Agricultural Productivity and Structural Transformation.
Evidence from Brazil
Paula Bustos Bruno Caprettini Jacopo Ponticelli
Abstract
We study the effects of the adoption of newagricultural technologies on structural transformation. To guide empiricalwork, we present a simple model where the effect of agricultural productivityon industrial development depends on the factor bias of technical change. Wetest the predictions of the model by studying the introduction of geneticallyengineered soybean seeds in Brazil, which had heterogeneous effects onagricultural productivity across areas with different soil and weathercharacteristics. We find that technical change in soy production was stronglylabor saving and led to industrial growth, as predicted by the model.
Keywords: Agricultural Productivity,Structural Transformation, Industrial Development, Labor Saving TechnicalChange, Genetically Engineered Soy.
The early development literature documentedthat the growth path of most advanced economies was accompanied by a process ofstructural transformation. As economies develop, the share of agriculture inemployment falls and workers migrate to cities to find employment in the industrialand service sectors [Clark (1940), Lewis (1954), Kuznets (1957)]. Thesefindings suggest that isolating the forces that can give rise to structuraltransformation is key to our understanding of the development process. Inparticular, scholars have argued that increases in agricultural productivityare an essential condition for economic development, based on the experience ofEngland during the industrial revolution. Classical models of structural transformation formalize their ideas byshowing how productivity growth in agriculture can generate demand for manufacturinggoods. However, several scholars notedthat the positive effects of agricultural productivity on industrializationoccur only in closed economies, while in open economies a comparative advantagein agriculture can slow down industrial growth. Despite the richness of the theoretical literature, there is scarceempirical evidence testing the mechanisms proposed by these models.
In this paper we provide direct empiricalevidence on the effects of technical change in agriculture on the industrial sectorby studying the recent widespread adoption of new agricultural technologies inBrazil. First, we analyze the effects of the adoption of genetically engineeredsoybean seeds (GE soy). This new technology requires less labor per unit ofland to yield the same output. Thus, it can be characterized aslabor-augmenting technical change. In addition, we study the effects of theintroduction of a second harvesting season for maize (milho safrinha). Thistechnique permits to grow two crops a year, effectively increasing the landendowment. Thus, it can be characterized as land-augmenting technical change.The simultaneous expansion of these two crops allows to assess the effect ofagricultural productivity on structural transformation in open economies.
To guide empirical work, we build a simplemodel describing a two-sector small open economy where technical change inagriculture can be factor biased. The model predicts that a Hicks- neutralincrease in agricultural productivity induces a reduction in the size of theindustrial sector as labor reallocates towards agriculture, as in Matsuyama(1992). Similar results are obtained when technical change is land-augmenting.However, if land and labor are strong complements in agricultural production,labor-augmenting technical change reduces labor demand in agriculture andcauses workers to reallocate towards manufacturing. In sum, the model predictsthat the effect of agricultural productivity on structural transformation inopen economies depends on the factor-bias of technical change.
In a first analysis of the data we findthat regions where the area cultivated with soy expanded experienced anincrease in agricultural output per worker, a reduction in labor intensity inagriculture and an expansion in industrial employment. These correlations areconsistent with the theoretical prediction that the adoption oflabor-augmenting agricultural technologies reduces labor demand in theagricultural sector and induces the reallocation of workers towards theindustrial sector. However, causality could run in the opposite direction. Forexample: an increase in productivity in the industrial sector could rise labordemand and wages, inducing agricultural firms to switch to less labor intensivecrops, like soy.
We propose to establish the direction ofcausality by using two sources of exogenous variation in the profitability oftechnology adoption. First, in the case of GE soy, as the technology wasinvented in the U.S. in 1996, and legalized in Brazil in 2003, we use this lastdate as our source of variation across time. Second, as the new technology hada differential impact on yields depending on geographical and weathercharacteristics, we use differences in soil suitability across regions as oursource of cross-sectional variation. Similarly, in the case of maize, weexploit the timing of expansion of second-harvest maize and cross-regionaldifferences in soil suitability.
