Curse or Cure?
An analysis of the impact of natural resource abundance
on the economic growth of developed countries
LINDA BODE s1059459 L.Bode@student.rug.nl November 24, 2006 Supervisor: dr. G. Péli Rijksuniversiteit Groningen Faculty of Economics
-CONTENTS
ABSTRACT 3
I INTRODUCTION 4
II LITERATURE REVIEW 7
III DATA AND MEASUREMENT 18
IV RESEARCH METHOD AND RESULTS 22
V DISCUSSION AND CONCLUSION 31
REFERENCES 35
ABSTRACT
Men of a fat and fertile soil, are most commonly effeminate and cowards; whereas contrariwise a barren country make men temperate by necessity, and by consequence careful, vigilant, and industrious. Jean Bodin (1576)1
I INTRODUCTION
In the 1990s, economic researchers discovered a striking relationship between the degree of endowment of a country in natural resources, and its economic growth and development. They found that this relationship was inverse, meaning that an abundance of natural resources tends to slow down economic growth. This is against intuitive thoughts of natural resources as sources of economic wealth. Economists call this phenomenon the “resource curse” or the “paradox of plenty” (Auty 1993).
Jeffrey Sachs and Andrew Warner were among the first to do in-depth research on the topic. They wrote an influential article about the resource curse, “Natural resource abundance and economic growth”, published in 1995. They came up with statistical evidence to prove the theory. They showed that developing countries had a lower economic growth due to their possession of natural resources. The term resource curse is used to describe how countries rich in natural resources are not able to use that wealth to boost their economies and how, counter-intuitively, these countries have lower economic growth than countries without an abundance of natural resources.
The cause of the curse is not natural resources, but government mismanagement when resources are present. Jean Bodin very early recognized the jeopardies of possession of natural resources (Bodin 1576). Interpreting his words, natural resources encourage imprudence, where as resource scarcity leads to more well-considered policies. Of course, the presence of many underdeveloped countries with no natural resources presumes that scarcity of natural resources does not automatically lead to better governance.
Nowadays, the thought of spoiling natural resources is directly linked to developing countries and most research is done with the focus on these countries. However, developing countries are not the only ones to be rich in natural resources. Many developed countries also have high endowments. So, somehow they seem to have found a way to overcome the curse. Nevertheless, the question remains if these countries still experience lower economic progression than their resource-poor counterparts.
To give a better impression and background information, the next section will very shortly summarize the economic development path of developed countries together.
Historical context of development
The First Industrial Revolution started around 1780 in the United Kingdom, with the introduction of the steam-engine. The steam-engine was made of iron and ran on coal. So along with the success of the engine, the coal and iron industries flourished as well. From the end of the 19th century, science would start to deeply penetrate industry. While the First Industrial Revolution passed off without scientific influence, the Second Industrial Revolution was the result of innovations in physics and chemical science.
In 1951, European Coal and Steel Community, was founded by France, West-Germany, Italy, Belgium, Netherlands and Luxembourg to pool the steel and coal resources. It was the start of a successful cooperation between West-European nations. Oil and electricity became competitors of coal, but the prosperous coal industry did not stop until in the 1960s (Caljé 1998).
Historically, increases in inputs and technological progress were important sources of economic growth in the industrialized nations. In the future some factors of production, such as labour, will not increase as rapidly as they have in the past. The effect of this decline on growth depends on the interplay among the law of diminishing marginal productivity, substitution possibilities, and technological progress (Tietenberg 2000).
A preliminary conclusion about the economic growth of developed countries, as compared to developing countries, is that the former used their natural resources as inputs for industrial evolution. Countries sold their natural resources, but also used them for their own industries and technological development. In addition, countries started to cooperate to benefit from economies of scale. This is contrasting with developing countries, whose resources are almost exclusively exported, instead of being used in the manufacturing sector, and who have to deal with internal struggle. With the notable exception of a relatively few oil-rich nations, most developing countries import a great deal of energy. Because this demand is relatively price inelastic (Taheri and Stevenson 2002), their expenditures on imports have risen tremendously without similar compensating increases in receipts from the sale of exports.
The situation is reversed in many of the oil-exporting countries, which are obtaining high prices for their oil. Their favourable terms of trade, however, have not always kept them away from development difficulties. Nigeria is a classic example. The oil exports affected the local wage structure and exchange rates in such ways, that they ended up severely harming agricultural production. Resources flowed out of agricultural production and into oil production. Even the income distribution was adversely affected, becoming much more unequally distributed (Tietenberg 2000).
A preliminary conclusion about what is most important for resource-based development is the nature of the learning process through which the economic potential of natural resources is achieved, and not the inherent character of the resources (Stijns 2005). This will be elaborated upon in the next section. The remainder of this thesis is structured as follows: after giving a more theoretical overview of the resource curse and an evaluation of existing literature about research done on the topic, economic growth of resource-rich developed countries will be compared with resource-poor developed countries to see if the resource curse still holds for developed countries.
II LITERATURE REVIEW
The historical record suggests, oddly enough, that countries with abundant natural resources tend to suffer a disadvantage in economic development. After reviewing existing literature about the topic and the formulation of hypotheses, this part will be concluded with a short analysis of the role of institutions, which are generally considered as a ‘cure’ for the resource curse.
II.A General literature
national income. The consequence of their paper was far reaching. During the years, many researchers retested and elaborated on their findings, but there was hardly any criticism expressed.
In a next paper, Sachs and Warner (2001) extended their previous research and state that also geographical or climate variables cannot explain the curse, and that there is no bias resulting from some other unobserved growth-slowing factor. The list of variables includes the percent of land area within 100 kilometres of the sea, kilometres to the closest major port, the fraction of land area in the geographic tropics and a malaria index from 1966. However, these four geography and climate variables are taken from Gallup et al. (1999) and their conclusion is simply copied by Sachs and Warner, instead of being tested again with natural-resource variables. Their other conclusion is that resource-abundant countries tend to be high-price economies and, maybe as a result, these countries do not focus on export-led growth in manufactures. It has to be noted that maybe, these countries have made some other growth-facilitating investments not directly linked to export, but to domestic progress.
I want to add that countries can also consume their own resources (domestic consumption is not taken into account by Sachs and Warner), without the need to develop and use their resources more efficient, because there is plenty. An example of this is the former Soviet Union.
To my opinion, an important shortcoming of Sachs and Warner is the absence of a control group consisting of countries which have no natural resource abundance. This would alter or strengthen their results more and give optional pathways of how economic development is possible without complete reliance on these resources. Instead of only ascertain the existence of the resource curse, Sachs and Warner could also give potential solutions, which would increase the academic and practical value of their findings. This still remains an option for future research.
