Who Will Feed China in the 21st Century? Income Growth and Food Demand and Supply in China

This paper uses resource-based cereal equivalent measures to explore the evolution of China's demand and supply for food. Although demand for food calories is probably close to its peak level in China, the ongoing dietary shift to animal-based foods, induced by income growth, is likely to impose considerable pressure on agricultural resources. Estimating the relationship between income growth and food demand with data from a wide range of countries, China's demand growth appears to have been broadly similar to the global trend. On the supply side, output of food depends strongly on the productivity growth associated with income growth and on the country's agricultural land endowment, with China appearing to be an out-performer. The analyses of income-consumption-production dynamics suggest that China's current income level falls in the range where consumption growth outstrips production growth, but that the gap is likely to begin to decline as China's population growth and dietary transition slow down. Continued agricultural productivity growth through further investment in research and development, and expansion in farm size and increased mechanization, as well as sustainable management of agricultural resources, are vital for ensuring that it is primarily China that will feed China in the 21st century.


Policy Research Working Paper 6926
This paper uses resource-based cereal equivalent measures to explore the evolution of China's demand and supply for food. Although demand for food calories is probably close to its peak level in China, the ongoing dietary shift to animal-based foods, induced by income growth, is likely to impose considerable pressure on agricultural resources. Estimating the relationship between income growth and food demand with data from a wide range of countries, China's demand growth appears to have been broadly similar to the global trend. On the supply side, output of food depends strongly on the productivity growth associated with income growth and on the country's agricultural land endowment, with This paper is a product of the Agriculture and Rural Development Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at efukase@worldbank.org or wmartin1@worldbank.org.
China appearing to be an out-performer. The analyses of income-consumption-production dynamics suggest that China's current income level falls in the range where consumption growth outstrips production growth, but that the gap is likely to begin to decline as China's population growth and dietary transition slow down. Continued agricultural productivity growth through further investment in research and development, and expansion in farm size and increased mechanization, as well as sustainable management of agricultural resources, are vital for ensuring that it is primarily China that will feed China in the 21st century. and created a foundation for family-based farming. Since then, China has sustained its output growth, largely benefiting from growing agricultural Research and Development (R&D) investment and high use of key inputs such as fertilizer and irrigation systems. 2 However, in recent years, China's demand for food appears to have been growing more quickly than its supply. As a result, China's trade position for food has turned from surplus to deficit, and this gap has been widening (Fukase and Martin, forthcoming). Some concerns about China's food self-sufficiency have arisen among Chinese policy makers and other stakeholders, especially since China is relatively poorly endowed with agricultural land and water supplies compared to its population base. 3 A World Bank project on urbanization and food security has recently devoted a great deal of research to the question of China's food demand and supply. This research looks in detail at key issues such as the impact of urbanization on China's food self-sufficiency and food security , on land availability (Deng, Huang and Rozelle, 2013) and on water availability (Wang, Huang and Rozelle, 2013). Using detailed structural models built up from estimated parameters of demand systems and production structures for China, Huang et al.'s study (2013) predicts that China will need to import feed grains and some other foods for some time but that its overall food self-sufficiency is likely to remain at above 90 percent level through 2030. Their results are consistent with those obtained from other studies, for instance, the forecast for China by 2022 prepared by OECD/FAO (2013) and a recent study using a multicountry multi-sector applied general equilibrium model (Anderson and Strutt, 2013). 4 The purpose of this paper is to analyze China's income-consumption-production dynamics for food using an entirely different approach to those used in the studies cited aboveeconometric techniques based on data for 154 countries during the period 1980-2009. The paper aggregates both the demand for and the supply of food into resource based cereal equivalents (Yotopoulos, 1985;Rask, 2004, 2011). This approach takes into account one of the central features of food demand behavior-the shift from reliance on direct consumption of grains and other sources of basic carbohydrates into a more diversified diet including edible oils and protein-rich animal products as incomes grow. On the supply side, agricultural output is specified as a function of income and land endowment, with agricultural output growing in response to the productivity growth that is associated with national output growth per person.
The econometric approach used relies on the experiences of a wide range of countries and is intended to complement, rather than replace, more detailed country-specific structural approaches.
Section 2 analyzes the changes that have occurred in dietary patterns in China. Section 3 presents the methodology used for the analysis and implements regression analyses. We first discuss the construction of the cereal equivalent measure of food output and demand. Next, we replicate and extend the income-consumption analysis of Rask (2004, 2011) to the period 1980-2009. Then, we modify Rask and Rask's approach on the production side, specifying a regression model relating both income and land endowment to production. Finally, we suggest the implications of these trends for China's likely net import demands in the future.

