THE PERFORMANCE OF NONRENEWABLE RESOURCE FUNDS

(1)

THE PERFORMANCE OF

NONRENEWABLE RESOURCE FUNDS

Abstract

Governments of oil-producing countries face many challenges. One of the largest challenges is in the field of fiscal policies. Oil revenues are volatile, and when this volatility becomes a characteristic of government expenditure, this can reduce economic growth. This research shows that government expenditure of oil-producing countries is procyclical and that the volatile oil revenues are an important factor in explaining government expenditure. The analysis also shows that the most proposed solution for this dilemma, a nonrenewable resource fund, is not able to reduce this procyclicality. Evidence can not prove the effectiveness of such funds.

Keywords: Oil-producing countries, Government expenditure, Procyclicality, Nonrenewable resource funds

Written by Supervised by

Erica Ross Dr. D.J. Bezemer

Student number: 1382454 d.j.bezemer@rug.nl

Email: e.m.ross@student.rug.nl Erica.m.ross@gmail.com

Telephone: 06-21204424

Master’s thesis International Economics and Business University of Groningen

(2)

List of abbreviations:

BP British Petroleum

DNB De Nederlandsche Bank

EIA Energy Information Agency GDP Gross Domestic Product

HIPC Heavily Indebted Poor Countries

HP Hodrick Prescott

IMF International Monetary Fund NRF Nonrenewable Resource Fund

OECD Organization for Economic Co-operation and Development OLS Ordinary Least Squares

OPEC Organization of Petroleum Exporting Countries SWF Sovereign Wealth Fund

(4)

1: Introduction

“All in all, I wish we had discovered water”

Sheik Yamani, Oil Minister of Saudi Arabia1

At first sight it seems like a genuine blessing; the discovery of oil fields. With the export of the oil, there can come an end to balance of trade deficits, the economy could prosper and the population should be able to benefit from the new resources that flow into the country. Unfortunately, this ideal picture is not the reality for many oil-exporting countries. As the citation above illustrates, even the largest oil producer in the world, Saudi Arabia, has not always profited from its richness.

Already in the 1950s, scholars reported on the negative effects for the economy when a country was highly dependent on the export of natural resources, including oil ( Prebisch, 1950; Singer, 1950; Hirschman, 1958). Their arguments were that the extraction of natural resources does not lead to forward and backward linkages with the rest of the economy, and that the terms of trade of these products decline over time compared to manufactured products.

In the 1970s and 1980s, the attention of scholars on the topic of problems for oil-exporting countries further increased. The phenomenon of the Dutch disease was widely discussed, most notably by Corden and Neary (1982) and Van Wijnbergen (1984). Corden and Neary sought to explain the consequences of a booming export sector for a small, open economy. Like in the Netherlands for the gas-sector, hence the terminology ‘Dutch’ disease, this booming export sector is usually of an extractive kind, like the oil industry. Apart from this oil-sector, in their framework the economy is divided in two other sectors: The non-booming export sector and the non-traded goods sector, which supplies domestic residents and includes services and construction. Corden and Neary show that through the mechanisms of the Dutch disease, the non-booming export sector gets crowded out by the two other sectors. When the booming export sector is a natural resource and the non-booming sector is manufacturing the Dutch disease will lead to de-industrialization.

1

(5)

The mechanisms of the Dutch disease are as follows. When a country experiences a sharp increase in oil exports or in the price it receives for its oil, this initially raises national income, because of the foreign exchange that flows into the country. The assumption is that not all new foreign exchange is spent on imports, so that the oil-boom has a direct impact on the country’s money supply and demand for domestically produced goods and services.

Because of the increase in income, there will be an increase in demand for services and other non-traded goods. This in turn will lead to an appreciation of the real exchange rate, both with fixed and floating exchange rates: With fixed exchange rates, the prices of non-traded goods will increase, but the price of the non-booming export sector cannot change because this price is set internationally. Therefore, the non-traded goods will become more expensive in terms of traded goods. With floating exchange rates, the increased supply of foreign currency drives up the value of the domestic currency. This causes an appreciation of the real exchange rate through a rise in the nominal exchange rate.

In either way, the appreciation of the real exchange rate weakens the competitiveness of the tradable (manufacturing) sector, causing it to shrink. This process is called the ‘spending effect’ of the Dutch disease. Another effect is the ‘resource movement effect’, which means that resources like capital and labour will shift from the manufacturing sector into the oil-sector or into the non-traded service industry that faces a higher demand. This too causes the production of the manufacturing sector to shrink.

(6)

achieve intergenerational equity and save an amount of the income in order to ensure that future citizens will also benefit (Davis et al., 2001).

Another problem that governments have to deal with is the high risk of volatility of the real exchange rate. Bleaney and Greenaway (2001) are among the many scholars that have found evidence for a negative relationship between exchange rate volatility and the growth in investment. For oil-exporting countries, when the oil price is low and it is profitable for the traded goods sector to expand production, this contains a high risk because it is likely that in the near future oil prices will increase. This will cause an appreciation of the currency, resulting in losses for the sector and making the investment unprofitable (Devlin and Lewin, 2004). Investment is necessary for future expansion of industries and for future economic growth, making exchange rate volatility possibly harmful.

One other crucial problem that governments have to face is the volatility of government revenues, which is caused by the volatility of the price of oil. It is on this aspect of the economics of oil that this research will focus.

The price of oil tends to be very volatile, as shown in graph 1. Another characteristic of oil prices is that they tend to be very hard to predict. Research done on this topic suggests that oil prices do not have well-defined time-invariant averages (Engel and Valdés, 2000).

GRAPH 1: OIL PRICE (CONSTANT 2004 US$)

0,00 10,00 20,00 30,00 40,00 50,00 60,00 70,00 80,00 90,00 100,00 197 0 1973 197 6 1979 1982 1985 1988 199 1 1994 199 7 2000 200 3 2006

(7)

Engel and Valdés use time series models to try to forecast future oil prices between 1957 and 1999. They found that none of the models performed significantly better than a simple random walk model. This means that the best prediction for the oil price in all future periods is the current oil price. The standard deviation grows over time, so the further one wants to predict in the future, the more difficult it will become to say anything sensible about the price. Especially the large fluctuations in price, for example in 1973 and 1979, could not be predicted at all.

Because governments of oil-exporting countries are usually very dependent on the revenues of oil, revenues of oil-exporting government are very volatile too. When a government decides to invest or consume these revenues immediately when they are obtained, volatility will also become a characteristic of fiscal policy.

For the purpose of this research, government spending in oil-producing countries which has this volatile characteristic and which rises with rising oil revenues and falls when oil revenues falls, will be defined as procyclical fiscal policy. This terminology is derived from research on fiscal spending reactions in the different stages of the business cycle, for example by Talvi and Végh (2000) and Lane (2003). For these scholars, government expenditure is procyclical when it rises when total national output rises, and falls when national output falls. For oil-producing countries, the terminology is strikingly appropriate when only looking at the total national output of oil, as such a substantial part of the economy is driven by the production of oil: Appendix 1 shows that an average of 72% of all merchandise exports is derived from the export of oil.

(8)

unanticipated changes that affect the government’s budget constraint. Neither Keynes nor Barro recommended a procyclical fiscal policy, however.

