## wpiea2019221-print-pdf - introduction of year and country fixed-effects in regressions resolve part of this problem. But

## Source details

**Canonical URL:** [wpiea2019221-print-pdf - introduction of year and country fixed-effects in regressions resolve part of this problem. But](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019221-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019221-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019221-print-pdf.pdf.json)

---

### Nightlights data: limitations and empirical patterns
- Heterogeneous errors of measurement in nightlights data have serious implications for prediction quality.
- Yemen: sum of nightlights (cleaned from regions with no urban centers or no human activity) shows a seasonal pattern before 2015 with relatively high radiance in winter months.
- Total nightlights in Yemen collapsed in March 2015 (beginning of the conflict) and plateaued at a low level afterward; composite VIIRS images confirm a dramatic decrease in nightlight intensity between end-2014 and end-2017, particularly in urban centers (Sanaa) and oil-producing regions.
- DMSP-OLS data limitation: inability to discriminate between light sources and to account for radiant heat from gas flaring; flares often produce much brighter light than urban electric lighting.
- Up to 40 percent of total nightlights in pre-conflict Yemen were generated by flaring activity while the oil sector represented less than 10 percent of total GDP.
- Disturbances/noise in VIIRS monthly data are unevenly distributed across time and countries; several countries (including Yemen) exhibit a massive 2017 increase in total nightlights that cannot be traced to economic changes.
- A cluster dummy (split by median change in nightlights between 2016 and 2017) is constructed to control for this 2017 noise pattern.

### Gas flaring, VIIRS, and measurement approach
- VIIRS instrument (with infrared sensors) allows discrimination of light sources based on radiant heat intensity and identification of gas flaring.
- VIIRS flaring data are retreated to eliminate detections under 1.500 degree Celsius to avoid ephemeral fires.
- Radiant heat from each recorded flaring event is averaged to construct monthly country-level observations.
- Gas flaring in Yemen since 2012 exhibited a decreasing pattern before the conflict (progressive depletion of wells); flaring collapsed more significantly after March 2015 due to withdrawal of foreign oil companies.
- Aggregate flaring data show a slight recovery through 2017 but remained well below pre-conflict levels.
- For country assignment, geographic coordinates and exclusive economic zone boundaries are used to relate onshore and offshore flaring sites to corresponding countries.

### Data sources and auxiliary variables
- Real GDP (oil and total) data from the World Economic Outlook (WEO) database in constant LCU and PPP (constant 2011 U.S. dollars) for 2012-2017.
- Oil production data from the U.S. Energy Information Administration (EIA).
- Fragile States Index (FSI) from the Fund for Peace used to account for fragility/non-linearity (higher FSI = more fragile).
  - Yemen had one of the highest FSI among 178 countries; FSI deteriorated in 2013, receded in 2014, and deteriorated further after the conflict; Yemen was ranked fourth in the 2017 FSI global ranking.
- Control for VIIRS noise: dummy clustering countries based on median change in nightlights between 2016 and 2017; interacted with year fixed-effects.

### Econometric specifications (panel OLS across 2012-17)
- Samples:
  - 72 countries for total real GDP.
  - 28 oil-producing countries for oil GDP (MENAP, CCA, SSA).
  - Period 2012-17.
- Oil flaring model (log form):
  - radiant heat RHit = (Git)^φ.
  - oil production OPit relates to RHit, gas-to-oil ratio GORit, and technology TECHit.
  - Assumption: GOR and TECH are treated as time-invariant at country level over the 6-year panel and absorbed by country fixed-effects.
- Estimated oil production/oil GDP equation (log-linear with country and time fixed-effects):
  - op_it = π·rh_it − por_i − tech_i + δ_t + ε_it
  - oil GDP analog with γ coefficient on rh_it.
- Total real GDP baseline equation (log form):
  - gdp_it = η·rh_it + λ·tnl_it + μ_i + τ_t + ξ_it
  - η = elasticity of real GDP to radiant heat; λ = elasticity of real GDP to total nightlights; μ country FE; τ time FE.
- Models include controls for noise (cluster×year interactions) and FSI and interaction ln(TNL)*FSI to capture fragility-driven non-linearity.

