Policy Research Working Paper 10414 D etox D evelopment : R epurposing E nvironmentally H armful S ubsidies Background Paper Fit for (re)Purpose? A New Look at the Spatial Distribution of Agricultural Subsidies Ebad Ebadi Jason Russ Esha Zaveri Sustainable Development Practice Group Office of the Chief Economist April 2023 Policy Research Working Paper 10414 Abstract Agricultural subsidies make up a large share of public bud- across 16 countries/regions and the distributional and select gets, exceeding 40 percent of total agricultural production environmental implications of input subsidies across 23 value in some countries. Subsidies are often important countries/regions. The findings show that, relative to the components of government strategies to raise agricultural spatial distribution of income, both types of subsidy are productivity, support agricultural households, and promote distributionally mixed. Output subsidies are relatively pro- food security. They do so by reducing production costs, gressive in 10 countries/regions and regressive in six, while promoting the use of inputs or modern farming techniques, input subsidies are relatively progressive in 11 countries/ encouraging the production of certain crops, and raising regions, regressive in nine, and neutral in three. The results household incomes. Given the magnitude of these subsidies, also show that input subsidy schemes significantly increase their distributional implications and the externalities they fertilizer use, particularly in richer regions within countries, impose on the environment are of significant consequence. leading to soil saturation of nitrogen, an indicator of accel- This paper uses a new spatial analysis to explore the dis- erated environmental degradation. tributional implications of agricultural output subsidies This paper is a product of the Office of the Chief Economist, Sustainable Development Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/ prwp. The authors may be contacted at xd2197@columbia.edu, jrentschler@worldbank.org, and jruss@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Fit for (re)purpose? A New Look at the Spatial Distribution of Agricultural Subsidies Ebad Ebadi, Jason Russ, Esha Zaveri Keywords: Agriculture subsidies, distribution, inequality, fertilizer, nitrogen pollution JEL codes: H23, Q01, Q18, Q53 1 Introduction Governments subsidize agriculture to the tune of US$635 billion per year (Gautam et al. 2022). These subsidies have many impacts; some intended, and some unintended. Agricultural subsidies are often intended to increase agricultural efficiency by incentivizing the increased use of inputs such as fertilizer, improved seeds, or water from irrigation, or encouraging increased production of certain crops to achieve domestic self-sufficiency goals (Ellis 1992). Subsidies are also often part of rural development, livelihood, or safety net programs. Indeed, with 80 percent of the world’s extreme poor and 75 percent of the moderate poor living in rural areas (Castaneda et al. 2016) and engaging largely in agriculture, such subsidies can act as crucial income support for the rural poor, reducing production costs through input subsidies and increasing the marginal revenue of production through output subsidies. In many countries, these subsidies make up significant shares of both government budgets and agricultural production value. In the 53 countries and regions 1 with available data on subsidies, the average estimated total subsidy to agriculture—including input and output subsidies, and public service provision such as research and development, infrastructure, and markets—is 18 percent of total agricultural production value. In some countries, it exceeds 40 percent (figure 1). Given the magnitude of these subsidies their distributional implications are of significant consequence. An important indicator of their impact on society, then, is their progressiveness, or the share that is captured by the lower versus higher income quantiles. But global estimates on the equity impacts of agricultural subsidies are lacking, particularly in developing countries. In richer countries, such as Canada, the United States, and European Union (EU) members, evidence suggests that support to agriculture is disproportionately distributed to larger farms (Moreddu 2011). This is a natural outcome when agricultural support is tied, as it often is, to production or related factors such as land (FAO, UNDP, and UNEP 2021). If inputs and outputs are subsidized in proportion to quantities used and produced, then larger and richer farmers will always receive a disproportionate share of the benefits. This can also have knock-on, policy-relevant consequences for other areas of society. For example, one study finds that female-headed farms, which tend to be smaller, are less likely to benefit from agricultural subsidies than male-headed farms (FAO 2011). In addition to household distribution, the spatial incidence of subsidies may also be of interest to policy makers. Output subsidy distribution is often based on producing a specific crop. By setting price floors or trade restrictions on certain crops, governments are (possibly inadvertently) supporting particular regions of their countries that are more suitable for growing those crop types. This will have social, economic, and political implications that extend beyond how progressive or regressive the subsidy is at the household level. For example, if they disproportionately benefit poorer regions, even if captured by richer households within those regions, subsidies may have positive distributional impacts through regional multiplier effects. Thus, from a regional development perspective, the spatial distribution of subsidies matters. 1This includes the EU-28 as a single, aggregate entity. From here on, where we talk about countries, we are referring to countries/regions, as the data include the EU-28. 2 Input subsidy schemes may also be responsible for unintended consequences over and above their distributional impacts. While not usually directed at a specific crop, such schemes encourage the production of crops that benefit most from the subsidized inputs. They also incentivize increased input use. This means that, as with output subsidies, regions that are suited to crops that benefit most from the subsidized inputs will gain the most from these subsidies. Second, the increased use of nitrogen fertilizer— a common target of input subsidy programs in many countries 2 —leads to important environmental spillovers. Exceptionally low prices encourage farmers to overuse fertilizers beyond optimal levels (Huang, Gulati and Gregory 2017; Cassou et al. 2017), which can cause nutrient imbalances in soils (Gautam 2015; Kurdi et al. 2020; Basu et al. 2022). This reduces soil fertility and increases nitrogen runoff into waterways, with potentially disastrous impacts on local ecosystems including algal blooms, hypoxia, and contaminated drinking water (James et al. 2005; Damania et al. 2019). According to the Food and Agriculture Organization of the United Nations (FAO), the most common chemical contaminant found in groundwater aquifers is nitrate from farming (Mateo-Sagasta et al. 2017). This paper presents a new global analysis that examines these unintentional distributional and environmental impacts of output and input subsidies, with estimates of their spatial distribution in 16 and 23 countries, respectively. Despite the lack of comprehensive microdata on the spatial distribution of output and input subsidies, we are able to assess—through the creative use of crop production data and crop-level national statistics on subsidies—the broad contours of the spatial distribution of these subsidies, even if these are not precise estimates. Next, we examine the spatial progressiveness or regressiveness of both types of subsidy within countries by overlaying our spatial subsidy maps with spatially disaggregated data on gross domestic product (GDP). We use Lorenz curves to quantify the spatial relationship between subsidies and incomes and determine the spatial progressiveness or regressiveness of