Lorebgmipllsumdolyorsltzmhhabc Land Governance POLICY BRIEF WORLD BANK ✦ DEVELOPMENT RESEARCH GROUP ISSUE 3 ✦ JANUARY 2018 Does large farm establishment create benefits for neighboring smallholders? Evidence from Ethiopia ✳✳ Daniel Ali, Klaus Deininger, and Anthony Harris Quantifying the direction and magnitude of from those higher up the value chain. Spill- too small to defray the costs, even without spillovers for small holders is an essential part overs can be positive or negative. Examples credit constraints. of the policy dialogue surrounding large-scale of the latter include land made available agricultural investment. We use intertemporal without proper valuation and compensa- The presence of large farms may assist variation in smallholders’ proximity or intensi- tion, or new investors monopolizing factor neighboring smallholders to overcome ty of exposure to large farms, while exploring markets or encroaching on land or water these obstacles. To the extent that they use the variation in large farm establishment over resources. these, large farms can affect smallholders’ time and space, to analyze the presence and use of technology and equipment, level magnitude of spillovers between large and Quantifying the direction and magnitude and intensity of input use, factor market small farms. Findings show that between of local spillovers is therefore essential. Yet, participation, and resilience to shocks. 2004 and 2014 new formation of commercial despite this interest, efforts at systematic farms did not contribute to job creation and quantification for agricultural production, In this study, we examine two types of provided at best modest benefits for neighbor- beyond individual case studies, have been spillovers (i) crop specific spillovers, where ing smallholders in terms of technology and rare. Careful empirical assessment based on transmission occurs between smallholders access to inputs. This implies that in Ethiopia a clear methodology, with a representative and large farms growing the same crop, a more strategic approach may be required and rigorous quantification of the under- and (ii) generic spillovers arising between to maximize smallholder benefits from large lying processes, is key to providing clarity smallholders and large farms regardless of farm formation. Our methodology has proven on spill overs. crops, which may capture improvements in to be robust and can be applied to study spill- input market access, marketing assistance over effects of large-scale commercial farming This policy brief presents a methodology for or increased local labor demand. more generally. Avenues to do so are outlined. quantifying spillovers for large-scale agri- cultural investments by analyzing spatial For crop specific spillovers, we focus on proximity as the main channel of trans- maize, wheat, sorghum, and teff, four key Introduction mission, while exploring variation in large cereals grown by many large and small farm establishment over time and space to farms across Ethiopia. Smallholders and The 2007/08 food price spike, together identify causal impacts. Using detailed data large farms report clear differences in with the recognition that some countries on commercial farms and smallholders for yields and input use intensities for these in Africa are endowed with large amounts Ethiopia, we estimate mean effects of estab- crops, except for teff. A sizable share of large of seemingly unoccupied or unclaimed lishing new large farms on key agricultural farms growing these crops report provid- land, triggered a marked increase in private indicators for neighboring smallholders. ing technical support to their smallholder sector demand for agricultural land to satisfy Ethiopia was selected, as it has a long his- neighbors. Both facts suggest significant increasing demands for food, fuel, and fiber. tory of large farm establishment, availability scope for spillovers. Although often described as ‘land grab’, the of good data, and plays a preeminent role expansion of large commercial farms also in recent debates on this topic. gave rise to expectations of private capital Data sources to complement public investment, making up for decades of underinvestment in agri- Inventory of possible A first set of data is the Ethiopia’s Central Sta- culture. It prompted a lively policy debate spillovers tistical Agency (CSA) smallholder agricultural regarding the direct or indirect effects from production survey conducted annually by such investments towards more rapid rural High transaction costs or ignorance of resident enumerators for a sample of some development and poverty reduction for potential benefits can discourage small- 1,400 kebeles nation-wide and covers the countries with ample land resources. holders from using certain inputs or period between 2003/4 and 2013/14. Data technologies, even if the benefits of doing on input use is collected from a random A key argument in the policy dialogue so exceed the cost. If transport or other sample of 20–40 farmers per kebele, i.e. a revolves around local spillover effects from transaction costs are high, smallholders total sample of 28,000 to 56,000 farmers private investment in agricultural produc- may be rationed out of input and output per year. However, data on yields is based tion, which are likely to be quite different markets as quantities demanded may be on crop cuts of randomly selected fields in each EA rather than of fields belonging to As CSA redrew the sample