## wpiea2025205-source-pdf

## Source details

**Canonical URL:** [wpiea2025205-source-pdf](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025205-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025205-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025205-source-pdf.pdf.json)

---

### Introduction and motivation
- State fragility constrains development in about a third of the world’s countries.
- About 65 countries are considered fragile by the World Bank, IMF or OECD as of 2024.
- Fragile states are home to close to a quarter of the world population; this share is expected to grow to a third by 2050.
- Paper proposes a parsimonious, response-based framework to identify fragility via system responses to shocks, remaining agnostic on root causes or manifestations.

### Conceptual framework and typology
- Fragility definitions and types:
  - Stress-induced fragility: disproportionate (supra-linear) adverse response to shocks (Taleb, 2012, 2013).
  - Chronic fragility: lack of responsiveness (sub-linear) to positive shocks or to time — inability to take off or sustain per capita growth over long periods.
  - Antifragility: accelerating benefit to large stressors (convex response).
- Operationalization focuses on:
  - Economic fragility: growth responses (per capita GDP dynamics).
  - Political fragility: incidents of internal conflict and coups d’état.
- Both chronic and stress-induced fragility can coexist and interact; response-based identification overlaps strongly with existing complex classifications.

### Identification methodology: chronic fragility
- Sample restriction and thresholds:
  - Only low and lower middle-income countries (LIC and LMIC) considered; pool reduced from 193 to 81.
  - Chronic fragility defined relative to the 50th percentile (median) of per capita growth among LICs and LMICs.
  - Median per capita growth averaged 1.5 percent during 1970-2023 and 1980-2023 and 1.8 percent during 1990-2023.
- Persistence and exclusion rules:
  - Countries classified as chronically fragile if per capita growth is below the median threshold, on average, during each of: 1970-2023, 1980-2023, and 1990-2023.
  - Exclude countries that showed sustained pickup over the past ten years (2015-2024) if growth remained above threshold for the entire ten-year window measured by trailing three-year moving averages (3yma).
  - First five years excluded for countries that broke off economically (South Sudan and former Soviet Union countries).
- Classification outcomes:
  - Below-median criterion initially identifies 43 countries as fragile.
  - Five countries excluded as exits due to sustained 2015-2024 growth on a 3yma basis: Cote d’Ivoire, Djibouti, Kenya, Senegal, and Togo.
  - Final count: 38 countries identified as chronically fragile (assessment based on end-2024 data as reported in IMF’s April 2025 WEO).
  - Of these 38 chronically fragile countries, 17 are low income and 21 are lower middle income.
- Interpretive note:
  - A per capita growth rate of 1.5 percent doubles GDP per capita in about 47 years and would reach current LMIC average income in about 78 years from the LIC average.

### Drivers of chronic fragility (constraints to growth)
- Two broad constraint categories:
  - Macroeconomic vulnerabilities (e.g., high debt).
  - Institutional constraints preventing markets from clearing efficiently (political, governance, regulatory, social factors).
- Supporting literature highlights:
  - Political constraints, corruption, civil war as major negative influences on growth and capacity building.
- Policy diagnosis approach recommended:
  - Diagnose binding constraints, design targeted policies, and institutionalize dynamic diagnosis and response (Rodrik; Hausman, Rodrik and Velasco).

### Identification methodology: economic fragility to stress
- Fragility definition and detection:
  - Fragility indicated by increasingly steeper GDP declines as shock size increases (supra-linear adverse responses).
  - Approach focuses on detecting manifested fragility (economic meltdowns) and then identifying drivers.
- Recession identification (1970–2024):
  - Recessions defined as periods of consecutive years—or a single year—of negative real GDP growth measured as a three-year moving average (3yma).
  - Measurement of steepness: (i) average cumulative decline between peak and end divided by recession years; (ii) largest annual decline during the recession.
- Fragile episode thresholds and exclusions:
  - Fragile episodes defined as recessions steeper than one standard deviation away from the mean (by average decline or fastest annual decline).
  - Recessions caused by conflicts or state breakoffs are excluded from being labeled as fragile responses.

### Aggregate statistics on recessions (1970–2024)
- Number of recessionary episodes identified: 448.
- Average duration of recessions: 2.6 years.
- Average cumulative decline in real GDP during recessions: 8.3 percent.
- Fragile responses:
  - Fragile responses accounted for about a quarter of all recessions.
  - Thresholds for fragile responses: average decline steeper than -3.5 percent during the recession OR one-year decline steeper than -5.1 percent (measured on 3yma basis).
  - Number of stress episodes identified: 106 (about 24 percent of total).
  - Of the 106 stress episodes, 52 reflected economic collapses due to conflict, state breakoffs or both; remaining 54 qualify as economically fragile responses.
- Comparative sample means (as reported):
  - All recessions: Number = 448; Length = 2.6 years; Average decline = -2.4 percent; Maximum decline = -3.7 percent.
  - Nonfragile responses: Number = 342; Length = 2.1 years; Average decline = -1.1 percent; Maximum decline = -1.5 percent.
  - Fragile responses: Number = 54; Length = 4.1 years; Average decline = -5.0 percent; Maximum decline = -7.6 percent.
  - Conflict or Breakoffs: Number = 45?52; Length = 4.4 years; Average decline = -7.5 percent; Maximum decline = -12.6 percent.
- During fragile episodes:
  - GDP losses averaged 5 percent a year, compared with 1.1 percent for non-fragile responses.
  - Average length of stress recessions: 4.1 years, versus 2.1 years for non-stress episodes.

### Recent stress recessions (2021–2024)
- Ten countries had stress recessions during 2021–2024:
  - Barbados, Fiji, Lebanon, Macao SAR, Palau, Samoa, South Sudan, St. Lucia, Suriname, Venezuela.
- Observed proximate drivers:
  - Highly concentrated activities hindering recovery from the pandemic or causing severe downturns (Barbados, Fiji, Macao SAR, Palau, Samoa, St. Lucia; South Sudan due to oil pipeline disruptions).
  - Debt-triggered crises and/or general economic mismanagement (Lebanon, Suriname, Venezuela).
- Note: Most countries showing economic fragility to stress are neither low income nor chronically fragile.

### Drivers of economic fragility (empirical and machine-learning findings)
- Categorization (percent of steepest recessions with main vulnerability due to):
  - Concentrated activity (oil, tourism): 42.6
  - Debt, mismanagement, vulnerabilities: 38.9
  - Political instability: 14.8
  - Natural disasters (drought): 3.7
- Macroeconomic and external amplifiers:
  - Conjunctural external shocks (high oil prices, higher Federal funds rate, heightened world uncertainty) increase probability of steep recession.
  - Weak external positions (large current account deficits, increases in external debt, low reserves) amplify risk.
  - Fiscal vulnerabilities (large deficits or higher public debt) amplify risk depending on debt-carrying capacity.
  - Higher incomes per capita are more strongly associated with economic recessions than lower incomes in the sample.
- Important predictors (listed in source) include:
  - Tourism expenditure (percent of GDP), natural resources rents (percent of GDP), oil prices (US$ per barrel), Federal funds rate (in percent), world uncertainty (index), current account balance (percent of GDP), one-year change in external debt (percent of GDP), reserves coverage (months of imports), fiscal balance (percent of GDP), public debt (percent of GDP), GDP per capita (US dollar), one-year change in inflation (percentage points).
- Machine learning framework:
  - Predicts probability of a steep economic recession within two to three years using macroeconomic, social and political factors (details in Annex V and Section 5).

