## 2050.  What are the macroeconomic consequences for fossil fuel producing countries?  We

## Source details

**Canonical URL:** [2050.  What are the macroeconomic consequences for fossil fuel producing countries?  We](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023097-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023097-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023097-print-pdf.pdf.json)

---

### Introduction and research question
- Reaching net zero carbon emissions by 2050 requires a 80 percent reduction in global fossil fuel extraction compared with 2021 levels, according to the International Energy Agency (2022).
- Research questions:
  - What are the macroeconomic impacts of large and persistent, exogenous declines in extraction activity?
  - Can maladies identified by the resource curse literature be reversed when the extractive sector shrinks?
- Scope and approach:
  - New data-set for 13 minerals (oil, gas, coal, metals) and 122 countries since 1950.
  - Identification of 35 episodes of persistent, exogenous declines in extraction.
  - Use of local projections to estimate effects on real output, external and domestic sectors.
  - Comparison with a stylized classical small-open-economy real model with and without full anticipation.

### Data, episode selection, and identification
- Dataset coverage:
  - 13 largest commodities in terms of worldwide production value for 122 countries from 1950 to 2020.
  - Country-level macroeconomic variables: real GDP in local currency, nominal GDP, investment, private and public consumption, exports and imports, the real exchange rate (relative to the US Dollar), the current account and employment; real value added by industry; measures of political stability and institutional quality.
- Episode selection criteria:
  - (a) The cumulative decline in the quantity of production is at least 20 percent from peak to trough.
  - (b) After reaching the trough, the production level has to remain below its pre-shock level for at least 5 years.
  - (c) The ADV is at least 0.05 percent.
- Episode counts:
  - Applying (a) and (b) isolates 154 episodes; adding (c) reduces that number to 70 macro-relevant episodes.
  - Focusing only on exogenous episodes yields 35 episodes of large and persistent declines.
- Exogeneity classification (narrative approach):
  - Six causes: (a) structural transitions or wars, (b) global recessions, (c) policy changes affecting all domestic industrial sectors, (d) policy changes affecting only the mining sector, (e) depletion, (f) undefined causes.
  - Episodes in categories (d) and (e) are classified as exogenous.
- ADV definition:
  - ADV_i = (1/T_i) * sum_{k=1}^{T_i} (ΔQ_{t+k,i} * P_{t+k,i}) / GDP_{t,i}

### Main empirical findings (aggregate effects)
- Typical episode magnitude and path:
  - Typical initial production shock: a 10 percent initial decline that accumulates to a 40 percent decline in the tenth year.
  - Median duration of exogenous declines: 7 years; mean duration: 8 years.
  - Median growth rate during the first ten years: -9 percent per year; thereafter median growth rate -3 percent.
- Real activity and demand:
  - Local projection estimates: such a decline leads to a persistent decline in real GDP of six percent after about six years.
  - Alternative IRF result: peak effect of seven percent after ten years (log-level).
  - Both private and public consumption and investment fall in line with the decline in GDP; private consumption declines with a four-year delay to -4% relative to pre-shock trend; public consumption falls with a four-year delay to -4.5% relative to pre-shock trend.
  - Investment responds with a delay and remains insignificantly different from zero over the majority of the ten-year horizon.
- External sector and exchange rate:
  - Real exports fall immediately, accumulating to a decline of 9% after ten years.
  - Real imports respond with a four-year lag and a peak decline of 3.5% (insignificant over the ten-year horizon).
  - Real exchange rate depreciates persistently by approximately 20 percent after four years.
  - Net exports fall in line with extraction; depreciation not sufficient to offset net export decline.
- Sectoral spillovers:
  - Mining value added: −2% at t=0 and −14.5% after ten years.
  - Manufacturing value added: −6% with a lag of three years.
  - Services & others value added: −5.5% with a lag of three years.
  - Depreciation does not stimulate reallocation toward manufacturing; negative spillovers outweigh potential depreciation benefits.
- Institutional effects:
  - A decline in extraction activity does not improve the quality of institutions even a decade after the shock.
  - Evidence consistent with hysteresis and asymmetric response: institutional damage is hard to reverse.
  - Starting with higher-quality institutions buffers negative economic effects.

### Anticipation and expectations
- Little anticipation effects for extraction declines overall, in contrast to extraction booms which are often anticipated.
- IMF Article IV Staff Report review:
  - In 29 decline episodes with Staff Reports: initial decline not anticipated in 24 episodes; fully anticipated in 4 episodes; commodity not mentioned in 1; 6 episodes had no Article IV Reports available.
- Country cases:
  - Chad: year before decline, projected production over 2005–09 projected to increase by 25 percent relative to pre-episode peak (2005); realized production was more than 50 percent below that projection. One year into the decline (2006), projected fall was 72 percent of the realized decline during 2006–10.
  - Gabon: year before decline, projected production over 1997–99 projected to increase by 5 percent relative to pre-episode peak (1997); by 1999 optimistic projection exceeded realized production by 15 percent. One year into decline (1998), production was projected to return to pre-episode peak.
- Modeling insight:
  - Empirical results match a small-open-economy real model when departing from full anticipation; modeling the decline as a sequence of perceived temporary shocks with slowly adjusting expectations replicates key features.