In particular, we obtain an exogenousmeasure of technological change in agriculture by using estimates of potentialsoil yields across geographical areas of Brazil from the FAO-GAEZ database.These yields are calculated by incorporating local soil and weathercharacteristics into a model that predicts the maximum attainable yields foreach crop in a given area. Potential yields are a source of exogenous variationin agricultural productivity because they are a function of weather and soilcharacteristics, not of actual yields in Brazil. In addition, the databasereports potential yields under traditional and new agricultural technologies.Thus, we exploit the predicted differential impact of the new technology onyields across geographical areas in Brazil as our source of cross-sectionalvariation in agricultural productivity. Note that this empirical strategyrelies on the assumption that goods can move across geographical areas ofBrazil, but labor markets are local due to limited labor mobility. Thisresearch design allows us to investigate whether exogenous shocks to localagricultural productivity lead to changes in the size of the local industrialsector. We use municipalities as our geographical unit of observation, whichare assumed to behave as the small open economy described in the model.
We find that municipalities where the newtechnology is predicted to generate a larger increase in potential yields ofsoy were indeed characterized by a faster adoption of GE soy. In addition,these regions experienced increases in the value of agricultural output perworker and reductions in agricultural labor intensity. Besides, the localindustrial sector was characterized by faster employment growth and reductionsin wages. Interestingly, the effects of technology adoption are different formaize. Regions where potential maize yields are predicted to increase the mostwhen switching from the traditional to the new technology did indeed experiencea higher increase in the area planted with maize. However, they alsoexperienced increases in agricultural labor intensity, reductions in industrialemployment and increases in wages.
The different effects of technologicalchange in agriculture documented for GE soy and maize indicate that thefactor-bias of technical change is a key determinant of the relationshipbetween agricultural productivity and structural transformation in open economies.Land-augmenting technical change, the case of second-harvest maize, leads to anincrease in the marginal product of labor in agriculture and a reduction inindustrial employment. However, labor-augmenting technical change, the case ofGE soy, leads to a reduction in the marginal product of labor in agricultureand employment growth of the industrial sector. Thus, in what follows we referto labor-augmenting technical change as labor-saving.
Our estimates can be used to quantify the effectof local labor-saving agricultural technical change on local structuraltransformation. In particular, we compute the elasticity of local sectoralemployment shares to changes in agricultural productivity induced by soytechnical change: 1 percent increase in agricultural labor productivity leadsto a 0.16 percentage points decrease in the agricultural employment share andan increase in the manufacturing employment share of a similar magnitude. Theseestimates can be used to understand to what extent the observed differences inthe speed of structural transformation across Brazilian municipalities can beexplained by labor- saving technical change in soy. In the year 2000, theaverage municipality had employment shares in agriculture and manufacturing of38 and 10 percent, respectively. During the next decade, the degree of laborreallocation across sectors varied extensively across municipalities. Ourestimates imply that labor-saving technical change in soy can explain 24percent of the observed differences in the reduction of the agriculturalemployment share across Brazilian municipalities and 31 percent of thecorresponding differences in the growth of the manufacturing employment share.
We assess the robustness of our estimatesto a number of deviations from our baseline framework. First, estimates arestable when we augment our empirical specification to allow municipalities withdifferent initial levels of development to be on differential structuraltransformation trends. Second, we obtain similar estimates in the subsample ofBrazilian municipalities where the agricultural frontier did not expand. Third,our estimates are not driven by pre-existing trends in manufacturing employmentnor migration flows. Fourth, our results are robust to using a larger unit ofobservation, micro-regions. Fifth, at least 60 percent of our estimated effectof agricultural technical change on the manufacturing employment share is notdriven by the processing of soy and maize in downstream industries nor largeragricultural sector demand for manufacturing inputs. Sixth, our estimates arenot driven by contemporaneous changes in commodity prices. Seventh, our mainestimates remain statistically significant when we correct standard errors toaccount for spatial correlation.
