OIL PRICES AND ECONOMIC GROWTH IN OIL-EXPORTING COUNTRIES Amany A. El Anshasy Collage of Business and Economics, United Arab Emirates University P.O. Box 17555 Phone: +971 05 6603536, E-mail: [email protected]

Abstract This paper investigates the effects of the highly volatile oil prices and the resulting fluctuations in government revenues on economic growth in oil-exporting countries. The evidence suggests that, oil price volatility does not seem to retard long run growth. In addition, after controlling for fiscal policy, higher oil prices have a fairly small long run positive effect on growth. In that sense, the resource rent is not a “curse” in itself. We find that fiscal policy is the main propagation mechanism that transmits the oil price shocks to the economy, and that seems to explain the differences in growth performance across oil exporters. JEL classification: 040; Q4; E62; C33. Keywords: Economic Growth; oil prices; Fiscal Policy; Models with Panel data.

1. Introduction Since the 1970‟s, oil-exporting countries have been facing large fluctuations in the prices of their oil, and hence in terms of trade. Such fluctuations often lead to higher output volatility which may lower long-run growth rates.1 In addition, oil price shocks are mainly a fiscal phenomenon because of their direct impact on government finances, resulting in a highly volatile revenue stream and an, often, pro-cyclical government expenditures, as well as pro-cyclical foreign and domestic finance. This suggests that oil shocks are transmitted to the economy mainly through fiscal policy.2 Observing the rates of growth of oil exporting countries during the period 1960 to 2000 reveals two stylized facts. First, oil-exporters experienced lower average growth rates than the group of developing countries and the group of non-oil exporting countries. This phenomenon has been referred to in the literature as “the resource curse” or the “paradox of plenty”. Second, these countries had relatively high average growth in per capita income when oil prices were rising in the seventies, after a period of relatively stable prices. On the contrary, the two decades of adverse oil price shocks, in the 1980s and 1990s, witnessed a huge implosion in oil-exporters‟ average growth rates compared to non-oil exporters and other developing countries.3 However, the growth performance varies widely across individual oil-exporters. Among this group, some do better than others. The resource curse literature has recently shifted attention 1

See for example: Easterly et al (1993), Sachs and Warner (1995), Deaton and Miller (1995), Deaton (1999), and Rodriguez and Sachs (1999). 2 See: El-Anshasy, Bradley, and Joutz (2005) and Husain et al (2008). 3 See for example: Hausmann and Rigobon (2002). 1

to explaining these variations. This paper is an attempt to contribute to that strand of research. It addresses the question of why some oil-exporters exhibit poorer growth performance than others at times of negative oil price shocks, and why some do better at times of positive oil shocks. It offers an answer: it is the fiscal policy channel. There can be several propagation mechanisms through which fiscal policy can interact with oil shocks in affecting long-run growth: (i) the composition of public expenditures; (ii) the method of financing these expenditures; and (iii) the response of fiscal policy to oil price shocks. Many studies have tried to explain the disappointing growth performance of resourceabundant economies in general, and to develop some theories of why they tend to grow slower than non-resource economies of similar levels of development. Oil- exporting economies are a striking example of such a “resource curse”.4 Two main lines of thought can be identified in the literature. Studies such as Basu and McLeod (1992), Sachs and Warner (1995), Mendoza (1997), Rodriguez and Sachs (1999), and Hausmann and Rigobon (2002), among others, emphasize a host of economic factors. The second group emphasizes political and institutional factors which come into play mainly through „policy choices‟. Examples of such studies can be found in Mahon (1992), Tornell and Velasco (1992), Auty (1994), Lane and Tornell (1998), Rodrik (1998), Sala-iMartin and Subramanian (2003), Isham et al (2004), and Mehlum et al (2006). This paper extends the existing literature in three ways. First, it investigates the direct growth effects of both oil price shocks and oil price volatility. Second, it controls for the fiscal channel which can play an important role in transmitting oil price shocks to the economy. In doing so, it investigates the growth implications of different methods of government finance. Third, using annual observations of a panel of 15 oil-exporting countries for the period 19702004, the study utilizes dynamic panel estimation techniques. In particular, it applies the Generalized-Method-of-Moment (GMM) system estimator. This estimator has the potential advantages of minimizing the bias resulting from estimating dynamic panel models, exploiting the dynamic and time series properties of the data, controlling for the unobserved country-specific effects, and correcting for the bias resulting from the possible endogeneity of the explanatory variables. In addition, employing annual observations allows exploiting the dynamic nature of the data and exploring the short- run as well as the long-run effects. The main findings can be summarized as follows: (i) The unanticipated increases in oil prices trigger higher growth rates but that effect is fairly small; (ii) Oil price volatility does not seem to trigger growth effects; (iii) Fiscal policy does play an important role in transmitting oil shocks to the economy. Oil revenue booms retard long-run growth; (iv) Countries that have a higher initial share of public investment to GDP can better cope with adverse oil shocks. However, they tend to benefit less from oil booms because the surge in capital spending tends to have very low return; (v) Adjusting to a negative shock by cutting capital or social expenditures can deepen the negative effect of the shock and hurt growth; (vi) Channeling some of the booming revenues to increase spending on infrastructure and public services tends to stimulate growth while government consumption in general tends to be less productive. These findings prompt some important policy implications: (i) diversification policies and expanding the non-oil tax base are growth-improving; (ii) At times of scarcity, governments should avoid cutting capital expenditure and should increase social spending to help stimulate

4

See: Gelb et. al (1988), and Sachs and Warner (1995). 2

growth; (iii) At times of plenty, the government should apply more prudent investment policies and stricter public projects criteria in order to enhance the productivity of the (larger) expenditures especially on infrastructure and on improving public services. (iv) Our findings also lend support to the importance of transferring the surplus to an autonomous sovereign wealth fund and use the receipts of that fund to smooth consumption. However, weighing the options of eliminating the fiscal surplus through a wealth fund versus reducing public debt will always remain a crucial policy choice. In addition to maximizing the social net gains of a revenue boom, eliminating the surplus would alleviate competitive rent-seeking pressures. The remainder of this paper is organized as follows. Section II is a review of the specific nature of fiscal policy in oil-exporting countries and the role it plays in growth. In section III the empirical model is discussed followed by the estimation methodology in section IV. The results and empirical findings are presented in section V. Section VI concludes.

