## wpiea2022010-print-pdf

## Source details

**Canonical URL:** [wpiea2022010-print-pdf](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022010-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022010-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022010-print-pdf.pdf.json)

---

### Executive summary and main findings
- Documents a "fiscal presource curse": giant resource discoveries tend to lead to government over-borrowing and deteriorations in debt sustainability.
- Debt-to-GDP ratios increase by about 1-2 percentage points of GDP per year in the first decade after a discovery.
- The average cumulative effect of discoveries on the debt-to-GDP ratio is about 15 percent of GDP in the first 10 years.
- Impact stronger for oil and gas discoveries; mineral discovery effect positive but imprecisely estimated.
- Discoveries increase the probability of fiscal stress and external debt distress in the first few years following discovery.
- Adverse debt effects significantly larger in countries with weaker political institutions and governance.

### Data, definitions, and sample
- Resource discovery datasets:
  - Oil and gas: discoveries of oil and natural gas (including condensate equivalent) fields with at least 500 million barrels (79,000,000 m3) of recoverable oil equivalent reserves, covering 1868-2012; Horn (2012).
  - Minerals: MinEx Consulting dataset (last update: February 2018) on giant deposits exceeding 1 million ounces of gold or 1 million metric tons of copper or equivalent.
  - Combined dataset constructed for 193 countries.
- Government debt:
  - Authors compiled government debt data from several sources; debt panel covers 1970-2017 and 171 countries.
  - Baseline debt metric: gross debt. Net debt = gross debt minus liquid financial assets used in robustness checks.
- Discovery indicators and additional measures:
  - Primary discovery indicator: dummy = 1 if at least one giant discovery in a country-year, 0 otherwise.
  - Intensive-margin measure: counts number of discovery episodes in the same year.
  - Net present value (NPV) of a discovery as percent of GDP used in some analyses.
- Panel coverage and descriptive statistics (selected):
  - Panel: 171 countries; time span: 1970-2017 (GDD primary).
  - Outcome means (selected): Government debt mean 49.57; SD 39.42; N and country counts as reported in Table B1.
  - Discovery dummies: Discovery dummy mean 0.08; Oil and gas discovery dummy mean 0.04; Mineral discovery dummy mean 0.04.

### Empirical strategy and identification
- Dynamic panel distributed lag model:
  - Baseline: one lag of dependent variable and 10 lags of discovery dummies; controls: country and year fixed effects and country-specific linear trends.
  - Debt_it = A(L) Debt_it + B(L) Disc_it + α_i + δ_t + γ X'_it + ε_it.
  - Impulse response function: IRF(L) = B(L) / (1 − A(L)) over horizons 1–10 years.
  - Standard errors: Driscoll and Kraay (1998), robust to spatial and temporal dependence.
- Crisis incidence:
  - Linear probability model for crisis_it+j with fixed effects: estimates for j ∈ {1,2,...,10}.
- Identification support:
  - Predictability tests (panel VAR Granger causality and likelihood-ratio tests) indicate lagged macro and political variables have negligible predictive power for discoveries once fixed effects are included; discoveries are serially correlated (past discoveries predict future discoveries).
  - Granger causality result: discoveries Granger cause debt accumulation at the 1 percent level.

### Main empirical results and key statistics
- IRF findings:
  - Pooling all discoveries: debt rises by about 1-2 percentage points of GDP per year in the 10 years immediately after a discovery; impact winds down in second decade.
  - Oil and gas discoveries: between 1 and 3 percentage points of GDP annually in the first decade following discovery.
  - Mineral discoveries: upward pressure but estimates imprecise and statistically indistinguishable from zero.
- Cumulative effects (Table 3):
  - All discoveries:
    - Time horizons 0-4 — Cumulative response 7.319 — p-value 0.022
    - Time horizons 5-10 — Cumulative response 7.752 — p-value 0.073
    - Time horizons 0-10 — Cumulative response 15.072 — p-value 0.025
  - Oil and gas discoveries:
    - Time horizons 0-4 — Cumulative response 7.916 — p-value 0.069
    - Time horizons 5-10 — Cumulative response 11.019 — p-value 0.055
    - Time horizons 0-10 — Cumulative response 18.935 — p-value 0.033
  - Mineral discoveries:
    - Time horizons 0-4 — Cumulative response 2.524 — p-value 0.592
    - Time horizons 5-10 — Cumulative response 2.203 — p-value 0.755
    - Time horizons 0-10 — Cumulative response 4.727 — p-value 0.606
- Crisis probabilities:
  - Fiscal crises (baseline indicator): discoveries significantly increase probability of fiscal crisis at the 5 percent level around t+2–t+4 years; effects last up to 6–7 years. Selected coefficients (Table C2):
    - t+3: 0.050 ∗∗ (standard error 0.021)
    - t+4: 0.052 ∗∗ (0.020)
    - t+5: 0.038 ∗∗ (0.018)
    - t+6: 0.041 ∗∗ (0.018)
  - External debt distress:
    - Positive effect over a 10-year horizon but imprecisely estimated overall.
    - Oil and gas discoveries significantly increase (at the 1 percent level) probability of external debt distress after t+3 to t+8 years (Table C3 summary).
- Additional numeric results (selected robustness / alternative debt measures, Table F1):
  - Gross debt cumulative responses:
    - All discoveries, 0-4: 13.613 — p-value 0.001
    - All discoveries, 5-10: 12.679 — p-value 0.004
    - All discoveries, 0-10: 26.291 — p-value <0.001
    - Oil and gas, 0-4: 13.834 — p-value 0.003
    - Oil and gas, 0-10: 24.087 — p-value 0.012
  - Net debt (GDD / WEO) and financial assets show mixed results:
    - Net debt, all discoveries, 0-10: 21.394 — p-value 0.087 (Table F1)
    - Financial assets, oil and gas, 0-4: cumulative response -8.624 — p-value 0.050