What distinguishes oil-producing countries from other countries regarding choices on fiscal policy is the much larger volatility of revenues, as described above. Pursuing a procyclical fiscal policy will transfer this volatility into the budget of government. Evidence seems to suggest that this can be very harmful for long-run economic growth. The negative relationship between government expenditure volatility and economic growth has been found by Furceri (2007), amongst others. He found that a 1% increase in government expenditure volatility causes a decrease of 0.78% in the long-run rate of growth. The question whether oil-exporting countries pursue a procyclical fiscal policy is therefore very relevant, and will be the first central question addressed in this thesis.

To counter a fiscal policy which is too procyclical and therefore very volatile the establishment of a Nonrenewable Resource Fund (NRF) has often been proposed. In 2008 the Saudi government made its intentions to establish such an NRF clear. Saudi Arabia is by no means the first large oil producer that is thinking of establishing a fund. Already in 1953 Kuwait started its Kuwait Investment Authority, which is a government investment vehicle that manages the oil revenues separately from the official reserves. Other oil-exporting countries were soon to follow, and especially in more recent times with rising oil prices from 2001 up to the second half of 2008, NRFs were seen as interesting options to make sure that not all windfalls are immediately spend.

(9)

There are a number of important characteristics of NRFs that become evident in the definition. First of all, the NRF is funded by foreign exchange assets, obtained through exports of natural resources. This means that only the countries with substantial natural resource exports are taken into account. This excludes Malaysia, for example, which does have a government investment vehicle, but most of its assets are obtained through the export of manufactured goods.

Secondly, the assets are managed separately from official reserves. This means that an actual entity has to be established which has the responsibility of managing the NRF. A country like Saudi Arabia, which does manage its oil revenues but at this stage not yet in a separate institution, is therefore not seen as a country with an NRF. Moreover, the condition that the assets are mainly invested abroad means that oil-exported countries that have funds in order to invest in domestic projects are not included either.

An NRF can only invest the assets obtained from oil that are received by the government. The percentage of total oil revenues that government receives varies from country to country, and from oilfield to oilfield, as royalty rates and the petroleum revenue taxes differ (Yergin, 1991; Mommer, 2002).

(10)

fund to make sure that not all government revenues of natural gas are spend by the ruling cabinet (De Nederlandsche Bank, 2008).

Not only is the functioning of NRFs relevant for oil-producing countries themselves; NRFs and other types of Sovereign Wealth Funds (SWFs)2 are a topic of heavy debate in Europe and the United States, where most of the reserves in the funds are being put into use. The popularity of the funds has possible impacts on the functioning of the global financial markets and some western policymakers are suspicious about the intentions of these funds when buying large shares of their domestic companies. This has led to very heavy debates recently, where Germany for example has decided not to permit non-European foreign entities to own more than 25% of shares in German companies and the French president Sarkozy called for the establishment of a European Sovereign Wealth Fund.

The remaining part of this paper will proceed as follows. In section 2, literature on procyclicality and on the functioning of NRF will be discussed. Section 3 discusses the methodology that will be used. Section 4 describes the data with which the analysis will take place. Section 5 analyzes the empirical results. Lastly, in section 6, conclusions, as well as important contributions and limitations of this research will be discussed.

2

(11)

2: Literature review

In this chapter, an overview of the relevant literature regarding procyclicality and NRFs is given. In paragraph 2.1, various researches on procyclicality will be analyzed. The second paragraph focuses on procyclicality in oil-producing countries, while the third paragraph discusses the studies on the effectiveness of Nonrenewable Resource Funds regarding the reduction of this procyclicality.

2.1: Procyclicality3

Much research has been done to answer the question to what extent governments in all types of countries pursue procyclical policies. Extensive studies on this topic took place in the early 1990s. Fiorito and Kollintzas (1994) and Fiorito (1997) did research on the procyclicality of government consumption and output in G-7 countries. They found that there seems to be no clear correlation between government consumption and output; the average correlation lies around zero. Therefore these authors concluded that procyclicality (or counter cyclicality) did not seem to exist, at least in G-7 countries.

Later research shifted to fiscal policy in other countries than G-7 countries, most notably in Latin America. Gavin et al (1996) and Gavin and Perotti (1996) found that in Latin America it is the case that fiscal policy tends to be procyclical. During economic expansions government expenditure increases while taxes fall, while the opposite is true during times of recession. It thus appears that fiscal policy in Latin America does conform to neither Keynesian prescriptions, which advocates countercyclical policy, nor Barro’s recommendations, who advises that fiscal policy should remain neutral over the business cycle.

The explanation that Gavin and Perotti (1996) give for this procyclicality is that low-income countries are not able to receive international credit in times of a recession, and are therefore forced to cut down expenditure. This view is opposed by Talvi and Végh (2000) and Lane (2003). Talvi and Végh argue that not all developing countries are cut

3

(12)

out of the international credit market during bad economic times. They have found that also OECD countries run procyclical policies, even though the policies of developing countries are more procyclical. The reason they give for this observation is that output in developing countries is far more volatile, and therefore tax revenues are more volatile too. Lane (2003) has also found that countries with volatile output are most likely to run procyclical fiscal policies.

This reasoning is very interesting in the light of the research proposed in this paper: Oil revenues, and therefore government revenues, are very volatile in the case of oil-producing countries too. It can be expected that procyclicality will exist in these countries, which leads to the first hypothesis of this paper:

Hypothesis 1: There exists a positive causality between government expenditure of

oil-exporting countries and their oil revenues.

2.2: Procyclicality in oil-producing states

One important research on procyclicality in oil-producing countries has been done by Gelb (1989). He describes the way in which oil price fluctuations have harmed producers of oil. He argues that the oil shocks were unprecedented, regarding scope and unexpectedness and therefore regarding the harm that they have caused. He finds that investment losses induced by oil price swings have been huge throughout history.

(13)

per country, but should roughly be two-thirds of the windfall. These savings should be invested abroad, one task that an NRF could of course easily perform.

The following table shows the ways in which windfall gains were used in the six oil-exporting countries analyzed by Gelb after the first rise of oil prices, in the period 1974-1978. The average percentage of the windfall that had been saved during this period of very high oil prices was only 17,9%. It is striking, moreover, that in these years of large revenues, two out of the six countries managed to actually run balance of payments deficits. These deficits increased even more when oil prices fell in the 1980s.

TABLE 1: USE OF WINDFALL GAINS

% of windfall gain

Country % of GDP from oil Private expenditure Public consumption Public investment Savings

Algeria 27.1 13.3 5,2 97,4 -15,9 Ecuador 16.7 17.4 32,9 28,7 21.0 Indonesia 16.0 2.5 15.0 49,4 33.1 Iran 36.9 -0,8 27.6 27,1 46,1 Nigeria 22.8 -16,2 18.4 85,5 12.3 Venezuela 10.7 48.6 15.0 45,8 -9,3 Average 21.7 6.2 19.4 56.5 17.9

Source: Talvi and Végh (2000), based on Gelb (1989)

2.3: The effectiveness of NRFs

The most influential research on the effectiveness of Nonrenewable Resource Funds is by Davis et al (2001), who studied the behaviour of governments of five countries with an NRF and compared them to seven countries without an NRF, for the period 1965 to 1999. Davis et al. did not find any convincing evidence for the effectiveness of NRF’s and did not find any evidence for a change in fiscal policy after the introduction of an NRF.