### Theoretical model and interpretation
- Model structure and calibration:
  - Two-sector small open economy with tradable (commodity/mining) and non-tradable sectors; CES consumption aggregator with home bias γ and substitutability η.
  - Calibration: β = 0.95; η = 6; initial steady state mining share set to 20 percent; B_{−1} = 0.
  - Exogenous output decline sequence calibrated to empirical estimates for the 35 episodes.
- Anticipated vs unanticipated declines (model results):
  - Permanent (anticipated) shock → trade surplus (savings increase to smooth consumption); consumption falls immediately and converges; output adjusts more gradually.
  - Sequence of transitory shocks → trade deficit (borrowing to smooth consumption); consumption declines more smoothly initially via borrowing while eventual GDP and welfare losses are larger; real exchange rate depreciates gradually and persistently.
  - Intuition: lack of anticipation delays adjustment and can worsen GDP and welfare even without additional frictions.

### Heterogeneity of effects across countries
- Mining sector size:
  - For mining shares at the 10th and 90th percentiles (2.5% and 58%):
    - Small mining sectors: decline of 3.7% in real GDP after ten years.
    - Large mining sectors: decline of 15.3% in real GDP after ten years.
  - Nonlinearity: impact of mining share is less than proportional — doubling the mining share increases GDP impact on average by about 40 percent.
- Income-level interaction:
  - For a 40% decline over ten years in a sector making up about 58% of aggregate value added:
    - Real GDP in low-income economies falls by 30% after ten years.
    - Real GDP in high-income countries falls by 10% after ten years.
  - Temporal pattern: initial impact stronger in high-income countries with large mining sectors, but by year 3 low-income countries fare increasingly worse.
  - Sample counts: 11 high and 24 low-income countries in the sample.
- Manufacturing sector size:
  - Mining-manufacturing ratio MR_i findings: larger manufacturing sector (lower MR_i) buffers extraction shocks.
  - Small MR_i (10th percentile): mining sector is 30% as large as manufacturing.
  - Large MR_i (90th percentile): mining sector is eight times as large as manufacturing.
  - Data limitation: manufacturing value added available for only 25 shock episodes.
- Institutions:
  - Persistent declines do not improve institutional quality.
  - Pre-shock institutional quality affects outcomes: five years after shock, GDP difference between high and low institutional quality about 5 percentage points; after ten years, democratic/high-quality country (indicator = 10) sees −2.5% GDP, autocratic/low-quality (indicator = −10) sees −10% GDP.

### Robustness, sensitivity, and complementary analyses
- Robustness to specifications, controls, and synthetic control methodology.
- Sub-sample results:
  - Non-oil shocks (15 metal episodes): 40% decline over ten years causes a real GDP slowdown between 5% and 6% after six years; impact levels off for commodities excluding oil.
  - Oil and gas shocks (20 episodes): negative impact more pronounced after 10 years.
- Leave-one-out sensitivity: results robust to excluding individual episodes.
- Anticipation test: anticipated declines show insignificantly smaller point estimates over 10 years than unanticipated ones (sample: 27 episodes — 24 unanticipated, 4 anticipated).
- Long horizon: after twenty years, real GDP still 4% lower; no reversed resource curve even at 20 years.
- Synthetic Control Method (SCM) median result: real GDP decreases by approximately 15 percent after six years relative to synthetic counterfactual (SCM trimmed to RMSPE below 75th percentile).
- Employment: employment falls slightly and is statistically significant between the six and eighth year horizon; SCM shows employment deviates with a lag of six years and returns by year 14.

### Quantitative accounting and notable statistics
- Identified exogenous episodes: 35.
- Initial candidate episodes and filtering: 154 (criteria a & b) → 70 macro-relevant (plus criterion c) → 35 exogenous.
- Typical episode path: 10 percent initial decline → 40 percent cumulative decline by year 10.
- Real GDP response: persistent decline of six percent after about six years; alternative IRF peak of seven percent after ten years.
- Real exchange rate: eventual depreciation of about 20 percent.
- ADV selection threshold: at least 0.05 percent.
- Largest revenue declines on average occurred in oil, copper and gold; declines in crude oil production come with large revenue losses of almost 20 percent of GDP on average.
- Dataset: 13 commodities, 122 countries, 1950–2020.
- Appendix A sample figures (selected exact entries from Table 2):
  - Crude Oil: # of Episodes 21; Decline of Extraction (% of Peak) -47; Decline of Extraction Quant. (Mil. USD) -82906; Duration (Years) -19.4
  - Copper: # of Episodes 17; Decline of Extraction (% of Peak) -80; Decline of Extraction Quant. (Mil. USD) -2269; Duration (Years) -3.8
  - Gold: # of Episodes 9; Decline of Extraction (% of Peak) -50; Decline of Extraction Quant. (Mil. USD) -3795; Duration (Years) -1.7
  - Total initial candidate episodes: 154

### Policy implications and recommendations
- Transition risks:
  - The clean energy transition may pose significant macroeconomic challenges for countries reliant on fossil fuel exports, particularly low- and middle-income countries.
- Policy priorities:
  - Improve the quality of institutions to reduce legacy effects and better absorb declines in extraction.
  - Reduce climate policy uncertainty which may lead to a lack of anticipation and a more expensive adjustment process.
- Broader contribution:
  - The paper provides empirical benchmarks for the economic impacts of the clean energy transition on fossil fuel producing countries and complements literature on stranded assets and fiscal implications.

