The mining boom and its impact on the manufacturing and agricultural sectors in Australia

(1)

The mining boom and its impact on the

manufacturing and agricultural sectors in

Australia

Master thesis international economics and globalisation

Niels Doodeman (11893990)

University of Amsterdam

MSc Economics

Amsterdam School of Economics

Supervisor: Dr. C. Sahin

July 10, 2018

(2)

Statement of Originality

This document is written by Niels Doodeman 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.

(3)

1 Introduction

Over the last decade Australia is experiencing a resource boom. High demand for mining products results in hard currency inflows associated with surging resource exports, leading to an appreciation of the real exchange rate. Real exchange rate appreciation will make export products more expensive abroad, while import prices in Australian dollars decline. The real exchange rate appreciation could thus affect both domestic import-competing producers and exporters. Import-import-competing sectors will suffer from competition of cheaper imports, while exporters see a deterioration of their international competitiveness. On the other hand, commodity ex-porters are presumably less affected. Demand elasticity is low for mining products and most commodities are invoiced in US dollars (Corden, 2012).

This thesis will estimate the effect of a resource boom on import-competing and exporting sectors. Australia will be the country of analysis. Australia is experiencing a resource boom if either the level of resource investment as a share of GDP or the terms of trade1are at high levels (Gregory and Sheehan, 2011). Over the 2005-2015 period, Australia had a mining boom in which both resource investment and the terms of trade where at record levels. What is more interesting is that Australia has both booming and declining sectors.

The research question will be: how does the 2005-2015 mining boom affect employment and production in Aus-tralia’s manufacturing and agricultural sector? I will only analyze manufacturing and agriculture because these sectors are typically affected the most by a mining boom: in Australia manufacturing is an import-competing sector, while agriculture is export-oriented. I will only analyze the 2005-2015 boom because the previous booms were different: the current boom is the first in a system of floating exchange rates and the boom last longer than five years, rather than being only a temporary boom.

On forehand the scope of the analysis seems to be quite broad. Therefore, I will ignore geographical and environmental aspects of the mining boom and asset markets. Asset markets are affected by mining booms (Connolly and Orsmond, 2011) but excluded from the analysis, because they only play a marginal role in the determination of sectoral employment and production changes.

The research question will be answered by using both theory and an empirical analysis. The paper starts with a theoretical overview of resource booms and their impact on the economy. Both classical and modern theories will be used. The theory will also be applied to Australia. Potential outcomes and weaknesses of the models will be addressed. The empirical analysis will consist of both descriptive data analyses and regression models. Patterns of structural change in employment and production will be analyzed and compared to non-booming periods.

There are various studies that analyze the impact of resource booms and appreciated currencies on specific sectors. Most papers focus on the tradeables/non-tradeables2 _{distinction, but I will do this differently. I will}

predominantly focus on non-booming tradeables sectors. I will apply Dutch disease theory especially to the non-booming sectors open to international trade (the manufacturing and agricultural sectors). I will show that the non-booming tradeables sectors are more heterogeneous than homogeneous, because the mining boom will affect import-competing and exporting sectors differently. In the academic literature, less attention is paid to the different non-booming tradeables sectors, they are often categorized as one homogeneous sector.

The structure of the paper is the following: the next chapter will provide more information about the resource booms in Australia. The third chapter covers a literature review about resource booms. Thereafter follows an empirical analysis of the impact of the resource boom on manufacturing and agriculture. The thesis ends with the conclusion.

1_{terms of trade is the ratio of export prices to import prices}

(5)

2 What are Dutch disease effects?

The term Dutch disease was originally used by describing the adverse effects of the discovery of gas on the several export and import-competing sectors in The Netherlands. The discovery and the resulting sales of gas led to an appreciated Dutch Guilder (the Dutch national currency before they implemented the Euro). The appreciation incurred several losses in the import- and exporting sectors. ’In case of Australia, ‘Dutch disease’ refers to the adverse effects through real exchange rate appreciation that the mining boom can have on various export-and import-competing industries’ (Corden, 2012, p. 1). The mining boom induces an appreciation of the real exchange rates, thereby affecting the terms of trade and the sectors open to international trade.

2.1 Resource booms in Australia

Australia is experiencing a resources boom if either the level of resource investment as a share of GDP or the terms of trade are at high levels (Gregory and Sheehan, 2011). Both price and volume effects play a role in the 2005-2015 resource boom, which makes the boom unique from previous booms (where not more than one of the two effects was at its record). As mentioned earlier, the main sector causing the resource boom is mining. Interestingly, not the discovery nor the production of mining is the main driver of the boom. The main driver of the 2005-2015 resource boom is Asia’s demand for mining products (Treasury, 2011). According to the report, growth in China and India in particular is considered as one of the main drivers for the boom. The increase in terms of trade, as well as the changes in relative (domestic to foreign) prices will presumably cause structural changes in the economy as competitiveness of many non-booming tradeables sectors is expected to deteriorate.

After the end of World War II there have been three main resource booms: there was a mineral and energy boom in the 1960s/early 1970s, an energy boom in the late 1970s/early 1980s and there is the current boom, which is again both a mineral and energy boom (Battellino, 2010)3. Of all permanent booms, the 2005-2015 boom is the only boom with a floating Australian dollar. This, combined with the different source of the boom (Asian demand) and the surge in mining investment will make the 2005-2015 boom different from the others.

Previous booms in demand for Australian resources were of different origin, namely from developed countries (e.g. Japan). The 2005-2015 boom is also the only boom that survived a global economic crisis. This was due to the fact that China and in particular India were less interconnected to the global economy, which reduced their exposure to global economic vulnerabilities. In one way this contributed to crisis resilience in Australia, since foreign demand for natural resources significantly contributes to capital flows and GDP in Australia. The first and the second boom both ended at the onset of an economic crisis. Terms of trade and the exchange rate went down. Another extraordinary development in the 2005-2015 boom is the swings in the terms of trade. This is mainly due to the large variations in commodity prices like oil (Atkin et al., 2014).

3_{There have also been a number of small booms, but these are considered as temporary, thereby not significantly affecting the}

(6)

Figure 1: Terms of trade and trade-weighted exchange rate, 1990-2015

Australia’s historical terms of trade index and trade weighted exchange rate. Source: Reserve Bank of Australia (RBA). Cat. no. 5302, table 19 and cat.no. 5206, table 01. Interpretation: an index of 100 means that the weighted export price is equal to the weighted import price. The real exchange rate has 100 as starting position (30 June 1970) and is defined as an index of the trade-weighted exchange rate. ToT here is the terms of trade, while RER is the trade-weighted exchange rate.

There are multiple effects of a resource boom on the economy. Gregory and Sheehan (2011) distinguish between higher real incomes and lower import prices, higher income of financial assets, higher mining investment and the competitive impact of the higher Australian dollar. Higher real incomes are the result of lower import prices: this depresses prices. Higher income of financial assets is the result of appreciated asset values and higher share values of (mining) firms. In the remaining of the paper, less attention will be spent to financial assets and mining investment. I will focus on the other effects of a boom.

