Policy Research Working Paper 11018 Crops, Conflict and Climate Change Erhan Artuc Guido Porto Bob Rijkers Development Economics Development Research Group January 2025 Policy Research Working Paper 11018 Abstract This paper studies the welfare impacts of agricultural shocks across developing countries by 2.90 percent on average, on households with detailed heterogeneity, by taking con- while changes in yields due to climate change will reduce sumption, land, and labor allocation choices into account. real incomes by 11.99 percent. The welfare impacts of both The underlying model is quantified with household survey shocks vary enormously across the income distribution, data from 51 developing countries, then used to analyze with already vulnerable households bearing the brunt of the welfare consequences of the food price hikes induced by their costs. Poor households suffer losses that are consider- the Russian Federation’s invasion of Ukraine and future cli- ably larger and much more dispersed than those predicted mate change. Both repress income and exacerbate inequality. by models that do not feature household heterogeneity and War-induced food inflation reduced real household incomes rely exclusively on aggregate data. This paper is a product of the Development Research Group, Development Economics. 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 eartuc@worldbank.org, guido.porto@depeco.econo.unlp.edu.ar, and brijkers@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 Crops, Conflict a nd C limate Change∗ † ‡ Erhan Artuc, Guido Porto, and Bob Rijkers§ Keywords: International trade policy, conflict, climate change, welfare, income inequality JEL codes : F1, F18, O11, O13, Q17 ∗ This paper has been supported by the World Bank’s Whole of Economy, the Umbrella Facility for Trade, and Knowledge for Change programs. 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 of Reconstruction and Development, the World Bank, and their affiliated organizations or those of the Executive Directors of the World Bank, or the countries they represent. We thank Nicolas Gomez-Parra, Guillermo Falcone, Isambert Leunga for excellent data work. Comments from seminar audiences at University of Colorado, Georgetown University, the USITC, the World Bank, Universidad Nacional de La Plata, the IMF, the Econometric Society European Winter Meeting, Society of Economic Dynamics Meetings, the Southern Economic Association Annual Conference and the Midwest International Trade Annual Conference are greatly appreciated. All errors are our responsibility. † Development Economics Research Group, World Bank, Washington, DC. ‡ Universidad Nacional de La Plata, La Plata. § Development Economics Research Group, World Bank, Washington, DC, and Utrecht University, Utrecht. 1 Introduction Households in low-income countries depend heavily on agriculture for income and spend a large share of their budget on food. This dependency on agriculture is even more pronounced among the poorest households (Banerjee and Duflo, 2010). Consequently, agricultural shocks can have large and highly uneven impacts on welfare, poverty and inequality. Since agricultural products are traded globally, a general equilibrium trade model is required to assess how such shocks propagate and impact prices in different countries. In addition, agrarian households make different consumption and farming decisions and these choices, which can be adjusted in response to shocks, modulate the impact of price changes on real incomes. The international trade linkages interact with the heterogeneity in household choices and together they determine how agricultural shocks reverberate through the income distribution. Yet, existing trade models typically do not feature detailed household heterogeneity, and have consequently been largely silent on the distributional consequences of agricultural shocks within countries. This paper aims to help fill this gap. We develop a discrete choice general equilibrium trade model of heterogeneous households as producers and consumers. As producers, households allocate their land and labor endowments to the production of different crops to maximize profits depending on their land and labor productivity. As consumers, households spend income on crops and products to maximize non-homothetic utility. Our framework captures household-level heterogeneity in labor income, crop sales and consumption allocations, similar to Deaton (1997), but with land supply decisions as in Sotelo (2020) and Costinot, Donaldson, and Smith (2016), and labor supply decisions inspired by Artuc, Chaudhuri, and McLaren (2010). The model is designed to leverage household survey data to assess the impacts of agricultural shocks across the entire income distribution, as well as to provide a more accurate measurement of the aggregation of those impacts relative to representative agent models. We exploit two major crises with global implications, notably the Russian Federation’s invasion of Ukraine (and the subsequent war) and climate change, to provide a fine-grained quantification of the impacts of agricultural shocks on real expenditures in 51 developing countries. For both the war and climate change shocks, we quantify the average effects on real household expenditures and the implications for income inequality. We study both shocks to illustrate how different mechanisms determine how agricultural shocks impact household welfare. The war shock allows us to validate the model and to highlight how changing prices impact the cost of living. We focus on the impact of war on third countries through agricultural production and consumption, and the impact of the war on 2 Ukraine is out of the scope of this paper.1 The climate change shock alters plot productivity and allows us to highlight how the income generation channel operates. The latter shock is also used to document several household-level adaptation mechanisms such as household expenditure adjustments, labor and land re-allocations and agricultural trade responses. To take the model to the data, we combine information on trade flows from the “International Trade and Production Database for Estimation” (henceforth ITPDE, introduced by Borchert, Larch, Shikher, and Yotov (2021)) with nationally representative household survey data from the “Household Impacts of Tariff” database (henceforth HIT, introduced by Artuc, Porto, and Rijkers (2020)). The HIT data is a key building block in our analysis because it contains information on income and expenditure shares for 24 different product categories and 100 representative households per country—each representing a percentile of that country’s income distribution. Using HIT, we are able to work with households in 51 low and middle-income countries.2 For the rest of the world, we work with a representative household using ITPDE data. Initial trade, factor allocation and consumption shares required to quantify the model are taken directly from these data. Importantly, the land and labor elasticities—parameters which govern household land and labor allocations—are estimated with a non-linear least squares estimator (similar to Costinot, Donaldson, and Smith (2016)) by combining the HIT database with the Global Agro-Ecological Zones database of the Food and Agriculture Organization (FAO and IIASA (2021), henceforth GAEZ). We study the Russian Federation’s invasion of Ukraine and climate change shocks with counterfactual simulations. The economic effects of the war on third countries are simulated by running the model conditional on the main trade restrictions on agriculture induced by the invasion. In our main simulation, Ukraine cannot export or import any agricultural products or inputs from the world. Russia, in turn, bans trading its major agricultural exports, namely wheat, rice, corn, other sugar, oilseeds, and fertilizers. Because the war shock has already occurred, we can validate the model by assessing its goodness of fit. We compare the price changes predicted by the model with those observed after the onset of the war (using data from the FAO’s Food Prices Monitoring and Analysis tool). Reassuringly, the price changes generated by the model correlate strongly with those observed in reality. The impact of climate change is simulated with a version of the model in which we input the productivity of different crops across countries based on climate change projections arising from the 1 For an assessment of welfare loss in Eastern Ukraine due to conflict see Artuc, Gomez-Parra, and Onder (2022). 2 The data covers all low and lower middle income countries for which nationally representative household surveys with both income and expenditure information are available. 3 main scenario of the UN’s Intergovernmental Panel on Climate Change (IPCC). Following Costinot, Donaldson, and Smith (2016), we use GAEZ data to implement the climate change shock. Turning to our main findings, the Russian Federation’s invasion of Ukraine represses real incomes and exacerbates inequality across the developing world through agriculture, with highly heterogeneous impacts both across and within countries. The overwhelming majority of households (99.6%) in our sample suffers a reduction in real incomes, with an average real income loss across countries of –2.90%. This is much larger than the –1.87% loss predicted by a representative household model. The range in average impacts across countries is wide, varying from a minimum of –10.94% to a maximum of -0.40%. The average income losses for the bottom 25% poorest households in a given country are –3.11%, while the average loss incurred by the top quartile is smaller at –2.78%. The aggregate losses are strongly correlated with reliance on imports from Ukraine and Russia. In particular, neighbors with large trade flows with the countries at war incur heavy losses (Azerbaijan, –10.94%; Mongolia, –10.49%; Georgia, –7.87%). The primary driver of these real income losses is higher consumption prices, especially on food. Furthermore, the losses are concentrated among the poor because they spend a higher share of their budget on food. Climate change has large and significantly heterogeneous impacts. On average, real incomes decline by –11.99%, but the variance in impacts across households is very large, ranging from a maximum average loss of –68.43% to a maximum average gain of 52.92%. In a model based on a single household, the average loss would be smaller (–9.93%) as in the war case. The poor are disproportionately impacted by climate change. Across all countries in our sample the bottom 25% poorest households within a given country experience losses (–14.63% of real income) which are approximately 1.5 times those incurred by households in the top quartile (–9.93%). The projected productivity impacts of climate change are the primary driver of differences across countries. Most countries experience negative agricultural productivity shocks and thus incur losses. These productivity changes cause large changes in labor and especially land incomes, which are the dominant mechanisms by which climate change operates. By contrast and unlike the war shock, changes in consumer prices are not as important. Adjustment also shapes the heterogeneous welfare impacts. Non-homothetic preferences yield larger welfare losses than homothetic ones. As average incomes decline, households are forced to allocate relatively larger expenditure shares to the agricultural goods most affected by the shock. Land and labor adjustments cushion households against the shock. Real income losses would be 2.79% percentage points higher if production factors were not allowed to reallocate (–14.78% versus –11.99%). Conversely, 4 allowing factor adjustment to vary across rich and poor households would dampen the impact of the shock. Trade is a major adaptation mechanism. Restricting imports and exports would increase average losses by 7.78 percentage points on average (–19.77%). Trade smooths out the changes in consumer prices and the responses of land and labor income. In all these scenarios, while adjustment can provide some relief, it neither offsets the losses entirely nor protects the poor from suffering greater harm than the rich. The paper builds on and complements several strands of existing literature. Our paper focuses on the relationship between agriculture and welfare, which is also the main theme in Costinot and Donaldson (2016), Costinot, Donaldson, and Smith (2016), and Sotelo (2020). A distinguishing feature of our approach is its focus on households which contrast with existing approaches that typically postulate a single household for each country. This not only allows us to quantify impacts on inequality but also improves estimates of the aggregate welfare effects. This is because of an inherent aggregation bias associated with using aggregate data instead of household level data. The average welfare impact across heterogeneous households making different production and consumption decisions does not coincide with the welfare impact for a single household characterized by aggregate level data. In the case of climate change, which causes significant household adjustment, the estimates of the welfare effect that a representative household model would yield are 23.96% lower than those obtained using our heterogeneous household model. For the war shock, the bias is 35.52%. These findings dovetail with the heterogeneous firm literature that has shown that microstructure matters for the aggregate ıguez-Clare, and Werning (2020)). gains from trade (Melitz and Redding (2015); Costinot, Rodr´ Second, our model accounts for a rich set of channels of impact and is characterized by product-level granularity. Just like Costinot and Donaldson (2016), Costinot, Donaldson, and Smith (2016), Fajgelbaum and Redding (2014), Sotelo (2020), and Tombe (2015) we depart from the common practice of aggregating agriculture into a broad (few) sector(s), and instead focus on 20 different crops. In our model, the land allocation problem of the household builds on Costinot, Donaldson, and Smith (2016) and Sotelo (2020), in which producers allocate plots to different crops to maximize their income in a discrete choice setup. Different from these papers, our unit of observation is the household rather than the plot, which allows us to incorporate heterogeneity in consumption baskets and differences in wage income. The labor allocation decision of the households is also modeled with a discrete choice framework, building on a static version of Artuc, Chaudhuri, and McLaren (2010), a la Lagakos and Waugh (2013), Lee (2020), Lee and Yi (2018) and Galle, Rodriguez-Clare, and Yi (2021). A paper close to ours with households in a trade model is Bergquist, Faber, Fally, Hoelzlein, Miguel, 5 ıguez-Clare (2022). Studies of the distribution of the gains from trade that are similar to and Rodr´ ıguez-Clare, and Yi (2023), who incorporate labor heterogeneity, our approach include Galle, Rodr´ Fajgelbaum and Khandelwal (2016) and Nigai (2016), who introduce non-homothetic preferences, and Adao, Carrillo, Costinot, Donaldson, and Pomeranz (2022), who account for heterogeneity across the earnings distribution. We also contribute to the burgeoning literature on the impact of climate change (Tol (2009)). Our paper is closest to Costinot, Donaldson, and Smith (2016), who use GAEZ projections to study aggregate welfare effects. We extend this analysis to accommodate household heterogeneity and show that impacts of climate change vary enormously not only across countries but also across households within countries. This also affects the aggregation of the average welfare effects. A related strand of literature includes Desmet and Rossi-Hansberg (2015) and Cruz and Rossi-Hansberg (2023). They quantify welfare effects, as we do, embedding climate change responses into the model instead of adopting GAEZ projections. Finally, this paper provides one of the first attempts to quantify the economic impact of the Russian Federation’s invasion of Ukraine on developing countries in general equilibrium. The remainder of this paper proceeds as follows. Section 2 introduces the theoretical model. Section 3 discusses the data and the calibration and estimation of the parameters of the model. Section 4 presents our analysis of the impact of the war shock, while section 5 analyses the impact of climate change. Section 6 concludes. 2 A Model of Trade and Agriculture with Households In this section, we introduce the model of agricultural trade that we use to explore the impacts of the war and climate change. A distinctive feature of our paper is that we focus on households in low-income countries and their heterogeneous decisions in terms of both the consumption and production of different agricultural products. This requires detailed household-level data, which we take from the Household Impacts of Tariffs (HIT) database (see Section 3). Our model is designed to capture and exploit the main features of this household survey data. To facilitate the exposition, we start by presenting the different building blocks of our theory. Countries. We consider N countries indexed with n. We divide countries into two groups based on the availability of household data. In those countries with household data in the HIT database (a total of 51 “HIT countries”), there are H n households which are engaged in both production and consumption. 6 These countries are low-income, agrarian economies, which are small players in international markets, accounting for only a small fraction of global trade.3 To address this issue, we also include major trade players such as the U.S., the E.U., China, Brazil, and India. We refer to these countries as the “central” economies as opposed to the low-income agrarian countries in HIT. In these countries, since we do not have suitable household-level data, we work with a representative household. Goods. Because we want to focus on agriculture, we exploit the HIT database and work with a set of 20 crops and agricultural products for which we have household-level consumption and production data. The details are provided in Section 3, which discusses the datasets we use. By focusing on disaggregated crop data at the household level, we are in a unique position to study the micro-level implications of price and productivity shocks, such as those generated by conflict and climate change, on agricultural outcomes. We assume that crops are differentiated across countries so that each country produces a different variety of each crop. Within countries, all households h produce the same variety of crop j (i.e., crops are differentiated by country but not by household). The rest of the non-agricultural economy is represented by three aggregate sectors. There is a manufacturing sector, M , which is traded. Manufactures are differentiated across countries. There is one agricultural input, F , (e.g., fertilizers), which is differentiated across countries and is traded. Lastly, there is a non-traded services sector, S . Households and firms. Households are heterogeneous in their consumption preferences. They have different endowments of land and labor. Land is not traded, while labor is freely mobile across crops and sectors. Households produce crops using their own land, but they can use their own labor, hire outside labor or work off-farm, either for other households or for manufacturing or services firms. Households are heterogeneous in land productivity and workers are heterogeneous in labor productivity. Even though we have information on thousands of households in the HIT database, for computational convenience and in order to facilitate data sharing we aggregate them to 100 households per country.4 The aggregation is done over 100 bins of the distribution of per capita expenditure. This means that each household, in each country, represents one percentile of the income distribution. This is a natural aggregation given our interest in inequality. There are 5100 representative households in our setting. Firms, operating in a perfect competition environment, produce manufactures and services using labor and a specific factor (capital or structures). For simplicity, we assume that the intermediate input (fertilizers) is produced using only a specific factor. 