2. Fiscal policy and economic growth in oil-exporting countries It is important to begin with laying out some stylized facts regarding the place of fiscal policy in the growth debate in general. Then, we turn our attention to discussing some specific features of fiscal policy in oil exporting countries, highlighting some major fiscal policy responses to the oil booms and busts of the 1970s and 1980s in our sample of 15 oil exporters.5

2.1. Fiscal Policy in growth models: In a standard neoclassical growth model, fiscal policy differences among countries may only explain the observed differences in income levels but not in long-run growth. On the contrary, the endogenous growth models predict that public policies could have permanent growth effects, and hence differences in fiscal policies may explain the observed differences in growth rates across countries. Barro (1990), King and Rebelo (1990), Rebelo (1991), Barro and Sala-i-Martin (1992), Jones et al. (1993), Mendoza et al (1995), and Greiner and Semmler (2000), are early examples of this class of models. On the empirical level, the growth effects of fiscal policy are far from being conclusive.6 However, one can draw a number of lessons from earlier literature. First, emphasizing one side of the budget (revenues or expenditures) leads to serious biases in the estimated fiscal coefficients since it fails to account for the fact that every increase in spending has to be financed through collecting more revenues or accumulating more debt. Conversely, changes in revenues have ramifications on expenditures or the budget balance. Thus, the growth effects of government spending and that of the government‟s financing methods are interdependent. That is: the growth effects of the different types of expenditures may strongly depend on the way they are being financed. Second, using aggregate measures of government spending may not capture many 5

The sample includes Bahrain, Cameroon, Colombia, Egypt, Indonesia, Iran, Kuwait, Malaysia, Mexico, Nigeria, Norway, Oman, Syria, UAE, and Venezuela. Although all sample countries are net exporters of oil, some are moderately oil-exporters and others are heavily dependent on oil exports. Also, for each country, oil exports constitute %15 of GDP; and (or) oil exports exceed 35% of total commodity exports; and (or) government oil revenues constitute at least 30% of government‟s total revenues. 6 See for example: Easterly and Rebelo (1993), Garrison and Lee (1995) and Mendoza et al (1995) Slemrod, J. (1995), Devarajan et al (1996), Kocherlakota and Yi (1997). Miller and Russek (1997) and Kneller et al. (1999 and 2001). 3

potentially important growth effects of fiscal policy, particularly those related to how government resources are being allocated. This can be of a greater concern in the context of developing countries, since the allocation of public funds among different uses and the way they are financed may matter more for growth than just the size of the government. Third, the composition of fiscal adjustments may have important implications for long-run growth.7

2.2. Fiscal policy in oil exporting countries: In oil exporting countries government finance is heavily dependent on the oil sector. Hence government revenues tend to be highly volatile, and will eventually dwindle and dry up in the future. In addition, oil price shocks tend to be persistent and the oil price cycles are highly unpredictable.8 These characteristics make fiscal management more challenging in such countries and have very important implications for their growth performance. We highlight some of these implications as follows. (i) The oil price volatility can be transmitted to the economy through the large fluctuations in government revenues. The uncertainty about future oil revenues and the variability of such revenues would result in changes in spending as the government reassesses its expected revenue stream, generating significant adjustment costs (Hausmann et al. 1993). Therefore, the resulting pro-cyclicality of government spending can ultimately lower growth rates. Carefully looking into some of the potential expenditure mechanisms, one can identify the following:  A positive revenue shock that is perceived as permanent typically leads to higher government spending, especially on non-tradables, creating incentives to shifting resources away from the (non-oil) tradable sector to the non-tradable sector.9 Such resource movements would lead to higher unemployment, output losses, and ultimately the de-industrialization of the economy; a phenomenon known as the “Dutch disease”. To the extent that the manufacturing sector provides positive spillovers to other sectors, the resource (government revenue) windfall would have a negative depressive effect on long-run growth (Sachs and Warner, 1995).  If a positive shock is perceived as temporary, accumulating the budgetary surpluses in developing economies is politically unpopular and the government will be subject to pressures to increase spending, especially on public projects. For example, during the period 1974-1978, 85%, 50%, and 46% of the windfall gains accrued to the governments of Nigeria, Indonesia, and Venezuela, respectively, were channeled to increasing public investment (Gelb et. al, 1988). Many studies found that most of the large surges in public capital spending during boom times are non-productive and typically have a very low return (See: Talvi and Vegh, 2000).  A negative shock, on the other hand, typically induces downward adjustments in government expenditures. This adjustment could be very costly. 10 On the one hand, cutting current

7

See for example: Alesina and Perotti (1996) and Ardagna (2001). See Cashin, Liang and McDermott (1999), Cashin, McDermott and scott (1999), Hausmann et al (1993), Engle and Valdes (2000). 9 It is difficult to distinguish, ex ante, between a temporary and permanent oil price shocks. So, if shocks are perceived as permanent then the best prediction for future period revenues is current period revenues. 10 For example, Hausmann et al (1993) estimated the (secondary) adjustment costs for Venezuela during the 1980s to be %10 of non-oil GDP; almost equal to the (primary) losses in absorption due to the initial external shocks. 8

4

expenditures is usually unpopular because of its negative social consequences. On the other hand, cutting capital expenditures would disrupt public projects, reducing the productivity of the initial investment and causing high social costs. (ii) In a downturn, it is not quite unusual that some governments delay a needed adjustment to avoid immediate spending cuts. If the shock turns out to be permanent, the persistent budget deficit and the growing public debt would put into question fiscal policy and current account sustainability, as well as government solvency. Ultimately, a larger adjustment at a higher cost would be inevitable at some point in the future. For example, in 1986, Venezuela did not allow for spending adjustment in response to the negative large oil shock. In 1989, the looming balance of payments crisis led to substantial costly adjustments (Hausmann et al., 1993). (iii) A fiscal consolidation in response to a permanent negative oil shock that aims to put fiscal policy on a sustainable path would adversely affect growth, leading to a more unsustainable path. A given level of primary deficit that may seem sustainable given a certain growth rate could be unsustainable at a lower rate of growth. This endogeneity of fiscal policy appears to be crucial in designing fiscal adjustments in shock-prone economies. (iv) Oil exporting countries tend to have higher borrowing capacity during boom times. Therefore, an oil boom could induce an expansion of easy borrowing, especially with the large growth in domestic absorption. That lately resulted in the phenomenon of highly-indebted oil-rich economies. The accumulation of debt during times of plenty makes the adjustment more costly and more difficult at times of scarcity because it implies larger adjustments. Therefore, at times of oil price downturns some oil economies may face foreign borrowing constraints, which would adversely hinder their development programs. In addition, this leaves the fiscal authorities with fewer options to finance the deficit. Sharp expenditure cuts may become inevitable, potentially harming long-run growth.

Table: 1. Changes in Government Expenditure (1970-2001) Variable

Total Government Expenditure Capital Expenditure Current Expenditure Wages Purchases of Goods and Services Current Transfers Government Consumption Social Expenditures Infrastructural Spending Public Services Total Government Revenues Tax Revenues Non Tax Revenues

1970-2000 % of GDP

Positive Oil Shocks (1970-1981) % of GDP

29.0 6.7 20.8 6.9 6.3 6.3 15.9 8.2 5.9 3.1 26.7 13.9 12.8

31.4 9.2 20.1 6.7 7.6 6.0 15.7 6.5 7.9 3.4 30.1 15.2 14.9

Negative oil Shocks (1982-2001) % of GDP

28.0 5.6 21.0 7.0 5.8 6.5 16.0 9.0 5.1 3.0 25.3 13.4 11.9

% Change

- 11 - 39 0.05 0.05 - 24 0.08 0.02 39 - 36 - 12 - 16 - 12 - 20

Elasticity1

0.65 2.4 - 0.003 - 0.003 1.5 - 0.005 - 0.001 - 2.4 2.3 0.75 -------

Source: author‟s calculations 1. The percentage change in expenditures relative to the percentage change in total revenues between the two sub-periods. Sample average (15 oil-exporting countries)