We complement our findings with an analysisof the service sector. For this purpose, we extend the theoretical model byincorporating non-traded services. A central feature of the analysis is thedistinction between two effects of agricultural technical change: the supplyeffect and the demand effect. When technical change is land-augmenting, thesupply effect is generated by an increase in the marginal product of labor inthe agricultural sector, which draws workers out of other sectors. In turn, thedemand effect is generated by the higher income resulting from technical changein agriculture which leads to increased consumption of non-traded services.Both effects lead to a reallocation of labor away from the manufacturingsector. However, when technical change is labor-saving, the supply effectreleases agricultural workers. As a result, in this case the net effect ofagricultural technical change on industrialization depends on the relativestrength of the supply and demand effects. In addition, the demand effect isonly driven by an increase in land rents. Thus, its strength depends on theextent to which land-owners consume services in the region where their land islocated. When we turn to the data, we find that local labor-saving technicalchange does not significantly affect local employment in the service sector.Note, however, that these findings do not necessarily imply that agriculturaltechnical change did not have an effect on the demand for services in theaggregate Brazilian economy. This is because the differences-in-differencesempirical strategy is not suitable to identify aggregate demand effects whenland owners do not reside locally or consume services in other regions. Thus, afurther investigation of the effect of agricultural technical change on theaggregate demand for services is left for future work.
Finally, we investigate the impact ofagricultural technical change on migration flows. In our model labor is assumedto be immobile across municipalities, thus all the adjustment to labor- savingtechnological change occurs through a reallocation of labor towards themanufacturing sector. However, if workers could reallocate to othermunicipalities, some of this adjustment would occur through out-migration.Indeed, we find that municipalities with larger increases in potential soyyields experienced a net outflow of migrants between 2000 and 2010. Ourestimates imply that the presence of migration flows across municipalitiesdampens the effects of technical change on sectoral employment shares, asaround a third of the adjustment occurs through migration flows.
Related Literature
There is a long tradition in economics ofstudying the links between agricultural productivity and industrialdevelopment. Nurkse (1953), Schultz (1953) and Rostow (1960) argued thatagricultural productivity growth was an essential precondition for the industrialrevolution. Classical models of structural transformation formalized theirideas by proposing two main mechanisms through which agricultural productivitycan speed up industrial growth in closed economies. First, the demand channel:agricultural productivity growth rises income per capita, which generatesdemand for manufacturing goods if preferences are non-homothetic. The higherrelative demand for manufactures generates a reallocation of labor away fromagriculture [Murphy, Shleifer, Vishny (1989), Kongsamut, Rebelo and Xie (2001),Gollin, Parente and Rogerson (2002)]. Second, the supply channel: ifproductivity growth in agriculture is faster than in manufacturing and thesegoods are complements in consumption, then the relative demand of agriculturedoes not grow as fast as productivity and labor reallocates towardsmanufacturing [Baumol (1967), Ngai and Pissarides (2007)].
The view that increases in agriculturalproductivity can generate manufacturing growth was challenged by scholarsstudying industrialization experiences in open economies. These scholars arguedthat high agricultural productivity can retard industrial growth as laborreallocates towards the comparative advantage sector [Mokyr (1976), Field(1978) and Wright (1979)]. Their ideas were formalized by Matsuyama (1992) whoshowed that the demand and supply channels are not operative in a small openeconomy that faces a perfectly elastic demand for both goods at world prices.The open economy model we present in this paper differs from Matsuyama’s in onekey dimension. In his model, there is only one input to production thustechnical change is, by definition, Hicks-neutral. In our model there are twofactors, land and labor, and the two are complements in agriculturalproduction. Thus technical change can be factor-biased. In this setting, a newprediction emerges: when technical change is labor augmenting, an increase inagricultural productivity leads to a reallocation of labor towards theindustrial sector even in open economies.
