Fetching primary parquet sources and computing exhibits.
structural change and the resource curse
Which resource-exporting economies show Dutch-disease symptoms?
Corden & Neary (1982, Economic Journal92(368)) formalised the mechanism: a resource-export boom appreciates the real exchange rate and crowds tradable manufacturing out through factor reallocation. Sachs & Warner (1997, Harvard CID; published in revised form 2001 in European Economic Review 45(4-6): 827-838) documented a negative cross-country growth correlation with natural-resource intensity. van der Ploeg (2011, Journal of Economic Literature 49(2): 366-420) surveyed three decades of follow-up evidence and flagged institutional quality as the key moderator. This page tests the textbook prediction on 2000-2024 BACI trade data: across 154 countries with at least one billion USD of exports in 2024, the change in resource-export share has a Pearson correlation of -0.35 with the change in manufacturing-export share, and the OLS slope is -0.19 (one percentage point more resources, 0.19 pp less manufacturing on average when slope is negative).
window2000-2024
resource HS225, 26, 27, 71
mfg HS284, 85, 86, 87, 88, 89, 90
sample154 countries
corr(Δres, Δmfg)-0.35
OLS slope-0.19
Setup: two baskets, two shares
For each country×year cell we compute the resource-export share as the value of HS2 chapters 25-27 plus 71 (salt, ores, mineral fuels, precious stones and metals) divided by total merchandise exports, and the manufacturing-export shareas HS2 chapters 84-90 (machinery, electricals, vehicles, aircraft, ships, instruments) divided by the same denominator. BACI values are stored in thousands of USD; we multiply by 1,000 for absolute amounts only, not for shares. This two-basket cut follows the empirical partition used by Harding & Venables (2016, IMF Economic Review 64(2): 268-302) in their resource-windfall test, and matches the exportable tradablesdistinction in Corden & Neary (1982).
Top-10 resource exporters: shares through time
The textbook Dutch-disease trajectory is a rising resource share paired with a falling manufacturing share in the same country. The next two panels plot the two shares for the ten countries with the largest resource-export value in 2024, conditional on resources making up at least 30% of their export basket. Commodity-price cycles (oil 2008, 2014, 2022) are visible as level shifts in the resource-share line.
The leading resource exporter by 2024 value is Australia (AUS) with a resource share of 72.3% and a manufacturing share of 3.9%. Series that trend up over the window are the Dutch-disease candidates on the export side; pair them with Figure 1b to see the manufacturing response.
Manufacturing-export share, same countries, 2000-2024
The Dutch-disease scatter
For every country with at least one billion USD in 2024 exports, we compute the change in resource share and the change in manufacturing share between 2000 and 2024. Under the Corden-Neary prediction, points should cluster along a downward-sloping line in (Δres, Δmfg)-space: countries that got more resource-heavy should have shed manufacturing share. The pooled OLS slope is -0.19 and the Pearson correlation is -0.35 on 154 countries, significantly negative and consistent with the prediction.
Figure 2
Change in resource share vs change in manufacturing share, 2000-2024
Resource-rich but diversified: Norway, Australia, Canada
The 'curse' is not destiny. Mehlum, Moene & Torvik (2006, Economic Journal 116(508): 1-20) showed that institutional quality flips the sign of the resource-growth relationship, and Frankel (2010, NBER WP 15836) documented several countries that used fiscal rules and sovereign-wealth vehicles to avoid the classical Dutch-disease path. Norway's Government Pension Fund Global (the former Petroleum Fund, 1990 Act) is the textbook case; Australia and Canada relied on institutional depth and federated fiscal adjustment. Their trajectories are plotted alongside the resource share to show how a commodity boom can coexist with a non-collapsing industrial base.
Figure 3a
Resource share, diversified benchmark (NOR, AUS, CAN), 2000-2024
The last panel ranks the current resource-rich set (resource share ≥ 30% in 2024) by a simple Dutch-disease-risk score: the change in resource share minus the change in manufacturing share over 2000-2024. A large positive score means the country has become materially more resource-concentrated while its manufacturing footprint has shrunk relative to its total export basket, the combination Corden & Neary warned about. This is a descriptive flag, not a causal claim; the Sachs-Warner 1997 critique, and later revisions such as Lederman & Maloney (2008, Economía 9(1): 1-39), stress that resource abundance and resource dependence are not the same thing, and institutions mediate the effect.