Changing Patterns of Chinese Diets
Since China embarked on its market-oriented reforms in 1978, it has achieved dramatic economic growth. China's per capita GDP in PPP 2005 prices, which was $524 in 1980, grew at an average annual growth rate of 8.5 percent to reach $7,958 in 2012 (the WDI, the World Bank). This rapid economic growth appears to have contributed greatly to changes in Chinese diets both in quantity and in composition.  Figure 1a shows that total calorie intake per person per day in China grew substantially, from 2,163 kcal in 1980 to 3,036 kcal in 2009. The decomposition of the source of this change reveals that a majority of the increase comes from a rise in the consumption of animal products, while the calorie intake from crops grew slowly and stabilized at around 2,300 kcal in recent years. China's increase in per capita calorie intake has been much faster than the world average, which grew 2,490 kcal in 1980 to 2,831 kcal in 2009. As of 2009, China's calorie intake was approaching the level in the Republic of Korea, although it remained lower than levels observed in the United States and the European Union (EU) countries. Figure 1a also shows that the total average individual annual calorie intake among high income countries, namely, the United States, Japan and EU countries, declined somewhat in the most recent years. Figure 1b shows that protein intake in China nearly doubled from 54 g per capita in 1980 to 94 g per capita in 2009 and that about three quarters of this growth came from consumption of animal products. Figure 1c shows that fat intake in China nearly tripled from 34 g per capita in 5 1980 to 96 g per capita in 2009, and that about two thirds of this growth came from increases in animal product consumption.
The changing dynamics of food consumption shown above affect supply and demand balances for food directly and indirectly. In particular, whereas direct demand for food grains such as wheat and rice, tends to decrease as income rises, the same driving force (the rise in per capita income) is likely to lead to an increase in indirect demand for feed grains as more grain and other feeds are needed for animal production. Figure 2a shows the self-sufficiency ratios for the key staple foods (rice, wheat, maize and soybeans combined) for the period 1960-2012 for Asian countries. Most strikingly, the self-sufficiency ratios declined sharply in higher income Asian countries, from around 75 to 27 percent for Japan, from about 88 to 21 percent for the Republic of Korea and from about 86 to 13 percent for Taiwan, China, during the period 1960/1961 data, the United States Department of Agriculture (USDA)). 5 Figure 2a also shows that whereas China tended to achieve self-sufficiency for grains for the 1960s, 1970s, 1980s and most of the 1990s, its selfsufficiency ratio has been declining in recent years. Figure 2b reports the evolution of the demand and supply gap by major grains for China.
It is clear that China's recent declining self-sufficiency ratio for grains is predominantly attributable to a large increase in soybean imports. The expansion of the livestock sector which increased the demand for protein meal, along with the rise in consumers' demand for vegetable oils, was a major factor leading to the growing demand. 6 The Chinese government appears to 5 However, further disaggregation of the data reveals that the self-sufficiency ratios for Japan, the Republic of Korea and Taiwan, China, vary by grain: whereas imports of corn contributed most to the widening gap for three countries, they have been relatively self-sufficient in terms of rice. 6 In 2009, out of 59 million tons of domestic soybean consumption, about 9 million tons was consumed as food. 49 million tons of soybean was crushed and made into 9 million tons of vegetable oil and 39 million tons of soymeal (PSD, USDA). 6 have responded to the rising demand by liberalizing soybean imports gradually (Weiming and Ying, 2013). 7 Some scholars view China's increasing imports of soybeans as a rational response to the rising resource constraints in China, especially because soybean is a crop which requires a large amount of land and water (Christiansen, 2012;Qiang, Liu, Cheng, Kastner and Xie, 2013;Weiming and Ying, 2013). For instance, calculating the "virtual" land use embodied in China's imports and exports of crops, Qiang et al. (2013) find that China has become a massive net importer in terms of virtual land during the period 1986-2009 and that the increase in virtual land imports was mainly driven by the rise in imports of soybean. By effectively freeing land, the soybean imports appear to have saved China's domestic cropland area for food grains such as wheat and rice which tend to be regarded as more important for food security objectives (OECD and FAO, 2013;Qiang et al., 2013). In terms of virtual water trade, Chapagain, Hoekstra and Savenije (2006) and Hoekstra and Hung (2005) find that China conserved its national water resources by importing water-intensive agricultural products.

Methodology and Regression Analyses
As incomes grow, consumers diversify their food consumption away from basic food staples.
This process includes a move to include more edible oils, vegetables, fruit and animal products.
Per capita consumption of staple foods declines during this process. Historically, the urbanization process that is inextricably linked with income growth appears to have reduced per capita food consumption slightly by reducing energy needs (Clark, Huberman and Lindert, 1995). In Asia, it also appears to have increased demand for wheat relative to rice (Huang and 7 China adopted a more liberal trade scheme for soybeans in 1996 which was locked in through negotiations to access the World Trade Organization (WTO) (Weiming and Ying, 2013). China became a full member of the WTO in 2001. David, 1993). However, it now appears that the key driving force behind changes in per capita food consumption is changes in real incomes-the same increases in real incomes that drive the urbanization process (Satterthwaite, McGranahan and Tacoli, 2010).