Our work builds on the empirical literaturestudying the links between agricultural productivity and economic development. Theclosest precedent to our work is Foster and Rosenzweig (2004, 2008) who studythe effects of the adoption of high-yielding-varieties (HYV) of corn, rice,sorghum and wheat during the Green Revolution in India. To guide empiricalwork, they present a model in which agricultural and manufacturing goods aretradable and technical change is Hicks-neutral. Consistent with their model, theyfind that villages with higher improvements in crop yields experienced lowermanufacturing growth. Our findings are in line with theirs in the case ofmaize, for which technical change is land-augmenting. However, we find theopposite effects in the case of soy, for which technical change is laborsaving. Thus, relative to theirs, our work highlights the importance of thefactor-bias of technical change in shaping the relationship betweenagricultural productivity and industrial development in open economies.
Our research also connects to theliterature studying the role of manufacturing in economic development. Thisliterature has shown that a reallocation of labor into manufacturing canincrease aggregate productivity: first, when labor productivity is lower inagriculture than in the rest of the economy [Gollin, Parente and Rogerson(2002), Lagakos and Waugh (2013) and Gollin, Lagakos and Waugh (2014)]; second,when the manufacturing sector is characterized by economies of scale generatedby on-the-job accumulation of human capital such as learning-by-doing [Krugman(1987), Lucas (1988), Matsuyama (1992)].
Our treatment of services in the modelfollows the literature on the Dutch Disease: Corden and Neary (1982) andKrugman (1987). In particular, Corden and Neary consider a three-sector openeconomy model with non-traded goods. One of the traded sectors is extractiveand experiences a boom, which leads to de-industrialization and an expansion ofthe service sector. We build on their distinction between two effects of theboom: the spending effect and the resource movement effect, which we call thedemand and supply effects. Our setting differs in that we consider labor-savingtechnical change which reduces the marginal product of labor in the booming sector,agriculture. Thus, in our model the net effect of agricultural technical changeon industrialization depends on the relative strength of these effects.
Finally, our work is related to recentempirical papers studying the effects of agricultural productivity onurbanization [Nunn and Qian (2011)], the links between structuraltransformation and urbanization [Michaels, Rauch and Redding (2012)], theeffects of agriculture on local economic activity [Hornbeck and Keskin (2012)],and the role of out-migration from rural areas in favoring the adoption ofcapital-intensive agricultural technologies [Hornbeck and Naidu (2014)].
The remaining of the paper is organized asfollows. Section I gives background information on agriculture in Brazil.Section II presents the theoretical model. Section III describes the data.Section IV presents the empirical strategy and results. Section V shows a setof robustness checks on our main results. Section VI concludes.
I. Agriculture in Brazil
In this section we provide backgroundinformation on recent technological developments in Brazilian agriculture. Inparticular, we focus on two new agricultural technologies for the cultivationof soy and maize. The first is the use of genetically engineered (GE) seeds insoy cultivation. The second is the introduction of a second harvesting seasonfor maize during the same agricultural year, which requires the use of advancedcultivation techniques.
A. Technical Change in Soy: Genetically Engineered Seeds
The main advantage of GE soy seeds relativeto traditional ones is that they are herbicide resistant, which facilitates theuse of no-tillage planting techniques.9 The planting of traditional seeds ispreceded by soil preparation in the form of tillage, the operation of removingthe weeds in the seedbed that would otherwise crowd out the crop or competewith it for water and nutrients. In contrast, planting GE soy seeds requires notillage, as the application of herbicide selectively eliminates all unwantedweeds without harming the crop. As a result, GE soy seeds can be applieddirectly on last season’s crop residue, allowing farmers to save on productioncosts since less labor is required per unit of land to obtain the sameoutput.10
The first generation of GE soy seeds, theRoundup Ready (RR) variety, was commercially released in the U.S. in 1996 bythe agricultural biotechnology firm Monsanto. In 1998, the Brazilian NationalTechnical Commission on Biosecurity (CTNBio) authorized Monsanto to field-testGE soy in Brazil for 5-years as a first step before commercialization. Finally,in 2003, the Brazilian government authorized the planting and commercializationof GE soy seeds. Prior to legalization,smuggling of GE soy seeds from Argentina was detected since 2001 according tothe Foreign Agricultural Service of the United States Department of Agriculture(USDA, 2001).
