## Annex I. (wpiea2020110-print-pdf)

## Source details

**Canonical URL:** [Annex I. (wpiea2020110-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020110-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020110-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020110-print-pdf.pdf.json)

---

### Definitions and data sources
- State-based conflict: conflict in which at least one party is the government of a state.
- Non-state conflict: conflict between two organized armed groups, neither of which is the government of a state.
- One-sided violence: the use of armed force by the government of a state or by a formally organized group against civilians.
- Data sources: Uppsala Conflict Data Program (UCDP) Georeferenced Event data set v. 19.1 and Battle-Related Deaths data set v. 19.1; IMF World Economic Outlook database; authors’ calculations.
- Conflict incidence (primary definition used in the paper): a country is coded as having experienced a conflict in a particular year if at least one hundred people were killed in battle that year for every one million people in the general population.
- Alternative incidence thresholds considered: 50 and 150 people killed per one million population.
- Conflict onset: the first year of conflict, defined as the year in which deaths exceed the chosen threshold (100, or 50 or 150) per one million population after at least four consecutive years without passing that threshold.
- Robustness check: onset defined after at least two consecutive years without passing the threshold.

### Conflict incidence statistics and sample
- Sample: 188 countries, 1989–2018 (5,640 country-year observations).
- Using the 100 deaths per one million population relative threshold:
  - 215 instances of conflict incidence (3.8 percent) in the sample.
  - 50 instances of conflict onset (0.9 percent) in the sample.
- Comparison with absolute threshold (1,000 deaths) — country-year counts and percentages (Table 1):
  - <1,000 deaths or no conflict and <100 deaths per million or no conflict: 5,240 (92.9%) country-years.
  - >1,000 deaths and <100 deaths per million or no conflict: 185 (3.3%) country-years.
  - <1,000 deaths or no conflict and >100 deaths per million: 36 (0.6%) country-years.
  - >1,000 deaths and >100 deaths per million: 179 (3.1%) country-years.
- Notes:
  - Rwanda in 1994 (0.5 million dead) is excluded in panel 1 of Figure 1.
  - In Figure 1, panel 2, a country is considered in conflict if in any year 100 (50 or 150) people or more are killed per 1 million population.

### Trends and geographic shifts
- Both the number of conflict-related deaths and the incidence of conflict based on the relative definition have risen in recent years from low levels in the early 2000s.
- Historical pattern: more interstate conflict between World War II and the 1990s; more internal civil war since the 1990s (Blattman and Miguel 2010).
- Geographic shift in major conflicts: from Sub-Saharan Africa in the 1990s to the broader Middle East region, especially since 2010.

### Share of global GDP affected by conflict
- Countries currently involved in conflict account for about 1.6 percent of global GDP (Figure 3).
- The share of global GDP affected by conflict has risen in recent years.
- Methodological note: to limit mechanical declines in the share when conflict reduces a country’s GDP, the percentage of global GDP in purchasing power parity terms that a country represents is recorded in the first year of the period (2002 or 2010) and kept constant throughout each period shown in Figure 3.
- Robustness: the conclusion that the share of global GDP affected by conflict has risen recently holds if alternative conflict definitions (50 or 150 killed per million) and other time periods (2002-2005, 2006-2009 and 2010-2016) are used.

### Methodology for estimating macroeconomic costs
- Objective: estimate macroeconomic costs of conflict on GDP per capita; components of real GDP (consumption, government spending, investment, trade in goods); sector value added; number of refugees.
- Estimation method: local projection method of Jordà (2005) and Teulings and Zubanov (2014).
- Regression specification:
  - y_{it+ℎ} − y_{it−1} = β_{1}^{ℎ} c_{it} + β_{2}^{ℎ} c_{it−1} + ∑_{j=1}^{ℎ−1} β_{3j}^{ℎ} c_{it+ℎ−j} + θ_{1}^{ℎ} ∆y_{it−1} + θ_{2}^{ℎ} x_{it} + μ_{i}^{ℎ} + γ_{t}^{ℎ} + ε_{it}^{ℎ},  h = 0,...,10
- Variable definitions and controls:
  - y_it is log GDP per capita; y_{it+ℎ} − y_{it−1} indicates cumulative growth between horizons t–1 and h.
  - c_it is the conflict variable.
  - x_it is a vector of controls (lagged log population; and, where dependent variable is number of refugees, lagged log GDP per capita).
  - μ_{i}^{ℎ} are country fixed effects; γ_{t}^{ℎ} are time fixed effects.
  - When controlling for pre-conflict GDP forecast, x_it includes y_{it+ℎ}^{forecast} − y_{it−1}, where y_{it+ℎ}^{forecast} is the IMF forecast of log GDP per capita in country i, h years ahead (WEO forecasts; forecasts used are the Spring WEO vintage of the same year).
- Technical note: the ∑_{j=1}^{ℎ−1} β_{3j}^{ℎ} c_{it+ℎ−j} term mitigates attenuation bias from omission of the shock variable between periods t and t+h.

### Effect on GDP per capita (key quantitative findings)
- Estimated immediate (contemporaneous) decline in GDP per capita at conflict onset: about 15 percent, on average (Table 2, column 2 and Figure 4, panel 1).
- Cumulative effect on per capita GDP is persistently negative:
  - Ten years after conflict onset, the cumulative reduction in the level of GDP is about 28 percent relative to the pre-conflict year.
- Robustness to pre-conflict expectations:
  - When controlling for pre-conflict GDP forecast (using WEO forecasts), estimated effects over the first five years are broadly similar to baseline estimates.
- Severity matters: estimated effect of conflict onset varies with conflict severity (less or more severe conflict episodes explored).

