeck, 2008). Furthermore, rather than fully taking over industries, governments should focus on solving inefficiencies and subsidising desirable firm behaviour. When governments deviate from these propositions, efficient private activity starts to be displaced and growth suffers. Hence, the reconciliation view proposes that the relationship between government size and economic growth is non-linear. Government spending increases growth up to a certain point, if the government size increases beyond that point, the relationship becomes negative.

2.4.1 The bidirectional causality view

The bidirectional causality view proposes a feedback response between government size and economic growth. Government spending and economic growth reinforce each other through the mechanisms considered in this thesis; thus, the causal relationship is bidirectional. Cheng and Lai (1997) utilised a vector autoregressive model (VER) to examine the causal relationship between government expenditures and economic growth in South Korea during the period 1954--1994 and found bidirectional causality. More recently, Granger causality was examined for a panel of 182 countries for the period 1950--2004 and bidirectional causality was found (Wu et all, 2010).

2.5 Hypotheses

Following the Neoclassical view, it is expected that countries with large public sectors perform more poorly than countries with small public sectors. This is because, as Keynsians point out, governments are most effective at certain activities such as education stimulation, innovation subsidisation and monopoly regulation (Mazzucato, 2011) (Landau, 1986). When the government grows beyond the scope of these activities private activity starts to be displaced which hampers growth. Therefore, it is also expected that government expenditures and real GDP growth are positively linearly and negatively quadratically related. A limited government can stimulate the economy, however, larger public sectors become obstructive to private activity. The same expectation holds for life expectancy, a limited government can subsidise healthcare but over intrusion in the economy can hamper private healthcare innovation and lead to lower life expectancy growth (Westmore, 2013). For the same reasons, a negative-positive relationship and a positive quadratic relationship between government expenditures and unemployment is expected. Relatively governments can boost job growth but too much government intervention will displace private activity (Bahal, Raissi, and Tulin, 2015).

13

institutions, welfare spending, and regulation will increase. Considering unemployment and life expectancy, it is expected that there is bidirectional causality. Higher life expectancy leads to a higher demand for healthcare, and other government programmes such as social benefits for the elderly, thus life expectancy causes the government to grow to meet this new demand. Similarly, a larger government can spend more on healthcare, and other social benefits which increase the health and wellbeing of the population and thus, life expectancy. Unemployment requires the government to spend more on welfare and/or create new jobs. Conversely, government spending can displace private activity which destroys jobs in the private sector.

Following these considerations, five hypotheses were formulated.

1. “Countries with relatively large public sectors perform more poorly than countries with relatively small public sectors”

2. “There is a positive linear, but negative quadratic relationship, between government expenditures and real GDP per capita growth”

3. “There are negative/positive linear, and quadratic, relations between government expenditures and, unemployment/life expectancy”

4. “There is unidirectional causality running from real GDP per capita growth to government expenditures”

14

3. Data and Methodology

3.1 Description of Data

This thesis utilised a panel data set on 32 (mostly) OECD countries1 _{from the period 1951--2019.}

Slovenia, Latvia, and Estonia were excluded since, including these countries would have reduced the observations by half due to the lack of available data. Variables used in the analysis are as follows. The dependent variables were real GDP per capita growth (RGDPCG), unemployment (UNEM) and life expectancy. The main independent variable was government expenditures as a percentage of GDP (GOVTEX). The control variables included, inflation (INF), the openness of the economy (OPEN), a human capital index (HC), capital productivity (TFP), foreign direct investment (FDI), rent on natural resources (NAT), fuel exports (FUELEX), total public capital investment/stock (IGOVT and KGOVT), and private capital investment/stock (IPRI & KPRI). The dataset was generated from a combination of data from the World Bank, the IMF, and PWT 9.1 (Feenstra, Timmer and Inklaar, 2015). Table 1 outlines a description of the data used in the generation of the data set.

Table 1: Description of data

This table presents a description of the data used in the analysis. Column one shows the name of the variable. Column two shows the abbreviation of the variable. Column three shows the quantification of the variable. Column four shows the period for which data on the variable were available. Column five shows the number of observations N. Column six shows the data source or derivation

Variable Abbreviation Quantification Period Observations N Source

Real GDP per capita RGDP GDP per capita adjusted for price changes in 2011 international dollars 1951-2019 2208 Madison Project Database 2018 Real GDP per capita growth RGDPG Yearly percentage growth of real GDP per capita 1951-2019 2179 Derived from Real GDP per capita

Inflation INF Yearly

consumer price index changes

1960-2019 1879 World Bank

15

Life expectancy LEX Life expectancy at birth in years

1960-2019 1916 World Bank

Unemployment UNEM Percentage of non-working persons in the labour force 1970-2019 1437 World Bank Government expenditure GOVTEX Government spending as a percentage of GDP 1951-2019 2048 IMF historical fiscal prudence dataset Economic openness

OPEN Combined share of imports and exports of GDP

1951-2019 2104 PWT 9.1

Human capital HC Human capital index based on years of schooling and returns to education 1951-2019 2104 PWT 9.1 Physical capital productivity TFP Welfare-relevant total factor productivity at current prices, indexed at 2011 (=1) 1954-2019 1939 PWT 9.1 Foreign direct investment

FDI Net inflow of foreign assets as share of GDP

1970-2019 1430 World Bank

Natural resource rent

NAT Value of capital services flows rendered by the natural resources, i.e., rents, as a percentage of GDP 1970-2019 1558 World Bank

16

investment in billions of dollars

stock data set 2019 Public capital stock KGOV Government capital stock in billions of dollars 1960-2019 1854 IMF investment and capital stock data set 2019 Private-public partnership capital investment IPPP Private-public partnership capital investment in billions of dollars 1960-2019 1218 IMF investment and capital stock data set 2019 Private-public partnership capital stock KPPP Private-public partnership capital stock in billions of dollars 1960-2019 1218 IMF investment and capital stock data set 2019