of kebeles for the for Ethiopians and 840 ha for foreigners). By the sampled farmers. This makes it impos- smallholder data only once (in 2007/08), we respondents’ own estimates, 55% of land sible to estimate farmer-specific production link survey rounds across years to obtain transferred remains unutilized, largely due functions and forces us to analyze yields at two kebele level panels for the 2004–2007 to technology and labor constraints. kebele level instead. The survey also lacks and the 2008–2014 period with about 500 data on farmer’s use of labor, preventing and 2,000 kebeles, respectively. Investments made focus on land clearing us from using these surveys to explore this and machinery rather than public goods, potentially important channel for large We link each large farm’s year of establish- and less than 20% of farms accessed credit farms to affect smallholders. Instead, we use ment to all rounds of the smallholder survey themselves. data on individual labor supply collected in to generate kebele-specific measures of the 2011/12 and 2013/14 LSMS-ISA panel proximity and exposure to large farms in to estimate local labor market effects from each year. The time variation in smallholder Measuring spillovers large farm establishment. kebeles’ proximity to large farms, overall or for specific crops, over the 2004–14 period Density functions for changes in distance The second data source is the 2014 round is used to identify if smallholders’ fertilizer between smallholder kebeles and the next of the annual CSA census of all operational use, yields, employment, or resilience are large farm over time overall illustrate the large farms. The category of large commer- affected by large farm establishment. rapid expansion of large farms over the cial producers is defined as all those with study period, with a marked change over a size over 10 hectares. It is important to the 2004–2010 period, corresponding to note that flower farms are not included in Large-scale land based new large farms establishment. Commercial the sample of CSA’s large and medium scale investment in Ethiopia maize farms spread from the center of the commercial farms survey. This second CSA country leaving only kebeles in Gambella survey covers a sample of all large farms in In Ethiopia, historic attempts at establish- and Tigray at a distance of more than 150 the 10–50 ha category and the universe ing large farms were mostly unsuccessful. km from the next large farm. In 2004, large of farms holding 50 ha or more. Farm Before 1974, subsidies were used to attract level information is provided on input use, commercial investment in so-called ‘model Table 1: Descriptive output, and year of establishment. GPS farms’, whose establishment was often coordinates taken for every field allow us associated with tenant evictions, mediocre statistics for large farms to map the universe of all farms above 50 productive performance and little employ- above 50 ha ha that were operational in 2014. ment generation. These failures led to a decision in the 1990s to rely on smallholder Total Finally, to account for inter-temporal agriculture as the main driver of develop- Size & establishment year variability in climatic conditions, we use ment. This changed in the Government’s Size cultivated (ha) 266.66 gridded 0.1’ rainfall data publicly available 2010–2015 Growth and Transformation Plan, from NOAA since 1980 to compute long- which views large farm investment as a stra- Before 91 (%) 4.68 term mean and standard deviation of tegic priority, asserting that capital intensive 1991–92 (%) 3.49 precipitation for each pixel. investment is the only way to bring such 1992–2002 (%) 21.50 land to productive use and generate spill- 2002–2006 (%) 22.61 over benefits for smallholders. Methodology for 07–2008 (%) 24.98 quantifying spillovers The Ethiopia’s CSA annual census of oper- 09–2010 (%) 14.54 ational large farms suggests that, since the 11–2013 (%) 8.20 As discussed, spillover channels rely on 1990s, about 1.3 mn. ha has been trans- Ownership type (%) physical proximity. Our measure of proxim- ferred to a total of 6,612 commercial farms, ity, constructed for each kebele and year, is some 78% of which own more than 50 Government 2.74 the distance to the nearest large farm or the ha. The annual rate of farm establishment Private 92.39 next large farm growing the same crop. Our dropped from a peak of close to 800 in Cooperative 4.61 alternative measure of exposure is the total 2007/08 back to the pre-2007 level of some large farm area in concentric circles or rings 250 in 2011/13. Ethiopian 96.62 around the kebele with inner and outer radii Foreign 2.67 of 0 to 25, 25 to 50 and 50 to 100 km. More land may have been transferred as Joint 0.71 only operational farms are included in Type of acquisition (%) Kebele centroids are used as a proxy for the the CSA survey. A review of some 10,600 location of households in the agricultural investment licenses by the Agricultural Direct negotiation 6.83 sample survey. For any kebele and year in Investment Agency suggests that less than Woreda 51.32 the smallholder sample, we compute the 20% of licensees ever established a farm. Region 37.17 distance from the kebele centroid to the Federal Government 4.68 centroid of the nearest large farm and the By 2014, 95% of land transferred for com- total area devoted to large farm cultivation mercial farms benefited Ethiopians or joint Number of obs. 3484 of a specific crop within a certain radius of ventures rather than foreigners (see Table Source: Own computation from 2013/4 CSA large farm the kebele centroid. 