### Channels, spillovers, and broader considerations
- Three reasons conflicts have weighed on global GDP growth in recent years:
  - 1) the number of economies experiencing strife has been increasing;
  - 2) some conflict episodes cause very severe negative effects on economic activity, leading to true economic collapses;
  - 3) some affected economies are large (examples noted: Iraq, Nigeria, Ukraine in some years).
- Macroeconomic costs are large and long-lasting for affected countries and can spill over to neighboring and non-neighboring countries via trade, refugee flows, and other channels; spillovers can continue to have effects over ten years or more following conflict onset.

---

### 2. Controlling for forecast

### Definitions and robustness checks
- Conflict onset definition: indicator = 1 in the first year of conflict after at least four consecutive years without passing the threshold of 100 killed per 1 million population.
- Alternative severity thresholds tested: 50, 100, and 150 conflict-related deaths per 1 million population.
- Alternative absolute threshold tested: 1,000 battle deaths.
- Alternative onset persistence tested: onset identified after at least two consecutive years without passing the threshold (in addition to four years).
- Key robustness findings:
  - Estimates are consistent across thresholds (50, 100, 150 killed per million), with the smallest estimated magnitude for the 50 killed per million definition.
  - The most severe conflicts (>150 killed per million) show the largest negative impacts.
  - The 1,000-death absolute threshold produces results most similar to the 50 deaths per million indicator.
  - Results remain "very strong negative and statistically significant" when switching to the two-year no-conflict onset definition.

### Main GDP impact results (selected estimates)
- Severe conflicts (>150 killed per million):
  - approximately 16 percent lower GDP per capita in the year of conflict onset,
  - about 34 percent lower 5 years on,
  - 38 percent lower ten years on.
- Aggregate (Table 2, conflict onset after 4 years of no conflict; selected coefficients shown):
  - t = 0: -12.446***; -14.888***; -16.358***; -11.650**; -10.591***; -13.716***; -14.730**; -9.742**
  - t = 1: -12.552***; -14.221***; -15.038***; -12.311***; -11.057***; -13.068***; -13.566***; -9.753***
  - t = 5: -25.940***; -27.048***; -34.034***; -22.748***; -20.452***; -21.246***; -27.162***; -15.543***
  - t = 10: -28.619***; -28.101***; -38.117***; -21.994***; -23.726***; -21.240***; -31.531***; -15.604***
- Controlling for forecast (Table 3, conflict onset after 4 years of no conflict; selected coefficients):
  - t = 0: -9.158**; -11.048***; -15.974***; -9.769**
  - t = 1: -7.650**; -8.324***; -14.146***; -9.790***
  - t = 5: -20.704**; -20.780**; -32.175***; -16.916*
- Regression controls and inference:
  - All regressions include controls for lagged GDP per capita growth and lagged log population, country and year fixed effects.
  - Standard errors clustered at the country level.
  - NA = not applicable. * p<0.1; ** p<0.05; *** p<0.01.

### Sub-samples and additional controls (Table 4)
- Sub-Saharan Africa (SSA) (100 deaths per million threshold):
  - GDP per capita declines by 8 percent on impact,
  - 25 percent cumulatively ten years after the shock.
- MENAP (Middle East, North Africa, Afghanistan and Pakistan):
  - No statistically significant effect detected (attributed to smaller sample size and data availability limitations).
- Commodity terms of trade (CTOT) control:
  - Including a commodities terms of trade control shows overall findings remain unchanged.
- Fragile vs. non-fragile states:
  - Negative and statistically significant effects for fragile states on impact and ten years after the shock.
  - For non-fragile states, negative and significant on impact but statistically indistinguishable from zero over longer horizons.
  - Fragility index is time-variant and limited in availability.

### Impact on expenditure components, trade, and sector value added (Tables 5 and Figure 5; 100 deaths per million threshold)
- Consumption:
  - Total consumption reduced by about 6 percent in the same year (t = 0).
  - Cumulative 12 percent one year later.
  - Cumulative 22 percent ten years after conflict onset.
  - Decline driven by private consumption rather than public spending.
- Trade in goods:
  - Exports: reduction of 26 percent in the same year; cumulative 35 percent five years later; cumulative 58 percent ten years after conflict onset.
  - Imports: reduction of about 17 percent in the same year; cumulative 23 percent five years later; cumulative 34 percent ten years after conflict onset.
  - Caveat: trade reductions may partly reflect rises in smuggling and underreporting of official trade statistics.
- Investment:
  - Evidence does not definitively show persistent declines in investment beyond a contemporaneous negative effect in the year of conflict onset; estimates are imprecise.
  - Data limitations and measurement issues in conflict-affected countries; separation of private and public investment not feasible.
- Sectoral value added:
  - Agriculture value added: decreases by 6 percent in the year of conflict onset; effect remains negative and statistically significant for the next six years.
  - Manufacturing: declines by 16 percent in the year of conflict onset and by 38 percent cumulatively ten years after the shock.
  - Services: decline by 12 percent on impact and by 30 percent ten years after the shock.
- Mechanisms and longer-term channels:
  - Conflict may reduce workforce, hamper labor productivity, reduce physical and mental health, induce refugee flows and migration, and prevent schooling—lowering human capital and causing large medium- and long-term negative effects.