### Key regression results and diagnostics
- Oil-output panel (28 countries, 6 years): within R2 for oil GDP specification with cluster interaction stands at 0.42 for the preferred oil GDP specification.
- Real GDP panel (72 countries, 2012-17): preferred specification (column 6) statistics:
  - Observations: 432
  - Countries: 72
  - Within R2: 0.407
  - ln(TNL) coefficient: 0.305*** (0.079)
  - ln(Radiant heat) coefficient: 0.099*** (0.020)
  - ln(TNL)*FSI coefficient: -0.002*** (0.001)
  - All regressions include country and year fixed-effects.
  - Standard errors reported in parentheses; significance codes: * p < 0.10, ** p < 0.05, *** p < 0.01.
- Oil GDP regressions (Table 1) indicate:
  - Preferred oil GDP specification (column 8) within R2 = 0.420.
  - ln(Radiant heat) in column 8: 0.116* (0.059).
  - Introduction of cluster interaction term decreases coefficient on radiant heat and increases that on total nightlights for oil GDP specifications.
- Appendix — Table A3 highlights coefficients on ln(TNL) across specifications (columns (1)–(6)):
  - (1) 0.080*** (0.026)
  - (2) 0.053** (0.026)
  - (3) 0.211*** (0.078)
  - (4) 0.156*** (0.033)
  - (5) 0.119*** (0.033)
  - (6) 0.296*** (0.078)
- Appendix — Table A3 coefficients on ln(Radiant heat):
  - (3) 0.107*** (0.022)
  - (4) 0.106*** (0.020)
  - (5) 0.099*** (0.022)
  - (6) 0.098*** (0.020)
- Appendix — Table A3 interaction ln(TNL)*FSI:
  - (3) -0.002** (0.001)
  - (5) -0.002** (0.001)
- Observations and sample sizes:
  - Observations: 432 in all columns.
  - Countries: 72 in all columns.

### Country-level model performance and examples
- Predictions versus official data:
  - Many countries: predicted values close to official series, but predicted series tend to be more volatile.
  - Yemen:
    - Model suggests oil GDP may have been overestimated before the conflict.
    - Oil GDP collapsed but less sharply than U.S. EIA production data suggest; slight recovery in 2017 indicated by model.
    - Real GDP for Yemen: model suggests pre-conflict overestimation of real GDP by some 24 percent (point estimate).
    - Predicted real GDP shows a dramatic collapse in 2015 and continued contraction at a decelerating pace; stagnation between 2016 and 2017.
  - Specific country mismatches:
    - Cameroon predicted contraction in 2016 (commodities crisis).
    - Uzbekistan predicted contraction/stagnation through 2015 while official data show steady growth.
    - Niger predictions differ from official oil GDP patterns.

### Yemen: aggregate and regional growth decomposition (2015-17)
- Aggregate predicted results for Yemen:
  - Point estimate: cumulative real GDP contraction over 2015-17 ≈ 24 percent.
  - Lower-bound estimate: cumulative contraction at most 34 percent.
  - Point estimates for 2017 suggest real GDP growth stood at -1 percent while oil production slightly recovered.
  - Cumulative oil GDP contraction throughout the conflict: 72 percent.
- Sectoral predicted contractions (2015-17):
  - Oil GDP contracted by 72 percent.
  - Non-oil GDP contracted by 21 percent.
- Regional (governorate-level) allocation method:
  - Predicted national oil and real GDP redistributed across governorates based on shares in total nightlights and gas flaring intensity; gas flaring data available for Marib, Hadramawt, Shabwah.
  - For governorate oil GDP, redistribution based on relative intensity of gas flaring.
  - Non-oil GDP obtained as aggregate predicted real GDP minus predicted oil GDP, then distributed by electric-source nightlights intensity.
  - Methodological caveat: assumes lit-pixels from flaring and electric sources generate the same light intensity; used as proxy.
- Governorate contributions to cumulative real GDP growth (2015-17) — selected entries from Appendix Table A4:
  - Aden: Total -5.8; Non-Oil -6.2; Oil -
  - Amanat Al Asimah (Sanaa city): Total -9.7; Non-Oil -10.4; Oil -
  - Hadramawt: Total -1.9; Non-Oil -1.3; Oil -10.1
  - Marib: Total 2.5; Non-Oil 2.6; Oil 1.2
  - Shabwah: Total -4.2; Non-Oil 0.2; Oil -63.2
  - Al Jawf: Total 3.1; Non-Oil 3.4; Oil -
  - Al Mahrah: Total 4.2; Non-Oil 4.6; Oil -
  - Totals for Yemen (2015-17):
    - Total GDP: -24.2 percent
    - Non-Oil GDP: -20.7 percent
    - Oil GDP: -72.0 percent
- Geographic heterogeneity noted:
  - Aden and Sanaa together contributed about 15 percentage points of the aggregate contraction.
  - Hadramawt and Shabwah combined contribution: -6 percentage points.
  - Western governorates Al Hudaydah, Ibb, and Taizz displayed massive contractions.
  - Al Jawf and Al Mahrah contributed positively to growth (around 7 percentage points combined), possibly due to proximity to neighbors and trans-national trade mitigating conflict impact.
  - Marib contributed positively to aggregate GDP growth (benefiting from business relocation and trade diversion); among the three oil-producing governorates, Marib was the only one with positive oil GDP growth.

