Does Mining Matter for Growth? An Empirical Evidence from Indonesia

45  Download (0)

Full text

(1)

Does Mining Matter for Growth? An Empirical Evidence from Indonesia

BSc Economics and Business Economics Ashila Ghitha – 12769991

Supervisor: Kees Haasnoot

June 2021

(2)

2 | P a g e

Statement of Originality

This document is written by Student Ashila Ghitha who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3 | P a g e

Table of Contents

Abstract ... 4

1. Introduction ... 5

2. Literature Review ... 8

3. Methodology and Data ... 12

3.1 Empirical Strategy ... 12

3.2 Data Description and Its Sources ... 15

3.3 Descriptive Statistics... 18

3.4 Outliers ... 21

4. Empirical Results ... 22

4.1 Share of Employment in the Mining Sector ... 24

4.2 Share of Credit in the Mining Sector ... 29

5. Discussion and Conclusion ... 32

References ... 35

Appendices ... 39

(4)

4 | P a g e

Abstract

Despite the abundance of natural resources in Indonesia, there have been economic growth differences in provinces with a high share of mining resources compared to provinces with fewer mining resources. This paper aims to find the relationship between the mining industry and economic growth across provinces in Indonesia after decentralization has taken place. Using panel data and various indicators of the mining industry, this paper finds that different ways of measuring the mining industry yield different outcomes in economic growth. The results show that the mining sector positively impacts growth if it is measured as the share of credit but may impede growth if it is measured as the share of employment, and the magnitude impact of this sector on the province’s growth is bigger in the provinces with higher mining resources. The results also show that the employees’ education is strongly negatively correlated with economic growth. These findings generate two main implications: (1) the mining sector is a relatively high capital-intensive industry, and (2) the high-skilled employees tend to work in high-productivity sectors such as manufacture and services rather than in primary sectors. This paper suggests that the choice of mining indicator measurement and its analysis on the relationship between the mining sector and economic growth should be performed carefully.

JEL Classification: O4, O11, O13

Keywords: economic growth, mining, natural resources.

(5)

5 | P a g e

1. Introduction

Indonesia is one of the natural resources-rich countries compared to the rest of the world. The country is known as the world’s largest Liquefied Natural Gas (LNG) exporter for 22 years until 2006, and since then it became the world’s leading steam coal exporter (IEA, 2008). In addition, Indonesia has become the leading exporter and producer of palm oil and the second-largest producer of rubber, robusta coffee, and fisheries products (Dutu, 2015). In 2007, the export earnings from palm oil reached over nine billion US dollars and the industry provided employment for about 3.8 people (Aldaz-Carroll, 2010). Despite high risk of being located within a ring of fire1, Indonesia is also estimated to hold 40 percent of the world’s geothermal reserves, primarily found in Sumatera, Java-Bali, and Sulawesi (IEA, 2008).

In the mining sector, Indonesia is home to 22% of the world’s known nickel reserves with 21 million tones in which the productions mostly take place on the Sulawesi Islands and North Maluku province (NSEnergy, 2021). Furthermore, the Freeport Indonesia Company also mines one of the world’s largest copper and gold deposits in Grasberg minerals district in Papua. From 1990 to 2019, the ditsrict has produced over 33 billion pounds of copper and 54 million ounces of gold which they predict to still increase in the following years (FM, 2021). In terms of natural resources rent as a percentage of GDP, Indonesia is one of the highest among Southeast Asian countries with its peak around 1979–1980 where more than 25 percent of GDP comes from natural resources rent (Figure 1).

1 The Ring of Fire, or also called as the Circum-Pacific Belt, is a path along the Pacific Ocean where the majority of earthquakes and Earth’s volcanoes take place.

(6)

6 | P a g e Figure 1. Total Natural Resources Rent (% of GDP) in Southeast Asian countries

Source: World Development Indicator (World Bank)

The abundance of natural resources in Indonesia, however, has not led to a uniform economic development outcomes within the country. In particular, some provinces with an abundance of mining resources have performed well in economic growth while others have not. An illustration can be seen from the case of East Kalimantan and South Sumatera in comparison to Bali and Yogyakarta. The first two are natural resources-rich provinces with one of the largest shares of GDP in the mining and quarrying sector compared to Bali and Yogyakarta as tourism-based provinces (Figure 2). However, the economic growth in East Kalimantan was low while the corresponding figure was high in South Sumatera (Figure 3). In contrast, the economic growth of low share of mining sector provinces such as Bali and Yogyakarta experienced relatively high economic growth. This raises a question: did the province’s economic growth come from the mining sector? Or, does the natural resources abundance even correlate with economic growth?

(7)

7 | P a g e Figure 2. Share of Mining and Quarrying Sector to GRDP (%)

Source: Indonesian Central Bureau of Statistics (BPS)

Figure 3. Growth GRDP per Capita (%)

Source: Author’s calculations from BPS

In this research paper, I will focus on the significance of the mining industry to Indonesia’s economic growth at the provincial level. As each province has a different deposit of natural resources and the arrangement has been decentralized to the province and district level since 2000, understanding the impacts of the mining

(8)

8 | P a g e sector at the provincial level becomes critically important for economic development strategies and the analysis at the provincial level will be relevant. This paper also measures the mining industry through four different ways: employment, credit, domestic direct investment, and natural resources revenue sharing. These different measurements will further help policy makers in understanding the role of the mining industry in Indonesia. Hence, this will also contribute to the research field as the significance of the mining industry depends on how it is defined and which measurement is being used. The rest of the paper is organized as follows. Section 2 will provide the discussions of the four transmission channel in which natural resources affect economic growth. Section 3 explains the data and the empirical strategy. Section 4 provides the results and analyses while the final section concludes.

2. Literature Review

Natural resources can affect economic prosperity through different mechanisms.