Total public capital investment

IGOVT Sum of public and private-public partnership capital investment 1960-2019 1854 IMF investment and capital stock data set 2019

Total public capital stock

KGOVT Sum of public and private-public partnership capital stock 1960-2019 1854 IMF investment and capital stock data set 2019

Private capital investment

IPRI Private capital investment in billions of dollars 1960-2019 1854 Private capital stock

KPRI Private capital stock in billions of dollars 1960-2019 1854

3.2 Methodology

3.2.1 Choice of variables

17

main variable to measure economic growth has since been taken up by most growth studies, (see, for example, Samuelson and Nordhaus, 2010). More specifically, although real GDP is a good indicator of a country’s general wealth, real GDP per capita is the superior measure. Specifically, GDP may rise due to population growth rather than other analyzed factors, using real GDP per capita inherently controls for this effect. Therefore, although real GDP per capita is a superior measure to real GDP, it is not perfect and common criticisms against using it as the only indicator of economic development exits.

Firstly, real GDP per capita, while more appropriate than nominal GDP, is not a completely accurate measure of the overall wealth of members of the population. real GDP per capita may increase because of the enrichment of a relatively low number of individuals. If such is the case, real GDP per capita reveals little about the overall wealth and welfare of the inhabitants of a country. An alternative measure of economic development that reveals more about the wellbeing of the general population is unemployment. High unemployment is related to loss of income, high crime rates, social problems, lost tax revenue, loss of human capital and political instability (Nichols, Mitchell and Lindner, 2013). Furthermore, individuals with a history of unemployment are more likely to experience physical and mental health problems and are often required to pay a higher interest rate on home loans (Brenner and Mooney, 1983). In summation, unemployment is an appropriate alternative indicator to measure a countries economic development.

Secondly rising GDP per capita and low unemployment do not guarantee an increase in the living situation of a countries inhabitants if the extra income is not spent on welfare-enhancing goods. Moreover, even if there is low unemployment in an economy the available jobs might be dangerous, of low quality, or pay poorly. This proposition is best illustrated with an example. In 1944, during World War 2, the American economy boomed, GDP was rising and unemployment was at a record low point of 1.2 percent (Tassava, 2008). However, the majority of the extra income was spent on the war effort rather than welfare increasing institutions domestically. Similarly, a large portion of the jobs in the economy involved fighting overseas or dangerous factory work at home. Another example is the industrial

18

development is correlated with life expectancy; for example, healthcare, education, and human capital accumulation (Cervellati and Sunde, 2009)

In summary, to measure overall economic development real GDP per capita growth, unemployment, and life expectancy are used.

Independent variable

The central independent variable in this thesis was the size of the public sector. The traditional measure of public sector size is government expenditures as a percentage of GDP. Because the government sector likely increases when the economy grows, using nominal government would always show an increasingly large absolute public sector. However, the public sector relative to the size of the full economy might have decreased, which is why government expenditures over GPD was used. This measure captures the relative size of the public sector in the economy and controls for effects of economic growth. Although there are other indicators of public sector size, such as public employment, government expenditures over GDP remains the most consistently used measure in growth studies (Di Matteo, 2013). Note that

government expenditures over nominal GDP rather than real GDP per capita is used This is because data on the latter measure are significantly less widely available.

Control variables

To obtain the most accurate results possible, several control factors must be considered. These factors are introduced to control for inherent differences between the countries in the sample.

Firstly, the degree to which an economy is open to trade influences economic growth and development. Trade restricting policies have been shown to negatively impact economic growth in terms of GDP (Sachs and Warner, 1995). Furthermore, Dollar and Kraay (2004), who studied the effects of economic

globalisation and thus, generally an increase in trade, on growth in the developing world, found that economic globalization has led the developing world to make an effort to catch up with the west. Wacziarg and Welch (2003) examined the relationship between openness indicators and trade liberalization policies in the period 1950—1998. They found that countries with policies aimed at increasing economic openness and trade liberalisation grew, on average, 1.5 percent more than countries that did not adopt such policies. The relationship between trade and growth is summarized quite

19

Closely related to economic openness is foreign direct investment (FDI). Moosa (2002) argues that an economically conductive environment attracts foreign investors and, as such, economic openness increases foreign direct investment. Consistent with this assertion, (Zaman et al, 2018) show that

economic openness and FDI are significantly positively related in their study on India, Iran, and Pakistan during the period 1982-2013. Furthermore, in their study on the relationship between foreign direct investment and economic growth Borensztein, De Gregorio and Lee (1998), found that FDI and economic growth were significantly positively related because of technology diffusion between nations. In

summary, to control the degree to which an economy is open to trade and receives foreign investment, the Impex rate and, foreign direct investment as a share of GDP are used respectively.

Secondly, the states of education, technology, and productivity are closely related to economic development. Regarding education, many macroeconomic growth models consider human capital accumulation a crucial engine of growth (Barro, 1996). This is because human capital increases labour productivity, production, and wages. The degree to which education increases both human capital

accumulation and labour productivity are captured in the human capital index of PWT 9.1. This index not only considers the average years of schooling (accumulation) but also the returns to education

(productivity). Nominal wages rise with productivity; hence, returns to education (wage increases due to schooling) is an acceptable proxy for labour productivity. (Tojerow et al., 2018). This notion is supported empirically by Feldstein (2008), who found that productivity grew proportionally to real wages in the USA over the period 1970--2006. To account for the state of technology in an economy, total factor productivity is used. Total factor productivity as defined as the ratio of aggregate output to aggregate input (Sickles and Zelenyuk, 2019). More specifically, total factor productivity is a measure of the efficiency with which an economy utilises the production factors, labour and capital, to produce output and indicates the overall state of production technology. Better technologies lead to higher productivity of capital and labour. Therefore, increased total factor productivity increases economic growth (Baier, Dwyer and Tamura, 2002). In summation to control for education, human capital, technology, and productivity, the human capital index and total factor productivity are used.