### Main conclusions and policy/research implications
- Satellite-recorded nightlights (VIIRS) provide a valuable alternative to official statistics for assessing economic developments in conflict settings and countries with weak statistical capacity.
- Accounting for gas flaring (via radiant heat) is critical when linking nightlights to real GDP, especially in oil-producing countries where flaring comprises a disproportionate share of nightlight intensity.
- Predicted cumulative contraction in Yemen (≈ 24 percent) is lower than official Ministry of Planning figures (~50 percent), but predicted collapse in oil GDP and suspension of public salary payments are primary drivers of the humanitarian crisis.
- Policy/research recommendations:
  - Include radiant heat (gas flaring) and controls for measurement noise (e.g., 2017 cluster dummy) when using nightlights to estimate GDP.
  - Use VIIRS flaring data to approximate oil production and oil GDP where official data are unreliable or absent.
  - Future research: leverage other satellite data (infrastructure, physical assets) to cross-check nightlights-based conclusions in conflict-afflicted countries.

*Source: wpiea2019221-print-pdf.*

### introduction of year and country fixed-effects in regressions resolve part of this problem. But

### wpiea2019221-print-pdf - introduction of year and country fixed-effects in regressions resolve part of this problem. But

### Nightlights data: limitations and patterns
- Heterogeneous errors of measurement in nightlights data have serious implications for prediction quality.
- Yemen: sum of nightlights (cleaned from regions with no urban centers or no human activity) shows a seasonal pattern before 2015 with relatively high radiance in winter months.
- Total nightlights in Yemen collapsed in March 2015 (beginning of the conflict) and plateaued at a low level afterward; composite VIIRS images confirm a dramatic decrease in nightlight intensity between end-2014 and end-2017, particularly in urban centers (Sanaa) and oil-producing regions.
- DMSP-OLS data limitation: inability to discriminate between light sources and to account for radiant heat from gas flaring; flares often produce much brighter light than urban electric lighting.
- Up to 40 percent of total nightlights in pre-conflict Yemen were generated by flaring activity while the oil sector represented less than 10 percent of total GDP.
- Disturbances/noise in VIIRS monthly data are unevenly distributed across time and countries; several countries (including Yemen) exhibit a massive 2017 increase in total nightlights that cannot be traced to economic changes. A cluster dummy (split by median change in nightlights between 2016 and 2017) is constructed to control for this 2017 noise pattern.

### Gas flaring, VIIRS, and measurement approach
- VIIRS instrument (with infrared sensors) allows discrimination of light sources based on radiant heat intensity and identification of gas flaring.
- VIIRS flaring data are retreated to eliminate detections under 1.500 degree Celsius to avoid ephemeral fires.
- Radiant heat from each recorded flaring event is averaged to construct monthly country-level observations.
- Gas flaring in Yemen since 2012 exhibited a decreasing pattern before the conflict (progressive depletion of wells); flaring collapsed more significantly after March 2015 due to withdrawal of foreign oil companies.
- Aggregate flaring data show a slight recovery through 2017 but remained well below pre-conflict levels.
- For country assignment, geographic coordinates and exclusive economic zone boundaries are used to relate onshore and offshore flaring sites to corresponding countries.