### Heterogeneity: role of institutions and governance
- Institutional indicators used:
  - Polity2 index (range -10 to 10), X-Polity, Control of corruption (Kaufmann et al., 2010), authors’ compiled governance indicator (coverage >90 percent).
- Main heterogeneity finding:
  - Fiscal presource curse stronger in countries with weaker political institutions/governance; differentiated impact statistically and economically significant, especially in first decade after oil and gas discoveries (Figure 6).
  - Some evidence oil and gas discoveries do not have statistically significant impact on debt sustainability in countries with better governance or lower corruption incidence.
- Country examples illustrating heterogeneity:
  - Mozambique: large gas reserves; public debt roughly tripled in the decade after discovery; culminated in economic stress and debt crisis.
  - Botswana: strong institutions helped manage diamond revenues; public debt remained relatively low.
  - Norway: Government Pension Fund Global assets under management exceeding 1.3 trillion US dollars; gross debts modest in recent decades.
  - Ghana: 2007 discovery, significant commercial borrowing pre-extraction, IMF assistance requested in late 2014; debt assessed at "high risk of distress."

### Robustness checks and alternative specifications
- Robustness across discovery measures and debt perimeters:
  - Intensive-margin (count) IRFs not statistically different from zero; the occurrence dummy matters more than count (Online Appendix Figure G3).
  - Arezki et al. (2017) NPV-as-percent-of-GDP measure: IRFs close to zero in some specifications but sample-selection and measurement concerns noted.
  - Alternative debt datasets (GDD, WEO net, WEO gross) produce qualitatively consistent patterns; gross-debt effects often statistically significant.
- Sample and period splits:
  - 1970–2000: stronger impact at shorter horizons (up to 4 years).
  - 1980–2010: stronger impact at longer horizons (up to 10 years).
  - Subsample by income group: impact strongest in MICs; advanced countries' estimates often not significant.
- Lag and specification checks:
  - Longer lag structures (up to 20) tested; IRFs less reliable as lag length increases but qualitative conclusions similar.
  - Including country-level quadratic trends yields impact magnitude comparable to baseline.
- Data limitations documented:
  - Financial asset data coverage low (about one quarter of country-years).
  - Oil discovery data collection halted in 2012; mineral data coverage extends to 2017.
  - Mineral dataset excludes “bulk” minerals (coal, iron ore, bauxite, potash, phosphate).

### Policy implications and recommendations
- Discoveries can justify some borrowing to finance extraction, but observed patterns indicate substantial risk of excessive borrowing that may lead to unsustainable debt.
- Suggested policy directions:
  - Improve quality of institutions and strengthen overall governance, including reducing corruption opportunities and preventing in-fighting over natural resources.
  - Strengthen fiscal governance: enforce good practices for budget execution, promote transparency and accountability, implement rules-based resource management frameworks.
  - Reduce risks to debt sustainability: develop and implement sound borrowing plans and medium-term debt management strategies.
  - Adhere to international best practices for managing resource wealth (e.g., Extractive Industries Transparency Initiative good practices).
  - Incentivize multinationals in resource sectors to adhere to socially responsible corporate practices to reduce conflict risk.
  - International financial institutions (IMF, World Bank) to continue providing policy advice and technical assistance; domestic ownership of reforms critical.

### Limitations and avenues for future research
- Paper does not disentangle relative importance of proposed mechanisms:
  - Over-optimism about future growth.
  - Distorted beliefs/policy decisions leading to unwise borrowing and resource misallocation.
- Other limitations highlighted: data gaps on government financial assets, differing coverage/stopping points across discovery datasets, exclusion of international reserves from asset measures.

*Content based solely on wpiea2022010-print-pdf - References25 and accompanying appendices and sections provided.*

### References25

### wpiea2022010-print-pdf - References25

### Executive summary and main findings
- The paper documents a "fiscal presource curse": giant resource discoveries tend to lead to government over-borrowing and deteriorations in debt sustainability.
- Debt-to-GDP ratios increase by about 1-2 percentage points of GDP per year in the first decade after a discovery.
- The average cumulative effect of discoveries on the debt-to-GDP ratio is about 15 percent of GDP in the first 10 years.
- The impact is statistically and economically stronger for oil and gas discoveries; the impact of mineral discoveries is positive but imprecisely estimated.
- Discoveries increase the probability of fiscal stress and external debt distress in the first few years following discovery.
- The adverse debt effects are significantly larger in countries with weaker political institutions and governance.

### Data, definitions, and sample
- Resource discovery datasets:
  - Oil and gas: discoveries of oil and natural gas (including condensate equivalent) fields with at least 500 million barrels (79,000,000 m3) of recoverable oil equivalent reserves, covering the period 1868-2012; compiled by Horn (2012) from AAPG datapages.
  - Minerals: MinEx Consulting dataset (last update: February 2018) on giant deposits of non-bulk minerals exceeding 1 million ounces of gold or 1 million metric tons of copper or equivalent.
  - Combined resource dataset constructed for 193 countries.
- Government debt:
  - Authors compiled government debt data from several sources to maximize observations.
  - Debt panel covers 1970-2017 and 171 countries, including advanced economies, MICs and LICs.
- Discovery indicators and additional measures:
  - Primary discovery indicator: dummy = 1 if at least one giant discovery in a country-year, 0 otherwise.
  - Intensive-margin discovery measure: counts number of discovery episodes in the same year.
  - Net present value (NPV) of a discovery as percent of GDP used in some analyses.
- Supplementary materials:
  - Summary statistics and data sources reported in online Appendix Table B1.
  - Extensive figures and tables map temporal and geographical distribution of giant discoveries, debt trajectories, and heterogeneity analyses.