Conceptually, a natural resource acts as an important foundation of economic growth since it raises wealth and purchasing power over imports thus; its abundance raises the economy’s investment (Rostow, 1960). However, since the 1960s, many studies start to question this positive relationship and find that high natural resources intensity actually inhibits economic growth (Sachs & Warner, 1995; Gylfason &

Zoega 2006; Gylfason 2001a; Gylfason 2001b; Rodriguez & Sachs 1999; Leite &

Weidemann 1999). High dependency on natural capital can lead to Dutch disease, rent-seeking behavior, crowding out investment, and undermining human capital in which all these consequences contribute to lower economic growth. Rather than the resource abundances per se giving a bad impact on economies, it is important to note that these indirect effects from natural resources affect activities that promote economic growth (Papyrakis & Gerlagh, 2004; Gylfason 2001a). This section will discuss the four main transmission channels from natural resources intensity to slow economic growth based on Gylfason (2002) classifications. Through these channels, natural resource-rich countries are argued to have slower economic growth.

(9)

9 | P a g e The first channel is through Dutch disease and foreign capital. Dutch disease2 starts when discovering a natural resource causes a surge of raw material exports followed by an increase in the real exchange rate. The currency’s fluctuations trigger the exchange rate volatility and uncertainty that can be harmful to exports, trade, and foreign investment (Gylfason, 2002; Gylfason 2001b; Sachs

&Warner 1995). In the fear of Dutch disease, the government may impose import restrictions to protect domestic producers and thereby lower the degree of openness in the long run (Sachs & Warner, 1995). Consequently, trade and investment are diminished, and so is exchange of goods and services, technology, ideas, and know- how (Gylfason, 2002). Furthermore, the higher demand for non-tradable goods also associates with the tradeable productions focused more on natural resources rather than the manufacturing sector. The manufacturing sector then tends to shrink and cause a socially inefficient decline in growth if externalities in production present (Sachs & Warner, 1995). Hence, Dutch disease is a major concern because of its potentially significant impact on economic growth. From the empirical evidence, Gylfason (2002) found that an increase in the natural resource abundance by ten percentage points is followed by a four-percentage point decrease in the openness index, leading to a decrease in annual per capita growth by around 0.3 percentage points. His cross countries findings are also supported by Papyrakis & Gerlagh (2004), who found that the negative impact of natural resources on openness and terms of trade contributes to over 40 percent decrease in growth.

Second, the abundance of natural resources may lead to a rampant rent- seeking behavior that distracts the interest and efforts of society from creating wealth to infertile interest pursuing (Sachs & Warner, 1995; Gylfason, 2001b;

Gylfason, 2002; Leite & Weidemann, 1999). As the returns of primary production generally in the form of rents, the owner of a resource (governments, enterprise, or individual) has an incentive to distort the markets so that they can earn more profits.

Governments, for instance, can give privileged access of common property rights or introduce imports restrictions to certain producers and create competitions among rent-seekers to gain those favors (Gylfason, 2001a; Gylfason, 2001b; Gylfason,

2 The term Dutch disease was used for the first time in 1977 when the discovery of vast natural gas deposits in the North Sea in 1959 was followed by a crisis in The Netherlands.

(10)

10 | P a g e 2002). These behaviors then can lead to corruption through the act of bribing businesses and public authorities to obtain access to resources rent, distorting the allocation of resources, and reducing both economic efficiency and social equity (Leite & Weidemann 1999; Sachs & Warner 1995; Torvik 2002; Gylfason, 2002).

Moreover, authorities in resource rich-countries tend to become overconfident and have a false sense of security, making them lose sight of the need for good and growth-friendly economic management (Gylfason, 2001a; Gylfason, 2001b). Hence, unless the authorities manage natural resources with the right regulations, i.e., the country has a good institutional quality, natural resources may not help economic growth. A study by Gylfason (2002) has found that an increase in the country’s primary labor share by 16 percentage points is followed by a decrease in the corruption perception index by one percentage point. The decrease in turn goes along with a decrease in per capita growth by one percentage point per year, on average (Gylfason, 2002). However, the rent-seeking behavior channel seems to have the smallest impact on economic growth as compared to other channels (Papyrakis & Gerlagh, 2004; Papyrakis & Gerlagh, 2007).

The third channel is through investment, saving, and physical capital. The abundance of natural resources may reduce public and private incentives to save, thereby hinder economic growth. An increase in the share of output that accrues to the owners of natural resources is followed by the fall of demand for capital, leading to lower real interest rates and less rapid growth (Gylfason & Zoega, 2006). Further, the quality and efficiency of physical capital are important (Gylfason, 2001b;

Gylfason, 2002), which make inefficient management of natural resources can retard the development of financial institutions in particular, thus discourage savings and investments (Gylfason, 2002; Sachs & Warner 1995). Moreover, Gylfason and Zoega (2006) argued that natural resource wealth indeed decreases the need for savings and investment because natural resources provide a continuous stream of future wealth that is less dependent on the transfer of human-made capital to future periods.

This channel was considered as the most important channel because of its highest impact on economic growth (Papyrakis & Gerlagh, 2004; Guo, Zheng, and Song, 2016). A country’s economic growth was found to fall by one percentage

(11)

11 | P a g e point from an increase in natural capital share by 25 percentage points, due to a decrease in the investment ratio by five percentage points. Natural dependence countries were found to have a saving rate of only five percent of GDP and 14 percent average gross investment rate, while natural resource-free and high growth countries can have an average gross saving rate and investment for 32 and 28 percent, respectively (Gylfason, 2002).