*Source: wpiea2023097-print-pdf*

### 2050.  What are the macroeconomic consequences for fossil fuel producing countries?  We

### 2050.  What are the macroeconomic consequences for fossil fuel producing countries?  We

### Introduction and research question
- Reaching net zero carbon emissions by 2050 requires a 80 percent reduction in global fossil fuel extraction compared with 2021 levels, according to the International Energy Agency (2022).
- The paper asks: What are the macroeconomic impacts of large and persistent, exogenous declines in extraction activity? Can the maladies identified by the resource curse literature be reversed when the extractive sector shrinks?
- Scope and approach:
  - New data-set for 13 minerals (oil, gas, coal, metals) and 122 countries since 1950.
  - Identification of 35 episodes of persistent, exogenous declines in extraction.
  - Use of local projections to estimate effects on real output, external and domestic sectors.
  - Comparison with a stylized classical small-open-economy real model with and without full anticipation.

### Data, episode selection, and identification
- Dataset coverage:
  - 13 largest commodities in terms of worldwide production value for 122 countries from 1950 to 2020.
  - Country-level macroeconomic variables include real GDP in local currency, nominal GDP, investment, private and public consumption, exports and imports, the real exchange rate (relative to the US Dollar), the current account and employment; real value added by industry; measures of political stability and institutional quality.
- Episode selection criteria (to identify large, persistent, macro-relevant declines):
  - (a) The cumulative decline in the quantity of production is at least 20 percent from peak to trough.
  - (b) After reaching the trough, the production level has to remain below its pre-shock level for at least 5 years.
  - (c) The ADV is at least 0.05 percent.
- Episode counts:
  - Applying (a) and (b) isolates 154 episodes; adding (c) reduces that number to 70 macro-relevant episodes.
  - Focusing only on exogenous episodes (policy changes specific to the mining sector or depletion) yields 35 episodes of large and persistent declines.
- Exogeneity classification (narrative approach):
  - Six causes: (a) structural transitions or wars, (b) global recessions, (c) policy changes affecting all domestic industrial sectors, (d) policy changes affecting only the mining sector, (e) depletion, (f) undefined causes.
  - Episodes in categories (d) and (e) are classified as exogenous.
- ADV definition:
  - ADV_i = (1/T_i) * sum_{k=1}^{T_i} (ΔQ_{t+k,i} * P_{t+k,i}) / GDP_{t,i}
  - ADV is closely related to NPV measures used in resource windfall work but NPV is not computed because duration is ex-ante uncertain and discount factors vary.

### Main empirical findings (aggregate effects)
- Typical episode magnitude:
  - A 10 percent initial decline that accumulates to a 40 percent decline in the tenth year.
- Output and demand:
  - Local projection estimates: such a decline leads to a persistent decline in real GDP of six percent after about six years.
  - Both private and public consumption as well as investment fall in line with the decline in GDP.
  - Consumption and the exchange rate have a delayed reaction, inconsistent with full anticipation of the persistent fall in extraction and related revenues.
- External sector and exchange rate:
  - The real exchange rate depreciates, eventually by about 20 percent, but not enough to offset the decline in net exports.
  - Net exports fall in line with extraction.
- Sectoral spillovers:
  - Significant negative spillover effects onto both the manufacturing and services sectors.
  - The depreciation of the real exchange rate does not stimulate a reallocation towards other tradable sectors such as manufacturing; negative spillovers to manufacturing and services more than offset potential benefits of depreciation.
- Institutional effects:
  - A decline in extraction activity does not improve the quality of institutions even a decade after the shock.
  - Evidence consistent with hysteresis and asymmetric response: once institutions are damaged it is difficult to improve them.
  - Starting with higher-quality institutions can help buffer negative economic effects of persistent declines in extraction activity.

### Heterogeneity of effects
- Effects depend on per capita income and the size of the mining sector relative to the economy.
- A large and persistent decline in extraction implies significantly larger declines in economic activity for low-income countries than for high-income countries.
- Initial manufacturing share matters:
  - Economies with a bigger initial manufacturing share fare better, suggesting non-convexities in the tradable sector (i.e., sunk costs) favoring the intensive margin over the extensive margin.

### Anticipation and expectations
- Little anticipation effects for extraction declines are found overall, in contrast to extraction booms which are often anticipated.
- IMF Article IV Staff Report review:
  - Evidence that the initial decline in production was not anticipated in 24 decline episodes out of 29 decline episodes with Article IV Staff Reports.
  - The initial decline was fully anticipated in only 4 episodes; for 1 episode the commodity production was not mentioned; 6 episodes had no Article IV Staff Reports available.
- Country-specific evidence:
  - Chad: the year before the decline started, projected production over the next five years (2005-09) was projected to increase by 25 percent relative to the pre-episode peak (2005); this optimistic projection was more than 50 percent above the actual realized production during the period. One year into the decline (2006), projected fall was 72 percent of the realized decline during 2006-10.
  - Gabon: the year before the decline started, projected production over the next three years (1997-99) was projected to increase by 5 percent relative to the pre-episode peak (1997); by 1999 this initial optimistic projection exceeded lower realized production by 15 percent. One year into the decline (1998), production was projected to return to the level of the pre-episode peak (i.e., seen as fully transitory).
- Modeling insight:
  - A stylized classical small-open-economy real model matches the empirical results qualitatively when departing from full anticipation, particularly when the extraction decline is perceived as a series of temporary negative shocks with slowly adjusting expectations.

### Magnitudes and notable statistics
- Identified exogenous episodes: 35.
- Initial candidate episodes and filtering: 154 (criteria a & b) → 70 macro-relevant (plus criterion c) → 35 exogenous.
- Typical episode path: 10 percent initial decline → 40 percent cumulative decline by year 10.
- Real GDP response: persistent decline of six percent after about six years.
- Real exchange rate: eventual depreciation of about 20 percent.
- ADV selection threshold: at least 0.05 percent.
- Largest revenue declines on average occurred in oil, copper and gold production; declines in crude oil production come with large revenue losses of almost 20 percent of GDP on average.
- Dataset: 13 commodities, 122 countries, 1950–2020.