### Empirical strategy and identification
- Econometric framework:
  - Dynamic panel distributed lag model used to estimate the impact of discoveries on government debt over different horizons.
  - Treats timing of discovery events as an exogenous source of within-country variation to identify causal effects.
- Identification support:
  - Authors show lagged macroeconomic and political variables, including government debt, have negligible predictive power for discoveries once year and country fixed effects are included.
- Robustness and alternative specifications:
  - Results are robust to alternative measures of government debt and resource discoveries, sub-samples and country groupings, lag and regression specifications.
  - Robustness checks and extensions presented in appendices (heterogeneity, alternative debt measures, alternative discovery measures, lag specifications, and regression specifications).

### Mechanisms and theoretical context
- Theoretical expectations:
  - Small open economy framework predicts initial borrowing to finance investment and consumption after discovery, later repayment once production starts (Arezki et al., 2017).
  - Over-optimism about future payoffs and politically distorted incentives can drive excessive borrowing and amplify vulnerability to fiscal and debt crises.
  - Political economy factors: weak institutions, patronage, and time-inconsistency can produce a "voracity effect" and persistent fiscal mismanagement.
- Empirical link to existing literature:
  - Extends studies treating discoveries as unanticipated "news shocks" (Tsui, 2010; Cust and Mihalyi, 2017; Arezki et al., 2017) by focusing on fiscal outcomes and government debt rather than growth alone.
  - Aligns with theoretical models where frontloaded public investment, declining investment efficiency, or disappointment in future resource revenues raise debt sustainability risks (Melina et al., 2016).

### Heterogeneity and country examples
- Heterogeneity:
  - The fiscal presource curse effect is larger where political institutions and governance are weaker.
  - Oil and gas discoveries drive most of the observed debt dynamics; mineral discovery effects are positive but less precisely estimated.
- Country vignettes and comparative examples:
  - Mozambique: large gas reserves and borrowing against future revenues later associated with a debt crisis and external debt rated "in debt distress" (IMF, 2020c).
  - Ghana: discovery in 2007 followed by significant commercial borrowing before extraction in 2011; subsequent request for IMF assistance in late 2014; debt assessed at "high risk of distress" (IMF, 2020b).
  - Mongolia and Uganda discussed as contrasting experiences with aggressive borrowing versus resistance to the curse.

### Policy implications and interpretation
- Discoveries can justify some borrowing to finance extraction, but observed patterns indicate a substantial risk of excessive borrowing that may lead to unsustainable debt.
- Strong political and fiscal institutions and effective governance appear critical to mitigating the fiscal presource curse.
- Monitoring of expectations, prudent borrowing practices, and safeguards on the use of borrowed funds (e.g., productive public investment, high public investment efficiency) are implied priorities to reduce debt distress risk following discoveries.

*Italic: Content based solely on wpiea2022010-print-pdf - References25*

### Appendix A.I, the primary source of information is the comprehensive Global Debt Database

### Appendix A.I — Primary data: Global Debt Database (GDD)

### Data sources and coverage
- Primary source: Global Debt Database (GDD) recently compiled by Mbaye et al. (2018), which accounts for over 80 percent of the observations in the sample.
- Panel coverage:
  - Time span: five decades (1970-2017).
  - Countries: 171 countries, including almost all advanced economies, and most MICs and LICs.
- Debt perimeter:
  - Main perimeter: Central Government (about 70 percent of the observations).
  - Broader perimeters (General Government and others) available for a smaller number of countries.
- Mineral and oil discovery data notes:
  - Collection of oil discovery data was halted in 2012; this yields 5 additional years of coverage for mineral discoveries.
  - The mineral data excludes ‘bulk’ minerals such as coal, iron ore, bauxite, potash, and phosphate.
  - Horn (2012) dataset is main source for oil discoveries; mineral data has more exhaustive coverage (almost complete for giant-sized deposits).

### Debt measures and net debt
- Gross debt is the baseline debt metric in the panel.
- Robustness checks incorporate government financial assets using a measure of ‘net debt’ defined as ‘gross debt minus liquid financial assets’.
- Sources for net debt and asset data:
  - Mostly from the IMF’s World Economic Outlook (WEO) Database.
  - Complemented by information from Arbelaez and Sobrinho (2017).
- Scope of financial assets:
  - Underlying financial assets are typically held at the Central or General Government levels.
  - These asset measures do not include international reserves at central banks.
- Data limitations on assets:
  - Information on government financial assets is not comprehensive across countries and across time.
  - Despite compiling asset data from multiple sources, information was obtained for only one quarter of the country-years in the panel.

### Fiscal stress and debt distress episodes
- Baseline distress indicator: a comprehensive indicator of distress that reflects situations of extreme fiscal duress (Medas et al., 2018).
- The appendix notes methodological discussions and limitations related to measurement and identification of resource discovery shocks in Sub-[text truncated in source], indicating further discussion elsewhere in the document.

### Key methodological caveats highlighted
- Coverage gaps for financial asset data (only one quarter of country-years).
- Exclusion of international reserves from government financial asset measures.
- Mineral dataset omissions: excludes ‘bulk’ minerals and has differing coverage and stopping points relative to oil discovery data.
- Recognition that the oil discovery dataset collection was halted in 2012, affecting comparability and time coverage.

*Source: Appendix A.I — primary source: Global Debt Database (GDD) as compiled by Mbaye et al. (2018); supplementary sources cited in the appendix (Arbelaez and Sobrinho (2017); Horn (2012); Medas et al. (2018)).*

### section IV.D. Also, note that this measure is only available for oilfields, but not for mineral deposits.

### Section IV.D

### Definitions and crisis measures
- Fiscal crises: episodes in which large fiscal imbalances lead to the adoption of extreme measures, including:
  - external credit events (accumulation of external arrears);
  - exceptionally large financial support from the IMF (above 100 percent of quota);
  - implicit default on domestic public debt (proxied by large inflation or accumulation of domestic arrears);
  - loss of market access (proxied by inability to issue external debt and by spikes in sovereign bond spreads).
- Alternative indicator: episodes of external debt crises identified using information on default on sovereign external debt. Main data sources: Catão and Milesi-Ferretti (2014) and the main credit ratings agencies.
- Rationale: alternative measure has smaller country-year coverage but better isolates distress associated with external over-borrowing; many MICs and LICs rely on external borrowing due to underdeveloped domestic debt markets.