Last, natural resources abundant countries tend to dis-incentivize investment in human capital, leading to lower economic growth in the long run. Higher educational attainment enables the labor force to have higher labor productivity and technological progress, contributing to the economy’s wealth (Nelson & Phelps, 1966). Natural resource-rich countries, however, find education not as important because there is less demand for high skilled labor in primary productions compared to learning-by-doing sectors such as manufacture, trades, and services (Gylfason, 2001a; Sachs & Warner, 1995; Gylfason & Zoega 2006). As a result, Gylfason (2001b) argued that countries with too much concentration on natural resources production have workers with less general education to offer in the modern sector and indirectly reduce the possibilities for domestic firms to enter foreign markets, impeding long-run economic growth. Furthermore, public expenditures on education relative to national income also appears to be inversely related to growth, showing the lack of incentive for public authorities to accumulate human capital (Gylfason, 2001b; Gylfason, 2002). “Awash in easy cash”, they see education as an investment that does not pay and more likely to make a mistake compared to nations without natural resources, which have a smaller margin for error (Gylfason, 2001a, p. 858)

The empirical research in this channel shows that a rise in the natural capital share by 25 percentage points is accompanied with a decrease in secondary school enrollment by 45 percentage points, which in turn followed by a decrease in the economic growth almost two percentage points (Gylfason, 2002). Although Guo, Zheng, and Song (2016) found that education to be the least significant transmission channel in their within-country study in China, Papyrakis & Gerlagh (2007) argued that this variable has the most significant relationship to natural resource abundance as it accounts for more than 15 percent variation in US education’s quality. Overall,

(12)

12 | P a g e all four transmission channels have proven that natural resources can danger economic growth if they are not handled properly.

3. Methodology and Data 3.1 Empirical Strategy

To measure the significance of the mining industry at the province level, this study needs a method that allows for controlling province-specific effects on growth caused by unobserved heterogeneity between provinces. Fixed Effects (FE) and Random Effects (RE) are both possible options because they capture the idiosyncratic level of the outcome variable for each individual province. The difference between the two methods is that in FE, the province-specific error term is correlated with the independent variables while in RE, it is not correlated with the independent variables.

I first run the Hausman test which compares RE and FE for each independent variable used to find the right method. The result shows that the null hypothesis is not rejected for all independent variables, meaning that there are no unobservable variables that affect the outcome variable but do correlate with the independent variables and the endogeneity from omitted variable bias is not present. In the next step, I also consider pooled OLS as one of the methods. However, there are some unobservable variables that might correlate with economic growth but cannot be measured. These variables are not correlated with any of the mining indicators so it can avoid the omitted variable bias issue, such as religion variable. Although the sign of the impact of religion is still debatable, studies found that religiosity does have an influence on GDP level (McCleary & Barro, 2006; Chase, 2015; Guiso, Sapienza, and Zingales, 2003). Barro (2004) pointed out that religious activities take time away from economically productive activity while Guiso, Sapienza, & Zingales (2003) state that religious beliefs promote a positive economic attitude thus higher economic growth. In Indonesia, although religious adherence can be measured, there are variations in terms of religious intensity3 across provinces. To the best of my

3The majority of population in Indonesia is Muslim with approximately 87 percent of the population followed by Christian and Hindu around 11 percent and 2 percent of the

(13)

13 | P a g e knowledge, there is no study that found the direct correlation between the religiosity with the mining industry in Indonesia and there is no logical connection between the two as well. Hence, religiosity in Indonesia can be an example of the variable that cannot be directly measured but influence GDP. Therefore, I use Random Effects panel regression as an empirical strategy with the following equation:

𝑦̂𝑖𝑡 = 𝛽0+ 𝛽1∙ 𝑀𝐼𝑁𝐼𝑁𝐺𝑖𝑡− 𝛽2∙ 𝑒𝑑𝑢𝑐𝑖𝑡+ 𝛽3∙ [𝑀𝐼𝑁𝐼𝑁𝐺𝑖𝑡∙ 𝑒𝑑𝑢𝑐𝑖𝑡] + 𝛽4 ∙ 𝑖𝑚𝑚𝑢𝑛𝑖𝑡+ 𝛽5∙ [𝑀𝐼𝑁𝐼𝑁𝐺𝑖𝑡∙ 𝑖𝑚𝑚𝑢𝑛𝑖𝑡] − 𝛽6∙ 𝑦𝑖,𝑡−1+ 𝛽7∙ 𝑋𝑖𝑡+ 𝑢𝑖 + 𝜀𝑖𝑡 (1)

Where 𝑖 represents province and 𝑡 represents time; 𝑦̂𝑖𝑡 denotes the per capita growth rate of Gross Regional Domestic Product from 𝑡−1 to 𝑡. The variable 𝑀𝐼𝑁𝐼𝑁𝐺𝑖𝑡 consists of four indicators that I use to measure the mining industry, namely share of employment, share of credit, share of domestic direct investment, and share of revenue sharing, and later will be regressed one by one. 𝑦𝑖,𝑡−1 represents the initial GDP per capita for each province, suggesting that there is a convergence story where provinces with lower initial GDP per capita grew faster than those with higher initial GDP per capita so the relatively poor provinces can catch up the relative rich provinces. 𝑒𝑑𝑢𝑐𝑖𝑡 or education represents the school enrollment rates at senior level and expected to have negative impact to economic growth as discussed in Gylfason (2002, 2001a, 2001b) while 𝑖𝑚𝑚𝑢𝑛𝑖𝑡, denoting the immunization coverage, is expected to have positive impact to growth. These two variables are also interacted with the share of employment as they represent the quality of employees in the mining industry and later be expected to positively correlate with growth. 𝑋𝑖𝑡 denotes two other control variables as proxy of infrastructure:

percentage of household with access to electricity, and the percentage of villages

population, respectively (BPS, 2010). During Islam’s fasting month Ramadhan, the government issued a regulation that the employees are allowed to leave the office earlier compared to the normal working hours (Anwar, 2021). However, the official cutting hours varies across provinces. For instance, in Aceh where religious intensity relatively higher compared to other provinces, the working hours are cut two and half hours (Setyadi, 2021) while in Bali, it is only cut by one hour (Supartika, 2021). These differences in cutting hours will affect the productivity of the workers which would also affect the economic growth.

