In a stylized story of green revolutions, improvements in agricultural technology are achieved through the introduction of improved land management techniques or improved inputs, including germplasm and fertilizer, all of which boost yields and labor productivity . If food is relatively non-tradable beyond local markets, then increased staple food production leads to reduced food prices, increased real wages and hence lower poverty. As staple yields jump and basic food needs are met, crop production begins to diversify, including to nonfood cash crops for export, and so the virtuous cycle of commercial farming begins. With greater savings and access to finance, farms begin to substitute capital for labor, and freed up workers begin to look for wage employment, typically in nearby cities. To the extent that other sectors enjoy higher labor productivity, this is welfare enhancing. It is also possible that this structural change triggers even further increases in non-agricultural labor productivity. One potential mechanism is that after subsistence is surpassed, savings rates increase, and the subsequent capital accumulation increases worker productivity . In parallel, governments are able to collect revenues to finance growth enhancing infrastructure, such as roads and ports, which increases the worker productivity of manufacturing and services. Another mechanism may be that increased incomes improve health outcomes,flood and drain tray which increase worker productivity, while also decreasing child mortality, reducing total fertility rates, increasing investment per child, and decreasing demographic pressures.
Or, it may simply be that the non-agricultural sector enjoys increasing returns to scale due to fixed costs or learning-by-doing, which would imply that a green revolution and the resulting labor shift would accelerate productivity growth in these non-agricultural sectors. Although our paper will not be able to pinpoint which of these mechanisms is at work, our contribution is to provide a causal framework to evaluate whether higher staple yields trigger labor shifts away from agriculture as well as faster growth in non-agricultural labor productivity. For the purposes of illustration and to motivate our empirical work more specifically, we describe agriculture-driven structural change with a simple model following the long theoretical tradition starting with works including Rostow ; Johnston and Mellor and formulated mathematically by Laitner ; Hansen and Prescott ; Gollin et al. , and others. We start with a country that has no trade in staple food products, and where the entire population works in either the agriculture or non-agriculture sector . The model is dynamic, but we dispense with the time subscript for simplicity of exposition. The stylized facts support the theoretical link between staple crop yields link and economic growth. Figure 1 shows indexed regional trends in food production per capita across the developing world from 1961-2001.The graph highlights the major growth in East Asia and the Pacific over the period, with per capita values nearly doubling, and considerable growth in Latin America and South Asia since the mid-1970s. Africa is the one region to have experienced a decline in per capita food production over the period, including a major decrease since the early 1970s and relative stagnation since 1980. These trends are mirrored in Figure 2, which presents cereal yields per hectare from 1961-2001. Again, all developing regions except Africa experienced major sustained growth rates in land productivity over the period, despite varying starting points, and all except Africa more than doubled yields by 2001.
East and Southeast Asia boosted yields from less than 1.5 tons per hectare in 1961 to more than 4 t/ha in 2001; Latin America’s yields grew from 1.3 t/ha to greater than 3 t/ha; and South Asia’s from 1 t/ha to nearly 2.5 t/ha. Africa had the lowest starting point at 0.8 t/ha, and still after 40 years had barely crossed the threshold of 1 t/ha, which was South Asia’s starting level in 1961. A simple Boserup hypothesis would argue that, relative to other regions, Africa’s yield stagnation is a product of its land abundance, and yields will increase as land becomes scarce. There are three main reasons why this hypothesis does not hold, as described in McArthur . First, the history of 20th century yield take-offs in the developing world was predominantly characterized by proactive public policies supporting a package of yield-boosting inputs, rather than by factor scarcity . These policies are thought to explain much of the regional variations in fertilizer use since 1960, as shown in Figure 3. Second, labor/land ratios vary tremendously across Africa but they are just as high or higher in many African countries than they were in pre-green revolution Asian countries. Third, land productivity is driven by the crucial latent variable of soil nutrients, which are being depleted at dramatic rates throughout Africa. High rates of soil nutrient loss strongly suggest that land pressures are not being surmounted by extensification.Figure 4 compares the growth of cereal yields to growth in GDP per capita over the 1965 to 2001 period, indicating a strong positive correlation between the two variables. A novel relationship is presented in Figures 5 and 6, which compare initial cereal yield levels to subsequent GDP growth across developing countries, excluding fuel exporters and socialist economies.Figure 5 covers the full 1965 to 2001 period and Figure 6 covers only the latter portion from 1985 to 2001. The horizontal line marks zero average growth and the vertical line marks 2 t/ha of cereal yields. In addition to the overall positive relationship between initial yield and economic growth, it is noteworthy that no country in the sample experienced negative average growth after reaching a yield threshold of 2 t/ha.3 Figure 7 presents a scatter plot similar to Figure 4 but shows growth in non-agricultural value added per non-agricultural worker on the vertical axis instead of GDP per capita, covering the period 1970-2001.