### Political institutions and governance (heterogeneity analysis)
- Four governance/institutions indicators used:
  - Polity2 index (Polity IV dataset): ranges from -10 (total autocracy) to 10 (total democracy); coverage: almost 90 percent of country-years in sample.
  - X-Polity index: subset of Polity2 components associated with the executive branch.
  - ‘Control of corruption’ indicator (Kaufmann et al., 2010).
  - Authors’ compiled governance indicator combining Kaufmann et al. (2010), ICRG, CNTS Data Archive, and Fraser Institute; coverage: over 90 percent of country-years.
- Definition used: governance as the set of “institutions, mechanisms, and practices through which governmental power is exercised in a country, including for the management of public resources and regulation of the economy” (IMF, 2017d).
- Purpose of heterogeneity analysis: test whether institutions/governance condition the impact of discoveries on government debt.

### Stylized facts on discoveries and macroeconomic salience
- Datasets: over a thousand giant discoveries — 335 of oil, 284 of gas, and 458 of minerals.
- Temporal patterns:
  - Mineral discoveries: somewhat evenly spread over time.
  - Oil and gas discoveries: spike in the late 1960s to early 1970s and increase after mid-1990s.
- Regional patterns:
  - Oil and gas: higher incidence in Persian Gulf, North Africa, Russia, North Sea.
  - Minerals: more evenly spread; prevalence in East Asia, Latin America, Southern Africa.
- Top countries by giant discoveries:
  - Oil and gas: Russia, Iran, United States, Saudi Arabia, China.
  - Minerals: United States, Canada, Australia, South Africa, Russia.
- Resource composition: mineral discoveries most prevalent for gold, silver, and copper.
- Macroeconomic magnitude: Arezki et al. (2017) computation — NPV of a median oil (or gas) discovery equals 9% of a country’s GDP.
- Country examples:
  - Mozambique: decade after giant natural gas discovery saw public debt roughly triple, culminating in economic stress and a debt crisis; weak fiscal institutions and absence of a full-fledged resource management framework cited as contributing drivers.
  - Botswana: strong institutions helped manage large diamond revenues; public debt remained relatively low.
  - Guyana: post-2015 large oil discovery; public debt remains moderate and framework to manage resource wealth is being put in place; too early to judge outcome.

### Methodology — empirical framework
- Dynamic panel distributed lag model:
  - Debt_it = A(L) Debt_it + B(L) Disc_it + α_i + δ_t + γ X'_it + ε_it
  - Debt_it: public-debt-to-GDP ratio of country i at time t.
  - Disc_it: dummy variable for a giant discovery event (oil, gas, or minerals).
  - Baseline specification: one lag of dependent variable and 10 lags of discovery dummies.
  - Controls: country and year fixed effects, country-specific linear trends; X'_it is vector of controls in robustness checks.
  - Standard errors: Driscoll and Kraay (1998), robust to spatial and temporal dependence.
- Impulse response function (IRF):
  - IRF(L) = B(L) / (1 − A(L)) over horizons (e.g., 1 to 10 years).
- Identification logic:
  - Discoveries treated as plausibly exogenous “natural experiments” because timing is argued to be hard to predict; after controlling for fixed effects, within-country timing variation used to identify causal effects.
- Linear probability model for crisis incidence:
  - crisis_it+j = α_i + δ_t + β_j · Disc_it + γ_j · pastDisc_it + ε_it
  - crisis_it+j: dummy for fiscal stress or debt distress at time t+j.
  - Controls: fixed effects and number of discoveries in previous 10 years (pastDisc_it).
  - Interpret β_j as marginal effect of a discovery on probability of fiscal stress or debt distress after j years; estimates presented for j ∈ {1,2,...,10}.
- Panel dimensions note:
  - Baseline T = 42 helps reduce Nickell bias in dynamic individual-effects models.

### Predictability tests (validating exogeneity)
- Granger causality panel VAR (equation 4) with explanatory variables considered one at a time: government debt, population, real GDP per capita, real GDP growth, CPI inflation, current account balance, fiscal deficit, real exchange rate, Polity2, governance indicator.
  - Standard errors clustered at country level; panel VAR fitted using GMM.
  - Result: estimated coefficients not statistically different from zero for most variables; cannot reject null that government debt does not Granger cause discoveries (column 1).
  - Discoveries Granger cause debt accumulation at the 1 percent level.
- Joint significance test (Mertens and Ravn, 2012): fixed effects logistic regressions with and without lagged macro variables; p-values from likelihood ratio tests reported in Table 2.
  - Result: cannot reject null of nonpredictability of discovery events; past discoveries are strongest predictor of future discoveries — discoveries are serially correlated, consistent with “learning-by-doing.”

### Baseline empirical results
- IRF findings (Figure 3):
  - Pooling all types of discoveries: debt rises by about 1-2 percentage points of GDP per year in the 10 years immediately after a discovery; impact gradually winds down in the second decade.
  - Oil and gas discoveries: bulk of impact in first decade — between 1 and 3 percentage points of GDP annually in first decade following discovery.
  - Mineral discoveries: upward pressure on debt trajectory but estimates imprecise and statistically indistinguishable from zero.
- Cumulative effects (Table 3):
  - All discoveries: cumulative effect on debt-to-GDP ratio of 15 percent of GDP in the first 10 years.
  - Oil and gas discoveries: cumulative effect almost 20 percent of GDP in the first 10 years.
- Crisis probability results:
  - Fiscal crisis probability (baseline indicator): discoveries significantly increase probability of fiscal crisis at the 5 percent level around 2–4 years after discovery; mainly driven by oil and gas discoveries and lasting up to 6–7 years (Figure 4).
  - External debt distress indicator: positive effect over a 10-year horizon but imprecisely estimated overall (Figure 5a); oil and gas discoveries significantly increase (at the 1 percent level) probability of external debt distress after 3 to 8 years (Figure 5b).
- Economic implications:
  - Observed haircuts in debt restructurings tend to be positively correlated with debt level prior to exchange (Online Appendix Figure B4).
  - Larger haircuts associated with persistently higher borrowing costs and longer exclusion from private capital markets.
  - Full-blown debt crises entail large output costs; fiscal presource curse may entail large indirect economic costs.