Nevertheless, even if the religious adherence in Indonesia can be measured, it is hard to measure religious intensity as the data is not publicly available.

(14)

14 | P a g e with asphalt road. The 𝑢𝑖 is the province random effect and 𝜀𝑖,𝑡 is the error term that is assumed to be independently and identically distributed (i.i.d).

The two interaction variables in equation (1) are used to examine how education and health associate with employment in the mining sector increasing growth. The first hypothesis is employment in mining industry increases growth when the employees have better education. Although in section 2 studies show how education is inversely correlated with growth in countries with abundant natural resource, I expect that an educated employment in the mining industry become more important in increasing productivity e.g. they are more familiar and skillful in operating high technology equipment, and thereby generating higher growth. A cross-countries study found that human capital influences the growth of total factor productivity positively and later will play a role in increasing domestic growth per capita (Benhabib & Spiegel, 1994). Through the interaction between employment and education, the previously negative correlation between education and growth is expected to turn into a positive impact to economic growth. This argument then shows reverse causality where employment in the mining sector can also affect education. A multicollinearity test is performed to assess whether this can pose a problem. The test shows that the Variance Inflation Factor (VIF) is equal to one for both enrollment ratio and share of employment in the mining sector, suggesting that there is no correlation between the two and multicollinearity issue can be avoided.

The quality of human capital can be measured in various indicators such as enrollment rates, literacy rates, or years of schooling. At best, the secondary enrollment rates are the most commonly used to represent investment levels in human capital and the effect of education on growth. However, other measures such as primary enrollment rates, senior enrollment rates, and years of schooling were also found to yield similar results (Gylfason, 2001a; Benhabib &Spiegel, 1994).

The second interaction is between employment in the mining sector and immunization coverage which is representing the health status of the employees.

Immunization is important for economic growth because exposure to diseases in childhood can have an impact on the adult’s productivity (Bleakley, 2010; Stack et al., 2011). For instance, an infection to malaria at young age can decrease adult income by over 50 percent, which could imply an effect on GDP around 30 percent

(15)

15 | P a g e if the childhood infection rates are about 60 percent (Bleakley, 2010). Furthermore, a study by Stack et al. (2011) also found that immunization at childhood in Indonesia can generate value-of-statistical life benefits of 21 billion dollars. This has pushed that disease prevention in childhood, i.e., vaccination and immunization, should not be underestimated as it affects the labor productivity in the future. Hence, by interacting immunization coverage under five years old to the number of employees in the mining sector, I hypothesize that employment with good health status increases productivity and thus will boost economic growth.

3.2 Data Description and Its Sources

In this paper, the coverage area is 31 provinces of Indonesia while the number of provinces currently is 34 provinces. The choice of 31 provinces is due to some provinces split from the parent provinces and formed new provinces after 2001;

meanwhile I do not have any data of these new provinces prior to 2001. Thus, I add back the three new provinces to their respective parent provinces as follows: Riau Islands as the 32nd province since 2002 is added back to Riau, West Sulawesi as the 33rd province since 2004 is added back to South Sulawesi, and North Kalimantan as the 34th province since 2013 is added back to East Kalimantan. The process is carried out by recoding the new provinces back to their parent provinces using identification code issued by the Indonesian Central Bureau of Statistics (BPS) in all datasets.

I intensively use secondary data from BPS and the Indonesia Database for Policy and Economic Research (INDO-DAPOER) prepared by the World Bank. I also use data from the National Single Window for Investment (NSWi) for domestic direct investment in mining. In this paper, I define the mining industry in various forms and the control variables, i.e., education, health, and infrastructure as a set of variables of interests on the right-hand side while growth as a dependent variable.

Economic Growth (Including Oil and Gas)

Economic growth is calculated using the real Gross Domestic Product (GDP) per capita from 2000 until 2018 which is after decentralization take place in Indonesia.

The decentralization in Indonesia started since 1999 but the law was effectively

(16)

16 | P a g e implemented since 2000. Hence, I choose to start the analysis from 2000 to avoid too many missing values and it was no longer a transition period from Soeharto regime. The real GDP from 2000 until 2009 is rebased to the most recent base year, 2010 constant price. The data is available at the province level issued by the BPS.

Mining Industry

In measuring the role of the mining industry in boosting economic growth, I utilize four indicators: the share of employment in the mining sector, the share of credit in the mining sector, the share of domestic direct investment in the mining sector, and the share of revenue sharing from mining sector in each province. Below is the definition of those indicators.

Share of employment in the mining sector: this variable represents the existence of the mining industry in each province by calculating the number of people employed in the mining and quarrying sector over the number of people in the labor force.

Both data is available in INDO-DAPOER from 2001 until 2018.

Share of credit in the mining sector: this variable is used as a proxy of the number of firms in the mining sector by calculating the amount of credit in the mining sector divided by total credit from all sectors. The data is originally issued by the Central Bank and publicly available in INDO-DAPOER from 2000 until 2009. Considering the data source, I assume that the borrowers are firms or businesses in the mining sector.

Share of domestic direct investment in the mining sector: this variable is used to measure investment in the mining sector by using the realization value of domestic direct investment (DDI) as a proxy. The share is calculated from the realization value of domestic direct investment in the mining sector over the total realization value of domestic direct investment from all sectors. As its counterpart, the foreign direct investment (FDI) should be added but this data is not available at the province-level for the mining sector. Domestic direct investment data is taken from the National Single Window for Investment (NSWi), available from 1990 until

(17)

17 | P a g e 2020. As I use the realization value, some data shows zero realization value even though there was a project, implying that no monetary values were generated from the project.