3 Together the 51 countries accounted for 4.80% of total global exports and imports recorded in United Nation’s COMTRADE database in 2021. 4 Many of the household survey datasets we use are subject to data access restrictions; by aggregating households into income percentiles we are able to circumvent such restrictions and create a dataset that is publicly available. 7 Trade and Prices. The N countries trade agricultural products, manufactures and intermediate n,m goods. The price of sector j product in country m is denoted with pm j . There are trade costs τj between countries n and m, which include transportation costs and tariffs. Services are not traded. 2.1 Household Preferences and Demand Households consume agricultural goods, manufactures and services. Household preferences are represented by a Stone-Geary utility function n n,h n νj U n,h = Cj − Cj , j n n . Here, j indexes the set of where C j > 0 are the subsistence requirements for all j , and 1 = νj n,h agricultural products, a manufacturing aggregate and services. Household h spends αj share of its total income E n,h on product j , where n n n,h n k n C k Pk C j Pjn αj = νj 1− + n,h , (1) E n,h E n,h and j αj = 1. The shares depend on household income and prices. If a product is required to be consumed in large quantities for subsistence, higher income household will have a smaller share for that product compared to lower income ones. This will generate heterogeneity across households in their consumption choices. Each country produces a different variety of the various crops and of the manufactured product. These varieties are then combined and consumed by the household in an Armington setup. The composite product j is created using the following aggregator function σ σ −1 σ −1 n,h ϑn,m n,h σ Cj = j Cm,j , (2) m where σ > 0 is the elasticity of substitution and each variety comes from a different country, indexed with m.5 ϑn,m j is a country-specific, but not household-specific, utility shifter. However, the shares spent on each composite j are household specific. Since households have the same Armington aggregator function (2) and face identical prices, the participation of each variety in the composite is the same across households. Concretely, the expenditure of household h residing in country n on a variety of n,m n,h agricultural good j produced in country m is pm j τj Cm,j = xn,m j n,h Ej n,h , where Ej is total household 5 This specification makes the import decision problem isomorphic to the Eaton and Kortum (2002), but provides a simpler formulation which facilitates the exposition. 8 expenditure in good j and the import share is 1−σ ϑn,m j pm n,m j τj xn,m j = 1−σ . (3) Pjn The price index for composite j is 1 1−σ 1−σ n,m′ m′ n,m′ Pjn = ϑj pj τj . (4) m′ Given endogenous expenditure shares, we have that n,h n,h n,h Ej = αj E , (5) where, again, E n,h is total household expenditure. n,h n,h n,h Similarly, total household expenditures on the manufacturing aggregate is: EM = αM E and n,m n,h n,m n,h the expenditure of h on the variety produced in country m is pm M τM Cm,M = xM EM where the import share and price index equations are identical to the agricultural goods. The non-traded good S is homogeneous within countries. Thus, there is no Armington aggregator n,h n,h for services, consumption is CS and the endogenous budget share is αS so that household expenditure n,h n,h n,h on services is ES = αS E . The index price for services is the equilibrium price Pin = pn S. 2.2 Production, Supply and Factor Demand The economy produces varieties of several crops, a variety of the manufacture aggregate, services and a variety of the intermediate input. While crops are produced by households, manufactures, services and the intermediate inputs are produced by firms. 2.2.1 Household Agricultural Production Households own land and labor. Land endowments are exogenous. There is a continuum of land in n n,h country n, with measure T . Land is divided across households and a measure T of this land belongs n,h n to household h, where hT = T . We index the zero measure land plots on this continuum with ωT . Households can allocate land freely to produce any of the crops. In this formulation, we assume there is no market for land.6 n Similarly, there is a continuum of workers with measure L in the economy and the labor endowment n,h of the household is denoted with L . We index the zero measure workers on this continuum with ωL . 6 As in Sotelo (2020), the solution of the model is the same if we allow for land to be rented. 9 Unlike land, labor is freely mobile, not only across crops, but also across the other aggregate sectors (manufactures and services) as well as across households. In other words, household members can work on their own plots, work on other farms, or work in manufacturing and services. Households are heterogeneous in land productivity. When allocated to crop j , the plot ωT has productivity ξT,j (ωT ), following Sotelo (2020) and Costinot, Donaldson, and Smith (2016). Unlike these papers, we also allow for labor heterogeneity, as do Lagakos and Waugh (2013), Lee (2020) and Galle, Rodriguez-Clare, and Yi (2021). A worker indexed with ωL has productivity ξL,j (ωL ) in sector j . This heterogeneous labor productivity applies not only to agricultural crops, but also to manufactures and services (as we will explain in more detail below). The combination of heterogeneous productivity in both land and labor is a novel feature of our work, and crucial for appropriate quantification of the distributional impacts of shocks. The household production decision problem consists of allocating different plots ωT of land to different crops. The production process combines land, labor and the intermediate input using a constant returns to scale Cobb-Douglas production function. Output for plot ωT when producing crop j is βL βF n qj (ωT ) = Fjn (ωT ) ˜ n (ωT ) L j [ξT,j (ωT )]βT , ˜ j (ωT ) are the effective units of labor (with productivity ξL,j (ωL )) demanded to work on plot where L ωT . The intermediate input, which is a composite of varieties purchased from the market, is denoted by Fj (ωT ). The variable ξT,j (ωT ) is the productivity shock for plot ωT . The Cobb-Douglas shares of production inputs add up to one, 1 = βF + βL + βT , and are common across crops. Due to the productivity shock ξT,j (ωT ), each plot will specialize in producing one crop. To solve the household land allocation problem, consider a plot ωT with productivity ξT,j (ωT ). Given this productivity and given the price of the national variety of crop j , pn j , the index price of n , and the effective wage, w n , the farmer can derive the optimal use of effective intermediate inputs, PF j labor and intermediate inputs to maximize the profits of producing qj (ωT ) units of output in said plot. Substituting in optimal factor use, the revenue derived from crop j is βL βF βT n 1 βF βT βL yj (ωT ) = pn j βT n n ξT,j (ωT ). (6) PF wj nξ Thus, the land revenue from this plot is equal to rj T,j (ωT ), where the effective return to land can be expressed as βF βL n βF βT βL βT 1 rj = βT pnj βT n n . (7) PF wj We assume that the distribution of the productivity shocks is Frechet, with scale and shape parameters 10 −1 given by γT An,h T,j and θT respectively, where γT ≡ Γ 1 − 1 θT . The household will allocate plot ωT with productivity ξT,j (ωT ) to crop j if this crop delivers the maximum land revenue. That is, if n n ξT,j (ωT )rj ≥ ξT,k (ωT )rk ; ∀k ̸= j . (8) Given the Frechet assumption, the probability of allocating a plot to product j can be expressed as (see Supplementary Material) n An,h )θT n,h (rj T,j πT,j = , (9) (Φn,h T ) θT 1 θT θT where Φn,h T = j ∈S n An,h rj T,j and S is the set of all crops. n,h n,h n,h n The effective units of land allocated to product j are Tjn,h = T πT,j ΦT /rj . The total return on land for household h is equal to n,h n,h RT = Φn,h T T . (10) We can now derive factor demands and output supply. To get the household demand for labor when producing crop j , we write optimal labor use at plot ωT . Then, we integrate it across all plots ωT allocated to crop j and sum over all households. Thus, we find βL +βT βF βT θT −1 n,h ˜n 1 βF βT βL n,h θT T Lj = pn j βT n n πT,j . (11) h PF wj An,h T,j The aggregate demand equation for the composite of the intermediate input is very similar. Finally, we can integrate (6) over ωT and sum over all households to derive the value of national output of product j βL βF βT θT −1 n,h n 1 βF βT βL n,h θT T yj = pn j βT n n πT,j . (12) h PF wj An,h T,j 2.2.2 Firms Firms produce manufactures, services and intermediate inputs. We assume there are many identical firms operating in perfect competition with zero profit. Manufactures (M ) and services (S ) combine (effective) labor and a specific factor, and the value of their output is n yi = pn ˜ n βL,i (K n )βK,i , i (Li ) i (13) where yi is the value of total aggregate output of good i = {M, S }, pn i is the price of one unit of ˜ i is effective units of labor demanded by the producer (with productivity ξL,i ) and Ki is a output, L fixed, specific factor. Given the price pn n i and the effective wage wi , the FOC of the firms profit maximization gives the 11 effective labor demand in sector i 1 ˜n pn i βL,k βK,i Li = n Kin . wi The return per unit of the fixed factor Kin is given by βL,i n 1 βL,i βK,i ri = βK,i (pn i) βK,i n . wi n = r n K n . We assume that households own the The total return to the specific factor in sector i is Ri i i n,h fixed factor, such that Kin = h Ki , and household rents are n,h n n,h Ri = ri Ki . (14) This holds for manufacturing M and services S . To better fit the available data (see Section 3 below), we adopt a linear production function for n , with revenues Rn = pn K n . intermediates F n = KF F F F 2.3 Factor Supply n , K n and K n is exogenous. This implies In each country n, the total supply of the specific factors KM S F that the factor supply to each sector is also exogenous. Factor rewards are given by the rents generated n n,h by this specificity. The total land endowment, T , as well as the household land endowments, T , are also exogenous. The land supply to different sectors, and the value of land rents, was addressed above. n n,h The total labor endowment, L , as well as the household labor endowments, L , are also exogenous. The remaining task is to determine labor supply to different sectors, which we do next. n,h Consider a household with labor endowment L . Each unit of labor ωL can be allocated to the production of J different products, including crops, manufactures and services. Households make their labor allocation decisions based on the productivity of labor for different crops and products, which n . When labor indexed with ω we denote ξL,j (ωL ), and the market wage for producing each good, wj L n . For each ω , the household will is allocated to crop/product j , the return is equal to ξL,j (ωL )wj L maximize labor income by allocating each labor unit to the sector with the highest return. If the optimal choice is a crop, then a given unit of labor can be allocated either to own-household plots or to off-farm plots. If the optimal choice is manufactures or services, then labor is hired by firms. We assume that ξL,j (ωL ) is Frechet distributed with scale and shape parameters γL An,h L,j and θL respectively. The scale parameter, which determines average productivity, depends on the product and −1 1 country, where γL ≡ Γ 1 − θL . Characterization of the optimization problem requires calculating the probability of choosing a specific production activity for labor and the effective units of labor based 12 on this allocation decision. A unit of labor will be allocated to sector j when n n ξL,j (ωL )wj ≥ ξL,k (ωL )wk ; ∀k ̸= j . (15) Based on the properties of Frechet distribution, we can write the probability of allocating one unit of labor to market j , given the parameters (γL An,h L,j , θL ), as θL An,h n L,j wj n,h πL,j = θL , (16) Φn,h L 1 θL θL where Φn,h L = j ∈S ′ An,h n L,j wj and S ′ is the set of sectors that employ workers, including all crops, manufacturing and services. The next step is to calculate the effective units of labor supply for each crop, taking productivity draws into account. From the Frechet assumptions, the total effective units of labor allocated to j , conditional on optimality, is equal to n,h n,h ΦL n,h Ln,h j = πL,j n L , (17) wj This delivers the productivity adjusted labor supply by household h to each crop/product j . Note that the probability of allocating labor to j is also equal to the share of the return to labor allocated to j n,h relative to the total return to labor, that is πL,j = Ln,h n j wj / k Ln,h n k wk . n,h Finally, the total wage income (return on labor) of household h defined as RL ≡ j Ln,h n j wj is equal to n,h n,h RL = Φn,h L L . (18) The proof for this statement is provided section A3 in the Supplementary Material. 2.4 Equilibrium Definition. The international trade equilibrium is given by a vector of crop prices for each crop variety in each country, pn n j ; a vector of manufacturing prices for each country variety, pM ; a vector of services prices in each country, pn n S ; a vector of intermediate input prices for each country variety, pF ; a vector n ; a vector of return of wages for each product (crops, manufacturing and services), for each country wj n ; and rental rates for the specific factors in manufacturing r n , services r n and intermediate on land rj M S n , such that: inputs rF Goods Market. For each product, global demand equals national supply. For crops j , we combine 13 total household expenditures (5) and the value of national output (12) to express the equilibrium condition as xn,m j n,h RT n,h + RM n,h + RS n,h + RL m = yj , (19) n h where xn,m j is the import share based on the price vector pn n,h j defined by (3) and (4), αj is the n,h n,h expenditure share on composite j given by (1) , RM and RS are the fixed factor revenues in sectors n,h n,h M and S (defined by (14)) accruing to household h, RT and RL are given by (10) and (18) and n,h n,h n,h n,h RT + RM + RS + RL = E n,h is total household expenditure. The revenue function of crops at the national level, (12), gives the right hand side of the equation. The equilibrium for manufactures is the same as (19) with j = M xn,m M αM n,h n,h RT n,h + RM n,h + RS n,h + RL m = yM , (20) n h where the revenue function (13) gives the right hand side of the equation. Supply and demand of the non-traded good requires that n,h n,h n,h n,h n,h αS (E n,h ) RT + RM + RS + RL n = yS . (21) h For the intermediate input F , we have n h j xn,m n,h F β F yj m. = yF Factor Markets. The labor allocation problem of households imply that θL − 1 n,h n n,h θL L βL,j yj πL,j = n , (22) h An,h L,j wj n,h n given equation for each sector j (any crop, manufacturing or services) where πL,j is a function of wj (16). The land allocation problem of the households imply that θT −1 n,h n n,h θT T βT,j yj πT,j = n , (23) h An,h T,j rj n,h n given equation (9). For services and manufacturing, we for any crop j where πT,j is a function of rj n y n /r n and for the inputs we have K n = y n /r n . The equilibrium price satisfies have Kin = βK,i i i F F F βF,j n βL,j n βT,j pn wj rj pn j = F n , (24) βF,j βL,j βT,j wjn βL,j n rj βT,j for crops, and pn j = βL,j βT,j for services and manufacturing, and finally pn n F = rF for inputs since there is only one factor of production for inputs (i.e. fertilizers). Note that the model could be solved in levels using the equations above by plugging in productivity 14 parameters for each choice. Alternatively, it can also be solved in changes, i.e. using hat algebra, which considerably reduces the data requirements. We provide the full solution method using hat algebra in detail in section A1 in the Supplementary Material.7 2.5 Welfare Due to the non-homothetic nature of preferences, we cannot simply divide nominal income to a common price index to calculate welfare or real income. Instead, household welfare can be captured by the indirect utility function   n n n νj E n,h − j C j Pj  V n,h = n νj .   n νj j j Pjn To explore distributional issues in the data, we work with a linear transformation of V n,h such that ρn V n,h + γ n = E n,h , that is, indirect utility is represented with household expenditures in the baseline. Then, the transformed indirect utility function will allow us to express changes in household welfare as equivalent changes in real income. The change in welfare, or real income, can be defined as the counterfactual change in expenditure at initial price levels that would give the equivalent change in utility after a shock. The aggregate household welfare effect comprises an income effect and a consumption or cost of living effect. The income effect is the change in welfare caused by a change in nominal income at constant prices. The cost of living effect is the change in welfare caused by a change in prices at constant nominal expenditures. See Appendix A2 for the derivation of expressions for welfare. 3 Data and Estimation To solve the model, we need to characterize household decisions using the initial allocations and response elasticities. Since the solution is in changes, using hat-algebra as in Dekle, Eaton, and Kortum (2008), the data requirements comprise household land and labor allocations, household utility function parameters, household and firm production function parameters, international trade shares, and elasticity parameters. In our setting, all of this is data, except for the elasticity parameters, which are estimated. These data requirements, data sources, and the estimated parameters are reported in Table 1. The calibrated and estimated parameters and coefficients then can be used to solve the model as explained in Appendix A1. 7 The equilibrium can alternatively be characterized without using the return on land directly, by substituting it out, as discussed in appendix A5. 