Neumayer (2004) tests if the resource curse still holds true for growth in real net domestic product (NDP) instead of GDP, and whether the negative effect of natural resource-intensity on growth is over- or underestimated by erroneously examining growth in GDP. This is important, because GDP is a particularly wrong measure of income for resource-intensive economies, because GDP contains an element of capital depreciation that should not be counted as income. This is corrected by using NDP. He finds that natural resource-abundant countries do suffer from a resource curse, but this is weaker in terms of growth of genuine income than growth of GDP. However, he does not question the existence of the curse itself; it is taken for granted.
the extent to which political opponents can challenge their power.The countries’ varied paths from resource wealth to political and economic outcomes suggest the need for ‘conditional theories’ of the resource curse. With this article, Dunning gives potential different pathways that reinforce conditional theories in a mathematical way. The following reviews are supported by his findings.
Papyrakis and Gerlagh (2004) try to explain empirically the direct and indirect effect of natural resource abundance on economic growth, because they want to investigate the causes for the underperformance of most countries rich in natural resources. They discover that natural resources have a positive influence on growth when considered in isolation, but when other variables such as corruption, investments, openness, terms of trade, and schooling are included and a country is badly performing on these variables, the presence of natural resources is even deteriorating the negative effect of these variables. An empirical analysis has been performed to show that natural resources increase growth, when abstracting from possible negative indirect effects. The analysis also made clear that, when accounting for the transmission channels, the overall effect of natural resource abundance on economic growth is strongly negative.
But they fail to explain why the influence of natural resources is increasing these negative effects and they do not prove that it is indeed caused by natural resources, and that it is not a coincidence. A country that suffers from corruption, low investments, protectionist measures, deteriorating terms of trade, and low educational standards will probably not experience economic growth at all, whether it has natural resources or not.
There is more than only a one-sided view on the relationship between natural resources and low economic growth; some papers have been published about possible explanations for the contrasting development paths of developed and developing countries with resource abundance.
For instance, Hodler (2004) develops a model to explain why natural resources can have such different effects for different countries. He looks at the impact of natural resources on the cohesion among the people, in terms of aggressive behaviour between rivalling groups. Fighting reduces productive activities and weakens property rights, making productive activities even less attractive. This negative effect exceeds the natural resources’ direct positive income effect if the number of rivalling groups is large enough. The model thus predicts that natural resources lower incomes in fractionalized countries like Nigeria and Angola, but increase income in a homogenous country like Botswana. However, the focus has always been on developing countries and no sufficient attention is paid to developed countries with natural resource abundance. I hope I can make a start in filling this gap.
II.B Four negative effects
There are several explanations for the existence of the resource curse, which can be summarized into four negative effects. This part will be concluded with a short analysis of the role of institutions, which are generally considered as a ‘cure’ for the resource curse.
1. Dutch Disease
‘Dutch Disease’ is a famous example of the phenomenon described by the Rybczynski theorem2. The term refers to the late 1960s, when huge reserves of gas were found in the Netherlands. Dutch Disease is an economic phenomenon in which the revenues from natural resource exports de-industrialise a nation’s economy by causing an increase of the real exchange rate and thus making the manufacturing sector less competitive in the
world market. Traditional agricultural or manufactured exports are crowded out, as experienced by the Netherlands and the United Kingdom.
The decrease in the manufacturing sector and dependence on natural resource revenue is bad, because it leaves the economy extremely vulnerable to price changes of the natural resource. Also, since productivity generally increases faster in the manufacturing sector, the economy will lose out on some of those productivity gains (Van de Ploeg 2006). Domestic price changes may encourage output and investment in non-traded goods and services, construction and retailing for example. There is a switch from products and internationally traded goods to services and non-traded goods, because of the worse international trade position (Murshed 2004).
So, as natural resources tend to increase the money demand, the exchange rate goes up and unemployment will increase because of a worse trade position due to more expensive exports (Van Wijnbergen 1984). When looking at Finland, this is supported because this country has high economic growth but also a high unemployment rate (OECD).
Hypothesis 1a: the exchange rate is inversely related to economic growth. Hypothesis 1b: the unemployment rate is inversely related to economic growth.
2. Extreme rent-seeking behaviour
The second aspect is extreme rent-seeking behaviour. Rent seeking is the process by which an individual, organization, or firm seeks to gain through manipulation of the economic environment, rather than through trade and the production of added wealth. Rent seeking generally implies the extraction of uncompensated value from others without taking actions which improve productivity, such as by gaining control of land and other pre-existing natural resources, or by imposing regulations or other government decisions that may affect consumers or businesses.
legal. In other instances, rent seeking takes more extreme forms, such as bribery, corruption, smuggling, and black markets (Krueger 1974).
In cases of extreme rent-seeking behaviour concerning natural resources, individual agents (politicians, entrepreneurs and elites) expose strategic behaviour to obtain resource rents, thereby disturbing the allocation of resources and reducing economic efficiency as well as social equity (Stiglitz 2004). Elites in control of resources resist industrialization, which would weaken their power base. The result is delayed modernization and lower levels of development (Bulte 2005).
Besides dampening the industrialization from inside the country, foreign investment is also expected to decrease or grow less, because political instability increases the risk and costs involving investment. Foreign companies are restrained to invest, when the return on their investments is precarious.
Thirty years ago, Indonesia and Nigeria – both depending on oil – had comparable incomes per capita. Nowadays, the income of Indonesia is four times as high as Nigeria’s. Nigeria's income even has decreased (Stiglitz 2004). Capturable resource rents can lead to rent seeking behaviour: revenues and royalties from oil or mineral resources are much more readily appropriable as compared to the income flows from agricultural commodities, because oil and mineral sources are more concentrated than agriculture.
Increases in the availability of resource rents following a boom in their world market prices can increase the greed for resource rents amongst certain individuals or groups within society. Consequently, these economies become weak and inefficient and eventually experience a growth collapse. Furthermore, the presence of competition over capturable or lootable natural resource rents can lead to civil war, like in Sierra Leone and Angola. In other cases oil, or the possible obstruction of oil-pipelines, fuel civil wars as in Sudan (Murshed 2004).
Summarizing: more natural resources lead to lower welfare. An increase in the level of corruption will lower the transparency, which implies less openness, communication and accountability, and decrease foreign inflow investment because of higher risks and costs (O’Higgins 2006).
Hypothesis 2a: transparency is positively correlated with economic growth.
Hypothesis 2b: foreign inflow investment is positively correlated with economic growth.
3. Overconfidence
The third negative effect is overconfidence; in this case the feeling of governments of more certainty than circumstances warrant. Governments feel overconfident about their future earnings, so they increase their spending and commit themselves to social arrangements which they cannot afford in the future. The real exchange rate increases through capital inflows from resource exports. This is tempting governments to accumulate debt because the interest payments are cheap, even though they are receiving natural resource revenues as well. But, if prices begin to fall, the real exchange rate falls as well and the debt payments increase (Stiglitz 2004). The generous social security system of the Netherlands, which has its base in the 1970’s, is currently the cause of many economy measures and resulting cuts in the welfare system (Van de Ploeg 2006).