5

Table (1) traces the changes in government expenditures in two distinctive episodes of oil price shocks: A period of oil booms in the 1970s and a period of low oil prices in the 1980s and 1990s. Public capital spending, especially on infrastructure, appeared pro-cyclical and was the most adversely hit by the negative shocks of the 1980s and 1990s. On the contrary, social expenditures on education, social welfare, health, and housing rose significantly to ease the negative social effects of the shocks. Government consumption and current spending as it appeared acyclical, remained almost neutral to the negative shock. The above aggregates naturally mask significant individual differences among the sample‟s countries in terms of fiscal policy responses to the oil shocks. For example, during the period 1970-2001, Nigeria and Venezuela experienced very similar boom-bust fiscal policy characteristics and very large swings in overall deficits, mainly driven by oil price developments. By the turning century, they became even more fiscally-dependent on oil. Indonesia, to the contrary, was more successful in insulating the budget and the economy against the same shocks. The share of oil revenues in total government revenues has gradually declined and the deficit was kept at modest levels. The key element to the Indonesian strategy has been spending the oil receipts to diversify the economy. Hence, they had a faster growth of public capital spending relative to current expenditures. The United Arab Emirates is another example of a country that took important measures to diversify its fiscal revenue base and to contain expenditures by the mid 1990s. However, the bulk of the fiscal retrenchment fell on capital expenditures, primarily, the infrastructural sector. Social expenditures on education and health has increased and spending on subsidies and transfers remained almost constant. Kuwait, to the contrary, has had a less diversified economy. By the end of the 1990s, the composition of government spending has shifted towards a growing share of wages, benefits, and transfers. Meanwhile, the relative share of capital expenditures in total government spending has declined. In addition, a large growing public sector (80% of labor force) contributed to a more rigid fiscal structure and to larger non-oil fiscal imbalances. However, Kuwait‟s fiscal strategy remained committed to building up sizable financial assets for future generations. The income generated from investing those assets has been successfully used to cushion revenue fluctuations.11

3. The Empirical Model Our goal is to estimate the direct effect oil prices play in oil exporters‟ growth, after controlling for the fiscal indirect channel, in a panel of 15 oil exporters during the period 19702004. Therefore, we consider the following dynamic panel-data model:

yit  yit 1   'Vit   ' Fit   ' X it  eit eit   i   it

; i = 1, 2, …, N;

t = 2,3,…., T

(1)

Where, y is the rate of growth of real GDP per capita; V is a vector that includes a measure of oil price uncertainty, oil price shocks, and oil shock dummies; F is a vector of fiscal policy variables (discussed below); X is a vector of current and lagged values of additional non-fiscal control variables, The above model represents an error component model where eit is an observationspecific error term that has two components: ηi, an unobserved time-invariant (fixed), country11

IMF Country Reports, various issues. 6

specific effect and ξit, an error term such that: E[ξit ξis] = 0 for i= 1, …N and all t ≠ s; E[ξit] = E[ηi] = 0; E[ηi ξit] = 0 for all i and t>2; and that E[yi1 ξi1] = 0. Γ = [α β‟ δ‟ φ‟]‟ is a vector of coefficients to be estimated. In order to capture the direct effect of oil prices on growth we include a measure of oil price shocks and a measure of oil price volatility. For oil price shocks, we use the yearly percentage change in oil prices, i.e., the growth rate in oil prices. If oil rents were truly a “curse” we would expect the relationship between the changes in oil prices and growth to be negative. The rationale for using this variable is that oil prices are known to follow a random walk, thus the best predictor for future price is current price. In that sense, the growth rate in oil prices reflects the “unanticipated” yearly changes in prices. However, some researchers would define an arbitrary threshold above which a change in prices is considered a shock.12 The downside of this approach is that the set of shocks will change by changing the threshold. To address the question of whether the increasing volatility of oil prices since the 1970‟s has hurt growth, we include a measure of oil price variability. The conditional variance of oil price shocks is obtained by employing a GARCH (1,1) model which utilizes monthly-frequency for crude oil prices (see table A.1 for GARCH results). The yearly-frequency conditional variance is then obtained by calculating the average volatility of the 12 month. In addition, in order to look for any asymmetric oil shock effects a group of dummy variables is used. Also time dummies are included. This oil price vector (V) is assumed to be strictly exogenous to the model. In addition to a host of potential non-fiscal factors, the differences in growth performance among oil exporters, and the “curse” outcome for many, could arise from the role fiscal policy plays in transmitting the shock. In that sense, fiscal policy could be the indirect channel through which oil windfalls my turn to a “curse” in some countries and a “bless” in others. Each of the following sub-sections explains the fiscal and non fiscal variables that are controlled for in the growth regression.

3.1. Controlling for Fiscal Policy: Every increase in spending has to be financed through collecting more revenues or issuing more debt. Conversely, changes in revenues will have implications either on expenditures or on the budget surplus. This financing interdependence, obviously, arises from the implications of the linear interdependence of the government budget constraint (GBC, thereafter). By introducing the GBC in the growth regression, both the sources of finance and expenditure composition are considered. Therefore, GBC= F 

k

 j 1

k

f j and  ' F    j f j , where f is the fiscal variables. j 1

However, since the GBC is an identity, such co-linearity requires one element to be omitted for the purposes of estimation. So, let variable k be omitted. It can be thus written as f k  

k 1

f j 1

j

.

By construction, this variable represents the compensating or freely varying method of finance of a change in one of the included variables, provided all other budget items remain unchanged.

12

For example, it is common to use two standard deviations as a threshold. 7

Omitting that k element will slightly change the usual interpretation of the fiscal coefficients. The estimated coefficients of the fiscal variables will now represent the growth effect of that particular variable, net of the effect of the method of financing (i.e., the omitted compensating variable). For example, if total revenue is eliminated from the GBC and total expenditures and the budget balance are included in the regression, the coefficient of total expenditures should be interpreted as the net effect of a change in total expenditures compensated by a change in revenues, leaving the budget balance unaffected. Formally, the fiscal vector of ' explanatory variables can be expressed as  Ft 

k 1

 j 1

* j

f jt ; where, k is the total number of

variables in the GBC, and the k element is assumed omitted. F is a vector of included fiscal variables

f,

and

 '  [ *j ,....... k*1 ] is a 1xk-1 vector of coefficients such that

 *j  ( j   k ) for all j=1, …, k-1. This approach allows: first, studying the growth effects of revenue booms that are transmitted through the changes in government expenditures; second, studying the growth effects of debt finance. In this case, the changes in (included) revenues or expenditures are compensated by changes in the (omitted) budget balance; third, studying the growth-effects of different expenditure types. So, expenditures can be decomposed into social, capital, current, infrastructural, government transfers, etc.13

3.2. Non-fiscal Controls: Levine and Renlet (1992) and Sala-i-Martin (1997) explored a wide range of variables to determine the most robust in the growth literature. Therefore, the non-fiscal control variables in X are picked in accordance with those found to be robust in these studies. Namely, we include: (i) Investment: since the fiscal vector includes capital spending by the government, private capital formation is included in X. (ii) Openness: the literature documents positive robust growth effects of the degree of openness. (iii) Democracy and institutions: Weak political institutions, a symptom of autocratic regimes, can be detrimental to growth. Polity Scores index is used to scale countries by the strength of their democratic institutions and authority traits. Countries with higher Polity scores tend to have more democratic regimes and stronger conflict management institutions. (iv) Education: Human capital is expected to have a positive long-run growth effects. (v) Inflation: it is expected that inflation would have a negative effect on growth.

