Presented by Bradford Mills (Virginia Tech), Genti Kostandini (University of Georgia), Anthony Murray (Economic Research Service, USDA), Jiangfeng Gao (Virginia Tech), Joseph Rusike (AGRA), Steven Omamo, Zhe Guo and Jawoo Koo (IFPRI) at the Livestock Systems and Environment (LSE) Seminar, ILRI, Nairobi, 28 January 2015
Integrated agricultural system, migration, and social protection strategies to reduce vulnerability to climate change in East Africa
1. Integrated Agricultural System, Migration, and Social
Protection Strategies to Reduce Vulnerability to Climate
Change in East Africa
Bradford Mills - Virginia Tech
Genti Kostandini – University of Georgia
Anthony Murray – Economic Research Service, USDA
Jiangfeng Gao – Virginia Tech
Joseph Rusike - AGRA
Steven Omamo
Zhe Guo - IFPRI
Jawoo Koo - IFPRI
LSE Seminar, ILRI Nairobi, 28 January 2015
2. Project Objectives
• Estimate potential costs climatic changes impose on
vulnerable rural households
– yield decreases
– yield variance increases
• Identify agricultural system strategies that mitigate climate
change costs
• Rural household use of integrated agricultural system, off-
on farm employment, migration, and formal and informal
safety net strategies to reduce vulnerability to climatic
change
• Policy briefs that assist policymakers to generate country-
specific interventions to mitigate the impacts of climatic
change
3. Project Components
• 6 months into an 18 month project
• Component one – Climate, crop, income
linkages
– Identify the monetary costs to households and
regions that climatic change is expected to have
on agricultural systems in two East Africa
countries: Ethiopia and Zambia
– Simulation modeling (Genti)
4. Component one
• Identify the monetary costs of climate changes on
the agricultural systems of Ethiopia and Zambia at
the regional and household level.
• Translate rainfall change patterns into climate shocks
for major crops using DSSAT crop model for the
2000-2011 period.
• Use a methodology that takes into account the effect
on mean yields and yield variance and higher
moments of yield distribution.
• Produce ex-ante estimates based on forward looking
plausible climate change scenarios.
6. Methodology cont.
• Benefits from mean yield decreases for each
household type
• Benefits from yield variance increases
iijjjjij PQPY )1(.Pr
(i = poor farm, average farm, rich farm: j = drought risk type)
)(5.0.Pr 22
jpijijijiij sRYRB
(i = poor farm, average farm, rich farm: j = PFS)
7. Data
• Geo-referenced rainfall and temperature data for
the 2000-2011 period to characterize drought
risk using planting and harvesting dates.
• Geo-referenced farm level panel household data
(The Ethiopian Rural Household Survey seven
waves from 1994 to 2009 and the Zambian
Central Statistical Office 2000, 2004 and 2008) to
estimate the benefits for different household
types.
• Use 11 years of DSSAT crop model data to isolate
the impact of changes in rainfall pattern on yield
and yield variability.
• Use 2005 baseline SPAM production data from
IFPRI.
8. Table 1. Simulations of potential impacts of a 20% Maize Yield decrease and a 10% yield variance increase.
Very Severe
Drought
Moderate
Drought
Mild Drought Incipient Drought No Drought
Annual welfare changes from a 20% mean yield decrease (Thousand US $) Total Total
PR CS PR CS PR CS PR CS PR CS Total Losses (MT) Losses (%)
Ethiopia (294) (111) (23,909) (9,022) (20,686) (7,806) (11,615) (4,383) (30) (11) (77,868) (569,078) (14.71)
Zambia (3,808.2) (1,878.0) - - (12,502.4) (6,165.6) (4,527.0) (2,232.5) (1.3) (0.7) (31,115.7) (156,065.7) (7.9)
Subtotal (4,102.0) (1,988.9) (23,909.1) (9,022.3) (33,188.8) (13,971.8) (16,141.8) (6,615.4) (31.4) (12.0) (108,983.) (725,143.)