Share of revenue sharing in mining sector: natural resources revenue sharing is the revenue from mining, fishery, forestry, geothermal energy, oil and gas that are collected then shared by the central government to each province where the natural resources is located and produced. The larger the natural resource available, the higher the share is. This money is used as the financial sources to support provincial development. To calculate the share of revenue sharing in mining sector, for each province, I firstly calculate the summation of revenue sharing from all those sub- sectors in the natural resources, then the amount of revenue from the mining sector is divided by the total amount of that summation. The data is taken from INDO- DAPOER and publicly available only from 2005 until 2009.

Education

As it has been discussed in the literature review and empirical strategy, education is an essential endowment for high labor productivity. Although education can have negative correlation with growth, I expect that it can turn into positive if it is interacted with employment. I use the net enrollment ratio at the senior level (in percentage) to measure the worker’s level of education. With a higher enrollment ratio in a province, it can be implied that workers in that province have better education so as to contribute to their productivity. The data is obtained from INDO- DAPOER from 1998 to 2018.

Health

I use the immunization coverage of children under five years old (as a percentage of children under five years old) to measure the health status of people. Immunization when young will decrease the probability of getting diseases at an older age. If workers were not immunized, they could have a lower productivity, affecting the country’s economic growth. The data is taken from INDO-DAPOER from 1996–

(18)

18 | P a g e 1999 and 2004–2017. The immunization covers BCG, DPT, Polio, and Measles for 1996–1999 and covers BCG, DPT, Polio, Measles and Hep B for 2004–2009.

Infrastructure

There are two indicators that I use to measure infrastructure. First, the percentage of villages with asphalt road as it is important for transporting mining products from sites to factories or markets within or across provinces. Further, the mining area is usually located in the rural area so the road availability across villages could represent the availability of passable roads in that province. The data is available every three years from 1996 until 2018 collected by BPS and obtained from INDO- DAPOER. The second indicator is electricity, measured using the household access to electricity (as a percentage of total households). Without electricity, the sector cannot perform optimally because it is the basic need for all activities. The data is available from INDO-DAPOER for 1996–2018.

3.3 Descriptive Statistics

Summary statistics for all data is shown in Table 1. I have 584 observations for the Growth GDP per Capita in 31 provinces while other variables are ranging from 143 to 550 observations. The least number of observations is in the share of revenue sharing as the data only available for four years. The distribution of the mining variables is all highly right-skewed except for the share of revenue sharing (Figure 4). This is also revealed by the skewness in Table 1 where the share of employment, share of credit, and share of domestic direct investment have skewness more than one. To overcome this skewed distribution, the value of these variables is transformed into log and should be checked in the empirical results. It should be noted that the zero (0) in the share of domestic direct investment means that there was a project from domestic investment with zero realization value, while the 99 percent of share of revenue sharing means that 99 percent of revenue sharing that the province received is from mining only, not from other natural resources sector (such as fishery, forestry, etc.).

(19)

19 | P a g e Figure 4. Distribution of the Mining Variables

Source: Author’s calculations from INDO-DAPOER (World Bank) and NSWi

The average of the share employment in the mining sector is two percent while it is around one percent for the average of credit share in the mining sector (Table 1).

When looking further into the distribution of these two mining indicators, Bangka Belitung Islands dominates the share as it counts more than 10 percent of share of employment and share of credit in the mining sector every year. South Kalimantan and East Kalimantan also have higher shares compared to other provinces. For the share of domestic direct investment in the mining sector, East Nusa Tenggara Timur accounts the two highest share with 100% and 98.5% in 2011 and 2014, respectively.

(20)

20 | P a g e Table 1. Summary Statistics

Source: Author’s calculations from BPS, INDO-DAPOER (World Bank), and NSWi

The correlations between the mining sector and growth per capita can be seen in Appendix A1. The four indicators of the mining industry show weak correlations with growth in which the share of employment and share of domestic direct investment are negatively correlated and the share of credit and revenue sharing are positively correlated. The strongest correlations are found in the education and health variable where both show strong negative and strong positive correlations with growth, respectively. Initial GDP per capita is also strongly negative with growth as discussed in the convergence theory. In a cross-country and within- country analysis, the per capita growth tends to be inversely related to initial level of output or income (Solow, 1956; Barro, 1992). Without holding the steady state level constant, Barro (1992) found that regions with poorer economies grow faster than richer regions, creating a convergence.

Variable N Median Mean SD Skewness Min Max

Growth GDP Per Capita (%) 584 3.19 2.81 4.40 0.90 –22.78 36.28 Initial GDP per Capita (Rp) 584 6.29 8.38 7.82 2.79 1.75 45.61 Share of Employment (%) 550 1.04 2.00 3.43 4.58 0.02 27.84 Share of Employment (log) 550 0.04 0.06 1.07 0.06 –4.16 3.33

Share of Credit (%) 284 0.15 1.19 4.74 8.67 0.00 59.47

Share of Credit (log) 284 –1.93 –1.81 1.83 0.38 –6.78 4.09 Share of Domestic Direct

Investment (%)

174 1.42 13.39 22.26 2.07 0.00 100.00

Share of Domestic Direct Investment (log)

174 0.35 0.57 2.73 –0.73 –9.05 4.61

Share of Revenue Sharing (%) 143 21.58 40.41 40.42 0.37 0.00 99.88 Net Enrollment Ratio at Senior

Level (%)

571 50.06 48.99 11.42 –0.16 19.62 72.99

Immunization Coverage (%) 431 93.26 92.63 4.72 –1.39 70.13 99.82 Villages with Road: Asphalt (%) 210 62.35 61.80 22.11 –0.12 13.32 100.00 Household access to electricity

(%)

542 91.73 86.06 15.19 –1.40 35.45 100.00

(21)

21 | P a g e

3.4 Outliers

The growth in GDP per capita has an average of around 2.8 percent with a moderately skewed distribution to the right (Table 1). Since there is a possibility of outliers, I check the data distribution through a boxplot as presented in Figure 5. The figure shows the first quartile (the lower box), the third quartile (the upper box), and the median at 3.18 percent which is the middle line between the two boxes. The observations above the upper fence and below the lower fence (the two vertical lines) are regarded as outliers, each pointed with the province’s name. The lowest and highest growth per capita can also be seen in the figure which is in Papua and West Papua at –22.7 percent and 36.2 percent, respectively. These two anomalies are both related to the mining sector, namely, (i) the high growth in West Papua in 2011 is caused by LNG production that has been operated since 2010 (Bappenas, 2015) and (ii) the very low growth in Papua was because of the world copper price fluctuations especially from 2005 to 2008 (WB, 2010), leading to the fluctuation in copper mine productions which contributed over 65.92 percent to the Papua’s economy (Pergub, 2009).