4. Econometric Methodology The dynamic error component model specified in (1) suggests three potential problems. First, the country-specific fixed effects are certainly correlated with the lagged dependent variable, and hence static panel techniques are not appropriate for such models since it results in a systematic bias in the estimates of the lagged dependent variable. Second, the country-specific 13

Table A.2 gives a full description of the variables and data sources. 8

fixed effects are likely to be correlated with some of the fiscal or non-fiscal explanatory variables in F or X. This implies that the current disturbance is correlated with the explanatory variables, rendering the OLS and random effect FGLS estimators inconsistent. Third, some of the model‟s explanatory variables may not be strictly exogenous. For example, a shock to output growth may determine future and/or current inflation, government spending, and private investment. Therefore, simultaneity bias arising from the joint determination of such variables and GDP growth is a concern. To control for all three problems, the study utilizes a Generalized-Methodof-Moment (GMM) System estimator developed by Arellano and Bover (1995), and Blundell and Bond (1998).14 The first step, as suggested by Arellano and Bond (1991) is to treat the model as a system of first-differenced equations, one for each time period, making use of all the linear-moment conditions implied by a dynamic panel-data model. So, we start by first-differencing the equation to remove the individual-specific effects.

y it  y it 1   ' Vit   ' Fit   ' X it   it ;

i=1, 2…, N; t = 3,…T

(2)

This transformation takes care of the first two problems as it eliminates the potential correlation between the regressors and the individual fixed effects that drops out as a result of differencing. However, that introduces a (first-order) autocorrelation in the transformed disturbances, (it - it1), and hence between this term and the lagged dependent variable (Yit-1 - Yit-2). Next, we need to define a matrix of valid instruments Zdi to instrument for the lagged dependent variable and any other potentially endogenous regressors in xit, then exploit the orthogonality conditions:

E ( Z di' ei )  E ( Z di'  i )  0 , where, ∆ei = ∆ξi = [∆i3, ∆i4,….., ∆iT]‟

(3)

Arellano and Bond (1991) suggest using all available lagged levels of the predetermined explanatory variables (including the lagged dependent) and differences of the strictly exogenous variables as internal instruments for the first-differenced equations. But, since the study‟s panel has a relatively long time dimension and a relatively small number of countries, we use a subset of the available moment conditions restricting the instrument matrix to only one lag for each predetermined explanatory variables. The last step is to define the instruments for the system estimator developed by Arellano and Bover (1995), and Blundell and Bond (1998). This estimator treats the model as a system of difference and level equations. Therefore, in addition to the moment conditions in (3), they suggest using the lagged differences of the predetermined variables as instruments for their levels. The strictly exogenous variables are instrumented, as in the standard procedure, by their own differences. This exploits the following extra moment conditions. E (y it 1eit )  0 ; for t =3,……T; and E (xit eit )  0 ; for t =2,……T (4) The validity of the extra orthogonality conditions in (4) requires an additional assumption. That is: E(ηi ∆yi2) = 0 and E(ηi ∆xi2) = 0, or in words, the changes in yit and xit are uncorrelated with the individual effects. Since –as mentioned earlier- the levels of yit and xit are potentially correlated with ηi,, The above assumption would hold only if this correlation remains constant across all time periods. Hence the system‟s set of instruments, comprising difference and

14

For a detailed description of this method, see: Blundell, Bond, and Windmeijer (2000).

9

some extra level instruments, can be simplified and expressed by a matrix Zsi of the following form:

Z

si

Z   di  0

0  Z li 

Where, Zdi is a block instrument matrix for the first-differenced equations, as previously defined, and Zli is a matrix of some extra instruments for the level equations. The System GMM estimator,

ei   i    , ∆ei = ∆ξi =  ei   ei 

thus, exploits the moment conditions: E ( Z si' i )  0 , where i  

[∆i3, ∆i4,….., ∆iT]‟ , and ei = [ei3, ei4, ….., eiT]‟. The validity of the above-suggested instruments hinges on two major assumptions. First, the disturbances it are not serially correlated; a requirement for the consistency of the GMM estimator. Therefore, we report the Arellano-Bond test for autocorrelation, applied to the residuals from the first-differenced equations. Second, the initial conditions yt1 are not correlated with the subsequent disturbances. We report the one-step Sargan test. The statistic is distributed as 2 under the null of valid instruments and utilizes the minimized criteria of the one-step estimators. It is also worth mentioning that the number of GMM instruments significantly increases as the number of endogenous variables grows. Meanwhile, if the number of groups is relatively small, as in the case of our sample, the resulting estimates can be seriously biased. Therefore, highly disaggregated fiscal variables in the growth system GMM regression will form a serious burden on the estimation if each of these variables is considered endogenous. In order to circumvent this problem, the GBC enters the model with a one period lag. This treatment is also consistent with the fact that fiscal policy is likely to affect growth after a period of delay because of possible institutional, legislative and /or implementation lags. Lastly, it is quite common in the growth literature to average the data over a number of years. Typically, the length of each period ranges from 3 to 10 years. One main advantage of averaging is to explore the long-run relationships and eliminate cyclical fluctuations. However, this comes at the cost of losing degrees of freedom and arbitrarily determining the length of the period. An alternative approach, the one used in this study, is to use annual observations and an appropriate number of lags in order to capture the long-run effects.15 The appropriate lag structure is determined using F-tests of the null that an additional lag on all explanatory variables is jointly insignificant. This approach has the advantage of distinguishing between the short and long run effects, exploiting the dynamic properties of the data. The panel is unbalanced and the annualfrequency data covers all or part of the period 1970-2004.

5. Empirical Results The growth model is estimated using annual observations, and therefore the long run effects are captured through adding a statistically significant number of lags. Starting from eight lags, the model is reduced to five lags on all variables. The fiscal variables of interest did not turn out to have any significant effects (individually and jointly) beyond five lags. Thus, in our sample 15