The new technology was characterized byfast adoption rates: in 2006 GE seeds were planted in 46.4 percent of the areacultivated with soy in Brazil, according to the last Agricultural Census (IBGE,2006). In the following years the technology continued spreading to the pointthat, according to the Foreign Agricultural Service of the USDA, it covered 85percent of the area planted with soy in Brazil by the 2011-2012 harvestingseason (USDA, 2012).
The timing of adoption of GE soy seedscoincides with an increase in labor productivity and a fast expansion in thearea planted with soy in Brazil. Figure 1(a) documents that soy laborproductivity has been increasing in Brazil since the early 1990s, andaccelerated sharply in the early 2000s: soy production per worker went from 100tones per worker in 2003 to around 300 tones per worker in 2011. Laborproductivity growth was accompanied by an expansion in the area planted with soy.Table 1 reports land use by agricultural activity according to the 1996 and2006 Agricultural Censuses. It shows that the area cultivated with seasonalcrops increased by 10.3 million hectares between 1996 and 2006. Out of these, 8.7 million hectares wereconverted to soy cultivation. Similarly, Figure 1(b) shows that the areaplanted with soy has been growing since the 1980s, and experienced a sharpacceleration in the early 2000s.
The adoption of GE soy can affect labordemand in the agricultural sector through two channels: the within-crop and theacross-crop effects. The first effect is due to a reduction in the amount ofagricultural workers per hectare required to cultivate soy: labor intensity ofsoy production fell from 29 workers per 1000 hectares in 1996 to 17 workers per1000 hectares in 2006 (Table 1). The timing of this change in labor intensityis illustrated by Figure 1(c), which shows a sharp increase in the area plantedper worker in soy production in the early 2000s. This reduction in labor intensity was strongenough to entirely offset the potential increase in labor demand for soy due tothe expansion in the area planted. As a result, employment in soy productionexperienced a constant decrease during the period under study [Figure 1(d)].
The second channel through which theadoption of GE soy can affect labor demand is the across- crop effect. Thiseffect is due to the expansion of soy cultivation over areas previously devotedto other crops. This effect reduces the labor intensity of production in theagricultural sector because soy production is one of the least labor-intensiveagricultural activities: its production required 17 workers per 1000 hectareswhile seasonal crops and permanent crops require 84 and 127, respectively (Table1).
B. Technical Change in Maize: Second Harvesting Season
During the last two decades Brazilianagriculture experienced also important changes in maize cultivation. Maize usedto be cultivated during the spring season, between August and December. At thebeginning of the 1980s a few farmers in the South-East region of Brazil startedproducing maize after the summer harvest, between March and July. This secondseason of maize cultivation spread across Brazil, where it is known as milhosafrinha (small-harvest maize). Figure 1(e) shows that the area devoted tosecond season maize has expanded steadily since the beginning of the 1990s,although the total area devoted to maize has increased only slightly.
Cultivation of a second season of maizerequires the use of modern cultivation techniques. First, more intensiveland-use removes nitrogen from the soil, which needs to be replaced byfertilizers. Second, the planting of a second crop requires careful timing, asyields drop considerably due to late planting. Third, herbicides are used toremove residuals from the first harvest on time to plant the second crop.Finally, the second season crop needs to be planted one month faster than thefirst, which usually requires higher mechanization.
The introduction of a second harvestingseason for maize can affect labor demand in the agricultural sector through thewithin-crop and across-crop effects described above. The first effect isdirectly due to the introduction of a second harvest which raises labor demandrelative to the benchmark of a single maize harvest. The second effect is dueto the expansion of maize over areas previously dedicated to less-laborintensive activities, which also tends to increase labor demand. According tothe 1996 Agricultural Census, maize cultivation is more labor intensive thanthe main agricultural activities in Brazil. In this year, labor intensity inmaize production was 100 workers per 1000 hectares, above the labor intensityof soy, other cereals and cattle ranching.