subsidies. For output subsidies, we distinguish between fiscal subsidies—direct government payments to farmers—and market price support (MPS), where consumers largely pay for subsidies to farmers through higher prices. Untargeted input and output subsidies will naturally be regressive. These are linked to production amounts, and since richer farmers always produce more output and use more inputs than poorer farmers, they will consequently receive a larger share of the subsidy. This analysis therefore focuses on the relative spatial progressiveness of subsidies. Here, a subsidy is considered spatially progressive (spatially regressive) if poorer regions in each country receive a relatively larger (relatively smaller) share of subsidies than their share of agricultural production value. For example, if the poorer half of a country produces 20 percent of total agricultural production, but receives 40 percent of the subsidy, the subsidy can be said to be spatially progressive. This is analogous to examining the progressiveness of income taxes, and testing whether richer households pay a larger share of their income in taxes than poorer households, instead of simply comparing the total amounts paid by rich versus poor households. Such examinations 2Studies find that subsidy regimes in Bangladesh, India, Nepal, and Sri Lanka incentivize excessive urea application and the underapplication of phosphorus and potassium fertilizers, micronutrients, and organic inputs (Islam and Beg, 2021; Kishore et al. 2021; Gulati and Banerjee 2015; Huang et al. 2017). Cassou et al. (2017) note that in China, maintaining low and stable fertilizer prices contributes to its overuse to a “moderate” or a “significant” extent. 3 are especially useful for addressing development challenges in lagging regions, where poverty tends to be spatially concentrated. In addition, because poorer farmers are more likely to live in poorer regions, assessing the broad contours of the spatial distribution of subsidies can give some indication of a subsidy scheme’s overall progressiveness or regressiveness at household level. Our analysis finds that output subsidies are spatially progressive in 10 countries and spatially regressive in six. We also find that if countries were to repurpose their output subsidies from rice—a very water- intensive crop responsible for exacerbating harmful externalities such as water scarcity, salinization, and carbon dioxide emissions—to maize, their subsidies would become significantly more spatially progressive. This finding stems from rice being produced in richer regions, probably because it is more input-intensive. For input subsidies, of the 23 countries examined, we find that 11 have spatially progressive subsidies, nine have spatially regressive subsidies, and three have spatially neutral ones. As well as distributional impacts, we examine environmental consequences. Using spatially disaggregated data on fertilizer use, we find a very close, positive relationship between input subsidies and fertilizer use, particularly in richer regions. We also show that subsidies are significant contributors to soil saturation of nitrogen, a sign of fertilizer overuse and an indicator of increased environmental degradation. The remainder of the paper is as follows. Section 2 briefly describes the literature on the impact of agricultural subsidies on distributional and environmental outcomes. Section 3 explains the main datasets we use and our strategy for estimating the spatial distribution of subsidies across regions. Section 4 describes our strategy of using Lorenz curves for estimating the progressiveness of subsidies. Section 5 presents our findings on the distribution of output subsidies, and section 6 on the distribution of input subsidies and their impact on fertilizer use and nitrogen concentrations in soils. Section 7 contains a discussion of the results and our conclusions. 2 Literature Much of the literature examining the distributional impacts of agricultural subsidies has focused on developed countries, and specifically the EU’s Common Agricultural Policy (CAP). Several studies find that CAP payment distribution has exacerbated economic and profit inequality by widening the gap between wealthy and poor farmers (Von Witzke and Noleppa 2007; Scown, Brady and Nicholas 2020). Others find more nuance in the relationship, showing that direct payments under CAP reduce farm-household income inequalities (Severini and Tantari 2013), but market-based subsidies such as MPS increase farm-household inequalities (El Benni and Finger 2013; Bojnec and Fertő 2019). Only a handful of studies examine the distributional impacts of subsidies in developing countries. Examining input and output subsidies in developing economies, Tang, Wang and Zhao (2022) find that output subsidies exacerbate the income gap among farmers, but input subsidies narrow it. However, some studies argue that subsidized fertilizer is disproportionately distributed to wealthier households, increasing income inequality in Kenya (Marenya and Barrett 2009), Malawi (Ricker-Gilbert, Jayne and Chirwa 2011; Chibwana et al. 2012), and Zambia (Mason and Tembo 2015; Kuteya et al. 2016). 4 Investigating China’s Rural Income Support Policy, Heerink, Kuiper and Shi (2006) find that tax abolition reduced income inequality within villages but exacerbated it between villages. Fan, Gulati and Thorat (2008) find that agricultural input and output subsidies contribute to high inequality in rural Indian states. This paper contributes to the literature in several ways. To our knowledge, this is the first semiglobal and empirical analysis of the income distribution of agricultural subsidies. Using a consistent methodology, this paper estimates and examines the spatial distribution of output subsidies across 16 developed and developing countries, and of input subsidies across 23 countries, which together account for more than 70 and 80 percent, respectively, of global agricultural production. We also show the differential distributional impacts between MPS and overall output subsidies and, through a policy experiment, demonstrate how our methodology can be useful for designing programs that repurpose subsidies to be more progressive. Finally, our methodology for estimating the distribution of input subsidies allows us to demonstrate their environmental consequences in terms of overfertilization and impacts on soil saturation of nitrogen. When subsidies lead to exceptionally low fertilizer prices, they can encourage farmers to deviate from optimal fertilizer levels, resulting in overuse (Schultz et al. 1964). This imbalance is often tilted on the side of nitrogen, since nitrogen fertilizers are much more heavily subsidized than other types (Islam and Beg 2021; Kishore et al. 2021; Gulati and Banerjee 2015; Huang, Gulati and Gregory 2017). Subsidy-driven applications of nitrogen fertilizer appear to be causing nutrient imbalances in many parts of the world, as farmers apply significantly more nitrogen compared to other primary nutrients, such as potassium, phosphorous, and other secondary and micronutrients (Kurdi et al. 2020; Gautam 2015). Such subsidies may also lead to a failure to supply the right amount of fertilizer at the right time, a problem that seems to be pervasive in developing countries, where technical know-how is low (Duflo, Kremer and Robinson 2011). This is particularly so in the case of nitrogen fertilizer, which is more complex than other fertilizer types when it comes to determining appropriate timings and scales for application. This is partly because the nitrogen forms available for crops are highly volatile and can leave the soil very quickly. Subsidized or distorted prices, combined with farmers’ inability to gauge amounts needed, may result in overapplication and incorrectly timed application (Islam and Beg 2021). Several studies show that fertilizer subsidies have increased both fertilizer use intensity and overuse (Huang, Gulati and Gregory 2017 and references therein). Some studies suggest that China’s maintenance of low and stable fertilizer prices has contributed to its overuse to a “moderate” or a “significant” extent (Cassou et al. 2017). 