20

impacts growth negatively or positively, it is not disputed that a relationship exists; thus, it needs to be considered. The natural resources of an economy were measured in two ways. Firstly, as most valuable natural resources are ultimately turned into fuel one way or another, fuel exports are used as a proxy for natural resource abundance. However, rising fuel exports may be a result of rising merchandise exports in general, which is why fuel exports were scaled proportionally to the change in merchandise exports to control for this effect. Additionally, the manner in which economies capitalize on natural resources also varies between countries. An illustrating example is a comparison between Norway and Saudi Arabia. Both nations have access to large oil reserves, but one (Norway), used it to benefit the population at large, while the other (Saudi Arabia), only saw the enrichment of a small group of individuals. Furthermore, some economies are more efficient in the extraction of natural resources. To control for resource capitalisation and allocation, natural resource rents were used. Natural resource rents are defined as the surplus-value of natural resources after total costs and normal returns. This measure captures the efficiency of extraction, the cost of refining, the profits of natural resources.

Fourthly, an economy’s infrastructure, both physical and digital, plays an important role in its development. Égert, Kozluk, and Sutherland (2009) examined the role of infrastructure investment in OECD countries and found that it is significantly positively related to economic growth high-quality infrastructure is needed to facilitate the production and transportation of goods Moreover, the quality and availability of public institutions, such as public hospitals, police and fire protection, prisons, and courts are also critically important for economic development. This is demonstrated by Yildirim and Gökalp (2016) who analyzed the relationship between institutions and macro-economic performance in developing countries and found that the integrity of public institutions like counts and central banks, positively impacts growth. They assert that reliable public institutions increase stability and willingness to invest. Both infrastructure and public institutions are part of public capital which is defined as the

aggregate body of government-owned assets (Ashauer, 1990). Thus, to control for the infrastructure and public institutions of countries, public capital was used.

Fifthly, inflation can reduce economic growth. As Jones and Manuelli, 1995 show, inflation can lead to reduced disposable income, reduced net exports, and lower investment spending through decreased consumer demand. Therefore, inflation was used in the analysis.

21

investment and economic growth. They further assert that private firms do the majority of equipment investing in market economies since firms produce most of the physical output, which is why private investment leads to growth. Furthermore, private investment leads to the accumulation of human capital through two main channels (Doyle, 1994). The first is direct private investment in human capital through workplace training programs. Workplace training improves the productivity of workers and subsequently increases the profitability of the firm. The second channel is private investment in public human capital through donations to the public school system. Managers do this because they are increasingly aware of the longterm positive externalities of a more educated workforce. An additional reason to include private capital investment in the model is to control for differing interest rates between countries. Historical data on interest rates is relatively rare and difficult to obtain, which is why it is excluded from the analysis. However, because of the significant correlation between private investment and interest rates, private investment partially controls for varying interest rates between countries as well. In summary, private capital investment is included in the model to control for differing levels of investment and varying interest rates between countries.

3.2.2 Preliminary analyses

3.2.2.1 Multicollinearity test

To test for multicollinearity (linear relation between regressors), correlation analysis and subsequent VIF test were performed. The correlations between the variables are outlined in Table 2.

Table 2: Correlations

This table shows the correlations between the regressors. Numbers are rounded down to 3 decimals. Negative numbers are shown in parentheses.

22

All correlations, except for the public and private investment and capital variables, were within reasonable bounds. This is to be expected because public/private investment directly causes the public/private capital stock to grow.

To illustrate this point further, consider Table 3 that presents the results of the VIF (Variance inflation factor) test, which provides a measure of the degree of multicollinearity present in the sample. VIF values below 10 are the general threshold for an acceptable degree of multicollinearity. As was expected from the correlation table, public/private capital investment/stock fell outside the bounds of acceptable multicollinearity. As a result, public and private capital stock were removed from the analysis.

Table 3: VIF values

This table shows the VIF values of the variables, which indicate the presence of multicollinearity. These values are based on pooled OLS estimation.

Variable VIF 1/VIF

KPRI 67.52* _0.015 IPRI 54.96* _0.018 KGOVT 38.01* _0.026 IGOVT 29.31* _0.034 INT 3.44 0.291 Inflation 2.82 0.354 HC 1.76 0.569 NAT 1.72 0.582 OPEN 1.63 0.612 FUELEX 1.61 0.623 GOVTEX 1.37 0.729 TFP 1.32 0.760 FDI 1.22 0.821

*VIF values above 10 lie outside of the acceptable multicollinearity window

3.2.2.2 Heteroskedasticity

A modified Wald test for groupwise heteroscedasticity in panel data was used, the results of which are presented in Table 4.

Table 5: Modified Wald test

This table presents the results of the modified Wald test. The test is based on a fixed effects regression. All independent variables are included. Column one shows the dependent variable of the regression on which the test is based. Column two shows the generated test statistic. The null-hypothesis assumed homoskedasticity.

Dependent variable Chi-squared statistic

RGDPCG 558.59*

UNEM 5933.20*

LEX 2.5e05*

23

As Table 5 indicates, the null hypothesis of no groupwise heteroskedasticity is rejected at the 1 percent level in all cases. Thus, groupwise heteroskedasticity was present in the sample.