### Heterogeneity: role of institutions and governance
- Empirical strategy: split sample into countries with “strong” institutions (country average index above sample median) and “weak” institutions (below median); sample average used, leaving out time effects and focusing on cross-country variation.
- Main finding:
  - Fiscal presource curse stronger in countries with weaker political institutions/governance; differentiated impact statistically and economically significant, especially for the two political institutions indicators and during the first decade after oil and gas discoveries (Figure 6).
  - Some evidence that oil and gas discoveries do not have statistically significant impact on debt sustainability in countries with better overall governance or lower corruption incidence.
  - For mineral discoveries, the impact on debt sustainability is not statistically significant in countries with stronger institutions/governance and remains statistically indistinguishable across governance levels for minerals (online Appendix Figures E1 and E2).
- Interpretation: fiscal presource curse appears to be a manifestation of a deeper political curse; strong institutions and governance mitigate the fiscal impact of discoveries.
- Supporting country cases:
  - Chile and Norway: large numbers of discoveries but no acute debt crisis clearly linked to fiscal presource curse; government gross debts remained modest in recent decades.
  - Norway: positive net worth with Government Pension Fund Global assets under management exceeding 1.3 trillion US dollars.

### Robustness checks
- Alternative measure of government debt — net debt (gross debt minus government financial assets):
  - Net debt increases following a discovery, with cumulative effect of more than 20 percent of GDP in 10 years (Online Appendix Figure G1).
  - IRFs of financial assets are negative and small; cumulative response does not differ statistically from post-discovery decade (Online Appendix Figure G2).
  - Caveat: low quality and coverage of government financial assets data.
- Alternative measures of discovery:
  - Count of discovery episodes (to capture multiple discoveries in a year): IRFs not statistically different from zero (Online Appendix Figure G3). Conclusion: intensive margin (number of discoveries) matters less than occurrence of a giant discovery (dummy).
  - Arezki et al. (2017) NPV-as-percent-of-GDP measure for oil and gas discoveries: IRFs negative but very close to zero; impact on debt not economically relevant (Online Appendix Figure G4a).
    - Concerns: measure relies on endogenous variables (risk-adjusted discount rates, GDP) and URR subject to measurement error.
  - Using alternative debt measures from IMF WEO (net debt and gross debt) with Arezki-style discovery measure: IRFs turn positive and statistically significant but economic relevance modest (Online Appendix Figures G4b, G4c); sample-selection concerns noted.
- Alternative controls:
  - Added controls with one lag each: log population, log real GDP per capita, inflation, Polity2.
  - Result: impact on government debt similar to baseline; peak around 8–11 years after discovery (Online Appendix Figure G5a).
  - Mineral discovery IRFs become more precisely estimated and statistically significant in this specification (Online Appendix Figure G5c).
- Alternative samples:
  - Restrict sample to countries with at least one giant discovery since 1970 (74 percent of countries from initial sample; 37 percent oil and gas; 41 percent minerals) — robustness explored (further details in continuation of robustness section beyond provided excerpt).
- Additional notes:
  - Longer discovery lags (up to 20) were tested; results reassure that short-term learning-by-doing-related autocorrelation does not pose major issue to baseline estimation.
  - The paper focuses on medium-term cumulative effects (up to 10 years) given intertemporal nature of debt sustainability and typical 4–7 year lag between discovery and production for oil and gas.

*Source: wpiea2022010-print-pdf - section IV.D.*

### Appendix Figure G6 presents the findings.   Even with this smaller sample,  we find that the

### wpiea2022010-print-pdf - Appendix Figure G6 presents the findings.   Even with this smaller sample,  we find that the

### Main empirical findings
- Giant oil, gas and mineral discoveries lead to large and persistent debt buildups, in the order of 15 percent of GDP during the first decade after the discoveries.
- The evidence supports a “fiscal presource curse”: discoveries themselves drive the effect rather than merely a country being (or turning) resource rich.
- The significant debt buildup is associated with an increased likelihood of fiscal stress and debt distress, implicating the fiscal presource curse as a possible indirect cause of sovereign debt crises in emerging and developing countries.

### Robustness to sample composition and period
- Smaller sample (Appendix Figure G6): magnitude of the impact is nearly the same as in the baseline estimation, though the IRFs seem estimated more imprecisely.
- Sample period splits tested:
  - 1970–2000: results show a stronger impact at shorter horizons (up to 4 years after a discovery).
  - 1980–2010: results show a stronger impact at longer horizons (up to 10 years after a discovery).
- Conclusion from these tests: findings have high external validity and are not undermined by alternative sample periods (online Appendix Figures G7 and G8).

### Alternative country groupings
- Sample split into advanced economies, emerging economies (mostly MICs), and developing countries (mostly LICs):
  - Impact strongest in MICs (online Appendix Figures G13-G15).
  - Estimated impact for advanced countries is not statistically significant, consistent with stronger political institutions and governance in these countries.
  - LICs show a muted impact; potential explanation is that many giant discoveries occurred when those countries were at different stages of development (some emerging economies today would have been LICs at the time of discoveries).

### Other specification checks
- Lag structure:
  - Tested longer lag specifications, e.g., two lags for government debt and up to 20 lags for the discovery dummies.
  - IRFs become less reliable as the lag structure increases, partly due to data availability (online Appendix Figures G9, G10 and G11).
- Secular trend in debt:
  - The secular increase in debt-to-GDP ratios starting in the 1970s could be a caveat to identification.
  - Including a country-level quadratic trend in the baseline model yields an impact magnitude comparable to the baseline and estimated slightly more precisely (online Appendix Figure G12).
  - Conclusion: findings are largely unrelated to the secular trend in debt ratios.