### Data sources and auxiliary variables
- Real GDP (oil and total) data from the World Economic Outlook (WEO) database in constant LCU and PPP (constant 2011 U.S. dollars) for 2012-2017.
- Oil production data from the U.S. Energy Information Administration (EIA).
- Fragile States Index (FSI) from the Fund for Peace used to account for fragility/non-linearity (higher FSI = more fragile). Yemen had one of the highest FSI among 178 countries; FSI deteriorated in 2013, receded in 2014, and deteriorated further after the conflict; Yemen was ranked fourth in the 2017 FSI global ranking.
- Control for VIIRS noise: dummy clustering countries based on median change in nightlights between 2016 and 2017; interacted with year fixed-effects.

### Econometric specifications (panel OLS across 2012-17)
- Samples: 72 countries for total real GDP; 28 oil-producing countries for oil GDP (MENAP, CCA, SSA); period 2012-17.
- Oil flaring model (log form): radiant heat RHit = (Git)^φ; oil production OPit relates to RHit, gas-to-oil ratio GORit, and technology TECHit. Assumption: GOR and TECH are treated as time-invariant at country level over the 6-year panel and absorbed by country fixed-effects.
- Estimated oil production/oil GDP equation (log-linear with country and time fixed-effects): op_it = π·rh_it − por_i − tech_i + δ_t + ε_it and oil GDP analog with γ coefficient on rh_it.
- Total real GDP baseline equation (log form):
  - gdp_it = η·rh_it + λ·tnl_it + μ_i + τ_t + ξ_it
  - η = elasticity of real GDP to radiant heat; λ = elasticity of real GDP to total nightlights; μ country FE; τ time FE.
- Models include controls for noise (cluster×year interactions) and FSI and interaction ln(TNL)*FSI to capture fragility-driven non-linearity.

### Key regression results and diagnostics
- Oil-output panel (28 countries, 6 years): within R2 for oil GDP specification with cluster interaction stands at 0.42 for the preferred oil GDP specification.
- Real GDP panel (72 countries, 2012-17): preferred specification (column 6) statistics:
  - Observations: 432; Countries: 72; Within R2: 0.407.
  - ln(TNL) coefficient: 0.305*** (0.079).
  - ln(Radiant heat) coefficient: 0.099*** (0.020).
  - ln(TNL)*FSI coefficient: -0.002*** (0.001).
  - All regressions include country and year fixed-effects.
  - Standard errors reported in parentheses; significance codes: * p < 0.10, ** p < 0.05, *** p < 0.01.
- Oil GDP regressions (Table 1) indicate:
  - Preferred oil GDP specification (column 8) within R2 = 0.420.
  - ln(Radiant heat) in column 8: 0.116* (0.059).
  - Introduction of cluster interaction term decreases coefficient on radiant heat and increases that on total nightlights for oil GDP specifications.

### Country-level model performance and examples
- Predictions versus official data:
  - Many countries: predicted values close to official series, but predicted series tend to be more volatile.
  - Yemen: model suggests oil GDP may have been overestimated before the conflict; oil GDP collapsed but less sharply than U.S. EIA production data suggest; slight recovery in 2017 indicated by model.
  - Real GDP for Yemen: model suggests pre-conflict overestimation of real GDP by some 24 percent (point estimate). Predicted real GDP shows a dramatic collapse in 2015 and continued contraction at a decelerating pace; stagnation between 2016 and 2017.
  - Specific country mismatches: Cameroon predicted contraction in 2016 (commodities crisis), Uzbekistan predicted contraction/stagnation through 2015 while official data show steady growth, Niger predictions differ from official oil GDP patterns.