3.2.2.3 Stationarity and co-integration

Table 6 outlines the results of the unit root non-stationarity test for panel data. As the table indicates, the variables of UNEM, LEX, GOVTEX OPEN, TFP IGOVT, KGOVT, IPRI, and KPRI were non-stationary according to both tests. Additionally, FDI, HC, NAT and, FUELEX were non-stationary according to one test.

Table 6: Unit root non-stationarity tests

This table presents the results of two tests for unit root non-stationarity for panel data. Column one shows the tested variable. Column two shows the test statistics of the Fisher (ADF) test. Column three shows the Levin-Lin-Chu test statistic. Based on a plot of the (stacked) time series, an individual intercept and/or trend was added. The null hypothesis in both tests assumed the presence of unit root non-stationarity. Negative values are in

parentheses.

Variable Fisher ADF Levin-Lin-Chu

RGDPCG 500.701* _(16.170)* UNEM 58.973 (.894) LEX 38.140 (.013) INF 119.733* _(6.873)* GOVTEX 35.504 2.241 OPEN 33.201 3.817 FDI 255.437* _.457 TFP 74.034 1.730 HC 31.842 (3.159)* NAT 369.791* _(32.293)* FUELEX 71.891 (2.655)* IGOVT 11.293 6.605 KGOVT 16.577 2.203 IPRI 3.965 11.897 KPRI 7.042 6.909

* Null hypothesis rejected at the 1% level

24

Table 7: Co-integration test

This table shows the results of the Kao test for panel co-integration. Column one shows the relevant dependent variable. Column two shows the generated test statistic. The null hypothesis of the test assumes no co-integration. Negative numbers are in parentheses. All independent regressors are used. Results based on the AIC criterion.

Dependent variable t-Statistic

RGDPCG (8.771)*

UNEM (6.898)*

LEX (8771)*

* Null hypothesis rejected at the 1% level

The Kao test is an extension of the Engel-Granger co-integration test for panel data. As Table 7 indicates, there is co-integration between all regressors and dependent variables. The lag length selection is based on the AIC criterion; however, selecting other criteria did not change the conclusion of the test.

Furthermore, an augmented Dicky-Fuller test which ran on the residuals of the Kao test, indicated that the variables were integrated to the degree one, I(1). This indicated that, while the variables are

non-stationary, a stationary linear combination exists.

In summation, all variables in the panel were integrated to the degree one I(1) Additionally, four variables were highly linearly related, two of which were removed from the model. Lastly, there were significant heteroskedasticity and autocorrelation present in the sample.

3.2.3 The models

Fully modified ordinary least squares (FMOLS) is the most appropriate technique for co-integrated panels. FMOLS is a method used to obtain unbiased estimators of long-run relationships when the underlying regressors are co-integrated to the degree one I(1) (Philips and Hansen, 1990). The FMOLS method works in two stages. In the first stage, the integrating vector is estimated by OLS. Subsequently, the co-integration vector estimates are used to form a “nuisance” or, “drift” parameter 𝜈𝑖𝑡. This parameter

is then used to remove bias in the coefficient estimates caused by co-integration. In the second stage, possible endogeneity between regressors because of correlation between the drift term and the error term 𝜖𝑖𝑡of the OLS estimation is corrected for. The Philips and Hansen method (FMOLS) takes care of this

problem through a semi-parametric correction of the regressand (see, Khan et al., 2019).

25

Where Yit is a (k x 1) vector of I(1) dependent variables, Xit is a (k x 1) vector of I(1) regressors,

and 𝜖𝑖𝑡 is the error term. And Xit is assumed to have the stationary process (II).

II. ∆𝑋𝑖𝑡= 𝜇 + 𝜈𝑖𝑡

Where ∆𝑋𝑖𝑡 is a (k x 1) vector of first differences of I(1) regressors, 𝜇 is the error term and 𝜈𝑖𝑡 is

the drift parameter

The following three relations are estimated using the method described above. For dependent growth variables, growth regressors are used. For level variables, logarithms are used.

1. 𝑅𝐺𝐷𝑃𝐶𝐺𝑖𝑡= 𝐼𝑛𝑓𝑙𝑎𝑡𝑖𝑜𝑛_𝐺𝑖𝑡+ 𝐺𝑂𝑉𝑇𝐸𝑋_𝐺𝑖𝑡+ (𝐺𝑂𝑉𝑇𝐸𝑋_𝐺)2𝑖𝑡+ 𝑂𝑃𝐸𝑁_𝐺𝑖𝑡+ 𝐹𝐷𝐼_𝐺𝑖𝑡+ 𝑇𝐹𝑃_𝐺𝑖𝑡+ 𝐻𝐶_𝐺𝑖𝑡+ 𝑁𝐴𝑇_𝐺𝑖𝑡+ 𝐹𝑈𝐸𝐿𝐸𝑋_𝐺𝑖𝑡+ 𝐼𝐺𝑂𝑉𝑇_𝐺𝑖𝑡+ 𝐾𝐺𝑂𝑉𝑇_𝐺𝑖𝑡+ 𝜖𝑖𝑡 2. 𝐿𝑛(𝑈𝑁𝐸𝑀)𝑖𝑡 = 𝐿𝑛(𝐼𝑛𝑓𝑙𝑎𝑡𝑖𝑜𝑛)𝑖𝑡+ 𝐿𝑛(𝐺𝑂𝑉𝑇𝐸𝑋)𝑖𝑡+ (𝐿𝑛(𝐺𝑂𝑉𝑇𝐸𝑋))2_𝑖𝑡+ 𝐿𝑛(𝑂𝑃𝐸𝑁)𝑖𝑡+ 𝐿𝑛(𝐹𝐷𝐼)𝑖𝑡+ 𝐿𝑛(𝑇𝐹𝑃)𝑖𝑡+ 𝐿𝑛(𝐻𝐶)𝑖𝑡+ 𝐿𝑛(𝑁𝐴𝑇)𝑖𝑡+ 𝐿𝑛(𝐹𝑈𝐸𝐿𝐸𝑋)𝑖𝑡+ 𝐿𝑛(𝐼𝐺𝑂𝑉𝑇)𝑖𝑡+ 𝐿𝑛(𝐾𝐺𝑂𝑉𝑇)𝑖𝑡+ 𝜖𝑖𝑡 3. 𝐿𝐸𝑋_𝐺𝑖𝑡= 𝐼𝑛𝑓𝑙𝑎𝑡𝑖𝑜𝑛_𝐺𝑖𝑡+ 𝐺𝑂𝑉𝑇𝐸𝑋_𝐺𝑖𝑡+ (𝐺𝑂𝑉𝑇𝐸𝑋_𝐺)2_𝑖𝑡+ 𝑂𝑃𝐸𝑁_𝐺𝑖𝑡+ 𝐹𝐷𝐼_𝐺𝑖𝑡+ 𝑇𝐹𝑃_𝐺𝑖𝑡+ 𝐻𝐶_𝐺𝑖𝑡+ 𝑁𝐴𝑇_𝐺𝑖𝑡+ 𝐹𝑈𝐸𝐿𝐸𝑋_𝐺𝑖𝑡+ 𝐼𝐺𝑂𝑉𝑇_𝐺𝑖𝑡+ 𝐾𝐺𝑂𝑉𝑇_𝐺𝑖𝑡+ 𝜖𝑖𝑡 3.2.3.2 Causality