### Conflict, political fragility, and interactions with economic fragility
- Of the 106 stress recessions, about a third were due to conflict.
- Conflict-induced recessions were steeper: average decline of 7.1 percent versus 5 percent for recessions due to economic fragility.
- Recessions associated with conflict are treated as political fragility for analysis; economic disruptions can drive political fragility and vice versa.
- Operational rules: conflicts identified via Uppsala Conflict Data Program (UCDP) for 1940–2023; where more than half of cumulative decline occurred before conflict, recession counted as economic stress (three cases cited).

### Empirical comparison: economic fragility vs. chronic fragility
- No evidence that economic fragility (frequency, steepness, duration of meltdowns) is at the root of chronic fragility in the sample.
- Average steepness of economic stress recessions: around 5 percent for both chronically fragile and non-chronically fragile countries (difference not statistically significant).
- Average duration: stress recessions last longer on average in chronically fragile countries (4.9 years) versus non-chronically fragile countries (3.9 years), but difference not statistically significant.
- Frequency: similar at about 1 stress recession per country during 1970–2024.
- Statistical testing: two-sided Wilcoxon rank-sum test (Mann–Whitney U test) used; significance levels reported as *** p < 0.001; ** p < 0.01; * p < 0.05; "not sign." = not significant.

### Political fragility: scope, data, and recent patterns
- Definitions and sources:
  - Civil wars defined as internal struggles with large loss of life; coups database: Powell and Thyne (2011) through 2024; internal conflict: Uppsala through 2023.
- Recent patterns:
  - Since the beginning of the decade, more than one in 10 countries have had a coup d’état or were affected by internal conflict.
  - Almost four in ten countries in sub-Saharan Africa have had a coup d’état or were affected by internal conflict.
  - Secular global decline in coups interrupted by surge in 2020–24: seventeen military takeovers or takeover attempts during 2020–24; all but six were in SSA.
  - Internal conflict counts hovered between 10-18 countries over the past decade with an upward trajectory in SSA.
- Countries identified with manifested political fragility (coup or internal conflict since 2020; list from source):
  - Afghanistan, Azerbaijan, Bangladesh, Bolivia, Burkina Faso, Burundi, Cameroon, Central African Republic, Chad, DR Congo, Ethiopia, Gabon, Guinea, Guinea-Bissau, Iraq, Israel, Libya, Mali, Mozambique, Myanmar, Niger, Sao Tome and Principe, Rwanda, Somalia, South Sudan, Sudan, Syria, Yemen.
- Twenty-eight countries had a coup or coup attempt or experienced internal conflict in any of the years since 2020; over half were in sub-Saharan Africa.
- Six countries experienced both coups and internal conflict: Burkina Faso, Chad, Mali, Myanmar, Niger, and Sudan.

### Drivers of political fragility (focus on coups d’état) and machine learning
- Data and sample:
  - Powell and Thyne (2011) database: 1950–2024, total of 493 coups in 97 countries.
  - Cross-database collection of 86 variables for 1970–2019 covering demographic, development, inclusion, governance, macro stability, policy, and sociopolitical stability indicators.
- Methodology:
  - Event studies to identify run-up dynamics and machine learning (random forest selected by AUC) to capture structural and conjunctural drivers and nonlinear interactions.
  - Model selection by AUC; robustness checks across regions, time, income subsamples.
- Structural drivers (most important):
  - Young demographics.
  - Weak structural fundamentals: low development, inequality, lack of inclusion, weak governance, dependence on natural resources.
  - Multiplicity of weak structural fundamentals amplifies coup risk in fragile or low-income countries.
- Conjunctural drivers (stressors):
  - Declines in growth, high inflation, weak external positions, political instability, security challenges, weaker macroeconomic policies.
  - Overlapping stressors amplify coup risk; global shocks of 2020–23 (COVID, war in Ukraine, commodity surges) increased coup probabilities via growth declines, higher inflation, and weaker external positions.
- Policy linkages:
  - Weaker fiscal positions increase risk.
  - Higher resource rents increase benefits of being in power and the likelihood of coups in many lower-income countries.
- Nonlinear interactions:
  - Structural weaknesses act as double-sided amplifiers: increase coup probability when stressors are present and reduce probability faster when stressors recede.
- Predictive performance:
  - Machine learning model trained on 1970–2019 predicted 2020–22 out-of-sample coups well.
  - Validation result: 9 out of 10 countries that experienced a coup during 2020-22 featured among the top 25 countries with highest coup probabilities estimated by the model.
  - Out-of-sample AUC for 2020-22 estimated at 0.878.

### Aggregate fragility stock and overlap across types
- Major finding: framework produces a list of 59 countries that have revealed recent fragility.
- Overlap across fragility types:
  - Only one country (South Sudan) exhibited all three types (chronic, economic, and political).
  - Thirteen countries display two types of fragility.
  - Few overlaps between chronic and economic fragility; overlaps between chronic fragility and political fragility are significant and would be larger if chronic fragility covered upper middle-income countries.
- Comparison with other classifications:
  - Of 81 countries on three lists compared, 29 are considered fragile by all three classifications.
  - Proposed list shows strong overlap (80 percent) with other fragility classifications.
  - Sixteen countries on proposed list not on other lists fall into four groups (economic stress cases, recent coups, chronic fragility, internal conflict).

### Policy implications and recommended priorities
- Overarching principle:
  - Prioritize eliminating sources of fragility (weak structural fundamentals) because they amplify stressors and increase breakdown likelihood.
  - Emphasize "via negativa" (Taleb): reduce existential risks and the risk of breaking before prioritizing efficiency or growth.
  - Follow growth-diagnostic logic: diagnose binding constraints, target policies, institutionalize dynamic diagnosis and response.
- Specific guidance:
  - Remove existential threats such as excessive concentration of activities, fiscal and external vulnerabilities, and natural disaster exposures.
  - Tailor interventions to country- and time-specific drivers (illustrated by cases like Niger and Gabon).
  - Protect vulnerable populations via strengthened social safety nets and redirect spending away from those better off.
  - In chronically fragile environments, addressing overlapping structural weaknesses helps reduce political susceptibility to destabilizing stressors.
  - Strengthening structural fundamentals can amplify positive shocks and yield higher returns in chronically fragile environments for averting political breakdowns.
- Role of multilateral institutions and donors:
  - Continued engagement likely yields high dividends because:
    - (i) Financial support can stabilize macroeconomic environment and mitigate fragile political responses.
    - (ii) Support can address sources of fragility through structural reforms.
    - (iii) Stronger policies targeted in programs reduce stressors and bring additional benefits.
- Analytical recommendation:
  - Shift from ex-post assessment of revealed fragility to identifying latent fragility for preemptive policy action, while noting latent indicators are more sensitive and often used internally.