### Yemen: aggregate and regional growth decomposition (2015-17)
- Aggregate predicted results for Yemen:
  - Point estimate: cumulative real GDP contraction over 2015-17 ≈ 24 percent.
  - Lower-bound estimate: cumulative contraction at most 34 percent.
  - Point estimates for 2017 suggest real GDP growth stood at -1 percent while oil production slightly recovered.
  - Cumulative oil GDP contraction throughout the conflict: 72 percent.
- Sectoral predicted contractions:
  - Oil GDP contracted by 72 percent (2015-17).
  - Non-oil GDP contracted by 21 percent (2015-17).
- Regional (governorate-level) allocation method:
  - Predicted national oil and real GDP redistributed across governorates based on shares in total nightlights and gas flaring intensity; gas flaring data available for Marib, Hadramawt, Shabwah.
  - For governorate oil GDP, redistribution based on relative intensity of gas flaring.
  - Non-oil GDP obtained as aggregate predicted real GDP minus predicted oil GDP, then distributed by electric-source nightlights intensity.
  - Acknowledged methodological caveat: assumes lit-pixels from flaring and electric sources generate the same light intensity; used as proxy.
- Governorate contributions to cumulative real GDP growth (2015-17):
  - Aden and Sanaa together contributed about 15 percentage points of the aggregate contraction.
  - Hadramawt and Shabwah combined contribution: -6 percentage points.
  - Western governorates Al Hudaydah, Ibb, and Taizz displayed massive contractions.
  - Al Jawf and Al Mahrah contributed positively to growth (around 7 percentage points combined), possibly due to proximity to neighbors and trans-national trade mitigating conflict impact.
  - Marib contributed positively to aggregate GDP growth (benefiting from business relocation and trade diversion); among the three oil-producing governorates, Marib was the only one with positive oil GDP growth.

### Main conclusions and implications
- Satellite-recorded nightlights (VIIRS) provide a valuable alternative to official statistics for assessing economic developments in conflict settings and countries with weak statistical capacity.
- Accounting for gas flaring (via radiant heat) is critical when linking nightlights to real GDP, especially in oil-producing countries where flaring comprises a disproportionate share of nightlight intensity.
- Predicted cumulative contraction in Yemen (≈ 24 percent) is lower than official Ministry of Planning figures (~50 percent), but predicted collapse in oil GDP and suspension of public salary payments are primary drivers of the humanitarian crisis.
- Geographic heterogeneity matters: some governorates contributed positively to aggregate growth despite overall national contraction.
- Policy/research implications:
  - Include radiant heat (gas flaring) and controls for measurement noise (e.g., 2017 cluster dummy) when using nightlights to estimate GDP.
  - Use VIIRS flaring data to approximate oil production and oil GDP where official data are unreliable or absent.
  - Future research: leverage other satellite data (infrastructure, physical assets) to cross-check nightlights-based conclusions in conflict-afflicted countries.

*Source: wpiea2019221-print-pdf.*

### REFERENCES

### REFERENCES

### Bibliographic sources cited
- Bandhari, Laveesh and Koel Roychowdhruy. 2011. “Night lights and economic activity in Inda: a study using DMSP-OLS night time images.” Proceedings of the Asia-Pacific Advanced Network 32: 218-236.
- Basihos, Seda. 2016. “Nightlights as a development indicator: the estimation of gross provincial product in Turkey.” WP Economic Policy Research Foundation of Turkey.
- Bundervoet, Tom, Laban Maiyo and Apurva Sanghi. 2015. “Bright lights, big cities: measuring national and subnational economic growth from outer space in Africa, with an application to Kenya and Rwanda.” World Bank Report ACS15584.
- Dai Zhaoxin, Dai, Yunfeng Hu and Guanhua Zhao. 2017. “The suitability of different nighttime light data for GDP estimation at different spatial scales and regional levels.” Sustainability 9-305.
- Do, Quy-Toan, Jacob N. Shapiro, Christopher D. Elvidge, Mohamed Abdel-Jelil, Daniel P. Ahn, Kimberley Baugh, Jamie Hansen Lewis and Mikhail Zhizhin. 2017. “How much oil is the Islamic State Group producing.” World Bank Policy Research WP 8231.
- Elvidge, Christopher, Kimberly E. Baugh, Sharolyn Anderson, Paul C. Sutton, and Tilottama Ghosh. 2012. “The night light development index (nldi): a spatially explicit measure of human development from satellite data.” Social Geography 7(1).
- Gosh, Tilottama, Rebecca L. Powell, Sharolyn Anderson, Paul C. Sutton and Christopher D. Elvidge. 2010. “Informal economy and remittances estimates of India using nighttime imagery.” International Journal of Ecological Economics and Statistics 17(10).
- Henderson, Vernon, Adam Storeygard, and David N. Weil. 2012. “Measuring economic growth from outer space.” American Economic Review 102(2): 994-1028.
- International Labour Organization. 2018. “Small and medium-sized enterprises damage assessment: Yemen.”
- McGregor, Thomas and Samuel Wills. 2017. “Surfing the wave of economic growth.” CAMA WP 31/2017.
- Ministry of Planning and International Cooperation of Yemen. 2019. “Prospects for Yemen’s economy and livelihoods priority.” Note from the Economic Studies and Forecasting Sector Division, Issue (40) February, 2019.
- Pfeifer, Gregor, Fabian Wahl and Martyna Marczak. 2018. “Illuminating the World Cup effect: night lights evidence from South Africa.” Journal of regional science 58(5): 887-920.
- Pinkovskiy, Maxim and Xavier Sala-i-Martin. 2016. “Newer need not be better: evaluating the Penn World Tables and the World Development Indicators using nighttime lights.” NBER WP No. 22216.
- Salisbury, Peter. 2017. “Yemen: national chaos, local order.” Chatham House Research Paper, December 2017.
- Skoufias, Emmanuel, Eric Strobl and Thomas Tveit. 2017. “Natural disaster damage indices based on remotely sensed data.” World Bank Poverty and Equity Global Practice WP No. 119.
- United Nations Development Programme. 2015. “Rapid business survey: impact of the Yemen crisis on private sector activity.”
- Zhao, Naizhuo, Feng-Chi Hsu, Guofend Cao and Eric L. Samson. 2017. “Improving accuracy of economic estimations with VIIRS DNB image products” International journal of remote sensing 38(21)” 5899-5918.