Granger causality between the three dependent variables and public sector size was estimated using a vector error correction model (VECM). The model functions in three stages. Firstly, an unrestricted vector autoregression (VAR) is estimated. The VAR method models a set of k variables as a linear function of only their past values, which are collected in a vector yt. Secondly, an error correction term is

added, which accounts for endogeneity and makes testing for long term relationships possible (Johansson, 1991). Lastly, a panel Granger causality test with a dynamic error correction term was ran.

26

I. VAR: 𝑿𝑖𝑡= 𝒄 + 𝝓1𝑿𝑖𝑡−1+ 𝝓𝑝𝑿𝑖𝑡−𝑝+ 𝜖𝑖𝑡

Where Xit is a (k x 1) vector of variables, c is a (k x 1) vector of constants serving as the intercept

terms, Xit-p is the pthe lag of the relevant variable and p is the optimal number of lags, 𝜙p is the

pthe (k x k) matrix of estimated coefficients, and 𝜖𝑖𝑡 is the error term.

Subsequently, an error correction term (II) is estimated

II. ECT: ∆𝑿𝑖𝑡= 𝚷𝑿𝑖𝑡−1+ ∑𝒑−𝟏𝒊=𝟏 𝝓𝒊∗∆𝑿𝒊𝒕−𝟏+ 𝝐𝒊𝒕

Where Π and 𝜙_𝑖∗ are functions of (III) and (IV) of the 𝜙 terms.

III. 𝝓_𝑗∗= − ∑𝑝𝑖=𝑗+1𝝓𝑖, 𝑗 = 1, … , 𝑝 − 1

IV. 𝚷 = −(𝐈 − 𝝓𝟏… − 𝝓𝒑) = −𝝓(𝟏) where “I” is an identity matrix.

Next, Granger causality is estimated according to the system of equations (V)

V. ∆𝒀𝒊𝒕 = 𝜶𝒋𝒊𝒕+ 𝝀𝒋𝒊𝑬𝑪𝑻𝒋𝒊𝒕+ ∑𝒌𝒋=𝟏𝜽𝒋𝒎𝒊𝒌∆𝒀𝒊,𝒕−𝒌+ ∑𝒌𝒋=𝟏𝜽𝒋𝒊 𝒊𝒌∆𝑿𝒊,𝒕−𝒌+ 𝝁𝒋𝒊𝒕

Where ∆𝒀𝒊,𝒕−𝒌 is a vector of first differences of dependent variables, 𝜶𝒋𝒊𝒕 is a vector of constant

terms, 𝝀𝒋𝒊 is a vector of error correction term (ECT) coefficients, 𝑬𝑪𝑻𝒋𝒊𝒕 is a vector of error

correction terms, 𝜽𝒋𝒎𝒊𝒌 is a vector of coefficients of the lagged variables, ∆𝒀𝒊𝒕,−𝒌 is a vector of

lagged dependent variables, ∆𝑿𝒊,𝒕−𝒌 is a vector of lagged independent variables, and, 𝝁𝒋𝒊𝒕 is a

vector of error terms. The subscripts i, t, j, and k indicate the observation, time, variable and lag length respectively.

The following six relations were estimated using the aforementioned method.