1). Mean farm size is about 200 ha (172 ha surveys. sorghum farms were clustered in Oromia, benefited or harmed neighboring small- incidence of fertilizer use due to spillovers eastern Afar, and the northwest of Amhara holders by affecting (i) their use of fertilizer; from large farm establishment. For teff, fer- spilling into Tigray. Although fewer than (ii) labor demand and job creation; and (iii) tilizer use by small producers is only slightly maize, their number expanded greatly by crop yields. We assume the main transmis- below that of commercial farms while their 2014. Large wheat farms expanded from sion channels are transfer of knowledge, yields are some 50% above those of the southern Amhara and Oromia and northern provision of access to markets for inputs and latter (table 3), suggesting the potential or SNNPR in 2004 to cover virtually the entire outputs, or implicit credit and insurance, the spillovers is low to begin with. Oromia and considerable parts of SNNPR former being particularly relevant if there and Amhara. Teff which, in 2004, was grown are gap in yields or input use between large While the above effects of commercial farms mainly in the central highlands, also has and small producers. establishment on incidence of chemical expanded greatly. fertilizer use may be driven predominantly Improved seeds and fertilizer have been by improved market access, impacts on Our data confirm a clear decrease in dis- identified as key to smallholder productivity neighboring smallholders’ yields would tance to the next commercial farm, from 78 in Ethiopia. Although our data suggest that reflect knowledge transfer or technology km in 2004 to 41 km in 2014 and an increase the use of improved seeds is significantly more directly. For maize and wheat, results of the number of large farms within a 0–25, higher in commercial than smallholder suggest that having a large farm area with 25–50, and 50–100 km radius. As a result, farming, the difficulty of unambiguously the same crop close to smallholders can the scope for interaction between large identifying improved seed types in stan- have yield-enhancing impacts. For teff and and small farms expanded. dard surveys leads us to focus our empirical sorghum coefficients are all insignificant. analysis only on fertilizer. Table 2 provides data on large farms over- all and the four crops considered. Mean For fertilizer use, a mar- Table 2: Descriptive statistics for large area for the farms considered here is 267 ginally significant positive farms above 50 ha for selected crops ha, with wheat farms largest and sorghum effects of large farm estab- and teff farms smallest. About 25% of farms lishment was found for Total Maize Sorghum Wheat Teff were established during 2007/08 alone. maize but not for other Size & establishment year With 55 ha per permanent employee, the crops. For smallholders Size cultivated (ha) 266.66 377.90 234.56 551.74 237.60 average farm employs less than 5 perma- producing maize, a reduc- nent workers. Employment intensity varies tion in mean distance to Before 2002 (%) 29.67 18.4 32.26 46.87 13.1 by crop, with wheat farms least and teff the next commercial farm, 2002–2006 (%) 22.61 19.70 27.26 8.52 18.70 farms most employment intensive. Over- would be predicted to 07–2010 (%) 39.52 54.31 34.35 39.78 57.92 all, 37% of farms report paying lease fees, increase the share of pro- ranging from 73% of wheat farms to 30% ducers applying fertilizer 11–2013 (%) 8.20 7.59 6.15 4.83 10.28 of sorghum farms. With only 6% reporting by 0.7 percentage points Employment current outstanding loans, access to formal with a further decrease Hectares per perm. worker 56 55 83 254 32 credit by commercial farms seems modest. from 40 km to 2 km pre- Temp. workers/ha 3.56 2.82 3.70 1.32 1.70 Provision of advice and technical assistance dicted to increase fertilizer as a channel for spillovers is supported by use by an additional 4.6 Fees, investment, credit reports from large farmers who state that percentage points. Effects Lease fee reported (%) 36.74 58.54 29.96 73.31 37.24 they provide advice to smallholders on a are more pronounced … if yes Birr/ha 530 327 274 1518 203 range of topics, topped by 20% on fertilizer for distance to the next Other payments. (%) 11.37 24.09 9.11 19.01 12.53 or seed varieties. large farm growing maize where equivalent shifts … if yes Birr/ha 376 464 218 180 170 Table 3 provides information on area cul- would be predicted to be Any investment (%) 93.11 90.18 98.71 80.45 89.43 tivated, yield and fertilizer use by crop for associated with increases Any loan last 5 years (%) 20.68 17.10 24.62 14.92 16.98 smallholders as well as commercial farms of 2.5 or 3.9 percentage of different sizes. The relationship between points, respectively, in the .. if yes Birr/ha 18,195 16,548 17,039 39,997 12,099 yields and farm size is non-linear; yields are propensity to use fertilizer. Provided advice in 2015 to smallholders on… highest for farms in the 10–20 ha range Results suggest that spill- Fertilizer 0.204 0.372 0.202 0.294 0.452 while those above 20 ha often obtain yields over effects are limited to Irrigation 0.071 0.154 0.045 0.066 0.173 20% below those attained by the top per- farms in very close proxim- forming group in each crop. For all crops ity; estimated coefficients New seed varieties 0.202 0.376 0.163 0.257 0.422 