3 Theories on Dutch disease effects

3.1 Classical theories

The theory on comparative advantages is probably the most well-known classical theory in international trade and the impact of trade on the economy. The theory was originally developed by Hecksher-Ohlin and stated that each country will export the good in which it has a comparative advantage (lower relative production costs). Total welfare will increase compared to the initial situation of closed economies, as both countries pro-duce most efficient and international trade ensures availability of the non-exported products. The sector owning a comparative advantage will expand and sell its products abroad (Feenstra, 2004). The sectors that do not have a comparative advantage will decline and may even disappear because foreign producers are more efficient. Australia, given its abundance in land and its highly productive agricultural sector is assumed to have a com-parative advantage in agriculture, so this sector must increase with international trade, whereas manufacturing should decline (high labour costs worsens the competitive position) (Anderson, 2017). A mining boom can also be analyzed by a model of comparative advantages, as in Leamer (1987). If countries are endowed with a large stock of natural resources, an increase in productive capital (for example by investment or technological

(7)

im-provement) per unit labour will increase the comparative advantage in products from the natural resource sector (because this sector is capital intensive). The country will export the booming sector’s product, at the cost of im-porting the other product (thus employment and production in the non-booming tradeables sectors will decline).

3.1.1 The core model of Dutch disease

The theoretical analysis of Dutch disease effects builds on a large body of research, of which most is written by Australian academics. Prominent papers include Gregory (1976), Corden and Neary (1982) and Corden (1984). They all come up with the same argument: a resource boom induces an appreciation of the real exchange rate, as the resource exports thrive up demand for the domestic currency. There will be two consequences: living standards go up (because of lower import prices; real income increases), but also competitiveness of other ex-porting and import-competing industries weakens. In my paper, I will predominantly focus on the weakened competitiveness of two sectors: agriculture (exporting sector) and manufacturing (import-competing sector). Agricultural prices are given in the world market and have not risen in the way that mining sector prices have. Hence, an exchange rate appreciation lowers prices of agricultural exporters in terms of Australian dollars. Fur-thermore, a real appreciation of the Australian dollar will make domestically produced products less competitive relative to imports.

All the core papers of Dutch disease effects (Corden (1984), Gregory (1976) and Corden and Neary (1982)) have a lot of similarities in their models. The base model used in literature is the core model of Corden and Neary (1982). The model assumes a small open economy producing 3 goods: two goods are traded at given world prices and a non-traded good. The price of the non-traded good is determined by domestic supply and demand forces. There are no market distortions, monetary considerations are ignored (there are only relative prices) and in line with Snape (1977) each of the three sectors uses one specific and one perfectly mobile factor of production. Labour is the mobile factor of production (as to equal wages over the three sectors) and all factor prices are flexible and perfectly mobile. This rules out the possibility of market distortions. The sector-specific factor is assumed to be capital in Corden (1984), Corden and Neary (1982) and Gregory (1976). Capital is sector-specific because is it not capable of moving across sectors. Trade policy (e.g.) tariffs are part of the theoretical models, but these policy options are less relevant today since Australia is part of the WTO and so has not changed tariffs significantly over the last decades. For the above mentioned reasons, tariffs and monetary policy will be neglected in both the theory overview and data analysis

Corden and Neary (1982) and Corden (1984) are proclaimed to be the ’founding fathers’ of Dutch disease the-ory. In this section I will give a theoretical description of the underlying mechanisms, without using all the sophisticated graphs as in Corden and Neary (1982) and Corden (1984). A theoretical review is sufficient in order to describe the process of a resource boom and its impact on the non-booming tradeables sectors. In the initial situation labour supply equals to labour demand. This equilibrium is expressed in the equilibrium wage rate. When there is a boom in the mining sector, profitability of hiring labour and thus labour demand goes up. As a result mining wages will go up at a given exchange rate (assuming the boom to have no effect on the supply side of labour). Labour responds to the higher wages in the mining sector by moving from the agricultural and manufacturing sectors to the mining sector. Employment falls in manufacturing and agriculture, thereby con-tributing to de-industrialization (production factors move from the non-booming sectors to the booming sector).

A distinction can be made between two effects of a resource boom, as in Corden (1984), Corden and Neary (1982) and Corden (2012): the resource movement effect and the spending effect. The spending effect reflects the changes in the output of different industries within the economy as the higher commodity export earnings are spent. Demand for non-tradeables and non-booming tradeables increases. Assuming prices of tradeables to be less elastic (an increase in national income has a larger impact on domestic demand than on global demand), this increase in demand raises the relative price of non-tradeables to tradeables (i.e. the real exchange rate ap-preciates). Thus the spending effect will be a second source of appreciation, which occurs after the start of the boom. In the resource movement effect a resource boom raises the aggregate prices and subsequently incomes of the production factors employed in the booming sector (higher marginal products of labour and capital). This attracts capital and labour from the other sectors. In case of Australia capital and labour move from agriculture and manufacturing to mining, thereby contributing to divergence in growth between the sectors. The resource

(8)

movement effect has two kinds of effects: direct and indirect de-industrialization.

In direct de-industrialization capital and labour moves from the non-booming tradeables to the booming sector, aiming for higher returns. Direct industrialization does not involve the market for non-tradeables and thus does not require an appreciation of the real exchange rate. The movement of production factors from the non-booming tradeables to the non-tradeables sector is called indirect de-industrialization. The increased demand for non-tradeables (driven by the spending effect) and the shift of production factors from the non-tradeables to the booming sector bring about an additional movement of labour out of the non-booming tradeables sector into the non-tradeables, reinforcing the indirect de-industrialization resulting from the spending effect. This is what happens: at first production factors aim for higher return and move to the booming sector. However, the spend-ing effects triggers demand for non-tradeables. The returns to production factors in the non-tradeables sector go up relative to the non-booming tradeables sector. Labour and capital will move from the non-booming tradeables sectors to non-tradeables. This is indirect industrialization, which supplements the direct de-industrialization. The theory review makes the impact of a boom on employment clear: employment in the manufacturing and agricultural sectors is projected to go down as labour moves to other sectors. The non-booming tradeables sectors decline when there is a resource movement and/or spending effect.

Even though the classical model is generally accepted and well-cited, a point of criticism is made by Snape (2011). Classical theories mention that all tradeables sectors other than the booming sector will decline after a resource boom. Some exceptions can however exist. The production of other tradeable goods could also in-crease with a resource boom, when the inputs of the booming sector are not a function of the factors other than the mineral resource itself. In this situation the resource movement effect is minimal and the other tradeables sectors are minimally affected by the boom. The extent of the spending effect depends on production changes, and thus relative price changes (but now determined by the supply side).

3.1.2 International capital mobility

Until now we have only analyzed an economy in which we assumed capital flows to be domestic. Suppose now that there is international capital mobility, as in one of Corden’s (1984) modification to the core model. We will see in the theory chapter about Australia that this is relevant for the analysis. In the modified model each sector employs sector-specific capital, but each of the three types of capital is to some extent internationally mobile. Thus (domestic) capital does not move between industries, but across borders. In a boom, capital rents rise in the booming sector and decline in the non-booming tradeables sector. Higher demand for products triggers production, which increases profitability of investment in the booming sector. Rents in non-tradeables could both rise and decline, this depends on the sector’s output and the impact of the spending effect. In-ternational capital mobility will cause capital inflow in the booming sector, thereby positively contributing to output changes (investment will increase). The non-booming tradeables sector will experience an (international) outflow of capital, since expected profitability decreases (and thus investment will go down). As a result, the de-industrialization effect will be even stronger, as output in the booming sector increases at the cost of the non-booming sectors.