### Effect on migration
- Global refugee context:
  - Number of refugees stood at around 20 million people in 2018, about 30 percent of which were refugees from Syria.
  - Vast majority of refugees go to neighboring countries; a small fraction go to advanced economies.
- Historical magnitude:
  - In 1995 about 15 percent of the population in conflict countries left as refugees.
- Regression findings on migration:
  - Refugees to neighboring countries: typically close to 2 percent, cumulatively, of the conflict country’s population 4 to 5 years following conflict onset.
  - The cumulative log difference of refugees to neighboring countries is mostly stable over time, indicating refugees tend to stay in neighboring countries for ten years or more after conflict onset.
  - Refugee flows to advanced economies are persistent and grow over time following conflict onset.
- Measurement and controls:
  - Migration outcomes analyzed as refugees in terms of conflict country’s population and as log differences to proxy growth over time.
  - Regressions control for lagged log GDP per capita and log population.

---

### 1. Number of refugees

### Patterns, data, and definitions
- Conflict incidence baseline: 100 killed per 1 million population. Alternatives: 50 and 150 per 1 million; traditional benchmark: 1,000 battle-related deaths per year.
- Refugee counts: UNHCR Population Statistics Reference Database. When UNHCR redacts figures ranging between 1 and 4, a value of 4 is assumed. Data where country of origin is "Various/Unknown" is dropped.
- Countries of asylum grouped by income (Advanced Economies and Emerging Market and Developing Economies per WEO April 2019) and by proximity (neighboring versus not neighboring per CEPII GeoDist).
- Conflict onset: 1 in first year of conflict after at least four consecutive years without exceeding threshold.

### Refugee dynamics and empirical findings
- Initial movements: most vulnerable move internally or to neighboring (non-advanced) countries; over time flows to advanced economies may rise.
- Bosnia example:
  - In 1992 more than 8 percent of Bosnia’s population fled as refugees to neighboring countries.
  - In 1992 about 1.7 percent of the Bosnian population sought refuge in an advanced economy.
  - By 1994 the number of Bosnian refugees in advanced economies exceeded that in neighboring countries.
- In most conflict-affected countries, the proportion obtaining refugee status in an advanced country is "much less than 1 percent" of the population.
- When dependent variable is refugees as percent of population of the conflict country, contemporaneous effect on refugees to advanced countries is generally not statistically significant.
- Cumulative log difference outcomes:
  - Growth of refugees in advanced economies is extremely persistent and continues even ten years after conflict onset.
  - Large refugee outflows and low return rates over the next ten years imply significant loss of present and future labor force.

### Select regression coefficients from Table 6 (Impact of Conflict Onset on the Number of Refugees)
- Reported coefficients (years after the shock):
  - t = 0: 0.433*** -0.013 0.446*** 0.455*** 1.175** -0.093 2.025*** 2.231***
  - t = 1: 0.811** -0.017 0.822** 0.912*** 1.784*** 0.342 2.595*** 2.733***
  - t = 2: 1.028** -0.020 1.043** 1.111*** 1.861*** 0.536** 2.545*** 2.524***
  - t = 3: 1.312*** 0.001 1.308*** 1.362*** 1.890*** 0.645*** 2.692*** 2.492***
  - t = 4: 1.899*** 0.084 1.810*** 1.830*** 2.210*** 0.800*** 3.291*** 2.857***
  - t = 5: 1.920*** 0.114 1.803*** 1.814*** 2.226*** 0.945*** 3.241*** 2.811***
  - t = 6: 1.831*** 0.064 1.763** 1.808*** 2.341*** 1.049*** 3.479*** 3.120***
  - t = 7: 1.786** 0.041 1.740** 1.927*** 2.251*** 1.056*** 3.322*** 2.900***
  - t = 8: 1.926** 0.029 1.896** 2.122*** 2.303*** 1.082*** 3.428*** 3.106***
  - t = 9: 1.989** 0.021 1.966** 2.225*** 2.382*** 1.181*** 3.457*** 3.169***
  - t = 10: 1.888* 0.035 1.857* 2.141*** 2.414*** 1.331*** 3.527*** 3.095***
- Table 6 notes: dependent variable is either cumulative change in total number of refugees as percent of population or cumulative log difference of total number of refugees; regressions include controls for lagged dependent variable growth, lagged log GDP per capita in U.S. dollars, lagged log population, country and year fixed effects; standard errors clustered at the country level.

### Macroeconomic implications
- Large refugee outflows and persistently lower populations imply long-lasting reductions in the labor force and productive capacity of conflict countries.
- Conflict can have spillover effects on other countries via refugee flows, trade disruptions, and other channels.
- Even though conflict directly affects a relatively small number of countries, measurable effects on global GDP warrant attention, especially given increases in conflict episodes in some large economies and regions like Sub-Saharan Africa and Middle East and North Africa.

### Policy implications and recommendations
- Reach ordinary people in conflict-affected countries—whose consumption falls and sometimes collapses for many years following conflict episodes—by extending public goods provision and social safety nets.
- Macroeconomic stabilization policies for conflict-affected fragile states often differ from those for countries hit by conventional severe economic shocks because trust in government institutions and delivery capacity are fundamentally eroded.
- The paper references approaches for adjusting macroeconomic stabilization policies to meet the specific challenges facing fragile states and for building inclusive approaches to recover from conflict and prevent future eruptions.

---

### Annex Table 1. Data Sources (indicator-level)
- Conflict-related deaths: UCDP Georeferenced Event Dataset v. 19.1; UCDP Battle-Related Deaths Dataset v. 19.1.
- Population: IMF WEO Database; United Nations, World Population Prospects, 2017 revision; World Bank, WDI Database.
- Number of refugees: UNHCR Population Statistics Reference Database.
- Real GDP, GDP per capita, consumption components, investment, imports and exports trade value: IMF WEO Database (vintages 1990-2019).
- Agriculture, manufacturing, services real value added: World Bank, WDI Database.
- Commodity export price index: Gruss and Kebhaj (2019).
- State Fragility Index (indicator for scores above 11): Center for Systemic Peace.
- Neighboring countries: CEPII GeoDist Dataset.