Economic diversification may be neglected by authorities or delayed in the light of the temporary high profitability of the limited natural resources. The attempts at diversification that do occur are often grand public works projects which may be misguided or mismanaged. Some have suggested that a more effective mechanism than state monopoly would be to simply distribute revenues from state-controlled natural resources evenly among the population, as is done in the oil-rich Alaska (Stiglitz 2004).
Concluding: governments will borrow more with future revenues of natural resources in mind (Sachs and Warner 2001).
Hypothesis 3: debt is negatively correlated with economic growth.
Another possible effect of the resource curse is the crowding out of human capital. Natural resources are a form of capital, which, if depleted, must be either replenished or substituted if countries are to expand their asset base and sustain their consumption levels (Tietenberg 2000). It needs to be emphasized that it is not the existence of natural wealth per se that is the problem, but rather the failure of governments to avoid dangers accompanying natural resource abundance. Countries that rely on natural resource exports may tend to neglect education because they see no immediate need for it (Gylfason 2000).
Resource-poor economies like Taiwan or South Korea spent enormous efforts on education, and this contributed in part to their economic success (see East Asian Tigers). These countries felt the need of investing in human capital in order to obtain economic growth (Gylfason 2000).
So public investment is also especially important for resource-rich countries. An early example of how to execute this is the 1879 United States Geological Survey, a detailed mapping of reserves and potential reserves, which was critical to development. Many state colleges offered mining degrees by the 1890's, including the University of California at Berkeley (Madrick 2004).
Public investment in human capital is expected to be lower for developed countries with an abundance of natural resources. It is indicated by three variables: public expenditure on education, labour productivity and research & development. The first is chosen as a general measure of investment in basic human capital. R&D indicates a high-level effort to develop human capital. Labour productivity can be seen as the output of development of human capital, since higher qualified skills are needed to increase this.
Hypothesis 4a: public expenditure on education is positively related to economic growth. Hypothesis 4b: labour productivity is positively correlated with economic growth.
II.C Role of institutions
Is there an inevitability that condemns point-sourced economies to poor economic performance? According to Sachs and Warner (1995), this is the case. However, rich mineral resource endowments did not prevent economic growth in Australia, Canada and the USA a century ago. Comparing how countries with diverse political systems use their natural resources suggests that systems of governance have important effects on good resource use.
The empirical evidence points to the importance of institutions that are the crucial link between endowments of natural resources and economic outcomes. Good institutions refer to aspects like voice and accountability, political stability, government effectiveness, regulatory quality, rule of law and control of corruption. Institutions determine the extent to which political incentives map into policy outcomes (Robinson et al. 2006). Three of the four negative effects of natural resource abundance can be averted by good institutions: overconfidence, investment in human capital and, above all, rent-seeking behaviour. Only Dutch Disease cannot be directly solved by institutions, since this is outside government control.
Institutional reform could, therefore, be the key to altering economic outcomes as expected by Sachs and Warner. This is confirmed by Bulte et al. (2005). They explore the impact of natural resources, possibly channeled through institutional quality, on several human development indicators. They conclude, after doing panel data analyses, that resource-intensive countries tend to suffer lower levels of human development, and that institutional reform may be a necessary condition for countries to develop.
The findings of Mehlum et al. (2006) contradict the claims of Sachs and Warner that institutions do not play a role. Their main hypothesis – that institutions are decisive for the resource curse – is confirmed statistically. Countries rich in natural resources constitute both growth ‘losers’ and growth ‘winners’. The combination of institutions that encourage rent seeking at the expense of production and resource abundance leads to low growth. Institutions that make production and rent seeking complementary activities, however, help countries to take full advantage of their natural resources. Sachs and
the role of institutions. Institutions may be decisive for how natural resources affect economic growth even if resource abundance has no effect on institutions. Panel data is used to prove this statement.
Robinson (2006) argues that the political incentives that resource endowments generate are the key to understanding whether or not they are a curse. Countries with institutions that promote accountability and state competence will tend to benefit from resource booms since these institutions ameliorate the perverse political incentives that such booms create. Countries without such institutions, however, may suffer from a resource curse.
Deacon (2004) gives a thorough analysis of a failing political system on the potential rents of natural resources. When a country’s political system is unstable or not representative, the individual’s claim to a resource stock’s future return can be rendered insecure. This reduces the payoff to natural resource conservation, leading to more rapid depletion of resource stocks. When insecurity is a general feature of an economy, however, it can have the secondary effect of raising the cost of resource extraction, rendering some stocks uneconomic and slowing rates of depletion. In addition, when a country’s natural resources are capable of generating significant rents, but institutions of democratic governance and the rule of law are not well-established, corruption by government officials responsible for resource management can encourage rent-seeking, dissipating the benefits those resources would otherwise confer.
Developed countries were able to establish a stable political system, with Norway as often-used example. “Oil wealth in many other countries has been used to finance colossal fortunes for the few, or bread and circuses for the many,” the Organization for Economic Cooperation and Development wrote in a recent report.3 Its system “'sets a powerful example of enlightened policies to other resource-rich countries,” the OECD said, even if many economists agree there are some elements that would be hard to copy, like Norway’s historic lack of serious corruption and tradition of consensus-based politics. To prevent their economy from the negative effects of the resource curse, the Norwegians used oil income for national debt payments. By 1995, Norway’s financial balances were stable and have been kept that way (Ekman 2005).
So, developed countries seem to have dealt successfully with these negative aspects of natural resource abundance, due to better institutional development. Still, the question remains whether the economic growth of resource-rich developed countries is lower than that of their non-abundant counterparts. Based upon the idea that the four negative effects will play a part in the economic growth, together with the idea that all developed countries have good institutions, so that is no aspect of differentiation, I expect developed countries with natural resources to experience lower economic growth than developed countries without natural resources.
Hypothesis 5: the economic growth of resource-rich developed countries is lower than the economic growth of resource-poor developed countries.
In addition, to what extent do the four negative aspects have an impact on the economic growth of these countries? These two questions are addressed in the next sections.
III DATA AND MEASUREMENT
The five hypotheses will be operationalized in this section. Several steps have to be taken in order to get useful datasets. To construct the data lists, I had to decide upon which countries and which variables to use, and how to define natural resource abundance. The time period chosen is from 1990 until 2003, because of the availability of data. The first choice to be made is to decide which countries to include in the sample. Three questions have to be answered: which countries are developed, what definition of resource abundance should be used, and which data about resource abundance are useful and representative?
Sample selection and classification
The country sample was straight forwardly composed. I combined the member countries of the OECD4 with the countries belonging to the advanced economies, as grouped by the
IMF5. There was already a huge overlap between the two lists and together they form a sample of 33 countries (table 1 appendix).