### Robustness and sensitivity
- Results are robust to:
  - Different specifications, inclusion of various control variables, and use of the synthetic control methodology.
  - Sensitivity checks to sub-samples and anticipation effects.

### Policy implications and recommendations
- The transition towards clean energy might be a challenge, especially for countries that rely on fossil fuel exports and are low to middle income.
- Policy priorities:
  - Improve the quality of institutions to reduce legacy effects and better absorb declines in extraction.
  - Reduce climate policy uncertainty which may lead to a lack of anticipation, risking a more expensive adjustment process.
- Broader contribution:
  - The paper provides empirical benchmarks for the economic impacts of the clean energy transition on fossil fuel producing countries and complements literature on stranded assets and fiscal implications.

*Source: wpiea2023097-print-pdf*

### 3.1    Mining Activity

### 3.1–6 The Impact of Extraction Declines: Mining Activity, Macroeconomic Effects, and Model Interpretation

### Mining activity: episodes, magnitudes, and temporal patterns
- Time convention and normalization:
  - Timet=0 corresponds to the peak of production; timet=1 is the first year with a negative production growth rate.
  - Log-levels of production are normalized relative to the median at time zero.
  - The analysis presents the 25th and the 75th pointwise percentiles for every horizon.
  - The time window ranges from 10 years before and 15 years after the shock.
- Sample and distributional facts:
  - A summary of all 35 declines is provided in Appendix B.
  - The exogenous declines have a median duration of 7 years.
  - The mean duration is 8 years.
  - The declines are skewed towards the lower end; the gap between the 25th percentile and the median increases after the peak with a skewness factor of 2 after year zero (presence of a few very large negative shocks).
- Growth rates and path:
  - The path of extraction declines follows a convex function.
  - Median growth rate during the first ten years: -9 percent per year.
  - Afterwards production declines at a median growth rate of -3 percent.
  - Some further declines occur up to the fourteenth year.

### Economic activity: aggregate and sectoral responses (empirical evidence)
- Aggregate GDP responses:
  - The log-level of real GDP declines persistently; the median does not return to its pre-shock trend even after 15 years.
  - Growth rate of GDP is persistently depressed for 15 years.
  - During a typical extraction decline episode, real GDP falls relative to pre-shock trend, with a peak effect of seven percent after ten years.
- Typical extraction shock used for identification:
  - The typical initial production shock is a ten percent decline (matches median first-year decline).
  - Persistence estimate: the initial 10 percent decline accumulates to a 40 percent decline after ten years.
- Domestic components:
  - Real private consumption declines with a four-year delay and falls to -4% relative to pre-shock trend.
  - Real public consumption declines with a four-year delay and falls to -4.5% relative to pre-shock trend.
  - Real investment responds with a delay and remains insignificantly different from zero over the majority of the ten-year horizon.
  - Interpretation: lagged consumption responses suggest agents perceived shocks as transitory and smoothed consumption, underestimating shock persistence.
- External components:
  - Real exports fall immediately, accumulating to a decline of 9% after ten years.
  - Real imports respond with a four-year lag and a peak decline of 3.5% (insignificant over the ten-year horizon).
  - Trade balance deteriorates during declines in extraction activity.
  - Real exchange rate depreciates persistently by approximately 20 percent after four years and remains at the lower level thereafter (response not precisely estimated).
  - Mechanisms: reduced export volumes from mining decline; lagged consumption sustains imports; borrowing to smooth consumption induces a trade deficit that temporarily mitigates consumption decline.
- Industry decomposition:
  - Mining value added: drops by 2% at timet=0 and falls by 14.5% after ten years (convex decline).
  - Manufacturing value added: falls by 6% with a lag of three years.
  - Services & others value added: falls by 5.5% with a lag of three years.
  - Lags in manufacturing and services consistent with domestic-demand driven channels and spillovers from mining.

### Theoretical framework: model setup and calibration
- Model structure:
  - Two-sector small open economy with tradable (commodity/mining) and non-tradable sectors.
  - Representative agents maximize lifetime utility U = sum_{t=0}^{∞} β^{t} (C_{t}^{σ} − 1)/σ with 0<β<1 and CES aggregation of tradable and non-tradable goods.
  - Consumption CES aggregator: C_{t} = [(1−γ)(c_{N,t})^{(η−1)/η} + γ(c_{T,t})^{(η−1)/η}]^{η/(η−1)} with home bias γ and substitutability η.
  - Tradable output y_{T,t} is exogenous; non-tradable output is constant.
  - Terms of trade exogenous and assumed constant and unitary for simplicity.
  - Flow budget constraint: B_{t} = (1+r_{t})B_{t−1} + y_{T,t} + p_{t} y_{N,t} − c_{T,t} − p_{t} c_{N,t}.
  - Market clearing implies c_{N,t} = y_{N,t}.
  - Small open economy interest rate r_{t} = r^{*} = β^{−1} − 1.
  - GDP definition: Y_{t} = p_{t} y_{N,t} + y_{T,t}.
  - Trade balance: TB_{t} = Y_{t} − p c_{N,t} − c_{T,t} = y_{T,t} − c_{T,t}.
- Calibration and empirical input:
  - Discount rate β = 0.95.
  - Intratemporal elasticity of substitution η = 6.
  - Initial steady state net foreign asset position B_{−1} = 0.
  - Home bias γ calibrated so extraction sector share in total output matches 20 percent in the initial steady state.
  - Exogenous output decline sequence for the extraction sector uses empirical estimates for the 35 extraction episodes.