### Annex Table 2. Country Sample (incidence and onset lists)
- Incidence (100 killed per one million population): Afghanistan (1989-2001, 2006-2018); Algeria (1995, 1998); Angola (1989, 1990, 1992-1994, 1999, 2001); Azerbaijan (1992-1994); Bosnia and Herzegovina (1992-1995); Burundi (1995-2003); Central African Republic (2013-2018); Chad (1990, 1992, 2000, 2006, 2008); Comoros (1997); Congo, Democratic Republic of the (1996-1999, 2002); Congo, Republic of (1997-1999); Croatia (1991, 1995); Djibouti (1990-1992); El Salvador (1989, 1990); Eritrea (1998-2000); Ethiopia (1989-1991, 1999, 2000); Georgia (1992, 1993, 2008); Ghana (1994); Guinea-Bissau (1998, 1999); Iraq (1991, 1997, 2003-2007, 2013-2017); Israel (2002, 2008, 2014); Kuwait (1990, 1991); Lebanon (1989, 1990, 2006); Liberia (1990-1996, 2001-2003); Libya (2011, 2014-2018); Moldova (1992); Mozambique (1989-1991); Nepal (2002); Nicaragua (1989); Panama (1989); Rwanda (1990-1995, 1997, 1998, 2001); Serbia (1998, 1999); Sierra Leone (1991-2000); Somalia (1989-1992, 1996, 2006-2012, 2016-2018); South Sudan (2011, 2013, 2014); Sri Lanka (1990-1993, 1995-2000, 2006-2009); Sudan (1989-1993, 1997, 1998, 2000-2004); Syria (2012-2018); Tajikistan (1992, 1993, 1996, 1998); Ukraine (2014); Yemen (1994, 2015, 2016, 2018).
- Onset (first year exceeding 100 per million after at least four consecutive no-conflict years): Afghanistan (1989, 2006); Algeria (1995); Angola (1999); Azerbaijan (1992); Bosnia and Herzegovina (1992); Burundi (1995); Central African Republic (2013); Chad (1990, 2000, 2006); Comoros (1997); Congo, Democratic Republic of the (1996); Congo, Republic of the (1997); Croatia (1991); Djibouti (1990); Eritrea (1998); Ethiopia (1999); Georgia (1992, 2008); Ghana (1994); Guinea-Bissau (1998); Iraq (1991, 1997, 2003, 2013); Israel (2002, 2008, 2014); Kuwait (1990); Lebanon (1989, 2006); Liberia (1990, 2001); Libya (2011); Moldova (1992); Nepal (2002); Panama (1989); Rwanda (1990); Serbia (1998); Sierra Leone (1991); Somalia (1989, 2006); South Sudan (2011); Sri Lanka (2006); Syria (2012); Tajikistan (1992); Ukraine (2014); Yemen (1994, 2015).
- Onset (after at least two consecutive no-conflict years): Afghanistan (1989, 2006); Algeria (1995, 1998); Angola (1999); Azerbaijan (1992); Bosnia and Herzegovina (1992); Burundi (1995); Central African Republic (2013); Chad (1990, 2000, 2006); Comoros (1997); Congo, Democratic Republic of the (1996, 2002); Congo, Republic of the (1997); Croatia (1991, 1995); Djibouti (1990); Eritrea (1998); Ethiopia (1999); Georgia (1992, 2008); Ghana (1994); Guinea-Bissau (1998); Iraq (1991, 1997, 2003, 2013); Israel (2002, 2008, 2014); Kuwait (1990); Lebanon (1989, 2006); Liberia (1990, 2001); Libya (2011, 2014); Moldova (1992); Nepal (2002); Panama (1989); Rwanda (1990, 2001); Serbia (1998); Sierra Leone (1991); Somalia (1989, 1996, 2006, 2016); South Sudan (2011); Sri Lanka (2006); Sudan (1997); Syria (2012); Tajikistan (1992, 1996); Ukraine (2014); Yemen (1994, 2015).

*Source: wpiea2020110-print-pdf - Annex I (https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020110-print-pdf.pdf)*

### Annex I.

### Annex I

### Definitions and data sources
- State-based conflict: conflict in which at least one party is the government of a state.  
- Non-state conflict: conflict between two organized armed groups, neither of which is the government of a state.  
- One-sided violence: the use of armed force by the government of a state or by a formally organized group against civilians.  
- Data sources: Uppsala Conflict Data Program (UCDP) Georeferenced Event data set v. 19.1 and Battle-Related Deaths data set v. 19.1; IMF World Economic Outlook database; authors’ calculations.  
- Conflict incidence (primary definition used in the paper): a country is coded as having experienced a conflict in a particular year if at least one hundred people were killed in battle that year for every one million people in the general population.  
- Alternative incidence thresholds considered: 50 and 150 people killed per one million population.  
- Conflict onset: the first year of conflict, defined as the year in which deaths exceed the chosen threshold (100, or 50 or 150) per one million population after at least four consecutive years without passing that threshold. Robustness check: onset defined after at least two consecutive years without passing the threshold.