3.2 Theories on Australia

Using general theories might give some discrepancies in analyzing a resource boom on the Australian economy. The usual pattern of structural change in growing economies is for the primary sector’s shares of GDP and employment to diminish as the industrial sector expands and for manufacturing to subsequently diminish as service sectors increasingly dominate the economy (Anderson, 2017). However, in Australia this effect is some-what different. Fluctuations in the terms of trade and large discoveries of minerals have distorted the process of structural change. By exporting more resources, terms of trade increases. This leads to a heavier decline in manufacturing relative to other developed countries. Over the last century the agricultural sector continued to have a strong comparative advantage, despite the resource export booms.

(9)

3.2.1 Application of trade theory to Australia’s Dutch disease effects

Using trade theory, a prediction could be made how the sectoral shares of manufacturing and agriculture have evolved over time. Agriculture is assumed to fall over time as the economy develops. Domestic prices and quantities of agriculture relative to other products will fall: income elasticity of demand for food is low and agricultural production is enhancing fast due to technological improvement (Anderson, 1987). In an open econ-omy as Australia, this development is less obvious. Australia has abundant land relative to labour and capital, leading to high productivity per worker. Interesting about agriculture is also that the shares in export are not necessarily falling if the world economy expands. This effect only works on a national scale. In global markets, agricultural exports depend on comparative advantages (Venables, 2004). Low trade costs and technological improvement strengthen comparative advantages. Both will keep production costs low. So a large agricultural sector in a developed country could exist, conditional on a strong comparative advantage in which the sector is export oriented, rather than producing for domestic consumption. This is especially relevant to Australia, since agriculture’s share in GDP and remained quite constant over the last decades (Downes et al., 2014). According to Anderson, it is obvious that Australia’s agricultural sector is larger compared to other countries. In order to be competitive, agriculture needed to be labour-saving (thus capital intensive) and innovative. As this was achieved, the comparative advantage was maintained, despite the appreciating Australian dollar. The opposite is true for manufacturing, which is more labour intensive and less innovative, and thus less competitive. Manufac-turing has declined significantly after a reduction in tariffs. Neither has manufacManufac-turing a comparative advantage (Anderson, 2017). Lower manufacturing (import) tariffs have increased foreign competition (and thus imports) in the Australian economy. The relative scarcity of labour (with its subsequent high labour costs) has limited the ability of manufacturing producers to compete with foreign, lower wage producers. As a result the man-ufacturing share in total production in Australia is smaller than in other resource abundant developed countries.

3.2.2 Dutch disease theory on Australia

Regarding traditional Dutch disease theory, Australia is also different. The resource movement effect is smaller in Australia than what theory would predict. When labour and capital are less mobile, capital and labour short-ages in the non-booming sectors are smaller after a boom, meaning that de-industrialization and subsequent output decline in non-booming tradeables sectors is smaller.

As mentioned in Corden (2012) the resource movement effect is smaller in Australia for to three reasons: labour movement is restricted by immigration policy (skilled immigrants can work directly only in the booming sector), mining employment is small relative to total employment and the movement of capital across sectors is replaced by high international capital mobility. Immigration policy limits labour outflow from the non-booming sectors, since the booming sector’s labour shortages are filled up by migrant workers. The fact that the mining sector is relatively small (on average 1.3 percent of total employment over the 1990-2015 period, see Table 1) means that there is less of a structural change in employment and capital. However, employment does move to the min-ing sector, but this concerns mainly specialized labour. International capital mobility reduces national capital mobility (national capital also has a factor specific constraint, as in Corden and Neary (1982)). International capital functions as a complement of domestic capital (Gregory and Sheehan (2011)), predominantly in the sense that most of the international capital is invested in the booming sectors (assuming higher rates of return in the booming sectors). The increased inflow of international capital automatically reduces capital rents in booming sectors, thereby reducing the movement of domestic capital from non-booming to booming sectors. As such, international capital mobility prevents mobile domestic capital from leaving the non-booming sectors. This is different from the findings in the core model of Corden (1984), who also assumes not only a strong capital inflow in the booming sector, but also a strong capital outflow in the non-booming tradeables sectors. Gregory and Sheehan (2011) find that most capital in the Australian non-booming tradeables sectors is sector specific and invested in long-term projects (and thus less mobile), which mitigates domestic capital outflow from the non-booming tradeables sectors.

These national characteristics do not make the existing theories less helpful in analyzing the impact of a resource boom on non-booming tradeables sectors, it only makes it harder to come up with the accurate results. The main signs of the resource movement and the spending effect are known, however the magnitude differs depends on several variables. As indicated by Connelly and Orsmond (2011) the resource movement effect is smaller the

(10)

easier the ability to import mining related labour and capital. Imported capital will prevent domestic capital from moving to booming sectors. Some domestic capital is also sector-specific, which makes capital less mobile and less likely to move to other sectors. Corden (2012) confirms the findings of Connelly and Orsmond (2011) considering labour migration. International labour migration is only allowed in the booming sectors, which limits labour movement from non-booming to the booming sectors. He furthermore finds that the spending effect is smaller the larger the share of mining industry is owned abroad, because profits are more likely to be spent or saved abroad. According to Connolly and Orsmond (2011) 80% of the companies in Australia’s mining industry are owned abroad. Consequently, large shares of profits will be shifted abroad. The resulting spending effect will be smaller, since more money will be spend abroad rather then consumed or invested in the domestic economy. Demand for domestically produced products (non-tradeables) will be lower, which mitigates the real appreciation. The second effect is also related to an appreciation: the fact that foreign owners of mining companies will move their profits abroad mitigates the real appreciation. Supply of the Australian dollar will go up, resulting in downward pressure on the currency.

3.3 Main empirical findings in the literature

There is a lot of literature about Dutch disease effects in Australia. Most papers do not use regression analyses. The data analyses are limited to descriptive statistics and graphs. Conclusions about the impact of the boom on the agricultural and manufacturing sector are mixed: this depends among other things on various factors as data selection, incorporated effects, analyzed time period etc.

Several models have been used to analyze the impact of a boom in Australia, as in Anderson (2017) and Downes et al. (2014) These papers build further on the models of Neary and Corden (1982) and Corden (1984), but make some adjustments. For example Downes et al. (2014) use a large structural model of the Australian economy. The model assumes that most output is determined by demand in the short run, but with some important exceptions (for example investment). They compared two scenarios: one in which there was no min-ing boom and one in which there was a minmin-ing boom. They used a large-scale structural econometric model, examining the sectoral and regional impacts of a transition to a higher terms of trade. Regressions are based on a 10-equation vector auto regressive model. The overall impact of the boom on the economy is positive: compared to a situation of no boom wages and disposable income raised. In 2015 mining investment raised with 3% of GDP and total unemployment is 1.3% lower than what it would have been without the boom. Export volumes show the same as what was predicted in theory: agricultural and manufacturing export declined (even though this effect was relatively small for agriculture), while mining export rose. Most interesting to see is the decline in industry output: the impact is larger for agriculture (12%) than for manufacturing (7%, 2015 prediction based on 2014 data). For employment these effects are smaller. Employment is less sensitive to the mining boom than production. The impact on manufacturing is smaller than what theory would predict. Manufacturing loses relatively to agriculture more export value, but less output. The non-tradeables part of manufacturing therefore increases relative to exports, otherwise output decline would be the same or larger as export decline. Manufacturing benefits from higher demand for mining inputs, these inputs are non-tradeables. So for manufacturing, the resource movement effect is smaller. This is interesting: Australia’s manufacturing sector is declining more than those of other countries, but their resource movement effect of a boom is smaller. This way, a boom could contribute to a smaller decline of the manufacturing sector after a boom than what would have happened when inputs were bought abroad.