(14)

Not all aspects of the research that Davis et al have done are entirely convincing. One problem is that the data used on which country when established an NRF are sometimes wrong. For example, the United Arab Emirates is used in the control panel of countries without an NRF, even though the emirate with most oil in its soil, Abu Dhabi, established an NRF in 1976. Another example is Kuwait, which was first to establish an NRF in 1953. To test whether Kuwait has changed its behaviour regarding procyclicality, however, Davis et al use the year of 1976.

An addition that can be made to the research of Davis et al. is the use of control variables which can also influence the procyclicality of government expenditure. Evidence suggests that GDP growth is an important factor in explaining procyclicality, and will be included in the analysis, as well as other control variables. The analysis of Davis et al will also be simulated without these control variables, to examine whether countries that have recently established an NRF have changed their behaviour following to this methodology.

Next to Davis et al., Fasano (1998) has reviewed the experience with oil and copper funds in Chile, Norway, Venezuela, Alaska, Oman and Kuwait in a qualitative manner. Contrary to Davis et al., he did find that the funds contributed to a decrease in government expenditure volatility, thereby enhancing the effectiveness of fiscal policy.

(15)

To state a hypothesis on the influence of an NRF on government expenditure volatility is problematic. As described above, the outcomes in the other literature are mixed and ambiguous. It is one of the main purposes of NRFs, however, to decrease volatility in government expenditure. It can be expected that when a fund is set up with the purpose of decreasing the volatility, government will strive to make this work properly (Bacon and Tordo, 2006). This leads to my second and third hypotheses.

Hypothesis 2: Oil-exporting countries with an NRF will experience less government

expenditure procyclicality than oil-exporting countries without an NRF.

Hypothesis 3: When an NRF is established, government expenditure procyclicality will

decrease.

(16)

3: Methodology

This research will use several econometric techniques to determine the truthfulness of the hypotheses. These econometric techniques are largely derived from the techniques in papers on procyclicality and on the effectiveness of NRFs, which were discussed in the literature review. In both this section and the following section discussing the empirical results, I will firstly discuss procyclicality of government expenditure in oil-producing countries in general (hypothesis 1); what follows are the with-without approach (hypothesis 2) and the before-after approach (hypothesis 3).

3.1: Hypothesis 1: procyclicality of oil-producing states

In accordance with Talvi and Végh (2000) procyclicality for oil-producing countries is defined as a positive correlation between yearly oil revenues and government expenditure of that same year. The following formula is used to determine procyclicality as an average for all examined countries:

(

)

( ) n X G n n i i

∑

1 , = ρ (1)

(17)

(

)

t X G t t t t

∑

=1 , ρ (2)

Where t represents time. To determine whether procyclicality is a common feature in all oil-producing states, a t-test is performed to test whether the average procyclicality is significantly larger than zero.

Talvi and Végh use Hodrick-Prescott (HP) filtered data, which is common for measurements of the business cycle (Sanchez de Cima, 2003; Agenor et al., 1999). HP filtered data are used to smooth a nonlinear time series, and are often used for measurements of the business cycle. The reasoning of HP filters is as follows: A time series y_t is made up from a trend component τ , and a cyclical component, c, such that

t

t c

y =τ_t + . The equation seeks for a τ that will minimize the equation

(

)

[

(

) (

∑

= − = + −

−

+

−

τ τ

τ

λ

τ

1 1 2 2 1 1 2 t t t t t t t t

y

)]

₍₃₎

Where λ is any adequately chosen number, which penalizes variations in the growth rate of the trend component. Usually 1600 is chosen as an adequate λ . The larger λ is, the more the trend component is penalized, and the smoother the adjusted time series will become. In some cases this can be preferred to the smoothing by first differences, for example because they show less fluctuation. Gelb (1988) has shown, however, that for measurements of the correlation between oil revenues and government expenditures first differences suffice. A big disadvantage of HP-filters is that if one-time permanent shocks occur, the filter will generate shifts in the trend that do not actually exist (French, 2001). This is problematic when estimating the impact of oil revenues, as large shocks in the oil price did occur in the period of our sample. Therefore, first differences will be used to estimate the correlation in this case.

(18)

Many economists have tried to measure the impact of a certain policy on an economic variable, for example the introduction of an IMF program on economic growth, or the implementation of the HIPC-initiative on poverty reduction. In this respect, the research on the impact of Nonrenewable Resource Funds on procyclicality of government expenditure is similar to many previous studies, and also faces some of the implications that many scholars have tried to solve in the past. Because desirable as these researches definitely are, it is difficult to estimate the impact of something, if you do not know what would have happened without this ‘something’. A lot of solutions have been proposed to deal with this problem, all with their own implications nonetheless. This research will use two of these solutions, namely the ‘with-without-’ and the ‘before-after’ approach.

The with-without approach compares cases that do have a certain characteristic with cases that do not. In this case, countries with an NRF will be compared with countries without an NRF. An advantage of this is that it can be done over the same time period, therefore excluding factors that have influenced all countries being considered and that do not have anything to do with the introduction of a fund. A problem, however, is that countries are not similar in all other respects (Bird, 2007). This can be partly solved by the introduction of control variables, but not all differences of countries can be observed. The before-after approach compares the period before a certain change, the establishment of an NRF in this case, with the period afterwards. An advantage here is that you do not compare a country with some other country that has fundamental characteristics that are entirely different. The before-after approach does implicitly assume, however, that all other things remain constant, an assumption that is not totally viable in most circumstances.

The two approaches vary from each other, and therefore it gives many additional insights to perform both with-without tests and before-after tests. A combination of these results will make it easier to draw firm conclusions out of the analyses.

3.3: Hypothesis 2: The with-without approach

(19)

extended version of the measurement of procyclicality specified in 4.1 will be used, as well as a regression-based measure of cyclicality, which is a variation of a model developed by Lane (2003).

The model of Talvi and Végh expressed in equation (1) measures the correlation between government spending and oil revenues (ρ(G

1,X1)). This model can also be used to test the hypothesis that procyclicality in certain countries is larger than in other countries. To test whether average correlations between government expenditure and oil revenues differs between countries that do have an NRF and countries that do not, an F-test will be performed. This is a ratio of the two correlation coefficients, each of which follows a t-distribution. For countries that have an NRF, only the years in which the NRF was actually operating will be used to measure the correlation. Data of Nigeria, Qatar and Russia are only up to the year in which they established an NRF (2004, 2003 and 2004 respectively); they are thus seen as countries without an NRF. For all other countries the dataset starts in 1970 or in the year that oil production started and ends in 2005.

Lane (2003) has used another type of model to measure procyclicality. In his model, two steps have to be followed. Firstly, a country-by-country regression is estimated. For countries with an NRF, this regression is only estimated for the years in which they have had the NRF. The regression is of the form

t t G

t X

G =α +β ∗ +ε

(4)

Where t represents time. The regression will be run in first differences, to measure the elasticity of government expenditure (d(G)) with respect to the growth of oil revenues (d(X)). The coefficient βGi is therefore the index of cyclicality; a positive value implies

procyclical behaviour.