### Policy implications and recommendations
- The fiscal presource curse can be as important quantitatively as cyclical drivers of debt accumulation and thus can pose a material threat to debt sustainability.
- Suggested policy directions:
  - Improve quality of institutions and strengthen overall governance, including reducing corruption opportunities and preventing in-fighting over natural resources.
  - Strengthen fiscal governance: enforce good practices for budget execution, promote transparency and accountability, implement rules-based resource management frameworks.
  - Reduce risks to debt sustainability: develop and implement sound borrowing plans and medium-term debt management strategies.
  - Adhere to international best practices for managing resource wealth (e.g., Extractive Industries Transparency Initiative good practices).
  - Incentivize multinationals in resource sectors to adhere to socially responsible corporate practices to reduce conflict risk.
  - International financial institutions (IMF, World Bank) to continue providing policy advice and technical assistance; domestic ownership of reforms is critical.

### Limitations and avenues for future research
- The paper does not disentangle the relative importance of the two proposed mechanisms:
  - Over-optimism about future growth.
  - Distorted beliefs/policy decisions leading to unwise borrowing and resource misallocation.
- These and other questions remain for future research.

*Source: wpiea2022010-print-pdf*

### REFERENCES

### REFERENCES

### Key empirical findings and analytical focus
- The paper analyzes the impact of giant resource discoveries (oil and gas, and minerals) on government debt, fiscal crises, and external debt distress using panel data and dynamic specifications with country-specific linear time trends.
- Estimation approaches:
  - Fixed effects regressions with Driscoll-Kraay standard errors.
  - Dynamic models that include 10 lags of discovery dummies and one lag of the dependent variable for some specifications.
  - Linear probability models for fiscal crises and external debt distress with Driscoll-Kraay standard errors.
  - Panel logistic regressions and panel VAR Granger causality Wald tests for discovery predictability.
- Data sources repeatedly cited:
  - Horn (2012) for oil and gas discovery data.
  - MinEx Consulting (last update: February 2018) for mineral discovery data.
  - IMF’s Global Debt Database (Mbaye et al., 2018) for government debt.
  - Gerling et al. (2017) for fiscal crises definition.
  - Polity IV dataset (Polity2 index) and Kaufmann et al. (2010) for governance/control of corruption indicators.

### Tables — tests, estimates, and exact reported statistics
- Table 1: Granger Causality Tests of Discovery Predictability
  - Dependent variable: dummy for giant resource discovery events.
  - Method: panel VAR Granger causality Wald test with country and time fixed effects; standard errors clustered at the country level.
  - Sample frame: 1970-2012.
  - Regressors and reported Chi-square statistics and P-values by lag:
    - 1 lag — Chi-square statistics: 0.026, 0.248, 3.140, 1.357, 1.417, 0.691, 4.630, 0.561, 2.819, 1.700. P-value: 0.871, 0.618, 0.076, 0.244, 0.234, 0.406, 0.031, 0.454, 0.093, 0.192.
    - 2 lags — Chi-square statistics: 0.043, 1.155, 4.063, 5.917, 1.824, 2.599, 4.765, 0.582, 2.937, 3.027. P-value: 0.979, 0.561, 0.131, 0.052, 0.402, 0.273, 0.092, 0.446, 0.230, 0.220.
    - 3 lags — Chi-square statistics: 0.237, 7.395, 5.249, 6.444, 4.900, 2.906, 3.972, 0.599, 3.212, 3.022. P-value: 0.971, 0.060, 0.154, 0.092, 0.179, 0.406, 0.265, 0.439, 0.360, 0.388.
  - Notes: explanatory variables included one at a time; null hypothesis is that distributed lags jointly equal zero.
- Table 2: Likelihood Ratio Tests of Discovery Predictability
  - Dependent variable: dummy for giant resource discovery events.
  - Explanatory variables: government debt (% of GDP), population (in log), real GDP per capita (in log), real GDP growth, CPI inflation, current account balance (3-year average), fiscal deficit (% of GDP), real exchange rate, Polity2 index and governance indicator.
  - Null hypothesis: coefficients on the first lag of each variable are jointly zero; based on panel logistic regressions with country and time fixed effects; standard errors clustered at the country level.
  - Reported statistics (as listed):
    - Chi-square statistics: 6.930, 23.574, 14.027, 20.545, 17.02025.974
    - P-value: 0.732, 0.015, 0.172, 0.038, 0.074, 0.007
    - N. of observations: 931, 931, 707, 707, 575, 575
    - N. of countries: 36, 36, 28, 28, 21, 21
    - Country and year fixed effects: X X X X X X
    - Previous discoveries in 10 years: X X X
  - Sample frame: 1970-2012. Discovery data sources as above.
- Table 3: Cumulative Impact of Giant Discoveries on Government Debt
  - Definition: cumulative responses equal to sum_{h=i}^{N} b_h where b_h is the impulse response at horizon h.
  - P-values computed using delta method (nonlinear combinations of OLS-estimated parameters from Equation 2).
  - Null hypothesis tested: H0: sum_{h=i}^{N} b_h = 0.
  - Sample: unbalanced panel of 169 countries for period 1970-2012 (1970-2017 for mineral discoveries). Debt data from IMF GDD (Mbaye et al., 2018) and authors’ compilation.
  - Exact reported cumulative responses and p-values:
    - All discoveries:
      - Time horizons 0-4 — Cumulative response 7.319 — p-value 0.022
      - Time horizons 5-10 — Cumulative response 7.752 — p-value 0.073
      - Time horizons 0-10 — Cumulative response 15.072 — p-value 0.025
    - Oil and gas discoveries:
      - Time horizons 0-4 — Cumulative response 7.916 — p-value 0.069
      - Time horizons 5-10 — Cumulative response 11.019 — p-value 0.055
      - Time horizons 0-10 — Cumulative response 18.935 — p-value 0.033
    - Mineral discoveries:
      - Time horizons 0-4 — Cumulative response 2.524 — p-value 0.592
      - Time horizons 5-10 — Cumulative response 2.203 — p-value 0.755
      - Time horizons 0-10 — Cumulative response 4.727 — p-value 0.606