1. ∆𝑅𝐺𝐷𝑃𝐶𝐺𝑖𝑡= 𝛼1𝑖𝑡+ 𝜆1𝑖𝐸𝐶𝑇1𝑖𝑡+ ∑𝑘𝑗=1𝜃11𝑖𝑘∆𝑅𝐺𝐷𝑃𝐶𝐺𝑖,𝑡−𝑘+ ∑𝑘𝑗=1𝜃12 𝑖𝑘∆𝐺𝑂𝑉𝑇𝐸𝑋𝑖,𝑡−𝑘+

27 2. ∆𝐺𝑂𝑉𝑇𝐸𝑋𝑖𝑡= 𝛼2𝑖𝑡+ 𝜆1𝑖𝐸𝐶𝑇1𝑖𝑡+ ∑𝑘𝑗=1𝜃21𝑖𝑘∆𝐺𝑂𝑉𝑇𝐸𝑋𝑖,𝑡−𝑘+ ∑𝑘𝑗=1𝜃22 𝑖𝑘∆𝑅𝐺𝐷𝑃𝐶𝐺𝑖,𝑡−𝑘+ 𝜇1𝑖𝑡 3. ∆𝑈𝑁𝐸𝑀𝑖𝑡= 𝛼1𝑖𝑡+ 𝜆1𝑖𝐸𝐶𝑇1𝑖𝑡+ ∑𝑘𝑗=1𝜃11𝑖𝑘∆𝑈𝑁𝐸𝑀𝑖,𝑡−𝑘+ ∑𝑘𝑗=1𝜃12 𝑖𝑘∆𝐺𝑂𝑉𝑇𝐸𝑋𝑖,𝑡−𝑘+ 𝜇1𝑖𝑡 4. ∆𝐺𝑂𝑉𝑇𝐸𝑋𝑖𝑡= 𝛼2𝑖𝑡+ 𝜆1𝑖𝐸𝐶𝑇1𝑖𝑡+ ∑𝑘𝑗=1𝜃21𝑖𝑘∆𝐺𝑂𝑉𝑇𝐸𝑋𝑖,𝑡−𝑘+ ∑𝑘𝑗=1𝜃22 𝑖𝑘∆𝑈𝑁𝐸𝑀𝑖,𝑡−𝑘+ 𝜇1𝑖𝑡 5. ∆𝐿𝐸𝑋𝑖𝑡= 𝛼1𝑖𝑡+ 𝜆1𝑖𝐸𝐶𝑇1𝑖𝑡+ ∑𝑘𝑗=1𝜃11𝑖𝑘∆𝐿𝐸𝑋𝑖,𝑡−𝑘+ ∑𝑘𝑗=1𝜃12 𝑖𝑘∆𝐺𝑂𝑉𝑇𝐸𝑋𝑖,𝑡−𝑘+ 𝜇1𝑖𝑡 6. ∆𝐺𝑂𝑉𝑇𝐸𝑋𝑖𝑡= 𝛼2𝑖𝑡+ 𝜆1𝑖𝐸𝐶𝑇1𝑖𝑡+ ∑𝑘𝑗=1𝜃21𝑖𝑘∆𝐺𝑂𝑉𝑇𝐸𝑋𝑖,𝑡−𝑘+ ∑𝑘𝑗=1𝜃22𝑖𝑘∆𝐿𝐸𝑋𝑖,𝑡−𝑘+ 𝜇1𝑖𝑡

4. Results

4.1 Descriptive statistics

Table 8 presents the descriptive statistics of the dataset.

Table 8: Descriptive statistics

This table presents the descriptive statistics of the dataset. Column one lists the variables. Column two shows the unit of measurement. Column three shows the number of observations. Column four displays the arithmetic mean. Columns five, six, and seven show the minimum value, maximum value, and standard deviation, respectively Column eight shows the percentage of outliers. Negative values are shown in parentheses. All values were generated from observations across all periods and all countries.

Variable Units N Mean Min Max Std. Dev. Outliers*

28

IGOVT Bn. $ 1854 38.62 .078 663.435 85.657 2.59 KGOVT Bn. $ 1854 637.60 1.422 11108.912 1451.336 3.61 IPRI Bn. $ 1920 159.65 .334 3456.697 330.538 1.82 KPRI Bn. $ 1920 1766.92 4.826 34881.591 3537.774 1.77

* An observation is counted as an outlier if its value falls outside of three standard deviations from the mean

Firstly, it should be noted that an acceptable outlier threshold of under five percent was adopted and that none of the variables exceeded that point. The outliers were nonetheless legitimate observations and considering this, none of the observations were removed. Secondly, a substantial variation between observations in all variables was revealed, particularly in the three dependent variables. Countries, on average, grew by 2.9 percent per year (-26.50% min, 29.4% max) and had an unemployment rate of 7 percent (0.2% min, 27.4% max). Furthermore, countries differed vastly in the size of the public sector, with governments being, on average, accountable for 37.26 percent of the GDP (3.4% min 85.3% max). Lastly, substantial differences can be observed for the control variables, most notably the differences in public and private investment.

Perhaps more revealing are the differences in performance between countries with relatively large or small public sectors. The sample was divided into three subsamples with regards to government expenditures (high, medium, and low) and subsequently analysed. Table 9 presents the results.

Table 9: Sub-sample analysis

This table shows the mean, minimum, and maximum value of the variables in column 1 for different government expenditures groups. Columns two, three, and four indicate the values of the variables for the subsamples, “high”, “medium”, and “low” government expenditures. Countries were assigned a label based on the value of average government expenditures over the period 1951-2019. Values of GOVTEX in the 75th percentile or higher gain the label “high”, values below the 25th percentile are labelled “low” and values in between are “medium”. Column four shows the difference in mean values of “high” and “low” countries.

High Medium Low H-L

Variable Mean Min Max Mean Min Max Mean Min Max Mean

RGDP 21940.17 11868.61 26841.05 23955.97 10460.93 35994.65 15112.28 6710.47 33345.17 6827.89 RGDPC 22577.17 3747.06 49388.49 150868.81 1349.26 1033080 8706.66 1182.92 48146.21 13870.51 RGDPCG 2.75 .33 4.09 2.93 2.16 3.44 3.32 2.29 5.64 (.38) RGDPCGY* _{409.38 21.73 822.27 335.01 23.28 798.39 376.09 14.91 1334.81 33.29} INF 8.10 3.25 30.34 6.95 2.61 29.83 16.05 2.45 42.90 (7.95) UNEM 6.65 3.75 9.15 7.22 3.30 14.63 7.264 2.76 14.63 (.62) LEX 75.34 70.97 77.06 75.74 72.36 78.09 71.93 63.36 77.90 3.40 LEXG** _{15.42 11.70 21.06 16.99 11.67 28.87 36.37 17.65 71.13 (5.64)} * Average RGDPC growth from 1951-2019 ** Average LEX growth from 1951-2019