(20)

have shown first-order correlation in most of the cases (not reported), a correction for first-order serial correlation in the error term is included.

The relationship is explored in first differences, rather than in levels. This is based on the Dickey Fuller unit roots tests done for each country with an NRF. These unit root tests for stationarity fail to reject the hypothesis of a unit root in almost all cases, like appendix 4 shows. When the hypothesis of a unit root cannot be rejected for either nonrenewable resource export earnings or central government expenditure, this suggests that the regression should be run in first differences, or the spurious regression could lead to results with no economic meaning (Davis et al., 2001). A caveat for using the first difference, however, is that the oil revenues and the government expenditure should not be cointegrated: They should not share similar stochastic trends and their difference should not be stationary (Davis et al., 2001). A Johansen cointegration test is performed in order to test for cointegration. The results of these tests are found in Appendix 5, and show that there seems no reason to suspect cointegration in most cases.

Nonstationarity in government expenditure and oil revenues variables in equation (4) does not come as a surprise, based on the properties of oil prices. As mentioned in the introduction, Engel and Valdés (2000) found that the best model to predict oil prices is a simple random walk model, so that the price of oil for this year is highly correlated with the price in the previous year. Mean and variance do not seem to be constant over time, and the condition that the covariance between two values does not depend on the actual times at which the variables are observed, does not seem to hold (Hill, Griffiths and Judge, 2001).

(21)

The second step of the model is to explain cross-country variation. This is done by a regression of the form

i i i i i α λ GDP λ OPENC λ NRF λ VOL ν βˆ = + 1* + 2* + 3* + 4* + (5)

Where are the set of estimated parameters from equation (4), GDPi is average Gross

Domestic Product growth per capita in country i over the years that have been measured;

OPENC is average trade openness; NRF is a dummy for the countries that have an NRF

and VOL is the volatility of GDP per capita. A clear description of the way in which these control variables are measured can be found in appendix 2; Descriptive statistics are provided in appendix 3.

i

βˆ

This second regression will be run in levels, and heteroskedasticity consistent. Reason for this is that it is likely that the variables are measured with different degrees of precision between countries (Hill, Griffith and Judge, 2001).

The most important variable for this research in equation 5 is the dummy for countries that have an NRF. λ is expected to have a negative sign, as hypothesis 2 3

expects an NRF to reduce procyclicality in producing countries compared to oil-producing countries that do not have an NRF. In previous researches, other variables have been proven to be important determinants forβˆias well and are therefore included.

According to Wagner’s law the demand for government services is income elastic and government services are a luxury good. Therefore, the share of public expenditure in GDP is expected to rise with GDP per capita. Lane (2003) found that GDP per capita could be of influence on procyclicality too; richer countries enjoy less procyclical government spending. He argues that richer countries are more capable to implement fiscal control procedures; therefore they are better able to adjust their revenues. Therefore,

1

λ is expected to be negative.

(22)

a country is more open, for example because of terms of trade volatility. Lane (2003) has found that the degree of openness can have a statistical significant negative influence on government expenditure procyclicality; λ2 is therefore expected to be positive and

significant.

A third possible determinant for β_i is volatility of GDP, which, following Rodrik (1998) is measured as the standard deviation of the GDP per capita growth rate. λ₄ is expected to be positive. High volatility of GDP per capita is the environment most conductive to generating procyclicality in oil-producing countries. When volatility is high, the tax base of government is volatile accordingly, which is expected to increase procyclical fiscal behaviour.

To check for multicollinearity, which means that explanatory variables are related to each other and move together in systematic ways, a Variance Inflation Factor (VIF) analysis will be performed. With a VIF analysis, all explanatory variables are individually regressed against all other variables. The VIF-value is

(

₂

)

1 1

R

− . There is no formal formula to determine when the VIF-value causes reasons for concern, but for larger dataset usually a value that exceeds 10 is reason to expect multicollinearity. With our relatively small dataset, a value that exceeds 5 for any of the variables will be reason to expect multicollinearity. The results of the VIF-analysis are shown in appendix 7. Multicollinearity is not found in this dataset. Therefore, it is justified to include all variables mentioned above.

(23)

multiple regression model is violated (Hill, Griffith and Judge, 2000). When running time series regressions for just one single country, omitted variables can cause a regression to be spurious when series are nonstationary (the covariance (yt, yt+s) depends not only on s

but also on t). Panel data analysis can control for these types of problems.

There are many different types of panel data analysis, all with their own advantages and disadvantages that might be of relevance for this research. The three most important models to choose from are the constant coefficients model, the fixed effects model and the random effects model. The constant coefficients model, which is also called the pooled regression model, is the simplest model to analyze panel data. In this model, all data are pooled and an ordinary least squares regression model is run. A constant coefficients model can be run when neither country nor temporal effects are statistically significant. A fixed effects model has constant slopes but intercepts that differ, over country and/or time. A fixed effects model is more specific that a constant coefficients model, but has the large drawback that it can have too many cross-sectional units of observation requiring too many dummy variables for their specification (Yaffee, 2003). This can lead to insufficient information to adequately perform statistical tests. A random effects model is a regression with a random constant term, and the assumption that the intercept is a random outcome variable (Yaffee, 2003). An important condition for the random effects model is that the cross-sectional specific error term is uncorrelated with the errors of the variables.

Talvi and Végh (2000) have used a constant coefficients model for their panel data when testing for procyclicality of government expenditure. The relatively small dataset that I will use will make any significant outcomes when using the fixed effects model very unlikely. Furthermore, country dummies cannot be used, because in the cases of countries with an NRF they will correspond to the NRF dummy. I will therefore follow the example of Talvi and Végh, and use the constant coefficients model for this panel data analysis.

(24)

represents oil revenues, the correlations ρ(G1,x1) and ρ(G2,x2) will depend on the standard deviation of oil revenues σ(X). It can be expected that this is not a tremendous problem when comparing oil revenues, as all countries face the same volatility of oil prices. However, countries differ in their volatility of oil production, which could make the correlation coefficients misleading. A comparison between the two variations can therefore provide us with a unique way to determine the actual extent to which countries with an NRF pursue less procyclical policies.

3.4: Hypothesis 3: The before-after approach

In accordance with Davis et al. (2001) countries with an NRF are examined individually, to see whether procyclicality has been reduced after the establishment of an NRF. 12 countries with an NRF are being examined. These countries are (in alphabetical order) Algeria, Azerbaijan, Brunei, Chile, Iran, Kazakhstan, Libya, Norway, Oman, Trinidad and Tobago, the United Arab Emirates, and Venezuela. Kuwait is excluded because its NRF was established before the start of the sample period. Nigeria, Qatar and Russia are excluded because these countries set up an NRF after 2001; therefore not enough observations of these countries with an NRF are available. Additional explanation on the choice of countries is provided in the following chapter.

For each country separately a regression is run of the form

( )

G_t d

( )

X_t d

(

X_t

)

_t

d =α +β1* +β2* −1 +ε (6)

Where G is government expenditure, X is oil revenues and t is the year index. Initially, both contemporaneous and lagged oil revenues will be included. This is the preferred specification if both are significant. Otherwise, the less significant of the two is excluded (Davis et al., 2001).