*Source: IMF Working Papers — State Fragility: Towards a Conceptual Framework (wpiea2025205-source-pdf).*

### Annex I. Identifying Chronic Fragility: Actual vs Benchmark Growth for LICs and LMICs ....................... 31

### Annex I. Identifying Chronic Fragility: Actual vs Benchmark Growth for LICs and LMICs

### Introduction and motivation
- State fragility constrains development in about a third of the world’s countries.
- About 65 countries are considered fragile by the World Bank, IMF or OECD as of 2024.
- Fragile states are home to close to a quarter of the world population; this share is expected to grow to a third by 2050.
- Fragility manifests across economic, political and social systems and involves interacting weaknesses that constrain development.
- The paper proposes a parsimonious conceptual framework to identify state fragility via system responses to shocks, remaining agnostic for identification purposes to root causes or manifestations.

### Conceptual framework (based on Taleb and extensions)
- Fragility definition (Taleb, 2012, 2013): disproportionate (supra-linear) adverse response to shocks (stress-induced fragility).
- Extension (Bar-Yam interpretation in Taleb, 2012): lack of responsiveness (sub-linear) to positive shocks or to time (chronic fragility).
- Fragility types:
  - Chronic fragility: sub-linear response to positive shocks or to time — inability to take off or sustain per capita growth over long periods.
  - Stress-induced fragility: supra-linear adverse response to large negative shocks — breakdowns under stress (economic collapses, political breakdowns).
- Antifragility contrasted: accelerating benefit to large stressors (convex response).
- The framework operationalizes fragility by focusing on responses of economic growth (for economic fragility) and incidents of internal conflict and coups d’état (for political fragility).

### Operationalization and typology
- Economic identification:
  - Chronic fragility: countries with historical growth below the median of their peers.
  - Stress-induced economic fragility: countries with steep recessions in response to recent shocks (revealed fragility measured ex-post).
- Political identification:
  - Political fragility to stress: instances of internal conflicts and coups d’état as evidence of breakdowns in political systems.
- Both chronic and stress-induced fragility can coexist and interact.
- The simple response-based identification produces a list that overlaps strongly with existing complex classifications.

### Drivers of fragility (summary of findings)
- Drivers are rooted in both structural and conjunctural factors and interact nonlinearly.
- Economic fragility drivers:
  - High concentration of economic activity.
  - Underlying macroeconomic imbalances such as high debt, which can amplify even moderate shocks.
- Political fragility drivers:
  - Conjunctural stressors that increase coup probabilities: destabilized economic, political, and security situations.
  - Structural drivers (sources of fragility): poverty, low economic/social/political inclusion (high inequality, low literacy rates, high ethnic fractionalization, weak democratization), poor governance, and recent/higher incidence of coups.
  - Structural drivers act as double-sided amplifiers: they increase coup probability when stressors or policy deterioration are present, and they also reduce coup probability faster when stressors recede or policies improve.
  - Stressors and weak structural fundamentals compound each other’s effects; chronically fragile states with multiple structural weaknesses are more prone to political breakdowns.

### Predictive application and validation
- Machine learning methods applied to coup drivers were used to estimate coup probabilities out-of-sample for 2020-22.
  - Validation result: 9 out of 10 countries that experienced a coup during 2020-22 featured among the top 25 countries with highest coup probabilities estimated by the model.
- Drivers differ by country and can be used to identify targeted policy interventions (illustrated by country cases such as Niger and Gabon in the paper).

### Policy implications and recommended priorities
- Structural fundamentals that act as sources of fragility amplify macroeconomic stressors and increase breakdown likelihood; priority should be given to eliminating these sources when they pose existential threats or constrain growth.
- In chronically fragile states where multiple structural weaknesses overlap, payoffs from a more stable macroeconomic environment and strengthened policies are particularly high.
- Engagement of multilateral institutions and donors is likely to yield particularly high dividends in fragile situations because financial and program support can address multiple fragility drivers.
- The analysis supports transitioning from ex-post assessment of revealed fragility to identifying latent fragility for preemptive policy action.

### Illustrative numeric evidence from examples
- Hurricane damages examples cited to illustrate disproportionate shocks on small states:
  - 2017 hurricane Maria inflicted 374 percent of GDP in damages on Dominica.
  - 2004 hurricane Ivan caused 184 percent of GDP in damages in Grenada.

*Source: IMF Working Paper — State Fragility: Towards a Conceptual Framework (Annex I excerpt).*

### 3.  Chronic Fragility and its Drivers

### 3.  Chronic Fragility and its Drivers

### Definition and conceptual rationale
- Chronic fragility is defined as the inability of an economy to generate growth—“the inability of an economy to take off”—emphasizing the economy as a system rather than individual markets, agents, or institutions.
- Growth is used as a sufficient summary metric for multidimensional fragility because per capita GDP is strongly associated with measures of the basics of human material wellbeing (nutrition, access to water and sanitation, shelter, health, education, natural environment), particularly at lower levels of development (Pritchett and Lewis (2022)): “economic growth is enough, and only economic growth is enough” for the basics.
- The framework favors a simple, general definition of fragility to capture system-level dysfunction while allowing for complex, interacting causes.

### Identification criteria and methodology
- Sample restriction:
  - Only low and lower middle-income countries (LIC and LMIC) are considered for chronic fragility; upper-middle-income and high-income countries are excluded.
  - This restriction reduces the pool of potentially chronically fragile countries from 193 to 81.
- Core threshold:
  - Chronic fragility is defined relative to the 50th percentile (median) of per capita growth among LICs and LMICs.
  - Per capita growth is measured dynamically as the median per capita growth for each year (using IMF’s WEO measures of real GDP per capita in national currency).
- Persistence requirement:
  - Countries are classified as chronically fragile if their per capita growth is below the median threshold, on average, during each of the three historical periods: 1970-2023, 1980-2023, and 1990-2023.
  - The median per capita growth averaged 1.5 percent during 1970-2023 and 1980-2023 and 1.8 percent during 1990-2023.
- Exclusion for recent structural break:
  - Countries that have shown an intrinsic and sustained pickup in growth over the past ten years (2015-2024) are excluded if their growth remains above the threshold for the entire ten-year window, measured by trailing three-year moving averages (3yma).
  - The first five years of growth are excluded for countries that broke off economically (South Sudan and former Soviet Union countries); the dynamic threshold helps benchmark countries with short historical data (Afghanistan, Somalia, South Sudan).
- Rationale for thresholds:
  - A per capita growth rate of 1.5 percent is characterized as very low: a country growing at 1.5 percent a year can double its GDP per capita in about 47 years and, if starting at the LIC average income, would reach the current LMIC average income in about 78 years.