Most of the papers however keep their analysis limited to descriptive data analysis. Well-cited papers are Ander-son (2017), McKissak et al. (2008) and Plumb et al. (2013). Even though most of the papers do not incorporate the 2012-2015 period, the papers give a clear insight in the dynamics of a boom. Plumb et al. (2013) described the boom and its consequences. The onset of a resource boom is a resource demand shock. What follows is a terms of trade shock. Higher investment in the resource sector will trigger resource investment. The result of this is a movement of capital and labour to the resource sector. Again the total effect on the economy is positive, but there are clearly losing sectors: these sectors are classifies as non-booming tradeables. Anderson (2017), Banks (2011), McKissak et al. (2008) and The Australian Treasury (2011) thoroughly analyzed the non-booming tradeables sectors. The higher terms of trade will move the economy to a knowledge based ser-vices and resource economy. These are the most competitive sectors. Agriculture is in some papers declining,

(11)

but not in all. Methodological differences are at the heart of these inconsistencies (e.g. measuring output by value added or by income of production factors etc). In some agricultural sub sectors (but obviously not all) productivity and technology improvements outweigh terms of trade deterioration, which enables the sector to grow.

4 Data and methodology

In this section I will analyze the Dutch disease effects on manufacturing and agriculture empirically. I will build further on the theory, as I will analyze Dutch disease effects empirically. The data analysis will start with a graphical illustration, where after follows the regression analysis. Both the data and the regression analysis focus on answering the research question: estimating the impact of a mining boom on the agricultural and manufacturing sectors. The data analysis is used to give an indication about structural change and the impact of the mining boom. The regression analysis will be used to directly estimate the impact of the boom on production and employment in the manufacturing and agricultural sector.

Most empirical analyses on Dutch disease do not include regressions. Descriptive data analyses are supported by theory. This widely used method provides a clear insight in structural change, but there is no direct estimation of Dutch disease effects on the non-booming tradeables sectors. This thesis’ research design will be different: there will be a regression analysis, since the focus is more on the direct effects of the mining boom on non-booming tradeables sectors, instead of the whole economy (as in most literature). Providing only a descriptive data analysis will make this empirical analysis incomplete, since it provides insufficient information to answer the research question. Regression analysis is also a good way to estimate the impact of the resource movement effect in Australia empirically, which enables me to critically analyze the literature findings. Furthermore, this empirical analysis also incorporates the pre-booming period. The analysis of structural change is more accurate when the pre-booming period is included because it is possible to compare the pre boom and booming period. Some papers only analyze structural change in the booming period, without taking into account the pre-boom years.

The regression analysis will consist of two regression models. The first model uses the difference-in-difference method in which the impact of the mining boom on the agricultural and manufacturing sectors will be analyzed. The second model is a panel data model which directly estimates the effects of mining production on agricultural and manufacturing production and employment. Special attention is given to imports (for manufacturing) and exports (for agriculture), since the tradeable components of these sectors (relative to the non-tradeables) are assumed to be more affected by the boom.

The resource movement will be analyzed by employment in- and outflow in the manufacturing and agricultural sectors. In the resource movement effect capital will be ignored, since Corden (2012) already proved that the capital movement effect is limited in Australia due to international capital mobility. The spending effect is not analyzed thoroughly, besides that price ratios and the exchange rate are included as control variables. Using these variables as controls means that they have no causal interpretation, but it is important to have these effects incorporated in the model since they play a vital role in literature.

After the difference-in-difference model follows a panel data analysis. The difference-in-difference estimator will give a clear insight in the impact of the mining boom on the different sectors. Absolute growth divergence will be present between the booming and non-booming sectors. In the panel data regression the impact of the mining boom will be analyzed further. The focus lies on the sectoral composition (tradeables versus non-tradeables) and international competitiveness.

4.1 Data

All data is retrieved from the Australian Bureau of Statistics. Data is quarterly and contains the 1990-2015 period. In the regression analysis models I have used the following variables:

(12)

Table 1: Summary statistics

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

VARIABLES N mean sd min max Classification Total emp 104 9,482 1,384 7,531 11,958 In thousands Non-mining emp 104 9,351 1,329 7,449 11,745 In thousands

Agriculture emp 104 0.0411 0.00962 0.0252 0.0587 Share in non-mining emp Agriculture emp 104 372.4 41.81 280.1 453.7 In thousands

Manufacturing emp 104 0.112 0.0208 0.0730 0.149 Share in non-mining emp Manufacturing emp 104 1,022 57.60 857.9 1,144 In thousands

Mining emp 104 0.0133 0.00447 0.00845 0.0241 Share in total emp Mining emp 104 130.9 62.80 75.39 274.4 In thousands Total gva 104 266,298 65,869 172,026 384,311 In millions Non-mining gva 104 252,987 61,862 163,582 359,971 In millions Agriculture gva 104 8,354 1,737 5,255 10,951 In millions

Agriculture gva 104 0.0334 0.00290 0.0271 0.0384 Share in non-mining gva Manufacturing gva 104 0.103 0.0167 0.0700 0.129 Share in non-mining gva Manufacturing gva 104 25,002 2,437 20,178 29,099 In millions

Mining gva 104 0.0494 0.00427 0.0436 0.0633 Share in total gva Mining gva 104 13,311 4,216 7,905 24,340 In millions

Agriculture Export 104 0.0880 0.0222 0.0448 0.154 Share in sectoral gva Manufacturing Import 104 0.420 0.177 0.173 0.872 Share in sectoral gva

RER 104 61.07 8.063 49.30 78.10 Trade Weighted Exchange rate Price Index 104 0.659 0.174 0.475 1.064 Export/Import price

Summary statistics of the data set. Data is quarterly and contains the 1990-2015 period. emp is an abbreviation for employment. Source: Australian Bureau of Statistics (ABS). ABS cat. no. 5204.0, table 05; 5368.0, table 32A, 32b, 35A and 35B & 6291.0.55.003, table 04. Employment is measured in thousands, while gross value added is measured in millions. Gross value added and employment are measured for all the 3 sectors. There is also a total in employment and gross value added: the sum of all sectors (either in employment or gross value added). The Australian Bureau of Statistics measures production by counting up the gross value added (gva). RER is the real exchange rate and price index is the weighted export price divided by the import price These statistics are also used to plot the figures in the following subsections.

Manufacturing and agricultural production and employment are measured as share in total non-mining gva and employment. Calculating these variables as share in total production/employment will cause endogeneity. If mining gva increases, total gva goes automatically up too, which will decline manufacturing and agricultural shares in total gva.

On average manufacturing is the sector with the highest employment (share), followed by agriculture and min-ing. There are however some differences visible over time, given the relatively large standard deviations and the below figures. Manufacturing is also the factor with the largest gva (share). However their average production share is smaller than their employment share. The same applies to agriculture. Mining’s average gva share (in total gva) is way larger than their employment share. This could imply labour efficient (or labour scarce) production and it could also be an indication that the resource movement effect is relatively small, as indicated by Corden (2012): the labour movement to the mining sector is small because its employment share is relatively small.