See for example: Kocherlakota and Yi (1997), and Kneller, Bleaney, and Gemmell (2001). 10

of oil-exporters, a five-year period would roughly capture any long-run growth effects of fiscal policy.16 F-test on the joint significance of fifth lag rejects the null that this lag is insignificant. The Arellano-Bond test for serial correlation is insignificant for all system estimations, hence the null of no second order autocorrelation cannot be rejected. Similarly, the one-step Sargan test is insignificant for all estimations hence the null that the moment conditions hold for the system cannot be rejected.17 The empirical investigation proceeds as follows. The long run growth effects, before and after including total government expenditures and the budget balance, are estimated first. Total government revenues are considered the financing element in the budget, hence omitted (results reported in table 2). But since the highly aggregated expenditures would mask the potential growth-effects of the different types of spending, a more disaggregated government budget constraint is used next. Expenditures are decomposed according to different classifications. For example, expenditures are disaggregated into capital and current. Then the latter is further decomposed into wages, government purchases, and transfers. Finally, we use the economic classification of expenditures by function. For all specifications in the second set of estimations (reported in table 3) we compare different assumptions regarding the government financing method. So, results for similar specifications are reported under different financing assumptions. The changes in expenditures are first assumed to be financed through changes in revenues, or compensated by changes in the budget balance (through debt finance), or finally through taxes. In addition, total revenues are further disaggregated to look into the effects of the tax and non-tax components.18 As a first step four basic growth regressions are estimated (table 2). Before introducing fiscal policy, in column 1, all other control variables except for private investment appeared significant and have the correct sign. The changes in oil prices seem to have a positive effect on long-run growth. A 1% increase in oil prices increases the growth rate by 0.03 percentage points. On the other hand, oil price volatility does not appear to have long run growth effects.19 The system is then re-estimated after including the GBC. Private investment and oil price volatility remained individually insignificant. Four important preliminary results need to be highlighted:  First: once fiscal policy is controlled for, the coefficient of democracy and the characteristics of political institutions loses significance and magnitude, lending some support to the notion that the political and institutional qualities exerts a negative effect on long run growth through the fiscal process. In other words, the detrimental effect of weak institutions works through fiscal policy.  Second: the positive effect of oil price changes on growth remained significant despite losing magnitude (dropped from 0.07 to 0.006). Thus, oil price shocks can explain 16

Bleany, Gemmell and Kneller (2001) found eight lags to be appropriate for OECD countries. Similarly, Kocherlakota and Yi (1997) found a same lag length for UK and US data. 17 The one-step Sargan test, despite being not robust to the presence of serial correlation and heteroskedasticity in the errors, it tends to have a higher frequency of rejection (of the null of appropriate instruments) than the robust (two-step) Sargan test (Arellano and Bond 1991).Therefore, the failure of a one-step Sargan test to-even weakly- reject the null lends strong support to the validity of the instruments. 18 The model was also estimated after adding post 1980 dummy, highly oil-dependent countries dummy. The results from these estimations did not seem to differ much from those reported. 19 However, volatility of oil prices appeared to be significant in the short run coefficients for first and second lags, which indicate that it may have transitory but not permanent effects on growth. 11

very little of the variability in the rate of growth, after controlling for fiscal policy. That clearly indicate that the decline in oil prices during the period 1981 to 1998 does not in effect explain much of the huge drop in average growth rates (by almost -7.4 percentage points) during the same period.20 Therefore, the decline in growth as oil prices declined is a mere correlation rather than causality. Table 2: Long-run growth effects: Aggregated Expenditures (Financing variable: government revenues)2 One-Step System GMM results3 (1) (2) Variables

Dlroilp

0.07** (2.54)

0.02* (1.80) 0.01** (3.04) -0.10 (1.65) 0.02 (1.09) -0.04*** (3.01) -0.12*** (8.07) 0.02*** (4.03) 0.006** (2.86)

VoloilP

-0.02 (1.51)

-0.01 (1.14)

Enrol Openness Polity PINV Inf

0.03** (2.31) 0.06*** (3.00) -0.13* (1.95) 0.06 (1.24) -0.05*** (2.81)

BBalance Totexp

Negshk_dum AR(2) test Sargan test F-test: 5th lag

Z=-1.58 [p-value= 0.12] Chi2=91 [p-value= 0.34] F= 2.36* [p-value= 0.06]

Z=-0.67 [p-value= 0.42] Chi2=90.96 [p-value= 0.48] F= 1.65 [p-value= 0.12]

(3) 0.03** (2.15) -0.006 (1.03)

0.06*** (3.80) -0.19*** (9.77) 0.02*** (4.47) 0.01 (0.59)

2.12 (0.31) Z= 0.09 [p-value= 0.97] Chi2=133.4 [p-value= 0.50] F= 1.65 [p-value= 0.12]

Note: The long run coefficients are calculated as the sum of the contemporaneous and lagged coefficients divided by one minus the coefficient of the lagged dependent variable. Five lags are included for all independent variables. F-statistics is reported under each coefficient. Fiscal policy variables entered with one period lag ***, **, * indicates 1% , 5%, and 10% significance levels, respectively. All estimations include dummies for the years 1974, 1979, 1986, 1991, 1998, 2000, and 2001.



20

Third: an interesting result is that accumulating budget surpluses as a result of a surge in government total revenues (the method of finance) retards long run growth. That effect is highly significant and strikingly large in magnitude. A one percentage point surge in the surplus (as a ratio of GDP) lowers growth rates by 0.12 percentage points. Many empirical studies have shown that high dependency on a primary commodity is negatively correlated with growth. Therefore, one may argue that oil-dependent economies would likely experience higher surpluses and, meanwhile, lower average growth rates. That is: the budget balance may mediate the negative correlation between oil dependency and growth. However, controlling for oil dependency (results not shown)

The average annual drop in oil prices during the period 1981-1998 was 7%.

12

revealed that a highly significant negative correlation between the budget surplus and growth would still hold.  Fourth: government spending –financed through a surge in revenues- is growth improving. The effect is highly significant and moderate in size. The remainder of this section aims to further investigate the above preliminary findings. An F-test on jointly dropping the political characteristics, private investment, and oil price volatility could not be rejected at the 5% level. Thus, all three insignificant variables were dropped in all estimations afterwards. The last estimation in Table (3) tests the hypothesis of asymmetric growth effects of negative and positive oil price shocks. A pulse dummy for negative shocks is added to capture any possible excess average growth-effects of negative shocks. The results, in column 4, show that this term is not significantly different from zero, indicating that oil price shocks may not have any significant asymmetric growth-effects in the long run. In addition, the coefficient of oil price shocks is not robust to the inclusion of a negative shock dummy. Next, we disaggregate the government budget constraint (results reported in table (3)). In line with the previous estimates, the changes in oil prices still maintain a significant positive but relatively small effect on growth. In what follows, we pool the results from the disaggregated estimation into two groups. The group of results related to governments‟ method of finance is discussed first, and then we turn to the growth effects of the different types of expenditures. The first three columns in table 3 represent estimations in which government revenue is omitted, hence assumed the compensating element in the budget constraint. The budget balance effects seem to be robust to the disaggregation of expenditure, and are still significantly negative with a relatively large marginal effect on growth in the range of (negative) 0.12 to 0.27 percentage points. This result needs more careful consideration, since it is seemingly at odds with known economic paradigms. In a Ricardian world, the budget surplus should have no significant effects on growth. The Keynesian paradigm, on the other hand, predicts- as many empirical studies have found for developing countries- that this effect would be positive since it implies less debt-financing. Note, however, that the negative effect found here has two components: the marginal effect of the budget balance net of the effect of government revenues. In other words, it measures the effect of accumulating government surplus as a result of a revenue boom, and vice versa. Therefore, the negative relationship could be rather arising from the source of the surplus (deficit). For example, Talvi and Vegh (2000) argue that higher government revenues at boom times create political pressures to increase spending. Such pressures ultimately result in weakening the fiscal control, extending spending on non-productive expenditures. In addition, experience in many resource-rich countries, e.g., Mexico, Nigeria, Venezuela and others, has revealed that revenue windfalls trigger wide-spread rent-seeking competitions over the state‟s oil revenues on all government levels. That effect is found to be detrimental to growth. Conversely, revenue downturns results in less surpluses (higher deficits), alleviating such rent-seeking pressures. Our analysis, however, accounts only for the first channel (expenditures) but not for the second (rent-seeking). To further investigate this, one suspect could be influential observations. A pooled OLS partial regression plot for budget balance and growth (Figure 1) reveals few observations with highly negative growth rates and exceptionally high surpluses (as a ratio of GDP). So, first these observations were removed, and then revenues are decomposed into tax and non-tax revenues. In some oil-rich countries, non-tax revenues comprise most of the direct revenues that accrue to the government from oil, including royalties and sales that were transferred to the budget. However, in others, especially those that have national major oil companies, the “budget” share of the oil sales is transferred to the budget through taxes. So, this decomposition does not clearly or necessarily correspond to a non-oil/oil revenue decomposition. It only serves as a rough approximation.