### Conflict incidence statistics and sample
- Sample: 188 countries, 1989–2018 (5,640 country-year observations).  
- Using the 100 deaths per one million population relative threshold: 215 instances of conflict incidence (3.8 percent) and 50 instances of conflict onset (0.9 percent) in the sample.  
- Comparison with absolute threshold (1,000 deaths): Table 1 tabulates country-years by absolute threshold (<1,000 deaths or no conflict; >1,000 deaths) versus relative threshold (<100 deaths per million or no conflict; >100 deaths per million). Exact counts and percentages from Table 1:  
  - <1,000 deaths or no conflict and <100 deaths per million or no conflict: 5,240 (92.9%) country-years.  
  - >1,000 deaths and <100 deaths per million or no conflict: 185 (3.3%) country-years.  
  - <1,000 deaths or no conflict and >100 deaths per million: 36 (0.6%) country-years.  
  - >1,000 deaths and >100 deaths per million: 179 (3.1%) country-years.  
- Note: In panel 1 of Figure 1, Rwanda in 1994 (0.5 million dead) is excluded.  
- Note on country-year conflict classification in Figure 1, panel 2: a country is considered in conflict if in any year 100 (50 or 150) people or more are killed per 1 million population.

### Trends and geographic shifts
- Both the number of conflict-related deaths and the incidence of conflict based on the relative definition have risen in recent years from low levels in the early 2000s.  
- Historical pattern: more interstate conflict between World War II and the 1990s; more internal civil war since the 1990s (Blattman and Miguel 2010).  
- Geographic shift in major conflicts: from Sub-Saharan Africa in the 1990s to the broader Middle East region, especially since 2010.

### Share of global GDP affected by conflict
- Countries currently involved in conflict account for about 1.6 percent of global GDP (Figure 3).  
- The share of global GDP affected by conflict has risen in recent years.  
- Methodological note: to limit mechanical declines in the share when conflict reduces a country’s GDP, the percentage of global GDP in purchasing power parity terms that a country represents is recorded in the first year of the period (2002 or 2010) and kept constant throughout each period shown in Figure 3.  
- The conclusion that the share of global GDP affected by conflict has risen recently holds if alternative conflict definitions (50 or 150 killed per million) and other time periods (2002-2005, 2006-2009 and 2010-2016) are used.

### Methodology for estimating macroeconomic costs
- Objective: estimate macroeconomic costs of conflict on GDP per capita; components of real GDP (consumption, government spending, investment, trade in goods); sector value added; number of refugees.  
- Estimation method: local projection method of Jordà (2005) and Teulings and Zubanov (2014).  
- Regression specification as presented in the source:
  푦_{it+ℎ} − 푦_{it−1} = 훽_{1}^{ℎ} c_{it} + 훽_{2}^{ℎ} c_{it−1} + ∑_{j=1}^{ℎ−1} 훽_{3j}^{ℎ} c_{it+ℎ−j} + 휃_{1}^{ℎ} ∆푦_{it−1} + 휃_{2}^{ℎ} x_{it} + 휇_{i}^{ℎ} + 훾_{t}^{ℎ} + 휀_{it}^{ℎ},  h = 0,...,10
- Variable definitions and controls:  
  - y_it is log GDP per capita; 푦_{it+ℎ} − 푦_{it−1} indicates cumulative growth between horizons t–1 and h.  
  - c_it is the conflict variable.  
  - x_it is a vector of controls (lagged log population; and, for regressions where the dependent variable is the number of refugees, lagged log GDP per capita).  
  - 휇_{i}^{ℎ} are country fixed effects; 훾_{t}^{ℎ} are time fixed effects.  
  - In regressions controlling for pre-conflict GDP forecast, x_it includes 푦_{it+ℎ}^{forecast} − 푦_{it−1}, where 푦_{it+ℎ}^{forecast} is the IMF forecast of log GDP per capita in country i, h years ahead (WEO forecasts; forecasts used are the Spring WEO vintage of the same year).  
- Technical note: following Teulings and Zubanov (2014), the ∑_{j=1}^{ℎ−1} 훽_{3j}^{ℎ} c_{it+ℎ−j} term is included to mitigate attenuation bias from omission of the shock variable between periods t and t+h.

### Effect on GDP per capita (key quantitative findings)
- Estimated immediate (contemporaneous) decline in GDP per capita at conflict onset: about 15 percent, on average (Table 2, column 2 and Figure 4, panel 1).  
- Cumulative effect on per capita GDP is persistently negative. Ten years after conflict onset, the cumulative reduction in the level of GDP is about 28 percent relative to the pre-conflict year.  
- Robustness to pre-conflict expectations: when controlling for pre-conflict GDP forecast (using WEO forecasts), estimated effects over the first five years are broadly similar to baseline estimates, indicating expectations about GDP are not an important source of bias.  
- Severity matters: the estimated effect of conflict onset varies with conflict severity (less or more severe conflict episodes are explored; details in the source).

### Channels, spillovers, and broader considerations
- Three reasons conflicts have weighed on global GDP growth in recent years:  
  - 1) the number of economies experiencing strife has been increasing;  
  - 2) some conflict episodes cause very severe negative effects on economic activity, leading to true economic collapses;  
  - 3) some affected economies are large (examples noted: Iraq, Nigeria, Ukraine in some years).  
- Macroeconomic costs are large and long-lasting for affected countries and can spill over to neighboring and non-neighboring countries via trade, refugee flows, and other channels; spillovers can continue to have effects over ten years or more following conflict onset.