4.2 Descriptive data analysis

4.2.1 Employment effects

Over the last decades, sectoral shares of employment have changed. Australia’s labour force composition is different than what general theory would predict. National characteristics like a comparative advantage in

(13)

agri-culture, a vulnerable manufacturing sector and a booming mining sector all have their impact on the sectoral composition of employment.

Figure 2 shows the sectoral shares of employment over the period 1990-2015. Manufacturing has declined over this period from 14% to about 7%. The trend in employment was already declining, but seems to accelerate lightly during the resource boom. Agriculture however is different: the sectoral share remained fairly constant until 2003, where after it dropped to 4%. The 2005 mining boom increased mining’s share in total employment. More labour was attracted from other sectors to mining, which could be a sign of the resource movement effect. However, based on Figure 2 the resource movement effect can not be attributed to the manufacturing and agricultural sector, since the trend in these sectors is not heavily affected in the booming period 2005-2015.

Figure 2: Shares of sectoral employment in total employment, 1990-2015

Australia’s employment shares of manufacturing, mining and agriculture. Source: Australian Bureau of Statistics (ABS). Cat no. 6291.0.55.003, table 04. Sectoral shares are based on own calculations.

Figure 3 shows the growth figures in employment per sector. On the left panel the trend in employment in the agricultural and manufacturing sectors seems again not heavily affect affected by the mining boom. Agriculture and manufacturing are declining. In manufacturing total employment decline in the boom is about 300 thou-sand in total, while this is 40 thouthou-sand in agriculture. Their growth is lagging behind the other sectors. The orange line, which shows total employment growth, shows an upward trend. Growth in total employment did not fluctuate much.

However, other factors of influence may also have played a role here, like the global financial crisis. Mining is booming. In the right panel we take into account only the ten years before the boom (1995-2004) and the in the literature mentioned years of the boom (2005-2015). According to the Australian Bureau of Statistics a sector is defined as tradeables if a threshold of 10% holds: if export and/or import as percentage of total production is larger than 10%. The analyzed tradeables sectors are mining, agriculture and manufacturing, while financial and insurance services, construction and retail trade are seen as non-tradeables. Non-market (public) sectors are irrelevant and thus excluded. The difference with the left panel is that it ignores the 1990-1995 period. This makes it more relevant to take a look at the impact of the boom, since it compares roughly the same period. There is no break-down of the trend visible in agriculture and manufacturing: pre-boom average annual growth

(14)

growth about the same as growth during the boom. Average annual manufacturing growth is 0.35 percent point lower during the boom, whereas agricultural growth is 0.49 percent point lower. Considering the spending effect, growth in the non-tradeables sector is not clearly increasing in the booming period, since there is no higher annual growth in financial retail trade and financial and insurance services after the boom. Mining growth obviously increased in the boom, with on average 5.06 percent point annually, compared to 2.4% on average before the boom.

Figure 3: Employment growth per sector.

Left: Growth of employment in the agricultural, manufacturing and mining sector. Employment on 1 January 1990 is indexed at 100. Interpretation: an index of 150 means an increase in employment of 50% compared to 1 February 1990. Right: Average annual percentage growth of sectoral employment in the ten years before and the period during the boom. Source: Australian Bureau of Statistics (ABS). Cat no. 6291.0.55.003, table 04. Based on original values. Growth figures based on own calculations.

4.2.2 Production effects

Besides the employment effects, production effects of the boom are also plotted. Theoretically the impact on employment is somewhat lagging behind production developments, since employment depends largely on production. Employment can also be seen as rigid on the short-term because employers and employees often establish fixed contracts. Factors like labour-saving technological progress also play a role, as mentioned by An-derson (2017). Production is measured by the quarterly gross value added per sector, as done by the Australian Bureau of Statistics.

Figure 4 shows the sectoral shares of production over time. Again manufacturing is declining, but this decline is now even stronger during the boom. The trend in agriculture remained constant, despite the large periodic swings in production. The trend during the boom is largely the same: agriculture did not decline much. This is different from agricultural employment, which declined over time. Mining production is again steeply increasing during the boom, but there are some swings in the production shares from 2010. Interesting to see is the 2013-2015 period: mining production increased, while mining employment declined. In 2013-2015 mining production was almost as large as manufacturing production. The production share of mining is way larger than its employment share: in some booming years the difference was close to 4 percentage points. This means indirectly that mining is less labour intensive than other sectors.

(15)

Figure 4: Shares of sectoral production in total production, 1990-2015

Australia’s production shares of manufacturing, mining and agriculture. Source: Australian Bureau of Statistics (ABS). ABS cat. no. 5204.0, table 05. Output calculations use original, chain volume measures data. Sectoral shares based on own calculations

Production growth per sector is shown in Figure 5. Figure 5 again has a left panel and a right panel. The trend in manufacturing production is increasing before the boom and flattens during the boom. This is different from agriculture, which grew after the boom (despite the large fluctuations from the trend) but at a lower pace. Compared to total production growth these sectors are lagging behind. Total production growth, especially in the boom, is higher. This is also visible in the right panel. Remarkable to see is the growth in construction: it is assumed to benefit from both pre-boom investment and booming production (McKissak et. al, 2008). Finally, the spending effect is also considered to be small: growth in the non-tradeables sectors (financial and insurance services, construction and retail trade) did not increase after the boom, which could indicate lower growth of the domestic to national price level (assuming that economic growth contributes to higher prices) and thereby mitigates further real appreciation.

(16)

Figure 5: Production growth per sector

Left: Growth of production in the agricultural, manufacturing and mining sector. Production on 1 February 1990 is indexed at 100. Interpretation: an index of 150 means an increase in production of 50% compared to 1 February 1990. Right: Average annual percentage growth of sectoral production in the ten years before and the period during the boom. Source: Australian Bureau of Statistics (ABS).

4.2.3 International competitiveness

Interesting to take into account are also the external demand factors. Export is probably affected the most by an appreciation of the currency. Figure 6 shows the export statistics of the agricultural and manufacturing sector, as well as the total export. Australian exports have risen more than threefold in the last twenty-five years. The main reason for the rise in exports is the move towards trade liberalization and the accompanied trade tariff reduction. This has led to a sharp increase in export volumes.

Except the 2008-2015 period, total exports have risen almost every year over the last twenty five years. Agri-culture is however scoring lower than average. The fact that agriAgri-culture is an export-oriented sector makes the export figures even more valuable. The dip between 2007 and 2009 in exports is also seen in the agricultural production. Agriculture is as export sector largely dependent on international economic conditions. This could be the main reason why agriculture saw a drop in production during the global financial crisis: foreign demand imploded. The right panel shows imports: manufacturing imports clearly increased during the boom. The spending effect could have played a role here: higher wealth induces higher spending.

For manufacturing, export is not as relevant as it is for agriculture. Manufacturing is seen as an import-competing sector. The export share of output is far smaller than for example agriculture, which makes the export channel less important to manufacturing. Manufacturing imports seemed to be co-integrated with total imports over the 1990-2005 period. This is because manufacturing imports were about 90% of the total imports over this period. Manufacturing imports clearly increased during the boom, which is according to theory, since imports were becoming more competitive relative to the import-competing manufacturing sector. A break-down is again visible in the years 2007 and 2008, which would have been affected by the global financial crisis.