The second question is what resources to pick. I made a list of the countries within the sample and checked them with lists with data about energy reserves per country, composed by British Petroleum6. Mineral abundance is left out, since statistical information about this resource is biased. Stijns (2005) presents convincing evidence that mineral reserves have unclear effects on economic growth, when split out into the four negative effects, whereas results are more unambiguous regarding coal, gas and oil. So these depletable, non-recyclable fuels oil, gas and coal remain. It has to be mentioned that none of the resource-poor countries is rich in minerals, so the sample would not alter if minerals were included. The division between resource-rich and resource-poor countries can be found in table 1 of the appendix. A country is classified as resource rich, when it has one or more of the selected resources, coal, gas and oil. A country is classified as resource poor, when it has none of the resources.
The third step is to label the developed countries as resource-abundant and as non-resource-abundant. So I had to deal with the definition of resource abundance. Different opinions exist about when a country can be considered as being resource-abundant. Sachs and Warner (2005) focused on primary export intensity, which is the ratio of primary exports to national income. Stijns (2005) instead looks at production and reserves data, which is also recommended by Deacon (2005). However, these studies dealt with the differences between countries concerning the degree of resource possession, instead of giving a threshold value dividing countries into resource rich and resource poor. As a guide line, I took charts of British Petroleum7 ranking countries which have most of the world reserves in oil, gas or coal. A country named in that list is labelled ‘resource rich’; countries not mentioned on any of the charts are labelled ‘resource poor’.
Stijns (2005) found a high degree of correlation between production and reserves data for oil, coal, gas and minerals. Correlations all range between 71 and 97 per cent, except for mineral production and reserves. I follow his suggestion partly and use production data as measure for resource abundance, because more specific data can be
5 http://www.imf.org/external/pubs/ft/weo/2005/02/data/dbcselm.cfm?G=110 6 British Petroleum: Statistical Review of World Energy (June 2005)
found on production. By dividing the sample into a resource-abundant and a non-resource-abundant subsample, I also have a control group to compare the results with. Most studies do not take this into account, but a control group will increase the validity and robustness of the results.
Dependent variable
In this paper, it is investigated if the possession of natural resources has an impact on the economic development of a developed country, so the dependent variable is economic growth. This is measured as the annual change in percentages.
Independent variables
The next problem to deal with is what variables to choose to determine economic growth. The problem with most studies is that they focus primarily on developing countries, thereby looking at their distinctive features. Sachs and Warner (1995) choose as independent variables share of primary exports in GDP (as measure of resource abundance), openness of the economy, access to the sea, investment, bureaucracy, and income inequity. They added a dummy variable to control for regional differences. Because I want to look specifically at developed countries, some variables have to be replaced and other variables have to be included as well. All the specifics of the variables can be found in table 2 of the appendix.
education and debate, where politicians and economists non-stop argue that Norway should avoid spending too much (Ekman 2005).8
Having this in mind, I want to see to what extent the four negative aspects of resource abundance can be assigned to resource-rich developed countries. In order to find this out, I have to translate the arguments as posed by Van de Ploeg and Stiglitz (among many others) into operational variables. My solution is the following, based on the argumentation in part II: I adopt unemployment rate (U) and exchange rate (XR) as variables for negative effect 1, higher exchange rates because of Dutch Disease, causing higher unemployment rates. There is a chance of lower growth affecting the unemployment rate; this will be dealt with in the next section.
Transparency (TR) and investment (INV) are the variables for negative effect 2, extreme rent-seeking behaviour. These variables also comprise the important factors by Rick van de Ploeg (2006), as mentioned in the former paragraph.
Debt (DE) is picked as variable for negative effect 3, overconfidence, because governments will borrow more with future revenues of natural resources in mind (Sachs and Warner 2001). Debt is used as a proxy for overconfidence; there are some difficulties with this choice because it does not cover social arrangements, but due to limited data availability, I see it as the best option.
Labour productivity (LP), public expense on education (ED; as % of GNI, because data in relation to GDP was not available) and expenses on R&D (RD) are variables for the last negative effect, neglect of investment in human capital.
Control variable
Another problem with most studies is that variables found to be significantly explaining why developing countries failed to develop, are not tested on developed countries to confirm that they really are determinants of economic growth. By testing to what degree these variables do account for economic growth in case of developed countries, by adding a dummy variable, I can judge upon their validity. The dummy variable indicates whether
a country is resource abundant or not, so one can see if the test results are influenced by this factor.
IV RESEARCH METHOD AND RESULTS
In this section, the five hypotheses will be tested and results will be discussed. The reflection on the research done will especially focus on heteroskedasticity, autocorrelation and causality, since these are roadblocks for the test results.
IV.A Economic growth differences
Step 1: Is the economic growth of resource-abundant developed countries significantly lower than the growth of their non-abundant counterparts?
GDP is a generally accepted measure of economic activity. Growth of real GDP, so GDP corrected for differences in price changes, is widely used to evaluate the performance of states in managing their economies.9 The country sample contains data of the countries’ GDP annual percentage change, in constant prices for the period 1991-2003. The data were published by the IMF10.
The average growth of the whole sample is 3,05 per cent. The average growth of developed countries rich in natural resources is 2,69 per cent. The average growth of non-resource abundant countries is 3,66 per cent. This seems to support the non-resource curse, but one cannot say that natural resources are the cause of lower economic growth. Another point of attention is that in there is no correction of initial GDP. This means that growth for countries with a relative low GDP will result in a higher percentage GDP growth than the case where GDP is relative high with equal economic growth. However, since all countries in the sample are developed and the time frame is recent, I expect the differences in economic growth no to be very large.
A two-sample t-test is useful to see if the average growth rates of the two groups are equal or not. Two hypotheses can be derived:
Hypothesis0: B=0 the economic growth of both groups are equal
Hypothesis1: B=0 the economic growth of both groups is not equal
The number of degrees of freedom is the number of both groups minus two, in this case 31. The t-value is 2,1111, the critical t-value (a = 0,05) is 2,045. Because the observed t-value is exceeding the critical t-value, H0 must be rejected in favour of the
alternative. The general conclusion from this is that the average economic growth of resource-rich developed countries is significantly lower than the average economic growth of resource-poor developed countries.
IV.B Testing the four hypotheses
Step 2: to what extent have the four negative aspects an impact on the economic growth of the resource-rich developed countries?
The fundaments of test 2 are based upon the work of Sachs and Warner. Their basic idea is that economic growth in economy I between time t=0 and t=T (in my case, 1991 and 2003) should be a function of initial income Yi0 and a vector of other structural
characteristics of the economy Zi. This is denominated as dependent variable G7089, because they chose a time span from 1970 to 1989. Sachs and Warner constructed the variables share of primary exports in GDP in 1971 (SXP), openness (SOPEN, dummy variable), investment to GDP ratio (INV), quality of bureaucracy (BUR), and initial income in 1970 (LGDP70).