Annual welfare changes from a 10% increase in yield variance (Thousand US $)
PR CS PR CS PR CS PR CS PR CS
Ethiopia (59.6) (83.7) (4,932.9) (6,924.2) (4,438.5) (6,230.2) (2,544.7) (3,571.9) (8.9) (12.6) (28,807.1)
Zambia (87.3) (263.6) - - (286.6) (865.6) (103.8) (313.4) - - (1,920.3)
Sub-total (146.9) (347.3) (4,932.9) (6,924.2) (4,725.1) (7,095.8) (2,648.4) (3,885.3) (8.9) (12.6) (30,727.4)
Total (4,248.8) (2,336.2) (28,842.0) (15,946.5) (37,913.9) (21,067.6) (18,790.2) (10,500.8) (40.4) (24.6) (139,710.9)
9. Table 2. Share of losses by drought risk zone.
Total
losses
('000 US$)
Share
losses
Very
Severe
Drought
Moderate
Drought
Mild
Drought
Incipient
Drought
No Drought
Variance
share in
total losses
Producer
surplus
share in
total losses
Total
production
losses (MT)
Production
Decrease
Share
production
losses
Ethiopia (106,675) 0.76 0.01 0.42 0.37 0.21 0.00 0.27 0.64 (569,078) (0.147) 0.8
Zambia (33,036) 0.24 0.18 0.00 0.60 0.22 0.00 0.06 0.65 (156,066) (0.079) 0.2
Total (139,711) 0.05 0.32 0.42 0.21 0.00 0.22 0.64 (725,144)
10. Table 3. Simulations of potential impacts of different maize yield decreases and a yield variance increases.
Very Severe
Drought
(50%)
Moderate
Drought
(40%)
Mild Drought
(30%)
Incipient
Drought
(20%)
No Drought
(10%)
Welfare changes from 50%-10% mean yield decrease (Thousand US $) Total Total
PR CS PR CS PR CS PR CS PR CS Total
Losses
(MT)
Losses (%)
Ethiopia (725) (274) (47,421) (17,895) (30,907) (11,663) (11,615) (4,383) (15) (6) (124,903) (918,205) (24)
Zambia (9,168) (4,521) - - (18,522) (9,134) (4,527) (2,232) (1) (0) (48,105) (245,657) (13)
Subtotal (9,893) (4,795) (47,421) (17,895) (49,429) (20,797) (16,142) (6,615) (16) (6) (173,008) (1,163,862)
Welfare changes from 50%-10% yield variance reductions in 2016 (Thousand US $)
PR CS PR CS PR CS PR CS PR CS
Ethiopia (460) (498) (27,356) (31,653) (16,572) (20,471) (5,680) (7,484) (9) (13) (110,195)
Zambia (699) (1,569) - - (1,092) (2,844) (234) (657) - - (7,096)
Sub-total (1,159) (2,067) (27,356) (31,653) (17,665) (23,315) (5,914) (8,141) (9) (13) (117,291)
Total (11,052) (6,862) (74,776) (49,548) (67,094) (44,112) (22,056) (14,756) (25) (19) (290,299)
11. Annual welfare changes from a 20% yield decreases (US $/year)
Small farms Average farms Big farms
Ethiopia -28.26 -52.08 -109.67
Zambia -39.19 -73.45 -182.60
Annual welfare changes from a 10% yield variance increase (US $/year)
Ethiopia -13.83 -14.85 -19.71
Zambia -4.52 -5.56 -10.19
Table 4. Simulations of household annual welfare changes from a 20% mean yield decrease and a 10% yield variance increase.
12. Summary
• Potential losses are considerable given that only
maize losses were simulated.
• Estimated losses vary widely depending on the level
of drought risk.
• Welfare changes due to yield variability are an
important part of the overall welfare changes.
13. On-going work on component one
• Finish all major crops and use DSSAT panel
data results to estimate regional losses.
• Complete household data analysis and
estimate household losses by drought risk.
• Estimate losses/benefits based on future
climate scenarios .