(17)

Figure 6: Export and Import growth per product type

Left: Growth of product exports in the agricultural, manufacturing and mining sector. Export on 1 January 1990 is indexed at 100. Interpretation: an index of 150 means an increase in production of 50% compared to 1 January 1990. Right: Average annual percentage growth of sectoral production in the ten years before and the period during the boom. Data is provided on a monthly basis. Source: Australian Bureau of Statistics (ABS). Cat. no 5368, tables 32a, 32b, 35a & 35b.

All in all there are quite some differences between production and employment effects. Agricultural employment declined slightly over the 1990-2015 period, while production increased at almost the average level of all sectors. Manufacturing employment also declined, even more than agriculture. Production increased, but at a slower pace than agriculture. From this we can conclude that agriculture is less declining than manufacturing, also in the boom.

Interesting to take into account is also the composition of labour in the sectors. In all the three relevant sectors (mining, manufacturing and agriculture) the labour share of production has declined (or grown slower), while production increased (faster). This indirectly means that productivity per worker has increased. As a result firms could be more competitive (sell products at lower prices), which could outweigh the disadvantage abroad of having an appreciated currency.

Finally, The spending effect seems to be fairly low as growth in both employment and production in the non-tradeables lagged behind. There could be some evidence for the resource movement effect, as labour moved away from the agricultural and manufacturing sectors and labour inflow increased in the booming sector. Cap-ital mobility is not analyzed, since it plays only a tiny role in Australia (Corden, 2012). More on the resource movement effect of labour in the regression analysis.

4.3 Difference-in-difference regression

4.3.1 Methodology

In the first regression analysis I will analyze the difference in growth between the sectors. As shown in sections 4.2 and 4.3, the mining sector is growing at a higher pace than the agricultural and manufacturing sectors. In this subsection I will use a difference-in-difference model in order to analyze the growth divergence between sectors, both before and during the boom.

The difference-in-difference model enables to distinguish between groups in a sample. The model is used to analyze trends over time per group during the boom. Usually there is a linear trend over time (before the boom). The key in this analysis is a changing variable at a certain moment. In this model the key variable is the boom. The trend in production of the analyzed sectors could differ during the boom, as it was assumed

(18)

to be the same before the boom. I am interested in this effect: the change in the production trend per sector during the boom, which enables me to go into deeper detail about the effect of the mining boom on the different sectors. The difference-in-difference model makes a distinction between two groups, a treatment group and a control group. The treatment group is affected by the boom, whereas the control group is not (corrected for exchange rate). Mining is in this model the control group. Manufacturing and agricultural production will be affected by the boom and so they are the treatment group. The model will be used to test the effects of the 2005-2015 mining boom boom on sectoral production. After all the most competitive sectors are more likely to grow than the by the boom affected sectors (mining exports will deteriorate their competitive position in international markets).

Sections 4.2 and 4.3 showed that mining production and employment have steeply increased during the boom, while agriculture and manufacturing have not. Manufacturing and agriculture are treatment groups. The im-pact of the mining boom is obviously different for these non-booming sectors. The difference-in-difference model will be used not only to estimate the impact of the mining boom per sector, but also to explain why some sectors grew different than others.

4.3.2 Validness of difference-in-difference approach

Figure 7 is used to show the relevance of the difference-in-difference model. The Figure shows that from 2005 mining exports exploded. Mining exports increased to over 15 fold during the boom. From the Figure it is also clear that 2005 is used as the starting point of the boom. Before the boom there is a small upward trend. During the boom there is no trend, for example during 2008-2010 mining export growth declined, presumable as a con-sequence of the global financial crisis. In order to still have a correct model, (booming) time fixed effects are used.

Figure 7: Mining export growth over time, 1990-2015

Australia’s exports of mining products. Mining exports on 1 January 1990 are indexed at 100. Source: Australian Bureau of Statistics (ABS). ABS cat. no. 5368.0, table 32A and 32B.

In order to be able to have a working difference-in-difference model, there has to be a control group in which there was no effect of the boom. The mining boom clearly increased mining production (as seen in Figures 4 and 5), but corrected for the exchange rate the boom effect is not significantly different from zero. This is showed in Table 2, where the following regression models are used:

(19)

∆Mining gva_t= α0+ α1∗ ∆Mining Exportt+ σ1∗ Real Exchange Ratet+ t (1)

and

∆Mining employment_t= α0+ α1∗ ∆Mining Exportt+ σ1∗ Real Exchange Ratet+ t (2)

The delta sign indicates that first differences are applied4_{. Normally demeaned variables are preferred in pooled}

OLS, but in this regression I was especially interested in the effect of (a) unit(s) change in export (over time) and its effect on mining employment and gva. This is after all exactly what happens in the mining boom. Furthermore, the difference between first difference and demeaned variables are very small and using within estimated will not lead to a different conclusion 5_{. Mining export is a proxy for foreign mining demand. The}

real exchange rate is used to control for mining production: because of the appreciated currency the effect of the boom on mining output is not significantly different from 0 (see column (1) in Figure 2). Column (2) shows that the difference-in-difference model is not applicable to employment since the coefficient is, even though it is smaller, significantly different from 0. The real exchange rate is also insignificant in the mining employment equation. Furthermore, Figure 3 also shows that there is no common trend in employment between the three sectors of analysis in the pre-booming period 1990-2005 (which is a prerequisite for an unbiased difference-in-difference model). The main conclusion that could be drawn is that mining demand, corrected for the exchange rate, has no impact on mining production. This means, in combination with a valid common trend assumption, that mining production (again measured by gva) is applicable to the difference-in-difference model.

Table 2: Regression analysis of relevance export boom on export production

(1) (2)

VARIABLES ∆ Mining gva ∆ Mining employment

∆ Mining export 0.00648 0.00288** (0.0195) (0.00115) RER 10.94*** 0.158 (2.345) (0.109) Constant -509.5*** -8.824 (135.3) (6.219) Observations 103 103 R-squared 0.207 0.142

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

Regression analysis of the effect of the mining demand boom on mining production. RER is the real exchange rate. Source: Australian Bureau of Statistics (ABS). ABS cat. no. 5204.0 table 05, 6291.0.55.003 table 04 & 5302.0 table 19.

At first sight the common trend assumption in manufacturing gva seems to be invalid (especially if one considers the left panel of Figure 5). However, figure 5 is not corrected for the real exchange rate. A regression for the manufacturing and mining gva is performed to show that the conditional common trend condition holds. The equation is the following:

∆Manufacturing gvat= α0+ α1∗ ∆Mining gvat+ α2∗ Yeart+ α3∗ Mining gva*Yeart+ σ1∗ RERt+ t

(3)

4_{Dickey-Fuller test statistics showed non-stationarity, t=-0.297 for mining employment, 2.246 for mining gva and -2.261 for}

mining export. The critical value is -2.958, so there is non-stationarity.