Other provinces that showed outliers in the growth per capita but all related to the mining sector are as follows: West Papua in 2001 and 2010–2012, Papua in 2001–2008, West Nusa Tenggara in 2011 and 2015, Bangka Belitung Islands in 2003, Central Sulawesi in 2013–2014, and Riau in 2001. These provinces also have a high share of mining on economic growth and some of them had productions of mining factories that started between these years. As a robustness check, these outliers will be removed in the empirical result to see whether they impact the results. There are also outliers in the mining variables such as in the share of employment, share of credit and share of domestic direct investment. However, since they are located in the provinces that I am interested in, i.e., natural resource- rich provinces such as Kalimantan and Bangka Belitung, they are not removed.

If the source of outliers in growth per capita is not related to the mining sector such as civil strife or natural disasters, I remove the observations from dataset. For instance, the high negative growth in Aceh in 2001 which was likely due to civil conflict or Papua in 2010–2011 where its low growth was estimated to be an after-effect of the earthquake in 2009. After checking one by one where the

(22)

22 | P a g e outliers come from, I finally removed 20 observations where their outliers are not related to the research.

Figure 5. Box Plots of Growth GDP per Capita (%)

Source: Author’s calculations from BPS

4. Empirical Results

This section presents and discusses the results of the relationship between the mining industry and economic growth at provincial level. As discussed in the previous section, the following empirical results do not contain the outliers in growth GDP per capita except the ones related with the mining sector. This section also discusses the robustness check for all of the regression results by performing the following methods. First, removing all outliers in growth GDP per capita including those that are related to mining. This is to assess whether the empirical results depend on the very high or low growth that comes from provinces with high dependence to mining. Second, using log transformation for the mining indicators that have highly skewed distribution: share of employment, share of credit and share of domestic direct investment. Since the share of domestic direct investment contains data with zero value, the log transformation in this variable is carried out by

(23)

23 | P a g e adding the observation with one before changing it into log. The share of revenue sharing is not transformed into log because the distribution is not skewed.

Table 2 presents the estimation results of the basic model where each of the mining variable, i.e., the share of employment, share of credit, share of domestic direct investment, and share of natural resources revenue sharing is regressed with the growth GDP per capita. The results show that only the share of employment and the share of credit are strongly correlated with the economic growth, and both coefficients are statistically significant at 5% and opposite signs. An increase in the share of credit by one percentage point could increase the economic growth by 0.0578 percentage points. In contrast, an increase of one percentage point in the share of employment may impede growth by 0.0573 percentage points. To get a sense of the magnitude of these coefficients, I compare Bali, a province with low mining resources, and East Kalimantan, a natural-resource rich province. When using credit, the contribution of the mining sector to growth is 0.004 percentage points in Bali while it is 0.154 percentage points in East Kalimantan. The magnitude then gets bigger when using share of employment where it can impede growth by 0.0231 percentage points in Bali and 0.358 percentage points in East Kalimantan.

Further, when using credit, the predicted growth Bali and East Kalimantan will be 2.983 percentage point and 3.133 percentage point, respectively; and when using employment, the corresponding figures will be 2.891 percentage point and 2.557 percentage point, respectively. In short, the impact of the mining sector on economic growth is larger when using the share of employment compared to the share of credit, and their magnitudes are getting bigger in natural resource-rich province.

Since the sign of employment and credit coefficients are opposite, the predicted growth using employment will be smaller than the one using the credit .

As for the other two mining indicators, the share of domestic direct investment is found to be negatively correlated with growth while the share of revenue sharing has a positive correlation, and both coefficients are small and insignificant. Given these results, Table 2 suggests that different ways of measuring the mining sectors would yield different outcomes when analyzing the significance of the mining industry to the province’s growth.

(24)

24 | P a g e Table 2. Relationship between Mining Industry and Growth (Basic Model)

(1) (2) (3) (4)

VARIABLES

Share of Employment –0.0573**

(0.0255)

Share of Credit 0.0578**

(0.0292)

Share of Direct Domestic –0.00633

Investment (0.0115)

Share of Revenue Sharing 0.00460

(0.00677)

Constant 2.915*** 2.980*** 1.975*** 3.387***

(0.262) (0.334) (0.269) (0.452)

Observations 539 264 173 138

Number of provinces 31 30 27 29

Province RE YES YES YES YES

Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1

In the remaining result section, I will further discuss the share of employment and share of credit as they are both statistically significant. The results of the other two mining indicators, namely the share of direct domestic investment and share for revenue sharing are not statistically significant even when changing the specifications or using different control variables. Therefore, I relegate the results of these two mining indicators to the Appendix A2–3 and A8–10.

4.1 Share of Employment in the Mining Sector

In the first sub-section, I regress the share of employment and growth per capita while adding the control variables, including the interactions. Table 3 shows the regression results when the control variables of education, health, and infrastructures are included. Overall, the results show that the share of employment is negatively correlated with province’s growth except when the model includes household access to electricity. When the immunization coverage is included as the control variable, an increase in the share of employment by one percentage point will decrease the growth per capita by 0.0767 percentage points, statistically significant at 1%. The increase of share of employment also decreases the growth per capita by 0.138 percentage points when all control variables are included, statistically significant at 1%. Taking the same reference provinces as comparison, Column 5 also shows that the predicted growth in Bali is higher than East Kalimantan. On average, the

(25)

25 | P a g e predicted growth in Bali is 4.211 percentage points while it is 3.241 percentage points in East Kalimantan which means there is about one percentage point difference. This is because the share of employment in the mining sector in Bali is very small on average, around 0.403 percent while in East Kalimantan, it is over 6.243 percent. As Bali is a tourist-based province, the negative impact of employment in the mining sector has smaller impact to the province’s growth.

