Spillover Effects of Large- Scale Commercial Farms in Ethiopia Daniel Ali, Klaus Deininger and Anthony Harris (World Bank – DECAR)
Background Heated debate on LSLBI often shallow and ill-informed Limited data on availability of land, actual transfers or use Even basic questions cannot be answered How much land has been transferred? What % is actually used/abandoned? Is use in line with investment plans? Do large farms provide opportunities for local communities? Using administrative data collected by Central Statistical Agency in Ethiopia we estimate spillovers from commercial farms
Measuring spillovers to small-holders Benefits from commercial farms to local small-holders key part of debate on large farms Positive: employment, technology transfer, marketing, access to inputs Negative: displacement and conflict, water resources, etc. Investors often receive land on the understanding they will create local employment and improve outcomes for local farmers Valuable for policy makers to assess whether this happens
Two possible channels: Employment & Technology Entry of commercial farms in an area could improve access to inputs, demonstrate technology or provide source of employment Value in good relationships with local communities We test this by looking at the relationship between proximity to commercial farms and yields, improved seed use, chemical fertilizer use, and employment
Outline Context Data & descriptive statistics Empirical identification Results
Context Evolution of large farms and patterns of land use: Slowed after 2011 Only 55% of land is utilized 95% of large farms Ethiopian owned Contribution to the economy Very little permanent employment: 1 worker/20 ha (1 worker /50ha for maize) Land clearing
Evolution of Commercial Farms
No. of farmsHolding size in haCultivated land in haShare utilized Total 6,612 1,552,262852, By Size (ha) < ,40923, , ,779129, , ,036176, , ,137396, ,97088, > ,211193, By national origin Ethiopian 6,287 1,570,323859, Foreign ,44547, Joint 36 11,9897, By major crop Maize ,99593, Sorghum ,61262, Wheat ,81685, Sesame 2, ,417314, Coffee ,152124, Cotton ,526163, Other ,021165, Level of Land Utilization – Commercial farms
Yield comparison by farm size (Quintal/ha) Farm sizeMaizeSorghumTeffWheatSesameCoffee Smallholder < >
Data (1) CSA Agricultural Sample Survey - 10 years of data on small-holder’s agricultural practices and yields (~44,000 parcels per year) CSA Commercial Farm Survey – 4 years of data on commercial farms, including start dates and location Demographic and Health Survey – 2000, 2005 and 2011 rounds include information on individual employment status
Data (2) Link individual observations across rounds in DHS and AGSS using geographic location Link start date of farm to the year of the survey Using commercial farm location to measure distance between location in AGSS/DHS and commercial farms Create kebele-level panel with measure of proximity by crop, changing over time
Descriptive (1) – Yields (Q/ha)
Descriptive (2) – Inputs
Descriptive (3) – Proximity (km)
Yield, improved seed use and proximity (Maize)
Yield, chemical fertilizer use and proximity (Wheat)
Estimating yield/technology spillovers
Regression results (yield/technology) Effect of Distance to nearest farm growing:Yield (kebele level) Chemical fertilizer use (HH Level) Improved seed use (HH Level) Coffee ( )( )( ) Maize-0.112*** ** * (0.0282)( )( ) Sorghum (0.0202)( )( ) Teff * (0.0148)( )( ) Wheat e ** (0.0903)( )( ) Observations18,729436,575 R-squared Year FEYYY Woreda FEYYY Standard errors are clustered at the Zone level *** p<0.01, ** p<0.05, * p<0.1 Increasing distance to nearest farm from 25km to 100km, decreases likelihood of using chem. fert. By 17%. Same change decreases likelihood of using Improved seeds for wheat or maize by 10%.
Estimating employment effects
Employment generation from commercial farms – (active farms compared to future farms) Dependent variable: Individual does Paid work, {1,0} (1) (2)(3)(4)(5)(6)(7)(8)(9)(10) Effect of active farm growing CROP, {1,0} *** ** ** (0.0484)(0.0344)(0.0570)(0.0475)(0.0857)(0.0399)(0.102)(0.0701)(0.0481)(0.0485) Observations4,3277,7832,5284,2371,8414,4851,1782,2892,7004,651 R-squared Share of DHS clusters with active farm in 00, 05, CropMaize Sorghum Wheat Coffee Sesame Distance band (km) Standard errors are clustered at the DHS cluster level. Controls include: head status, age, age2, education, female, hhsize, religion Fixed effects for survey year and month as well as area fixed effects (defined as 100x100 grid) *** p<0.01, ** p<0.05, * p<0.1
Proximity to cities and paid work (DHS) – check validity of measure
Conclusions 1.Employment effects are low Little job creation – large farms will not replace small holders, but depends on crop Commercial farms report 1 permanent worker / 20 ha Effects dissipates with distance 2.Evidence of technology spillovers, improved yields and improved access to inputs for some crops Identifying potential mechanisms for increased yields Greater use of improved seeds or chemical fertilizer may be channel for effect on yields
Conclusions & next steps Administrative data is the only way to make sense of LSLBI Drawing on and improving upon existing data sources can get you a long way Highly relevant policy questions for government can be addressed: Small improvements to CSA Commercial Farm survey we find evidence of technology spillovers and assess employment effects of commercial farms We hope to expand this work to other countries facing high demand for LSLBI