5_{There is no need here to use yearly fixed effects: this will filter out the yearly effect of the boom in exports on gva (which is}

(20)

The regression concerns the pre-booming period, 1990-2005 6. The coefficient of the interaction term is of in-terest, since this variable measures whether the trend in mining has an impact on manufacturing production. If this is the case, mining gva will affect manufacturing differently over time, and so the common trend assumption will be violated. The output is presented in Table 3:

Table 3: Proof of validness of the common trend assumption of manufacturing (1)

VARIABLES ∆ Manufacturing gva

∆ Mining gva 0.115 (0.189) Year*Mining gva -1.48e-05

(3.52e-05) Year 18.78 (24.76) RER -21.27*** (5.420) Constant -35,937 (48,729) Observations 59 R-squared 0.222

Proof of validness of the common trend assumption of manufacturing. Data is quarterly and contains the 1990-2005 period, pre-boom period. Source: Australian Bureau of Statistics (ABS). ABS cat. no. 5204.0, table 05.

Table 3 shows that the real exchange rate (RER) is an important predictor of the model. The coefficient is highly significant. Where I am interested in is the coefficient of the interaction term, Year*Mining gva. This variable is an interaction term of mining gva and the year. This coefficient provides an estimate of the impact of the trend in mining on (the first difference of) manufacturing gva. So this coefficient will show whether the mining trend affects (or is different from) manufacturing gva. The coefficient is not significantly different from 0, indicating that the mining production trend does not affect manufacturing gva. Therefore, there is no different trend between manufacturing and mining gva before the boom and thus the common trend assumption is still valid. The interaction term shows that the pre-booming trend in mining has no impact on manufacturing production differences over time.

Now let’s discuss the real difference-in-difference models. The difference-in-difference model is used for both manufacturing and agriculture. The model is described as:

Gvat= α0+ α1∗ Agriculture + δ1∗ Year*It+ δ2∗ Agriculture*Year*It+ σ1∗ RERt+ t (4)

and:

Gvat= α0+ α1∗ Manufacturing + δ1∗ Year*It+ δ2∗ Manufacturing*Year*It+ σ1∗ RERt+ t (5)

The dependent variable is total gross value added (gva) in millions7_{. In the first difference-in-difference model}

only mining and agriculture are analyzed. The second equation uses manufacturing and mining as sectors of

6_{I have again used first differences to control for serially correlation and robust standard errors are used.}

7_{I did not use gva as share values, because this will cause endogeneity. In the second and third regression analyses (subsections}

(21)

analysis8. HAC standard errors correct for serially correlated and heteroskedasticity consistent standard errors. It allows for serial correlation within groups (the sectors) but not between them. The variables Manufacturing and Agriculture are dummies, which are 1 if it concerns the sector. α0is the estimated gva of the mining sector

in the year 1990. α1 is the estimated difference of gva between agriculture (or manufacturing in the second

regression) and mining. δ1 is the trend for mining during the boom. δ2 provides an estimate of the trend for

agriculture (or manufacturing) during the boom. I is a boom dummy, which is 1 if year ≥ 2005 is true and 0 -if it concerns the period before the boom. Where I am interested in is the effect of the boom on manufacturing and agricultural output, δ2.

Based on theory and the data-analysis in the previous subsections, I would expect both α1> 0 in (1) and (2):

in 1990 agricultural and manufacturing gva was larger than mining gva. For the same reason, I would expect δ1 > 0, since mining gva clearly grows in the boom. The coefficient of δ2 is unclear for both agriculture and

manufacturing. In Figure 5 these sector’s gva increase in the boom, but Figure 5 is not corrected for the real exchange rate. The exchange rate coefficient is assumed to be negative in theory: an appreciating currency will lower both manufacturing and agricultural gva (agricultural exports go down, manufacturing products are being substituted by cheaper imported goods). However, for mining the exchange rate is positively related to mining gva (even though it is a control variable), as seen in Table 2. Therefore the impact of the real exchange rate is on forehand not completely clear.

4.3.3 Results

Regression output is presented in Table 4. In the agricultural sector coefficients are as expected. The dummy of agriculture is negative, indicating that production is 3243 million lower than mining. Agricultural production in 1990, with RER = 0 (the constant for agriculture) is 7669-3243=4423 million. The boom-dummy variable (I) is 3.059, indicating that mining production increases with 3.059 million per year during the boom, corrected for the other variables. The total effect of the mining boom on agriculture is negative: δ2, the coefficient of

I*Year*Agriculture, is -2.015, which means that the mining boom decreases agricultural production with 2.015 million per year. All coefficients are strongly significant. Interesting is the role of the real exchange rate (RER). Even though it is a control variable (it has thus no causal interpretation) and the coefficient is not significant, it provides a positive relationship with gva. The R squared is quite strong, 0.776. 77.6% of the variance in the model is explained by the variables. So it has come clear that in the boom mining production grew and agriculture production declined.

Column (2) of Table 4 shows the regressions with manufacturing as treatment group. The dummy of manu-facturing is positive in (2), indicating that manumanu-facturing gva is on estimation 13178 higher than mining gva. The boom effects are kind of the same for agriculture, with a boom-dummy value of 3.079: mining production grew with 3.079 million per year before the boom. The total effect of the boom is here again δ2, which is

-1.749: mining production lowered manufacturing production with 1.749 million per year during the boom. All coefficients are again strongly significant, and the role of the real exchange rate is largely the same as in the agriculture model. The R squared is now quite high: apparently the coefficients explain most of the variance of the model. On the other hand a high R squared is also a risk, since it could indicate weaknesses in the model.

8_{The model uses HAC standard errors because the initial standard errors were serially correlated. Dickey-Fueller test statistics}

were all higher than the critical values. Test statistics are -1.848 for manufacturing, -2.523 for agriculture and 2.246 for mining; critical t-value is -2.890.

(22)

Table 4: Regression coefficients of the difference-in-difference model

(1) (2)

VARIABLES gva gva

Agriculture -3,243*** (0) I*Year 3.059** 3.079** (0.122) (0.142) I*Year*Agriculture -2.015*** (0) RER 49.79 46.61 (18.80) (22.00) Manufacturing 13,178*** (0) I*Year*Manufacturing -1.749*** (0) Constant 7,669* 7,846* (1,045) (1,223) Observations 208 208 R-squared 0.776 0.897 Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Regression output of the difference-in-difference analysis. Data is quarterly and contains the 1990-2015 period. Source: Australian Bureau of Statistics (ABS). ABS cat. no. 5204.0, table 05.

After all it is quite clear that both manufacturing and agriculture were declining during the boom, and that their sectoral shares were declining (since booming flourished). Both the difference-in-difference regression model and Figure 5 support these findings. Mining gva was growing, while manufacturing and agriculture were declining. The real role of the mining boom on these sector is not entirely clear yet. This will be analyzed further in the following regression models. The model analyzed the two different periods, but there is no clear dependent variable. It has come clear that manufacturing and agriculture were declining during the boom, but it is not entirely sure that the mining boom was the only reason for this. Neither is the extent of the impact of the mining boom clear. For example the global financial crisis (which occurred in the booming period) could also have affected sectoral production. For these reasons I will also include a regression in which I directly analyze the impact of the mining boom on manufacturing and agriculture.

4.4 The mining boom and sectoral growth

4.4.1 Methodology

For the panel data regressions, data will be somewhat modified. Employment and gva estimates are now share values. Agricultural and mining employment are the sectoral share in total non-mining employment. Mining employment is defined as total mining employment as share of total employment. The same definition applies to the sectoral gva variables. The calculation of the shares is done in order to analyze the impact of the boom on the sectoral composition (the essence of Dutch disease models and resource movement effects). Using non-mining employment and gva as the denominator in sectoral shares for manufacturing and agriculture is done to prevent endogeneity. The variable price index is a control variable, which defines average export/import price (a modified version of the terms of trade, which is a weighted export/import price and assigns more weight to products that are less relevant to this analysis). Price index is used here and not the exchange rate because export/import prices are assumed to have a larger impact on sectoral production than the exchange rate (an example is food: an appreciated exchange rate does have to mean that food production declines since price elasticity of food is generally low). Data is again quarterly, over the 1990-2015 period.