3 Data We use three datasets to examine the distributional impact of output subsidies. We take disaggregated crop-production data from the Spatial Production Allocation Model (SPAM) (IFPRI 2019). SPAM estimates the spatial distribution of crop production for 42 crops by combining crop production statistics from national or subnational administrative regions with crop-specific suitability information based on local landscape, climate, and soil conditions. In doing so, it disaggregates coarse data from countries or 5 subnational units into grid cells with a 10x10-kilometer resolution. We use the most recent SPAM statistics, for 2010. The OECD database (OECD 2022) provides the crop-level data on output subsidies. We use the Producer Single Commodity Transfer (PSCT) indicator, which measures the annual monetary value of gross transfers from consumers and taxpayers to agricultural producers. PSCT includes both MPS—which results largely from policies (especially trade policies such as import or export bans) that create a gap between internal market prices and border prices—and output subsidies that are fiscal in nature, such as direct transfers to farmers per unit of crop produced. PSCT can be positive or negative: while fiscal subsidies can only be positive, MPS can be positive or negative, depending on whether trade policies increase or decrease market prices. Positive values represent benefits to farmers, and negative values represent costs to farmers. To be comparable with the SPAM dataset, we use 2010 PSCT data. OECD reports PSCT for 65 commodities in 20 countries. After matching SPAM with the OECD dataset, 23 commodities in 16 countries remain in our analysis. They are wheat, rice, maize, barley, potatoes, sorghum, cassava, lentils, other pulses, soybeans, groundnuts, coconut, palm oil, sunflowers, rapeseed, sugar, cotton, coffee, cocoa beans, tea, tobacco, bananas, and plantains. These 23 commodities represent 91 percent of PSCT paid to all commodities in 2010 in countries in the OECD dataset, excluding livestock, meat, and dairy products. The intersection of the 23 crops in the 16 countries account for 34 percent of total agricultural production values worldwide. The analysis rests on the assumption that the subsidy for each crop is spatially neutral. This means that the spatial distribution of output subsidies in a country for a particular crop is equal to the spatial distribution of that crop. Put another way, if a particular region in a country produces 5 percent of that country’s maize, then that region also likely receives 5 percent of the output subsidy for maize. This assumption will generally be true enough to make inferences about the aggregate distribution of subsidies, but may not always hold. We discuss reasons why it may not hold, and the implications of this, in section 7. Based on the above assumption, we spatially allocate PSCT transfers for each crop in the same way as crop production, aggregating data into 0.5x0.5-degree grid cells (approximately 56x56 kilometers at the equator) and summing them across all crops in the country for which there are PSCT and crop production data. The result is a PSCT distribution map within each of the 16 countries in our analysis. To isolate grid cells with significant crop production, we only include those where land use is at least 50 percent cropland according to the European Space Agency’s land-cover/land-use database. Figure 2 shows total PSCT transfers measured for each grid cell and summed by crop. We construct the spatial distribution of MPS, a subset of PSCT, in the same way. With a few additional assumptions, we extend the analysis to examine the spatial incidence of input subsidies. While crop-specific input subsidy data are not available, the OECD dataset provides country- level data on input subsidies. Since these are reported at country, rather than crop level, we make two additional assumptions to estimate their spatial incidence. First, we assume that input subsidies are mainly used to produce five crops: potatoes, sugarcane, maize, rice, and wheat. This assumption is based on literature surveys (Dorward 2009; Hemming et al. 2018), which found that these crops are most likely to be targeted by such programs, and the fact that they tend to be among the highest-yielding crops. 6 Second, we assume that these crops receive the same share of input subsidies as the share of their values. So, if maize accounts for 10 percent of a country’s overall production value among these crops, it would receive 10 percent of total input subsidies. So, if a grid cell produces 20 percent of the country’s total maize production, we would estimate it to receive 10 percent x 20 percent—that is, 2 percent—of the country’s total input subsidy. In sum, these assumptions assign input subsides based on the value produced by selected crops in each region. Based on these assumptions, figure 3 shows the distribution of input subsidies across the world. Finally, we use data on income per capita for each grid cell from Kummu, Taka and Guillaume (2018) to rank grid cells within each country from poorest to richest. We then use this ranking to examine whether relatively poorer grid cells within a country receive a share of subsidies that is proportionate to their agricultural production. 4 Methodology Lorenz curves (Lorenz 1905) are commonly used in equity analyses to graphically portray a population’s cumulative distribution function of income. Although the technique is most commonly used to show income distribution, it is possible to use it to represent any variable distributed across a population, including subsidies. Since Kakwani (1977) expanded and generalized the concept to cover the analysis of correlations between the distributions of different economic variables, studies have used Lorenz curves to investigate poverty and income inequality (Lucky and Sam 2018), the horizontal and vertical equity of electric vehicle rebate allocation (Guo and Kontou 2021), the equity of transit service accessibility (Lope and Dolgun 2020), and the distributional impact of climate change across income deciles and regions (Mideksa 2010; Tol 2021). In agricultural economics, the Lorenz curve has been used to study the equity effects of the US agricultural subsidy program (Kirwan 2007), inequalities in the distribution of estimated income among urban farmers in northern Thailand (Jaitiang, Huang and Yang 2021), inequality between CAP beneficiaries in the EU (Garcia-Bernardo, Jansky and Misak 2021), income inequality due to the adoption of agricultural technology in Northern Western Ethiopia (Shita, Kumar and Singh 2021), and inequality between rural households under the Conversion of Croplands to Forest and Ecological Welfare Forest Programs in rural Anhui, China (Zhang et al. 2019). Samman (2005) also uses Lorenz curves to study farm subsidies in Europe, the United States, and Brazil, to compare the distribution of subsidies across countries. We calculate Lorenz curves to plot the cumulative subsidy received for a given percentile of per capita GDP. These are illustrated in figure 4. In this example, the cumulative amount of total subsidy (plotted in blue) measures the share of subsidy at each point received by grid cells in the GDP per capita percentile on the x axis, or lower. For example, by tracing upwards from the 20th percentile on the x axis, we see that the bottom 20 percent of grid cells in terms of income receive about 90 percent of the total subsidy. The total value of all agricultural production (red line) is plotted in the same way. The red line shows that these 20 percent of grid cells produce about 10 percent of agricultural production. Borrowing from the risk aversion literature, the subsidy is said to first-order stochastically dominate production value if the blue line sits above the red line for the full distribution. This implies that the share 7 of subsidies received by the country’s poorer regions is larger than their share of agricultural production, across the full distribution of income. In this case, we would denote the subsidy as