14. Component two – integrated coping
mechanisms
Identify broader set of household
adaptations with long-term panel datasets
in Zambia and Ethiopia
(Brad for Anthony and Jiangfeng)
15. • Focus on agricultural adaptation: crop shares and yields
• Three-wave national representative panel:
– Created by Zambian Central Statistical Office, Ministry of Agriculture
and Cooperatives, and the Food Security Research Project
• Survey focuses on agricultural production and household
characteristics
– Survey rounds cover 1999/2000, 2002/2003, and 2006/2007
agricultural seasons
– 4,286 households successfully interviewed in all 3 panels (out of
original 7,699)
• Past research not found attrition bias (Mason and Jayne, 2013)
– This analysis only looks at households growing maize
• Climate Data:
– African Drought and Flood Monitor:
• Daily rainfall (mm)
• Aggregated for Planting (Nov-Dec), Growing (Jan-Mar), and Harvest (Apr-May)
seasons
Integrated Household Coping
Mechanisms – Zambia Data
16. Empirical Specification
• Assume a household fixed effects model to
exploit panel dataset:
• Two dependent variables of interest:
– Share of maize grown by farm household
– Maize yield per hectare
it it i itXy
17. Variable Mean Std. Dev. N
Maize Share of Cropped Land 0.60 0.27 6098
Number of Adult Equivalent 5.37 2.68 6098
Female Headed Household 0.21 0.41 6098
Net Income (less Maize) (US Dollars) 553.80 979.62 6098
Total Assets 71.40 188.60 6098
Owns Livestock (dummy) 0.83 0.38 6098
Total Landholdings 2.53 3.05 6098
Hectares Cultivated 1.99 2.03 6098
Had Fallow Land (dummy) 0.33 0.47 6098
Use Fertilizer (dummy) 0.40 0.49 6098
2008 Dummy 0.52 0.50 6098
Lag Groundnut Price 0.28 0.03 6098
Lag Sweet Potato Price 0.05 0.01 6098
Lag Maize Price 0.09 0.01 6098
Grew Cash Crops previous Survey Year 0.21 0.41 6098
Grew High Value Crops previous Survey Year 0.45 0.50 6098
Grew Other Staple Crops previous Survey Year 0.52 0.50 6098
Lagged Planting Season Coef. of Var. (5 yr) 0.69 0.12 6098
Lagged Growing Season Coef. of Var. (5 yr) 0.68 0.14 6098
Lagged Planting Season 10 day Rainfall (5 yr avg.) 53.14 12.55 6098
Lagged Growing Season 10 day Rainfall (5 yr avg.) 59.59 10.91 6098
Summary Statistics: Maize Share
18. Variable Mean Std. Dev. N
Maize yield per hectare 1619.39 1308.20 9929
Number of Adult Equivalent 5.30 2.74 9929
Female Headed Household 0.21 0.40 9929
Net Income (less Maize) (US Dollars) 465.08 859.41 9929
Total Assets 273.26 1702.91 9929
Owns Livestock (dummy) 0.83 0.38 9929
Total Landholdings 2.63 3.07 9929
Hectares Cultivated 2.03 2.06 9929
Had Fallow Land (dummy) 0.37 0.48 9929
Use Fertilizer (dummy) 0.36 0.48 9929
Fertilizer per Hectare 101.10 203.68 9929
Grew Cash Crops (dummy) 0.21 0.40 9929
Grew High Value Crops (dummy) 0.49 0.50 9929
Grew Other Staple Crops (dummy) 0.51 0.50 9929
Share of Maize planted/Total Cropped land 0.60 0.29 9929
Total Rainfall (mm) over growing season 621.80 155.40 9929
Planting Season Coef. of Var. (5 yr) 0.70 0.12 9929
Growing Season Coef. of Var. (5 yr) 0.62 0.12 9929
Harvest Season Coef. of Var. (5 yr) 1.51 0.36 9929
Summary Statistics: Yield/ha
19. Climate: By Year & Specification
Maize yield per hectare 1999/2000 2002/2003 2006/2007
Climate Variables Mean Std. Dev Mean Std. Dev Mean Std. Dev
Total Rainfall (mm) over growing season 603.13 89.91 597.68 177.64 663.95 170.26
Planting Season Coef. of Var. (5 yr) 0.69 0.14 0.69 0.13 0.71 0.08
Growing Season Coef. of Var. (5 yr) 0.57 0.10 0.64 0.10 0.63 0.14
Harvest Season Coef. of Var. (5 yr) 1.65 0.39 1.36 0.24 1.53 0.37
Maize Share Specification 2002/2003 Season 2006/2007 Season
Climate Variables Mean Std. Dev Mean Std. Dev