(25)

Chile are run heteroskedasticity consistent. All other regressions are run OLS. The results of the White-Heteroskedasticity tests are reported in Appendix 7.

After the regressions in the form of equation (6) and the decision whether contemporaneous, lagged or both nonrenewable resource export earning terms are included, two tests will be performed in order to test the hypothesis that procyclicality is reduced after the introduction of an NRF. Firstly, a Chow test for structural stability will be performed. Hereby, it is tested whether the slopes and/or intercepts for NRF countries are different before and after the introduction of a fund. Basically, the Chow test uses two regressions:

d

( )

G_t =α₁+β1₁*d

( )

X_t +β2₁*d

(

X_t −1

)

+ε_t

d

( )

G_t =α₂ +β1₂*d

( )

X_t +β2₂*d

(

X_t −1

)

+ε_t (7) And tests the null-hypothesis that α1 is α2, that β11 is β12, and that β21 is β22. The

null-hypothesis is rejected when one of these tests shows significant results to prove that the coefficients are different.

Secondly, a test of the statistical significance of a post-NRF dummy variable will be performed. Regressions for each country and each dataset will be run of the form

( )

(

)

( )

(

t

)

t i t i t NRF X d NRF X d NRF X d X d G d ε β β β β β α + + + + + + = − − * * * * * * * 1 5 4 3 1 2 1 (8)

(26)

expenditure, and is expected to be negative. The distinction between the NRF-dummy and the interaction variable is made clear in the picture below:

(27)

4: Data description.

4.1: The dependent variable

The dependent variable in this research is government expenditure. There are, however, many ways in which government expenditure can be measured. Following Rodrik (1998), both real government consumption from the Penn World Tables, and government expenditure from the WDI will be used to examine the extent to which an NRF influences its volatility. Appendix 2 provides an overview of the definitions of the two measurements of government expenditure, as well as the definitions of the explanatory variables. Appendix 3 shows the descriptive statistics of all variables.

Compared to the WDI data, more data are available from the Penn World tables, but they run up only to 2004 compared to 2005 for WDI. The disadvantage of the Penn data is that they exclude public investments, which of course are of interest in this study. Their main advantage is that they are free of biases arising from cross-country differences in the relative price of government purchases (Rodrik, 1998). The fact that this is not the case for the WDI data is the main criticism on these data, and it is likely that this is one of the most important reasons for the fact that the correlation between the two types of measurements is not always as high as one would expect. Appendix 8 gives an overview of this correlation for each country separately. The following graphs stipulate the differences between countries in correlation of the two variables clearly.

GRAPH 2: CORRELATION GOVERNMENT EXPENDITURE

As 2 4 6 8 10 12 Pe n n

Correlation Government Expenditure Nigeria

1 2 3 4 5 6 7 Pe n n

(28)

The graphs on the previous page show the correlation of government expenditure in Oman and Nigeria. The Penn and WDI data of government expenditure of Oman have a correlation of 0,97, whereas the correlation the two measurements of government expenditure in Nigeria is only 0,41. High correlation in the two measurements of government expenditure increases the likelihood that the measurements are robust. This high correlation does not mean, however, that correlation between the first differences of these variables, which is the measurement used in this study, is also high. This shows in the following graphs:

GRAPH 3: CORRELATION d(GOVERNMENT EXPENDITURE)

The correlation of the difference of the logs of government expenditure in Oman is 0.33, whereas for Nigeria it is 0.09.

4.2: Explanatory variables

(29)

in a country by the average Brent oil prices during that year. The Brent price is the standard price for oil that is used when discussing the price of oil (Brooke et al., 2004). This is oil coming from the Brent oilfield in the North Sea, which came into use in 1983. Before 1983, the Saudi Arabian ‘Ras Tanura’ oil price, which has the same characteristics and quality as the Brent oil, is used.

The second proxy to measure government oil revenues is by using the variable fuel exports, available from the World Development Indicators. A disadvantage here is that the availability of data is very limited. An additional limitation is that the oil-production used for the domestic economy is not taken into account.

The data used will not exactly represent that a government receives from the oil revenues. One reason is that the quality and thus the price of oil differs from country to country, and from oilfield to oilfield4. Another reason, already mentioned in the introduction, is that is that the royalty rates and the petroleum revenue taxes differ between countries. (Yergin, 1991; Mommer, 2002). However, because the interesting part for our research is the extent to which government expenditure responds to changes in oil prices, the data are perfectly usable when assuming that the quality of the oil within a country does not change and the royalties as a percentage of total oil revenues do not change over time either. These are reasonable assumptions to make, and also made by Davis et al. (2001), for example.

For the copper-producing countries Chile and Peru, there are sufficient data on the exports of their copper available from the World Development Indicators. The data from the WDI do not only include copper, however, but also other ores and metals. These are nevertheless the most specific data available for copper revenues. For Chile, copper is by far the largest ore or metal that is being produced, for Peru also gold and zinc are to some extent included in the variable.

4

(30)

GRAPH 4: COPPER PRICE

Copper price (constant 1998 US$) 1970 - 2005

0 1.000 2.000 3.000 4.000 5.000 6.000 1970 1974 1978 1982 1986 1990 1994 1998 2002 $ copper price

Just like with the two measurements of government expenditure, a high correlation between the two types of oil revenue proxies serves as a check for robustness. Again, the correlation varies between different countries. Correlation of oil revenues estimations by country is included in appendix 8.

GRAPH 5: CORRELATION OIL REVENUES

0 2,000 4,000 6,000 8,000 10,000 12,000 0 1,000 2,000 3,000 4,000 5,000 6,000 WDI BP

Correlation oil revenues Ecuador

The graphs above make clear that correlation of the two types of oil revenues can be very different between countries. Correlation for Ecuador is 0.85, whereas for Algeria it is

0 4,000 8,000 12,000 16,000 20,000 24,000 28,000 32,000 0 10,000 20,000 30,000 40,000 50,000 BP

Correlation oil revenues Algeria

(31)

0.37. Correlation of the differences of the logs of the BP measurement and the WDI measurement is in general quite high, as the following graphs illustrate:

GRAPH 6: CORRELATION d(OIL REVENUES)

Correlation for Ecuador is 0.86; for Algeria it is 0.63.

4.3: The sample

I will focus on oil-producing countries where oil revenues accounted for at least 20% of total fiscal revenues in 2004, and for which sufficient information is available. The countries for which oil revenues account for 20% or more of fiscal revenues are derived from Ter-Minassian (2007), where they are defined as the thirty oil-producing countries of the world. The thirty initially chosen countries are (in alphabetical order) Algeria, Angola, Azerbaijan, Bahrain, Brunei, Cameroon, Chad, Republic of Congo, Ecuador, Equatorial Guinea, Gabon, Indonesia, Iran, Kazakhstan, Kuwait, Libya, Mexico, Nigeria, Norway, Oman, Qatar, Russia, Saudi Arabia, Sudan, Syria, Trinidad and Tobago, United Arab Emirates, Venezuela, Vietnam, and Yemen5.

5

(32)

After a study on the data availability of these thirty countries, it was decided to drop the countries for which not enough information was available. These countries are Angola, Chad, Equatorial Guinea, and Sudan. This is very unfortunate, because the four countries have similar characteristics in that they are all African countries with a very low GDP per capita compared to the other countries.