The graph shows a clearly positive relationship between the two variables, even amidst a considerable degree of variation, and suggests that higher rates of progress in agricultural productivity are structurally correlated with higher growth rates in non-agricultural sectors.This paper’s empirical strategy proceeds in two parts. The first focuses on establishing a country-level physical production function for cereal yields , in order to motivate the emphasis on agronomic inputs in a study of structural change. The second part focuses on identifying the impact of increased yields on economic outcomes and structural change, measured by GDP per capita, labor shares and non-agricultural value added per worker.It was chosen over log-linear and log-log approaches since neither of the latter were found to provide a better fit with the data, and indeed most countries with significant input use have pursued relatively linear fertilizer-yield trajectories, as shown in Figure 8. This linear relationship is somewhat at odds with the field-level agronomic data that show decreasing returns, but is likely an inherently limited product of the country-level unit of aggregation. This paper aims to present a first approximation of a country-level agricultural production function, which to our knowledge has not been previously done in the economics literature. Future research would be well placed to provide more refined estimates anchored in more specific crop types and input combinations, the latter captured for example through a range of possible interaction terms. With these points in mind, this paper’s regression results provide information only on marginal additive effects of various inputs. One might hesitate to interpret associations between agronomic inputs and yields in a causal framework; indeed, omitted variables such as farmers’ agronomic know-how might be correlated with both yields and inputs and thus bias coefficients in the estimation. In order to assuage these concerns and improve identification in the case of fertilizer use, we construct a novel time-varying instrument. Our approach follows a similar spirit to the instrument presented in Werker et al. . A valid instrument needs to be correlated with countries’ fertilizer use and satisfy the exclusion restriction . We use fluctuations in the global fertilizer price to generate temporal variation exogenous to conditions in any one developing country. In order to generate the cross-sectional variation in the instrument we exploit the fact that the production of nitrogen fertilizer is intensive in natural gas usage and therefore produced in only a select group of facilities around the world, most of which are in developed countries. We contend that the distance fertilizer travels from these facilities to the agricultural heartlands of each developing country is valid cross-sectional variation that can be interacted with the global fertilizer price to generate a valid instrument for fertilizer use in developing countries. Specifically, we hypothesize that countries closer to fertilizer plants are more sensitive to the commodity’s price variation relative to the transport costs that farmers incur. The instrument satisfies reverse causality concerns ,nft hydroponic and the omitted variable bias concern is assuaged since a problematic omitted variable would need be to correlated with the global fertilizer price and have the same distance decay function from agricultural heartlands to global fertilizer production facilities. A specific concern that a reader might have is that fertilizer price fluctuations might be correlated to fossil fuel prices, which might affect economic outcomes through many channels. However, the correlations between crude oil prices and phosphate, DAP, urea and potash prices are only between 0.11 and 0.38 over the period . Moreover, the correlation is only problematic if the specific distance decay function we use from agricultural centroids to nitrogen facilities matches the pattern of cross-country differences in fossil fuel prices, and there is no reason to believe that this will be the case. We use a Geographic Information System to calculate the agriculturally weighted centroid of each country, using data on percentage of each 5 arc-minute grid cell’s area planted to staple crops from Monfreda et al. . Next, we geolocate 63 of the production facilities of the top fertilizer producers in the world .
Although these are present-day facilities , we remind the reader that most facilities are located in developed countries not in our sample, and many locate in proximity to natural gas deposits, so the issue is unlikely to have a big effect on our results. We then calculated the minimum cost adjusted distance from each country’s agriculturally weighted centroid to the nearest fertilizer production site. In order to adjust for relative transport cost between land and water, we use Limão and Venables’ result that shipping a standard 40-foot container from Baltimore to different destinations around the world in 1990 costs $190 for an extra 1,000 km by sea and $1,380 for an extra 1,000 km by land. This indicates roughly a 1:7 cost ratio, which we use to optimize travel over sea and navigable rivers versus travel over land. The centroids, fertilizer production sites and optimal cost-distance function are mapped in Figure 9. The distance component of the instrumental variable is itself strongly correlated with fertilizer use across countries, as shown in Figure 10, which plots the log of fertilizer use per hectare at the 1985 sample midpoint against the indexed distance measure. The correlation between the two variables in the graph is -0.63. Towards the top left of the scatter plot, a country like Vietnam has an distance index value of 3,954 and a fertilizer value of 84 kg/ha, while Rwanda , towards the bottom right, has a distance value of 13,083 and a fertilizer value of 1.7 kg/ha. It is trivial for higher agricultural productivity to be linked to higher economic growth in the same period, since agricultural output is included directly in national accounts. For example, if one holds fixed all prices and production levels in other sectors, a green revolution-style five-year doubling of output in a low-income country with 30 percent of GDP in food production would translate mechanically to a 5.4 percent annual real GDP growth rate.For a country with only 15 percent of GDP in food production, the same yield doubling would translate to 2.8 percent annual growth. Of course a major supply expansion would be expected to decrease the price of food, and the nominal measured growth rate would be much smaller—so 5 or 6 percent could be considered an upper bound on the direct contribution of increasing yields to economic growth.