(1) G7089 = α0 + α1*SXP + α2*SOPEN + α3*INV7089 + α4*BUR + α5*LGDP70 + ε
I want to use the same type of equation form as Sachs and Warner, but since their variables are more applicable to distinct between developed and developing countries, I use variables which make it easier to differentiate between developed countries, based on the discussion in section III. For instance, the variable SXP is not precise, because most
11t =X1-X2 / √ (S
developed countries have a high-developed manufacturing industry, using their natural resources in this sector. (Stijns 2005). Natural resources are indirectly exported in this way, but this is not accounted for in SXP, since only raw-natural resource exports are counted, so instead I look at the reserves data.
The dummy SOPEN is controversial, because countries get a 1 when their economies are open for the entire period 1965-1989, and a 0 when they are not. This invokes a bias for the 0, because all countries becoming open during the time period are grouped under 0. Instead, I made a dummy variable NRD, grouping countries into possession and non-possession of natural resources.
BUR is a measure for quality of bureaucracy. I think it is more suitable to use the transparency index of Transparency International, because this index includes bureaucrational level.
The investment variable INV is also used by me, but initial income LGDP not. The main critic I have regarding this variable is the equal start of all countries, while natural resources or additional reserves can have been found earlier or later on. The linked profits, therefore, start accumulating on different moments. For example, gas was found in the Netherlands in the 1960s. Gains were made shortly after that, so the initial income year had to be chosen earlier. Maybe most profits were already made, or still had to be made because of technological progress in later periods.
I want to see to what extent the negative aspects of resource abundance have an impact on economic growth. The following equation expresses the link between economic growth and all negative aspects (also summarized in table 2):
(2) GDP9103 = ß1 + ß2*U + ß3*XR + ß4*TR + ß5*INV + ß6*DE + ß7*ED + ß8*LP +
ß8*RD + ß8*NRD + ε
rent-seeking behaviour, thereby generating an environment with lower risk coupled with lower costs, which makes it more attractive for foreign investors to make investments. The argument can also be reversed; higher corruption positively influencing foreign investment, because rent-seeking institutions, groups or persons can give lucrative contracts to foreign investors and keep the benefits for themselves or share them with the foreign investors. So there is also a chance of TR influencing INV, which is excluded by terminating TR.
Fixed effects for cross section was not possible, because it would result in a near singular matrix. Fixed period effects were accounted for in the regression. This approach is relevant when one expects that the averages of the dependent variable will be different for each cross-section unit, or each time period, but the variance of the errors will not.
In table 5, I excluded education (ED), because of its high p-value. Data from the OECD also show that public expenditure on education has not enormously changed per country over the last two decades.12 So its merit for the model can be doubted. The yield of education can also be acknowledged when looking at labour productivity, which I consider the output of educational expenses. This exclusion increases the significance of the other variables, except for the already non-significant ones. It also increases the explanatory power of the equation as a whole, because less regressors give more strength to the equation.
According to table 5, the next equation is adequate:
(3) GDP9103 = 0,693 - 0,041*U + 0,024*XR + 0,001*INV – 0,009 *DE + 0,683*LP – 0,505*RD – 2,228*NRD
The exchange rate (XR), debt (DE), labour productivity (LP), research & development (RD) and the natural-resource dummy (NRD) are accepted at the 5% level. Investment (INV) has no significant influence, but its coefficient is also very low. Attention needs to be paid to the dummy variable NRD, which indicates that the possession of natural resources has a negative impact on the growth of GDP, a very important proposition of this thesis. This partly confirms the theory of Sachs and Warner, with support for
hypotheses 3 and 4. Hypotheses 1 and 2 are rejected, because XR (as represented by H1 about Dutch Disease) has a positive parameter, and extreme rent-seeking behaviour (H2) is rejected, because investment (INV) is not significant. The parameter of research and development (RD) (H4) also has the opposite sign than expected. It could be the case that the initial expenditure in R&D leads directly to lower economic growth, since resources put into R&D cannot be used elsewhere. Before output is produced out of R&D efforts, many years can pass. So there is a time delay between input and output. Labour productivity (LP) seems to be a very important indicator of economic growth (GDP9103). This would suggest that active government policy to improve labour productivity will result in higher economic development.
Assumptions of the simple linear regression model
The simple linear regression model is based upon six assumptions (Carter Hill 2001) . SR1: Linear relationship between xt and yt. The value of y, for each value of x, is: Y = b1 + b2x + e
SR2: The error term e is zero on average. This is assumed, because the mean value is denoted in terms of independent variables: E(y) = b1 + b2x
SR3: The variance of the error term var(et) is constant over time. This implies that at each level of an independent variable, we are equally uncertain about how far values of the dependent variable may fall from their mean value. If this assumption is violated, data are said to be ‘heteroskedastic’. This will be further explained below.
SR4: The errors are uncorrelated. The data collected must be statistically independent.
SR5: x must take at least two different values; regression analysis is used to measure the effects of changes of the independent variable on the dependent variable. To obtain this, the independent variable most at least take two different values within the sample of data.
Correlation
The correlation matrix (table 10) shows that NRD, the dummy for resource abundance, is negatively correlated with debt and R&D, and positively correlated with labour productivity and unemployment. The relationship between natural resource abundance and research & development, and natural resource abundance and labour productivity are as expected, nevertheless these are statistical, not causal, relationships.
Covariance is another measure for correlation, to show the range of two variables. According to tables 11 and 12, covariance values are very low, indicating a low correlation between the variables. Table 12 gives values corrected for period heteroskedasticity and general correlation of observations within a given cross-section. Its numbers are in general more extreme; correlations are stronger under Period SUR (Seemingly Unrelated Regression), a function within Eviews which is an example of the Parks estimator.13. Both tables confirm the evidence of table 10, about the relatively low correlations between the variables, leaving some exceptions there.
Autocorrelation and heteroskedasticity
Two important problems can arise when analyzing a regression: autocorrelation and heteroskedasticity (Carter Hill 2001). The former is linked to time-series analyses, while the latter is suffered by cross-sectional analyses. Since panel data has both cross-sectional and time dimensions, the problems could be very serious. According to Carter Hill et al. (2001) error terms of panel data cannot be correlated, because the randomness of the sample implies that the error terms for the different observations are uncorrelated. However, in time-series data observations follow a natural ordering through time, generating a possibility that the successive errors will be correlated with each other. This means a violation of the assumption of a simple linear regression model that covariances are zero. The Durbin-Watson test statistic is a good indication of this autocorrelation. A value around 2 indicates no autocorrelation; below 2 positive autocorrelation. As can be
13
seen in tables 4 and 5, the Durbin-Watson statistic is 1,11, indicating the existence of positive autocorrelation. This means that the error term of the equation is affected by shocks earlier in time and not only by current shocks.