15 3.1 Data Household data. We begin with household-level variables. Information on households’ expenditure shares, crop-specific land allocation shares, and sector-specific labor shares is taken from the Household Impacts of Tariffs (HIT) database (Artuc, Porto, and Rijkers (2020)). This dataset contains highly disaggregated information on household budget and income shares from representative harmonized household surveys for 51 low- and middle income countries.8 These surveys cover 300 million households and 1.6 billion people. For each country, households are grouped into 100 income bins n,h each representative of a percentile of the income distribution. Consumption shares αj and labor allocation shares, π n,h n,h L,j , are taken directly from the data. Land shares π T,j are calculated by using the fact that a crop’s total output share must equal it’s total land allocation share in equilibrium. ιn,h The initial shares of household income from land and labor, denoted by ¯T ιn,h and ¯L , are also taken directly from the data. These shares enable us to calculate the changes in household income based on changes in land and labor income implied by the model. Fixed factor shares are also calculated based on households’ sales of manufacturing and services in the HIT database. To close the model, for each country we construct a residual household which is excluded from the 100 bins per country we use in our analysis, and allocate fixed factors that are not owned by households to it. n,h The initial shares of each household in the national supply of land and labor, denoted by ηT,j and n,h n,h ηL,j , and the initial share of each household in the national demand of crop j , denoted by Dj , are also taken from the HIT database. This information is used to calculate the change in aggregate labor and land supply to each specific crop, as well as the change in aggregate expenditures. Trade and production data. The second piece of data is the trade data, which we take from the “International Trade and Production Database for Estimation” project by Borchert, Larch, Shikher, and Yotov (2021) and Borchert, Larch, Shikher, and Yotov (2022). ITPD-E provides product specific import and export shares xn,m j and xn,m j respectively. The ITPD-E data also provides domestic absorption rates, which are necessary to calculate the total output. In order to harmonize the HIT and ITPDE data, it is necessary to create consistent product categories. Since the HIT data is more detailed for agricultural products, we aggregate some products 8 The 51 countries are Armenia, Azerbaijan, Bangladesh, Benin, Bhutan, Bolivia, Burkina Faso, Burundi, Cambodia, Cameroon, the Central African Republic, Comoros, Cˆ ote d’Ivoire, Ecuador, the Arab Republic of Egypt, the Gambia, Georgia, Ghana, Guatemala, Guinea, Guinea-Bissau, Indonesia, Iraq, Jordan, Kenya, Kyrgyzstan, Liberia, Madagascar, Malawi, Mauritania, Moldova, Mongolia, Mozambique, Nepal, Nicaragua, Niger, Nigeria, Pakistan, Papua New, Guinea, Rwanda, Sierra Leone, South Africa, Sri Lanka, Tajikistan, Tanzania, Togo, Uganda, Uzbekistan, Viet Nam, the Republic of Yemen and Zambia. 16 in HIT to match ITPD-E.9 Since the HIT countries do not cover a significant proportion of the world production for the crops we consider in the model, we add 47 relatively large countries for which we have ITDPE-data to account for the rest-of-the-world production and trade.10 We assume a single representative household for countries that were not included in HIT data. This representative household produces all agricultural products as well as services and manufacturing outputs using fixed factors. A key parameter that we need to solve for the international trade equilibrium is the trade elasticity. We use the estimate from Simonovska and Waugh (2014), which gives us 1 − σ = −4.0. 3.2 Calibration of the utility function parameters In this section, our goal is to calibrate utility function parameters, which in turn will facilitate the calculation of endogenous expenditure shares and changes in welfare. n,h Letting κn n n n j ≡ C j Pj , we write equation (1) as function of νj and κj such that n,h n γn κn j αj (Θ) = νj 1− + , E n,h E n,h where Θ = {κn n n j , νj | j ∈ {1, 2, ., J }} and γ = j κn j . This expression gives us the implied expenditure n and κn by minimizing the share given coefficients of interest, i.e. elements of Θ. Then, we calibrate νj j square of Euclidean distance between implied shares and the shares observed in the data: 2 n,h n,h Θ∗ = arg min α ¯j − αj (Θ) , (25) h j n,h subject to γ n = j κn n j , βj ≥ 0 and n k νj ¯j = 1, where α are expenditure shares in the data, and n,h αj (Θ) are expenditure shares implied by calibrated coefficients. The calibrated coefficients κn j and n allow us to calculate the endogenous expenditure shares, αn,h , which are needed to solve the model νj j given changes in household expenditures and prices. 9 Table A4 in the Supplementary Material presents the concordance between HIT and ITPD-E we developed. The final set of sectors is as follows: 1. Wheat, 2. Rice, 3. Corn, 4. Other cereals, 5. Soya, 6. Other oilseeds, 7. Sugar, 8. Legumes, 9. Fruits and vegetables, 10. Nuts, 11. Eggs, Meat and Dairy, 12. Confectionery and Cocoa, 13. Oils and Fats, 14. Other staple food, 15. Beverages, 16. Cotton, 17. Tobacco, 18. Spices/herbs, 19. Alcohol, 20. Fish, 21. Manufacturing, 22. Services, and 23. Fertilizers and other chemicals as agricultural inputs. 10 These additional economies are Argentina, Australia, Austria, Belgium, Bulgaria, Brazil, Canada, Switzerland, Chile, China, Colombia, Czechia, Germany, Denmark, Spain, Finland, France, United Kingdom, Greece, Hungary, India, Israel, Italy, Japan, Kazakhstan, the Republic of South Korea, the Lao People’s Democratic Republic, Morocco, Mexico, Malaysia, Netherlands, Norway, Peru, Philippines, Poland, Portugal, Romania, the Russian Federation, Saudi Arabia, Singapore, the Slovak Republic, Sweden, Thailand, Tunisia, T¨ urkiye, Taiwan, China and the United States. 17 Table 1: Parameters and initial shares Initial Household Allocations (data) νjn , κn Utility function coefficients HIT (Calibrated) j π n,h T,j π L,j , n,h Land and labor shares HIT ιn,h ¯T , ¯ ιn,h L Share of income from land and labor HIT n,h n,h ηT,j , ηL,j Households’ share in total factor supply HIT n,h Dj Households’ share in demand HIT Initial International Trade Shares (data) xn,m j , xn,m j Import and export shares ITPDE Parameters 1−σ -4.00 Trade elasticity Simonovska and Waugh (2014) βL,j 0.55 Labor elasticity agriculture Sotelo (2020) βT,j 0.22 Land elasticity agriculture Sotelo (2020) βF,j 0.23 Fertilizer elasticity agriculture Sotelo (2020) βL,j 0.75 Labor elasticity manufacturing and services Cobb-Douglas Response Elasticities θT 1.34 Frechet shape - land elasticity Own estimate (1.12, 1.57) 95% confidence interval θL 2.15 Frechet shape - labor elasticity Own estimate (1.57, 5.20) 95% confidence interval Notes: HIT=Household Impacts of Tariffs Database (Artuc, Porto, and Rijkers (2020)), ITPDE=International Trade and Production Database for Estimation, (Borchert, Larch, Shikher, and Yotov (2021)). Note that utility function parameters and initial shares vary both across countries and by income percentile which is why it is not practical to display them in one table. For the same reason, import and and export shares, which vary by country and product, are not displayed. 18 3.3 Estimation of land and labor elasticities Central to the model are the Frechet shape parameters θT and θL which govern land and labor allocation decisions. These are the two most important parameters of the model. We estimate the elasticities with a non-linear least squares estimator by matching revenues from each crop across countries. The estimation of the elasticities requires information on crop productivity and crop prices. Since these data are not available (on a consistent basis) at the household level, we use aggregate country-level data instead. Note that the revenue from crop j as a function of θL and θT , for a given aggregate labor πn n L,j and land π L,j allocation is βT βL θT −1 1−βF θL −1 1−βF n Yjn (θT , θL ) = pn j Aj πn T,j θT πn L,j θL , (26) where pn n n j is the price of crop j in country n. Assuming AT,j and AL,j are common across households within a given country, the crop-specific productivity is βM βL βT n βM 1−βM Aj = n An L,j 1−βM An T,j 1−βM . (27) PM These productivities are calculated using Food and Agriculture Organization’s Global Agro-Ecological Zones (GAEZ) data set. The aggregate land and labor allocations, π n n L,j and π T,j , are computed using the Household Impacts of Tariffs database (Artuc, Porto, and Rijkers (2020)). The elasticities are estimated as 2 ∗ ∗ n Yjn (θT , θL ) { θT , θL } = arg min Y −j n . (28) θT ,θL n j k Yk (θT , θL ) n We are thus implicitly matching the share of output j , Y , taken from Artuc, Porto, and Rijkers (2020), with the share of output predicted by the model (given the parameters). The sample size includes the 51 HIT countries and 13 crops.11 To improve the estimates, we impute the production function shares for the agricultural products from Sotelo, that is we set βT = 0.22, βL = 0.55 and βF = 0.23. We assume βL = 0.75 for the manufacturing and services sectors following the original Cobb-Douglas estimate. The estimated elasticities are reported in the bottom panel of Table 1. Our procedure yields ∗ = 1.34 and of the labor elasticity θ ∗ = 2.15, which are both estimates of the land elasticity θT L significantly larger than 1. The confidence intervals are calculated by bootstrapping the sample 5000 times. These estimates are in line with those obtained in related literature. For example, the estimated 11 We exclude a few crops that cannot be matched to GAEZ (eggs/meat, cocoa, oils/fat, other staple food, spices, alcohol, and fish). 19 land elasticity in Costinot, Donaldson and Smith is 2.46, which is higher than ours (1.34). Our estimate of the labor elasticity (2.15) is well within the range of elasticities reported by Galle, Rodriguez-Clare, and Yi (2021) which vary between 1.42 and 2.79, as well as those reported by Hiesh et al. (2013) and Burnstein et al. (2019) which range from 1.20 to 3.44. 4 Agriculture and the Russian Federation’s invasion of Ukraine The Russian Federation’s invasion of Ukraine has a myriad of political, economic, and humanitarian consequences. We focus on its impacts on agricultural trade and production on third countries. In particular, this section uses our model to assess how the agricultural trade disruptions caused by the war have impacted household welfare in developing countries. The onset of the war coincided with a surge in food prices. Ukraine and Russia are both important agricultural suppliers. On the eve of the war, Russia accounted for about 15% of global wheat exports, while Ukraine accounted for about 17% of global corn exports. Russia was also one of the largest fertilizer exporters globally, accounting for about 18% of potassic, 14% of phosphatic, and 16% of nitrogenous exports in the world. The war resulted in substantive supply disruptions in 2022, higher uncertainty, increased stockpiling, and various trade bans and retaliatory trade sanctions. The observed surge in food prices is thus plausibly, at least in part, attributable to the war. The price hikes affect real incomes. Exposure to war-induced supply disruptions varies with how reliant countries are on food and fertilizer supplied by Ukraine and Russia. In our sample, the majority of all wheat imports by Armenia, the Arab Republic of Egypt, Georgia, Jordan and Nicaragua was supplied by Ukraine and Russia prior to the onset of the war. Ukraine and Russia also supplied over half of all fertilizers imported by Georgia, Mongolia, Moldova and Azerbaijan. These countries may experience severe losses due to the war. By contrast, countries that rely less on imports from Ukraine and Russia, such as Comoros or Madagascar, may lose little, while countries that are themselves net exporters of wheat and fertilizer might yet benefit. Within countries, household exposure depends on their consumption and income earning patterns. If food prices increase, net food consumers lose while net producers gain. The overwhelming majority of households in developing countries is net buyers of agricultural products and fertilizers (61% in our sample). On average, across the countries in the HIT dataset, households spend 44% of their income on agricultural products and derive 39% of their income from selling agricultural goods. Poorer households spend a larger share of their budget on food and are thus systematically more exposed to food price 20 inflation. The high variation in household income and consumption portfolios is at the root of the heterogeneity in the impact of the war on different households. We assess the welfare and inequality effects of the war on third countries with counterfactual simulations. We focus primarily on the agricultural trade and production disruption in the countries directly involved in the war. Our main simulation assumes (i) that Ukraine cannot import or export any agricultural products, or fertilizers and (ii) that Russia bans exporting wheat, rice, corn, other cereals, sugar, other oilseeds, and fertilizers to the world. These bans were extensively discussed in the media, including coverage from CNBC (2022), NPR (2022), and The Wall Street Journal (2022) among others. Export bans are conceptualized as prohibitive increases in transportation costs that push exports of banned products to zero.12 The remainder of this section is organized as follows. In section 4.1, we evaluate the goodness of fit of our model by assessing whether its predictions correlate with actual food price increases observed in developing country markets. The opportunity to validate the model’s predictions with real data is in fact an important reason to study the war shock. In section 4.2 we present the results of the simulations and in section 4.3 we unpack the mechanisms by which the war has impacted household welfare in developing countries. 4.1 Goodness of Fit The fact that the war shock has already occurred can be exploited to assess the goodness of fit of our model. To do that, observed country and crop specific price changes are compared with price changes predicted by our model. Observed price changes come from the FAO’s Global Information and Early Warning System (GIEWS) Food Price Monitoring and Analysis (FPMA) dataset, which collects price changes for the most commonly consumed staple commodities. These data cover 19 different products that we can match to the harmonized HIT-ITPDE data for 37 of the 51 HIT countries.13 The sample is not balanced, i.e., the number of products covered varies across countries. In interpreting the goodness of fit results, it is important to bear in mind that our model abstracts from many relevant aspects of reality, such as uncertainty, stockpiling, and speculation, which influence the pricing of agricultural commodities. 12 The spectrum of potential simulations to run with our model is vast. To illustrate, we present in the Supplementary Material a number of additional complementary simulations. In particular, we examine the role of supply disruptions in Ukraine production and we explore the role of retaliatory protectionism by other countries. 13 The concordance between the HIT-ITPD-E data and the WFP food prices dataset is presented in Table A5 in the Supplementary Material. 21 Table 2: Goodness of Fit Dependent Variable Observed Price Change Sample All products Selected Products corn, wheat, rice, soya, sugar, and other cereals No outliers No outliers (1) (2) (3) (4) Predicted price change 0.842∗∗∗ 1.281∗∗∗ 0.849∗∗∗ 1.589∗∗∗ (0.257) (0.226) (0.311) (0.458) Country FE Yes Yes Yes Yes R-squared 0.319 0.409 0.667 0.736 Adj. R-squared 0.223 0.321 0.475 0.577 Partial correlation (Observed vs Predicted) 0.160 0.211 0.313 0.487 Obs. 299 287 102 94 Notes: *, **, and *** denote significance at the 10%, 5%, and 1% significance level respectively. Standard errors are reported in parentheses and are clustered two ways, by country and product. All estimates are obtained using Ordinary Least Squares (OLS) estimation. Columns 1 and 2 present regressions for the entire sample of products for which we have price data, spanning 19 products from 37 different countries. The dependent variable is changes in average monthly prices observed between April 2022 and February 2023 versus April 2021 and February 2022. Note that these are annual averages excluding March, the first full month after the war started, to limit the impact of potential overshooting of initial price responses. Columns 3 and 4 present regressions for a selected sample of products, notably corn, wheat, rice, soya, sugar, and other cereals. Columns 2 and 4 present regressions in which observations in the top and bottom 1% of the realized price change distribution are excluded. The partial correlation reflects the correlation between observed price changes, residualized on country fixed effects, and predicted price changes, also residualized on country fixed effects. Nonetheless, the model has strong predictive power, as is shown in Table 2. We report the results of regressions in which observed price changes are regressed on the price change predicted by our model. The observed price changes are computed as the difference in average monthly price observed between April 2022 and January 2023 relative to the average monthly price observed between April 2021 and January 2022. We use a 10 month average to minimize the impact of seasonality and temporary price fluctuations.14 All regressions include country fixed effects to account for differences in overall inflation associated with country-specific conditions (such as monetary policy, exchange rate shocks etc.). The model predicted price changes have high explanatory power and strongly and significantly predict realized price changes. The coefficient on predicted price changes is 0.842 and the partial correlation between observed and predicted price changes is 0.16015 . This coefficient rises to 1.281 once we exclude the top and bottom 1% of the observed price changes, as is done in column 2. Note also that the explanatory power of the model and the partial correlation between observed and predicted price changes improves when outliers are excluded. The explanatory power of the model is even more striking when attention is confined to a subset of 14 February and March are excluded to minimize the impact of initial price spikes which partially reflect elevated uncertainty. In robustness tests that are not presented here to conserve space but available upon request we verify that the results are also robust to using different time windows. 15 The model predicts 16% of the residual variation in price movements once country fixed effects are controlled for. 22 product categories—corn, wheat, rice, soya, sugar, and other cereals— less susceptible to measurement error due to being fairly homogeneous.16 These results are reported in columns 3 and 4. The coefficient on model predicted price changes is 0.849 when raw data are used (column 3), and rises to 1.589 once outliers are excluded (column 4). The partial correlation between predicted price changes and observed price changes for this last specification is 0.487. The model thus accurately predicts observed agricultural price responses to the war. 4.2 Results The war leads to substantial economic losses which are strikingly heterogeneous across developing countries.17 This heterogeneity is shown in Table 3 and in Figure 1. The top panel presents a kernel density of the real income gains associated with the war pooling all households in our sample. The bottom panel reports a ridgeline plot of kernel densities for each country separately—sorted in order of their average gains (with the highest gains at the top). The collage of welfare effects is apparent. The overwhelming majority of households in our sample (99.6% of observations) experiences a reduction in their real incomes as a consequence of the war. The average loss across households is –2.90% with a standard deviation of 2.17% (column 1). Almost three quarters of all households experience losses in the range of –3 to 0 percent, and roughly 90% suffer losses of up to –6%, but the distribution of losses is skewed and has a large left tail. The largest average losses are incurred by households living in countries located close to Russia such as Azerbaijan (–10.94%), Mongolia (–10.49%), Georgia (–7.87%) and Armenia (-7.61%). On the other end of the spectrum, the lowest average losses are observed in Iraq (–0.40%), Pakistan (–0.66%) and Bhutan (–0.82%). There is also a very large within-country variation in the losses, with some households even experiencing positive welfare gains. This pattern cannot be captured by most existing trade models based on a single aggregate agent. The within-country heterogeneity can be seen in panel b) of Figure 1, which reveals the wide dispersion in the household welfare effects in each country and, in turn, how this dispersion varies across countries. In Azerbaijan for instance, all households lose from the war, but the losses range from a minimum of –11.72% to a maximum of –10.28%. In Togo, all households lose, too, but the range of losses goes from –3.34% to –1.62%. By contrast, in Pakistan, impacts are mixed. About 82% of the population loses real income, but the remainder gains slightly. The impacts 16 The other product categories are Beverages, Confectionery/Cocoa, Eggs/Meat/Dairy, fertilizers, fruits and vegetables, fish, other oilseeds, oils/fats, nuts, manufacturing, legumes, spices/herbs. 