### Appendix — Table A1: Summary Statistics (sample indicators and moments)
- Variables: ln(TNL), ln(GDP), cst LCU, ln(GDP), PPP cst 2011 US$, ln(Radiant heat), ln(Oil GDP), ln(Oil prod.), Fragile States Index, Statistical Capacity Index.
- ln(TNL): Mean 14.645; SD 2.475; Min 1.665; Max 18.749; Count 432; Sample Full.
- ln(GDP), cst LCU: Mean 6.480; SD 3.049; Min -1.685; Max 15.758; Count 432; Sample Full.
- ln(GDP), PPP cst 2011 US$: Mean 3.806; SD 1.760; Min -0.686; Max 7.395; Count 432; Sample Full.
- ln(Radiant heat): Mean 1.675; SD 0.812; Min -0.225; Max 3.514; Count 168; Sample Oil producers.
- ln(Oil GDP): Mean 5.216; SD 3.262; Min -0.041; Max 14.288; Count 168; Sample Oil producers.
- ln(Oil prod.): Mean 6.016; SD 1.777; Min 2.398; Max 9.424; Count 168; Sample Oil producers.
- Fragile States Index: Mean 85.687; SD 15.114; Min 43.700; Max 114.900; Count 432; Sample Full.
- Statistical Capacity Index: Mean 61.470; SD 15.307; Min 20.000; Max 95.556; Count 396; Note: N/a for Bahrain, Kuwait, Oman, Qatar, Saudi Arabia and UAE.

### Appendix — Table A2: List of Countries for Panel Regressions
- Real GDP equation (72 countries): Afghanistan; Algeria; Angola; Armenia; Azerbaijan; Bahrain; Benin; Botswana; Burkina Faso; Burundi; Cabo Verde; Cameroon; Central African Republic; Chad; Comoros; Côte d'Ivoire; Democratic Republic of the Congo; Djibouti; Egypt; Equatorial Guinea; Eritrea; Ethiopia; Gabon; Georgia; Ghana; Guinea; Guinea-Bissau; Iran; Iraq; Jordan; Kazakhstan; Kenya; Kuwait; Kyrgyzstan; Lebanon; Lesotho; Liberia; Libya; Madagascar; Malawi; Mali; Mauritania; Morocco; Mozambique; Namibia; Niger; Nigeria; Oman; Pakistan; Qatar; Republic of Congo; Rwanda; Saudi Arabia; Seychelles; Sierra Leone; Somalia; South Africa; Sudan; Swaziland; São Tomé and Príncipe; Tajikistan; Tanzania; Gambia; Togo; Tunisia; Turkmenistan; United Arab Emirates; Uganda; Uzbekistan; Yemen; Zambia; Zimbabwe.
- Oil GDP equation (28 countries): Algeria; Angola; Azerbaijan; Bahrain; Cameroon; Chad; Democratic Republic of the Congo; Egypt; Equatorial Guinea; Gabon; Ghana; Iran; Iraq; Kazakhstan; Kuwait; Libya; Niger; Nigeria; Oman; Qatar; Republic of Congo; Saudi Arabia; Sudan; Tunisia; Turkmenistan; United Arab Emirates; Uzbekistan; Yemen.