29

GDP and real GDP per capita. However, countries with smaller public sectors grew, on average, .38% per year more than their counterparts. Thus, it is probable that countries with a relatively large public sector are more likely to be high-income countries; however, these countries perform worse in terms of growth per year. Note that large public sector countries grew more when examined over the period 1951-2019, which could be attributed to initial GDP differences. Additionally, countries with large public sectors experienced less inflation and unemployment and had, on average, higher life expectancy. However, countries with small public sectors saw a more radical increase in life expectancy over the whole period. Most interestingly, countries with a medium-sized public sector outperformed with regards to real GDP per capita growth, inflation, and life expectancy. More insight can be gained through an example.

Consider the Netherlands and Spain, the former had a large public sector over the period 1951-2019 while the latter had a relatively small public sector. The public sector of the Netherlands grew by 62 percent, from 25.9 percent of GDP in 1951 to 42 percent of GDP in 2019. Spain's public sector grew by 326 percent from 9 percent of GDP in 1951 to 39 percent of GDP in 2019. The Netherlands’ real GDP per capita grew by 321 percent from 7200 dollars in 1951 to 30000 in 2019. Spanish real GDP increased by 340 percent from 1627 dollars in 1951 to 7175 in 2019. These results are relatively representative for the entire set, countries with larger public sectors performed better, but the public sectors of small countries grew more sharply, as did their GDP’s.

4.2 Regressions

Table 10 presents the results of the FMOLS regressions.

Table 10: Fully modified OLS regressions

This table presents the results of the FMOLS regressions. Column one shows the regressors. Columns two, three, and four show the dependent variables brackets. GOVTEX-SQ is a square term of government expenditures. For all dependent growth variables, growth regressors were used. For the level variables, logarithms were used. Newey-west heteroskedasticity standard errors are in square brackets. Negative numbers are in parentheses. N = 1334

30 GOTEX-SQ .1995* [0.06775] (.11101)* [.01309] (.00019)*** [.000108] OPEN 4.0133* [1.33629] (.38063)* [.03463] .00904** [.004219] FDI (.0020) [.00218] .02314*** [.01273] .000005*** [2.77E-06] TFP 26.7376* [3.76271] (.55674)* [.10303] (.02224)** [.010622] HC 218.8412* [14.59878] (1.33779)* [.08406] (.09405) [.08672] NAT .1863* [.06067] 0.02938* [.00835] .00001 [5.64E-05] FUELEX (.0969)** [.04054] 0.19516* [.01299] (.00003) [5.99E-05] IGOVT 6.4304* [.75891] (0.28451)* [.03186] .00078 [.002085] IPRI 17.0988* [1.26712] 0.30741* [.03017] (.00010) [.003675] Adjusted R2 _{.245986 .200686 .007706}

* significant at the 1% level * significant at the 5% level *** significant at the 10% level

As table 10 presents, there was a negative linear, and positive quadratic relation between government expenditures and real GDP per capita (−1.292 + 0.399𝐺𝑂𝑉𝑇𝐸𝑋_𝐺)2_{. Between government expenditures}

and unemployment/life expectancy the relation was linearly positive and quadratically negative (1.53 − 0.222𝐿𝑛𝐺𝑂𝑉𝑇𝐸𝑋) and (0.00254 − 0.0038𝐺𝑂𝑉𝑇𝐸𝑋_𝐺). Thus, a 1 percent change in the ratio of government expenditures over GDP leads to a change of real GDP per capita growth of -1.29 + 0.1995GOVTEX percent, a 1.53 – 0.111LnGovtex change in unemployment, and a 0.00254 – 0.0019GOVTEX_G change in life expectancy. Regarding the control variables, most of them had the expected signs and were highly significant. Economic openness was positively related to growth and life expectancy and negatively to unemployment. Total factor productivity human capital and natural resource rents were positively related to growth and negatively to unemployment. Both private and public

31

investment displayed a positive relationship with growth; however, only public investment had a negative relationship with unemployment and a positive relationship with life expectancy.

4.3 Causality

Table 10 presents the results of the VECM Granger causality test.

Table 11: VECM granger causality

This table presents the results of the VECM Granger causality test. Column one shows the tested direction of causality. Column two presents the test statistic. Column three presents the optimal lag length, selected with the AIC criterion.

Direction of causality Chi-square statistic Lag length N

GOVTEX  RGDPCG 7.549849 7 1841 RGDPCG  GOVTEX 13.64336** GOVTEX  UNEM 40.43124* 13 1134 UNEM  GOVTEX 26.88012* GOVTEX  LEX 0.907338 3 1738 LEX  GOVTEX 0.923697

* significant at the 1% level ** significant at the 5% level

The Granger test indicated causality running from real GDP per capita growth to government expenditures. It also concluded that there was bidirectional causality between unemployment and

government expenditures. There was no causality between life expectancy and government expenditures.

4.4 Discussion

32

(2013). However, it could also show that the faster growth of countries with a small public sector and low GDP is a consequence of economic convergence due to the utilisation by developing economies of technological innovation of more developed economies (Rassekh and Thompson, 1998). Sub-sample analysis, while insightful, is insufficient to gain more insight into the relationship between government and development.