17 It is important to emphasize that we study the economic consequences of the war in terms of real income/expenditure gains or losses. We do not discuss impacts on Ukrainian households themselves, since they suffer from the war in a multitude of ways that our model does not capture. 23 range from –1.79% to 0.33%. A unique advantage of our approach is that it not only yields granular estimates of the distribution of impacts, but also allows us to pinpoint who these accrue to. In particular, since we have information on each household’s per capita expenditure, we can examine the welfare effects and their distribution across the developing world income spectrum. To show this, Figure 2 presents country-specific scatter plots of real income gains against the pre-war log of household expenditure per capita (in real 2010 USD) using all observations in our data. Each observation represents a percentile of the population in a particular country. The graph shows widespread losses across the entire income distribution of low- and middle-income countries. There are, however, a few households with gains in a few selected countries (such as Iraq and Pakistan). The heaviest losses show up towards the right tail because the countries with the highest losses tend to have higher per capita incomes (relative to the other countries in our sample) as well. The local polynomial regression of real income gains against log per capita expenditure is relatively flat, showing an average loss of slightly more than 2% for most of the left tail of the income spectrum. However, it dips to about 5% at the upper-right tail and steadily inclines afterwards, implying lower losses for the comparatively richest households. Within countries, poor households tend to incur the largest losses. This is illustrated in Figure 3 which presents a local polynomial plot of the welfare gains against the household rank in their country’s own income distribution. Rank 1 is the bottom percentile and rank 100 is the top percentile. The graph is clearly upward sloping. On average households at the bottom of the income distribution suffer larger losses than households at the top. The losses for the bottom 25% (-3.11% on average) are 11.8% higher than the losses for the top 25% (-2.78% on average). See columns 2 and 3 of Table 3. On average, within-country inequality increases. Since our model features household heterogeneity, we can measure the aggregate welfare effects with more precision than single-household models. This is because the average welfare impact across heterogeneous households making different production and consumption decisions does not coincide with the welfare impact for a single household characterized by aggregate level data. There are two drivers of this bias.18 The structure of the crop production technology and of preferences prevents linear aggregation of the individual household welfare effects. On top of that, the differences in household choices from the choice of a single household affect the equilibrium of the model (because household heterogeneity interacts with the international trade linkages in equilibrium—see also section 4.3). Column 4 of Table 3 presents the average welfare effects calculated from a version of our model 18 We show this bias formally and discuss the mechanisms underpinning it in the supplementary material A4. 24 Figure 1: Distribution of welfare effects of the Russian Federation’s invasion of Ukraine (a) Across countries Density -15 -12 -9 -6 -3 0 3 welfare gain (b) By country Iraq Pakistan Bhutan Nepal Comoros Bangladesh Cambodia Kenya Guinea Malawi Togo Madagascar Benin Sri Lanka Burundi Sierra Leone Uganda Vietnam Papua New Guinea Guatemala Ghana Nigeria Bolivia Niger Indonesia Jordan Liberia Tajikistan Yemen Guinea-Bissau South Africa Tanzania Zambia Egypt, Arab Rep. Rwanda Gambia Cote d'Ivoire Mozambique Mauritania Cameroon Ecuador Uzbekistan Nicaragua Burkina Faso Kyrgyzstan Central African Republic Moldova Armenia Georgia Mongolia Azerbaijan -15 -10 -5 0 welfare gain Notes : The top graph depicts the kernel density distribution of the estimated welfare impacts of the war via its impacts on agricultural trade, expressed in real income gains (in percent), across all countries in our sample. The bottom graph presents kernel density graphs by country, sorting countries in terms of the average real income gain (in percent). Countries with the highest average gains, such as Pakistan and Iraq are at the top, and those with the lowest gains, such as Azerbaijan and Mongolia, at the bottom. Darker shades denote greater average losses. 25 Table 3: Impacts of the Russian Federation’s invasion of Ukraine ∆Welfare ∆Income ∆Cost of Exposure Average Bottom Top Single Total Labor Land Living Imports 25% 25% HH. Panel A: All countries (pooled) Average -2.90 -3.11 -2.78 -1.87 0.14 0.12 0.11 -3.04 5.46 Pop w. average -2.10 -2.34 -1.93 -1.09 0.56 0.44 0.70 -2.64 4.31 Std. dev. 2.17 2.32 2.12 2.03 1.56 1.18 2.16 1.25 5.36 Minimum -10.94 -11.03 -11.03 -9.65 -6.53 -4.59 -8.25 -9.07 0.30 Median -2.26 -2.40 -2.11 -1.36 0.54 0.42 0.67 -2.76 3.73 Maximum -0.40 -0.38 0.04 1.07 2.22 2.05 2.85 -1.33 19.23 Panel B: By country Azerbaijan -10.94 -11.03 -10.76 -9.65 -6.53 -4.59 -8.05 -4.70 18.18 Mongolia -10.49 -10.48 -11.03 -9.06 -5.07 -3.39 -8.25 -5.68 6.68 Georgia -7.87 -8.99 -7.13 -6.27 -1.53 -1.26 -1.95 -6.42 19.02 Armenia -7.61 -9.48 -6.26 -6.07 1.56 1.20 2.30 -9.07 17.52 Moldova -6.92 -7.38 -6.34 -4.03 -2.88 -2.05 -5.31 -4.14 19.23 Cent. Afr. Rep. -4.03 -3.87 -4.23 -3.73 -1.11 -1.25 -1.11 -2.95 1.12 Kyrgyzstan -3.84 -4.18 -3.57 -3.05 -0.30 -0.24 -0.37 -3.55 11.76 Burkina Faso -3.80 -3.95 -3.69 -2.22 -1.02 -0.95 -1.04 -2.81 1.35 Nicaragua -3.69 -4.19 -3.41 -2.79 -1.37 -1.24 -1.53 -2.35 1.42 Uzbekistan -3.43 -3.69 -3.49 -1.72 1.42 0.26 2.65 -4.78 17.35 Ecuador -3.34 -3.81 -2.99 -2.35 -1.09 -0.96 -1.31 -2.27 14.18 Cameroon -3.34 -3.64 -3.09 -2.49 -0.78 -0.76 -0.79 -2.58 0.63 Mauritania -3.07 -3.16 -2.99 -1.67 0.65 0.53 0.67 -3.70 2.85 Mozambique -3.04 -2.66 -3.32 -2.79 0.34 0.19 0.40 -3.37 6.77 ote d’Ivoire Cˆ -2.94 -3.25 -2.73 -2.06 -0.92 -0.73 -0.98 -2.03 4.16 Gambia, The -2.81 -2.64 -2.91 -2.01 -0.03 -0.03 -0.02 -2.78 2.19 Rwanda -2.75 -2.72 -2.81 -0.71 0.21 0.15 0.30 -2.95 1.77 Egypt, Arab Rep. -2.73 -3.26 -2.33 -2.27 0.98 0.76 1.30 -3.69 7.06 Zambia -2.67 -2.77 -2.62 -1.73 0.74 0.54 1.00 -3.38 2.96 Tanzania -2.64 -2.97 -2.41 -1.19 -0.13 -0.13 -0.15 -2.51 5.04 South Africa -2.41 -3.19 -2.05 -0.93 -0.02 -0.02 0.18 -2.39 3.69 Notes : This table presents the results of a simulation in which (i) Ukraine cannot import or export any agricultural products, or fertilizers and (ii) Russia bans exporting wheat, rice, corn, other cereals, sugar, other oilseeds, and fertilizers. Statistics in Panel A refer to average welfare impacts across countries. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure measures what share of imports of a given country are accounted for by imports from Ukraine and Russia before the war. Countries are ordered in terms of average real income gains (from lowest to highest). 26 Table 3: Impacts of the Russian Federation’s invasion of Ukraine (continued) ∆Welfare ∆Income ∆CL Exposure Average Bottom Top Single Total Labor Land HH. Imports 25% 25% HH. Guinea-Bissau -2.36 -2.38 -2.36 -0.90 0.92 0.88 0.94 -3.25 3.78 Yemen, Rep. -2.35 -2.75 -2.08 -1.67 0.65 0.55 0.80 -2.99 3.76 Tajikistan -2.30 -2.38 -2.26 -1.61 0.80 0.78 0.92 -3.09 9.54 Liberia -2.29 -2.26 -2.30 -1.71 0.49 0.41 0.53 -2.76 2.05 Jordan -2.26 -2.53 -2.04 -1.92 0.41 0.33 0.68 -2.66 0.68 Indonesia -2.26 -2.40 -2.15 -1.21 -0.12 -0.11 -0.18 -2.14 2.62 Niger -2.23 -2.08 -2.37 -1.10 0.66 0.51 0.69 -2.88 5.88 Bolivia -2.17 -2.50 -1.95 -1.57 0.19 0.13 0.24 -2.36 1.54 Nigeria -2.14 -2.35 -2.08 -1.38 0.54 0.43 0.65 -2.67 2.11 Ghana -2.13 -2.62 -1.88 -0.82 -0.81 -0.69 -1.26 -1.33 3.27 Guatemala -2.08 -2.10 -2.06 -1.38 0.14 0.12 0.19 -2.21 1.02 Papua New G. -2.07 -2.07 -2.11 -0.71 1.20 0.85 1.36 -3.23 1.29 Viet Nam -2.06 -2.25 -1.93 -1.31 0.03 -0.00 0.06 -2.10 2.72 Uganda -2.06 -2.20 -2.02 -0.74 0.65 0.59 0.69 -2.69 2.13 Sierra Leone -1.94 -1.84 -2.05 -1.14 0.95 0.73 0.98 -2.87 4.12 Burundi -1.90 -2.11 -1.86 -0.98 0.70 0.55 0.73 -2.59 4.74 Sri Lanka -1.86 -2.04 -1.74 -0.51 0.44 0.31 0.64 -2.29 15.57 Benin -1.84 -1.85 -1.87 -1.03 0.62 0.52 0.67 -2.44 1.78 Madagascar -1.84 -1.78 -1.92 -0.46 1.01 0.84 1.11 -2.83 0.59 Togo -1.82 -1.86 -1.81 -1.36 0.39 0.42 0.38 -2.20 2.97 Malawi -1.64 -1.44 -1.78 -0.73 0.74 0.59 0.83 -2.36 8.34 Guinea -1.51 -1.53 -1.54 -0.76 1.11 0.88 1.14 -2.59 0.96 Kenya -1.48 -1.50 -1.52 -0.73 1.30 1.15 1.48 -2.75 4.81 Cambodia -1.42 -1.61 -1.27 -0.62 0.69 0.45 0.84 -2.09 3.32 Bangladesh -1.42 -1.36 -1.46 -0.87 0.98 0.70 1.16 -2.38 4.65 Comoros -1.38 -1.42 -1.47 -0.76 1.67 1.50 1.77 -3.01 0.30 Nepal -0.85 -1.20 -0.58 0.28 2.07 2.05 2.11 -2.87 7.64 Bhutan -0.82 -0.91 -0.69 0.19 1.56 1.36 1.94 -2.35 3.79 Pakistan -0.66 -1.39 0.04 1.07 2.22 2.01 2.75 -2.83 3.77 Iraq -0.40 -0.38 -0.42 -0.10 1.59 1.04 2.85 -1.96 3.73 Notes : This table is a continuation from the table on the previous page. It presents the results of a simulation in which (i) Ukraine cannot import or export any agricultural products, or fertilizers and (ii) Russia bans exporting wheat, rice, corn, other cereals, sugar, other oilseeds, and fertilizers. Statistics in Panel A refer to average welfare impacts across countries. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure measures what share of imports of a given country are accounted for by imports from Ukraine and Russia before the war. Countries are ordered in terms of average real income gains (from lowest to highest). 27 Figure 2: Impacts of the war by initial income - all countries 2 Pakistan 0 Nepal Bhutan Iraq -2 -4 welfare gain Moldova -6 Georgia -8 -10 Azerbaijan Mongolia -12 -14 2 4 6 8 10 12 log per capita expenditure Notes : The graph plots the estimated welfare impacts of the war via its impacts on agricultural trade, expressed in real income gains, across all countries in our sample against pre-war log per capita household expenditure. Each observation (denoted by a “+”) represents an income percentile in a different country. Each country is demarcated by a different color—with lighter colors denoting higher average gains. The blue line is a fitted polynomial regression. Figure 3: Impact of the war by initial (within-country) income rank -2.6 -2.8 welfare gain -3.2 -3.4 -3.6-3 0 20 40 60 80 100 income rank Notes : The graph depicts a local polynomial fitted line of the percentage change in welfare, measured as real income, associated with the war’s impact on agricultural trade against a household’s rank in the initial per capital income distribution, with rank 100 denoting the richest percentile and rank 1 denoting the poorest percentile. The shaded area depicts the 95% confidence interval. Darker colors depict larger average losses. 28 based on a single household for each country.19 According to this model the average loss across countries is –1.87%, which is 35.52% smaller than the average loss estimates based on heterogeneous households (-2.90%). There are, however, instances where the bias is much bigger. Some notable examples are Pakistan, Nepal and Bhutan. 4.3 Mechanisms Why does the war cause real incomes to decline for most households? Why are these losses so heterogeneous? Why are they so much larger for certain countries (mostly at the upper stratum of world income)? Why does within-country inequality increase? Our theoretical model offers insights to illuminate the answers to these questions. First, we have the “trade channel.” The war led to the imposition of trade barriers that resulted in changes in prices across countries. The trade relationship with Ukraine and Russia determines how exposed to the war shock each country is. Second, the household decisions in terms of consumption, land, and labor allocation choices determine how exposed each household is to the war-induced price changes. This is the “household choices channel.” These mechanisms are intertwined, and jointly determine the impact of the war on a given household. The household choices channel is the main driver of the vastly heterogeneous welfare losses created by the war. The best way to see this is to split the welfare losses into its main constituent parts. The portfolio of income generating activities a household undertakes determines how changes in producer and input (fertilizer) prices impact its income. The pattern of household expenditures, in turn, determines how changes in consumer prices impact the household cost of living. We explore these mechanisms in Figure 4. For each household, we plot the cost of living effect (panel a) and the overall income effect from both land and labor (panel b). The heterogeneity in both the consumption and income effects is evident. The consumption effects, however, dominate the income effects. The war increases consumer prices which result in welfare losses for all households in all countries (panel a). The average losses (given by the local polynomial regression—the blue line) range between 2% and 4% and are slightly increasing with income for most of the world income spectrum. However, the average losses dip lower for upper-middle income households in relatively rich countries. In fact, the consumption losses for the poorest and richest households are comparable. The average consumption loss across all households is –3.05% (see Table 3). The pattern of income effects is similar, but the magnitudes are smaller. Unlike the cost of living effect, the 19 This single household is characterized by the country-level aggregate land and labor allocation shares as well as by the budget shares of the median household in HIT. 29 income effect can be positive or negative (depending on the changes in producer and fertilizer prices). The average income gain is positive, very small, and flat across most of the income distribution. The exceptions are once again the upper middle income countries at the right tail, where households suffer sizeable losses. The average income gain across all households is a mere 0.14%.20 We can further examine these mechanisms by turning to the trade channel. Figure 5 depicts the welfare impacts against the pre-war exposure measured by the share of imports coming from Ukraine and Russia. We plot the average welfare effect for each country against exposure with a blue dot; we also plot the distribution of estimated welfare impacts (the kernel densities) against exposure. The fitted line between the average welfare effects and exposure is negatively sloped, indicating that the average losses are larger in countries that imported more from Ukraine and Russia before the war started. These are the countries with the highest per capita income in our sample (e.g., Azerbaijan, Georgia, Armenia and Moldova), which, because of the trade mechanism, suffer the largest welfare losses. Inspection of the relationship between the kernel densities and trade exposure reveals that the countries with the largest average losses also have a more unequal distribution of such losses. This can be seen by noticing that the density plots of the distribution of the gains widen with reliance on imports from the countries at war. The trade mechanism thus plays a role in increasing within-country inequality as well. As an alternative way of showing this, we plot in Figure 5 fitted lines of the average impact of the top 25% richest (the long-dashed dark blue line) and bottom 25% poorest (the short-dashed light blue line) households against exposure. The line reflecting losses for the bottom 25% is always below the one reflecting losses for the top 25% and, importantly, more steeply downwards sloping; the war tends to widen income disparities. Greater reliance on imports from the two countries at war is not only associated with reduced incomes but also with exacerbated inequality. The trade mechanism primarily affects households through the consumption channel rather than the income channel. This is shown in Figure 6, which displays the distribution of the consumption losses (panel a) and of the income effects (panel b) at different levels of trade exposure to Ukraine and Russia. Both average consumption and income effects intensify with higher exposure to these countries, 20 We explored this in more detail by running two additional simulations in which respectively only trade in agricultural goods and only trade in fertilizers are banned. The results are reported in section A6 in the Supplementary Material. In the former case, labor and land income effects are positive; higher profits for farmers partially offset the deleterious effects of higher food prices. By contrast, in a scenario in which only trade in fertilizers is restricted the land and labor effects are even more negative. The negative impacts on land and labor income are thus due to the increase in the cost of fertilizers. Households also partially mitigate the shock by changing their income earning portfolio, choosing different crops and different wage jobs, but this type of adjustment is quantitatively limited in the case of Ukraine shock since on only 0.55% of households change their crop choice (i.e. land income) and 0.72% their wage income. 30 Figure 4: Impact of the war by initial income All countries (a) Consumption channel 2 0 Iraq -2 Bhutan Nepal Pakistan consumption effect Moldova -4 Azerbaijan Mongolia Georgia -10 -8 -6 -12 -14 2 4 6 8 10 12 log per capita expenditure (b) Income channel Nepal Pakistan 2 Bhutan Iraq 0 Georgia -2 Moldova income effect -6 -4 Mongolia Azerbaijan -8 -10 -12 -14 2 4 6 8 10 12 log per capita expenditure Notes : The graph plots the estimated welfare impacts of the war via the consumption channel (cost of living effect), panel a), and the income channel (sales effect net of input cost), panel b), expressed in real income gains, across all countries in our sample against pre-war log per capita expenditure. Each observation (denoted by a “+”) represents an income percentile in a different country. Each country is demarcated by a different color—with lighter colors denoting higher average gains. The blue line is a fitted polynomial regression. 31 Figure 5: Impacts by reliance on imports from Ukraine and Russia 0 -5 welfare gain -10 -15 0 5 10 15 20 share of imports coming from Russia and Ukraine Notes : The graph plots the estimated distribution of welfare impacts (real income changes) of the war (via its impacts on agricultural trade), for all 51 countries, against exposure (proxied by the pre-war share of a country’s total imports coming from Ukraine and Russia). Blue dots denote average impacts by country. The blue line is the fitted line of these averages; the long-dashed dark blue line is the fitted lines of the average impact of the top 25% richest households and the short-dashed light blue line is the corresponding fitted line for the bottom 25% poorest households. 