except teff, smallholders’ yields amount to on area beyond 25 km are Pests 0.146 0.258 0.166 0.125 0.341 75% or less of those attained by commer- not significantly different Soil problems 0.115 0.213 0.106 0.118 0.286 cial farms, creating prima facie potential for from zero while those on General cropping problems 0.177 0.297 0.171 0.191 0.351 spillovers, either through adoption of better both total and maize area technology or more intensive input use on in the 0–25 km range are Water issues 0.101 0.162 0.092 0.092 0.265 the part of smallholders. highly significant. Animal services 0.103 0.175 0.126 0.099 0.232 Marketing 0.106 0.192 0.111 0.072 0.255 We used the data to test if and to what Results for other crops extent large farm formation in Ethiopia suggest no increases in Number of obs. 3484 822 1659 194 323 With growing population and limited farm density sufficiently high to facilitate Methodologically, our paper demonstrates absorptive capacity in the non-agricultural interaction with small farmers. that linking information on start date and sector, the ability to create gainful employ- location of large farms to georeferenced ment is a key concern for policy makers These findings are subject to a few caveats. household or farm survey data and using and considerable hopes are pinned on First, our large farm data exclude flower and inter-temporal variation in large farm pres- employment creation from large farms horticultural farms, so spillovers estimated ence for identification allows to rigorously being integrated into agro-processing here are only for field crops. Second, as assess spillovers from commercial farm value chains. Using large farm area culti- data on commercial farms below 50 ha is establishment in ways that transcend the vated within different distance bands as the collected only on a sample basis, estimated limitations of case studies. In many countries independent variable, our analysis supports spillovers are from farms above 50 ha. Third, where commercial farming has become a the notion of large farms having no impact the benefits we compute can be interpreted policy priority, the data required to estimate on local labor markets. Data suggests that as social gains if the land now used by large spillovers in this way are available or could in marked contrast to, for example, flower farms was earlier unutilized. Survey data easily be generated for analysis at rather farms, large farms fail to significantly con- suggest this was not the case in about 30% low cost. This is a worthwhile investment to tribute to generation of paid jobs and, even of cases. Fourth, while our methodology inform policy given the far-reaching impact in their immediate vicinity, such farms can in principle be applied to any outcome of large farm expansion on agricultural have no perceptible effect on casual labor variable, lack of data on socio-economic structure and local livelihoods, reputational demand. outcomes precludes us from making infer- risk, and the possibility of shaping private ences on broader welfare effects. sector investment decisions by adapting and fine-tuning the regulatory framework Conclusion and policy Modest investments in additional data col- in light of actual outcomes. implications lection would allow to address these issues and enhance the policy relevance of these Exploring size and direction of spillovers findings for Ethiopia by (i) comparing size of from large farm establishment in Ethiopia, impacts of investing in large-scale farming contributes to the debate on commer- to those of other investments (irrigation, cial farming in Africa. During the 2004–14 establishment of agro-processing facilities, period, a rather uncoordinated process of direct technology transfer) to enhance commercial farm establishment provided smallholder integration into value chains; (ii) neighboring smallholders with at best exploring heterogeneity of effects by nature modest spillover effects in technology, of contractual provisions, investor type, or input market access, and resilience, but no farm size class, in particular for farms in the job creation benefits. Spillover effects are 10–20 ha group; and (iii) assessing spillover largely limited to maize, a crop where tech- effects on measures of welfare. nology is similar across farm sizes and large Table 3: Productive performance of smallholders vs. commercial farms in different farm size classes Maize Sorghum Teff Wheat Yield (Q/ha) Smallholder 27.07 21.29 13.62 21.85 Commercial farmers 37.69 27.01 8.29 25.58 Use fertilizer (%) Smallholder farmers 38.76 15.40 65.24 73.18 Commercial farmers 67.73 23.48 73.52 65.25 Observations Smallholder kebeles 1,368 910 955 634 Commercial farmers 1,659 3,077 826 464 Source: Own computation from 2013/4 CSA large farm and smallholder farm surveys This policy brief is based on Ali, Daniel Ayalew; Deininger, Klaus W.; Harris, Charles Anthony Philip. 2016. Large Farm Establishment, Smallholder Productivity, Labor Market Partic- ipation, and Resilience: Evidence from Ethiopia . World Bank Policy Research Working Paper; No. 7576. http://bit.ly/2DThF4G Daniel Ali is a senior economist and Klaus Deininger a lead economist at the World Bank, Washington DC; Anthony Harris works at Mathematica Policy Research, Cambridge MA. ✦  CONTACT: kdeininger@worldbank.org. The views presented are those of the authors and do not necessarily represent those of the World Bank, its Executive Directors or the member countries they represent.