In addition to these 26 countries, Chile, a country not dependent on oil but on copper which has set up an NRF to stabilize its copper revenues, will be included. This is in line with Davis et al. (2001). To compare the fiscal performance of Chile with other copper exporting countries, Peru is also included in the country sample.6 7 The regressions will also be run without these two countries, to make sure they do not bias the outcomes.

The final sample of countries contains a lot of variety in characteristics. All are oil (or copper) producing countries, but the size of their oil-production varies tremendously. Furthermore, some countries are much richer than others, and it is likely that the way in which their government expenditures are spend will vary greatly as well. The following graph shows a simple comparison between Congo Brazzaville, Saudi Arabia and Norway regarding their size of oil production and GDP per capita.

GRAPH 7: SELECTED COUNTRIES: OIL PRODUCTION

Annual Oil Production (m illions of barrels) 1999-2004

0 2.000 4.000 6.000 8.000 10.000 1999 2000 2001 2002 2003 2004 Norw ay Saudi Arabia Congo (Brazzaville) 6

With a production of 5.7 million tons of copper in 2007, Chile is the largest copper producer worldwide. Peru is the second producer, with 1.2 million tons in 2007. Source: US Geological survey, 2008.

7

(33)

GRAPH 8: SELECTED COUNTRIES: GDP

GDP per capita (current US$) 1999 - 2004

0 5.000 10.000 15.000 20.000 25.000 30.000 35.000 40.000 1999 2000 2001 2002 2003 2004 Norw ay Saudi Arabia Congo

Appendix 1 gives an overview of the sample of countries, the degree to which they are dependent on oil, measured by the fuel exports as a percentage of total merchandise exports in 2005, and for the countries that have set up an NRF (14 in total) the year that the fund was established. Not all the data are available for all countries of my sample for the whole period 1970-2005, making it an unbalanced panel; an overview of the years that will be used for each country of the sample is given in appendix 9.

(34)

5: Empirical results

5.1: procyclicality of oil-producing states

Table 2 shows the correlation between government expenditure and oil revenues for each country. As two sources for government expenditure and two sources for oil revenues are used, four different datasets are available for most countries. In this table, as well as in the rest of this chapter, gp stands for government expenditure obtained from the Penn World Tables; gw for government expenditure obtained from the WDI; BP for oil revenues from BP and the EIA; and WDI for oil revenues from the WDI.

TABLE 2: PROCYCLICALITY

Country d(gp)d(BP) D(gp)d(WDI) d(gw)d(BP) d(gw)d(WDI)

Algeria 0,31 0,23 -0,02 -0,06 Azerbaijan 0,19 -0,10 0,47 0,77 Bahrain 0,00 0,66 -0,01 0,54 Brunei 0,65 0,39 Cameroon 0,14 -0,18 -0,13 0,41 Chile -0,25 0,74 Republic of Congo -0,31 0,78 -0,12 0,02 Ecuador 0,07 -0,05 0,07 0,02 Gabon -0,20 -0,04 -0,06 0,20 Indonesia 0,01 0,41 0,13 0,44

Iran, Islamic Rep. 0,16 0,14 0,06 0,40

Kazakhstan 0,46 0,45 0,73 0,80 Kuwait 0,01 0,28 0,01 0,02 Libya 0,09 0,36 Mexico 0,14 0,17 0,26 0,19 Nigeria -0,27 -0,41 0,29 0,42 Norway -0,10 -0,14 0,09 0,25 Oman 0,05 0,21 0,40 0,17 Peru 0,19 0,41 Qatar 0,24 0,65 0,63 0,52 Russia -0,32 0,61 0,47 0,65 Saudi Arabia -0,26 -0,14 0,24 0,37 Syria 0,34 0,24 0,28 0,27

Trinidad and Tobago 0,25 0,26 -0,14 0,09

United Arab Emirates -0,09 0,16 0,04 0,00

Venezuela, RB 0,13 0,17 0,10 0,14

Vietnam -0,18 -0,21 -0,09 0,06

Yemen -0,51 -0,78 0,35 0,74

Average 0,04 0,14 0,17 0,33

(35)

The average of all four datasets is larger than zero. Simple hypothesis testing with a null hypothesis of a zero mean confirms that these averages are significantly larger than zero in all but one of the cases. Overall, the data confirm the hypothesis that oil-producing countries run procyclical policies with respect to their oil revenues. As has been described in the literature review, oil revenues are in general very volatile. Procyclical policies can therefore lead to volatility in government expenditure, which can be harmful for long-term economic growth.

5.2: The impact of NRFs: the with-without approach

To estimate the extent to which countries with an NRF have less procyclical policies than countries without a fund, two variations of the with-without approach are tested. The first variation of the with-without approach is similar to the model used in paragraph 5.1. The correlation of government expenditure and oil revenues for countries with an NRF (in the years in which this NRF had been established) and without an NRF are specified in table 3. In this table, the countries that have an NRF established are colored in red.

(36)

TABLE 3: WITH-WITHOUT APPROACH (CORRELATIONS)

Country d(gp)d(BP) d(gp)d(WDI) d(gw)d(BP) d(gw)d(WDI)

Algeria -0,13 -0,18 0,02 -0,15 Azerbaijan 0,81 0,74 0,96 0,93 Bahrain 0,00 0,66 -0,01 0,54 Brunei 0,24 0,23 Cameroon 0,14 -0,18 -0,13 0,41 Chile -0,51 0,83 Republic of Congo -0,31 0,78 -0,12 0,02 Ecuador 0,07 -0,05 0,07 0,02 Gabon -0,20 -0,04 -0,06 0,20 Indonesia 0,01 0,41 0,13 0,44

Iran, Islamic Rep. 0,22 0,21 0,47 0,48

Kazakhstan -0,13 -1,00 0,67 0,72 Kuwait 0,01 0,28 0,01 0,02 Libya 0,13 0,34 Mexico 0,14 0,17 0,26 0,19 Nigeria -0,27 -0,41 0,29 0,42 Norway -0,22 -0,24 0,08 0,17 Oman 0,23 0,21 0,36 0,17 Peru 0,19 0,41 Qatar 0,24 0,65 0,63 0,52 Russia -0,32 0,61 0,47 0,65 Saudi Arabia -0,26 -0,14 0,24 0,37 Syria 0,34 0,24 0,28 0,27

Trinidad and Tobago 0,32 0,25 -0,14 -0,10

United Arab Emirates -0,08 0,16 0,05 0,00

Venezuela, RB 0,42 0,32 0,40 0,33

Vietnam -0,18 -0,21 -0,09 0,06

Yemen -0,51 -0,78 0,35 0,74

Average 0,02 0,09 0,21 0,33

Average NRF countries 0,17 0,02 0,3 0,34 Average non-NRF countries -0,07 0,14 0,15 0,32

The second variation of the with-without approach, derived from Lane (2003), is a cross-country regression. To determine βGi, the index of cyclicality, equation (4) is calculated

for each country. The results of this regression can be found in appendix 10.