Heteroskedasticity, or the existence of different variances in the data (violation of assumption SR3), does not affect the values of the estimated coefficients, but it has an impact on the standard errors of the coefficients and consequently on the reported t-statistics. It is dealt with in table 6. It shows the regression with Period SUR. It corrects for both period heteroskedasticity and general correlation of observations within a given cross-section.
According to table 6, the following equation is adequate:
(4) GDP9103 = 0,593 - 0,046*U + 0,024*XR + 0,026*INV – 0,002 *DE + 0,642*LP – 0,680*RD – 1,322*NRD
Only the exchange rate (XR), investment (INV), labour productivity (LP) and research & development (RD) are not rejected at the 5% level.
Compared to the results of the regression in table 5, overall significance of the variables has decreased, but the explaining power has increased, and the positive autocorrelation has disappeared. It seems that heteroskedasticity and general correlation within cross-sections play a large role in determining the results. This can be explained by looking at the type of data. Economic data are inherently vulnerable to correlation, which can also be seen when looking at R-squared. It has increased compared to table 5. R-squared is still relatively high; although other possible influential factors are still missing.
Normality
Jarque-Bera test gives insufficient evidence to conclude that the normal distribution assumption is unreasonable. When the residuals are normally distributed, the Jarque-Bera statistic has a chi-squared distribution with 2 degrees of freedom. Thus, the hypothesis of normally distributed errors is rejected when a calculated value exceeds a critical value. In this case, the 5% critical value from a chi-squared distribution with 2 degrees of freedom is 5,99. Because Jarque-Bera 285,7719 (table 7) and Jarque-Bera 21,09130 (table 8) both exceed the critical value 5,99, the hypothesis of normally distributed errors cannot be accepted. This conclusion means that SR6 is violated.
The equation quite fits the actual observations (table 9), because a balance can be seen of positive and negative residuals throughout the sample, except for two negative peaks.
Causality
Another possible roadblock for my research is the matter of causality, in other words, the question that resource-rich developed countries have lower economic growth does indeed depend on their possession of natural resources.
Granger Causality is a method to see if one time series is useful in predicting another. Normally, regressions show just correlations. However, Clive Granger (1969) stated that there is an interpretation of a set of tests as indicating something about causality. It is important to know that the statement ‘x Granger causes y’ does not indicate that y is the effect or the result of x. Granger causality measures precedence and information content, but does not by itself indicate causality in the more common use of the term.
INV Granger causes RD, meaning that inflow foreign direct investment is affecting investment in R&D. This seems logical, because more foreign expenditure points to a higher interest in a country to develop economic activities within this country. R&D is necessary to maintain or improve this progression.
A danger with using labour productivity is the risk that it is also caused by the dependent variable, economic growth, since higher economic growth can also boost labour productivity. A reason for this can be that an increase in economic growth will provide funds to invest in public health and education. This, in turn, will increase labour productivity (Mahmud and Rashid 2006). However, by looking at table 13, it can be seen that this is not the case. So, maybe it only works the other way around: economic growth can make people ‘lazy’, because there is no urge to improve.
In conclusion, the negative effects extreme rent-seeking behaviour (H2) and investment in human capital (H4) are confirmed. However, the dummy variable NRD (H5) has a higher p-value, which means that the significance of the claim that natural resource abundance is diminishing. The negative sign remains, indicating a negative relationship between GDP change and the possession of natural resources. Hyothesis 3 about overconfidence resulting in higher debt is not confirmed. The exchange rate XR (H1) and research & development RD (H4) still have the opposite parameters.
A possible explanation why investment in research and development (RD) can influence GDP negatively is because it can take many years of input and costs before progress is made and the fruits can be reaped.
V DISCUSSION AND CONCLUSION
Some countries have managed to turn their resource wealth into economic wealth; other countries have failed to do so and still struggle with their policy. The question is if the presence of natural resources causes lower economic growth. My research is directed towards developed countries. A two-sample t-test showed that there is in fact a significant difference between the economic growth of resource-poor and resource-rich developed countries, for benefit of the former. Additional tests involving regression analysis support this finding. However, significance of the results decreases when taking period heteroskedasticity and general correlation of observations into account.
The resource curse has four main negative effects, which I transformed into four hypotheses. The fifth hypothesis combined economic growth with natural resources.
Hypothesis 1a: the exchange rate is inversely related to economic growth. Hypothesis 1b: the unemployment rate is inversely related to economic growth.
Both hypotheses are rejected. The first hypothesis is not accepted, because the correlation between the exchange rate and economic growth is positive instead of negative. Indirectly, Dutch Disease is not supported, possibly because an appreciation of the exchange rate does not necessarily mean a lower growth of GDP. I expected a negative relationship, and still believe that the chosen variable is correct, because its relation with economic growth is significant. The problem is that the dampening effect is hard to measure. What would the economic development have been without the exchange rate appreciation? That is difficult to predict.
Hypothesis 2a: transparency is positively correlated with economic growth.
Hypothesis 2b: foreign inflow investment is positively correlated with economic growth.
The first hypothesis is rejected, because transparency was to complex to measure and made the equation instable. So this variable was left out during further tests.
The second hypothesis is confirmed by the tests; it can be seen when looking at the variable INV corresponding with direct foreign inflow investment. The direction of the sign is negative and significance is high. Both variables indirectly represent extreme rent-seeking behaviour, the second negative aspect of natural resource abundance according to the theory. However, developed countries are characterized by well-developed instititutional structures, so the occurrence of extreme rent-seeking behaviour is less likely in these countries than in developing countries.
Hypothesis 3: debt is negatively correlated with economic growth.
This hypothesis is rejected when looking at the debt of a country, since the result was not significant. This is not surprising, since in developed countries one can assume that debt problems do not play a significant role. In addition, debt may have been not the perfect variable to cover the term overconfidence, as was mentioned in part II. Furthermore, the presence of high-developed institutions is characteristic for developed nations, as was explained in section II. This leads to stable political systems, which are a required condition for sustainable economic growth with the exploitation of natural resources.
Hypothesis 4a: public expenditure on education is positively related to economic growth. Hypothesis 4b: labour productivity is positively correlated with economic growth.
Hypothesis 4c: investment in R&D is positively correlated with economic growth.
Hypothesis 4a is rejected, because the relation between public expenditure on education and economic growth was highly insignificant. For this reason, and the fact that H4b and H4c are also good measures of human capital, the variable was left out during further tests.
parameters. This is one of the most important results of this paper. The correlation is unilateral, so the dependent variable economic growth is not affecting labour productivity.
Hypothesis 4c is rejected; research and development is significant, but negatively correlated with economic growth. It seems that the economic results of R&D are more long-term orientated; costs on the short run might results in profit on the long run, but the time period in this paper is too short to account for that.