To measure access road as basic infrastructure for mining industry, I introduce villages with asphalt road as the control variable. However, this variable is not significant and costs me around three quarter of the data (Table 3 Column 3).

After including all control variables, this village road variable remains insignificant and reduces the number of the observations further (Column 5). Having these results, I remove the “villages with asphalt road” variable from the regressions for the rest of the result section.

Table 3. The Relationship between Employment in Mining and Growth

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

VARIABLES

Share of Employment –0.0454 –0.0767*** –0.0618 0.00397 –0.137***

(0.0388) (0.0263) (0.0596) (0.0427) (0.0489) Enrollment Ratio at Senior

Level

–0.0511**

(0.0217)

–0.106***

(0.0383)

Immunization Coverage 0.138*** 0.0489

(0.0403) (0.178)

Villages with Road: Asphalt –0.0151

(0.0148)

–0.00850 (0.0348) Household Access to

Electricity

0.00189 (0.0288)

0.0689* (0.0382) Initial GDP per Capita –0.0471 –0.0712* –0.0360 –0.0821 –0.0283 (0.0519) (0.0427) (0.0495) (0.0645) (0.0448)

Constant 5.837*** –8.813** 4.418*** 3.237 0.0667

(1.061) (3.724) (1.020) (2.443) (15.69)

Observations 536 424 178 507 91

Number of provinces 31 31 31 31 31

Province RE YES YES YES YES YES

Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1

As discussed previously, I create interactions between employment in the mining sector with either education or health to see how education and health associate with employment in increasing economic growth. Before interacting them, I center the three variables by subtracting each observation with their respective mean. This

(26)

26 | P a g e centering method is used to simplify and give more meaningful interpretations later on. The results can be seen in Table 4 where education is represented as enrollment ratio at senior level and immunization coverage represents health indicator. When interaction presents, the impact of employment in the mining sector on economic growth is found to always be negative and strong correlation. Take for example in Column 1 where the interaction coefficient shows that if enrollment ratio increases by one percentage point, the impact of employment in the mining sector to growth becomes 0.0248 percentage points smaller, statistically significant at 1%. Thus, an increase in the share of employment by one percentage point, when the enrollment ratio is on average, is associated with a decrease of 0.2638 percentage points in growth per capita, statistically significant at 1%. This level of significance and the sign of impact are consistent for all models in Table 4.

However, the magnitude of the impact tends to become smaller when interaction between education and employment is not present. Column 4 shows the estimates when employment is interacted only with immunization and includes all control variables. The impact of share of employment to growth per capita when immunization is equal to average is now at negative 0.1413 percentage points, statistically significant at 1%. Looking into the immunization coverage data, the provinces with a high share of mining to GDP tend to be higher coverage than the provinces with lower share of mining to GDP. For instance, in Bali and Yogyakarta, two provinces with a very low share of mining to GDP, the immunization coverage is almost over 100 percent on average, 98.53 percent and 99.19 percent, respectively. However, in the provinces with high share of mining to GDP like East Kalimantan and South Sumatera, the immunization coverage on average is slightly lower, 95.39 percent and 93.82 percent, respectively. These differences might explain why the magnitude of share of employment is smaller when it is only interacted with immunization.

(27)

27 | P a g e Table 4. The Relationship between Employment in Mining and Growth

(Interacted with Education and Health)

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

VARIABLES

Share of Employment –0.239*** –0.272*** –0.122*** –0.119*** –0.287***

(0.0859) (0.0950) (0.0429) (0.0435) (0.0950) Enrollment Ratio at Senior –0.00454 –0.0458* –0.0221 –0.0507*

Level (0.0480) (0.0278) (0.0278) (0.0284)

Employment*Enrollment –0.0248*** –0.0252** –0.0286***

(0.00941) (0.0106) (0.0110)

Immunization Coverage 0.00144 0.0531 –0.0189 0.0119

(0.0510) (0.0714) (0.0512) (0.0552)

Employment*Immunization –0.0151 –0.0223 0.0150

(0.0148) (0.0194) (0.0250)

Household Access to 0.0175 0.0187 0.0175

Electricity (0.0443) (0.0444) (0.0443)

Initial GDP per Capita –0.0571 –0.0564 –0.0574 –0.0716 –0.0558 (0.0450) (0.0522) (0.0408) (0.0561) (0.0529)

Constant 3.177*** 1.908 3.698*** 2.059 1.887

(0.349) (3.815) (0.313) (3.847) (3.815)

Observations 536 395 424 395 395

Number of provinces 31 31 31 31 31

Province RE YES YES YES YES YES

Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1

Robustness Check

Table 5 shows the combined regression results of two robustness checks for share of employment and its interactions: removing outliers in growth GDP per capita and using log transformation. Only model that includes all control variables are presented in Table 5 while the rest of regression results are put in the Appendix A4–

A5. First, I compare the results that have no outliers in growth GDP per capita as follows: (i) Table 3 (Column 5) with Column 1 in Table 5 and (ii) Table 4 (Column 5) with Column 2 in Table 5. Overall, removing outliers in growth per capita still generates a negative and strong correlation employment in the mining sector with economic growth. In Column 1, an increase of share of employment by one percentage point is associated with a decrease in growth per capita by 0.113 percentage points, statistically significant at 1%. Similarly, Column 2 shows that a one percentage point increase in share of employment when enrollment is equal to average will decrease growth by 0.3103 percentage points, statistically significant at 1%. The rest of the results are similar with Table 3 and Table 4 which can be seen in Appendix A4. These results suggest that when outliers in economic growth are

(28)

28 | P a g e caused by mining related factors, they do not affect the relationship between the mining industry and economic growth.