(37)

TABLE 4: CROSS-COUNTRY VARIATION

Dependent

variable: beta gpBP beta gwBP beta gpWDI beta gwWDI

C -0,002557 0,078867 -5,37E-03 4,06E-01

[0,023205] [0,120765] [0,024769] [0,182196]**

GDP -3,70E-07 -2,65E-06 -5,13E-07 -5,46E-06

[6,94E-07] [0,00000361] [7,81E-07] [0,00000565]

OC 3,43E-03 -0,039899 0,005701 -0,379366

[0,024662] [0,137129] [0,031889] [0,270229]

VOL -8,55E-03 -5,56E-02 0,002591 0,263965

[0,108738] [0,635156] [0,119266] [1,011899]

NRF 0,017812 8,68E-02 0,018849 -0,05969

[0,014549] [0,078651] [0,016120] [0,118091]

N 24 23 23 22

R-squared 0,076148 0,092714 0,080234 0,15497 * 10% level; **5% level; *** 1% level of significance. Standard errors are given in parentheses

The most important variable for this research is the NRF dummy. The hypothesis is that this variable has a negative sign, as one of the objectives of an NRF is to reduce procyclicality in government expenditure. Again, however, the data give evidence of the contrary; in all but one of the datasets the NRF dummy has a positive sign and for the dataset where the dummy does have a negative sign, it is not significant at the 10% level.

Other interesting facts come to light when examining table 4 in more detail. GDP growth per capita, for example, is negative in all cases. This is in line with the finding of Talvi and Végh (2000), who claim that poor countries are most procyclical because they have less other options to obtain revenues other than from their primary export products.

(38)

The fact that in all but one of the regressions countries with an NRF seem to have more procyclical fiscal policies than countries without an NRF is puzzling. One likely explanation is that countries that suffer from a government unable to be strict in fiscal matters set up an NRF to deal with their problems. But as Bacon and Tordo (2006) found in their research, an NRF is only able to perform when other fiscal policies are in place. When policies are insufficient to deal with volatile revenues, a fund will not make matters any better. Therefore, a fund is not a panacea to deal with less than prudent fiscal policies for oil-producing countries, but can only work properly when integrated in a well functioning fiscal structure.

Another possible explanation for the lack of performance of NRF’s in the regression above, is that explanatory variables are missing. Therefore, two other variables are being included. The first is oil dependency (OILDEP), which shows the degree to which a country is depending on oil revenues to obtain foreign currency. It is expected to be positive, because if government has little other export earnings, government spending will more readily respond on oil price changes than when there are many other taxable sectors available. The second additional control variable is the quality of government (GOVQ), measured as a composite index of the six government quality measurements of Kaufmann, Kraay and Mastruzzi (2008). I expect that when the quality of government is high, fiscal policies will be more prudent and therefore less procyclical. The definitions of these variables can be found in appendix 2.

(39)

TABLE 5: CROSS-COUNTRY VARIATION (2)

Dependent variable: beta gpBP Beta gwBP Beta gpWDI beta gwWDI

C -0.009553 0.051801 -0.005516 0.486673

[0.029590] [0.160011] [0.029926] [0.235540]*

GDP 2.74E-07 -1.81E-07 4.17E-07 -9.12E-06

[9.11E-07] [4.87E-06] [1.01E-06] [7.60E-06]

OC 0.005607 -0.037536 0.003645 -0.338462 [0.025375] [0.142351] [0.032854] [0.281388] VOL 0.007759 0.042869 0.033407 0.286670 [0.114919] [0.722693] [0.124124] [1,145281] NRF 0.024351 0.109363 0.033228 -0.095360 [0.016738] [0.089288] [0.019635] [0.146569] OILDEP -0.008108 -0.030707 -0.021818 -0.106493 [0.039305] [0.223647] [0.044711] [0.353465] GOVQ -0.016467 -0.063681 -0.021621 0.082326 [0.014392] [0.078280] [0.015085] [0.115234] N 24 23 23 22 R-squared 0,144798 0,131311 0,185762 0,290948 * 10% level; **5% level; *** 1% level of significance. Standard errors are given in parentheses

Using the two extra control variables, the NRF-dummy is still positive in three of the four cases. As expected, the quality of government has a negative sign in all but one of the cases, but fails to be significant as well. Surprisingly, oil dependency has a negative sign, but no firm conclusion can be drawn as this variable is not significant either. Similarly to the other with-without test, this test cannot confirm the hypothesis that countries with an NRF perform better than countries without a fund.

5.3: The before-after approach

(40)

5.3.1: Results for individual countries

Algeria

For Algeria, only contemporaneous oil revenues are significant, suggesting that oil revenues are spend only after one year. The Chow test shows a structural break in two of the cases for the year 2000, when the fund got introduced. When looking at the signs of the NRF-dummy and the interaction variable, results are very interesting. In the two cases where a structural break is evident, the interaction variable has the expected negative sign, but the NRF-dummy is positive and significant. This means that government expenditure rose after the establishment of the fund, but procyclicality decreased.

Azerbaijan

A problem with Azerbaijan is that only a limited number of observations is available, especially after the establishment of an NRF. Still, a structural break is evident in two of the datasets. In these datasets the NRF-dummy is negative and significant, meaning that government expenditure fell after the introduction of the fund. The procyclicality, on the other hand, seems to have increased somewhat after 1999.

Brunei

In the case of Brunei, the structural break test does not provide any interesting results. The interaction variable has the expected negative sign, but is not significant. The NRF-dummy is significant, and suggests that government expenditure increased after the establishment of the NRF.

Chile

(41)

Iran

The case of Iran is problematic, as none of the datasets provides any significant results. R-squared is very low in all of the cases, suggesting that other factors are more important in predicting the government expenditure than the oil revenues obtained in the previous year.

Kazakhstan

Just like with Azerbaijan, a problem here is the limited availability of data. In Kazakhstan a structural break is evident after the introduction of the NRF. However, the evidence suggests that both government spending and procyclicality actually increased after the introduction of the NRF.

Libya

In the case of Libya, there are no significant variables. The NRF-dummy and the interaction variable actually have a positive sign, in contradiction to the hypothesis, but they are not significant.

Norway

Similar to Davis et al, the Chow test does not find evidence for a structural break after the establishment of the fund took place. In two of the cases, the NRF-dummy is significant, but not with the expected sign. The interaction variable does have the expected sign in these cases, but is not significant. This is also in line with what Davis et al. have found.

Oman

(42)

Trinidad and Tobago

For Trinidad and Tobago only contemporaneous oil revenues are used, suggesting that oil revenues are immediately spend. The structural break test is significant in two cases, but the NRF-dummy and the interaction variable do not have the expected sign.

UAE

For the United Arab Emirates, no conclusions can be derived. R-square is very low, suggesting that government expenditures can not be explained very well based on the changes in oil revenues.

Venezuela

For Venezuela, it becomes clear that oil revenues are an important variable explaining government expenditure, even though only the lagged oil revenues are significant. Oil revenues have only become more important after the introduction of the NRF. Government expenditure increased significantly in two of the datasets, and the evidence of one of the datasets shows that procyclicality also significantly increased.