So indirect evidence was found for the link between natural resource abundance and neglect of investment in human capital .Economic policies that result in the productive reinvestment of a substantial portion of resource rents are strongly recommended to avoid this trap.
H5: the economic growth of resource-rich developed countries is lower than the economic growth of resource-poor developed countries.
This hypothesis is justified, because of the negative sign of the dummy variable for natural resource abundance, NRD. However, significance was strongly decreasing when accounting for heteroskedasticity and autocorrelation. This was not only the case for the dummy, but also for the other variables. Shocks earlier in time have consequences for the current situation. It seems very hard to prove the existence of the resource curse, because it is hard to isolate and analyse the independent variables outside their context.
It seems that natural wealth is not per se deteriorating economic development in developed countries. If exploited wisely, resource abundance can be turned into a growth industry that provides a solid and even long-term foundation for economic growth. This is hardly a curse.
Some shortcomings in exploring the resource curse remain, like the ignorance of the degree of natural resource possession. I simply divided countries in resource-poor and resource-rich groups, but more research needs to be done on the impact of more or less abundance on economic growth. Currently, most industrialized countries depend on oil and natural gas for most of their energy needs. In the United States, for example, these two resources together supply 67% of all energy consumed. Both are depletable, non-recyclable sources of energy. Crude oil proven reserves peaked during the 1970s and natural gas peaked in the 1980s in the United States and Europe, and since that time, the amount extracted has exceeded additions to reserves (Tietenberg 2000).
In addition, it is hard to involve the speed of resource depletion, possible increase of proven reserves, or the moment of resource discovery, which variates between countries. Maybe some methodological solution can be found for this; increasing the relatively low R-squared result obtained now.
REFERENCES
Auty, Richard M., Resource-based industrialization: sowing the oil in eight developing
countries (New York: Oxford University Press 1990).
Auty, Richard M., Sustaining development in mineral economies: the resource curse
thesis (London: Routledge 1993).
Bodin, Jean, The Six Books of the Commonwealth (1576; reprint Cambridge: Cambridge University Press 1992).
Bulte, Erwin H., Richard Damania and Robert T. Deacon, ‘Resource intensity, institutions, and development’, World Development 33 (2005) 1029-1044. Caljé, P.A.J. en J.C. den Hollander, De nieuwste geschiedenis (Utrecht: Het Spectrum
1998).
Carter Hill, R., William E. Griffiths and George G. Judge, Undergraduate econometrics (Hoboken: John Wiley & Sons, Inc. 2001).
Deacon, Robert T. and Bernardo Mueller, Political economy and natural resource use, Economics Working Paper Series 1169 (2004).
Dunning, Thad, ‘Resource dependence, economic performance, and political stability’,
Journal of Conflict Resolution 49 (2005) 451-482.
East Med, ‘Growth To Moderate Further’, Middle East Monitor 15 (2005) 4.
Ekman, Ivar, ‘Trouble brewing in oil-rich Norway; after years of restraint, some want to tap forbidden fund’ The International Herald Tribune (November 19, 2005). Freedman, David, ‘From Association to Causation: Some Remarks on the History of
Statistics’, Statistical Science 14 (1999) 243-258.
Gallup, J.L., J.D. Sachs and A.D Mellinger, ‘Geography and economic development’,
International Regional Science Review 22 (1999) 179-232.
Gelb, A.H., Windfall gains: blessing or curse? (New York: Oxford University Press 1988).
Granger, C.W.J., ‘Investigating Causal Relations by Econometric Methods and Cross-Spectral Methods’, Econometrica 34 (1969) 424-438.
Gylfason, Thorvaldur, Natural resources, education, and economic development, For the 15th Annual Congress of the European Economic Association (2000).
Hodler, Roland, The curse of natural resources in fractionalized countries, Diskussionsschriften University of Bern (2004).
Isham, Jonathan, et al, The varieties of resource experience. How natural resource export
structures affect the political economy of economic growth, Middlebury College
Economics Discussion Paper no. 03-08R (2004).
Krueger, Anne O., ‘The political economy of the rent-seeking society’, American
Economic Review 64 (1974) 291-303.
Madrick, Jeff, ‘Far from a 'curse,' natural resources can form the basis for economic growth’, New York Times (February 19, 2004).
Mahmud, Muhammad and Abdul Rashid, ‘Labor productivity and economic growth, what causes what: an empirical analysis’, Journal of Management and Social
Sciences 2 (2006) 69-81.
Mehlum, Halvor, Karl Moene and Ragnar Torvik, ‘Institutions and the resource curse’,
The Economic Journal 116 (2006) 1-20.
Messemer, Clarisse and Richard W. Parks, ‘Bootstrap methods for inference in a SUR model with autocorrelated disturbances’ University of Washington, Department of
Economics (2004).
Murshed, S. Mansoob, et al, Natural resource endowment types and longterm growth, Institute of Social Studies Working Paper no. 391 (2004).
Neumayer, Eric, ‘Does the ‘Resource Curse’ hold for growth in genuine income as well?’
London School of Economics (2004).
OECD, Economic Surveys: Norway (2005).
O’Higgins, Eleanor R.E., ‘Corruption, underdevelopment, and extractive resource industries: addressing the vicious cycle’, Business Ethics Quarterly 16 (2006) 235-254.
Papyrakis, Ellisiasos and Reyer Gerlagh, ‘The resource curse hypothesis and its
transmission channels’, Journal of Comparative Economics 32 (2004) 181–193. Ploeg, Rick van der, ‘De vloek van natuurlijke rijkdommen’, Het Financieele Dagblad
(February 11, 2006).
Robinson, James A., Ragnar Torvik and Thierry Verdier, ‘Political foundations of the resource curse’, Journal of Development Economics 79 (2006) 447-468.
Sachs, Jeffrey D., and Andrew M. Warner, Natural resource abundance and economic
growth, NBER Working Paper Series no.5398 (1995).
Sachs, Jeffrey D., and Andrew M. Warner, ‘Natural resources and economic
development. The curse of natural resources’, European Economic Review 45 (2001) 827-838.
Stiglitz, Joseph, ‘Rijkdom is vaak een vloek ; Diamanten en olie leiden vaak tot oorlog en corruptie’, Volkskrant (August 12, 2004).
Stijns, Jean-Philippe, ‘Natural resource abundance and economic growth revisited’,
Resources Policy 30 (2005) 107-130.
Taheri, Abbas A. and Rodney Stevenson, ‘Energy price, environmental policy, and technological bias’, Energy Journal 23 (2002) 85-107.
Tietenberg, Tom, Environmental and natural resource economics (Reading: Addison-Wesley 2000).
Torvik, Ragnar, ‘Learning by doing and the Dutch disease’, European Economic Review 45 (2001) 285-306.
Torvik, Ragnar, ‘Natural resources, rent seeking and welfare’, Journal of Development
Economics 67 (2002) 455-470.