Table 5. Robustness Check: The Relationship between Employment in Mining and Growth (No Outliers and Log Transformation)

No Outliers Log Transformation

VARIABLES (1) (2) (3) (4)

Share of Employment –0.113*** –0.282***

(0.0310) (0.0797)

Log Share of Employment –0.157 –0.357*

(0.277) (0.192) Enrollment Ratio at Senior Level –0.130*** –0.0238 –0.147***

(0.0191) (0.0261) (0.0266)

Log Enrollment Ratio at Senior –0.197

Level (1.334)

Employment*Enrollment –0.0283***

(0.00874)

Log Employment*Log Enrollment –1.723***

(0.667)

Immunization Coverage 0.162*** 0.0616 0.0734

(0.0399) (0.0648) (0.0644)

Log Immunization Coverage –0.371

(5.091)

Employment*Immunization 0.00478

(0.0183)

Log Employment*Log Immunization 1.198

(3.929) Household Access to Electricity 0.0346* –0.0311** 0.0939** 0.0129 (0.0189) (0.0124) (0.0428) (0.0424)

Initial GDP per Capita –0.0422 –0.0232 –0.0774 –0.0748

(0.0289) (0.0271) (0.0481) (0.0565)

Constant –7.632** 5.922*** –3.751 2.553

(3.450) (1.178) (3.939) (3.665)

Observations 382 382 395 395

Number of provinces 31 31 31 31

Province RE YES YES YES YES

Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1 Note: This table is a summary of Appendix A4-–A5

The second robustness check is performed by transforming the share of employment into log. First, I compare Table 3 (Column 5) with Column 3 in Table 5. With log transformation, the share of employment becomes insignificant to growth per capita.

The results are still not significant when including the control variables one by one which can be seen in Appendix A5. In the second comparison, between Table 4 (Column 5) and Column 4 in Table 5, the share of employment is still significant but at a lower level. As the share of employment increases by one percent, the growth

(29)

29 | P a g e per capita decreases by 0.00357 percentage points, statistically significant at 10%.

The full results of the model with interaction in the log form are in Appendix A5 where it shows that the estimates are now only significant at 10%. Hence, using the log transformation in the share of employment and its interactions turns the impact on economic growth to become less significant.

4.2 Share of Credit in the Mining Sector

In this sub-section, I look into the relationship between the share of credit in the mining sector with economic growth when including the control variables. The estimation results can be seen in Table 6. The results in general show that the share of credit is positively correlated with growth and statistically significant when either education or electricity present. Unlike previous results when using employment as the mining indicator (Table 3–Table 5), enrollment ratio at senior level, in particular, and now becomes positively correlated with economic growth. When education is included as the control variable, an increase of credit in the mining sector by one percentage point increases the growth per capita by 0.126 percentage points, statistically significant at 5%. The impact becomes more significant when introducing household access to electricity as control variable. An increase of share of credit by one percentage point is associated with an increase of growth per capita by 0.147 percentage points, statistically significant at 1%. The coefficients of enrollment ratio at senior level and household access to electricity also show positive and significant results, implying that these two variables support credit in the mining sector to increase economic growth.

Taking different provinces, I compare Yogyakarta, a province with low mining resources with South Sumatera, a mining resources-rich province to see the magnitude of these estimates in those provinces. Based on Column 1, the predicted growth of South Sumatera and Yogyakarta is –4.829 and –5.892 percentage points, respectively, showing a 1.063 percentage points difference between the two provinces. Looking through the average share of credit on these respective provinces, South Sumatera has 0.22 percent share of credit in the mining sector, slightly higher compared to Yogyakarta, which counts 0.159 percent. It appears that this difference of credits in these respective provinces could explain the higher role

(30)

30 | P a g e of credit in boosting economy in the mining-resource rich province like South Sumatera.

Table 6. The Relationship between Credit in Mining and Growth

(1) (2) (3) (4)

VARIABLES

Share of Credit 0.126** –0.104 0.147*** –0.0537

(0.0522) (0.0638) (0.0438) (0.0486)

Enrollment Ratio at Senior Level 0.0876* 0.0764

(0.0508) (0.104)

Immunization Coverage 0.134 –0.0896

(0.0943) (0.120)

Household Access to Electricity 0.0635* 0.127**

(0.0341) (0.0615)

Initial GDP per Capita –0.149* –0.0866 –0.135 –0.207

(0.0837) (0.0653) (0.0872) (0.137)

Constant 0.413 –8.280 –1.428 –0.819

(1.964) (8.816) (2.826) (6.382)

Observations 256 165 230 139

Number of provinces 30 29 30 29

Province RE YES YES YES YES

Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1

Robustness Check

The regression results of the robustness check are presented in Table 7 where the first two models are the results with no outliers in growth GDP per capita while the latter is when the share of credit is in the log form. The table only includes the results that are significant while the full regression results are in Appendix A6–A7.

In the first robustness check, I compare Table 6 (Column 2 and 4) with Column 1 and 2 in Table 7. Contrary to the estimates in Table 6 where share of credit is negative correlated with economic growth and insignificant as in Colum 2 and 4, the coefficient of credit is now significant only when either immunization or electricity is included as the control variables as in Table 7 Column 1 and 2.

However, in the presence of immunization, an increase of credit in the mining sector by one percentage point is associated with a decrease of growth per capita by 0.092 percentage points, statistically significant at 1%. The immunization itself is also significant and positively correlated with growth. Furthermore, economic growth will decrease by 0.116 percentage points when share of credit increases by one percentage point and all control variables are included, statistically significant at

Figure

Updating...

References

Related subjects :