### Data sources and construction (online appendix excerpt)
- Government gross debt data:
  - Primary source: IMF’s Global Debt Database (GDD), available at Global Debt Database (IMF).
  - The repository covers 190 countries over the period 1950- (text ends).
- Discovery datasets:
  - Oil and gas discovery data: Horn (2012).
  - Mineral discovery data: MinEx Consulting (last update: February 2018).
- Other databases and indicators used across analyses:
  - IMF’s Global Debt Database (Mbaye et al., 2018).
  - Polity IV dataset (Polity2 index).
  - Kaufmann et al. (2010) governance/control of corruption indicators.
  - Gerling et al. (2017) fiscal crises definition.

*Content derived from: wpiea2022010-print-pdf - REFERENCES*

### 2018.  The database was compiled by IMF staff using information from almost 100 different

### The ‘Fiscal Presource Curse’: Giant Discoveries and Debt Sustainability (Working Paper No. [WP/22/XX]) — 2018 (selected appendix and tables)

### Data and sample construction
- Database composition and coverage
  - The Global Debt Database (GDD) accounts for about four fifths of all country-years in the sample.
  - Final sample: 171 countries comprising 31 advanced economies, 77 emerging market economies, and 63 low-income countries (WEO classification).
  - Sample period: 1970-2017 (noted in Table B1 notes).
- Debt measures and assets
  - Government net debt defined as gross debt minus government holdings of liquid financial assets; WEO is the main source for net debt.
  - Financial assets data and net debt data cover only one quarter of the country-years in the panel of 171 countries.
  - When WEO net debt unavailable, sources include OECD, Eurostat, and IMF’s Government Finance Statistics Yearbook.
- Discovery data
  - Oil and gas discovery data from Horn (2012).
  - Mineral discovery data from MinEx Consulting (last update: February 2018).
  - Net present value of oil and gas discoveries (Arezki et al., 2017) used in intensive-margin specifications.

### Governance indicator construction
- Core source: Kaufmann et al. (2010) Worldwide Governance Indicators (WGI).
- Selected five WGI subcomponents: Political Stability and Absence of Violence/Terrorism; Government Effectiveness; Regulatory Quality; Rule of Law; Control of Corruption.
- Strategy to extend coverage back to 1970:
  - (i) Aggregate WGI index by averaging the five indicators and normalizing for 1996 onward (with linear interpolation for 1997, 1999, 2001).
  - (ii) For 1970-1995, construct proxies for subcomponents using multiple sources:
    - Political Stability proxied by ICRG Political Risk Rating (PRR) Government Stability and Internal Conflict when available; otherwise CNTS Data Archive indicators combined to approximate PRR subcomponents.
    - Government Effectiveness proxied by PRR Bureaucracy Quality.
    - Regulatory Quality proxied by PRR Investment Profile or Fraser Institute’s Regulation (Economic Freedom of the World Index). Missing annual values for Economic Freedom index (available in 1970, 1975, 1980, 1985, 1990, 1995, 2000) filled using cubic spline.
    - Rule of Law proxied by PRR Law and Order or Economic Freedom’s Legal System and Property Rights.
    - Corruption proxied by PRR Corruption or V-Dem Political Corruption Index.
  - Indicators normalized and adjusted by sample standard deviation by country.
  - Result: coverage increased to over 90 percent of country-years in the panel.

### Descriptive statistics (selected, from Table B1)
- Outcome variables (sample means, SD, min, max, N, countries, years):
  - Government debt: Mean 49.57; SD 39.42; Min 1.12; Max 224.75; N 76, 95; Countries 169; Years 47
  - Government net debt: Mean 26.45; SD 64.02; Min -344.21; Max 222.46; N 1469; Countries 69; Years 36
  - Government gross debt: Mean 49.85; SD 37.85; Min 1.57; Max 224.75; N 3525; Countries 110; Years 36
  - Government financial assets: Mean 23.40; SD 43.95; Min 0; Max 349.02; N 1292; Countries 67; Years 36
- Main regressors (discovery indicators):
  - Discovery dummy: Mean 0.08; SD 0.27; N 18299; Countries 193; Years 43
  - Oil and gas discovery dummy: Mean 0.04; SD 0.20; N 18299; Countries 193; Years 43
  - Mineral discovery dummy: Mean 0.04; SD 0.20; N 19071; Countries 193; Years 47
  - Net present value of oil and gas discoveries: Mean 3.04; SD 82.24; Min 0.63; Max 14.45; N 6969; Countries 184; Years 43
- Selected controls:
  - Polity2 index: Mean 2.86; SD 7.06; Min -10; Max 10; N 69011; Countries 165; Years 47
  - Governance indicator: Mean 51.61; SD 13.71; Min 17.44; Max 87.34; N 76271; Countries 169; Years 47
  - Real GDP per capita, PPP (2011 USD): Mean 74.27; SD 220.12; Min 0.45; Max 1369.44; N 71261; Countries 169; Years 45