5.3.2: Overall conclusions from the before-after approach

(43)

(44)

6: Conclusions

In October 2008, delegates from the Organization of Petroleum Exporting Countries (OPEC) met in Vienna. Their concern was the recent dramatic fall of oil prices; 60% from its highest price in July. This fall meant an abrupt end to the rocketing oil prices ever since 2001, and was hardly foreseen by any analyst in their midst. Talks were about the mutual cutting of production, to avoid prices falling even more. For many of these countries, falling oil prices meant that they would have to cut in their budgets drastically. This in turn could lead to political unrest internally, investments that were agreed upon which would have to be cancelled again, and the inability to support other states in their region.

This example clearly illustrates the problems that have been discussed in this paper. Oil prices fluctuate strongly, which allows governments to lift expenditure in times of rising prices, but also forces them to cut back when prices are falling. This leads to large uncertainty, and research suggests that this fluctuation and uncertainty leads to a decline in economic growth.

The research done in this paper confirms the hypothesis that government expenditure is positively correlated with the revenues governments receive from oil; in this respect, it acts in a procyclical manner.

(45)

negative using the with-without approach. Secondly, there is no evidence found that countries become less procyclical after the establishment of an NRF.

There are two possible reasons why the research done does not find any evidence for the good performance in reducing procyclicality of countries that have established NRFs. The first, obvious, reason is that countries with NRFs do not perform any better than countries without NRFs. The funds are neither necessary nor sufficient tools to improve fiscal policy in oil-producing countries, and even though the establishment of an NRF could be a sign of government that it wants to pursue less procyclical fiscal policies, in practice this does not happen.

The lack of evidence could also be explained by some of the limitations of the research in this paper. The main limitation in this respect is the poor availability of high quality data, mainly for African countries. Some funds have been set up over the last ten years, making the availability of high quality data for these countries even more difficult to obtain. A quantitative analysis does not take into account the large differences in the types of funds that have been established throughout the world. Funds differ greatly in their governance structure, the percentage of oil revenues that they put into the fund, the rules that allow government to withdraw money from the fund, in how the money is invested, in how secretive they are; even in the degree to which avoiding procyclicality is their main objective. Even if all the data was of high quality, there are only 32 countries in the world that rely on oil revenues for at least 20% of fiscal revenues, making the sample, and therefore the chance to find significant results, rather small. The fact that proxies had to be used to estimate government oil revenues does not help to improve the quality of data either. On the other hand, in this study the impact of NRFs of many more countries compared to what has been done in previous research was examined, thereby increasing the likelihood that the results presented are just.

(46)

that government has been receiving for the exploitation of oilfields is better known for certain countries. Furthermore, single-country analysis can take the unique characteristics of each NRF into account.

For oil-exporting countries themselves, the results in this research can help them deal with the question of why exactly their funds did not help them pursue more prudent and less procyclical fiscal policies. Fiscal policies that are now being executed are less than optimal, and it will be of importance to these countries to develop long-term strategies how to develop their fiscal systems so that long term growth and investments can be guaranteed, and change their NRFs accordingly.

(47)

References

Agenor, P., J. McDermott and E. Prasad, 1999, “Macroeconomic fluctuations in developing countries: some stylized facts,” Working Paper 99/35 (Washington: International Monetary Fund).

Bacon, Robert and Silvana Tordo, 2006, “Experience with oil funds: Institutional and financial aspects,” Energy Sector Management Assistance Program (ESMAP) report 321/06 (Washington: World Bank).

Barro, Robert J, 1979, “On the determination of the public debt,” The Journal of Political

Economy, 87 (5) 940-971.

Bird, Graham, 2007, “The IMF: A Bird’s eye view of its role and operations,” The

Journal of Economic Surveys, 21 (4) 683-745.

Bleaney, Micheal and David Greenaway, 2001, “The impact of terms of trade and real exchange rate volatility on investment and growth in sub-Saharan Africa,” The Journal of

Development Economics, 65 (2) 491-500.

Brooke, Anne-Marie, Robert Price, Douglas Sutherland, Niels Westerlund and Christophe Andre, 2004, "Oil Price Developments: Drivers, Economic Consequences and Policy Responses," OECD Economics Department Working Paper No. 412 (Paris: OECD).

Brüderl, Josef, 2005, “Panel data analysis,” University of Mannheim, available at http://www.sowi.uni-mannheim.de/lehrstuehle/lessm/veranst/Panelanalyse.pdf

(48)

Davis, Jeffrey, Rolando Ossowski, James Daniel, and Steven Barnett, 2001, “Stabilization and savings funds for nonrenewable resources. Experience and fiscal policy implications,” Occasional Paper 205 (Washington: International Monetary Fund).

De Nederlandse Bank, 2008, De Nederlandse gasbaten en het begrotingsbeleid: Theorie

versus praktijk, Occasional studies 6 (5).

Devlin, Julia and Michael Lewin, 2004, “Managing oil booms and busts in developing countries,” in Managing economic volatility and crises, eds. Aizenman, Joshua and Brian Pinto, (Cambridge, Cambridge University Press).

Engel, Eduardo, and Rodrigo Valdés, 2000, “Optimal fiscal strategy for oil exporting countries,” Working Paper 00/118 (Washington: International Monetary Fund).

Fasano, Ugo, 2000, “Review of the experience with oil stabilization and savings funds in selected countries,” Working Paper 00/112 (Washington: International Monetary Fund).

Feldstein, Martin, 1999, “A self-help guide for emerging markets,” Foreign Affairs, vol. 78 (2) 93-100.

Fiorito, Ricardo, 1997, Stylized facts of government finance in the G-7, Working Paper 97/42 (Washington: International Monetary Fund).

Fiorito, Ricardo and Kollintzas, Tryphon, 1994, “Stylized facts of business cycles in the G7 from a real business cycles perspective,” European Economic Review, 38 235-269.

(49)

French, Mark W., 2001, Estimation changes in trend growth of total factor productivity:

Kalman and H-P filters versus a Markov-switching framework, United States Federal

Reserve, available at www.federalreserve.gov/pubs.

Furceri, Davide, 2007, “Is government expenditure volatility harmful for economic growth? A cross-country analysis,” Fiscal Studies, vol 28 (1) 103-120.

Gary, Ian and Nikki Reisch, 2005, Chad’s oil, miracle or mirage? Following the money

in Africa’s newest petro-state, Catholic Relief Services and the Bank Information Centre,

available at www.bicusa.org.

Gavin, Michael, Ricardo Hausmann, Roberto Perotti, and Ernesto Talvi, 1996, Managing

fiscal policy in Latin America and the Caribbean: Volatility, procycliclity and limited creditworthiness, OCE Working Paper 326, (Washington: Inter-American Development

Bank).

Gavin, Michael and Roberto Perotti, 1996, “Fiscal policy in Latin America,” NBER

Macroeconomics Annual, (Chicago, The University of Chicago Press).

Gelb, Alan, 1989, Oil windfalls: blessing or curse?, (Washington, The World Bank).

Heller, Peter S. and Jack Diamond, 1990, “International comparisons of government expenditure revisited: the developing countries, 1975-86,” Occasional Paper 69 (Washington: International Monetary Fund).

Heston, Alan, Robert Summers and Betina Aten, 2006, Penn World Tables Version 6.2, Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania.

Hill, R. Carter, William E. Griffiths and George G. Judge, 2001, Undergraduate

THE PERFORMANCE OF NONRENEWABLE RESOURCE FUNDS