32 as expected. As we showed above, more exposed households are hurt the most and this also shows up in the underlying operating mechanisms. While the dispersion of the consumption channel widens with trade exposure, the dispersion of the income effect does not. The consumption effect is thus driving both the severity of the average welfare losses and the decrease in across-country inequality. Finally, this relationship is also evident in Figure 7, which depicts average welfare, consumption and income effects against the household’s rank in their own country’s income distribution. The plot confirms that the income effect (black long-dash line) is much smaller than the consumption effect (orange small-dash line) across all income strata. In fact, the average welfare effect (purple solid line) and the average consumption effects are close together. Furthermore, poor households are hurt more by the increase in the cost of living than richer households. For instance, while the cost of living reduces the average real income by 3.05%, the consumption losses for the bottom 25% poorest households are -3.29% and the losses for richest quartile are –2.87% instead. The income loss is on average only 0.14% and appears to be distributionally neutral. This confirms that the average losses and the within-country disequalizing impacts are mostly driven by the consumption impacts. 5 Agriculture and Climate Change The shock generated by the Russian Federation’s invasion of Ukraine on third countries is structurally different from the climate change shock. Climate change is a productivity shock: changes in temperature, precipitation, wind patterns, humidity, soil degradation and other environmental factors influence crop yields across the globe. In contrast to the war shock, which we modeled as aggregate quantitative trade restrictions, climate change is a direct technological shock affecting output. This distinction allows us to delve into additional dimensions of the distributional impacts of agricultural shocks. In particular, we explore in more depth the mechanics of the income generation process of agrarian households, and we investigate adaptation and adjustment to climate change. To study the impact of climate change, we follow Costinot, Donaldson, and Smith (2016) and use GAEZ data to quantify the implied productivity shock of different crops. Specifically, we focus on the main scenario adopted by the UN’s Intergovernmental Panel on Climate Change (IPCC), the FAO GAEZ Hadley CM3 A1 model. This model predicts a future world of accelerated economic growth and also accounts for the introduction of more efficient fossil technologies. We feed these GAEZ productivity projections in our model to explore household behaviour and the attendant welfare, poverty and inequality effects. 33 Figure 6: Consumption and Income Impacts by reliance on imports from Ukraine and Russia (a) Consumption channel 0 -5 consumption effect -10 -15 0 5 10 15 20 share of imports coming from Russia and Ukraine (b) Income channel 2 0 income effect -4 -2 -6 -8 0 5 10 15 20 share of imports coming from Russia and Ukraine Notes : The graphs plot the estimated distribution of the consumption effects (panel a) and income effects (panel b) of the war (via its impacts on agricultural trade), for all 51 countries, against exposure (proxied by the pre-war share of a country’s total imports coming from Ukraine and Russia). Blue dots denote average impacts by country. The blue line is the fitted line of these averages; the long-dashed dark blue line is the fitted lines of the average impact of the top 25% richest households and the short-dashed light blue line is the corresponding fitted line for the bottom 25% poorest households. 34 Figure 7: Impact of the war by initial income rank Welfare, income and consumption effects 0 -1 welfare change -2-3 -4 0 20 40 60 80 100 income rank welfare income consumption Notes : The graph presents the percentage change in welfare, measured as real income, associated with the war (via its impacts on agricultural and fertilizer trade) against a household’s rank in the initial per capita income distribution, with rank 100 denoting the richest percentile and rank 1 denoting the poorest percentile. Our experiments complement Costinot, Donaldson, and Smith (2016) in two ways. First, we quantify heterogeneity in the impact of climate change within countries, as well as across countries (as they do). Second, their focus is on 50 countries which are the top agricultural producers and account for almost 90% of global agricultural output. Our paper focuses instead on the implications of climate change for low-income and potentially more vulnerable countries.21 5.1 Results Climate change impacts a much broader set of products than the war shock and has very sizable direct productivity effects across countries (Table 4), whereas the war shock only indirectly affects third countries. The majority of countries in our sample, 40 of 51, is projected to have lower yields. Productivity changes range from a minimum of –61.97% in Bolivia to a maximum of 112.03% in Mongolia. Across countries the median change in productivity is –30.93% and the average is –19.60%. The collage of household-level welfare effects of these climate change productivity shocks is presented in Figure 8, which displays “ridgeline” estimates of the distribution of gains by country. The 21 Only 18 of the countries in our sample are included in the analysis of Costinot, Donaldson, and Smith (2016) 35 country-specific impacts are also presented in Table 4. Overall, the average real income gains are negative in 37 of 51 countries in our sample. As in the war case, there are widespread average losses (accruing to 72.8% of households) but, unlike the war, there are many more instances of gains (accruing to 27.2% of households). On average across all countries, households in our sample see their incomes decline by 11.99% but there is very large standard deviation of 22.67%. Countries located close to ote d’Ivoire, the equator, where average temperatures are already high, such as Guinea Bissau, Cˆ the Gambia, Central African Republic, Bolivia, Nigeria, Mozambique, Benin and Papua New Guinea experience average losses exceeding 30%. By contrast, average gains in Kyrgyzstan, Kenya, Burundi, Mongolia and Rwanda exceed 10%. Within countries, gains vary a lot. To illustrate the wide disparities, note the cases of Guinea-Bissau and the Gambia where all household lose from climate change and the losses range from –92.06% to –50.24% and –75.61% to –20.45% respectively. In Bangladesh, there are also widespread losses, but they are less dispersed, ranging from –27.08% to –18.35% of real household income. Conversely, in Mongolia, all households benefit, with real income gains ranging between 33.40% and 84.11%. Similarly, in Kyrgyzstan, everyone gains, though with a narrower range of welfare effects (from 13.30% to 27.86%). In most countries, however, there are both winners and losers from climate change. Within-country heterogeneity in the impact of climate change is thus of first order importance. There are also biases in single-household models in the case of climate change (see column 4 of Table 4). Solving the model with only one household in each country, the average worldwide loss would be –9.12%, which is 23.96% smaller than our estimate (–11.99% in column 1). As expected, the magnitude of the bias varies across countries, with some countries showing sizable discrepancies, such as Mozambique and the Central African Republic. In Figure 9, we assess the inequality impacts of climate change. Panel a) presents a scatter plot of the corresponding welfare impacts against present day (log) per capita expenditure pooling all countries in our sample (each demarcated with a different color). The kernel regression is negative and subtly upward sloping across most of the developing world income distribution, except at the very extreme of the left tail. At this tail, the kernel averages are in fact positive but this is because of one country, Burundi, which displays large gains for the poorest households in our world sample. Leaving aside Burundi, poor households tend to lose close to 20% of their incomes, while the average impacts for the richest households at the right tail are closer to 10%. Panel b), which presents the non-parametric averages across bins of the per capita expenditure spectrum, shows the same patterns within countries. On average, households at lower percentiles of each country’s well-being lose more than household at 36 Figure 8: Distribution of the effect of climate change By country Rwanda Mongolia Burundi Kenya Kyrgyzstan Tajikistan Armenia Comoros Georgia Yemen Nepal Uganda Egypt, Arab Rep. Uzbekistan Jordan Bhutan Moldova Azerbaijan Iraq Mauritania South Africa Madagascar Tanzania Pakistan Ecuador Liberia Cameroon Zambia Malawi Cambodia Bangladesh Burkina Faso Sierra Leone Indonesia Guatemala Nicaragua Vietnam Ghana Togo Niger Sri Lanka Guinea Papua New Guinea Benin Mozambique Nigeria Bolivia Central African Republic Gambia Cote d'Ivoire Guinea-Bissau -100 -50 0 50 100 income gain Notes : The graph presents kernel density plots by country, sorting countries in terms of their average real household income gains, with countries with the highest average gains, such as Mongolia and Rwanda at the top, and those with the lowest gains, such as the Gambia and Guinea-Bissau at the bottom. Darker shades denote greater average losses. 37 Table 4: Impacts of climate change ∆Welfare ∆Income ∆CL Exposure Average Bottom Top Single Total Labor Land HH. Yield 25% 25% HH. Panel A: All countries (pooled) Average -11.99 -14.63 -9.93 -9.12 -7.26 -6.09 -8.11 -4.42 -19.60 Pop w. aver. -16.03 -18.75 -13.89 -12.03 -9.43 -7.25 -13.40 -6.91 -27.15 Std. dev. 22.67 25.36 20.84 20.34 24.61 21.00 28.44 5.25 36.94 Minimum -68.43 -76.66 -65.10 -66.68 -70.48 -66.84 -72.76 -16.34 -61.97 Median -11.19 -13.50 -10.51 -9.23 -6.48 -5.29 -8.74 -4.88 -30.93 Maximum 52.92 54.32 54.77 48.39 61.46 54.27 72.01 7.82 112.03 Panel B: By country Guinea-Bissau -68.43 -76.66 -65.10 -66.68 -70.48 -66.84 -72.76 7.82 -42.03 Cˆote d’Ivoire -43.30 -57.29 -33.25 -34.67 -45.19 -34.20 -48.25 4.18 -44.09 Gambia, The -42.24 -56.76 -31.53 -35.63 -44.44 -44.13 -44.57 3.89 -39.11 Cent. Afr. Rep. -39.50 -43.27 -34.65 -35.50 -35.46 -32.43 -35.49 -5.70 -45.40 Bolivia -39.07 -50.38 -32.77 -30.91 -31.61 -24.32 -37.27 -7.83 -61.97 Nigeria -37.18 -40.15 -34.32 -32.99 -27.43 -22.81 -32.46 -12.53 -48.67 Mozambique -37.12 -44.20 -26.79 -27.97 -30.18 -17.61 -34.75 -9.55 -56.04 Benin -36.29 -37.00 -34.39 -34.06 -35.77 -31.17 -38.38 -0.98 -44.66 Papua New G. -32.71 -37.90 -28.30 -27.24 -28.20 -19.38 -32.19 -5.88 -44.11 Guinea -28.60 -33.66 -23.54 -24.52 -24.40 -19.07 -25.04 -4.95 -42.60 Sri Lanka -27.84 -31.56 -23.59 -20.46 -23.16 -17.44 -31.90 -5.18 -45.66 Niger -27.76 -29.59 -28.34 -26.08 -25.72 -21.65 -26.86 -2.35 -27.87 Togo -25.80 -36.32 -18.83 -20.43 -25.79 -27.95 -24.71 0.04 -47.37 Ghana -25.58 -31.35 -21.69 -21.50 -28.99 -25.53 -45.41 5.38 -46.34 Viet Nam -24.92 -29.70 -20.81 -16.98 -14.09 -9.44 -17.79 -11.95 -49.60 Nicaragua -24.34 -31.22 -18.75 -16.75 -17.28 -14.69 -20.49 -6.98 -41.90 Guatemala -24.27 -28.36 -21.34 -19.39 -18.56 -17.63 -19.85 -6.23 -42.14 Indonesia -22.89 -27.42 -18.88 -14.89 -11.99 -9.81 -25.58 -11.95 -32.16 Sierra Leone -22.76 -25.45 -18.87 -19.59 -23.12 -17.81 -24.07 0.34 -48.72 Burkina Faso -21.61 -20.16 -21.79 -17.97 -16.77 -15.19 -17.32 -5.96 -43.96 Bangladesh -21.55 -23.49 -20.11 -18.92 -17.48 -12.65 -20.57 -4.77 -54.96 Notes : This table presents the results of a simulation of the impacts of climate change based on the FAO GAEZ Hadley CM3 A1 model. Welfare changes and changes in income are expressed as percentage changes in real household income relative to the status quo. Statistics in Panel A refer to average welfare impacts across countries. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure refers to the change in agricultural productivity (i.e. crop yields) relative to the status quo. Countries are ordered in terms of average real income gains (from lowest to highest). 38 Table 4: Impact of Climate Change (continued) ∆Welfare ∆Income ∆CL Exposure Average Bottom Top Single Total Labor Land HH. Yield 25% 25% HH. Cambodia -21.46 -23.47 -19.80 -15.96 -5.93 -3.92 -7.31 -16.34 -57.45 Malawi -17.87 -17.60 -17.75 -14.15 -11.23 -8.95 -12.54 -7.58 -42.39 Zambia -14.04 -14.68 -13.06 -10.91 -6.48 -4.77 -8.74 -8.04 -32.41 Cameroon -12.18 -13.50 -10.92 -9.23 -5.46 -5.29 -5.57 -7.07 -30.93 Liberia -11.19 -12.11 -10.51 -10.08 -14.20 -12.54 -15.04 3.30 -45.75 Ecuador -10.97 -14.10 -8.86 -7.02 -7.01 -5.82 -8.95 -4.14 -36.38 Pakistan -10.37 -12.83 -8.95 -7.28 -7.22 -6.57 -8.62 -3.18 2.21 Tanzania -9.97 -9.72 -9.60 -5.09 -3.04 -2.21 -4.01 -7.17 -30.49 Madagascar -9.59 -10.42 -9.01 -5.82 -0.71 -0.65 -0.75 -8.93 -25.55 South Africa -6.93 -11.33 -4.37 -2.69 -1.74 -1.62 -9.52 -5.28 -19.83 Mauritania -6.90 -10.38 -4.37 -3.79 -3.03 -1.68 -3.19 -3.97 -18.71 Iraq -6.12 -8.49 -4.49 -4.87 -4.44 -3.09 -7.53 -1.68 -14.66 Azerbaijan -3.73 -5.88 -3.19 -2.35 -0.67 -0.28 -0.94 -3.08 -24.74 Moldova -1.59 -1.79 -1.39 -2.50 -2.83 -1.85 -5.72 1.29 -15.95 Bhutan -1.38 -1.59 -1.28 0.31 0.83 0.38 1.74 -2.20 -7.69 Jordan -1.28 -2.16 -0.34 -0.47 3.74 3.08 6.00 -4.88 7.20 Uzbekistan 0.71 1.37 0.55 0.14 -1.37 0.24 -3.08 2.09 1.99 Egypt, Arab Rep. 0.81 0.97 0.70 0.49 -0.48 -0.38 -0.61 1.30 0.00 Uganda 1.91 2.93 0.82 5.41 13.46 12.90 13.85 -10.08 -11.02 Nepal 1.97 2.25 2.21 5.63 13.09 13.15 12.87 -9.79 -24.38 Yemen, Rep. 3.58 4.89 2.97 3.47 9.34 8.15 11.02 -5.25 0.00 Georgia 3.98 3.91 3.99 4.13 6.38 5.29 8.17 -2.27 1.40 Comoros 5.31 7.19 4.03 4.67 9.19 8.40 9.75 -3.46 -8.11 Armenia 5.48 6.33 4.32 5.23 8.51 7.01 11.81 -2.68 72.36 Tajikistan 8.26 8.88 7.75 8.21 9.22 9.11 10.00 -0.85 11.83 Kyrgyzstan 18.97 20.85 17.32 18.32 21.77 17.17 29.22 -2.04 34.31 Kenya 28.01 25.37 25.54 25.70 49.58 45.26 56.27 -14.15 13.88 Burundi 28.29 31.99 23.53 24.81 49.32 34.48 50.39 -14.12 65.69 Mongolia 47.54 44.48 54.77 45.54 45.91 31.85 72.01 1.36 112.03 Rwanda 52.92 54.32 50.51 48.39 61.46 54.27 71.14 -5.52 73.25 Notes : This table is a continuation from the table on the previous page. It presents the results of a simulation of the impacts of climate change based on the FAO GAEZ Hadley CM3 A1 model. Welfare changes and changes in income are expressed as percentage changes in real household income relative to the status quo. Statistics in Panel A refer to average welfare impacts across countries. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure refers to the change in agricultural productivity (i.e. crop yields) relative to the status quo. Countries are ordered in terms of average real income gains (from lowest to highest). 39 the upper percentiles. For example, the losses for the bottom 25% are –14.63%, while the losses for the top 25% are –9.93%. Climate change exacerbates inequality. 5.2 Mechanisms The mechanics of the impacts bear resemblances and differences with those of the war discussed in section 4. The productivity shock of climate change interacts with the household choices to determine the impacts. Similar to the war scenario, the patterns of household consumption and production shape the heterogeneity in impacts. But there is a major difference: in the climate change scenario, the income channel is large and in fact significantly larger than the consumption channel. On average, while the cost of living effect across all countries is –4.42%, the income loss is –7.26%, about 64.25% larger. This is because climate change operates as a negative productivity shock to yields in most countries, resulting in direct output declines and direct losses in income from land and labor. Another key difference is that, unlike the war scenario, the magnitudes of the land and labor income effects differ substantially, with the losses on the return to land (–8.11% on average) being bigger than the losses on the return to labor (–6.09% on average). To explore how these effects vary across the income distribution, we plot the non-parametric averages of the three mechanisms against initial within-country income rank in Figure 10. The solid red line is the cost of living effect; the thick color-graded line is the land income effect, and the thin color-graded line is the labor income effect. Two results emerge, both pointing towards a heavier burden of climate change on the poor. First, the land income losses are bigger than the labor income losses and both labor and land income losses are more substantial for poorer households. The land income loss is roughly 10% for the poorest households, while the labor income loss is about 7% instead. However, the magnitudes of both income losses tend to converge for the richest households. Second, due to higher consumer prices, the cost of living effect is negative and, due to non-homotheticity, this is especially so for the poorest households. As a result, the losses for the poor are much bigger than the losses for the rich. These results illustrate the disproportionate burden of climate change on poorer households. The model offers insights as to why all this happens. The higher cost of living effect on the poor occurs because of non-homothetic preferences combined with climate change induced price increases in food products. Climate change increases the cost of the subsistence bundle and, conditional on income, this subsistence bundle accounts for a bigger share of expenditures for the poorest households. The different patterns of labor and land income losses are driven by a combination of factors. Land is 40 Figure 9: Impacts of climate change by income (a) Across countries 80 60 40 20 Mongolia welfare gain Nepal Georgia Bhutan Azerbaijan Moldova -20 0 Iraq Pakistan -40 -60 -80 2 4 6 8 10 12 log per capita expenditure (b) Within countries -8 -10 -12 welfare gain -14 -16 -18 0 20 40 60 80 100 income rank Notes : Panel a) depicts the kernel density distribution of the estimated welfare impacts of climate change, expressed in real income gains, across all countries in our sample against current log per capita household expenditure. Each observation (denoted by a “+”) represents an income percentile in a different country. Each country is demarcated by a different color—with lighter colors denoting higher average gains. The blue line is a fitted polynomial regression. Panel b) depicts a local polynomial fitted line of the percentage change in real income due to climate change against a household’s rank in the initial per capital income distribution, with rank 100 denoting the richest percentile and rank 1 denoting the poorest percentile. The shaded area depicts the 95% confidence interval. Darker colors depict larger average losses. 41 Figure 10: Impact of climate change by initial income rank Consumption and income effects -2 -4 cost of living welfare gain -6 labor income land income -8 -10 0 20 40 60 80 100 income rank Notes : The graph presents the percentage change in welfare, measured as real income, associated with climate change against a household’s rank in the initial per capita income distribution, with rank 100 denoting the richest percentile and rank 1 denoting the poorest percentile. The solid red line represents the cost of living effects; thin color-graded line is the labor income effect and the thick color-graded line is the land effect. used for crop production, while labor can be allocated to crops, manufacturing, and non-traded goods. Climate change directly impacts crop production but not manufacturing and non-traded goods so that jobs in these sectors offer some protection for labor against these shocks. Moreover, the labor of the poor tends to be more specialized in agriculture than in manufacturing or services, so that this buffer effect is ameliorated for them. Furthermore, the land elasticity is much lower than the labor elasticity, meaning that households can adjust labor more easily than they can adjust land. As a consequence land is more specific and thus more severely affected than labor. In turn, the land income of the poor is affected more because they rely more on agriculture and because, within agriculture, they tend to specialize in crops more severely affected by climate change. 