Lagged Planting Season Coef. of Var. (5 yr) 0.69 0.11 0.68 0.12
Lagged Growing Season Coef. of Var. (5 yr) 0.64 0.12 0.71 0.14
Lagged Planting Season Total Rainfall (5 yr avg.) 56.73 13.39 49.79 10.68
Lagged Growing Season Total Rainfall (5 yr avg.) 59.42 10.84 59.74 10.98
N = 2943 N = 3155
20. Zambia: Maize Share (1 of 2)
Variable Coefficient Std. Err
Number of Adult Equivalent -0.006*** 0.002
Female Headed Household -0.002 0.017
Net Income (less Maize) (US Dollars) -1.80E-05*** 4.30E-06
Total Assets 3.73E-05 2.48E-05
Owns Livestock (dummy) -0.023** 0.011
Total Landholdings -0.001 0.001
Hectares Cultivated -0.012*** 0.003
Had Fallow Land (dummy) -0.010 0.008
Use Fertilizer (dummy) 0.026*** 0.010
2006/2007 Agricultural season (dummy) 0.101* 0.056
Changes in Household Characteristics:
• Changes in number of adults significantly decreases maize share planted
• Increases in income and ownership of livestock associated with crop diversification
• Adding more hectares cultivated reduces maize share
• Adopting fertilizer use increases maize shares
• Households in 2006/2007 ag. season grew more (p = 0.10) maize as a share of crops
21. Zambia: Maize Share (2 of 2)
Changes in Price & Climate Variables:
• Higher sweet potato prices in previous survey year decreases maize share
• Growing other crops (Cash/High Value/Staple) in previous survey year all show
significantly higher share of maize in current survey year
• Higher mean rainfall over past 5 years (excluding current year) for planting and
growing (p = 0.10) lead to increased share of maize
Lag Groundnut Price -7.03E-06 3.86E-05
Lag Sweet Potato Price -0.001*** 0.000
Lag Maize Price 9.89E-05 0.000
Grew Cash Crops previous Survey Year 0.099*** 0.011
Grew High Value Crops previous Survey Year 0.069*** 0.008
Grew Other Staple Crops previous Survey Year 0.080*** 0.009
Lagged Planting Season Coef. of Var. (5 yr) -0.072 0.045
Lagged Growing Season Coef. of Var. (5 yr) 0.057 0.077
Lagged Planting Season Total Rainfall (5 yr avg.) 0.002** 0.001
Lagged Growing Season Total Rainfall (5 yr avg.) 0.003* 0.001
Constant 0.368 0.123
22. Zambia: Maize yield/ha (1 of 2)
Changes in Household Characteristics:
• Increasing the number of adults significantly (p = 0.10) increases yield/ha
• Increases in income increases maize yield/ha
• Adding more hectares cultivated reduces yield/ha
• Fertilizer/ha increases yield/ha, but at a decreasing rate
• Households in 2002/2003 season have lower maize yield/ha relative to 1999/2000
Dependent Variable: Maize (kg) per Hectare Coefficient Std. Err
Number of Adult Equivalent 17.527* 8.974
Female Headed Household -22.608 74.699
Net Income (less Maize) (US Dollars) 0.087*** 0.022
Total Assets 0.014 0.016
Owns Livestock (dummy) 28.712 42.012
Total Landholdings 4.618 6.831
Hectares Cultivated -103.778*** 13.262
Had Fallow Land (dummy) -9.961 34.326
Use Fertilizer (dummy) -36.883 66.145
Fertilizer per Hectare 3.134*** 0.296
Fertilizer per Hectare Squared -0.001*** 0.000
2002/2003 Agricultural season (dummy) -167.697*** 38.137
23. Zambia: Maize yield/ha (2 of 2)
Changes in Household Characteristics & Climate Variables:
• Households in 2006/2007 season have lower maize yield/ha relative to 1999/2000
• Changes in growing cash/other staple crops led to significantly lower yield/ha
• Households with increasing shares of maize had lower yield/ha
• Increased rainfall during growing season increases yield/ha, but at a decreasing rate
• Higher Coefficients of Variation in Planting and Growing season associated with
lower yield/ha, while higher CV for Harvest associated with higher yield/ha
2006/2007 Agricultural season (dummy) -133.739*** 41.391