Figure 4
Dutch-disease risk ranking: (Δres − Δmfg) over 2000-2024, resource-rich countries
#
Country
Res share 2024
Δ res
Mfg share 2024
Δ mfg
Risk score
1
SLE Sierra Leone
79.5%
+67.6%
2.9%
-48.0%
+115.66pp
2
TCD Chad
91.4%
+91.1%
0.1%
-6.4%
+97.54pp
3
BFA Burkina Faso
88.6%
+84.9%
0.9%
-10.5%
+95.38pp
4
LBR Liberia
60.5%
+42.2%
The cross-section in levels: a 2-way quantile read
Figure 2 plotted changes over 2000-2024; Figure 5 plots levels in 2024. Each dot is a country placed by its resource share (x-axis) and manufacturing share (y-axis), with dot area proportional to total export value. On top we overlay two sets of conditional medians: blue dots RQ1-RQ5 mark the median of manufacturing share within each resource-share quintile (a conditional median of mfg given res), and red dots MQ1-MQ5 mark the median of resource share within each manufacturing-share quintile (a conditional median of res given mfg). If the Corden-Neary prediction holds in levels, both sets of medians should trace a downward-sloping curve. This is the non-parametric 2-way quantile version of what a quantile-regression estimator (Koenker & Bassett 1978, Econometrica 46(1): 33-50) would identify under smoother functional-form assumptions.
Figure 5
Resource share vs manufacturing share, 2024, with conditional quintile medians
Causal identification: IV using world oil-price shocks
Figures 1-5 are descriptive: they tell us the observed correlation between resource and manufacturing shares but not the Corden-Neary effectof a resource boom on manufacturing. Harding & Venables (2016) used giant-oil-discovery timing as an IV; Caselli & Michaels (2013) used municipal oil rents in Brazil. We cannot replicate either in this workbench at the country level. What we can do is a Bartik-style instrument (Goldsmith-Pinkham, Sorkin & Swift 2020, AER 110(8)): multiply the country's year-2000 HS 27 share (a pre-determined measure of oil-export exposure) by the log change in the World Bank Pink Sheet Brent price from 2000 to 2022. The exogeneity argument is Hamilton (2009, BPEA): world oil prices are close to exogenous for any single non-giant producer once we hold fixed its initial specialisation. Figure 6a is the first stage (instrument vs Δres_share), Figure 6b is the reduced form (instrument vs Δmfg_share). Both windows run 2000-2022 to use the strongest world price variation (2000 to 2022 peak) before the 2023-24 retreat which weakens the instrument.
Figure 6a
IV first stage: Δ resource share vs Bartik oil-shock instrument, 2000-2022
Figure 6b
IV reduced form: Δ manufacturing share vs Bartik oil-shock instrument, 2000-2022
Institutional moderation: rule of law and the resource-mfg trade-off
Mehlum, Moene and Torvik (2006, Economic Journal 116(508): 1-20) argue that the resource curse is conditional on institutions: grabber-friendly institutions turn rents into a curse, producer-friendly institutions reverse the sign. Robinson, Torvik and Verdier (2006, Journal of Development Economics 79(2): 447-468) make the same point in a political-economy complementary frame. The descriptive scatter in Figure 2 and the IV in Figure 6 ignore this moderation. Figure 7 adds it. We pull the Worldwide Governance Indicators rule-of-law score (RL.EST, Kaufmann, Kraay and Mastruzzi 2010, World Bank Policy Research WP 5430) for the earliest available year close to 2000as a pre-determined institutional proxy, sort the resource-rich subsample (resources ≥ 30% in 2024) into RL tertiles, and report the mean change in manufacturing share over 2000-2024 per tertile. If the Mehlum-Moene-Torvik conditional- curse story holds, weak-RL countries should show the steepest mfg-share decline; strong-RL countries should be flat or positive.
Figure 7
Mean Δ manufacturing share by WGI rule-of-law tertile, resource-rich exporters, 2000-2024
Within the resource-rich subsample (n = 62), the weak-RL tertile shows a mean Δmfg of -4.8%; the strong-RL tertile shows -2.5%. The gap +2.3% percentage points is the Mehlum-Moene-Torvik conditional-curse moderator at the country aggregate. A negative gap (strong-RL countries doing worse) would reject the institutions- save-you story; a positive gap is consistent with it. This is a single-cut moderator on a small sample and should be read alongside Figure 6 (causal IV) and the per-country trajectories in Figures 1 and 3 rather than on its own. WGI is survey-based and has its own measurement controversies (Apaza 2009, PS: Political Science & Politics42(1): 139-143; Langbein & Knack 2010, Journal of Development Studies 46(2): 350-370).
Sources: CEPII BACI 202501 (retrieved 2026-04-28) country_year_product for resource and manufacturing baskets; World Bank Worldwide Governance Indicators (Kaufmann, Kraay and Mastruzzi 2010, WPS 5430), indicator RL.EST (rule of law, estimate). Resource-rich threshold and HS2 baskets per the page setup. Authors calcs.