5.3 Adjustment and Adaptation Since the productivity shock of climate change is large, it offers a good opportunity to examine the role of adjustment and adaptation. We study four counterfactual scenarios. First, we keep the budget shares fixed, thus cancelling the role of non-homothetic preferences that cause expenditure shares to 42 Figure 11: Impact of climate change by initial income rank Adjustment scenarios -5 homothetic preferences -10 baseline welfare gain varying elasticities no factor adjustment -15 -20 no trade adjustment -25 0 20 40 60 80 100 income rank Notes : The graph presents the percentage change in welfare, measured as real income, associated with climate change against a household’s rank in the initial per capita income distribution, with rank 100 denoting the richest percentile and rank 1 denoting the poorest percentile. The color-graded line is the baseline effects of climate change. The solid black line keeps budget shares constant (shutting down non-hometheticity adjustments). The long-dash black line allows for the different land and labor elasticities for the poor (bottom 50%) and the rich (top 50%). The short-dashed black line does not allow for land and labor adjustment at the household level. The dotted black line represents a model where trade (imports and exports) is not allowed to respond. vary with household income. Second, we consider a fixed factors model in which households are not allowed to adjust their land and labor allocations. Third, we shut down the trade mechanism, keeping trade volumes fixed. Finally, we explore a model in which the land and labor elasticities are different for the rich and the poor. To implement this, we use the same procedure as in section 3.3 to estimate two land and two labor elasticities, for the bottom and top 50% of households based on income. The land elasticities are 1.81 for the poor and 2.24 for the rich (compared to 2.15 in the baseline). The labor elasticities are 1.58 for the poor and 1.34 for the rich (compared to 1.34 in the baseline). The results are presented in Figure 11. We summarize our findings with a plot of the average welfare effect across countries at different income ranks. The main conclusion from this analysis is that, in all scenarios, the average welfare effects of climate change remain negative across all income strata. In addition, the losses are bigger for the poor than for the rich, as in the baseline. Consequently, these various sources of adjustment and adaptation do not fully offset the losses caused by climate change and do not suffice to protect the poor from suffering higher losses than the rich. 43 There are nevertheless some interesting differences with the baseline scenario to highlight. First, assuming fixed budget shares leads to lower real income losses (solid black line). This is because the climate change shock is typically negative making households poorer. As households see their incomes decline, they adjust budget shares towards food and agricultural products that see higher prices due to the negative productivity shock. This leads to higher losses in the baseline, compared to the fixed budget share case. This is especially so for the poorest households. Non-homothetic preferences thus amplify the impact of climate change. Similarly, allowing for factor adjustment of varying intensity for the poor and the rich (i.e., varying land and labor elasticities) leads to slightly lower losses than in the baseline scenario of common elasticities (long-dash black line). This is because the heterogeneity in the factor elasticities allows for more heterogeneous adjustments for rich and poor households, who are thus not restricted as in the baseline. The differences are, however, small. The fixed factors counterfactual yields much bigger losses for all household income bins (short-dash black line). If land and labor cannot be adjusted, the land and labor returns absorb all the price shock while factor reallocation instead smooth this out. As a result, the losses are much larger and can represent an additional reduction of real income of more than 2.79% on average. These additional losses are similar across income bins (i.e., for the poor and the rich). Cancelling the trade channel (i.e., keeping trade flows fixed) creates the largest losses among our experiments (dotted black line). International trade helps in averaging the shocks across crops and countries. The impacts are large, with the additional losses being 7.78% on average. Trade is particularly beneficial for rich households, which would suffer the largest additional income reduction if trade flows were not allowed to adjust. 6 Conclusion Agriculture is central to development but inherently subject to shocks, which is being aggravated by conflict and climate change. Examining how agricultural shocks reverberate through the income distribution is challenging because agricultural products are traded globally and because households in different segments of the income distribution are impacted in multiple ways—as consumers, producers, and laborers—and have vastly different income and consumption patterns, which they can adjust. This paper has introduced a discrete choice general equilibrium trade model of heterogeneous households making consumption, land, and labor allocation choices, designed to leverage household 44 survey data. The model is distinctive in being able to generate both highly granular, income-percentile specific, and highly comprehensive assessments of the impact of agricultural shocks. It accounts for changes in land income, labor income and the cost of consumption, and accommodates factor adjustment. A striking result from our simulations is the overwhelming heterogeneity in the welfare gains and losses, both across and within countries, and for both the war and the climate change shocks. This heterogeneity cannot be accurately captured by representative household models. The two agricultural shocks tend to create losses that are larger for more vulnerable families, thus increasing inequality across the developing world. In the case of the war, almost all households (99.6%) lose real income with a worldwide average loss of –2.90% for third countries. The losses for the bottom income quartile are bigger, –3.11%, while the losses for the top income quartile are smaller, –2.78%. The countries most impacted are those with higher pre-war reliance on imports from Ukraine and Russia. The primary driving mechanism is the cost of living effect generated by the food price hikes induced by the war. This factor explains the prevalence of average losses and the bigger losses for the poor. Climate change was shown to have even more deleterious and more dispersed effects on the income distribution. While 37 countries lost average real income, the remainder 14 countries gained. The average change in real income due to climate change is –11.99%. The losses for the bottom 25% and the top 25% are –14.63% and –9.93%, respectively, implying that climate change exacerbates inequality, too. These welfare effects are strongly correlated with the projected productivity effect of climate change on agriculture. Unlike the war case, the primary drivers are the land and labor income effects (rather than the cost of living effect). The adjustment mechanisms we studied offer some relief but neither fully offset losses nor protect the poor from greater harm than the rich. References Adao, R., P. Carrillo, A. Costinot, D. Donaldson, and D. Pomeranz (2022): “Imports, exports, and earnings inequality: Measures of exposure and estimates of incidence,” The Quarterly Journal of Economics, 137(3). Agri-Pulse (2022): “Russian halt to fertilizer exports expected to boost prices, spur shortfalls,” Accessed: 2023-1-13. Artuc, E., S. Chaudhuri, and J. McLaren (2010): “Trade shocks and labor adjustment: A structural empirical approach,” American Economic Review, 100(3). 45 Artuc, E., N. Gomez-Parra, and H. Onder (2022): “The True Cost of War: The Impact of Conflict in Eastern Ukraine,” World Bank Policy Working Paper Series. Artuc, E., G. Porto, and B. Rijkers (2020): “Household Impacts of Tariffs: Data and Results from Agricultural Trade Protection,” The World Bank Economic Review, 35(3). Bergquist, L. F., B. Faber, T. Fally, M. Hoelzlein, E. Miguel, and A. Rodr´ ıguez-Clare (2022): “Scaling agricultural policy interventions,” Discussion paper, NBER. Borchert, I., M. Larch, S. Shikher, and Y. V. Yotov (2021): “The International Trade and Production Database for Estimation (ITPD-E),” International Economics, 166. (2022): “The International Trade and Production Database for Estimation - Release 2 (ITPD-E R02),” USITC Working Paper 2022–07–A. CNBC (2022): “India isn’t the only one banning food exports, these countries are doing the same.,” Accessed: 2023-1-13. Costinot, A., and D. Donaldson (2016): “How Large Are the Gains from Economic Integration? Theory and Evidence from U.S. Agriculture, 1880-1997,” mimeo, MIT. Costinot, A., D. Donaldson, and C. Smith (2016): “Evolving Comparative Advantage and the Impact of Climate Change in Agricultural Markets: Evidence from 1.7 Million Fields around the World,” Journal of Political Economy, 124(1). Costinot, A., A. Rodr´ ıguez-Clare, and I. Werning (2020): “Micro to Macro: Optimal Trade Policy with Firm Heterogeneity,” Econometrica, 88(6). Cruz, J.-L., and E. Rossi-Hansberg (2023): “The Economic Geography of Global Warming,” Review of Economic Studies, forthcoming. Deaton, A. (1997): The Analysis of Household Surveys: A Microeconometric Approach to Development Policy. Washington D.C.: World Bank. Dekle, R., J. Eaton, and S. Kortum (2008): “Global Rebalancing with Gravity: Measuring the Burden of Adjustment,” IMF Staff Papers, 55(3). Desmet, K., and E. Rossi-Hansberg (2015): “On the spatial economic impact of global warming,” Journal of Urban Economics, 88. Fajgelbaum, P. D., and A. K. Khandelwal (2016): “Measuring the unequal gains from trade,” The Quarterly Journal of Economics, 131(3). Fajgelbaum, P. D., and S. J. Redding (2014): “External integration, structural transformation and economic development: Evidence from argentina 1870-1914,” CEPR Discussion Paper No. DP10026. FAO and IIASA (2021): “Global Agro Ecological Zones,” Accessed: 2023-1-13. 46 Galle, S., A. Rodriguez-Clare, and M. Yi (2021): “Slicing the Pie, Quantifying the Aggregate and Distributional Effects of Trade,” NBER Working Paper 23737. Galle, S., A. Rodr´ ıguez-Clare, and M. Yi (2023): “Slicing the pie: Quantifying the aggregate and distributional effects of trade,” The Review of Economic Studies, 90(1). Lagakos, D., and M. E. Waugh (2013): “Selection, Agriculture, and Cross-Country Productivity Differences,” American Economic Review, 103(2). Lee, E. (2020): “Trade, inequality, and the endogenous sorting of heterogeneous workers,” Journal of International Economics, 125. Lee, E., and K.-M. Yi (2018): “Global Value Chains and Inequality with Endogenous Labor Supply,” Journal of International Economics, 115. Melitz, M. J., and S. J. Redding (2015): “Heterogeneous Firms and Trade,” Handbook of International Economics, 4. Nigai, S. (2016): “On Measuring the Welfare Gains from Trade Under Consumer Heterogeneity,” The Economic Journal, 126(593). NPR (2022): “Russia has blocked 20 million tons of grain from being exported from Ukraine,” Accessed: 2023-1-13. Reuters (2022): “Russia temporarily bans grain exports to ex-Soviet countries,” Accessed: 2023-1-13. Simonovska, I., and M. E. Waugh (2014): “The Elasticity of Trade: Estimates and Evidence,” Journal of International Economics, 92(1). Sotelo, S. (2020): “Domestic Trade Frictions and Agriculture,” Journal of Political Economy, 128(7). The Wall Street Journal (2022): “Russian Ministry Recommends Suspending Fertilizer Exports.,” Accessed: 2023-1-13. Tol, R. S. J. (2009): “The economic effects of climate change,” Journal of Economic Perspectives, 23(2). Tombe, T. (2015): “The missing food problem: Trade, agriculture, and international productivity differences,” American Economic Journal: Macroeconomics, 7(3). 47 Supplemental Material for “Crops, Conflict and Climate Change” E. Artuc, G. Porto and B. Rijkers A1 Solution Method The model is solved in changes using hat-algebra following Dekle, Eaton, and Kortum (2008) which n,h n,h utilizes factor allocation shares (i.e. πT,j and πL,j ) instead of productivity parameters (i.e. An,h T,j and An,h T,j ). We omit manufacturing and services sectors here to make the exposition simpler. Note that we perform the analysis using households grouped by their income as the units of observation, as opposed to plots. This prevents us from directly using the underlying plot-level productivity data of GAEZ in the solution. From Table 1 in the main text, the main inputs to solve the model are the initial shares and a set of key parameters. In particular, we need the labor and income allocation shares, π n,h T,j and π n,h n,h L,j ; the household h’s labor supply share in total effective labor allocated to j , ηL,j ; the household n,h h’s land supply share in total effective land allocated to j in country n, ηT,j ; and the shares of land ιn,h and labor and income in household h’s total income, ¯ ιn,h T and ¯L . With respect to trade, we need the share of exports of m in n’s imports, xn,m j , and the share of imports of n in m’s exports, xn,m j . To n,h characterize demand, we need the share of household h in total demand for j in country n, Dj . The set of calibrated or estimated parameters are: the Cobb-Douglas labor, land, and intermediate n and κn ; the labor inputs production shares, βL,j , βT,j , and βF,j ; the utility function coefficients, νj j and land elasticities (Frechet shape), θL and θT ; and the implicit trade elasticity (negative), 1 − σ . We introduce two different shocks in this system. In the case of the war, we impose export bans on selected agricultural products from Ukraine and Russia. We model this as follows. In the case of Ukraine, we shut down imports and exports of all agricultural products and fertilizer. That is, we impose prohibitively high τH trade costs on imports U,m τj = τH , ∀j, F &∀m, where U stand for Ukraine. We also set prohibitively high trade costs on Ukrainian exports n,U τj = τH , ∀j, F &∀n. In addition, we impose bans on Russian exports of wheat, rice, corn, other cereals, sugar, other oilseeds, and fertilizers to the world. That is, we set n,R τj = τH , ∀n, 1 where R stands for Russian and j is wheat, rice, corn, other cereals, sugar, other oilseeds, and fertilizers. To see how we solve the model with hat-algebra, let x denote the proportional change in a variable n , rental rates r n and x. We proceed as follows. First, start with a guess for the change in wages wj j input prices at origin pn F. With these guesses, we calculate the change in input price index faced by producers 1 1−σ 1−σ n PF = xn,m F pm n,m M τM ; m the change in consumer prices βF βT βL pn n j = PM n rj n wj ; and the change in price index faced by consumers 1 1−σ 1−σ Pjn = xn,m j pm n,m j τj . m This allows to calculate the change in trade (import shares) for consumed goods and intermediate input 1−σ n,m pm j τj xn,m j = 1−σ . Pjn These trade shares plunge to zero for the shocked products and countries. In the next step, we calculate the change in household labor supply, Ln,h j , via (17), and we calculate Ln j at national level by adding them across households   1 θL θL Φn,h L =  π n,h n L,j wj  ; j (wjn )θL Φn,h Ln,h j = θ L n ; n,h L wj ΦL Ln j = Ln,h n,h j ηL,j , h n,h n,h Ln,h j where ηL,j is the share of household h in total labor income of sector j , ηL,j = ′ . h′ Ln,h j Similarly, we calculate household land supply, Tjn,h , and then calculate Tj at national level by adding them up across households   1 θT Φn,h T =  π n,h T,j (rj ) θT  ; j 2 n )θT (rj Φn,h Tjn,h = θT T n ; rj Φn,h T Tjn = n,h Tjn,h ηT,j , h n,h n,h n,h Tj where ηT,j is the share of household h in total return on land in sector k , ηT,j = h ′ h′ Tj This permits the calculation of the change in household income E n,h = Φn,h L ¯ ιn,h n,h n,h L + ΦT ¯ιT . The change in expenditure shares is n 1− γn γn κnj Pjn νj E n,h E n,h + E n,h E n,h n,h αj = κn , n 1− γn j νj E n,h + E n,h where γ n = j κn Pjn Pjn / j κn j . The change in total demand for each product by each country consumers is n n,h n,h n,h Dj = αj E Dj , h the change in demand for each product produced by each country is n n,m n,m ∆m j = Dj xj xj , n and the change in demand for inputs produced by each country is n,m n,m ∆m F = ∆n j xF xF . n n , rental rates r n , These updated values are employed to refine our guesses of the changes in wages wj j and input prices at origin pn M based on n wj = ∆n n j /Lj , n rj = ∆n n j /Tj , pn n F = ∆F . We keep iterating and adjusting these guesses until reaching equilibrium. For the case of the climate change shock, we input changes in yields from GAEZ. These yield n be the GAEZ projection projection changes are country n and crop j specific. To implement this, let zj in country n referring to crop j . Then we set n n Aj = z j , 3 n for all h, where Aj is given by (27). A2 Welfare To measure household welfare, we consider a linear transformation, with parameters ρn > 0 and γ n , of the indirect utility function   n n n νj E n,h − j C j Pj  V n,h = ρn n νj + γn.   n νj  j j Pjn This is convenient because we can set the linear transformation parameters γ n and ρn arbitrarily such that V n,h = E n,h for every household at the initial equilibrium (in the baseline). To achieve this, we n νj n νj n set ρn = j Pjn / j n νj and γ n = j C j Pjn . The percentage change in welfare implied by the money metric utility in the new equilibrium relative to the reference equilibrium is given by    E n,h E n,h − γ n γ n  1 V n,h E n,h , Pn =  n νj + γ n  n,h , n E j Pj n n where Pn is the vector of change in prices and γ n = j C j Pjn Pjn / j C j Pjn is the change in the cost of the subsistence bundle. Note that E n,h E n,h ≥ γ n γ n . We can also decompose the overall household welfare effect into its constituent parts, the income effect and the consumption or cost of living effect. The income effect is the change in welfare caused by a change in nominal income at constant prices. This is given by the percentage change in nominal income/expenditures E n,h . To calculate the changes in household-specific cost of living, we compute the indirect utility keeping nominal expenditure constant at the new prices V n,h 1, Pn . A3 Proofs for the labor and land supply problems For simplicity, we are only considering the labor allocation problem because the land allocation problem is isomorphic, and all proofs will hold for the land allocation problem if one replaces L with T , and w with r. We define the following variables for convenience: xj ≡ An,h n θL n,h θL L,j wj , and Bj ≡ (xj ) /(ΦL ) . Recall 4 1 Φn,h J θL θL that L = j =1 xj . We use an additional random variable zj with scale 1 and shape θL , instead of using the random variable ξj (ωL ) to make the algebra easier to follow. Define the pdf and −1−θL cdf of zj as f (zj ) = θL zj exp −(zj )−θL and F (zj ) = exp −(zj )−θL respectively. Recall that ξL,j (ωL ) has scale γL An,h n,h L,j and shape θL , thus we are setting γL AL,j zj = ξL,j (ωL ). Proposition 1 