progressive. Conversely, if the red line stochastically dominates the blue line—that is, the red line sits above the blue line for the full distribution—we would consider the subsidy to be regressive. If the lines intersect at least once, then neither first-order stochastically dominates the other, and we can turn to second-order stochastic dominance, where we compare the area under each curve. In this case, the larger one second- order stochastically dominates the other. In cases where the subsidy is negative (as is the case of output subsidies for Argentina, India, and Ukraine), then the analysis is flipped. If the blue line stochastically dominates the red line (first- or second-order), this implies that the subsidy is regressive, and vice versa. This is because a negative subsidy acts more like a tax. 5 Output subsidy results For each country, we plot the distribution of the PSCT subsidy and agriculture production values following the methodology described in section 4. The results of the analysis are shown in table 1 (and figure B.1 in appendix B reports the Lorenz curves by crop and country). In table 1, the first column shows the area under the curve (AUC) for production value (the red line in figure 4). The second column shows the AUC for PSCT (the blue line in figure 4). The third column is the difference between the two. If the difference is negative (positive)—that is, the AUC for PSCT is smaller (greater) than the AUC for production value, this implies that the subsidy is spatially regressive (spatially progressive).3 The results show that PSCT is spatially progressive in 10 countries and spatially regressive in six, according to second-order stochastic dominance. From the Lorenz curves shown in figure B.1 (appendix B) we can also see the results with respect to the stronger set of requirements for first-order stochastic dominance. The subsidy distribution first-order stochastically dominates agricultural production value in five countries (Brazil, Indonesia, India, Kazakhstan, and Ukraine), implying a spatially progressive subsidy throughout the income distribution. The reverse is found in three countries (Canada, Mexico, and the Philippines), implying a spatially regressive subsidy throughout the income distribution. In the other eight countries (Argentina, China, Colombia, Japan, Republic of Korea, Türkiye, the United States, and Vietnam) the lines intersect at least once. For these, we rely on the results presented table 1, which are based on second-order stochastic dominance. The countries in table 1 are ranked according to how regressive their subsidies are (column 3), based on second-order stochastic dominance, where larger positive numbers are more progressive and larger negative numbers are more regressive. The Philippines has the most regressive subsidies, and Brazil has the most progressive. While the differences in the areas under the curve may seem small, even a value of 0.01 is meaningfully different from zero. A score of zero implies that the subsidy is received in equal proportion to production value. Consequently, if a region produces 10 percent of total agricultural 3 The analysis for Argentina, India, and Ukraine is reversed because they have negative PSCTs. 8 production, it also receives 10 percent of the output subsidy.4 A score of 0.01 is the equivalent of starting with this equal proportion distribution and shifting 50 percent of the subsidy received by one quintile of the distribution to the quintile below (for example, from the 40th income percentile to the 20th income percentile). Likewise, a score of 0.02 is the equivalent of shifting 50 percent of the subsidy two quintiles below, or 100 percent of the subsidy one quintile below. Figure B.2 (appendix B) plots the differences between AUCs and GDP per capita levels. While there is a slight inverted-U trend, the relationship is not very strong, implying the progressiveness of output subsidies is not linked to country income levels. Rice is far the largest recipient of PSCT, both in terms of number of countries that subsidize rice, and share of subsidies going to rice (see table B.1 in appendix B for the share of PSCT by crop, by country). Indeed, Indonesia, Japan, the Republic of Korea, the Philippines, and Vietnam all direct more than 70 percent of all output subsidies to rice production. The crops that receive the next largest shares of PSCT are maize, wheat, sugar, and soybeans. Though the distribution of PSCT transfers varies considerably across countries, the distribution for maize production tends to be more progressive than other crops—that is, it tends to be produced in poorer parts of countries. One policy experiment is to estimate the impact on distribution of shifting subsidies from rice to maize. Rice production has huge environmental impacts, in terms of both water use and emissions, with subsidies often encouraging rice cultivation in very water-stressed regions (Damania et al. 2017), and rice cultivation estimated to be responsible for 16 percent of agricultural greenhouse gases and 2.2 percent of total greenhouse gases (Searchinger et al. 2020). Reducing government support to rice cultivation may, therefore, lead to significant environmental gains. Columns 4, 5, and 6 of table 1 show the spatial distributional impacts of such an experiment where all rice support is shifted to maize. Column 4 recalculates the AUC of PSCT if each country were to switch their allocation of PSCT from rice to maize; column 5 recalculates the difference between the production value AUC and PSCT AUC (column 4 minus column 1); and column 6 shows the overall impact of this policy change (column 5 minus column 3). Output subsidies in nine countries would become more progressive with this policy change, while only Canada and Kazakhstan would see more regressive results. Japan and Vietnam would move from overall spatially regressive subsidies to spatially progressive subsidies. Japan, the Philippines, and Vietnam would see the largest distributional shifts in the progressive direction from moving their PSCT from rice to maize. Another policy experiment is to examine the differential distributional impacts of MPS—where consumers largely finance the subsidies to farmers by paying higher prices—and fiscal subsidies, which come from government budgets and, consequently, tax payers. PSCT includes both of these types of subsidy. Thus, we explore the differential effects by repeating the analysis described in section 4 for only MPS (table 2). Here, column 1 shows the AUC for production value; column 2, the AUC for MPS; and column 3, the difference between the two. Column 4 compares these results with table 1. A negative (positive) difference in column 3—that is, where the AUC for MPS is smaller (greater) than the AUC for production value—implies that the subsidy is spatially regressive (spatially progressive).5 The results show that MPS 4 It could also imply offsetting impacts. That is, if the subsidy is regressive in the lower half of the income distribution, but progressive in the upper half, these deviations can offset, resulting in an AUC difference of zero. 5 The analyses for Argentina and India are reversed because they have negative MPS. 9 is spatially progressive in six countries and spatially regressive in seven, based on second-order stochastic dominance. The findings from comparing MPS and PSCT distributions suggest that MPS subsidies are more spatially regressive. Agriculture subsidies in Argentina, Kazakhstan, and India have flipped from being progressive with PSCT to being regressive with MPS. Our analysis of MPS distribution was limited to 13 countries, compared to the 16 with PSCT, because several countries do not have MPS, and restricted to eight crops, compared to 23 for PSCT, because countries have MPS for fewer crops than PSCT.6 Nevertheless, the results of these analyses suggest that fiscal subsidies tend to be more spatially progressive than MPS. 6 Input subsidy results Next, we analyze the spatial distribution of input subsidies within countries. Using our estimate of spatially explicit input subsidy data described in section 3, we repeat the methodology