### Application and classification outcomes
- Initial below-median criterion:
  - According to the below-median growth criterion, 43 countries are considered fragile (illustrated by red lines in the source’s figures).
- Exits based on recent sustained growth:
  - Five countries are counted as having exited fragility due to sustained above-threshold growth over 2015-2024 (on a 3yma basis): Cote d’Ivoire, Djibouti, Kenya, Senegal, and Togo.
  - The exclusion criterion is strict: Cote d’Ivoire and Djibouti are the most robust exit cases, having grown above the threshold for each of the past eleven years (at least) and not just on a 3yma.
  - Senegal is noted as the least robust exit case because growth weakened significantly in 2022-23.
  - Marginal near-exit cases include Benin and Kiribati (grew on average above the threshold over the past ten years but did not sustain above-threshold growth for all ten years on a 3yma basis).
  - Additional marginal cases with strong recent performance on an 8-year 3yma basis: Guinea and Guinea-Bissau (Guinea’s strong performance attributed to oil discovery and therefore not “intrinsic”; Guinea-Bissau’s strength characterized as marginal).
- Final count:
  - Applying the exclusion criterion results in 38 countries identified as chronically fragile (assessment based on end-2024 data as reported in IMF’s April 2025 WEO).
  - Of these 38 chronically fragile countries, 17 are low income and 21 are lower middle income.
- Observations on scope:
  - Chronic fragility, as defined here, is not restricted to low-income countries; however, many advanced economies or oil-dependent economies with low recent growth are excluded because they previously achieved higher standards of living through earlier growth.
  - If the same below-median criterion were applied to all countries, about a third of upper-middle-income countries (including some oil producers) and about half of advanced economies would be added to the list.

### Drivers of chronic fragility (constraints to growth)
- The literature frames the primary drivers of chronic fragility as constraints to growth; these constraints typically interact and compound each other.
- Two broad categories of constraints:
  - Macroeconomic vulnerabilities.
  - Institutional constraints preventing markets from clearing efficiently (political, governance, regulatory, and social factors).
- Supporting findings from the literature:
  - Political constraints influence business decisions and macroeconomic outcomes, affecting growth (Henisz (2000)).
  - Corruption’s negative effect on growth is especially pronounced in autocracies, transmitting to growth via decreased FDI and increased inflation (Gründler and Potrafke (2019)).
  - Civil war is a major risk: long duration, high costs for the economy and neighbors, high risk of relapse, and reduced incentives to invest in state capacities (Collier (2007); Besley and Persson (2010)).
- Policy diagnosis approach:
  - Rodrik (2006) emphasizes the absence of universal blueprints: the recommended sequence is (i) diagnose the most significant constraints on growth in a given setting, (ii) develop creative policy design to target identified constraints, and (iii) institutionalize diagnosis and policy response to ensure dynamism.
  - Hausman, Rodrik and Velasco (2005) and Rodrik (2010) identify two broad constraint types: low returns to economic activity (low human capital, bad infrastructure, government failure) and high cost of finance (high risk, poorly functioning markets).

*Source: IMF WORKING PAPERS State Fragility: Towards a Conceptual Framework — Chapter 3*

### 4.  Economic Fragility to Stress and its Drivers

### 4.  Economic Fragility to Stress and its Drivers

### Definition and measurement of economic fragility
- Fragility is defined following Taleb’s notion as a disproportionate response to large adverse shocks: an increasingly steeper decline in GDP as the size of the shock increases indicates fragility.
- A linear (proportional) GDP decline to shock size is not fragile; increasing concavity of the GDP response is indicative of fragility.
- Detection approaches:
  - Measure likely response to a specific stressor (predict whether shock induces a fragile response). Example: Taleb et al. (2012) measure concavity of bank capital and public debt responses to growth shocks via a “fast-and-frugal, model-free, probability free heuristic”.
  - Use empirical/econometric models to detect past nonlinearities in responses to specific shocks (identification complexity can be significant; nonlinearities can arise from large shocks or amplification of smaller shocks).
- The chapter’s approach: detect already manifested fragility (economic meltdowns) and then identify drivers to uncover latent fragility.

### Identification of recession episodes and fragile responses (1970–2024)
- Recession identification:
  - Recessions defined as periods of consecutive years—or a single year—of negative real GDP growth.
  - GDP measured as a three-year moving average (3yma) to smooth temporary recoveries.
- Measurement of steepness:
  - (i) Average cumulative decline in GDP between peak and end of recession divided by the number of years in recession; or
  - (ii) Largest annual decline during the recession.
- Fragile episode definition:
  - Recessions steeper than one standard deviation away from the mean, measured either by average decline or by fastest annual decline.
  - Recessions caused by conflicts or by state breakoffs are excluded from being labeled as fragile responses.
  - Thresholds thus determined exclude conflict- or breakoff-associated recessions.

### Aggregate statistics on recessions (1970–2024)
- Number of recessionary episodes identified: 448.
- Average duration of recessions: 2.6 years.
- Average cumulative decline in real GDP during recessions: 8.3 percent.
- Fragile responses share and thresholds:
  - Fragile responses accounted for about a quarter of all recessions.
  - Thresholds for fragile responses: average decline steeper than -3.5 percent during the recession OR one-year decline steeper than -5.1 percent (measured on 3yma basis).
  - Number of stress episodes identified: 106 (about 24 percent of total).
  - Of the 106 stress episodes, 52 reflected economic collapses due to conflict, state breakoffs or both (these are not considered economic fragility for this exercise); remaining 54 qualify as economically fragile responses.
- Comparative statistics (Table 3 summary, sample means):
  - All recessions: Number = 448; Length = 2.6 years; Average decline = -2.4 percent; Maximum decline = -3.7 percent.
  - Nonfragile responses: Number = 342; Length = 2.1 years; Average decline = -1.1 percent; Maximum decline = -1.5 percent.
  - Fragile responses: Number = 54; Length = 4.1 years; Average decline = -5.0 percent; Maximum decline = -7.6 percent.
  - Conflict or Breakoffs: Number = 45?52 (table formatting indicates 45 and 52 in two columns; the narrative states 52 reflected economic collapses due to conflict/state breakoffs); Length = 4.4 years; Average decline = -7.5 percent; Maximum decline = -12.6 percent.
- During fragile episodes:
  - GDP losses averaged 5 percent a year, compared with 1.1 percent for non-fragile responses.
  - Average length of stress recessions: 4.1 years, versus 2.1 years for non-stress episodes.

### Recent manifestations: stress recessions in 2021–2024
- Using the thresholds derived from the full sample, ten countries had stress recessions during 2021–2024:
  - Barbados, Fiji, Lebanon, Macao SAR, Palau, Samoa, South Sudan, St. Lucia, Suriname, Venezuela.
- Observed proximate drivers for these downturns:
  - Highly concentrated activities that hindered recovery from the pandemic or caused severe downturns (Barbados, Fiji, Macao SAR, Palau, Samoa, St. Lucia; South Sudan due to disruptions in oil pipelines).
  - Debt-triggered crises and/or general economic mismanagement (Lebanon, Suriname, Venezuela).
- Note: Most countries showing economic fragility to stress are neither low income nor chronically fragile, supporting a distinction between chronic fragility and stress fragility.