*Source: wpiea2020110-print-pdf - Annex I (https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020110-print-pdf.pdf)*

### 2. Controlling for forecast

### 2. Controlling for forecast

### Definitions and robustness checks
- Conflict onset definition: indicator = 1 in the first year of conflict after at least four consecutive years without passing the threshold of 100 killed per 1 million population.
- Alternative severity thresholds tested: 50, 100, and 150 conflict-related deaths per 1 million population.
- Alternative absolute threshold tested: 1,000 battle deaths.
- Alternative onset persistence tested: conflict onset identified after at least two consecutive years without passing the threshold (in addition to four years).
- Key robustness findings:
  - Estimates are consistent across thresholds (50, 100, 150 killed per million), with the smallest estimated magnitude for the 50 killed per million definition.
  - The most severe conflicts (>150 killed per million) show the largest negative impacts.
  - The 1,000-death absolute threshold produces results most similar to the 50 deaths per million indicator, suggesting the traditional 1,000-death definition may understate the negative effects by focusing on lower-intensity conflicts.
  - Results remain "very strong negative and statistically significant" when switching to the two-year no-conflict onset definition.

### Main GDP impact results (selected estimates)
- Severe conflicts (>150 killed per million):
  - approximately 16 percent lower GDP per capita in the year of conflict onset,
  - about 34 percent lower 5 years on,
  - 38 percent lower ten years on.
- Aggregate (Table 2, conflict onset after 4 years of no conflict; selected coefficients shown):
  - t = 0 (contemporaneous): -12.446***; -14.888***; -16.358***; -11.650**; -10.591***; -13.716***; -14.730**; -9.742**
  - t = 1: -12.552***; -14.221***; -15.038***; -12.311***; -11.057***; -13.068***; -13.566***; -9.753***
  - t = 5: -25.940***; -27.048***; -34.034***; -22.748***; -20.452***; -21.246***; -27.162***; -15.543***
  - t = 10: -28.619***; -28.101***; -38.117***; -21.994***; -23.726***; -21.240***; -31.531***; -15.604***
- Controlling for forecast (Table 3, conflict onset after 4 years of no conflict; selected coefficients):
  - t = 0: -9.158**; -11.048***; -15.974***; -9.769**
  - t = 1: -7.650**; -8.324***; -14.146***; -9.790***
  - t = 5: -20.704**; -20.780**; -32.175***; -16.916*
- Note: All regressions include controls for lagged GDP per capita growth and lagged log population, country and year fixed effects. Standard errors clustered at the country level. NA = not applicable. * p<0.1; ** p<0.05; *** p<0.01.

### Sub-samples and additional controls (Table 4)
- Sub-Saharan Africa (SSA) (conflict onset based on 100 deaths per million):
  - GDP per capita declines by 8 percent on impact,
  - 25 percent cumulatively ten years after the shock.
- MENAP (Middle East, North Africa, Afghanistan and Pakistan):
  - No statistically significant effect detected (attributed to smaller sample size and data availability limitations, especially for severely affected countries such as Syria).
- Commodity terms of trade (CTOT) control:
  - Regressions including a commodities terms of trade control (price fluctuations weighted by export shares) show overall findings remaining unchanged.
- Fragile vs. non-fragile states:
  - Conflict effect is negative and statistically significant for fragile states both on impact and ten years after the shock.
  - For non-fragile states, the effect is negative and significant on impact but becomes statistically indistinguishable from zero over longer horizons.
  - Fragility index is time-variant and limited in availability, constraining further exploration.

### Impact on expenditure components, trade, and sector value added (Tables 5 and Figure 5; using 100 deaths per million threshold)
- Consumption:
  - Total consumption reduced by about 6 percent in the same year (t = 0),
  - cumulative 12 percent one year later,
  - cumulative 22 percent ten years after conflict onset.
  - Decline driven by private consumption rather than public spending.
- Trade in goods:
  - Exports: reduction of 26 percent in the same year; cumulative 35 percent five years later; cumulative 58 percent ten years after conflict onset.
  - Imports: reduction of about 17 percent in the same year; cumulative 23 percent five years later; cumulative 34 percent ten years after conflict onset.
  - Caveat: trade reductions may partly reflect rises in smuggling and underreporting of official trade statistics.
- Investment:
  - Evidence does not definitively show persistent declines in investment beyond a contemporaneous negative effect in the year of conflict onset; estimates are imprecise.
  - Data limitations and measurement issues in conflict-affected countries noted; separation of private and public investment not feasible in sample.
- Sectoral value added:
  - Agriculture value added: decreases by 6 percent in the year of conflict onset; effect remains negative and statistically significant for the next six years.
  - Manufacturing: declines by 16 percent in the year of conflict onset and by 38 percent cumulatively ten years after the shock.
  - Services: decline by 12 percent on impact and by 30 percent ten years after the shock.
- Mechanisms and longer-term channels:
  - Conflict may reduce workforce, hamper labor productivity, reduce physical and mental health, induce refugee flows and migration, and prevent schooling—lowering human capital and causing large medium- and long-term negative effects.

### Effect on migration
- Global refugee context:
  - Number of refugees stood at around 20 million people in 2018,
  - about 30 percent of which were refugees from Syria.
  - Vast majority of refugees go to neighboring countries; a small fraction go to advanced economies.
- Historical magnitude:
  - In 1995 about 15 percent of the population in conflict countries left as refugees.
- Regression findings on migration:
  - Refugees to neighboring countries: typically close to 2 percent, cumulatively, of the conflict country’s population 4 to 5 years following conflict onset.
  - The cumulative log difference of refugees to neighboring countries is mostly stable over time, indicating refugees tend to stay in neighboring countries for ten years or more after conflict onset.
  - Refugee flows to advanced economies are persistent and grow over time following conflict onset.
- Measurement notes:
  - Migration outcomes analyzed as refugees in terms of conflict country’s population and as log differences to proxy growth over time.
  - Regressions control for lagged log GDP per capita and log population.