employed for output subsidies. The results in table 3 show that the spatial incidence of input subsidies is regressive in nine countries, progressive in 11, and neutral in three. As with output subsidies, results are mixed and vary significantly between countries. But some patterns emerge: upper-middle and high-income countries tend to have progressive input subsidy schemes, while the spatial distribution tends to be more regressive in African countries. The clear exceptions to these patterns are Kenya, which has the most progressive spatial incidence of all countries, and the EU-28, which has among the most regressive. The EU result is unsurprising, given that the input-subsidy spatial distribution correlates with output and value. The EU CAP has also received significant criticism for how regressive it is, particularly in its early stages (Kaditi and Nitsi 2011; OECD 2003). To examine the environmental consequences of input subsidies—more specifically, the relationship between input subsidies and fertilizer use—we use a cross-sectional model and spatially disaggregated data on average annual nitrogen fertilizer used per hectare of cropland from Lu and Tian (2017). To match our data on subsidies, we use fertilizer use data for 2010 and aggregate data over our 0.5-degree grid cells. We then use regression analysis to estimate the relationship between subsidies and fertilizer use. Given the cross-sectional nature of the analysis, unobserved correlation between subsidies and other factors may influence fertilizer use. To alleviate some of these concerns of omitted variable bias, we include additional controls. These include country-fixed effects that account for time-invariant characteristics that are specific to a country, such as geographic or institutional features, and grid-level characteristics, such as terrain roughness by using the standard deviation of elevation in a grid cell, and “wetness index”, the compound topographic index (CTI) derived from HYDRO1k, which correlates with soil moisture. Figure 5 shows a positive relationship between input subsidies and nitrogen fertilizer use. The histograms for input subsidies and nitrogen levels in fertilizer on the axes depict their distribution across different values. Because we have demeaned the values for fertilizer and input subsidies within countries, these may take negative values. Positive values are above the country average, and negative values are below the country average. The trend indicates a positive correlation between input subsidies and nitrogen 6 Sugar, wheat, rice, maize, barley, rapeseed, coffee, and tea are among the crops studied in the MPS analysis. 10 fertilizer use, with the latter increasing when the former does. But the pace of increase in nitrogen levels in fertilizer varies, depending on the level of input subsidies. We further examine the relationship between input subsidies and fertilizer nitrogen levels in table 4. The log of applied nitrogen fertilizer is the dependent variable, and all independent variables are also in log form, allowing us to interpret the estimated coefficients as elasticities. Results with country-fixed effects and additional cell-level characteristics are shown in columns 2 and 3. According to the estimated coefficients in column 3, raising input subsidies by 10 percent results in a 3.1 percent increase in nitrogen fertilizer use. Next, we investigate whether there are heterogeneities between richer and poorer areas in the subsidies- fertilizer relationship. We define these as the areas within each country whose GDP per capita values are below (poorer) or at or above (richer) the country’s median. The results in column 4 show that the relationship between input subsidies and nitrogen levels in fertilizer is stronger in richer areas. Specifically, richer areas see a 4.7 percent greater increase in nitrogen fertilizer levels from a 10 percent increase in subsidies than poorer areas. In turn, these higher increases in fertilizer use can have environmental consequences on soil and water pollution. To examine the extent to which subsidies can impact groundwater quality, we use spatially disaggregated data on nitrate stored in the vadose (unsaturated) zone from Ascott et al. (2017). Before pollution can be detected in groundwater, contaminants that accumulate in the subsurface spread vertically and laterally in the vadose zone, long before reaching the water table. As such, the amount of stored nitrate here provides a first glimpse into likely impacts on groundwater pollution over time. Table 5 first examines the impact of increased nitrogen fertilizer use on nitrate storage in the vadose zone. Again, since we express fertilizer and soil nitrate values in logs, we can interpret the estimated coefficients as elasticities. Columns 1 and 2 demonstrate that a rise in nitrogen fertilizer use correlates with a rise in nitrate storage in the vadose zone, whether or not we account for country-fixed effects. The results are robust when we also consider land elevation and CTI, with the estimated coefficients in column 3 suggesting that increasing nitrogen fertilizer use by 10 percent would result in a 2.9 percent increase in nitrate storage in the vadose zone. Finally, we examine the role of subsidies in amplifying this expected impact by distinguishing between areas with high and low input levels (that is, where subsidy inputs are above or below each country's median). Column 4 shows that, in regions with high input subsidy levels, a 10 percent increase in fertilizer use causes a 5.7 percent greater increase in nitrate stored in the vadose zone. As a result, increased fertilizer use due to a rise in input subsidies can have an even stronger impact on groundwater pollution in areas with high subsidy levels. This shows that subsidies are exacerbating environmental degradation by incentivizing the use of nitrogen fertilizers beyond efficient levels. 7 Conclusions Our global empirical analysis provides evidence of the spatial distribution of output and input subsidies in the agriculture sector. Using a generous definition of what would be considered “progressive”, we find 11 very mixed results for both output and input subsidies, with a slightly higher percentage of countries having subsidies that are spatially progressive. We also find that input subsidies contribute significantly to nitrogen overfertilization, and that this is leading to increased nitrogen concentrations in soils, an important indicator of both soil saturation and groundwater contamination. Within countries, it is the richer regions—which are both more likely to increase fertilizer use and to have increased soil saturation of nitrogen due to subsidies—rather than the poorer regions, that drive this relationship. And as subsidies often disproportionately benefit wealthier areas, this is also where their unintended harmful consequences are amplified. The evidence here contributes to a growing body of literature that shows the inefficiency, distributional and environmental impacts of agricultural subsidies, and suggests the need to re-evaluate their role. But even if richer farm-households benefit more from subsidies in absolute terms, the impacts of reform on relative income may be greater for poorer households. Thus, before reforming subsidies, policy makers must take great care to analyze and mitigate possible impacts on vulnerable households, seeking to repurpose subsidies in ways that reduce environmental impacts while improving distributional outcomes and protecting vulnerable households (Gautam et al. 2022). The results of this paper suggest that for many countries, shifting from subsidizing rice to maize would go a long way towards accomplishing these goals. In some countries, a shift from MPS to fiscal subsidies may also improve distributional outcomes. But there is no one-size-fits-all policy; reform efforts should reflect individual country characteristics and priorities. Our findings are subject to several important caveats. First, we must again acknowledge that the bar for a subsidy being spatially progressive in this exercise is quite low and is focused more on regional