Subject to optimality, the probability of allocating labor to product j is equal to n,h (xj )θL πL,j = . (Φn,h L ) θL Proof. Optimality requires that zj xj ≥ zk xk for every k ̸= j , if the labor is allocated to j . This x j means zk ≤ zj xj /xk . For a given zj , the probability of this outcome is k̸=j F ( xk zj ). We need to integrate this probability over its domain after multiplying with f (zj ), to calculate the unconditional probability: ∞ n,h xj πL,j = f (zj ) F( zj )dzj , 0 xk k̸=j ∞ −θL −θL xj = θL (zj )−1−θL exp −zj exp − zj dzj , 0 xk k̸=j ∞ J −θL xj = θL (zj )−1−θL exp − zj dzj , 0 xk k=1   −θL ∞ xj = θL (zj )−1−θL exp − zj  dzj , 0 Φn,h L ∞ = θL (zj )−1−θL exp −(Bj )−1 (zj )−θL dzj , 0 define u = exp −(Bj )−1 (zj )−θL then 1 n,h πL,j = Bj du, 0 = Bj . Proposition 2 Subject to optimality, the effective units of labor allocated to production j is equal to n,h (xj )θL ΦL n,h Ln,h j = n L . (Φn,h )θL wj L Proof. Note that effective units of labor is the sum of labor units when taking the productivity shock ξL,j (ωL ) into account. Based on the previous proof, we can express Ln,h j as 5 ∞ Ln,h j = Ln,h γL An,h L,j zj θL (zj )−1−θL exp −(Bj )−1 (zj )−θL dzj , 0 ∞ −θL zj = Ln,h γL An,h L,j zj θL (zj )−1−θL exp − dzj , 0 S −1 where S ≡ Bj θ . Note that the expression above closely resembles the expected value of a random Frechet variable with scale S and shape θL . In fact, consider a Frechet draw z ′ with scale S and shape θL , then ∞ −1−θL −θL θL z′ z′ E (z ′ ) = z′ exp − dz ′ , 0 S S S Ln,h j 1 1 −1−θL = n,h . Ln,h γL A S L,j S Note that mean of Frechet drawn z ′ is E (z ′ ) = S γL , therefore S Ln,h j 1 1 −1−θL = n,h , γL Ln,h γL A S S L,j and Ln,h j = Ln,h An,h L,j (S ) 1−θL , θ −1 = Ln,h An,h L,j (Bj ) θL , n,h (xj )θL ΦL n,h = n L . (Φn,h )θL wj L Proposition 3 Return on labor allocated to production of j as a share of total return on labor is equal to the probability of allocating labor to production of j . Proof. The return on labor allocated to production of j is equal to Ln,h n j wj , and the total return on labor is equal to k Ln,h n k wk , therefore 6 (xj )θL Ln,h n Φn,h L L n,h j wj (Φn,h )θL L = , k Ln,h k wk n (xk )θL n,h n,h k (Φn,h )θL ΦL L L (xj )θL Φn,h L L n,h (Φn,h L ) θL = , Φn,h L L n,h (xj )θ = , (Φn,h L ) θL n,h = πL,j . Proposition 4 Expected average returns from labor allocated to each crop are equalized, such that Ln,h n n,h n,h n n,h ′ n,h n,h j wj /πL,j = Lj ′ wj ′ /πL,j ′ for any j and j as long as both πL,j > 0 and πL,j ′ > 0. Proof. This directly follows from the expressions derived in previous propositions. Ln,h j wj n n,h = Φn,h L L n,h , πL,j Ln,h n j ′ wj ′ = n,h , πL,j ′ n,h Hence it holds for to any j ′ as long as the expression is not divided by zero, i.e. πL,j ′ > 0. n,h Proposition 5 The total wage income (return on labor) of household h can be calculated as RL = Φn,h L L n,h , if labor is allocated optimally to crops. Proof. n,h RL = Ln,h n j wj , j n,h Φn,h L n,h n = πL,j n L wj , wj j = Φn,h L L n,h n,h πL,j , j = Φn,h L L n,h . 7 A4 A note on the upward bias in factor income when using aggregate data It is possible to show that the change in factor income caused by using aggregate data is biased upwards for a given vector of factor price changes. Therefore, if the aggregate welfare impact is negative, it will be underestimated; and if the impact is positive, it will be overestimated. However, it is important to note that this is only true for a given vector of factor price changes, and the direction of the overall bias will be ambiguous in a complex general equilibrium setting, such as our model. This is because factor prices (wages and rental rates), final good prices and intermediate input prices are subject to general equilibrium effects. Nevertheless, the proposition below is useful for demonstrating the existence of a bias and for understanding its underlying mechanisms. Proposition 6 When an aggregate version of our model is used (i.e. H n = 1) with aggregate factor allocations, instead of varying factor allocations with household, the change in total factor income is biased upwards for a given vector of factor price changes (i.e. wage or land rental rate changes) when θL > 1 and θT > 1. Proof. n,h The initial return on labor for household h is denoted as RL , the share of labor income from crop j is π n,h j .The share of household h in total labor income of the economy is denoted with η ˜Ln,h = n,h n,h′ R L /( RL ). The new labor income for household h after a change in wages is   1 θL n,h θL n,h RL = RL  π n,h j n wj  . j n > 0 by construction (i.e. 0 < w n < 1 when price change is Note that θL > 1 by assumption and wj j n ≥ 1 otherwise). The function above is strictly concave in π n,h , as all elements of the negative and wj j Hessian are negative such that 1 θL −2 n,h 1 1−θ n n θL Hjk = RL w w π n,h n wl < 0, ∀j, k . θL θ L j k l l Recall that the change in total labor income is    1  θL n,h θL n,h  n RL = RL η ˜L π n,h n wj .     j h h j 8 If we were to use aggregate data, i.e. economy-wide allocations rather than household specific n = n,h n,h allocations, shares would be πj ˜L hη πj , then we could write the change in labor income using aggregate labor income data as   1  θL n,h θL n∗ n,h n,h n RL = RL ˜L η πj wj .     h j h n∗ ≥ Rn due to Jensen’s inequality, or more precisely Rn∗ > Rn as the function is Note that RL L L L strictly concave. The proof for land income is identical. A5 Alternative characterization of the equilibrium The equilibrium can alternatively be characterized without using the return on land. We preferred to use the version that uses the return on land in the main text because it simplifies the equations and the solution method significantly and makes the exposition easier to follow. The alternative characterization is provided below. Definition. The international trade equilibrium is given by a vector of crop prices for each crop variety in each country, pn n j ; a vector of manufacturing prices for each country variety, pM ; a vector of services in each country, pn n S ; a vector of intermediate input prices for each country variety, pF ; a vector n ; and rental rates of wages for each product (crops, manufacturing and services), for each country wj n , services r n and intermediate inputs r n , such that: for the specific factors in manufacturing rM S F Goods Market. For each product, global demand equals national supply. For crops j , we combine total household expenditures (5) and the value of national output (12) to express the equilibrium condition as xn,m j n,h αj n,h RT n,h + RM n,h + RS n,h + RL m = yj , (29) n h where xn,m j is the import share based on the price vector pn n,h j defined by (3) and (4), αj is the n,h n,h expenditure share on composite j given by (1) , RM and RS are the fixed factor revenues in sectors n,h n,h M and S (defined by (14)) accruing to household h, RT and RL are given by (10) and (18) and n,h n,h n,h n,h RT + RM + RS + RL = E n,h is total household expenditure. The equilibrium for manufactures is the same as (29) with j = M xn,m M αM n,h n,h RT n,h + RM n,h + RS n,h + RL m = yM . (30) n h Supply and demand of the non-traded good requires that 9 n,h n,h n,h n,h n,h n αS RT + RM + RS + RL = yS . (31) h For the intermediate input F , we have that xn,m n,h F β F yj m = yF . (32) n h j Factor Markets. The equilibrium wage in sector j equates labor supply with labor demand βL +βT βF θL −1 n,h βT θT −1 n,h n,h θL L 1 βF βT βL n,h θT T πL,j = pn j βT n n πT,j . (33) h An,h L,j h PF wj An,h T,j For manufactures and services, we have θL −1 n,h 1 n,h θL L pn k βL,k βK,k n πL,k = n Kk , (34) An,h L,k wk for k = M, S . Given prices, the rental rates for the specific factors can be recovered from (14). Finally, while land is not traded, households allocate their own plots to different crops and, given crop prices, there is an equilibrium return for land given by (10). Note. The equilibrium outlined in the main text, in section 2.4, and the alternative characterization of equilibrium given above are isomorphic, and therefore give the same outcome. Proof. We can substitute out the return on land, by combining equations (24), (22) and (23) from the main definition of the equilibrium, then obtain precisely the same conditions described in the alternative definition. A6 Supplementary simulations This section presents the results of supplementary simulations to assess the mechanisms that drive the results of the war shock in more detail. Inputs channel (the war). This scenario is similar to the export ban simulation in the main text, but only fertilizers are considered: (i) Ukraine cannot import or export fertilizers; (ii) Russia bans exporting fertilizers. Final goods channel (the war). This scenario is similar to the main simulation, but fertilizers are excluded: (i) Ukraine cannot import or export any agricultural products, but not fertilizers; (ii) Russia bans exporting wheat, corn, other cereals, sugar, other oilseeds, but not fertilizers. Retaliatory protectionism (the war). In an additional simulation we consider the policy responses by other countries based on the news and policy reports (see CNBC (2022), The Wall Street Journal (2022), Reuters (2022) and Agri-Pulse (2022): (i) Ukraine cannot import or export any agricultural products, or fertilizers; (ii) Russia bans exporting wheat, rice, corn, other cereals, sugar, 10 other oilseeds, and fertilizers; (iii) Other countries respond (as documented in the news): Argentina bans exporting soya, Georgia bans exporting wheat, Ghana bans exporting corn, India bans exporting wheat and rice, Indonesia bans exporting oils, Japan bans exporting rice, Kazakhstan bans exporting wheat, Kyrgyzstan bans exporting wheat, Moldova bans exporting wheat, Tanzania bans exporting urkiye bans exporting corn, meats, and oils. corn, Tunisia bans exporting fruits and vegetables, T¨ 11 Table A1: Impact of the war - Robustness Inputs channel ∆Welfare ∆Income ∆CL Exposure Average Bottom Top Single Total Labor Land HH. Imports 25% 25% HH. Panel A: All countries (pooled) Average -1.67 -1.77 -1.52 -1.57 -0.40 -0.31 -0.57 1.30 5.42 Pop w. aver. -1.24 -1.37 -1.02 -1.08 -0.01 -0.02 -0.02 1.25 4.30 Std. dev. 1.61 1.74 1.55 1.60 1.57 1.17 2.19 0.37 5.35 Minimum -9.25 -10.18 -8.30 -8.94 -7.59 -5.43 -9.28 -0.13 0.30 Median -1.21 -1.26 -1.09 -1.16 0.11 0.10 0.13 1.28 3.66 Maximum 0.05 0.13 0.15 0.13 1.02 0.90 1.26 2.08 19.97 Panel B: By country Azerbaijan -9.25 -10.18 -8.30 -8.94 -7.59 -5.43 -9.28 1.84 18.25 Mongolia -7.27 -7.25 -7.65 -7.33 -5.34 -3.64 -8.56 2.08 6.10 Moldova -4.54 -4.97 -3.86 -4.45 -4.14 -2.94 -7.65 0.42 17.72 Cent. Afr. Rep. -3.68 -3.58 -3.65 -3.66 -2.05 -2.05 -2.05 1.69 1.12 Nicaragua -2.85 -3.20 -2.45 -2.64 -1.69 -1.51 -1.92 1.19 1.42 Cameroon -2.80 -3.04 -2.45 -2.55 -1.21 -1.19 -1.22 1.64 0.63 Georgia -2.76 -3.10 -2.43 -2.60 -1.88 -1.55 -2.37 0.91 19.97 Armenia -2.61 -2.59 -2.52 -2.59 -0.93 -0.76 -1.30 1.72 17.85 Ecuador -2.52 -3.01 -1.89 -2.19 -1.16 -1.03 -1.40 1.39 14.18 Burkina Faso -2.37 -2.46 -2.25 -2.24 -1.31 -1.22 -1.34 1.08 1.36 Cˆote d’Ivoire -2.34 -2.56 -2.05 -2.14 -1.22 -0.98 -1.29 1.14 4.09 Kyrgyzstan -2.32 -2.48 -2.18 -2.25 -1.13 -0.90 -1.50 1.22 11.60 Gambia, The -1.80 -1.54 -1.81 -1.81 -0.10 -0.09 -0.10 1.74 2.13 Mauritania -1.63 -1.97 -1.17 -1.35 0.01 -0.00 0.01 1.67 2.85 Tanzania -1.59 -1.78 -1.30 -1.32 -0.55 -0.50 -0.62 1.05 5.04 Bolivia -1.55 -1.63 -1.26 -1.32 -0.02 -0.03 -0.01 1.56 1.54 Liberia -1.43 -1.30 -1.47 -1.46 0.21 0.18 0.22 1.66 1.70 Indonesia -1.38 -1.65 -1.05 -1.16 -0.18 -0.16 -0.32 1.21 2.62 Notes : This table presents the results of a simulation in which (i) Ukraine cannot import or export any fertilizers and (ii) Russia bans exporting fertilizers. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure measures what share of imports of a given country were accounted for by imports from Ukraine and Russia before the war. Countries are ordered in terms of average real income gains (from lowest to highest). 12 Table A1: Impact of the war - Robustness (continued) Inputs channel ∆Welfare ∆Income ∆CL Exposure Average Bottom Top Single Total Labor Land HH. Imports 25% 25% HH. Ghana -1.37 -1.76 -1.10 -1.17 -1.50 -1.31 -2.29 -0.13 3.23 Mozambique -1.34 -0.78 -1.87 -1.65 0.11 0.05 0.13 1.47 6.77 Viet Nam -1.30 -1.29 -1.20 -1.25 -0.04 -0.04 -0.04 1.28 2.72 Sierra Leone -1.28 -1.45 -1.16 -1.20 0.37 0.30 0.38 1.67 4.12 Togo -1.27 -1.18 -1.24 -1.24 0.09 0.11 0.08 1.38 2.90 Jordan -1.26 -1.39 -1.09 -1.18 0.16 0.13 0.27 1.44 0.76 Nigeria -1.25 -1.02 -1.27 -1.27 0.18 0.14 0.23 1.45 2.11 Guatemala -1.21 -1.21 -1.12 -1.16 0.01 0.01 0.03 1.24 1.02 Yemen, Rep. -1.19 -1.25 -1.04 -1.07 0.47 0.41 0.56 1.68 3.76 Kenya -1.19 -1.33 -0.97 -0.96 -0.06 -0.06 -0.06 1.14 4.81 South Africa -1.18 -1.72 -0.52 -0.63 -0.05 -0.05 -0.11 1.15 3.66 Zambia -1.18 -1.26 -1.13 -1.15 0.46 0.34 0.61 1.66 2.96 Comoros -1.11 -1.16 -1.04 -1.06 0.56 0.50 0.60 1.69 0.30 Egypt, Arab Rep. -1.09 -1.10 -1.03 -1.04 0.37 0.29 0.49 1.48 6.91 Burundi -1.03 -1.02 -0.89 -0.94 0.45 0.29 0.47 1.50 4.74 Bangladesh -1.02 -1.16 -0.84 -0.93 0.46 0.34 0.54 1.50 4.65 Rwanda -0.89 -1.11 -0.65 -0.74 -0.07 -0.08 -0.07 0.82 1.77 Iraq -0.87 -1.06 -0.55 -0.68 0.28 0.18 0.49 1.15 4.14 Guinea -0.86 -0.75 -0.95 -0.89 0.50 0.39 0.51 1.38 0.96 Sri Lanka -0.86 -1.28 -0.36 -0.52 0.25 0.19 0.34 1.13 15.57 Guinea-Bissau -0.86 -0.98 -0.55 -0.56 0.82 0.80 0.84 1.70 3.65 Niger -0.82 -0.57 -1.02 -0.90 0.48 0.38 0.50 1.32 5.40 Uganda -0.82 -0.89 -0.85 -0.82 0.11 0.10 0.12 0.94 2.13 Papua New G. -0.72 -0.67 -0.70 -0.68 0.64 0.46 0.73 1.37 1.29 Benin -0.70 -0.45 -0.90 -0.82 0.44 0.37 0.47 1.14 1.75 Uzbekistan -0.69 -0.57 -0.92 -0.78 0.34 -0.05 0.76 1.04 17.31 Pakistan -0.61 -1.11 0.03 -0.17 0.37 0.33 0.50 0.99 3.77 Tajikistan -0.61 -0.53 -0.72 -0.63 0.46 0.45 0.49 1.07 9.41 Cambodia -0.59 -0.61 -0.50 -0.50 0.68 0.47 0.83 1.28 3.32 Madagascar -0.53 -0.47 -0.61 -0.53 0.67 0.56 0.73 1.21 0.59 Malawi -0.41 -0.28 -0.67 -0.54 0.47 0.37 0.53 0.89 8.34 Nepal -0.41 -0.53 -0.28 -0.33 0.57 0.57 0.58 0.98 7.72 Bhutan 0.05 0.13 0.15 0.13 1.02 0.90 1.26 0.97 3.80 Notes : This table is a continuation from the table on the previous page. It presents the results of a simulation in which (i) Ukraine cannot import or export any fertilizers and (ii) Russia bans exporting fertilizers. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure measures what share of imports of a given country were accounted for by imports from Ukraine and Russia before the war. Countries are ordered in terms of average real income gains (from lowest to highest). 13 Table A2: Impact of the war - Robustness Final goods channel ∆Welfare ∆Income ∆CL Exposure Average Bottom Top Single Total Labor Land HH. Imports 25% 25% HH. Panel A: All countries (pooled) Average -0.31 -0.37 -0.23 -0.26 0.45 0.36 0.58 0.77 5.42 Pop w. aver. -0.05 -0.11 0.03 0.00 0.46 0.38 0.59 0.52 4.30 Std. dev. 0.88 1.12 0.67 0.78 0.48 0.41 0.68 1.11 5.35 Minimum -4.17 -5.06 -2.74 -3.47 -0.11 -0.11 -0.13 0.02 0.30 Median -0.10 -0.06 -0.09 -0.08 0.25 0.20 0.32 0.39 3.66 Maximum 0.76 0.88 1.35 1.09 2.28 1.80 3.29 6.27 19.97 Panel B: By country Georgia -4.17 -5.01 -2.74 -3.47 0.61 0.47 0.81 5.00 19.97 Armenia -3.74 -5.06 -2.67 -3.25 2.28 1.80 3.29 6.27 17.85 Mongolia -1.88 -2.27 -1.47 -1.71 -0.00 -0.01 0.02 1.92 6.10 Egypt, Arab Rep. -1.30 -1.55 -0.98 -1.13 0.60 0.48 0.77 1.92 6.91 Azerbaijan -1.03 -1.54 -0.41 -0.82 0.83 0.57 1.03 1.88 18.25 Mauritania -0.90 -1.14 -0.61 -0.72 0.09 0.06 0.10 1.01 2.85 Yemen, Rep. -0.78 -1.18 -0.36 -0.54 0.09 0.07 0.12 0.88 3.76 Jordan -0.73 -1.05 -0.46 -0.59 0.17 0.14 0.27 0.91 0.76 Zambia -0.64 -0.70 -0.58 -0.61 0.20 0.15 0.26 0.84 2.96 South Africa -0.62 -0.89 -0.21 -0.28 0.01 0.01 0.16 0.63 3.66 Kyrgyzstan -0.55 -0.66 -0.38 -0.48 0.86 0.68 1.16 1.41 11.60 Mozambique -0.52 0.05 -1.36 -1.05 0.18 0.10 0.21 0.71 6.77 Tajikistan -0.48 -0.42 -0.52 -0.47 0.55 0.54 0.60 1.04 9.41 Uzbekistan -0.34 -0.21 -0.61 -0.44 0.96 0.16 1.82 1.31 17.31 Bolivia -0.33 -0.46 -0.18 -0.23 0.10 0.08 0.12 0.44 1.54 Guinea-Bissau -0.26 -0.28 -0.19 -0.18 0.19 0.17 0.20 0.45 3.65 Ecuador -0.26 -0.33 -0.18 -0.21 -0.11 -0.11 -0.13 0.14 14.18 Guatemala -0.24 -0.24 -0.19 -0.21 0.08 0.06 0.10 0.32 1.02 Niger -0.23 -0.15 -0.29 -0.26 0.04 0.05 0.04 0.27 5.40 Notes : This table presents the results of a simulation in which (i) Ukraine cannot import or export any agricultural product and (ii) Russia bans exporting wheat, rice, corn, other cereals, sugar, and other oilseeds. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure measures what share of imports of a given country were accounted for by imports from Ukraine and Russia before the war. Countries are ordered in terms of average real income gains (from lowest to highest). 14 Table A2: Impact of the war - Robustness (continued) Final goods channel ∆Welfare ∆Income ∆CL Exposure Average Bottom Top Single Total Labor Land HH. Imports 25% 25% HH. Nicaragua -0.14 -0.14 -0.10 -0.11 0.25 0.19 0.32 0.39 1.42 Burundi -0.14 -0.17 -0.05 -0.08 0.11 0.13 0.12 0.25 4.74 Gambia, The -0.13 -0.10 -0.12 -0.12 0.07 0.06 0.07 0.19 2.13 Liberia -0.13 -0.02 -0.18 -0.16 0.24 0.20 0.26 0.36 1.70 Malawi -0.11 0.02 -0.27 -0.20 0.16 0.13 0.18 0.27 8.34 Bhutan -0.11 -0.13 -0.02 -0.05 0.31 0.28 0.37 0.43 3.80 Cambodia -0.10 -0.11 -0.09 -0.09 -0.08 -0.06 -0.10 0.02 3.32 Sri Lanka -0.09 -0.13 -0.03 -0.05 0.02 0.00 0.05 0.10 15.57 Papua New G. -0.08 -0.06 -0.11 -0.07 0.39 0.29 0.44 0.48 1.29 Togo -0.07 -0.00 -0.09 -0.08 0.24 0.24 0.24 0.31 2.90 Nigeria -0.06 -0.06 -0.09 -0.06 0.32 0.27 0.38 0.38 2.11 Viet Nam -0.06 -0.05 -0.05 -0.05 0.02 -0.00 0.03 0.07 2.72 Indonesia -0.02 -0.02 -0.02 -0.02 0.04 0.03 0.11 0.06 2.62 Burkina Faso -0.01 -0.01 -0.01 0.01 0.21 0.18 0.22 0.21 1.36 Cˆote d’Ivoire 0.01 0.03 -0.00 0.00 0.09 0.06 0.09 0.08 4.09 Bangladesh 0.02 0.02 0.02 0.02 0.35 0.25 0.41 0.33 4.65 Rwanda 0.05 0.06 0.02 0.02 0.22 0.17 0.29 0.17 1.77 Madagascar 0.06 0.08 0.01 0.05 0.23 0.18 0.25 0.17 0.59 Cameroon 0.10 0.11 0.08 0.09 0.35 0.34 0.36 0.25 0.63 Uganda 0.11 0.13 0.02 0.07 0.45 0.41 0.47 0.33 2.13 Moldova 0.13 0.03 0.09 0.17 1.06 0.72 2.08 0.93 17.72 Tanzania 0.13 0.13 0.11 0.13 0.35 0.30 0.40 0.22 5.04 Sierra Leone 0.19 0.37 0.02 0.12 0.53 0.40 0.55 0.34 4.12 Guinea 0.20 0.30 0.13 0.17 0.53 0.42 0.54 0.32 0.96 Kenya 0.22 0.42 0.14 0.22 1.24 1.11 1.40 1.02 4.81 Benin 0.28 0.42 0.14 0.23 0.62 0.53 0.67 0.34 1.75 Ghana 0.32 0.37 0.26 0.28 0.54 0.46 0.90 0.22 3.23 Cent. Afr. Rep. 0.33 0.54 0.21 0.23 1.13 1.09 1.13 0.80 1.12 Comoros 0.34 0.33 0.27 0.32 0.95 0.86 1.00 0.61 0.30 Nepal 0.59 0.64 0.52 0.55 1.36 1.35 1.39 0.76 7.72 Iraq 0.69 0.88 0.61 0.64 1.25 0.86 2.16 0.56 4.14 Pakistan 0.76 0.28 1.35 1.09 1.51 1.38 1.82 0.75 3.77 Notes : This table is a continuation from the table on the previous page. It presents the results of a simulation in which (i) Ukraine cannot import or export any agricultural product and (ii) Russia bans exporting wheat, rice, corn, other cereals, sugar, and other oilseeds. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single. HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure measures what share of imports of a given country were accounted for by imports from Ukraine and Russia before the war. Countries are ordered in terms of average real income gains (from lowest to highest). 