(23)

The regression model is straightforward: there will be just one explanatory variable, combined with yearly fixed effects and the price index as control variables. By using yearly fixed effects, the impact of the boom will be captured by the year dummies. The other control variable is the price index. Even though it is a control variable and so it has no causal interpretation, it could still provide some insights in the spending effect. The resource movement effect can be tested in the employment regression equation, as mining production is assumed to affect manufacturing and agricultural employment. Again capital is ignored in the resource movement effect, for the same reason as in the difference-in-difference model.

The regressions for employment are the following:

∆Agricultural Employmentt= α0+ α1∗ ∆Mining gvat+ σ1∗ Yeart+ σ2∗ Price Indext+ t (6)

and

∆Manufacturing Employmentt= α0+ α1∗ ∆Mining gvat+ σ1∗ Yeart+ σ2∗ Price Indext+ t (7)

For gva the regression models are the same:

∆Agricultural gva_t= α0+ α1∗ ∆Mining gvat+ σ1∗ Yeart+ σ2∗ Price Indext+ t (8)

and

∆Manufacturing gva_t= α0+ α1∗ ∆Mining gvat+ σ1∗ Yeart+ σ2∗ Price Indext+ t (9)

First differences are applied in all the models in both the dependent as the independent variable since the error terms were originally auto correlated. Using within (demeaned) estimates was also possible, but first differences allowed me to directly estimate the impact of the boom, since I am particulary interested in the changes of mining production over time 9_{. The year 2005 is used as reference year dummy, as it provides insight in the}

effects of the boom on the various outcome variables.

On forehand I expect the Mining gva variable α1 to be negative in all the four models. Based on theory, an

increase in mining production will decrease non-booming production and employment. However, the manufac-turing sector will presumably be less affected by the mining boom, as we keep in mind the findings of Downes et al. (2014) who find that manufacturing is an input sector of mining. Some (non-tradeables) manufacturing products are intermediate inputs in the mining sector. So production of the intermediate mining inputs are assumed to grow when mining production grows. Year dummies are relative to the year 2005-dummy. Since the boom started in 2005, years prior to the boom are assumed to have a negative coefficient, years after will have a positive coefficient (perhaps the years in the global financial crisis will have a negative coefficient). The price index will presumably be negative, as a higher export/import price will negatively affect output and so on employment.

4.4.2 Results

The interpretation of α1 in column (1) of Table 5 is the following: a one-unit increase in the first difference of

(mining gva/total gva) will decrease the first difference of (agricultural gva/total non-mining gva) with 0.584. Correctly spoken an increase in mining production in period t relative to period t-1 will decrease agricultural production with 0.584 in period t compared to period t-1. Putting this in perspective, mean (mining gva/total gva) is 0.0494 (see Table 1). So if Mining goes up from from 0.0494 to 0.0594 (first difference is 0.01), agricultural production will go down to 0.0334-0.584/100= 0.0328 (0.0334 is the average agricultural gva, 0.584 is α1 and

divided by 100 since the first difference of Mining is 0.01). Thus a 1 percent point increase in mining gva will in-duce a decrease of agricultural production by 0.584 percent point. For manufacturing the coefficient is positive.

9_{The Dickey-Fueller test statistics were all higher than the 5% critical level. T-statistic is -0.323 for mining gva, -0.530 for}

agriculture gva, 0.094 for manufacturing gva, -0.821 for agriculture employment and 0.061 for manufacturing employment. So the H0-hypothesis of non-stationarity will not be rejected. Trend-stationarity is, given Figures 3 and 5, not plausible. There is no reason to apply IV-estimates, since there is no source of endogeneity. Heteroskedasticity is avoided by using robust standard errors.

(24)

So an increase in mining production will increase manufacturing production. This coefficient is also significant, but now at the 95% level. Using the mentioned interpretation, a 1 percent point increase in mining production (since this is a share in total production) will increase the share of manufacturing to non-mining production with 0.299 percent point. This is very interesting, since it is contrary to international trade and Dutch disease theory. The next regression will get into more detail about this. The year dummy is in almost all years as expected: positive during, and negative before the boom. All year dummy coefficients are in appendix 1; 2005 is the reference dummy. The price index is small and insignificant for agricultural gva, for manufacturing gva the effect is larger and significant. A 1-unit higher price ratio (export/import price) lowers manufacturing gva growth with 0.0144 percent point. This effect is quite small, given that the variable has a range of 0.475 to 1.064.

The employment effects are far smaller in the model and less robust. The R squared is low and so the effect of mining gva on non-booming tradeables employment is weak (also given the insignificant and small coefficients). A one percent point increase in mining production from period t relative to period t-1 will decrease the agricul-tural employment share by 0.002 percent point. This effects is very low and insignificant. Suppose the share of mining production in total production goes up from 0.0494 to 0.594 (suppose again the average increases with 1 percent point), agricultural employment as share in total employment will go down with 0.002 percent point from 4.11 percent to 4.108 percent. So this effect is very small. This also applies to manufacturing employment, where the coefficient is -0.115. Again this coefficient is insignificant. The effect of a one percent point increase in mining gva from period t to period t-1 will decrease manufacturing employment as a share in non-mining employment with 0.115 percent point.

Table 5: Regression coefficients of the gva/employment model

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

VARIABLES ∆ Agric. gva ∆ Manuf. gva ∆ Agric. emp ∆ Manuf. emp

∆ Mining gva -0.584*** 0.299** (0.152) (0.134) ∆ Mining emp -0.00215 -0.115 (0.256) (0.315) Price Index -0.00103 -0.0144*** 0.000693 -0.0102 (0.00158) (0.00275) (0.00462) (0.00695) Constant 0.00101 0.00866*** -0.000915 0.00595 (0.00106) (0.00187) (0.00327) (0.00459) Observations 103 103 103 103 R-squared 0.714 0.585 0.168 0.188

Year Dummies yes yes yes yes

Regression output of the gva/employment regression analysis. First differences are applied. Agric is an abrreviation for agriculture, manufa for manufacturing and emp for employment Data is quarterly and contains the 1990-2015 period. Source: Australian Bureau of Statistics (ABS). ABS cat. no. 5204.0, table 05 & 6291.0.55.003, table 04.

.

The main conclusion that can be drawn here is that mining production has a clear impact on gva in the non-booming tradeables sectors manufacturing and agriculture. The impact on agricultural gva is negative (-0.584), whereas mining production positively affects gva in the manufacturing sector (0.299). The effect of the mining boom on agricultural and manufacturing employment is weak: coefficients are small and insignificant. This again proves that the resource movement effect (for employment shares) in Australia is small. The R squared is fairly low in both employment equations. The impact of export to import price changes (a change in the price index) is also small in the model, indicating low price elasticities of foreign demand. This also indicated that evidence for the spending effect is weak.

The mining boom and its impact on the manufacturing and agricultural sectors in Australia