than household equity. The analysis examines whether poorer areas receive a larger proportion of subsidy relative to their proportion of agricultural production. But in absolute terms, poorer farmers tend to receive an overall lower share of an output subsidy because their output is lower, and they use fewer inputs than richer farmers. So, if a subsidy is marked as being progressive, this is only true in a regional sense. The analysis will still likely overestimate a subsidy’s relative progressiveness, because the main assumption is that the distribution of output subsidies for a single crop is the same as the distribution of its production. But poorer farmers are less connected to markets and more likely to produce for household consumption, and so are less likely to benefit from output subsidies, which are only received on marketed goods. For input subsidies, however, some countries have programs that target lower-income households (Mason and Ricker-Gilbert 2013). In these cases, we are likely to underestimate their progressiveness. But studies have also found that even subsidies that target lower-income households tend to disproportionately benefit higher-income households (Chibwana et al. 2010; Ricker-Gilbert and Jayne 2012). Finally, our analysis is based on equity and environmental objectives. But policy makers may consider other objectives—such as food security, self-sufficiency, efficiency gains, or other political goals—when allocating subsidies across regions. In this case, they may be willing to accept a distributional trade-off to fulfill other policy goals. 12 References Angelucci, F, Gourichon, H, Mas Aparisi, A and Witwer, A. 2013. Monitoring and Analysing Food and Agricultural Policies in Africa. MAFAP Synthesis Report. Rome: Food and Agriculture Organization. Anríquez, G, Foster, W, Ortega, J, Falconi, C and De Salvo, C P. 2016. Public Expenditures and the Performance of Latin American and Caribbean Agriculture. IDB Working Paper IDB-WP-722. Washington DC: Inter-American Development Bank. Ascott, M. J., Gooddy, D. C., Wang, L., Stuart, M. E., Lewis, M. A., Ward, R. S., & Binley, A. M. (2017). Global patterns of nitrate storage in the vadose zone. Nature Communications, 8(1), 1-7.Basu, N. B., Van Meter, K. J., Byrnes, D. K., Van Cappellen, P., Brouwer, R., Jacobsen, B. H. et al. (2022). Managing nitrogen legacies to accelerate water quality improvement. Nature Geoscience, 15(2), 97-105 Bojnec, Š and Fertő, I. 2019. “Farm household income inequality in Slovenia.” Spanish Journal of Agricultural Research 17(4): e0112–e0112. Cassou, E., S. M. Jaffee, and J. Ru. 2018. The Challenge of Agricultural Pollution: Evidence from China, Vietnam, and the Philippines. Washington, DC: World Bank. Castaneda, R, Doan, D, Newhouse, D L, Nguyen, M, Uematsu, H and Azevedo, J P. 2016. Who are the Poor in the Developing World? World Bank Policy Research Working Paper (7844). Chibwana, C, Fisher, M, Jumbe, C, Masters, W A and Shively, G. 2010. Measuring the Impacts of Malawi's Farm Input Subsidy Program. Available at SSRN 1860867. Chibwana, C, Fisher, M and Shively, G. 2012. “Cropland allocation effects of agricultural input subsidies in Malawi.” World Development 40(1): 124–133. Damania, R, Berg, C, Russ, J, Federico Barra, A, Nash, J and Ali, R. 2017. “Agricultural technology choice and transport.” American Journal of Agricultural Economics 99(1): 265–284. Damania, R, Desbureaux, S, Rodella, A S, Russ, J and Zaveri, E. 2019. Quality Unknown: The Invisible Water Crisis. World Bank Publications. Dorward, A. 2009. Rethinking Agricultural Input Subsidy Programmes in a Changing World. Report prepared for the Food and Agricultural Organization of the United Nations, Rome. Duflo, E., Kremer, M., & Robinson, J. (2011). Nudging farmers to use fertilizer: Theory and experimental evidence from Kenya. American economic review, 101(6), 2350-90. El Benni, N. and R. Finger (2013). “The effect of agricultural policy reforms on income inequality in Swiss agriculture: An analysis for valley, hill and mountain regions.” Journal of Policy Modeling 35(4): 638–651. Ellis, F. 1992. Agricultural Policies in Developing Countries. Cambridge University Press. Fan, S, Gulati, A and Thorat, S. 2008. “Investment, subsidies, and pro-poor growth in rural India.” Agricultural Economics 39(2): 163–170. 13 FAO. 2011. State of Food and Agriculture 2010-11: Women in Agriculture-Closing the Gender Gap for Development. Food and Agriculture Organization of the United Nations. FAO, UNDP and UNEP. 2021. Multi-billion-dollar Opportunity: Repurposing Agricultural Support to Transform Food Systems. Rome: Food and Agriculture Organization. Garcia-Bernardo, J, Jansky, P and Misak, V. 2021. Common Agricultural Policy Beneficiaries: Evidence of Inequality from a New Data Set. Technical report, IES Working Paper. Gautam, M. 2015. “Agricultural subsidies: Resurging interest in a perennial debate.” Indian Journal of Agricultural Economics 70 (1): 83–105. Gautam, M, Laborde, D, Mamun, A, Martin, W, Pineiro, V and Vos, R. 2022. Repurposing Agricultural Policies and Support. Gulati, A., and P. Banerjee. 2015. “Rationalizing Fertilizer Subsidy in India: Key Issues and Policy Options.” Working Paper 307, Indian Council for Research on International Economic Relations. https://www.academia.edu/14970315/Rationalizing_Fertilizer_Subsidy_in_India- Key_Issues_and_Policy_Options. Guo, S and Kontou, E. 2021. “Disparities and equity issues in electric vehicles rebate allocation.” Energy Policy 154: 112291. Heerink, N, Kuiper, M and Shi, X. 2006. “China’s new rural income support policy: Impacts on grain production and rural income inequality.” China & World Economy 14(6): 58–69. Hemming, D J, Chirwa, E W, Dorward, A, Ruffhead, H J, Hill, R, Osborn, J, Langer, L, Harman, L, Asaoka, H, Coffey, C and Phillips, D. 2018. “Agricultural input subsidies for improving productivity, farm income, consumer welfare and wider growth in low-and lower-middle-income countries: A systematic review.” Campbell Systematic Reviews 14(1): 1–153. Huang, J, Gulati, A and Gregory, I (eds). 2017. Fertilizer Subsidies: Which Way Forward? IFDC/FAI Report. Washington DC: IFDC. IFPRI. 2019. Global spatially-disaggregated crop production statistics data for 2010. Version 2.0. International Food Policy Research Institute. Islam, M., & Beg, S. (2021). Rule-of-Thumb Instructions to Improve Fertilizer Management: Experimental Evidence from Bangladesh. Economic Development and Cultural Change, 70(1), 237-281. Jaitiang, D, Huang, W-C and Yang, S H. 2021. “Does income inequality exist among urban farmers? A demonstration of Lorenz curves from northern Thailand.” Sustainability 13(9): 5119. James, C S, Fisher, J, Russell, V, Collings, S and Moss, B. 2005. “Nitrate availability and hydrophyte species richness in shallow lakes.” Freshwater Biology 50: 1049–63. doi:10.1111/j.1365-2427.2005.01375. Kaditi, E A and Nitsi, E I. 2011. “Vertical and horizontal decomposition of farm income in equality in Greece.” Agricultural Economics Review 12 (389-2016-23441). Kakwani, N C. 1977. “Applications of Lorenz curves in economic analysis.” Econometrica 45(3): 719–727. Kirwan, B E. 2007. The Distribution of US Agricultural Subsidies. Available at SSRN 1117342. 