### Theoretical model results: anticipated vs unanticipated declines
- Two scenarios compared:
  - "Permanent shock": an anticipated persistent decline in mining (tradable) output known as of t=0.
  - "Sequence of transitory shocks": the same empirical decline implemented as a sequence of annual transitory unanticipated declines where, in the model, each period k’s decline is expected to reverse in subsequent periods.
- Key comparative outcomes:
  - Trade balance:
    - Permanent (anticipated) shock → trade surplus (savings increase to smooth consumption).
    - Sequence of transitory shocks → trade deficit (borrowing abroad to smooth consumption).
  - Output vs consumption adjustment:
    - Permanent shock → output adjusts more gradually than consumption (consumption falls immediately and converges).
    - Transitory shocks → consumption declines more smoothly initially (via borrowing) while both consumption and output eventually decline more than in the anticipated case.
  - Exchange rate dynamics:
    - Under transitory shocks, real exchange rate depreciation is gradual and persistent; depreciation under anticipated shock occurs differently (price of non-tradables declines relative to tradables as country becomes poorer).
- Intuition:
  - Lack of anticipation delays economic adjustment and can lead to lower GDP and welfare even without other frictions.
  - The model’s differences stem from expectations: anticipated future income declines induce saving and immediate consumption adjustment; unanticipated persistent declines elicit borrowing and delayed adjustment, worsening eventual outcomes.

### Empirical framework and identification
- Episode definition and first stage:
  - For each commodity- and country-specific time series i, binary dummy δ_{it} = 1 if the year is part of a decline episode and 0 otherwise for the set Ω_{X} of exogenous episodes.
  - First-stage regression: ∆q_{it} = α_{i} + β_{i} δ_{it} + u_{it}, where ∆q is percentage change in extraction.
  - Fitted values ∆ˆq_{it} from the first stage are used to create a shock series removing production noise.
- Local projections baseline (Jordà (2005); Stock and Watson (2018)):
  - y_{t+h,i} − y_{t−1,i} = α + β^{h} ∆ˆq_{t,i} + sum_{j=1}^{p} Γ^{h}_{j} y_{t−j,i} + sum_{j=1}^{p} Π^{h}_{j} ∆ˆq_{t−j,i} + ψ_{n} + φ_{t} + u_{t+h,i}.
  - p = 3 lags of dependent variable and shock series included.
  - Country fixed effects ψ_{n} and time fixed effects φ_{t} included.
  - Standard errors clustered around every shock.
- Persistence estimation for typical shock:
  - Specification: ∆ˆq_{t+h,i} = α_{A} + β_{A,h} ∆ˆq_{t,i} + e_{t+h,i} for h = 0, . . . , 10.
  - Result: initial 10 percent decline accumulates to a 40 percent decline after ten years (used as normalized shock in IRFs).

### Baseline quantitative results and accounting
- Normalized shock used for IRFs: accumulated exogenous extraction decline of 40 percent over ten years (based on initial 10 percent decline and estimated persistence).
- Aggregate and sector accounting comparison:
  - GDP accounting identity used: Y_{t} = C_{t} + G_{t} + I_{t} + (X_{t} − M_{t}).
  - Weighted sum of sectoral point estimates: 4.1%*0.54 + 5.3%*0.16 + 5.8%*0.2 + (8.7%−3.5%)*0.1 = 4.8%.
  - Total GDP response from IRFs: 6.9%.
  - The difference is not statistically significant; likely due to sector share differences across countries versus unweighted IRFs.
- Timing and significance:
  - Private and public consumption responses delayed by four years.
  - Investment response delayed and mostly insignificant over ten years.
  - Employment falls slightly and is statistically significant between the six and eighth year horizon (appendix G).
  - Real exports decline accumulates to 9% after ten years; real imports peak decline 3.5% (lagged four years).
  - Real exchange rate depreciates approximately 20 percent after four years (persistent, imprecisely estimated).
- Industry-specific magnitudes:
  - Mining: −2% at t=0, −14.5% after ten years.
  - Manufacturing: −6% (lag of three years).
  - Services & others: −5.5% (lag of three years).

*Source: IMF working paper chapter "3.1 Mining Activity" (content unit: wpiea2023097-print-pdf - 3.1    Mining Activity).*

### 6.4    Heterogeneity Across Countries

### 6.4    Heterogeneity Across Countries

### Income and Mining Sector Size Effects
- Augmented local projection specification includes interaction terms for per capita income and mining sector share MS_i (value added of mining averaged over the 10 years preceding each episode):
  - y_{t+h,i} − y_{t−1,i} = α + β_h Δˆq_{t,i} + γ_h (Δˆq_{t,i} * MS_i) + θ_h MS_i + Σ_{j=1}^p Γ_{h j} X_{t−j,i} + ψ_n + φ_t + u_{t+h,i}.
- Empirical responses for mining sector sizes at the 10th and 90th percentiles (mining share of 2.5% and 58%, respectively):
  - Countries with relatively small mining sectors: decline of 3.7% in real GDP after ten years.
  - Countries with large mining sectors: decline of 15.3% in real GDP after ten years.
- Nonlinearity: impact of mining share is less than proportional — doubling the mining share increases the impact on real GDP on average by only about 40 percent.
- Normalization for impulse responses: shock episode normalized to an accumulated exogenous extraction decline of 40% over ten years.