13

Table 3: Long-run growth effects: disaggregated expenditures and revenues under different financing assumptions. One-Step System GMM results. Dependent Variable: real GDP per capita growth rate Different Assumptions Regarding The Method of Finance Variable REVENUES DEBT FINANCE TAXES Enrol Openness Inf Dlroilp BBalance

(1) 0.03* (2.10) 0.01*** (3.50) -0.05*** (3.53) 0.05** (2.74) -0.12*** (5.81)

(2) 0.02 (1.81) 0.01*** (3.29) -0.04* (1.87) 0.03** (2.62) -0.17*** (7.12)

(3) 0.03** (2.79) 0.01*** (4.15) -0.05* (1.89) 0.02* (1.96) -0.27*** (8.35)

Tax NonTax Capexp Curexp CapXdoilp CurXdoilp

0.17 (0.79) 0.01 (1.52) -0.92** (2.44) 0.18 (0.96)

Wages

0.41** (2.90)

-0.35* (1.96)

Curtransf

-0.02 (1.17) -0.07*** (5.76) 0.37 (1.37) 0.07 (0.93) -0.71** (2.86)

-0.09 (1.01) -0.08*** (7.88) 0.52** (2.28)

-0.03* (2.12) -0.09*** (7.53)

-0.34** (2.72)

Z =-0.53 [P = 0.59] Chi2=90.4 [P = 0.44]

Z=-1.12 [P = 0.20] Chi2=89.2 [P= 0.47]

Z=-0.91 [P = 0.31] Chi2=81.6 [P = 0.56]

Z=0.01 [P = 0.98] Chi2=90.0 [P = 0.45]

Z= -0.62 [P = 0.53] Chi2=92.1 [P = 0.41]

0.20* (1.97) 0.42*** (4.23) 1.1*** (3.85) 0.03* (2.05) Z= 0.89 [P = 0.37] Chi2=83.5 [P = 0.59]

F= 2.96*** [P = 0.00]

F= 2.23** [P = 0.03]

F= 1.98** [P = 0.05]

F= 2.37** [P = 0.02]

F= 1.92** [P = 0.04]

F= 1.75* [P = 0.08]

Publicserv Otherexp

(7)

0.03 (1.33) 0.01*** (3.34) -0.04** (2.38) 0.06*** (2.99) -0.07 (1.01)

-0.01** (2.22) 0.34 (1.46) -0.004 (0.17) -0.60** (2.46) 0.11 (0.89)

0.03 (1.38) 0.07 (1.51) 0.12** (2.57) -0.10** (2.62) 0.29** (2.45) 0.93*** (3.89) 0.08 (1.82)

Infra

F-test: 5th lag

(6)

0.04* (1.89) 0.009*** (3.62) -0.05* (1.95) 0.03 (1.66)

0.09 (1.10)

Socialexp

Sargan test

(5)

0.02 (1.48) 0.008*** (3.21) -0.02 (1.65) 0.02*** (3.34)

-0.09** (2.29) -0.07** (2.50) -0.001 (1.82)

Goods&srv

AR(2) test

(4)

0.02 (1.10) 0.008*** (3.42) -0.03* (1.84) 0.08*** (3.58)

Notes:

Z=-0.15 [P = 0.88] Chi2=90.6 [P = 0.43] F= 2.57*** [P = 0.01]

All fiscal policy variables enter with one period lag. Long run effects are calculated as the sum of the contemporaneous and lagged coefficients divided by one minus the coefficient of the lagged dependent variable. Fstatistics is reported under each coefficient. ***, **, * indicates 1% , 5%, and 10% significance levels, respectively. All estimations include dummies for the years 1974, 1979, 1986, 1991, 1998, 2000, and 2001.

14

The estimation in column 7 is identical to that in one except that taxes are now assumed the method of finance instead of total revenues. The results show that, indeed, once non-tax revenues are included, the budget balance turned out to be highly insignificant, despite the sign remains negative. In addition, higher non-tax revenues at the expense of dwindling tax revenues hurts growth. These findings prompt two main conclusions. First: the seemingly negative growth effects of the budget balance are indeed driven by some influential observations and by the negative effect of non-tax revenues that in part reflect oil revenue windfalls. Second: reducing the fiscal dependency on oil by widening the non-oil tax base and increasing its share in total revenues is growth-improving. In estimations 4 to 6, the budget balance is omitted, assuming that the changes in other fiscal variables will be accommodated by changes in the size of debt. Non-tax revenues still maintain a negative and a strongly significant effect on long-run growth even when they lower the debt (or are used to accumulate a surplus), confirming our previous insights. This, again, supports the preceding argument that oil revenue windfalls are harmful for growth when they accrue to the government. Furthermore, taxes have no significant effect on growth.21 One possible explanation is that countries that are highly dependent on oil in their fiscal structures have a relatively small non-oil tax base. Next, we turn to the expenditure side of the budget. Expenditures are first decomposed into current and capital components. The evidence- as presented in table 3- is mixed. Government capital spending, whether financed through revenues or debt, seems to have a positive large effect on long-run growth. The magnitude of this effect could be as high as 0.5 a percentage point for a one percentage point increase in capital expenditures to GDP. However, this variable is not robust to the decomposition of current expenditures. In addition, it could be argued that countries with a larger initial share of capital expenditures would be better able to cope with a negative revenue shock. A better infrastructural foundation, for example, may help protect productivity from declining for some time when adverse shocks occur. To test this hypothesis, an interaction term of capital spending with the changes in oil prices is introduced. We expect this term to have a significantly negative coefficient. So, at high levels of initial public investment (as a ratio of GDP) the negative shock would have a smaller adverse effect on growth.22 The results seem to lend support to this hypothesis as the multiplicative term of capital spending and oil price shock turned out to be significantly negative. This result can also be interpreted in a number of interesting ways. First, it, conversely, means that the higher is the ratio of government capital spending to GDP the lower is the (net) positive effect of a favorable change in oil prices. One possible interpretation can be found in many oil countries‟ experiences at boom times where surpluses are used to fund non-productive investments. Therefore, countries with higher levels of public capital spending would benefit less from an oil boom. Second, capital expenditures have non-linear effects on growth with respect to oil price shocks. When an adverse shock occurs, cutting capital spending would sharply deepen the negative effect of the shock. This „additional‟ negative effect is larger the larger is the size of 21