### Drivers of economic fragility (empirical and machine-learning findings)
- Categorization of main sources of fragility (Table 4; percent of steepest recessions with main vulnerability due to):
  - Concentrated activity (oil, tourism): 42.6
  - Debt, mismanagement, vulnerabilities: 38.9
  - Political instability: 14.8
  - Natural disasters (drought): 3.7
- Machine learning exercise:
  - Uses macroeconomic, social and political factors to predict probability of a steep economic recession within two to three years (details in Annex V and Section 5).
  - Identified drivers broadly consistent with observed categorizations above.
- Key macro and external amplifiers highlighted:
  - Conjunctural external shocks: high oil prices, higher Federal funds rate, heightened world uncertainty increase probability of steep recession.
  - Amplification by weak external positions: large current account deficits, increases in external debt, low reserves.
  - Amplification by fiscal vulnerabilities: large deficits or higher public debt (up to thresholds depending on debt carrying capacity).
  - Higher incomes per capita are more strongly associated with economic recessions than lower incomes—likely because deep recessions occur in more complex or more financially developed economies.
- Specific predictors emphasized in figure discussion (contribution to probability of a steep recession, relative to sample average):
  - Tourism expenditure (percent of GDP) — top predictor in both two- and three-year exercises.
  - Natural resources rents (percent of GDP).
  - Oil prices (US$ per barrel).
  - Federal funds rate (in percent).
  - World uncertainty (index).
  - Current account balance (percent of GDP).
  - One-year change in external debt (percent of GDP).
  - Reserves coverage (months of imports).
  - Fiscal balance (percent of GDP).
  - Public debt (percent of GDP).
  - GDP per capita (US dollar).
  - One-year change in inflation (percentage points).

### Conflict, political fragility, and interactions with economic fragility
- Conflict as driver:
  - Of the 106 stress recessions, about a third were due to conflict.
  - Conflict-induced recessions were significantly steeper: average decline of 7.1 percent versus 5 percent for recessions due to economic fragility.
- Conceptual stance:
  - Conflict-induced recessions are treated as political fragility rather than economic fragility for this analysis; political fragility can directly affect the economy, and economic disruptions can in turn drive political fragility.
  - Jihadist insurgencies are counted as internal conflicts for these purposes.
- Operational exclusion criteria described:
  - Recessions associated with conflicts identified using Uppsala Conflict Data Program (UCDP) conflict intensity for 1940–2023; where conflict occurred late in a recession such that more than half of cumulative decline happened before conflict, the recession was counted as due to economic stress rather than conflict (three cases: Democratic Republic of Congo (1990–2002), Libya (1981–88), Trinidad and Tobago (1983–1990)).
  - Recessions associated with state breakoffs identified by focusing on 1989–1993 breakups in former Soviet/socialist bloc republics and South Sudan/Sudan recessions of 2011–12.

### Economic fragility and chronic fragility: empirical comparison
- Question investigated: whether economic fragility (steepness, duration, frequency of recessions) is at the root of chronic fragility (lack of convergence of lower income countries).
- Findings:
  - No evidence that economic fragility prevents countries from sustaining growth over time through more frequent or longer-lasting meltdowns (recovery differences not investigated).
  - Average steepness of economic stress recessions: around 5 percent for both chronically fragile and non-chronically fragile countries (difference not statistically significant).
  - Average duration: stress recessions last longer on average in chronically fragile countries (4.9 years) versus non-chronically fragile countries (3.9 years), but difference not statistically significant.
  - Frequency: similar at about 1 stress recession per country during 1970–2024.
- Statistical testing:
  - Significance assessed using two-sided Wilcoxon rank-sum test (Mann–Whitney U test), comparing mean steepness between chronically fragile and non-chronically fragile countries within each recession type; tests conducted separately for "economically fragile" and "conflict or breakoffs" recession groups; significance levels defined as *** p < 0.001; ** p < 0.01; * p < 0.05; "not sign." = not significant.

*IMF Working Papers — State Fragility: Towards a Conceptual Framework*

### Annex V Table 1 for a comparison of the countries in conflict under different thresholds).

### wpiea2025205-source-pdf - Annex V Table 1 for a comparison of the countries in conflict under different thresholds)

### Definitions and data sources
- Civil wars definition: struggle between the state and one or more armed groups seeking political goals and resulting in significant loss of life; includes wars for territorial independence, ideological wars, ethnic wars and often coups d’état. Operational definitions often use a sustained fatalities threshold of 500-1000 deaths.
- Coups database: Powell and Thyne (2011), annual coverage up to 2024 (de facto covering the last five years).
- Internal conflict database: Uppsala (UPPSALA) database, annual coverage up to 2023; extra-systemic conflicts (group 1) and interstate conflicts (group 2) excluded.
- World Bank and IMF current (2022-24) classification of countries fragile due to conflict:
  - (a) absolute number of conflict deaths above 250 according to ACLED and 150 according to UCDP; and
  - (b) above 2 per 100,000 population according to ACLED and above 1 per 100,000 population (i.e. 10 per million) according to UCDP;
  - or (b) countries with a rapid deterioration of the security situation, measured by:
    - (a) absolute number of conflict deaths above 250 according to ACLED and 150 according to UCDP;
    - (b) a lower number of conflict deaths relative to the population between 1 and 2 (ACLED) and 0.5 and 1 (UCDP); and
    - (c) more than a doubling of the number of casualties in the last year.

### Political fragility: scope and recent patterns
- Since the beginning of the decade:
  - More than one in 10 countries in the world have had a coup d’état or were affected by internal conflict.
  - Almost four in ten countries in sub-Saharan Africa (SSA) have had a coup d’état or were affected by internal conflict.
- Coup trend:
  - Secular decline globally was interrupted by a surge in 2020–24.
  - Of the seventeen military takeovers or takeover attempts during 2020–24, all but six were in SSA.
- Internal conflict trend:
  - Number of countries experiencing conflict hovered between 10-18 over the past decade.
  - Upward trajectory in SSA offset declines elsewhere.
  - SSA has seen a particularly large increase in fatalities since 2021 due to conflict in Ethiopia and the intensity of conflict in the Sahel.

### Countries identified as politically fragile (manifested fragility)
- Twenty-eight countries had a coup or coup attempt or had experienced internal conflict in any of the years since 2020; over half were in sub-Saharan Africa.
- Six countries experienced both coups and internal conflict: Burkina Faso, Chad, Mali, Myanmar, Niger, and Sudan.
- Countries listed in Table 5 (conflict or coups 2020-24 in the source):
  - Afghanistan, Azerbaijan, Bangladesh, Bolivia, Burkina Faso, Burundi, Cameroon, Central African Republic, Chad, DR Congo, Ethiopia, Gabon, Guinea, Guinea-Bissau, Iraq, Israel, Libya, Mali, Mozambique, Myanmar, Niger, Sao Tome and Principe, Rwanda, Somalia, South Sudan, Sudan, Syria, Yemen.
- The fragility list excludes countries with known conflicts that are external or have limited fatalities relative to population size, e.g., Mexico, Nigeria, Pakistan, Turkey, Ukraine.