Grew Cash Crops (dummy) -117.635** 58.140
Grew High Value Crops (dummy) 36.012 44.143
Grew Other Staple Crops (dummy) -175.742*** 45.782
Share of Maize planted/Total Cropped land -664.952*** 93.240
Total Rainfall (mm) over growing season 3.228*** 0.739
Total Rainfall (mm) over growing season Squared -0.002*** 0.001
Planting Season Coef. of Var. (5 yr) -659.932*** 150.862
Growing Season Coef. of Var. (5 yr) -811.947*** 231.816
Harvest Season Coef. of Var. (5 yr) 202.038*** 64.982
Constant 1534.644 280.125
24. Ethiopia:
• Focus on broader set of adaptions
• Migration
• Off-farm labor
• Transfers
• Family
• Informal networks
• Formal networks
• Rainfall shocks influence these decisions through levels and
variance:
• Higher rainfall levels increase mean agricultural income
• Make on-farm activities more attractive
• Higher rainfall variability increases variance of agricultural
income and household vulnerabilty
• Make urban and off-farm jobs more attractive
25. Data
• ERHS household level data
– 3 waves (1999-2004-2009), unbalanced panel
– sample size: 1836 (1999)+1263 (2004) +1467 (2009)
– demographics, assets, expenditures, migration,
remittance, social safety networks, and off-farm
activities
• Climatic data
– precipitation (mm) on a daily basis
– mean and variance are calculated for each
Belg/Kiremt planting and growing season
28. Preliminary results: variables
Variable name Label
dlabmig Dummy for HH with migrated member due to labor market reasons
sharlabmig Share of HH members who migrated due to labor market reasons
valfhhmem Monetary value of transfers from former HH members
valfgov Monetary value of public transfers
valfissn Monetary value of transfers from informal social safety nets
nddwkp4m Number of person-days worked off-farm in the past 4 months
fhhsize Household size before migration
ratiorainbp Ratio of mean rainfall over past 5 years to mean rainfall over past 30 years in Belg planting season
ratiorainbg Ratio of mean rainfall over past 5 years to mean rainfall over past 30 years in Belg growing season
ratiorainkp Ratio of mean rainfall over past 5 years to mean rainfall over past 30 years in Kiremt planting season
ratiorainkg Ratio of mean rainfall over past 5 years to mean rainfall over past 30 years in Kiremt growing season
nsdrainbp
Revised standard deviation of rainfall during Belg planting season over last 5 years (=std. dev. for
above historical average rainfall, =-std. dev. for below historical average rainfall)
nsdrainbg
Revised standard deviation of rainfall during Belg growing season over last 5 years (=std. dev. for
above historical average rainfall, =-std. dev. for below historical average rainfall)
nsdrainkp
Revised standard deviation of rainfall during Kirmet planting season over last 5 years (=std. dev. for
above historical average rainfall, =-std. dev. for below historical average rainfall)
nsdrainkg
Revised standard deviation of rainfall during Kirmet growing season over last 5 years (=std. dev. for
above historical average rainfall, =-std. dev. for below historical average rainfall)
36. Component three – distilling policy relevant implications
• Evidence of broad adaptation
• Crop choice
• Off-farm labor
• Migration
• Safety net utilization (formal vs. informal)
• Evidence of real welfare costs of
• Mean rainfall decreases
• Variance Increases
• Further research
• Simulations linking historic and predicted rainfall to crop
changes
• Better (varying) timeframes for climate impacts
• Better specification of variance impacts
• Relative size of benefits from adaptation alternatives
• Guidelines for integrated policy support for adaptation
• Partnering with AGRA