### Income Level Interaction (per capita income)
- Augmented local projection with per capita income dummy I_n (0 for high income; 1 for middle and low income, World Bank (2022) classification) and triple interaction:
  - y_{t+h,i} − y_{t−1,i} = α + β_h Δˆq_{t,i} + γ_h (Δˆq_{t,i} * I_n) + σ_h (Δˆq_{t,i} * MS_i) + ρ_h (Δˆq_{t,i} * I_n * MS_i) + θ_h MS_i + Σ_{j=1}^p Γ_{h j} X_{t−j,i} + ψ_n + φ_t + u_{t+h,i}.
- Key findings:
  - Declines in mining activity are particularly detrimental for low-income countries with large mining sectors.
  - For a 40% decline in extraction over ten years in a sector that makes up about 58% of aggregate value added:
    - Real GDP in low-income economies falls by 30% after ten years.
    - Real GDP in high-income countries falls by 10% after ten years.
  - Temporal pattern: initial impact stronger in high-income countries (with large mining sector), but by year 3 the ranking flips and low-income countries experience increasingly worse outcomes over time.
- Sample notes:
  - Small sector size corresponds to 10th percentile in sample: 5 percent for HICs and 2.5% for LICs.
  - Large sector size corresponds to 90th percentile: 58% for HICs and 50% for LICs.
  - There are 11 high and 24 low-income countries in the sample.

### Manufacturing Sector Size Effects
- Construct mining-manufacturing ratio MR_i = MS_i / Manf_i and estimate:
  - y_{t+h,i} − y_{t−1,i} = α + β_h Δˆq_{t,i} + γ_h (Δˆq_{t,i} * MR_i) + θ_h MR_i + Σ_{j=1}^p Γ_{h j} X_{t−j,i} + ψ_n + φ_t + u_{t+h,i}.
- Data limitation: manufacturing value added available for only 25 shock episodes; small sample reduces statistical power when including multiple interactions.
- Findings:
  - A larger manufacturing sector (lower MR_i) can buffer the shock to extraction.
  - Small mining-manufacturing ratio (10th percentile): mining sector is 30% as large as the manufacturing sector.
  - Large mining-manufacturing ratio (90th percentile): mining sector is eight times as large as the manufacturing sector.
- Impulse responses normalized to an accumulated exogenous extraction decline of 40% over ten years; shaded areas report 90 percent confidence intervals.

### The Role of Institutions
- Research questions:
  - Do declines in extraction improve institutional quality (reverse of resource-boom deterioration)?
  - Does pre-shock institutional quality affect the economic impact of extraction declines?
- Institutional quality measure: Center for Systemic Peace (2022) indicator (-10 to 10).
- Results:
  - Persistent declines in extraction do not cause improvements in institutional quality in either high or low-income countries; impact statistically insignificant from zero (Figure 13a).
  - Institutional quality before the shock affects GDP responses:
    - Five years after the shock, difference in real GDP between countries with high and low institutional quality is about 5 percentage points.
    - After ten years, a democratic country with high institutional quality (indicator = 10) experiences a decline in real GDP of 2.5%; an autocratic country with bad institutions (indicator = -10) exhibits a decline of 10% after ten years (Figure 13b).
  - Interpretation: higher institutional quality may enable more diversified economies or faster enactment of structural policies to address shortfalls in mining output and revenues; institutional quality acts as a proxy for ability to cushion negative effects with public policy.

### Robustness and Complementary Analyses
- Sub-sample analysis (non-oil vs oil shock episodes):
  - Non-oil shocks (15 decline episodes in metals): a 40% decline in extraction over ten years causes a real GDP slowdown between 5% and 6% after six years; impact levels off for commodities ex. oil.
  - Oil and gas shocks (20 decline episodes): the negative impact becomes even more pronounced after 10 years.
- Leave-one-out sensitivity: results robust to excluding one of the 35 decline episodes at a time.
- Anticipation effects:
  - Local projection including anticipation dummy: anticipated declines lead to an insignificantly smaller response in point estimates over the 10-year horizon than unanticipated ones.
  - Sample in anticipation test: 27 shock episodes included; 24 unanticipated and 4 anticipated; no data for the remaining seven.
  - Interpretation: anticipated episodes are still hard to predict and leave little time for adjustment.
- Long-term horizon:
  - Impulse response over 20 years: after twenty years, real GDP is still 4% lower due to the decline in extraction; no reversed resource curve even 20 years after the initial shock.
- Synthetic Counterfactual Method (SCM):
  - SCM used as complementary method; trimmed to observations with RMSPE below the 75th percentile for fit.
  - Median results: real GDP decreases by approximately 15 percent after six years relative to the synthetic counterfactual.
  - Other SCM median deviations:
    - Real exports and real imports: declines consistent with GDP gap.
    - Employment: deviates with a lag of six years, returns to synthetic counterpart by year 14.
    - Real private consumption falls less than real GDP.
    - Real investment declines as much as GDP (but SCM fit for investment may reduce reliance on its estimates).