### Drivers of political fragility (focus on coups d’état)
- Data and sample:
  - Powell and Thyne (2011) database: period 1950–2024, total of 493 coups in 97 countries.
  - Cross-database collection of 86 variables capturing demographic, development, inclusion, governance, macro stability, policy, and sociopolitical stability characteristics for 1970–2019.
- Methodology:
  - Event studies to identify factors with significantly different dynamics in the runup to coups.
  - Machine learning (random forest selected by AUC) to identify structural and conjunctural drivers and nonlinear interactions; AUC used to select model and tuning parameters.
  - Robustness checks across regional, time, and income subsamples.
- Structural drivers (most important):
  - Young demographics.
  - Weak structural fundamentals: low levels of development, inequality, lack of inclusion, weak governance, and dependence on natural resources.
  - Fragile or low-income countries where multiple weak structural fundamentals amplify each other’s effects are especially prone to coups.
- Conjunctural drivers (stressors):
  - Declines in growth, high inflation, weak external positions, political instability, security challenges, and weaker macroeconomic policies.
  - Overlapping stressors amplify coup risk.
  - Global shocks of 2020–23 (covid pandemic, war in Ukraine, commodity price surges) produced lower growth, higher food, energy and overall inflation, and weaker external positions for food or oil price importers—all factors increasing coup probabilities.
- Policy linkages:
  - Weaker policies—especially weaker fiscal positions—increase risk.
  - Higher resource rents (e.g., higher oil prices) increase benefits of being in power and with them the likelihood of coups in many lower-income countries.
- Nonlinear interactions:
  - Structural weaknesses act as double-sided amplifiers of stressors:
    - They increase probability of coups when stressors are present.
    - They reduce probability of coups faster when stressors recede.
  - Strengthened structural fundamentals across macroeconomic, institutional, political, and social fronts yield higher returns in structurally fragile environments for staving off political breakdowns than in structurally robust environments.

### Predictive assessment and model performance
- Machine learning model trained on 1970–2019 used to predict coups for out-of-sample period 2020–22 (a high coup incidence period).
- Model findings:
  - Assigned high probability of coups to SSA countries that experienced military takeovers during 2020-23, including Mali, Burkina Faso, Niger, Gabon, among others.
  - Predicted probabilities more reflective of cross-country structural fundamentals than time-bound stressors.
- Predictive performance:
  - Out-of-sample AUC for 2020-22 estimated at 0.878 (where AUC of 0.5 = random guessing and 1 = perfect prediction), indicating overall good model prediction performance.

*Source: Annex V Table 1 from wpiea2025205-source-pdf - Annex V Table 1 for a comparison of the countries in conflict under different thresholds).*

### 6. Taking Stock: What Countries are Fragile?

### 6. Taking Stock: What Countries are Fragile?

### Major findings on manifested fragility and the proposed fragility list
- The fragility framework produces a list of 59 countries that have revealed recent fragility.
- Overlap across fragility types:
  - Only one country (South Sudan) exhibited all three types (chronic, economic, and political).
  - Thirteen countries display two types of fragility (those in green font in Figure 18).
  - There are surprisingly few overlaps between chronic and economic fragility; overlaps between chronic fragility and political fragility are significant and would be larger if chronic fragility covered upper middle-income countries.
- Time windows used for identification:
  - Political and economic fragility to stress refer to the last four years only (2020/21-23/24).
  - Chronic fragility assessed based on 1970-2024 data.
- Manifested fragility only is reported in the list (latent fragility is discussed separately).

### Comparison with other fragility classifications
- The proposed list shows significant overlap with existing IFI and OECD fragility lists:
  - Of the 81 countries on the three lists compared, 29 are considered fragile by all three classifications (countries in red font in Figure 20).
  - The proposed list shares another 14 fragile states with either the OECD or the IMF/WB classification only.
  - The paper reports a strong overlap (80 percent) with other fragility classifications, largely by capturing chronic and political fragility.
- The 16 countries on the proposed list that do not appear on the other lists fall into four groups:
  - (a) Display economic fragility to stress (a concept not captured in the other two frameworks): Barbados, Fiji, Macao SAR, Palau, Sri Lanka, St. Lucia, Suriname.
  - (b) Had recent coups or coup attempts and are considered politically fragile in this framework: Bolivia, Gabon.
  - (c) Exhibit chronic fragility in this framework: Jordan, Malawi. Samoa, Vanuatu.
  - (d) Have experienced internal conflict: Azerbaijan, Israel, Rwanda.

### Latent fragility vs. manifested fragility
- The conceptual framework can be used to produce lists of latent fragility—where prevailing drivers signal a high probability of a stress response—rather than only realized (manifested) fragility.
- Reporting latent fragility would largely affect economically and politically fragile country lists, since chronic fragility is a longer-term concept.
- Latent fragility indicators are more sensitive to report publicly and are more likely to be used for internal purposes (e.g., IMF vulnerability exercise).
- The paper reports manifested fragility only for sensitivity reasons, not as an intrinsic feature of the framework.

### Conceptual framework and typology of fragility
- Core conceptual point: a system is fragile when it shows a supra-linear (disproportionate) response to negative stressors and a sub-linear (underwhelming) response to positive stressors or to time as a stressor (extension of Taleb 2012).
- Simple typology:
  - Fragility to stress: breakdowns in political or economic systems in response to negative shocks (measured for economic systems by growth responses and for political systems by coups d’etat and conflict).
  - Chronic fragility: inability of economic systems to generate growth across time (measured by growth relative to benchmarks).

### Drivers of fragility (manifested and latent) — empirical insights
- The paper focused on drivers behind fragility to stress using machine learning for coups d’etat and other analyses.
- Key driver patterns:
  - Conjunctural drivers (destabilized economic, political, or security situations; weaker policies) increase probability of political breakdowns.
  - Structural weaknesses (sources of fragility)—poverty, poor inclusion, weak governance, high political and security risks—act as shock amplifiers and make fragile responses more likely.
  - In economic systems, underlying fragilities (high concentration of economic activity, macroeconomic imbalances such as high debt) amplify shocks.
- Latent fragility assessment:
  - Machine learning was used to assess coup probabilities for all countries in 2021-23 based on contemporaneous drivers; the fragility ranking captured well countries that experienced coups during this period.
  - Comparable latent assessment for economic fragility and exit from chronic fragility is left for future research.

### Policy principles and recommendations
- Overarching policy principle:
  - Focus on eliminating sources of fragility (weak structural fundamentals) because they are the most significant contributors to systemic breakdowns by amplifying stressors.
  - Emphasize “via negativa” (Taleb 2012): reduce existential risks and the risk of breaking before prioritizing efficiency, success, or growth.
  - Aligns with Rodrik’s growth diagnostic: identify and address the most significant constraints in a dynamic, evolutionary setting.
- Specific guidance from driver analysis:
  - Design policies to remove existential threats such as excessive concentration of activities, fiscal and external vulnerabilities, and threats from natural disasters.
  - Tailor interventions to country- and time-specific drivers (e.g., lessons from recent coups in Niger and Gabon).
  - Protect vulnerable populations through strengthened social safety nets and reorientation of spending away from benefiting those better off.
  - In chronically fragile environments, addressing overlapping structural weaknesses helps reduce political susceptibility to destabilizing economic and policy stressors.
  - Strengthening structural fundamentals can amplify positive shocks: improvements can have higher returns in chronically fragile environments in terms of averting political breakdowns.
- Role of multilateral institutions and donors:
  - Continued engagement likely yields high dividends, particularly via structural programs, for three reasons:
    - (i) Financial support can help stabilize the macroeconomic environment and mitigate risks of fragile political responses.
    - (ii) Support can address sources of fragility through structural reforms.
    - (iii) Stronger policies targeted in programs reduce stressors and bring additional benefits.