*Source: Authors’ calculations (from "2. Controlling for forecast").*

### 1. Number of refugees

### 1. Number of refugees

### Patterns and definitions
- Conflict incidence is defined as 100 killed per 1 million population. Alternative thresholds considered include 50 and 150 people killed per 1 million, and the traditional literature benchmark of 1,000 battle-related deaths per year.
- Refugee counts are from the United Nations High Commissioner for Refugees’ (UNHCR) Population Statistics Reference Database. When UNHCR redacts figures ranging between 1 and 4, a value of 4 is assumed. Data where country of origin is "Various/Unknown" is dropped.
- Countries of asylum are grouped by income into Advanced Economies and Emerging Market and Developing Economies (based on the WEO definition as of April 2019), and by proximity into neighboring countries (share a land border) and not neighboring countries (CEPII GeoDist, updated).
- Conflict onset takes the value of 1 in the first year of conflict after at least four consecutive years without exceeding the conflict threshold.

### Refugee dynamics and empirical findings
- At conflict onset, the most vulnerable populations tend to move internally or to neighboring (non-advanced) countries; over time flows to advanced economies may rise as people search for better opportunities.
- Case example (Bosnia):
  - In 1992 more than 8 percent of Bosnia’s population fled as refugees to neighboring countries.
  - In 1992 about 1.7 percent of the Bosnian population sought refuge in an advanced economy.
  - By 1994 the number of Bosnian refugees in advanced economies exceeded that in neighboring countries.
- In most conflict-affected countries, only a tiny proportion of people manage to obtain refugee status in an advanced country, "much less than 1 percent" of the population.
- When the dependent variable is refugees as percent of population of the conflict country, the contemporaneous effect of conflict onset on refugees to advanced countries is generally not statistically significant (Figure 7, panel 3).
- Focusing on the cumulative log difference of refugees over time shows that the growth of refugees in advanced economies is extremely persistent and continues even ten years after conflict onset (Figure 7, panel 4). This suggests initial obstacles to obtaining refugee status in advanced economies, with continued efforts becoming more fruitful over time.
- Large refugee outflows and low return rates over the next ten years imply a significant loss of present and future labor force, negatively affecting productive capacity well after conflict ends.

### Key quantitative points reported in the paper
- Example: "GDP per capita is about 28 percent lower ten years after conflict onset."
- Bosnia example: "more than 8 percent" (to neighboring countries in 1992) and "about 1.7 percent" (to advanced economies in 1992).
- Persistence statement: refugee growth in advanced economies "continues even ten years after conflict onset."
- Conflict incidence threshold used in baseline: "one hundred per one million population."

### Select regression coefficients from Table 6 (Impact of Conflict Onset on the Number of Refugees)
- For years after the shock, reported coefficient values include (as presented):
  - t = 0: 0.433*** -0.013 0.446*** 0.455*** 1.175** -0.093 2.025*** 2.231***
  - t = 1: 0.811** -0.017 0.822** 0.912*** 1.784*** 0.342 2.595*** 2.733***
  - t = 2: 1.028** -0.020 1.043** 1.111*** 1.861*** 0.536** 2.545*** 2.524***
  - t = 3: 1.312*** 0.001 1.308*** 1.362*** 1.890*** 0.645*** 2.692*** 2.492***
  - t = 4: 1.899*** 0.084 1.810*** 1.830*** 2.210*** 0.800*** 3.291*** 2.857***
  - t = 5: 1.920*** 0.114 1.803*** 1.814*** 2.226*** 0.945*** 3.241*** 2.811***
  - t = 6: 1.831*** 0.064 1.763** 1.808*** 2.341*** 1.049*** 3.479*** 3.120***
  - t = 7: 1.786** 0.041 1.740** 1.927*** 2.251*** 1.056*** 3.322*** 2.900***
  - t = 8: 1.926** 0.029 1.896** 2.122*** 2.303*** 1.082*** 3.428*** 3.106***
  - t = 9: 1.989** 0.021 1.966** 2.225*** 2.382*** 1.181*** 3.457*** 3.169***
  - t = 10: 1.888* 0.035 1.857* 2.141*** 2.414*** 1.331*** 3.527*** 3.095***
- Table 6 notes: dependent variable is either cumulative change in total number of refugees as percent of population or cumulative log difference of total number of refugees; regressions include controls for lagged dependent variable growth, lagged log GDP per capita in U.S. dollars, lagged log population, country and year fixed effects; standard errors clustered at the country level.

### Macroeconomic implications
- Large refugee outflows and persistently lower populations imply long-lasting reductions in the labor force and productive capacity of conflict countries.
- Conflict can have spillover effects on other countries via refugee flows, trade disruptions, and other channels.
- Even though conflict directly affects a relatively small number of countries, measurable effects on global GDP warrant attention, especially given increases in conflict episodes in some large economies and regions like Sub-Saharan Africa and Middle East and North Africa.

### Policy implications and recommendations
- Finding ways to reach ordinary people in conflict-affected countries—whose consumption falls and sometimes collapses for many years following conflict episodes—by extending public goods provision and social safety nets is essential.
- Macroeconomic stabilization policies for conflict-affected fragile states often differ from those for countries hit by conventional severe economic shocks (banking or currency crises) because trust in government institutions and delivery capacity are fundamentally eroded.
- The paper references Chami et al. (2020) for ways to adjust macroeconomic stabilization policies to meet the specific challenges facing fragile states, and United Nations and World Bank (2018) for policies on building inclusive approaches to recover from conflict and prevent future eruptions.