The empirical growth literature provides no conclusive evidence on the growth-effects of taxes. Many studies found insignificant effects of taxes (See: Levine and Renelt, 1992). 22 The marginal effect of a negative oil price shock: change in oil prices (-)* coefficient (-) * capital expenditures to GDP (+) = effect on long run growth (+). So, this term mitigates the adverse direct effect of the negative shock on growth. 15

the shock and the sharper is the cut in capital expenditures.23 Clearly, this non-linear effect will also be negative when capital spending is extended to lavish and non-productive investments during booms and will become larger the larger is the size of the positive shock. The results also show that aggregate current expenditures do not have a robust growth effect, regardless of the way it is being financed. However, when current expenditures are further disaggregated into wages, purchases of goods and services and current transfers, the components seemed to have different implications for growth. Changes in spending on wages and on purchases of goods and services are harmful for growth (column 2), or at best neutral (column 5). The implication is that the growth effects of those types of spending range from negative to zero depending on the financing method, indicating that those expenditures are mostly unproductive. However, in times of scarcity, a deficit financed increase in current transfers by a one percentage point would lead to higher growth rates by 0.12 percentage points (column 5). Lastly, total expenditures are grouped into social expenditures (spending on education, health, social welfare and housing), infrastructure (spending on energy, transportation, communication, forestry, water and sewerage), public services, and all other remaining expenditures. The results (columns 3 and 6) show that spending on infrastructure and on public services, irrespective of the method of finance, has positive effects on growth. This effect is, however, implausibly large for public services. But it generally indicates that spending more on public services pays off in terms of faster growth rates and improvement in the quality of life in these countries. On the other hand, more spending on social expenditures at boom times (through revenues) seems to be harmful for growth. However, increasing such expenditures at times of scarcity, even when it means higher deficits, would have positive effects on growth since it would partly ameliorate the negative effect of the shock.

6. Conclusion In this paper, the evidence from 15 oil-exporters shows that oil price shocks are not detrimental to long run growth. Higher oil prices tend to increase growth rates in the long run as well as the short run, and vice versa. However, the magnitude of the direct effect cannot explain much of the collapse in growth in many oil exporters during the 1980s and 1990s. Oil price volatility does not retard long run growth as well. Therefore, the resource rent is positively related to growth. However, the way the government spends its resource windfalls, and the way it adjusts its spending in a downturn seem to explain the differences in growth performance across oil exporters. The main policy implications are: (i) diversification policies and expanding the non-oil tax base are growth-improving. (ii) At times of scarcity, governments should avoid cutting capital expenditure and should increase social spending to help stimulate growth (iii) At times of plenty, the government should apply more prudent investment policies and allocate more resources to productive expenditures such as infrastructure as well as to improving the public services. In addition, it is extremely curtail to weigh the policy options of eliminating the fiscal surplus through transferring the excess revenues to an autonomous stabilization (wealth) fund versus reducing public debt. In addition to maximizing the social net gains of the revenue surge, eliminating the surplus would alleviate competitive rent seeking pressures. 23

The marginal effect of a decline in capital spending in the wake of a negative oil price shock= The change in capital expenditures (-) * the change in oil prices (-)* the coefficient of the interaction term (-) = (-) effect. 16

References Alesina, A. and Perotti, R. (1996), Fiscal Adjustments in OECD Countries: Composition and Macroeconomic Effects, WP/96/70, Washington DC: IMF. Ardagna, S. 2001, “Fiscal Policy Composition, Public Debt, and Economic Activity”, Public Choice, 109, 301-325. Arellano, M and Bond, S. (1991), “Some Tests of specification for Panel Data: Monte Carlo Evidence and Application to Employment Equations”, Review of Economic Studies, 58, 277-297. Arellano, M. and Bover, O. (1995), “Another Look at the Instrumental Variable Estimation of Error-components models”, J. Econometrics, 68, 29-51. Auty, R. (1994), “Industrial Policy Reform in Six Large Newly Industrializing Countries: The Resource Curse Thesis”, World Development, 22, 11-26. Barro, R. (1990), “Government Spending in a Simple Model of Endogenous Growth”, J. Political Economy, 98(1), s103-s117. ________ (1991), “Economic Growth in a Cross- Section of Countries”, Quarterly J. Economics, 104, 407-444. Barro, R. and Sala-I-Martin, X. (1992), “Public Finance in Models of Economic Growth”, Review of Economic Studies, 59 (4), 645-661. Basu, P. and McLeod, D. (1992), “Terms of Trade and Economic Fluctuations in Developing Countries”, J. Development Economics, 37, 89-110. Bleaney, M., Kneller, R., and Gemmell, N. (2001), “Testing the Endogenous Growth Model: Public Expenditure, Taxation and Growth over the Long Run”, Canadian J. Economics, 34(1), 36-57. Blundell, R. and Bond, S. (1998), “Initial Conditions and Moment Restrictions in Dynamic Panel data models”, J. Econometrics, 87, 115-43. Blundell, R. and Bond, S., and Windmeijer, F. (2000), “Estimation in Dynamic Panel Data Models: Improving on the Performance of the Standard GMM Estimator”, Advances in Econometrics: Nonstationary Panels, Panel Cointegration, and Dynamic Panels, 15, 53-91. Cashin, P., Liang, H. and McDermott, C. (1999), How Persistent Are Shocks to World Commodity Prices?, WP/99/80, Washington DC: IMF. Cashin, P., McDermott, C. and Scott, A. (1999), Booms and Slumps in World Commodity Prices, WP/99/155, Washington DC: IMF. Davarajan, S., Swaroop, V. and Zou, H. (1996), “The Composition of Public Expenditure and Economic Growth”, J. Monetary Economics, 37(3), 313-344. Deaton, A. (1999), “Commodity Prices and Growth in Africa”, J. Economic Perspectives, 13(3), 23-40. Deaton, A. and Miller, R (1995), “International Commodity Prices, Macroeconomic Performance, and Politics in SubSaharan Africa”, Princeton Studies of International Finance, 79. Easterly, W. and Rebelo, S. (1993), “Fiscal Policy and Economic Growth: an Empirical Investigation”, J. Monetary Economics, 32 (3), 417-458. Easterly, W., Kremer, M., Pritchett, L. and Summers, H. (1993), “Good Policy or Good Luck? Country Growth Performance and Temporary Shocks”, J. Monetary Economics, 32 (3), 459-48. Elanshasy, A., Bradley, M., and Joutz, F. (2005), “Evidence on the Role of Oil Prices in Venezuela‟s Economic performance,” the 25th Annual North American Conference Proceedings of the International Association of Energy Economics, Denver: September 18-21. Engel, E. and Valdes, R., 2000, Optimal Fiscal Strategy for Oil Exporting Countries, IMF WP/00/118, Washington D.C.: IMF. Garrison, C. and Lee, F. (1995), “The Effects of Macroeconomic Variables on Economic Growth Rates: A CrossCountry Study”, J. Macroeconomics, 17 (2), 303-317. Gleb, A and associates, (1988), Oil Windfall: Blessing or Curse, New York: Oxford University Press. Greiner, A. and Semmler, W. (2000), “Endogenous Growth, Government Debt and Budgetary Regimes”, J. of Macroeconomics, 22(3), 363-384. Hausmann, R., Powell, A. and Rigobon, R., 1993, “An Optimal Spending Rule Facing Oil Income uncertainty (Venezuela),” In External Shocks and Stabilization Mechanisms, eds. E. Engel and P. Meller, Washington DC: Inter-American Development Bank. Hausmann, R. and Rigobon, R. (2002), An Alternative Interpretation of the ‘Resource Curse’: Theory and Policy Implications, NBER working paper No. 9424. Husain, A., Tazhibayeva, K. and Ter-Martirosyam, A. (2008), Fiscal Policy and Economic Cycles in Oil-exporting Countries, WP/08/253, Washington DC: IMF Isham, J., Prichett, L., Woolcock, M., and Busby (2004), 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.