### Annex highlights: economic fragility episodes and methodology comparisons
- Annex I (Chronic Fragility) describes the identification of chronic fragility by comparing actual growth (3-year moving average) to benchmark threshold (median) growth for LICs and LMICs; countries below threshold on average since the 1970s, 1980s, and 1990s are considered fragile unless they showed growth above threshold for each of the past ten years. Note: for South Sudan and former Soviet Union countries, the first four years after breakaway are not shown.
- Annex II (Economic Fragility to Stress) — Table II.1 lists recessions with fragile economic responses, reporting for each episode:
  - Recession begins / Recession ends / Length of recession / Cumulative decline in GDP / Average decline in GDP / Max annual decline / Chronic fragility indicator.
  - Examples from Table II.1 (exact entries preserved as in the source):
    - Antigua and Barbuda 2009 2012 4 -16.9 -4.2 -7.3
    - Argentina 2000 2003 4 -10.4 -2.6 -5.4
    - Aruba, Kingdom of Neth. 2020 2021 2 -10.1 -5.1 -8.7
    - Bahamas, The 2020 2021 2 -9.2 -4.6 -6.7
    - Barbados 2019 2021 3 -10.2 -3.4 -5.2
    - ... (Table II.1 contains many additional country recession episodes, including South Sudan 2024 2024 1 -10.1 -10.1 -10.1 and Venezuela, República Bolivariana d 2015 2022 8 -115.1 -14.4 -25.8)
  - Footnote describing steepest recessions: average or maximum declines steeper than one standard deviation away from the mean of all recessionary episodes (excluding those due to conflict or country breakoffs). Recessions identified as years of negative growth measured on a 3-year moving average basis.
- Annex III summarizes methodologies used by major institutions for fragility identification:
  - Fund For Peace / Fragile States Index (FSI): 12 indicators grouped into cohesion, social indicators, economic indicators, political, and cross-cutting; scoring 0–10 per indicator; categories based on total score with thresholds 0.0–29.9 Sustainable, 30.0–59.9 Stable, 60.0–89.9 Warning, 90.0–120.0 Alert.
  - World Bank / IMF List of Fragile and Conflict Affected States: classifies fragility by (1) high institutional and social fragility (CPIA below 3.0 for disclosed IDA countries; presence of a UN peacekeeping operation; or flight across borders of 2,000 or more per 100,000 population regarded as refugees), and (2) countries affected by violent conflict using conflict-death thresholds from ACLED and UCDP (absolute numbers and per-capita thresholds described in the source).
  - OECD / States of Fragility: multidimensional fragility across six dimensions (Economic, Environmental, Human, Political, Security, Societal) using 56 indicators; principal component analysis reduces indicators, fragility score uses first two components from each dimension; classification thresholds: High fragility score below –1.18, Extreme fragility score below –2.7.

*Source: IMF Working Paper — chapter "6. Taking Stock: What Countries are Fragile?"*

### Annex IV.   Drivers of Economic and Political

### Annex IV.   Drivers of Economic and Political Fragility: Machine Learning Methodology

### Data and sample
- Political fragility exercise:
  - Total set of 55 predictors (as described in Cebotari, et al (2024)).
  - Training sample: 192 countries during the period 1970–2019.
  - Out-of-sample testing period: 2020-2022.
  - Prediction framework uses year t-1 data to predict coup in year t.
- Economic fragility exercise:
  - Predictor set slightly modified to include additional variables to capture economic drivers such as tourism expenditures as a share of GDP.
  - Prediction framework uses year t data to predict the occurrence of a steep economic recession within the following two years (i.e., year t+1 and t+2) or the following three years (i.e., year t+1, t+2, and t+3).
- Predictors capture demographic, development, inclusion, governance, macro stability, and sociopolitical stability characteristics (including first difference transformations to capture within country dynamics).

### Machine learning algorithms: tree-based models
- Modeling approach:
  - Two tree-based supervised machine learning models employed for prediction of coups: Random Forest and XGBoost.
  - A horse race between models is conducted to select the winning model based on AUC scores.
  - Both are tree-based ensemble methods that combine many single trees and introduce sampling and randomness to mitigate overfitting common in single binary classification trees.
- Binary classification tree (Breiman et al., 1984) — core concepts:
  - Decision tree structure: root node, parent and child nodes, branches, terminal nodes (leaves).
  - Leaves represent flagged classes (determined by the class with the most votes within one leaf).
  - Observations split to left or right child nodes according to splitting rules identifying indicators and thresholds; splitting chosen based on impurity measures such as the Gini impurity index.
  - Recursive partitioning allows modelling of non-linearities, non-monotonicities, and interactions among indicators.
  - Deep trees that fit training data perfectly can overfit and fail to generalize; ensemble methods address this.
- Random Forest (Breiman, 2001):
  - Uses bootstrap aggregating (bagging): multiple binary classification trees grown on random samples selected with replacement from the training sample.
  - Performs random feature sampling: a random subset of predictors considered at each split.
  - Final class predictions by majority vote of trees; scores by averaging scores from individual trees.
  - Bagging plus feature sampling decrease variance and prevent strong correlations among trees, improving prediction performance and reducing overfitting.
- XGBoost (Chen & Guestrin, 2016):
  - Ensemble learning via gradient boosting: binary trees trained sequentially, each new tree trained to learn residuals of previous trees.
  - Adds a penalty term to the loss function to control model complexity and prevent overfitting.
  - Complexity measured by sum of depths of all trees and the number of trees in the model.
  - Depth of a tree refers to the number of splits; more splits allow capturing complex interactions but increase overfitting risk.
  - Algorithm builds new trees to minimize training errors while controlling tree complexity for generalization.

### Shapley values and interpretability
- Shapley values used to unpack model predictions and measure contributions of predictors.
  - Based on Shapley values from cooperative game theory (Shapley, 1953; Young, 1985) and adapted in Strumbelj and Kononenko, 2010; Lundberg and Lee, 2017.
  - Measure the additive contribution of each predictor to the likelihood of a coup relative to the sample-average predicted probability of a coup.
  - Employed to interpret and report predictor importance and contributions from the tree-based models.

*Source: Annex IV, "Drivers of Economic and Political Fragility: Machine Learning Methodology", IMF Working Papers.*