### Appendix — Table A3: Real GDP (PPP, constant 2011 U.S. dollars) and Total Nightlights — regression highlights
- Models (columns (1)–(6)); all regressions include country and year fixed effects. Standard errors in parentheses. Significance: * p < 0.10, ** p < 0.05, *** p < 0.01.
- Coefficients on ln(TNL):
  - (1) 0.080*** (0.026)
  - (2) 0.053** (0.026)
  - (3) 0.211*** (0.078)
  - (4) 0.156*** (0.033)
  - (5) 0.119*** (0.033)
  - (6) 0.296*** (0.078)
- Coefficients on ln(Radiant heat):
  - (3) 0.107*** (0.022)
  - (4) 0.106*** (0.020)
  - (5) 0.099*** (0.022)
  - (6) 0.098*** (0.020)
- FSI (Fragile States Index) coefficients:
  - (3) 0.005 (0.012)
  - (5) 0.008 (0.012)
- Interaction ln(TNL)*FSI coefficients:
  - (3) -0.002** (0.001)
  - (5) -0.002** (0.001)
- Cluster in 2017: No for columns (1)–(3); Yes for columns (4)–(6).
- Observations: 432 in all columns.
- Countries: 72 in all columns.
- Within R2:
  - (1) 0.212
  - (2) 0.262
  - (3) 0.373
  - (4) 0.250
  - (5) 0.291
  - (6) 0.411

### Appendix — Table A4: Governorate contributions to cumulative real GDP growth in Yemen: 2015-17
- Governorate contributions (Total GDP in percent; Non-Oil GDP in percent; Oil GDP in percent):
  - Abyan: Total -0.1; Non-Oil -0.2; Oil -
  - Aden: Total -5.8; Non-Oil -6.2; Oil -
  - Al Bayda: Total 0.4; Non-Oil 0.4; Oil -
  - Al Dali: Total 0.0; Non-Oil 0.0; Oil -
  - Al Hudaydah: Total -3.4; Non-Oil -3.6; Oil -
  - Al Jawf: Total 3.1; Non-Oil 3.4; Oil -
  - Al Mahrah: Total 4.2; Non-Oil 4.6; Oil -
  - Al Mahwit: Total 0.0; Non-Oil 0.0; Oil -
  - Amanat Al Asimah (Sanaa city): Total -9.7; Non-Oil -10.4; Oil -
  - Amran: Total -0.5; Non-Oil -0.6; Oil -
  - Dhamar: Total -0.4; Non-Oil -0.4; Oil -
  - Hadramawt: Total -1.9; Non-Oil -1.3; Oil -10.1
  - Hajjah: Total -0.1; Non-Oil -0.1; Oil -
  - Ibb: Total -1.0; Non-Oil -1.1; Oil -
  - Lahij: Total -0.8; Non-Oil -0.9; Oil -
  - Marib: Total 2.5; Non-Oil 2.6; Oil 1.2
  - Raymah: Total 0.2; Non-Oil 0.2; Oil -
  - Saadah: Total -0.5; Non-Oil -0.5; Oil -
  - Sanaa: Total -3.6; Non-Oil -3.8; Oil -
  - Shabwah: Total -4.2; Non-Oil 0.2; Oil -63.2
  - Taizz: Total -2.6; Non-Oil -2.8; Oil -
- Totals for Yemen (2015-17):
  - Total GDP: -24.2 percent
  - Non-Oil GDP: -20.7 percent
  - Oil GDP: -72.0 percent
- Note: Estimates based on the methodology described in section IV.

*Source: wpiea2019221-print-pdf - REFERENCES*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019221-print-pdf.pdf_