Oil dependency and happiness in net oil-exporting countries: is it a curse or blessing?

ABSTRACT

The resource curse hypothesis postulates that countries endowed with and dependent on abundant natural resources tend to underperform in socioeconomic and development outcomes than those with fewer natural resources. Recently, a few studies argued that this curse also manifests in lower life satisfaction or happiness. Focusing on 31 net oil-exporting countries over the 2006–2019 period, we find no evidence that oil rents (and aggregate and disaggregate resource rents) have an adverse effect on happiness or subjective well-being. This contrasts with recent studies using a global sample. We further contribute to this debate by examining the channels of resource curse or blessing along with income, unemployment, inflation, levels of human development, and governance. We show that oil rent enhances the positive marginal effects of income on happiness. We find no evidence of this conditional effect through other channels. Being rich in oil or natural resources is not necessarily a curse on happiness, but, if any, it is a blessing through income-generating well-being.

KEYWORDS:

• Resource blessing
• resource curse
• happiness
• life satisfaction
• subjective well-being
• net oil-exporting countries

JEL CLASSIFICATION:

• Q34
• I31
• C33

Acknowledgement

Earlier version of this paper was presented at the 24th Malaysian Finance Association International Conference 2022.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 We use ‘subjective life satisfaction’, ‘subjective well-being’, ‘well-being’, and ‘happiness’ interchangeably (Veenhoven, 2012; Helliwell et al., 2013).

2 Ayelazuno (2014) provided a critical analysis of the case of Ghana. He argued that despite having met many conditions (e.g. good quality of governance of oil wealth, and the political institutions) prescribed by the ‘orthodox oil curse’ approach to turn oil into a blessing, the oil rents have not trickled down to the average local people in Ghana as oil multinational corporations have created the enclaved resource sector that is detached from the rest of the economy and hence there is no real spillovers and job creation.

3 Oil rents (as a share of GDP), a measure of oil dependency – estimated based on production cost, price, and quantities of oil – capture more closely the economic rents accrued from oil that is usually used to support the oil-rich countries’ current consumption and investment, for example, expenditures on welfare supports and programs, and welfare-enhancing investments.

4 Toews (2015) modeled the links between resource boom (oil price), aspiration, and satisfaction with income. In the model, increasing oil prices would also increase people's expectations about future income (and aspirations). The model predicts that if actual realized income does not live up to this expectation (i.e. it is smaller than the expected future income) people's satisfaction would be reduced.

5 This finding implies that exclusive focus on net oil-exporting countries is crucial as their subjective wellbeing seems to response differently to oil price booms than the ones of oil-importing countries.

6 ${\mu }_i$ is assumed to be fixed in fixed effect model or random (and becomes part of $e_{it}={\mu }_i+{\epsilon }_{it}$) in random effect model, depending on the outcome of the diagnostic tests, i.e. Hausman specification test.

7 Classification follows the U.S. Energy Information Administration (EIA) and the World Factbook.

8 Other net oil-exporting countries, e.g. Algeria, Brunei, Iraq, Oman, Norway, among others, are excluded due to data unavailability on LS.

9 Please imagine a ladder with steps numbered from 0 at the bottom to 10 at the top. The top of the ladder represents the best possible life for you, and the bottom represents the worst possible life for you. If the top step is 10 and the bottom step is 0, on which step of the ladder do you feel you personally stand at the present time? (See Helliwell et al., 2013).

10 These variables enter the main Equation (1) at a later stage to robust check the main results reported in Table 1. The outcomes of this extensive robust checks are reported in Table 3. The summary statistics for these additional covariates are reported in Appendix A.

11 Thanks to an anonymous referee for this suggestion.

12 Recent literature has raised concern about the presence of cross-sectional dependency in panel data, especially the large T and large N panel, that is due to omitted common and spatial effects, among other causes (Chudik & Pesaran, 2015a). Although Pesaran CD test (reported in Table 2) shows no evidence of the presence of cross-sectional dependency in the residuals, we also alternatively attempted to robust check our finding by estimating Equation (1) using Chudik and Pesaran’s (2015b) dynamic common correlated-effects estimator that accounts for cross-sectional dependency. Unfortunately, this estimator is for a large T and large N panel data, and, when it is applied to our small T ( = 13) and small N ( = 31) panel data, it generates a much larger parameter than the degree of freedom available in our panel dataset (see Thombs, 2022). Thanks to an anonymous referee for directing us to this estimator.

13 System-GMM estimation was conducted using Roodman’s (2009) strategies (collapsing the instrument matrix and using only one lag as instruments) to minimize instrument proliferation (see Slesman et al., 2015; Slesman, 2022). It should be noted that System-GMM is suitable for the large sample with a significantly large cross-sectional dimension N (and usually employed in dynamic panel model settings). Since our N = 31 is very small, System-GMM is unsuitable for this study. For example, in this robust check of the baseline model, we observed that, although we use both collapsing the instrument matrix and using a minimum one lag of all available lags as instruments in the estimation (as recommended by Roodman (2009) to deal with instrument proliferation), the number of instruments (J = 34) is easily larger than N. This may pose challenges for statistical inference, see further detail in Roodman (2009).

14 We also re-estimate Table 3 using global sample (reported in Appendix E) and found the findings remain intact.

15 A time plot (not reported but available upon request) of the number of giant oil fields (discovered each year between 2006 and 2019 – using an extended Horn's dataset provided by Cust et al. (2021)) and the annual group mean of LS – also further support our main finding as both variables show no discernable relationship. Thanks to an anonymous referee for suggesting this.

16 Equation (1.1) gives the marginal effect of LRGDPC on LS as $\mathrm{\partial }L{S}_{it}/\mathrm{\partial }LRGDP{C}_{i,t-1}={\beta }_7+{\alpha }_{1}Oilren{t}_{i,t-1}.$ Furthermore, we can also compute the threshold level of Oilrent as $Oilren{t}_{i,t-1}\ge -{\beta }_7/{\alpha }_1$.

17 $\mathrm{\partial }L{S}_{it}/\mathrm{\partial }LRGDP{C}_{i,t-1}=-0.0278+0.0028Oilren{t}_{i,t-1}$which give $Oilren{t}_{i,t-1}\ge 9.93$.