Caveats
Export shares are not output shares.Corden & Neary (1982) model factor reallocation across sectors in GDP; export shares are an observable proxy and move with global prices, not just domestic structure. A spike in the oil price raises the resource share mechanically even at fixed production quantities.
Two-HS2 baskets are coarse. HS 71 bundles pearls, precious stones, and precious metals; countries with large refining sectors look more 'resource' than their raw endowment warrants. HS 84-90 excludes textiles, basic metals, chemicals, and food processing, which matter for some resource-poor manufacturing countries. Narrower partitions change the scatter slope but not the sign in our sample.
Descriptive, not causal. The correlation in Figure 2 is a conditional association. Identification of a Dutch-disease effectin the Corden-Neary sense requires an instrument for the resource boom (Harding & Venables 2016 use a giant-oil-discovery instrument; Caselli & Michaels 2013, AEJ: Applied Economics 5(1): 208-238, use Brazilian municipal oil rents).
Policy read and open questions
Institutions as the moderator.Mehlum, Moene & Torvik (2006) showed that grabber-friendly institutions turn resource rents into the curse; producer- friendly institutions (rule of law, secure property rights) reverse the sign. Our risk ranking is institution-blind; cross-referencing with Worldwide Governance Indicators (loaded as wgi.parquet) would let us split the at-risk list into institutionally fragile and institutionally robust resource exporters. This is a natural next step.
Real exchange rate as the mechanism. The Corden-Neary (1982) transmission channel is RER appreciation. Frankel (2012, HKS Faculty Research WP RWP12-014) and van der Ploeg (2011) both emphasise the RER as the mediator. The workbench holds WDI REER for a subset of countries; adding it as a covariate in Figure 2 would test whether the Δmfg response to Δres is mediated by RER appreciation, as the theory predicts.
Sovereign-wealth discipline.Norway, Chile, and Botswana use rules-based fiscal frameworks to sterilize commodity windfalls. Harding & Venables (2016) argue that the non-resource current-account response is what separates cursed and uncursed economies, not the resource share itself. A non-resource current-account series from IMF BOP would sharpen the diagnostic.
Identification.Our scatter is descriptive. Caselli & Michaels (2013) and Harding & Venables (2016) used giant-oil-discovery timing as an IV. Without an instrument, the (Δres, Δmfg) correlation conflates structural-shift effects with common global commodity-price shocks.
References
Caselli, F., & Michaels, G. (2013). 'Do Oil Windfalls Improve Living Standards? Evidence from Brazil.' American Economic Journal: Applied Economics 5(1): 208-238.
Corden, W. M., & Neary, J. P. (1982). 'Booming Sector and De-Industrialisation in a Small Open Economy.' Economic Journal 92(368): 825-848.
Frankel, J. A. (2010). 'The Natural Resource Curse: A Survey.' NBER Working Paper 15836.
Frankel, J. A. (2012). 'The Natural Resource Curse: A Survey of Diagnoses and Some Prescriptions.' In R. Arezki, C. Patillo, M. Quintyn & M. Zhu, eds., Commodity Price Volatility and Inclusive Growth in Low-Income Countries, IMF; also HKS Faculty Research Working Paper RWP12-014.
Harding, T., & Venables, A. J. (2016). 'The Implications of Natural Resource Exports for Nonresource Trade.' IMF Economic Review 64(2): 268-302.
Koenker, R., & Bassett, G. (1978). 'Regression Quantiles.' Econometrica 46(1): 33-50.
Lederman, D., & Maloney, W. F. (2008). 'In Search of the Missing Resource Curse.' Economía 9(1): 1-39.
Mehlum, H., Moene, K., & Torvik, R. (2006). 'Institutions and the Resource Curse.' Economic Journal 116(508): 1-20.
Sachs, J. D., & Warner, A. M. (1997). 'Natural Resource Abundance and Economic Growth.' Harvard Center for International Development Working Paper; originally circulated as NBER Working Paper 5398 (1995); revised version published 2001 in European Economic Review 45(4-6): 827-838.
van der Ploeg, F. (2011). 'Natural Resources: Curse or Blessing?' Journal of Economic Literature 49(2): 366-420.
Manufacturing share in 2024 for the set ranges from 0.1% to 25.3%. Flat-near-zero lines mean the country never had a manufacturing base to lose; downward slopes in countries that started above 10% are the visual fingerprint of classical Dutch disease.