*Source: Authors’ calculations; content as presented in the supplied IMF working paper excerpt.*

### Annex Table 1. Data Sources

### Annex Table 1. Data Sources

### Indicator sources
- Conflict-related deaths
  - Uppsala Conflict Data Program (UCDP) Georeferenced Event Dataset v. 19.1; Uppsala Conflict Data Program (UCDP) Battle-Related Deaths Dataset v. 19.1
- Population
  - International Monetary Fund, World Economic Outlook Database; United Nations, Department of Economic and Social Affairs, Population Division, World Population Prospects, the 2017 revision; World Bank, World Development Indicators Database.
- Number of refugees
  - United Nations High Commissioner for Refugees (UNHCR), Population Statistics Reference Database.
- Real GDP, GDP per capita, total consumption, private consumption, government spending, investment; imports and exports trade value
  - International Monetary Fund, World Economic Outlook Database (all vintages from years 1990-2019).
- Agriculture, manufacturing, services real value added
  - World Bank, World Development Indicators Database.
- Commodity export price index
  - Gruss and Kebhaj (2019)
- State Fragility Index (indicator for scores above 11)
  - Center for Systemic Peace
- Neighboring countries
  - CEPII GeoDist Dataset

*IMF Working Paper annex tables*

---

### Annex Table 2. Country Sample

### Incidence: 100 killed per one million population
- Afghanistan (1989-2001, 2006-2018)
- Algeria (1995, 1998)
- Angola (1989, 1990, 1992-1994, 1999, 2001)
- Azerbaijan (1992-1994)
- Bosnia and Herzegovina (1992-1995)
- Burundi (1995-2003)
- Central African Republic (2013-2018)
- Chad (1990, 1992, 2000, 2006, 2008)
- Comoros (1997)
- Congo, Democratic Republic of the (1996-1999, 2002)
- Congo, Republic of (1997-1999)
- Croatia (1991, 1995)
- Djibouti (1990-1992)
- El Salvador (1989, 1990)
- Eritrea (1998-2000)
- Ethiopia (1989-1991, 1999, 2000)
- Georgia (1992, 1993, 2008)
- Ghana (1994)
- Guinea-Bissau (1998, 1999)
- Iraq (1991, 1997, 2003-2007, 2013-2017)
- Israel (2002, 2008, 2014)
- Kuwait (1990, 1991)
- Lebanon (1989, 1990, 2006)
- Liberia (1990-1996, 2001-2003)
- Libya (2011, 2014-2018)
- Moldova (1992)
- Mozambique (1989-1991)
- Nepal (2002)
- Nicaragua (1989)
- Panama (1989)
- Rwanda (1990-1995, 1997, 1998, 2001)
- Serbia (1998, 1999)
- Sierra Leone (1991-2000)
- Somalia (1989-1992, 1996, 2006-2012, 2016-2018)
- South Sudan (2011, 2013, 2014)
- Sri Lanka (1990-1993, 1995-2000, 2006-2009)
- Sudan (1989-1993, 1997, 1998, 2000-2004)
- Syria (2012-2018)
- Tajikistan (1992, 1993, 1996, 1998)
- Ukraine (2014)
- Yemen (1994, 2015, 2016, 2018)

### Onset: first year of conflict in which the number of deaths exceeds 100 killed per one million population after at least four consecutive years without passing that threshold
- Afghanistan (1989, 2006)
- Algeria (1995)
- Angola (1999)
- Azerbaijan (1992)
- Bosnia and Herzegovina (1992)
- Burundi (1995)
- Central African Republic (2013)
- Chad (1990, 2000, 2006)
- Comoros (1997)
- Congo, Democratic Republic of the (1996)
- Congo, Republic of (1997)
- Croatia (1991)
- Djibouti (1990)
- Eritrea (1998)
- Ethiopia (1999)
- Georgia (1992, 2008)
- Ghana (1994)
- Guinea-Bissau (1998)
- Iraq (1991, 1997, 2003, 2013)
- Israel (2002, 2008, 2014)
- Kuwait (1990)
- Lebanon (1989, 2006)
- Liberia (1990, 2001)
- Libya (2011)
- Moldova (1992)
- Nepal (2002)
- Panama (1989)
- Rwanda (1990)
- Serbia (1998)
- Sierra Leone (1991)
- Somalia (1989, 2006)
- South Sudan (2011)
- Sri Lanka (2006)
- Syria (2012)
- Tajikistan (1992)
- Ukraine (2014)
- Yemen (1994, 2015)

### Onset: first year of conflict in which the number of deaths exceeds 100 killed per one million population after at least two consecutive years without passing that threshold
- Afghanistan (1989, 2006)
- Algeria (1995, 1998)
- Angola (1999)
- Azerbaijan (1992)
- Bosnia and Herzegovina (1992)
- Burundi (1995)
- Central African Republic (2013)
- Chad (1990, 2000, 2006)
- Comoros (1997)
- Congo, Democratic Republic of the (1996, 2002)
- Congo, Republic of (1997)
- Croatia (1991, 1995)
- Djibouti (1990)
- Eritrea (1998)
- Ethiopia (1999)
- Georgia (1992, 2008)
- Ghana (1994)
- Guinea-Bissau (1998)
- Iraq (1991, 1997, 2003, 2013)
- Israel (2002, 2008, 2014)
- Kuwait (1990)
- Lebanon (1989, 2006)
- Liberia (1990, 2001)
- Libya (2011, 2014)
- Moldova (1992)
- Nepal (2002)
- Panama (1989)
- Rwanda (1990, 2001)
- Serbia (1998)
- Sierra Leone (1991)
- Somalia (1989, 1996, 2006, 2016)
- South Sudan (2011)
- Sri Lanka (2006)
- Sudan (1997)
- Syria (2012)
- Tajikistan (1992, 1996)
- Ukraine (2014)
- Yemen (1994, 2015)

*IMF Working Paper annex tables*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020110-print-pdf.pdf_