The regression analyses revealed a negative linear and positive quadratic relationship between

government expenditures and real GDP growth contrary to hypothesis 2. The values of the coefficients indicate that the public sector starts to positively impact economic growth when it grows larger than 3.2 percent of GDP. Although the public sector negatively impacts growth when it is extremely limited, it starts to positively impact growth at a relatively small size, which is in line with the Keynesian viewpoint. Public-sector expenditures are positively linearly and negatively quadratically related to unemployment and life expectancy in opposition to hypothesis 3. When the public sector is limited, expenditures increase unemployment, however, when the public sector grows larger than 6.89 percent the relationship becomes negative. This result is also consistent with the Keynsians. The relationship between public expenditures and life expectancy is also positive linear and negative quadratic. However, the explanatory power of the model was extremely limited and the coefficients were tiny, which is why a conclusion regarding this relationship is difficult to obtain in this case. Thus, while the regression results mostly confirm the Keynsian viewpoint it should be noted that the regression models imply that the relationship between government expenditures and economic growth continues to be positive ad infinitum. In reality, however, this is unlikely to be true as the four schools of thought agree that too large a government impedes healthy economic development. Thus, while the four economic viewpoints differ in their assertion regarding the optimal public sector size, they are in agreement that such a size exists (Nyasha and Odhiambo, 2019). When the model was re-estimated with a cubed government expenditures term however, the results were insignificant. As such, the point at which the relationship between public sector size and development becomes negative again was unable to be identified.

Regarding the control factors; human capital and total factor productivity strongly positively impact growth and decrease unemployment as is consistent with most economic growth theories. This is because human capital and productivity increase production and promote innovation (Corvers, 1996).

33

sample prioritized profitable financial investment over human or physical capital investment which creates jobs.

The Granger causality test revealed unidirectional causality running from growth to government

expenditures, consistent with the Neoclassical viewpoint and hypothesis 4. This result indicates that it is economic growth that causes the public sector to grow. This is likely because of additional demand for public institutions, regulation, and social welfare spending (Wagner, 1958). Additionally, bidirectional causality between unemployment and government expenditures was found. Government expenditures cause unemployment but, unemployment also cause government expenditures as the bidirectional

causality school of thought asserts. This causality could be explained as follows; unemployment increases the need for social welfare spending while government spending displaces private activity which results of fewer jobs, especially since public firms tend to be less efficient than private firms (Boycko, Shleifer, and Vishny, 1996). No causality was found between government expenditures and life expectancy contrary to hypothesis 5.

5. Conclusion

34

smaller public sectors. Countries with medium-sized public sectors performed better still in terms of life expectancy, unemployment, and inflation. The regression analyses uncovered that public sector size and real GDP per capita growth, were negatively linearly, and positively quadratically related, consistent with Keynesian theory. Government size and unemployment/life expectancy were positively linearly, and quadratically negatively, related. This finding is also mostly in line with Keynesian theory. While two regression models had acceptable explanatory power, the explanatory power of the life expectancy model was low, and the coefficients were negligible. Additionally, causation running from real GDP per capita growth to public sector size was found consistent with the Neo-classical view of the relationship. In terms of unemployment, there was bidirectional causality which conforms to the bidirectional causality view. No causality between government size and life expectancy was found.

The results indicated that public policymakers cannot unequivocally promote economic development through government expenditures. Additionally, government spending can cause unemployment which is not a result that is often considered. Policymakers, therefore, need to carefully consider which economic sectors can benefit from government involvement.

This study presented evidence for a Keynsian correlation between government spending and

35

References

Alesina, A and Wacziarg, R. (1997). Openness, Country Size and the Government. NBER Working Paper Series. 6024 (1), p1-27.

Arezki, R and Van der Ploeg, F. (2007). Can the Natural Resource Curse Be Turned Into a Blessing? The Role of Trade Policies and Institutions. IMF Working Paper. 7 (55), p1-36.

Aschauer, D. (1990). Why is infrastructure important?. Conference Series ; [Proceedings]. 34 (1), p21-68.

Bahal, G and Raissi, M and Tulin, V. (2015). Crowding-out or crowding-in? Public and private investment in India. IMF Working Paper. 15 (264), p1-23.

Baier, S and Dwyer, G and Tamura, R. (2002). How important are capital and total factor productivity for economic growth?. Economic Inquiry. 44 (1), p1-66.

Barnett, S. (2006). Evidence on the Fiscal and Macroeconomic Impact of Privatization. IMF Working Paper. 00 (130), p1-25.

Barro, R. (1989). The Ricardian Approach to Budget Deficits. Journal of Economic Perspectives. 3 (2), p37-54.

Barro, R. (1991). Economic Growth in a Cross-Section of Countries. The Quarterly Journal of Economics. 106 (2), p407-433.

Barro, R. (1996). Determinants of Economic Growth: A Cross-Country Empirical Study. NBER Working Paper Series. 5698 (1), p1-79.

Bartel, A. and Harrison, A (2005). Ownership Versus Environment: Disentangling the Sources of Public-Sector Inefficiency. The Review of Economics and Statistics. 87 (1), p135-147.

Baxter, M and King, R. (1993). Fiscal Policy in General Equilibrium. American Economic Review. 83 (3), p315-334.

Bezemer, D. (2019). Functioning. In: Errington, C Money and its uses: development or wealth?. Utrecht: Libertas/Pascal. p47-69

Boardman, A and Vining, A. (1989). Ownership and Performance in Competitive Environments: A Comparison of the Performance of Private, Mixed, and State-Owned Enterprises. Journal of Law and Economics. 32 (1), p1-33.

Bohl, M. (1996). Some International Evidence on Wagner's Law. Public Finance. 51 (2), p185-200. Borensztein, E and De Gregorio and Lee, J. (1998). How does foreign direct investment affect economic 1 growth?. Journal of International Economics. 45 (1), p115-135.

Boycko, M and Shleifer, A and Vishny, R. (1996). A Theory of Privatisation. Economic Journal. 106 (435), p309-319.