17

Jones, L.; Manuelli, R. and Rossi, P. (1993), “Optimal Taxation in Models of Endogenous Growth”, J. Political Economy, 101 (3), 485-517. King, R. and Rebelo, S. (1990), “Public Policy and Economic Growth: Developing Neoclassical Implications”, J. Political Economy, 98 (5), s126-s150. Kneller, R., Bleaney, M. and Gemmell, N. (1999), “Fiscal Policy and Growth: Evidence from OECD Countries”, J. Public Economics, 74, 171-190. Kocherlakota, N. and Yi, K. (1997), “Is there endogenous Long-Run Growth? Evidence from the United States and the United Kingdom”, J. Money, Credit and Banking, 29 (2), 235-262. Lane, P. and Tornell, A. (1998), Voracity and Growth, Discussion Paper No. 654, Harvard Institute for International Development, Harvard University. Levine, R. and Renlet, D. (1992), “A Sensitivity Analysis of Cross- Country Growth Regressions”, American Economic Review, 82(4), 942-963. Mehlum, H., Moene, K. and Torvik, R. (2006) ”Institutions and the resource curse”, Economic Journal 116, 1-20. Mendoza, E., Milesi-Ferreti, G. M. and Asea, P. (1995), “Do Taxes Matter For Long-Run Growth?”, International Finance Discussion Papers, No. 511, Board of Governors of the Federal Reserve System. Mendoza, E. (1997), “Terms-of-Trade Uncertainty and Economic Growth”, J. Development Economics, 54 (2), 323356. Miller, S. and Russek, F. (1997), “Fiscal Structures and Economic Growth: International Evidence”, Economic Inquiry, 35, 603-613. Rebelo, S. (1991), “Long-Run Policy Analysis and Long-Run Growth” , J. Political Economy, 99(3), 500-521. Rodriguez, F. and Sachs, J. (1999), “Why Do Resource Abundant Economies Grow More Slowly? A New Explanation and an Application to Venezuela”, J. Economic Growth, 4(3), 277-303. Rodrik, D. (1998), “Where Did All the Growth Go? External Shocks, Social Conflicts, and Growth Collapses”, NBER Working Paper No. 6350. Sala-I-Martin, X. (1997), “I Just Ran Two Million Regressions”, American Economic Review, 87(2), 178-183. Sala-i-Martin, X. and Subramanian (2003), Addressing the Natural Resource Curse: An Illustration from Nigeria, WP/03/139, Washington DC: IMF Sachs, J.D. and Warner, A. (1995), “Natural resources Abundance and Economic growth”, NBER Working Paper No. 5398. Slemrod, J. (1995), “What Do Cross- Country Studies Teach about Government Investment, Prosperity and Economic Growth?”, Brooking Papers on Economic Activity, 2, 373-431. Talvi, E. and Vegh, C. (2000), “Tax Base Variability and Procyclical Fiscal Policy”, NBER Working Paper No. 7499. Tornell, A. and Velasco, A. (1992), “The Tragedy of the Commons and Economic Growth: Why Does Capital Flow from Poor to Rich Countries?”, J. of Political Economy, 100(6), 1208-31.

18

Appendix: Figure A.1: Partial Regression Plots (Pooled OLS) (2) Oil price shocks

20 10 0 -10 -20

-20

100

-50

100

-10 0 10 e( curexp | X )

20

coef = .01522769, se = .03478924, t = .44

(4) Current Expenditures

0 20 40 e( bbalance | X )

60

coef = -.1868855, se = .04588223, t = -4.07

20 10 0 -10 -20

-10

0

10

20

e( rgdppcgrowth | X )

30

-20

-20

coef = .02483157, se = .01196101, t = 2.08

-20

-20

-10

0

10

20

e( rgdppcgrowth | X )

30

0 50 e( dlroilp2 | X )

30

-50 0 50 e( oildpnd | X )

coef = -.03295237, se = .00848925, t = -3.88

e( rgdppcgrowth | X )

e( rgdppcgrowth | X )

-10

0

10

20

e( rgdppcgrowth | X )

20 10 0 -10 -20

e( rgdppcgrowth | X )

-100

(3) Budget Balance

30

30

(1) Oil dependency

-10

0 10 e( capexp | X )

20

coef = .15982839, se = .08256105, t = 1.94

(5) Capital Expenditures

19

-20

-10 0 10 e( pcapf or | X )

20

coef = .13712618, se = .05127728, t = 2.67

(6) Private Capital Formation

Table A.1: Conditional variance of oil price shocks: GARCH (1, 1) Model ARCH GARCH

0.4 (0.00) 0.6 (0.00)

Sample 1957 – 2008 Q-stat for autocorrelation in mean eq. Q-stat for ARCH in variance eq.

622 obs.

0.97 (0.81) 0.37 (0.95) Heteroskedasticity Test: ARCH F-statistics= 0.17 (0.97) Obs*R-squared = 0.86 (0.97) P-values in parentheses. Bollerslev-Wooldridge robust standard errors & covariance are used. Mean equation: dloilp on dloilp(-1) dloilp(-1)2 dloilp2. Where: dloilp is the log difference of crude world oil prices.

Table A.2: Variables and Data Sources Variable

Definition

Real GDP per capita growth rate Dlroilp Voloilp

The log difference of GDP per capita, deflated by countries‟ GDP deflators. The log difference of real world crude oil prices. Conditional standard deviation from a GARCH(1,1) model (see table:1)

Inf Polity

The growth rate of the consumer price index Polity scores. A composite scale that ranges from 10 (strongly democratic) to -10 (strongly autocratic). The log difference of private capital formation as a % of GDP. Private capital formation is calculated as the difference between gross fixed capital formation (obtained from WDI) and government capital formation as % of GDP (obtained from GFS). The log difference of total (primary + secondary) school enrolment ratio. The log difference of a measure of openness, defined as exports plus imports (ratio of GDP) A composite measure of dependency on oil, defined as the sum of oil as a ratio of total commodity exports and oil as a ratio of GDP. Pulse dummy =1 if the growth in real oil prices