14 Kishore et al. 2021 Kummu, M, Taka, M and Guillaume, J H. 2018. “Gridded global datasets for gross domestic product and Human Development Index over 1990–2015.” Scientific Data 5(1): 1–15. Kurdi, S, Mahmoud, M, Abay, K A and Breisinger, C. 2020. Too Much of a Good Thing? Evidence That Fertilizer Subsidies Lead to Overapplication in Egypt. MENA RP Working Paper. Washington DC: International Food Policy Research Institute. Kuteya, A N, Sitko, N J, Chapoto, A and Malawo, E. 2016. An In-depth Analysis of Zambia’s Agricultural Budget: Distributional Effects and Opportunity Cost. Indaba Agricultural Policy Research Institute Working Paper (107). Lope, D J and Dolgun, A. 2020. “Measuring the inequality of accessible trams in Melbourne.” Journal of Transport Geography 83: 102657. Lorenz, M O. 1905. “Methods of measuring the concentration of wealth.” Publications of the American Statistical Association 9(70): 209–219. Lu, C., & Tian, H. (2017). Global nitrogen and phosphorus fertilizer use for agriculture production in the past half century: shifted hot spots and nutrient imbalance. Earth System Science Data, 9(1), 181-192. Lucky, L A and Sam, A D. 2018. “Poverty and income inequality in Nigeria: An illustration of Lorenz curve from NBS survey.” American Economic & Social Review 2(1): 80–92. Marenya, P P and Barrett, C B. 2009. “State-conditional fertilizer yield response on western Kenyan farms.” American Journal of Agricultural Economics 91(4): 991–1006. Mason, N M and Ricker-Gilbert, J. 2013. “Disrupting demand for commercial seed: Input subsidies in Malawi and Zambia.” World Development 45: 75–91. Mason, N M and Tembo, S T. 2015. Do Input Subsidy Programs Raise Incomes and Reduce Poverty Among Smallholder Farm Households? Evidence from Zambia. Technical report. Mateo-Sagasta, Javier; Zadeh, S. M.; Turral, H.; Burke, J. 2017. Water pollution from agriculture: a global review. Executive summary. Rome, Italy: FAO; Colombo, Sri Lanka: International Water Management Institute (IWMI). CGIAR Research Program on Water, Land and Ecosystems (WLE) Mideksa, T K. 2010. “Economic and distributional impacts of climate change: The case of Ethiopia.” Global Environmental Change 20(2): 278–286. Moreddu, C. 2011. Distribution of Support and Income in Agriculture. OECD. 2003. Farm Household Income: Issues and Policy Responses. Paris: Organisation for Economic Co- operation and Development. OECD. 2020. Agricultural Policy Monitoring and Evaluation 2020. Paris: Organisation for Economic Co- operation and Development. https://doi.org/10.1787/928181a8-en. Ricker-Gilbert, J, Jayne, T S and Chirwa, E. 2011. “Subsidies and crowding out: A double-hurdle model of fertilizer demand in Malawi.” American Journal of Agricultural Economics 93(1): 26–42. 15 Ricker-Gilbert, J and Jayne, T S. 2012. Do Fertilizer Subsidies Boost Staple Crop Production and Reduce Poverty Across the Distribution of Smallholders in Africa? Quantile Regression Results from Malawi. Samman, E. 2005. Gini Coefficients for Subsidy Distribution in Agriculture. Technical report, Human Development Report Office (HDRO). New York: United Nations Development Programme. Schultz, T.W. 1964. Transforming Traditional Agriculture. New Haven, CT: Yale University Press. Scown, M W, Brady, M V and Nicholas, K A. 2020. "Billions in misspent EU agricultural subsidies could support the Sustainable Development Goals.” One Earth 3(2): 237–250. Searchinger, T D, Malins, C, Dumas, P, Baldock, D, Glauber, J, Jayne, T, Huang, J and Marenya, P. 2020. Revising Public Agricultural Support to Mitigate Climate Change. Washington DC: World Bank. https://openknowledge.worldbank.org/handle/10986/33677. Severini, S and Tantari, A. 2013. “The impact of agricultural policy on farm income concentration: The case of regional implementation of the cap direct payments in Italy.” Agricultural Economics 44(3): 275–286. Shita, A, Kumar, N and Singh, S. 2021. “Technology, poverty and income distribution nexus: The case of fertilizer adoption in Ethiopia.” African Development Review 33(4): 742–755. Tang, C S, Wang, Y and Zhao, M. 2022. “Input- vs. Output-based Farm Subsidies in Developing Economies: Farmer Welfare and Income Inequality.” In Agricultural Supply Chain Management Research, edited by O. Boyabatlı, B Kazaz and C S Tang, 265–286. Springer. Tol, R S. 2021. “The distributional impact of climate change.” Annals of the New York Academy of Sciences, 1504(1): 63–75. Von Witzke, H and Noleppa, S. 2007. Agricultural and Trade Policy Reform and Inequality: The Distributive Effects of Direct Payments to German Farmers under the EU’s New Common Agricultural Policy. Working Paper. Zhang, Q, Bilsborrow, R E, Song, C, Tao, S and Huang, Q. 2019. “Rural household income distribution and inequality in China: Effects of payments for ecosystem services policies and other factors.” Ecological Economics 160: 114–127. 16 Figures Figure 1. Total agricultural total support estimate and GDP, 2016–2018 120% estimate(TSE)/agricultural production 100% NOR 80% ISL CHE SEN 60% KOR JPN KEN Total support 40% UGA JAM value BEN BDIPHL SLV IDN HTI DOM PAN TUR RWA BHS EU USA 20% MOZ BOL BFA GUY PER CHN SUR COLRUS ISR NIC TZA HND ECU BLZ MEX KAZ CAN GTM ZAF BRA PRY CRI CHL URY NZL AUS IND 0% VNM UKR MWI ETH ARG -20% MLI -40% GHA -60% $0 $20,000 $40,000 $60,000 $80,000 $100,000 GDP per capita, 2016 (constant 2015 US$) Sources: TSE data are from OECD 2020, Anríquez et al. 2016, and Angelucci et al. 2013; agricultural value added and GDP per capita are from the World Bank’s World Development Indicators database. Notes: Figure shows TSE as a share of agricultural value added (2016–18 mean) versus GDP per capita (2016). Figure 2. Global distribution of PSCT 17 Notes: Data is aggregated into 0.5x0.5 degree grid cells (approximately 56x56km at the equator) to map the distribution of PSCT globally and measure PSCT transfers to each region. Grid cells with less than 50% cropland are excluded from the analysis. Original data is from IFPRI 2019 and OECD 2020 Figure 3. Global distribution of input subsidy Notes: Data are aggregated into 0.5x0.5 degree grid cells (approximately 56x56km at the equator) to map input subsidy distribution globally and measure subsidies at regional level. Grid cells with less than 50% cropland are excluded from the analysis. 18 Figure 4. Example of Lorenz curve for subsidies Notes: The figure shows an example of the analysis for a single country. The blue line plots the cumulative amount of total subsidies received by each income percentile along the x axis. The red line plots the same information for total agricultural production value. 19 Figure 5. Relationship between input subsidy and nitrogen levels in fertilizer Notes: We have excluded the outliers (top 10% values) in estimated data for input subsidy (US$ million) and nitrogen levels (grams per square meter) in fertilizers. For illustrative purposes, we have excluded the values above 4 grams per square meter from the histogram of nitrogen in fertilizer. We have demeaned the values for fertilizer and input subsidies using the country's average values. 20 Tables Table 1. Distribution of PSCT relative to agricultural production value Policy experiment (moving subsidies Analysis from rice to maize production) Country AUC AUC PSCT Diff AUCs Policy production AUC PSCT Diff AUCs with policy with policy impact value Philippines 0.64 0.51 -0.13 0.62 -0.02 0.11 Mexico 0.58 0.52 -0.05 0.52 -0.05 0.00 Canada 0.59 0.54 -0.05 0.51 -0.08 -0.03 Japan 0.36 0.33 -0.02 0.69 0.33 0.36 Vietnam 0.47 0.46 -0.01 0.55 0.08 0.09 Türkiye 0.55 0.54 -0.01 0.54 -0.01 0.00 Argentina* 0.50 0.49 0.01 0.49 0.01 0.00 Korea, Rep. 0.51 0.51 0.01 0.55 0.05 0.04 China 0.58 0.60 0.02 0.59 0.02 0.00 Colombia 0.52 0.55 0.02 0.61 0.09 0.06 United States 0.50 0.55 0.05 0.55 0.05 0.00 Ukraine* 0.61 0.56 0.06 0.55 0.07 0.01 India* 0.57 0.51 0.06 0.49 0.08 0.02 Indonesia 0.63 0.71 0.08 0.72 0.08 0.01 Kazakhstan 0.67 0.77 0.10 0.74 0.08 -0.02 Brazil 0.48 0.62 0.15 0.64 0.16 0.02 Notes: Our analysis of PSCT distribution covered 16 countries and 23 crops. * identifies countries with negative PSCT values. Diff = difference between. The analysis is flipped for these countries to be consistent with measured values for other countries. In the case of negative PSCT, differences between AUCs are multiplied by the negative one. The last column shows the spatial distributional impacts of an experiment where all rice support is shifted to maize. Table 2. Distribution of MPS relative to agricultural production value Country AUC production value AUC MPS Diff AUCs Diff AUCs from table 1 Philippines 0.64 0.50 -0.14 -0.13 Argentina* 0.50 0.56 -0.07 0.01 Japan 0.36 0.29 -0.06 -0.02 Kazakhstan 0.67 0.61 -0.06 0.10 Mexico 0.58 0.53 -0.05 -0.05 India* 0.57 0.62 -0.05 0.06 Vietnam 0.47 0.46 -0.01 -0.01 Korea, Rep. 0.51 0.52 0.01 0.01 China 0.58 0.59 0.02 0.02 Indonesia 0.63 0.69 0.05 0.08 Colombia 0.52 0.59 0.07 0.02 United States 0.50 0.63 0.13 0.05 Brazil 0.48 0.69 0.22 0.15 Notes: Our analysis of MPS distribution was limited to the 13 countries that have MPS and the eight crops covered by MPS. * identifies countries with negative MPS values. Diff = difference between. The analysis is flipped for these countries to be consistent with measured values for other countries. In the case of negative MPS, differences between AUCs are multiplied by the negative one. 21 Table 3. Distribution of input subsidies relative to agricultural production value Country AUC production value AUC input subsidy Diff AUCs Brazil 0.34 0.48 -0.14 Philippines 0.51 0.63 -0.13 EU_28 0.54 0.61 -0.07 Burkina Faso 0.48 0.52 -0.03 Malawi 0.56 0.59 -0.03 Kazakhstan 0.64 0.67 -0.03 Bolivia 0.45 0.46 -0.01 Ukraine 0.60 0.61 -0.01 South Africa 0.50 0.51 -0.01 Vietnam 0.47 0.47 0.00 Korea, Rep. 0.52 0.52 0.00 Mozambique 0.71 0.71 0.00 India 0.58 0.57 0.01 Türkiye 0.56 0.54 0.01 Canada 0.62 0.60 0.02 United States 0.52 0.50 0.02 Argentina 0.53 0.50 0.03 Colombia 0.57 0.53 0.04 Indonesia 0.69 0.63 0.06 China 0.64 0.58 0.06 Japan 0.44 0.36 0.09 Mexico 0.67 0.58 0.09 Kenya 0.83 0.68 0.15 Note: EU-28 is considered as a single unit. Diff = difference between. 