The most extreme Dutch-disease trajectory in the resource-rich subset is Sierra Leone (SLE): resource share +67.6%, manufacturing share -48.0%. Orange dots are resource-rich countries (resource share ≥ 30% in 2024); blue dots are the rest. Dot area is proportional to total 2024 export value.
Source: CEPII BACI 202501 (retrieved 2026-04-28). Each dot is a country; x = 2000-2024 change in resource share (pp), y = same-window change in manufacturing share (pp). Authors calcs. Slope -0.19, correlation -0.35, n = 154.
All three sit in the 15.4% to 79.2% resource-share band over the window. Norway moves most with the oil cycle; Canada and Australia have more stable baskets because metals and agricultural commodities trade on different cycles.
Source: CEPII BACI 202501 (retrieved 2026-04-28). Benchmark set selected ex ante from the literature on resource-curse avoidance (Mehlum, Moene & Torvik 2006; Frankel 2010).
Canada's manufacturing share has trended down from the mid-40s (2000) to the mid-20s (2024), Australia from low double digits to below 5%, Norway broadly around 10% but with a sharp dip in 2022. Contrast with Figure 1b: several top-10 resource exporters that started above 10% have slid below 5%.
Top of the list is Sierra Leone (SLE) with a score of +115.7pp. The table shows 12 countries ranked by this score, alongside their current resource share, current manufacturing share, and window-over-window change.
Sample of 154 countries with at least USD 1B of exports in 2024. Spearman rank correlation between resource share and manufacturing share in levels is -0.60 (n = 154), negative as the Corden-Neary prediction requires but far from one: many resource-rich economies have never built significant manufacturing, and many manufacturing hubs have no resource base, so the two shares are negatively but not tightly related. The blue quintile markers (RQ1-RQ5, binned by resource share) step down from high-mfg/low-res at RQ1 to low-mfg/high-res at RQ5, which is the Dutch-disease signature in levels. The red markers (MQ1-MQ5, binned by mfg share) trace the same relationship from the orthogonal direction. The two median curves approximately agree on shape, which is the non-parametric version of the OLS slope in Figure 2 being robust to the conditioning direction.
Source: CEPII BACI 202501 (retrieved 2026-04-28) country_year_product, 2024. Resource = HS2 25+26+27+71. Manufacturing = HS2 84-90. Quintile medians computed non-parametrically: partition ranked sample into five equal bins on the binning axis, take the median of the partner axis within each bin. Koenker & Bassett (1978) is the parametric quantile-regression reference.
The Brent price log-change over 2000-2022 is 1.26 (nominal USD: Brent rose from 28 to 100USD/bbl, source World Bank Pink Sheet annual). The first-stage slope π of Δres_share on the Bartik instrument is -0.23 with R² = 0.15 on n = 160 countries. A strong positive slope means the instrument actually picks up the resource-share shift: countries with higher year-2000 oil exposure moved more strongly into resources by 2022, as the mechanism requires. F-statistic on the first stage is not shown (requires heteroskedasticity-robust variance computation); by Stock-Yogo (2005) convention, an F below 10 signals a weak instrument.
Source: CEPII BACI 202501 (retrieved 2026-04-28) country_year_product for 2000 and 2022 shares (HS 27 share in 2000 as exposure, HS 25+26+27+71 as total resource basket). World Bank Pink Sheet annual (Crude oil, Brent, USD/bbl). Bartik instrument construction per Goldsmith-Pinkham, Sorkin & Swift (2020) AER 110(8). Dot area proportional to 2022 total export value.
The reduced-form slope θ is 0.02 with R² = 0.01. The 2SLS Dutch-disease elasticity implied by dividing reduced form by first stage is βIV = -0.10: a one percentage-point increase in the resource share, instrumented by the world oil shock weighted by year-2000 oil exposure, causes a 0.10 percentage-point decrease in the manufacturing share. The descriptive OLS slope in Figure 2 was -0.19; the IV point estimate is less negative (weaker) than the descriptive slope, consistent with the direction of bias expected if unobserved demand shocks (post-2008 global mfg slowdown, Covid) attenuate the true Dutch-disease channel in OLS. Caveats: this is a single-instrument Bartik with one pre-determined weight (oil share in 2000); Goldsmith-Pinkham et al. (2020) show the exogeneity assumption rests on the pre-determined share's orthogonality to the residual, defensible for the commodity-price mechanism but not for institutional or geographic confounders. No robust SE is reported here; a proper implementation would use heteroskedasticity- consistent and cluster-robust variance.
Source: CEPII BACI 202501 (retrieved 2026-04-28) country_year_product; World Bank Pink Sheet annual (Brent crude). Instrument construction as in Figure 6a. OLS closed-form slope/intercept/R²; 2SLS point estimate = θ/π. Authors calcs.