*Source: 6.4 Heterogeneity Across Countries, wpiea2023097-print-pdf*

### References

### References

### Major literature cited
- Key methodological and empirical works on resource booms, Dutch Disease, and synthetic control methods:
  - Abadie, A., A. Diamond, and J. Hainmueller (2010). “Synthetic control methods for comparative case studies: Estimating the effect of California’s tobacco control program”. Journal of the American Statistical Association 105.490, pp. 493–505.
  - Abadie, A. and J. Gardeazabal (2003). “The economic costs of conflict: A case study of the Basque Country”. American Economic Review 93.1, pp. 113–132.
  - Acemoglu, D., S. Naidu, P. Restrepo, and J. A. Robinson (2019). “Democracy does cause growth”. Journal of Political Economy 127.1, pp. 47–100.
  - Allcott, H. and D. Keniston (2018). “Dutch Disease or agglomeration? The local economic effects of natural resource booms in modern America”. The Review of Economic Studies 85.2, pp. 695–731.
  - Brunnschweiler, C. N. and E. H. Bulte (2008). “The resource curse revisited and revised: A tale of paradoxes and red herrings”. Journal of Environmental Economics and Management 55.3, pp. 248–264.
  - Feenstra, R. C., R. Inklaar, and M. P. Timmer (2015). “The next generation of the Penn World Table”. American Economic Review 105.10, pp. 3150–82.
  - Jordà, Ò. (2005). “Estimation and inference of impulse responses by local projections”. American Economic Review 95.1, pp. 161–182.
  - Montiel Olea, J. L. and M. Plagborg-Møller (2021). “Local projection inference is simpler and more robust than you think”. Econometrica 89.4, pp. 1789–1823.
  - Ploeg, F. Van der and A. J. Venables (2012). “Natural resource wealth: The challenge of managing a windfall”. Annual Review of Economics 4.1, pp. 315–337.
  - Sachs, J. D. and A. M. Warner (2001). “The curse of natural resources”. European Economic Review 45.4-6, pp. 827–838.
- Data sources and institutional reports:
  - BP (2022). Statistical review of world energy. British Petroleum, London.
  - International Energy Agency (2022). World Energy Outlook. International Energy Agency. Paris, France.
  - United Nations (2022). Statistical Database. United Nations, New York.
  - World Bank (2022). New World Bank country classifications by income level: 2022-2023. World Bank, Washington, DC.
  - Center for Systemic Peace (2022). The Polity Project. Center for Systemic Peace, Vienna, VA.
  - U.S. Geological Survey (2012). Historical statistics for mineral and material commodities in the United States. U.S. Geological Survey, Reston, V.A.
  - Federal Institute for Geosciences and Natural Resources (2012). Production database. Federal Institute for Geosciences and Natural Resources, Hanover.

### Appendix A — Summary statistics: entire dataset (averages)
- Table 2: Summary of episodes of declining extraction activity by commodity (in averages).
- Key figures (exact values preserved):
  - Bauxite: # of Episodes 10; Decline of Extraction (% of Peak) -69; Decline of Extraction Quant. (Mil. USD) -476; Duration (Years) -3.4
  - Brown Coal: # of Episodes 2; Decline of Extraction (% of Peak) -76; Decline of Extraction Quant. (Mil. USD) -670; Duration (Years) 0.0
  - Copper: # of Episodes 17; Decline of Extraction (% of Peak) -80; Decline of Extraction Quant. (Mil. USD) -2269; Duration (Years) -3.8
  - Crude Oil: # of Episodes 21; Decline of Extraction (% of Peak) -47; Decline of Extraction Quant. (Mil. USD) -82906; Duration (Years) -19.4
  - Gold: # of Episodes 9; Decline of Extraction (% of Peak) -50; Decline of Extraction Quant. (Mil. USD) -3795; Duration (Years) -1.7
  - Hard Coal: # of Episodes 11; Decline of Extraction (% of Peak) -77; Decline of Extraction Quant. (Mil. USD) -1135; Duration (Years) -0.2
  - Lead: # of Episodes 30; Decline of Extraction (% of Peak) -75; Decline of Extraction Quant. (Mil. USD) -259; Duration (Years) -0.1
  - Natural Gas: # of Episodes 7; Decline of Extraction (% of Peak) -60; Decline of Extraction Quant. (Mil. USD) -60807; Duration (Years) -0.5
  - Nickel: # of Episodes 3; Decline of Extraction (% of Peak) -46; Decline of Extraction Quant. (Mil. USD) -493; Duration (Years) -0.5
  - Iron: # of Episodes 14; Decline of Extraction (% of Peak) -79; Decline of Extraction Quant. (Mil. USD) -555; Duration (Years) -1.5
  - Silver: # of Episodes 10; Decline of Extraction (% of Peak) -64; Decline of Extraction Quant. (Mil. USD) -249; Duration (Years) -0.2
  - Tin: # of Episodes 11; Decline of Extraction (% of Peak) -85; Decline of Extraction Quant. (Mil. USD) -2211; Duration (Years) -0.4
  - Zinc: # of Episodes 9; Decline of Extraction (% of Peak) -60; Decline of Extraction Quant. (Mil. USD) -1536; Duration (Years) -0.2
  - Total: 154

- Note: Column 5 (the average ADV) is the result of dividing column 3 by column 4 and the GDP at time t=0 for every shock.

### Appendix B — Summary statistics: 35 extraction decline episodes (selected entries)
- Table 3: Summary of the 35 episodes of exogenous extraction declines. Countries sorted by estimated value of the extraction decline as share of GDP (ADV). Anticipation indicates whether the initial decline was anticipated in IMF article IV consultation reports; there are no reports for 6 out.
- Selected rows with exact entries:
  - Kuwait — Oil — 1979-1982 — Duration 4 — ADV 23.3 — Reason Policy change — Anticipated No
  - Brunei Darussalam — Oil — 1979-1981 — Duration 3 — ADV 13.5 — Reason Policy change — Anticipated NA
  - Chad — Oil — 2005-2014 — Duration 10 — ADV 4.1 — Reason Depletion — Anticipated No
  - Suriname — Bauxite — 1974-1977 — Duration 3 — ADV 3.8 — Reason Policy change — Anticipated NA
  - Gabon — Oil — 1997-2002 — Duration 6 — ADV 3 — Reason Depletion — Anticipated No
  - Norway — Oil — 2003-2013 — Duration 11 — ADV 1.5 — Reason Depletion — Anticipated Yes
  - Denmark — Gas — 2008-2013 — Duration 6 — ADV 0.3 — Reason Depletion — Anticipated Yes
  - United Kingdom — Oil — 1986-1990 — Duration 4 — ADV 0.2 — Reason Depletion — Anticipated Yes
  - Netherlands — Gas — 2013-2019 — Duration 6 — ADV 0.2 — Reason Depletion — Anticipated NA
  - Note: The table lists 35 episodes in total.