22 Table 4. Impact of subsidies on nitrogen in fertilizer No controls Country FE Country FE and Country FE, grid cell grid cell controls, and median controls GDP indicator Subsidy 0.141*** 0.032*** 0.031*** 0.020*** (0.003) (0.002) (0.002) (0.002) (0.000) (0.000) (0.000) (0.000) At or above median GDP 0.028*** (0.008) (0.000) Subsidy × at or above 0.019*** median GDP (0.003) (0.000) Country fixed effects N Y Y Y Grid cell controls N N Y Y Observations 13,978 13,978 13,971 13,971 Adjusted R-squared 0.1215 0.8401 0.8406 0.8411 Notes: The dependent variable is the level of nitrogen in fertilizers. Values are in log formats. The first parentheses under the estimated coefficients show standard deviations; the second parentheses show p-values. Grid cell controls include mean elevation and CTI. Significance levels: * 10%, ** 5%, *** 1%. Y = yes; N = no. Table 5. Impact of nitrogen fertilizer use on nitrate storage in the vadose zone No controls Country FE Country FE and Country FE, grid cell grid cell controls controls, and high subsidy indicator Fertilizers 0.304*** 0.217*** 0.290*** 0.175*** (0.014) (0.031) (0.030) (0.032) (0.000) (0.000) (0.000) (0.000) High subsidy 0.481*** (0.052) (0.000) Fertilizers × high 0.085*** subsidy (0.025) (0.001) Country fixed effects N Y Y Y Grid cell controls N N Y Y Observations 13,483 13,483 13,480 13,480 Adjusted R-squared 0.0346 0.1875 0.2326 0.2586 Notes: The dependent variable is the level of nitrogen in fertilizers. Values are in log formats. The first parentheses under the estimated coefficients show standard deviations; the second parentheses show p-values. Grid cell controls include mean elevation and the CTI. Significance levels: * 10%, ** 5%, *** 1%. Y = yes; N = no. 23 Appendix A. Spatial distribution of nitrogen in fertilizers and soil nitrate Figure A.1 Nitrogen in fertilizers: global map Source: Lu and Tian 2017 Figure A.2. Soil nitrate: global map Source: Ascott et al. 2017 24 Appendix B. Country-specific results of PSCT distributional analysis This appendix presents country-specific results for the PSCT distributional analysis, with two diagrams for each country in the analysis. The left panels show the distribution of crops produced, and the right panels compare PSCT distribution on all crops with production value. For each country, we note whether PSCT stochastically dominates production value to identify the progressiveness of subsidies. We use the results in table 1 define whether PSCT second-order stochastically dominates production value. Figure B.1. Country-specific results of PSCT by crop (left) and total production (right) (a) Argentina Notes: Negative PSCT does not first-order stochastically dominate production value, but second-order stochastic dominance suggests a spatially progressive distribution. Dashed lines show negative PSCT values. (b) Brazil Notes: PSCT first-order and second-order stochastically dominates production value. The distribution of subsidies is spatially progressive. 25 (c) Canada Notes: Production value first-order and second-order stochastically dominates PSCT. The distribution of subsidies is regressive. (d) China Notes: PSCT does not first-order stochastically dominate production value, but the second-order stochastic dominance suggests a spatially progressive distribution. Dashed lines show negative PSCT values. (e) Colombia 26 Notes: PSCT does not first-order stochastically dominate production value, but the second-order stochastic dominance suggests a spatially progressive distribution. (f) Indonesia Notes: PSCT first-order and second-order stochastically dominates production value. The distribution of subsidies is spatially progressive. Dashed lines show negative PSCT values. (g) India Notes: Negative PSCT first-order and second-order stochastically dominates production value. The distribution of subsidies is spatially progressive. Dashed lines show negative PSCT values. (h) Japan 27 Notes: PSCT does not first-order stochastically dominate production value, but the second-order stochastic dominance suggests a regressive distribution. 28 (i) Kazakhstan Notes: PSCT first-order and second-order stochastically dominates production value. The distribution of subsidies is spatially progressive. Dashed lines show negative PSCT values. (j) Korea, Rep. Notes: PSCT does not first-order stochastically dominate production value, but the second-order stochastic dominance suggests a spatially progressive distribution. (k) Mexico Notes: Production value first-order and second-order stochastically dominates PSCT. The distribution of subsidies is regressive. 29 (l) Philippines Notes: Production value first-order and second-order stochastically dominates PSCT. The distribution of subsidies is regressive. Dashed lines show negative PSCT values. (m) Türkiye Notes: PSCT does not first-order stochastically dominate production value, but the second-order stochastic dominance suggests a regressive distribution. (n) Ukraine 30 Notes: Negative PSCT first-order and second-order stochastically dominates production value. The distribution of subsidies is spatially progressive. Dashed lines show negative PSCT values. (o) United States Notes: PSCT does not first-order stochastically dominate production value, but the second-order stochastic dominance suggests a spatially progressive distribution. (p) Vietnam Notes: PSCT does not first-order stochastically dominate production value, but the second-order stochastic dominance suggests a spatially regressive distribution. Dashed lines show negative PSCT values. 31 Figure B.2. Differences between AUCs and GDP per capita levels 32 Table B.1. Share of PSCT received by each crop, 2010 Korea, United Crop Argentina* Brazil Canada China Colombia Indonesia India* Japan Kazakhstan Rep Mexico Philippines Turkey Ukraine* States Vietnam Banana 0.0% 24.2% Barley 11.8% 1.0% -2.3% 1.4% 1.9% 12.6% 6.9% 0.3% Cassava 0.0% Cocoa 0.0% Coconuts 0.2% Coffee 2.9% 2.1% 0.0% 0.1% -2.6% Cotton 37.4% 29.4% 9.6% 39.1% 0.4% 6.3% 4.9% Groundnuts 3.4% Lentils 18.1% Maize 9.8% 8.5% 8.2% 29.5% 45.8% 23.5% 4.1% -8.2% 27.4% -6.5% 6.3% 20.3% 41.7% 4.9% Palm 7.4% -5.6% Plantains 10.1% Potatoes 12.6% 0.1% 28.1% Rapeseed 34.5% 3.9% 3.6% 0.0% Rice 8.3% -0.9% 32.9% 76.2% 33.4% 89.0% -17.5% 88.3% 0.2% 70.7% 1.5% 82.3% Sorghum 15.7% 1.4% Soybeans 71.3% 11.9% 1.1% 4.0% 0.6% 3.9% 0.4% 10.3% 0.7% 25.2% Sugar 24.7% 4.1% 1.6% 5.2% 5.2% 7.5% 28.4% 35.6% 11.1% 16.0% Sunflower 4.9% -17.6% 3.4% 39.8% Tea -0.6% Tobacco 0.1% Wheat 14.0% 6.4% 26.4% 29.9% 2.0% 106.4% 25.1% 43.3% 33.0% 13.9% Total PSCT (US$, - millions) -5,097 1,074 272 28,337 112 5795 21,348 1551 80 2,422 610 2,129 4,170 -647 3,334 2,142 Notes: A green box indicates that a crop receives more than 20% of the total PSCT in that country; a yellow box indicates that a crop received less than 20% of the total PSCT paid in a country, it is shown by the yellow color. A red box indicates that a crop contributes negatively to the total PSCT—that is, the country paid subsidies to all other commodities except that crop. * identifies countries with negative PSCT values. 33