15 Table A3: Impact of the war - Robustness Retaliatory protectionism ∆Welfare ∆Income ∆CL Exposure Average Bottom Top Single Total Labor Land HH. Imports 25% 25% HH. Panel A: All countries (pooled) Average -2.24 -2.43 -1.95 -2.05 0.23 0.18 0.22 2.55 5.42 Pop w. aver. -1.43 -1.68 -1.06 -1.18 0.67 0.53 0.85 2.13 4.30 Std. dev. 2.30 2.66 2.03 2.20 1.90 1.48 2.67 1.68 5.35 Minimum -10.87 -12.56 -9.55 -10.25 -6.32 -4.56 -8.65 -0.60 0.30 Median -1.79 -1.61 -1.61 -1.60 0.49 0.35 0.51 2.36 3.66 Maximum 0.99 0.98 1.77 1.08 4.21 3.88 7.26 10.04 19.97 Panel B: By country Azerbaijan -10.87 -12.56 -8.97 -10.25 -6.32 -4.56 -7.69 5.14 18.25 Mongolia -9.72 -10.28 -9.55 -9.55 -5.42 -3.72 -8.65 4.76 6.10 Georgia -8.50 -9.91 -6.25 -7.42 -1.24 -1.08 -1.50 7.98 19.97 Armenia -7.33 -9.06 -5.82 -6.68 1.95 1.51 2.87 10.04 17.85 Moldova -4.24 -4.58 -3.71 -4.18 -3.85 -2.76 -7.03 0.40 17.72 Cent. Afr. Rep. -3.70 -3.43 -3.67 -3.70 -0.82 -0.81 -0.82 2.98 1.12 Kyrgyzstan -3.26 -3.55 -2.93 -3.11 -0.72 -0.61 -0.89 2.63 11.60 Mauritania -3.18 -3.89 -2.25 -2.61 -0.04 -0.08 -0.04 3.24 2.85 Nicaragua -3.13 -3.51 -2.65 -2.87 -1.42 -1.32 -1.56 1.76 1.42 Ecuador -2.92 -3.51 -2.17 -2.51 -1.37 -1.23 -1.62 1.59 14.18 Egypt, Arab Rep. -2.84 -3.17 -2.39 -2.59 0.95 0.74 1.24 3.90 6.91 Cameroon -2.77 -3.01 -2.47 -2.53 -0.47 -0.46 -0.48 2.36 0.63 Ghana -2.62 -3.23 -2.15 -2.27 -3.21 -2.79 -5.03 -0.60 3.23 ote d’Ivoire Cˆ -2.54 -2.79 -2.21 -2.31 -1.12 -0.91 -1.18 1.45 4.09 Yemen, Rep. -2.44 -3.12 -1.64 -1.95 0.65 0.54 0.81 3.18 3.76 Burkina Faso -2.32 -2.33 -2.23 -2.19 -0.92 -0.90 -0.93 1.43 1.36 Tajikistan -2.25 -1.94 -2.47 -2.27 1.40 1.37 1.55 3.74 9.41 Mozambique -2.23 -0.91 -3.82 -3.20 0.29 0.13 0.35 2.60 6.77 Jordan -2.20 -2.65 -1.73 -1.96 0.71 0.59 1.14 2.98 0.76 Notes : This table presents the results of a simulation in which (i) Ukraine cannot import or export any agricultural products, or fertilizers and (ii) Russia bans exporting wheat, rice, corn, other cereals, sugar, other oilseeds, and fertilizers. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure measures what share of imports of a given country were accounted for by imports from Ukraine and Russia before the war. Countries are ordered in terms of average real income gains (from lowest to highest). 16 Table A3: Impact of the war - Robustness (continued) Retaliatory protectionism ∆Welfare ∆Income ∆CL Exposure Average Bottom Top Single Total Labor Land HH. Imports 25% 25% HH. Gambia, The -2.19 -1.86 -2.16 -2.18 -0.10 -0.11 -0.07 2.14 2.13 Bolivia -2.10 -2.35 -1.57 -1.71 -0.01 -0.07 0.04 2.13 1.54 Zambia -2.08 -2.22 -1.97 -2.01 0.67 0.50 0.92 2.82 2.96 South Africa -2.04 -2.93 -0.84 -1.05 -0.09 -0.09 0.03 2.00 3.66 Liberia -1.97 -1.51 -2.12 -2.05 0.91 0.76 0.98 2.93 1.70 Tanzania -1.92 -2.04 -1.61 -1.60 -0.63 -0.56 -0.71 1.31 5.04 Nigeria -1.79 -1.55 -1.79 -1.79 0.61 0.50 0.73 2.44 2.11 Togo -1.75 -1.68 -1.64 -1.66 0.32 0.34 0.32 2.11 2.90 Guatemala -1.58 -1.57 -1.41 -1.48 0.09 0.06 0.14 1.69 1.02 Guinea-Bissau -1.56 -1.78 -1.07 -1.07 0.97 0.91 1.01 2.57 3.65 Burundi -1.50 -1.54 -1.17 -1.25 0.49 0.39 0.51 2.03 4.74 Viet Nam -1.43 -1.44 -1.29 -1.36 0.02 -0.02 0.05 1.47 2.72 Bangladesh -1.37 -1.58 -1.10 -1.23 1.08 0.77 1.27 2.49 4.65 Niger -1.32 -0.96 -1.57 -1.41 0.40 0.35 0.42 1.75 5.40 Uzbekistan -1.13 -0.86 -1.64 -1.33 1.81 0.31 3.42 2.98 17.31 Sri Lanka -1.06 -1.60 -0.44 -0.63 0.28 0.19 0.44 1.36 15.57 Kenya -0.99 -0.89 -0.80 -0.71 1.82 1.57 2.09 2.84 4.81 Rwanda -0.99 -1.20 -0.74 -0.84 0.11 0.06 0.20 1.11 1.77 Comoros -0.96 -1.12 -0.90 -0.88 1.82 1.62 1.93 2.80 0.30 Uganda -0.92 -1.06 -1.02 -0.94 0.68 0.61 0.72 1.62 2.13 Papua New G. -0.85 -0.78 -0.85 -0.78 1.24 0.88 1.40 2.10 1.29 Indonesia -0.82 -0.90 -0.74 -0.75 -0.13 -0.13 -0.15 0.69 2.62 Cambodia -0.71 -0.76 -0.59 -0.60 0.68 0.45 0.82 1.40 3.32 Malawi -0.71 -0.38 -1.21 -0.98 0.57 0.44 0.65 1.29 8.34 Madagascar -0.61 -0.53 -0.75 -0.61 1.05 0.86 1.16 1.68 0.59 Benin -0.50 -0.05 -0.88 -0.69 1.24 1.04 1.33 1.74 1.75 Sierra Leone -0.49 -0.29 -0.72 -0.51 2.46 1.96 2.52 2.97 4.12 Bhutan -0.49 -0.59 -0.09 -0.20 1.50 1.32 1.83 1.99 3.80 Guinea -0.43 0.00 -0.73 -0.54 1.96 1.52 2.02 2.40 0.96 Pakistan -0.09 -1.61 1.77 1.08 2.57 2.32 3.21 2.67 3.77 Iraq 0.06 0.20 0.51 0.31 4.21 2.88 7.26 4.15 4.14 Nepal 0.99 0.98 0.95 0.96 3.91 3.88 3.97 2.89 7.72 Notes : This table is a continuation from the table on the previous page. It presents the results of a simulation in which (i) Ukraine cannot import or export any agricultural products, or fertilizers and (ii) Russia bans exporting wheat, rice, corn, other cereals, sugar, other oilseeds, and fertilizers. Welfare change (∆Welfare), income effect (∆Income), and cost of living effect (∆CL) are expressed as equivalent percentage changes in real household income relative to the (pre-shock) status quo, as defined in section 2.5. Bottom 25% (top 25%) refers to the poorest (richest) 25% of the population within a given country. Single HH. denotes a representative household and presents the results of a single agent model in which all households are aggregated into one representative household. Exposure measures what share of imports of a given country were accounted for by imports from Ukraine and Russia before the war. Countries are ordered in terms of average real income gains (from lowest to highest). 17 A7 Concordance Table A4: Concordance ITPDE-HIT (1/2) Code Description ITPD-E Product(s) 1 Wheat Wheat 2 Rice Rice (raw) 3 Corn Corn 4 Other cereals Other cereals, Cereal products, Grain mill products 5 Soya Soybeans 6 Other oilseeds Other oilseeds (exc. peanuts) 7 Sugar Raw, Refined sugar, Sugar crops, Other sweeteners, Sugar 8 Legumes Pulses, Legumes(dried, preserved) 9 Fruits and vegetables Fresh fruit, Fresh vegetables, Processing/preserving of fruit and vegetables 10 Nuts Nuts 11 Eggs/Meat/Dairy Live Cattle, Live Swine, Other meats, livest. pr. live animals, Processing/preserving of meat, Eggs, Dairy products 12 Cocoa Cocoa and cocoa products, Cocoa chocolate and sugar confectionery 13 Oils/Fats Vegetable and animal oils and fats 14 Other staple food Animal feed ingredients and pet foods, Prepared fruits, fruit juices, Prepared vegetables, Other ag. products, nec, Starches and starch products, Prepared animal feeds, Bakery products, Macaroni noodles and similar products, Other food products n.e.c. 15 Beverages, nec Beverages (nec), Soft drinks, mineral waters 16 Cotton Cotton 17 Tobacco Tobacco leaves and cigarettes, Tobacco products 18 Spices/herbs Spices 19 Alcohol Wines, Malt liquors and malt, Distilling rectifying and blending of spirits 20 Fish Processing/preserving of fish 21 Manufacturing Mining of hard coal, Mining of lignite, Extraction crude oil and gas, Mining of iron ores, (continues Other mining and quarring, Electricity prodcn, collcn, and distr., Gas production and on next page) distribution, Coke oven products, Refined petroleum products, Processing of nuclear fuel, Textile fibre preparation; textile weaving, Made-up textile articles except apparel, Carpets and rugs, Cordage rope twine and netting, Other textiles n.e.c., Knitted and crocheted fabrics and articles, Wearing apparel except fur apparel, Dressing and dyeing of fur, processing of fur, Tanning and dressing of leather, Luggage handbags etc., saddlery and harness, Footwear, Sawmilling and planing of wood, Veneer sheets plywood particle board etc., Builders’ carpentry and joinery, Wooden containers, Other wood products; articles of cork/straw, Furniture, Domestic appliances n.e.c., Office accounting and computing machinery, Electric motors generators and transformers, Electricity distribution and control apparatus, Insulated wire and cable, Accumulators primary cells and batteries, Lighting equipment and electric lamps, Other electrical equipment n.e.c., Electronic valves tubes etc., TV/radio transmitters; line comm. apparatus, TV and radio receivers and associated goods, Medical surgical and orthopedic equipment, Measuring/testing/navigating appliances and equipment, Optical instruments and photographic equipment, Watches and clocks, Pulp paper and paperboard, Corrugated paper, and paperboard, Other articles of paper and paperboard, 18 Table A4: Concordance between HIT and ITPDE (2/2) Code Description ITPD-E Product(s) 21 Manufacturing (continued) Publishing of books and other publications, Publishing of newspapers journals etc., Publishing of recorded media, Other publishing, Printing, Service activities related to printing, Reproduction of recorded media, Basic chemicals except fertilizers, Fertilizers and nitrogen compounds, Plastics in primary forms; synthetic rubber, Pesticides and other agro-chemical products, Paints varnishes printing ink and mastics, Pharmaceuticals medicinal chemicals etc., Soap cleaning and cosmetic preparations, Other chemical products n.e.c., Man-made fibers, Rubber tires and tubes, Other rubber products, Plastic products, Glass and glass products, Pottery china and earthenware, Refractory ceramic products, Struct.non-refractory clay; ceramic products, Cement lime and plaster, Articles of concrete cement and plaster, Cutting shaping and finishing of stone, Other non-metallic mineral products n.e.c., Basic iron and steel, Basic precious and non-ferrous metals, Casting of iron and steel, Structural metal products, Tanks reservoirs and containers of metal, Steam generators, Cutlery hand tools and general hardware, Other fabricated metal products n.e.c., Engines and turbines (not for transport equipment), Pumps compressors taps and valves, Bearings gears gearing and driving elements, Ovens furnaces and furnace burners, Lifting and handling equipment, Other general purpose machinery, Agricultural and forestry machinery, Machine tools, Machinery for metallurgy, Machinery for mining and construction, Food/beverage/tobacco processing machinery, Machinery for textile apparel and leather, Weapons and ammunition, Other special purpose machinery, Motor vehicles, Automobile bodies trailers and semi-trailers, Parts/accessories for automobiles, Building and repairing of ships, Building/repairing of pleasure/sport. boats, Railway/tramway locomotives and rolling stock, Aircraft and spacecraft, Motorcycles, Bicycles and invalid carriages, Other transport equipment n.e.c., Jewelery and related articles, Musical instruments, Sports goods, Games and toys, 22 Services Transport, Travel, Health services, Education services, Telecommunications, computer and information services, Manufacturing services on physical inputs owned by others, Maintenance and repair services n.i.e., Construction, Insurance and pension services, Financial services, Charges for the use of intellectual property n.i.e., Other business services, Heritage and recreational services, Government goods and services n.i.e., Services not allocated, Trade-related services, Other personal services 23 Fertilizers Fertilizers and nitrogen compounds 19 Table A5: Concordance between HIT-ITPDE and WFP food prices data (1/3) Description WFP Commodity Code 1 Wheat Buckwheat, Buckwheat grits, Bulgur, Couscous, Feed (flour), Feed (rakhel), Feed (wheat bran), Noodles (short), Pasta, Pasta (macaroni), Pasta (spaghetti), Semolina, Wheat, Wheat (food aid), Wheat (imported), Wheat (mixed), Wheat (white), Wheat flour, Wheat flour (first grade), Wheat flour (high quality), Wheat flour (imported), Wheat flour (local), Wheat flour (Turkey), Wheat meal 2 Rice Rice, Rice (aromatic), Rice (basmati broken), Rice (basmati), Rice (broken imported), Rice (carolina 1st), Rice (carolina 2da), Rice (coarse BR-8/ 11/ Guti Sharna), Rice (coarse Guti Sharna), Rice (coarse), Rice (denikassia imported), Rice (estaquilla), Rice (good quality), Rice (Grano de Oro), Rice (high quality local), Rice (high quality), Rice (imported Pakistan), Rice (imported Tanzanian), Rice (imported), Rice (local), Rice (long grain high quality local), Rice (long grain imported), Rice (long grain), Rice (low quality local), Rice (low quality), Rice (medium grain imported), Rice (medium grain), Rice (medium quality), Rice (milled local), Rice (mixed low quality), Rice (ordinary first quality), Rice (ordinary second quality), Rice (paddy long grain local), Rice (paddy), Rice (red nadu), Rice (red), Rice (short grain low quality local), Rice (small grain imported), Rice (white imported 551), Rice (white imported JPC SK Gold), Rice (white imported), Rice (white), Rice (milled 80-20) 3 Corn Corn Soy Blend (CSB++ food aid), Cornstarch, Maize, Maize (crushed), Maize (food aid), Maize (local), Maize (white biofortified), Maize (white dry), Maize (white East), Maize (white North), Maize (white), Maize (yellow biofortified), Maize (yellow), Maize flour, Maize flour (imported), Maize flour (white), Maize meal, Maize meal (white breakfast), Maize meal (white first grade), Maize meal (white roller), Maize meal (white with bran), Maize meal (white without bran), Tortilla (maize) 4 Other cereals Barley, Barley (mixed), Barley (white), Fonio, Millet, Millet (bulrush), Millet (finger), Millet flour, Oat flakes, Quinoa, Sorghum, Sorghum (brown), Sorghum (food aid), Sorghum (local), Sorghum (mixed), Sorghum (red), Sorghum (r’haya), Sorghum (taghalit), Sorghum (white), Teff, Teff (mixed), Teff (red), Teff (Sergegna), Teff (white) 5 Soya Soybeans 6 Other oilseeds Lin seed, Rape seed, Sesame 7 Sugar Cane juice (light), Cane juice (strong), Honey, Sugar, Sugar (brown imported), Sugar (brown local), Sugar (local), Sugar (premium), Sugar (white) 8 Legumes Beans, Beans (black East) Beans (black North) Beans (black) Beans (butter) Beans (catarino) Beans (dolichos) Beans (dry) Beans (fava dry) Beans (fava) Beans (green fresh) Beans (green) Beans (haricot red) Beans (haricot white) Beans (haricot) Beans (kidney red) Beans (kidney pinto) Beans (kidney), Beans (magnum), Beans (mung), Beans (niebe white), Beans (niebe), Beans (pod), Beans (red East), Beans (red North), Beans (red), Beans (rosecoco), Beans (spotted), Beans (white East), Beans (white North), Beans (white), Beans (yardlong green), Beans (yellow), Beans(mash), Chickpeas, Cowpea leaves, Cowpeas, Cowpeas (brown), Cowpeas (Red), Cowpeas (white), Cowpeas (whole average), Lentils, Lentils (masur), Lentils (moong), Peas, Peas (fresh), Peas (green dry), Peas (mixed), Peas (split dry), Peas (yellow split), Peas (yellow), Pigeon peas, Pulses (Diamond Masoor Dal), Lentils (broken) 20 Table A5: Concordance between HIT-ITPDE and WFP food prices data (2/3) Code Description WFP Commodity 9 Fruits and vegetables Apples, Apples (dried), Apples (red), Avocados, Avocados (Hass medium size), Bananas, Bananas (imported), Bananas (local), Bananas (medium size), Beetroots, Blackberry, Broccoli, Cabbage, Cabbage (chinese flowering), Carrots, Cashew fruit, Cassava, Cassava (cossette), Cassava (dry), Cassava (fresh), Cassava flour, Cassava leaves, Cassava meal, Cassava meal (attieke), Cassava meal (gari fine), Cassava meal (gari yellow), Cassava meal (gari), Cassava meal (tapioca), Cauliflower (medium size), Coconut, Coconut (dried), Cocoyam (macabo), Cucumbers, Cucumbers (greenhouse), Dates, Eggplants, Garlic, Garlic (medium), Grapes (black), Grapes (pink), Guava, Kale, Leafy vegetables, Lemons, Lemons (Criollo medium size), Lemons (Persa medium size), Lentils (red), Lettuce, Mandarins, Mangoes, Melons (cantaloupe), Naranjilla (hybrid), Okra (dry), Okra (fresh), Onions, Onions (dry), Onions (imported), Onions (red dry), Onions (red imported), Onions (red local), Onions (red), Onions (shallot medium), Onions (shallot), Onions (white dry), Onions (white), Oranges, Oranges (big size), Oranges (Pi˜ na), Oranges (Valencia medium size), Papaya, Passion fruit, Peach (medium size), Pineapples, Plantains, Plantains (apem), Plantains (apentu), Plantains (barraganete green), Plantains (barraganete mature), Plantains (big size), Plantains (dominico green), Plantains (dominico mature), Plantains (medium size), Potatoes, Potatoes (imported), Potatoes (Irish imilla), Potatoes (Irish red), Potatoes (Irish white), Potatoes (Irish), Potatoes (local), Potatoes (red), Potatoes (super chola), Potatoes (unica), Pumpkin, Radish, Spinach, Squashes, Strawberries, Sweet potatoes, Swiss chard, Taro, Tomatoes, Tomatoes (big size), Tomatoes (bitter), Tomatoes (greenhouse), Tomatoes (local), Tomatoes (medium size), Tomatoes (navrongo), Tomatoes (paste), Tree tomatoes, Watermelons, Wax gourd, Yam, Yam (Abuja), Yam (dry), Yam (florido), Yam (flour), Yam (puna), Yam (white), Yam (yellow) 10 Nuts Cashew nut, Groundnuts, Groundnuts (Bambara), Groundnuts (large shelled), Groundnuts (paste), Groundnuts (shelled), Groundnuts (small shelled), Groundnuts (small unshelled), Groundnuts (unshelled), Peanut, Walnuts 11 Eggs/Meat/Dairy Butter, Butter (cow milk), Butter (goat milk), Cheese, Cheese (dry), Cheese (fat), Cheese (local), Cheese (low-fat), Cheese (picon), Cheese (white boiled), Chicken, Eggs, Eggs (broiler), Eggs (duck fermented), Eggs (duck), Ghee (artificial), Ghee (natural), Kefir, Meat, Meat (antelope smoked), Meat (beef canned), Meat (beef chops with bones), Meat (beef first quality), Meat (beef second quality), Meat (beef with bones), Meat (beef without bones), Meat (beef), Meat (camel), Meat (chicken broiler), Meat (chicken fillet), Meat (chicken fresh), Meat (chicken frozen imported) -, Meat (chicken frozen), Meat (chicken local), Meat (chicken whole), Meat (chicken), Meat (gazelle smoked), Meat (goat), Meat (lamb), Meat (mixed sausage), Meat (mutton), Meat (pork first quality), Meat (pork second quality), Meat (pork with fat), Meat (pork), Meat (sheep second quality), Meat (sheep), Milk, Milk (camel fresh), Milk (camel), Milk (condensed), Milk (cow fresh), Milk (cow pasteurized), Milk (fresh), Milk (non-pasteurized), Milk (pasteurized), Milk (powder), Milk (UHT), Poultry, Sour cream, Yogurt 21 Table A5: Concordance between HIT-ITPDE and WFP food prices data (3/3) Code Description WFP Commodity 12 Confectionery/Cocoa Cocoa, Cocoa (powder) 13 Oils/Fats Cooking fat, Fat (salo), Oil, Oil (coconut), Oil (cooking), Oil (cotton), Oil (groundnut), Oil (maize), Oil (mixed), Oil (olive), Oil (palm nut), Oil (palm refined), Oil (palm), Oil (sunflower), Oil (vegetable bulk), Oil (vegetable Himani Best Choice), Oil (vegetable imported), Oil (vegetable local), Oil (vegetable Mahakosh), Oil (vegetable packaged), Oil (vegetable), Oil (mustard), Oil (soybean) 14 Other staple food Bitterball, Bread, Bread (first grade flour), Bread (high grade flour), Bread (khoboz), Bread (pita), Bread (rye), Bread (wheat), Curd, Gari, Gari (white) 15 Beverages, nec Coffee, Coffee (instant), Tea, Tea (black), Tea (green), Water (drinking), Water spinach 16 Cotton Cotton 18 Spices/herbs Chili (bird’s eye green), Chili (bird’s eye red), Chili (bird’s eye), Chili (red curly), Chili (red dry raw), Chili (red large), Chili (red), Jalape˜ no pepper (big size), Jalape˜ no pepper (medium size), Niger seed, Peppers (dried), Peppers (fresh), Peppers (green), Peppers (red dry), Peppers (red), Salt, Salt (iodised), Peppers (sweet) 20 Fish Fish, Fish (appolo), Fish (barbel sole), Fish (bonga), Fish (catfish), Fish (dry katta), Fish (dry sprats), Fish (dry), Fish (fresh silvi), Fish (fresh), Fish (frozen), Fish (goldstripe sardinella), Fish (herring), Fish (jack), Fish (latesdryimported) -, Fish (latesdrylocal) -, Fish (mackerel fresh), Fish (mullet catfish), Fish (omena dry), Fish (sail fish), Fish (sardine canned), Fish (skipjack tuna), Fish (smoked), Fish (snake head dry), Fish (snake head), Fish (striped catfish), Fish (tilapia salted dried) -, Fish (tilapia), Fish (trenched sardinella), Fish (tuna canned), Fish (yellowfin tuna), Panga, Prawn, Shrimps 21 Manufacturing Antibacterial wipes, Antibiotics (imported), Antibiotics (local), Antipyretic (imported), Antipyretic (local), Basin, Batteries, Batteries (big), Batteries (small), Candles (big), Candles (small), Disinfecting solution, Hand sanitizer (gel), Handwash soap, Hoe, Jerrycan (20 L), Jerrycan (5 L), Laundry detergent, Laundry soap, Mug (plastic), Nails, Pen, Pencil, Plate (plastic), Rope, Sanitary pads, Shampoo, Surgical mask, Toothbrush, Toothpaste, Torch, Underwear 23 Fertilizers Fuel (diesel), Fuel (gas), Fuel (kerosene paraffin), Fuel (kerosene), Fuel (LPG), Fuel (petrol), Fuel (petrol-gasoline 92 octane), Fuel (petrol-gasoline 95 octane), Fuel (petrol-gasoline), Fuel (Super Petrol) 22