### Appendix C — Model: first-order conditions (two-sector model)
- Intra-temporal Euler equation relating non-tradable and tradable consumption, with relative price p, home bias γ and substitutability η:
  - (c_{T,t} / c_{N,t})^{1/η} = p^{* γ/(1−γ)} (Equation (17))
- Inter-temporal Euler equation relating current and future tradable consumption via relative price r and discount factor β:
  - (c_{T,t+1} / c_{T,t})^{1/η} = β∗(1+r) (Equation (18))

### Appendix D — Synthetic control methodology (formal statement)
- Observed variable y_it with determinants Z_it, i=1 treated country, i=2,...,N control countries, t ∈ [−10, ..., T].
- Synthetic counterfactual is a weighted average ∑_{i=2}^{N−1} w_i y_it with weights W = (w_2, ..., w_{N+1}), w_i ≥ 0 and ∑_{i=2}^{N−1} w_i = 1.
- Pre-shock matching conditions:
  - ∑_{i=2}^{N−1} w_i y_it = y_1t
  - ∑_{i=2}^{N−1} w_i Z_it = Z_1t
- Optimal weights W* solve:
  - min (S_1 − S_C W)' M (S_1 − S_C W) subject to w_i ≥ 0 and ∑_{i=2}^{N−1} w_i = 1,
  - where S_1 is (k×1) vector of treated country characteristics, S_C is (k×N) for control nations, M is (k×k) symmetric, semi-definite relevance matrix.

### Appendix E — Synthetic control example: Gabon
- GDP synthetic counterfactual constructed from weighted average of eight oil-producing countries chosen to mimic Gabon’s pre-shock oil production and real GDP.
- Weighted countries (exact weights): Angola 8.1%; Ecuador 18.8%; India 29.4%; Iran 5%; Italy 18.9%; Saudi Arabia 0.9%; Thailand 15.1%; Vietnam 3.7%.
- Key outcomes (exact figures reported):
  - Real GDP drops by 16 percent after two years compared to the counterfactual.
  - The gap between actual and synthetic GDP widens to a peak size of 52 percent after eleven years.
- Figure 18 shows synthetic counterfactuals for: (a) Real GDP; (b) Employment; (c) Real Consumption; (d) Real Investment; (e) Real Exports; (f) Real Imports. Shaded area denotes duration of the oil production shock.

### Appendix F — Estimated shock persistence
- Persistence estimated following Caselli, Grigoli, and Sandri (2022) with specification:
  - ∆ˆq_{t+h,i} = α_A + β_{A,h} ∆ˆq_{t,i} + e_{t+h,i} (Equation (19))
- Interpretation:
  - Coefficient β_{A,h} is the autocorrelation coefficient; Figure 19 reports these coefficients.
  - A shock at time t tends to persist with an average duration of eight years.
- Iterative accumulation of persistence for baseline production level Y_t:
  - Y_{t+h} = Y_{t+h−1} * (β_{A,h} + α_A) = Y_{t−1} * (β_{A,h} + α_A) * ··· * (β_{A,0} + α_A) (Equation (20))

### Appendix G — Employment response
- Impulse response of the employment rate to shock episodes:
  - Employment rate defined as employed in the labour force (older than 15 and younger than 65) over the population.
  - Data quality caveat: data not reliable for many middle and low-income countries in the sample.
  - Responses are generated using a shock episode normalized to an accumulated exogenous extraction decline of 40% over ten years.
  - Figure 20 shows impulse responses with shaded 90 percent confidence intervals.

### Appendix H — Linearity restriction sensitivity
- Impulse responses for real GDP when restricting effect of sector size via:
  - y_{t+h,i} − y_{t−1,i} = α + β_h ∆(∆ˆq_{t,i} * MS_i) + ∑_{j=1}^p Γ^h_j X_{t−j,i} + ψ_n + φ_t + u_{t+h,i}
- Responses generated for a shock normalized to an accumulated exogenous extraction decline of 40% over ten years; shaded areas denote 90 percent confidence intervals (Figure 21).

### Appendix I — Sub-samples (leave-one-out)
- Impulse responses for real GDP excluding one decline episode at a time (sensitivity analysis).
- Responses normalized to an accumulated exogenous extraction decline of 40% over ten years; shaded areas report 90 percent confidence intervals (Figure 22).

### Appendix J — Industry-specific analysis
- Sectoral impulse responses of real value added to a shock episode (normalized to 40% accumulated decline over ten years):
  - Agriculture, value added (Figure 24a)
  - Construction, value added (Figure 24b)
  - Transportation, value added (Figure 24c)
  - Wholesale, value added (Figure 24d)
- Shaded areas in the figures report 90 percent confidence intervals.

*Economic Consequences of Large Extraction Declines — Working Paper No. WP/2023/097*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023097-print-pdf.pdf_