### References

### References (Selected works cited in the chapter)

### Theoretical foundations of growth, institutions, and development
- Bar-Yam, Y., and Epstein, I. R. 2004. “Response of complex networks to stimuli,” Proceedings of the National Academy of Sciences of the United States of America, 101(13), 4341–4345. https://doi.org/10.1073/pnas.0400673101
- Besley, T. J. and Persson, T. 2014. “The Causes and Consequences of Development Clusters: State Capacity, Peace, and Income,” Annual Review of Economics, Vol. 6, pp. 927-949, 2014, Available at SSRN: https://ssrn.com/abstract=2477767 or http://dx.doi.org/10.1146/annurev-economics-080213-041128.
- Besley, T. and Persson, T. 2010. “State Capacity, Conflict, and Development,” Econometrica 78(1), 1-34.
- Besley, T. and Persson, T. 2011. “Pillars of Prosperity: The Political Economics of Development Clusters,” Princeton University Press
- Henisz, W.J. 2003. “The Institutional Environment for Economic Growth,” Economics and Politics, Wiley Blackwell, 12(1), 1-31.
- Lucas, R.E. Jr. 1988. “On the Mechanics of Economic Development,” Journal of Monetary Economics 22: 3-42
- Romer, P.M. 1990. “Endogenous technological change,” Journal of Political Economy 98 (5): S71-S102
- Solow, R.M. 1956. “A Contribution to the Theory of Economic Growth,” Oxford Review of Economic Policy 23 (1): 3-14
- Rodrik, D. 2006. "Goodbye Washington Consensus, Hello Washington Confusion? A Review of the World Bank's Economic Growth in the 1990s: Learning from a Decade of Reform." Journal of Economic Literature, 44 (4) 973-987.
- Rodrik, D. 2010. “Diagnostics before Prescription”, Journal of Economic Perspectives—Volume 24, Number 3, Summer, 33–44.
- Pritchett, L. 1997. "Divergence, Big Time." Journal of Economic Perspectives, 11 (3): 3-17.
- Pritchett, L. 2000. “Understanding Patterns of Economic Growth: Searching for Hills among Plateaus, Mountains, and Plains”, The World Bank Economic Review, 14(2), 221–250. http://www.jstor.org/stable/3990235
- Pritchett, L., and Lewis, A. 2022. “Economic growth is enough, and only economic growth is enough.” Basics-legatum-paper_short.pdf

### Fragility, conflict, and macroeconomic costs
- Bosetti, L., A. Ivanovic, and M. Munshey. 2016. “Fragility, Risk, and Resilience: A Review of Existing Frameworks,” United Nations – Center for Policy Research, United Nations University.
- Collier, P. 2007. "Africa's Economic Growth: Opportunities and Constraints," African Development Review, African Development Bank, vol. 19(1), pages 6-25.
- Cebotari, A., E. Chueca-Montuenga, Y. Diallo, Y. Ma, R. Turk, R. W. Xin, and H. Zavarce. 2024. “Political Fragility: Coups d’État and their Drivers,” IMF Working Paper 2024/034.
- Chami, R., M. Coppo, R. Espinoza, and P. Montiel. 2021. “Macroeconomic Policy Issues in Fragile States,” In Macroeconomic Policy in Fragile States, Oxford University Press, edited by R. Chami, R. Espinoza, and P. Montiel.
- Novta, N. and Pugacheva, E. 2021. "The macroeconomic costs of conflict," Journal of Macroeconomics, Elsevier, vol. 68(C).
- Leepipatpiboon, P.; C. Castrovillari, and T. Mineyama. 2023. “Macroeconomic Shocks and Conflict”, Working Paper 23/68, International Monetary Fund.
- Ray, D. and Esteban, J. 2017. “Conflict and Development,” Annual Review of Economics, 9, Free Review Article.
- Rohner, D. and Thoenig, M. 2021. “The Elusive Peace Dividend of Development Policy: From War Traps to Macro Complementarities,” Annual Review of Economics, 13 (1), 111–131.
- Mueller, H., and Rauh, C. 2022. “The Hard Problem of Prediction for Conflict Prevention”, Journal of the European Economic Association, Volume 20, Issue 6, December 2022, Pages 2440–2467, https://doi.org/10.1093/jeea/jvac025
- Taleb, N. N., Canetti, E., Kinda, T., Loukoianova, E., and Schmieder, C. 2012. “A New Heuristic Measure of Fragility and Tail Risks: Application to Stress Testing,” IMF Working Paper, WP/12/216.

### Measurement, vulnerability, and country risk assessment
- International Monetary Fund. 2021. "How to Assess Country Risk: The Vulnerability Exercise Approach Using Machine Learning," IMF Technical Notes and Manuals 2021/003, International Monetary Fund.
- Laeven, L. and Valencia, F. 2020. "Systemic Banking Crises Database II," IMF Economic Review, Palgrave Macmillan; International Monetary Fund, vol. 68(2), 307-361, June.
- Cerra, V. and Saxena, S. 2008. "Growth Dynamics: The Myth of Economic Recovery," American Economic Review, American Economic Association, vol. 98(1), 439-457, March.
- Forni, M., Gambett, L., Maffei‐Faccioli, N. and Sala, L. 2024. "Nonlinear Transmission of Financial Shocks: Some New Evidence," Journal of Money, Credit and Banking, Blackwell Publishing, vol. 56(1), 5-33, February.
- Novta, N. and Pugacheva, E. 2021. "The macroeconomic costs of conflict," Journal of Macroeconomics, Elsevier, vol. 68(C).

### Machine learning, model interpretation, and methodology
- Breiman, L. 2001. “Random forests,” Machine Learning, 45(1): 5-32.
- Breiman, L, Friedman, J,, Stone, C.J., and Olshen, R.A. 1984. “Classification and regression trees,” CRC Press.
- Chen, T., & Guestrin, C. 2016. “Xgboost: A scalable tree boosting system,” In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining. 785–794.
- Lundberg, S. M., & Lee, S. I. 2017. “A unified approach to interpreting model predictions.” Advances in neural information processing systems, 30.
- Strumbelj, E., & Kononenko, I. 2010. “An efficient explanation of individual classifications using game theory,” The Journal of Machine Learning Research, 11, 1-18.
- Shapley, L. S. 1953. “А Value for n-Person Games,” Annals of Math Studies, (28).
- Taleb, N. N. 2011. “A Map and Simple Heuristic to Detect Fragility, Antifragility, and Model Error”, NYU-Poly working paper, SSRN.
- Taleb, N. N. 2012. Antifragile: Things That Gain from Disorder, Harlow, England: Penguin Books.
- Taleb, N. N. 2013. “Philosophy: 'Antifragility' as a mathematical idea.” Nature, Feb 28; 494(7438): 430. doi: 10.1038/494430e. PMID: 23446406.
- Strumblj, E., & Kononenko, I. 2010. “An efficient explanation of individual classifications using game theory,” The Journal of Machine Learning Research, 11, 1-18.

### Institutional economics, corruption, and planning theory
- Gründler, K. and Potrafke, N. 2019. “Corruption and economic growth: New empirical evidence,” European Journal of Political Economy, Volume 60, 1-12.
- Lloyd, P. and Lee, C. 2018. "A Review of the Recent Literature on the Institutional Economics Analysis of the Long Run Performance of Nations," Journal of Economic Surveys, Wiley Blackwell, vol. 32(1), pages 1-22, February.
- Leeson, P. T. 2020. "Logic is a harsh mistress: welfare economics for economists," Journal of Institutional Economics, Cambridge University Press, vol. 16(2), 145-150, April.
- Rittel, H. W., & Webber, M.1973. "Dilemmas in a General Theory of Planning." Policy sciences, 4(2), 155-169. https://urbanpolicy.net/wp-content/uploads/2012/11/Rittel+Webber_1973_PolicySciences4-2.pdf
- Shapley, L. S. 1953. “А Value for n-Person Games,” Annals of Math Studies, (28).
- Young, H.P. 1985. “Monotonic solutions of cooperative games.” International Journal of Game Theory, 14:65–72

*State Fragility: Towards a Conceptual Framework — Working Paper No. WP/2025/205*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025205-source-pdf.pdf_