### Empirical results — dynamic panel estimates and impulse responses
- Dynamic panel distributed lag estimates (Table D1):
  - Regressions: OLS with country and year fixed effects; Driscoll-Kray standard errors reported.
  - Lagged dependent variable coefficients (y_{t−1}) are consistently positive and significant (examples):
    - GDD dataset: 0.759*** (0.101)
    - WEO (Net): 0.655*** (0.053)
    - WEO (Gross): 0.806*** (0.027)
- Cumulative impacts on government debt (Table F1 — alternative debt measures)
  - Net debt:
    - All discoveries, 0-4 horizons: cumulative response 0.140; p-value 0.122
    - All discoveries, 5-10 horizons: cumulative response 11.254; p-value 0.092
    - All discoveries, 0-10 horizons: cumulative response 21.394; p-value 0.087
    - Oil and gas discoveries, 0-4: -2.228; p-value 0.762
  - Gross debt:
    - All discoveries, 0-4: cumulative response 13.613; p-value 0.001
    - All discoveries, 5-10: cumulative response 12.679; p-value 0.004
    - All discoveries, 0-10: cumulative response 26.291; p-value <0.001
    - Oil and gas discoveries, 0-4: 13.834; p-value 0.003
    - Oil and gas discoveries, 0-10: 24.087; p-value 0.012
    - Mineral discoveries: not statistically significant in most gross-debt cumulative tests (e.g., 0-4: -6.863; p-value 0.271)
  - Financial assets:
    - All discoveries, 0-4: cumulative response -6.022; p-value 0.109
    - Oil and gas discoveries, 0-4: cumulative response -8.624; p-value 0.050
- Intensive-margin and NPV specifications (Table F2)
  - Intensive margin and NPV specifications yield mixed results:
    - Net present value of oil and gas discoveries and net/gross debt outcomes show some significant cumulative responses:
      - Net present value → Government Debt (GDD) 5-10: -0.251; p-value 0.002; 0-10: -0.297; p-value 0.013
      - Gross Debt (WEO) 0-4: 0.404; p-value 0.040; 0-10: 0.715; p-value 0.056
      - Net Debt (WEO) 5-10: 0.928; p-value <0.001; 0-10: 1.291; p-value 0.005
- Crisis probabilities — Linear probability models (Table C2: Fiscal crises)
  - All discoveries: Discovery event coefficients (selected horizons):
    - t+3: 0.050 ∗∗ (standard error 0.021)
    - t+4: 0.052 ∗∗ (0.020)
    - t+5: 0.038 ∗∗ (0.018)
    - t+6: 0.041 ∗∗ (0.018)
    - Some negative and statistically significant coefficients appear at longer horizons (e.g., t+7 onward), indicating complex dynamics.
  - Number of discoveries in previous 10 years generally negative and often significant (e.g., t+4: -0.019 ∗∗∗ (0.007))
- Crisis probabilities — External debt distress (Table C3)
  - All discoveries: Discovery event coefficients include positive and significant effects at some horizons:
    - t+5: 0.021 ∗ (0.012)
    - Several later horizons show significance in extended specifications.
  - Oil and gas discoveries show stronger and more persistent positive effects on external debt distress probabilities:
    - Multiple horizons statistically significant (e.g., many coefficients significant at 5% or 1% across t+3 to t+10).
  - Number of discoveries in previous 10 years is positively associated with higher external debt distress probability and statistically significant across many horizons (e.g., 0.012 ∗∗∗ (0.002) at t+1).

### Heterogeneity and institutional quality
- Heterogeneity results (Figures E1 and E2)
  - Impact of giant discoveries on government debt varies by institutional quality:
    - Countries with institutions below the median (weaker institutions) show larger increases in government debt following oil and gas or mineral discoveries.
    - Measures used include Polity2, X-Polity, control of corruption indicator (Kaufmann et al., 2010), and the compiled governance indicator.
  - These heterogeneity results are estimated via fixed effects regressions with Driscoll-Kraay standard errors, dynamic model including 10 lags of discovery dummies and one lag of dependent variable, controlling for country-specific linear time trends.

### Robustness checks
- Robustness across datasets and specifications (Figures G1–G15; Tables F3–F5)
  - Alternative debt datasets (GDD, WEO net, WEO gross) produce consistent qualitative patterns: gross debt increases following discoveries with statistically significant cumulative responses in multiple specifications.
  - Subsample analyses:
    - Emerging economies: All discoveries, 0-4 cumulative response 8.741; p-value 0.048; 0-10 cumulative response 31.999; p-value 0.008 (Table F3)
    - Advanced countries: weaker and often insignificant cumulative responses in many specifications.
  - Alternative lag specifications (Table F4): results robust across p=2; q=10 and p=1; q=15 in many cases, with oil and gas discoveries often showing significant cumulative gross-debt responses.
  - Alternative regression specifications (Table F5):
    - Including quadratic time trends, controlling for debt defaults, and including other controls generally preserve the main finding that discoveries — especially oil and gas — are associated with cumulative increases in gross government debt in many specifications.
  - Net debt results are more mixed and less consistently significant than gross debt results; financial assets sometimes decline after oil and gas discoveries (e.g., financial assets 0-4 for oil and gas: -8.624; p-value 0.050).

### Key empirical points (concise)
- Gross government debt:
  - All discoveries → cumulative increase in gross debt significant: 0-4: 13.613; p-value 0.001; 0-10: 26.291; p-value <0.001 (Table F1).
  - Oil and gas discoveries strongly associated with cumulative increases in gross debt (e.g., 0-4: 13.834; p-value 0.003; 0-10: 24.087; p-value 0.012).
  - Mineral discoveries less consistently associated with gross-debt increases.
- Net government debt and financial assets:
  - Net debt cumulative responses for all discoveries are not consistently statistically significant across horizons (Table F1).
  - Financial assets tend to decline following discoveries in some specifications; oil and gas 0-4: -8.624; p-value 0.050 (Table F1).
- Crises:
  - Discoveries increase the probability of fiscal crises at certain short-to-medium horizons (Table C2).
  - Oil and gas discoveries increase the probability of external debt distress at multiple horizons (Table C3).
- Institutional quality:
  - Countries with weaker political institutions and governance indicators tend to experience larger debt increases following giant discoveries (Figures E1 and E2).

*Selected appendix material and empirical tables/figures from “The ‘Fiscal Presource Curse’: Giant Discoveries and Debt Sustainability” (Working Paper No. [WP/22/XX]).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022010-print-pdf.pdf_