Wijnbergen, Sweder van, ‘Inflation, employment, and the Dutch Disease in oil-exporting countries: a short-run disequilibrium analysis’, Quarterly Journal of Economics 99 (1984) 233-250.
http://www.bp.com http://www.imf.org http://www.oecd.org
http://www.transparency.org
APPENDIXES
Table 1: classification
Resource rich
Country Natural resource
Table 2: Variables
Code Variable Unit Negative effect / hypothesis
Source Predicted sign
GDP9103 GDP growth Annual change
in % - IMF dependent variable U Unemployment % of labour force 1 IMF / WDI -XR Exchange rate change Annual change in % 1 OECD
-TR Transparency Index, value
between 0-10
2 Transparency
International +
INV Investment Inflow FDI as
% of GDP 2 WDI + DE Debt % of GDP 3 OECD -ED Education Public expenditure on education as % of GNI 4 WDI + LP Labour productivity GDP per hour worked, annual change in % 4 OECD +
RD R&D Gross domestic
-Table 3: Statistics GDP U XR INV DE LP RD Mean 2.778805 7.922814 100.8458 5.135342 65.20259 2.366760 1.818099 Median 2.800000 7.200000 101.7714 1.866020 62.23801 2.303438 1.870000 Maximum 11.70000 24.20000 137.3661 349.9594 154.0327 8.341024 4.290000 Minimum -6.900000 1.300000 55.36202 -0.334853 5.191555 -4.122943 0.360000 Std. Dev. 2.424346 4.356115 10.64047 22.79311 30.04402 1.852164 0.764000 Skewness 0.135127 1.233935 -0.716606 13.53796 0.579586 0.027515 0.236790 Kurtosis 5.581490 4.683304 5.796312 201.7232 3.390090 4.225904 2.670172 Jarque-Bera 73.82767 97.79102 108.1967 440788.1 16.39201 16.50181 3.649840 Probability 0.000000 0.000000 0.000000 0.000000 0.000276 0.000261 0.161231 Sum 730.8257 2083.700 26522.45 1350.595 17148.28 622.4579 478.1600 Sum Sq. Dev. 1539.892 4971.643 29663.55 136115.8 236492.6 898.7942 152.9284 Observations 263 263 263 263 263 263 263
Table 4: LS regression with fixed effects
Dependent Variable: GDP Method: Panel Least Squares Date: 07/27/06 Time: 15:05 Sample: 1991 2003
Cross-sections included: 27
Total panel (unbalanced) observations: 261
Variable Coefficient Std. Error t-Statistic Prob. U -0.042882 0.030831 -1.390887 0.1655 XR 0.025700 0.012013 2.139375 0.0334** INV 0.001565 0.004958 0.315758 0.7525 DE -0.008409 0.004279 -1.965046 0.0506** ED 0.021638 0.088363 0.244883 0.8068 LP 0.677566 0.064856 10.44717 0.0000* RD -0.519331 0.182234 -2.849806 0.0048* NRD -2.192956 0.530000 -4.137651 0.0000* C 0.444829 1.504691 0.295628 0.7678 Effects Specification Period fixed (dummy variables)
Sum squared resid 755.8947 Schwarz criterion 4.348979 Log likelihood -509.1143 F-statistic 12.36399 Durbin-Watson stat 1.114327 Prob(F-statistic) 0.000000
One star indicates a variable statistically different from 0 at a 1% level of significance, 2 stars at 5%, 3 stars at 10%.
Table 5: LS regression with fixed effects, ED excluded Dependent Variable: GDP
Method: Panel Least Squares Date: 07/27/06 Time: 15:07 Sample: 1991 2003
Cross-sections included: 27
Total panel (unbalanced) observations: 263
Variable Coefficient Std. Error t-Statistic Prob. U -0.041353 0.030231 -1.367916 0.1726 XR 0.024194 0.011940 2.026251 0.0438** INV 0.001412 0.004934 0.286134 0.7750 DE -0.008900 0.004184 -2.126997 0.0344** LP 0.683014 0.064374 10.61016 0.0000* RD -0.504837 0.178592 -2.826768 0.0051* NRD -2.228301 0.523079 -4.259972 0.0000* C 0.693403 1.457770 0.475660 0.6347 Effects Specification Period fixed (dummy variables)
R-squared 0.503607 Mean dependent var 2.778805 Adjusted R-squared 0.464795 S.D. dependent var 2.424346 S.E. of regression 1.773596 Akaike info criterion 4.056894 Sum squared resid 764.3911 Schwarz criterion 4.328541 Log likelihood -513.4815 F-statistic 12.97537 Durbin-Watson stat 1.119041 Prob(F-statistic) 0.000000
One star indicates a variable statistically different from 0 at a 1% level of significance, 2 stars at 5%, 3 stars at 10%.
Table 6: EGLS regression, ED excluded
Dependent Variable: GDP
Method: Panel EGLS (Period SUR) Date: 07/27/06 Time: 16:29 Sample: 1991 2003
Cross-sections included: 27
Linear estimation after one-step weighting matrix
Variable Coefficient Std. Error t-Statistic Prob. U -0.046261 0.037638 -1.229101 0.2202 XR 0.024065 0.012034 1.999695 0.0466** INV 0.026051 0.011201 2.325740 0.0208** DE -0.002120 0.004769 -0.444457 0.6571 LP 0.642287 0.058224 11.03138 0.0000* RD -0.679512 0.209208 -3.248018 0.0013* NRD -1.322047 1.319908 -1.001621 0.3175 C 0.593386 1.451552 0.408794 0.6830 Weighted Statistics
R-squared 0.610046 Mean dependent var 0.943520 Adjusted R-squared 0.599341 S.D. dependent var 1.279584 S.E. of regression 0.809947 Sum squared resid 167.2834 Durbin-Watson stat 1.953644
Unweighted Statistics
R-squared 0.287324 Mean dependent var 2.778805 Sum squared resid 1097.445 Durbin-Watson stat 0.939275
One star indicates a variable statistically different from 0 at a 1% level of significance, 2 stars at 5%, 3 stars at 10%.
Table 7: Histogram Residuals
0 4 8 12 16 20 24 28 32 36 -10.0 -7.5 -5.0 -2.5 0.0 2.5
Table 8: Histogram Residuals, Period SUR 0 10 20 30 40 50 60 70 80 -3 -2 -1 0 1 2
Series: Standardized Residuals Sample 1991 2003 Observations 263 Mean -0.020391 Median 0.000000 Maximum 2.497075 Minimum -2.776394 Std. Dev. 0.798792 Skewness -0.196757 Kurtosis 4.330349 Jarque-Bera 21.09130 Probability 0.000026
Table 9: Fitted, actual and residual values graph
-12 -8 -4 0 4 8 -8 -4 0 4 8 12 25 50 75 125 175 227 261 300 350
Residual Actual Fitted