Cereal Equivalent (CE) Measures of Food
Some scholars argue that the dietary shift from crop based to animal based products may increase total food demand sharply relative to supply, due to the inefficient conversion of plant based feeds (typically cereals 8 ) into animal based foods. 9 Yotopoulos (1985) argues that the supply of cereals available for food may decline as developed and middle-income countries consume cereals disproportionately as feed, raising world prices of cereals. He suggested that this "Food-Feed Competition" may have contributed to the world food crisis of 1972-74. More recently, a number of scholars have explored the implications of this dietary shift on agricultural resources and the environment (e.g., Elferink and Nonhebel, 2007;Garnet, 2009;Gerbens-Leenes, Nonhebel and Ivens, 2002ab;Gerber, Steinfeld, Henderson, Mottet, Opio, Dijkman, Falcucci and Tempio, 2013;Steinfeld et al. 2006;Williams, Audsley and Sandars, 2006;Wirsenius, 2003;Wirsenius, Azar and Berndes, 2010). For instance, Wirsenius (2003) argues that the idea of competition for grains between animals and humans does not capture fully the resource implications of dietary change, since the core issue is the competition for land rather than competition for consumption of cereals. Analyzing comprehensively total feed 10 and land requirements in their land-minimizing model, Wirsenius et al. (2010) suggest that greater feed-to-food efficiency in animal production, decreased food wastage and dietary changes towards less land-demanding foods would help to reduce agricultural land use.
Animal production competes for land directly, for instance, for grazing and fattening, and indirectly through the need to produce animal feeds. 11 Gerbens-Leenes et al. (2002ab) show that foods associated with affluent lifestyles, especially animal products, oils and fats, and beverages, tend to require more land for their production than foods associated with less affluent lifestyles. For China, following Gerbens-Leenes et al. (2002ab)'s methodology, Li et al. (2013) find that China's urbanization appears to have increased pressure on limited arable land resources, since urban residents consume more animal based and other land-demanding food than their rural counterparts.
The approach adopted in this paper is based on the methodology developed by Rask (2004, 2011) which converts crop and animal products into cereal equivalents. The CE coefficients for crop-based products are computed very simply by matching their caloric content to those of an equal weight of cereals, assuming broadly similar efficiencies across commodities but taking into account the greater resource use associated with producing foods that contain more calories per unit of weight (e.g., vegetable oils) relative to those that have, for instance, a higher water content (e.g., starchy roots). For animal products, the CE coefficients reflect the feedstuff used to produce one unit of animal products in terms of the dietary energy equivalent of a unit of corn, considering not only grains consumed but also other types of feed such as protein supplements, forages (including pasture) and other feeds. 12 11 Livestock is the world's largest user of land resources, with grazing land and cropland dedicated to the production of feed crops and fodder representing about 70 percent of all agricultural land. About 33 percent of arable land is used to produce livestock feed (Steinfeld et al., 2006). 12 The CE coefficients developed by Rask (2004, 2011) are based on a study published by the USDA (1975). The USDA study is unique in covering all types of feeds including forage crops and pasture in calorie equivalents of corn over the period 1964-1973. Wirsenius (2003 challenges the view that the use of non-edible feeds is "free" arguing that it involves opportunity costs such as production of feedstock for biofuels, preservation of Table 1 shows the CE coefficients used to convert crop and animal products into cereal equivalents. The coefficients in Table 1 reflect the high resource costs of producing animal products relative to cereals, and illustrate the great differences among animal products. The CE coefficient of 19.8 for carcass beef, for instance, takes into account the large amount of feed used directly to produce beef; the relatively low dressing weight percentage for live cattle (0.59); and feed for breeding cows and young calves needed to supply production animals and replacement breeding stock. 13 Pork, poultry and fish are more efficient both because of generally higher feeding efficiencies and the lower costs involved in maintaining their breeding stock. Within crop products, CE coefficients range from 2.7 for vegetable oils to 0.07 for vegetables.
The magnitudes of the CE coefficients appear to be broadly consistent with other estimates of land requirements in the literature (e.g., Gerbens-Leenes et al., 2002ab; Williams et favorable soil condition and restoration of ecosystems and habitats. Wirsenius et al. (2010) show that the global use of the non-edible type feeds is substantial. 13 Using beef as an example, Rask and Rask (2011) use a feed conversion ratio of 11.7 for producing live cattle, taking into account feed for both animals slaughtered and for breeding animals. This figure is converted to carcass weight using a dressing weight percentage of 59 percent, which gives rise to the final value of 19.8 (11.7/0.59). Carcass weight is used in order to conform to FAOSTAT meat consumption coefficients which are presented in carcass weight format (e-mail communication with Norman Rask). Wirsenius, 2003Wirsenius, , 2010. For instance, using data for the Netherlands in 1990, Gerbens- Leenes et al. (2002a) estimate the land requirement for beef to be 20.9 m 2 year per kilogram of meat which is more than twice that for pork (8.9 m 2 year per kilogram of meat) whereas their land requirement estimate for cereals turns out to be relatively small (1.4 m 2 year per kilogram).
For China, land requirement estimates obtained by Li et al. (2013) and Zhen et al. (2010) are broadly comparable with those from other studies. 14 The implications for environmental burdens implied by the CE coefficients 15 are also generally in line with the findings of other studies. 16 Researchers tend to find that beef production has the most severe GHG impact per kilogram of meat, followed by pork and chicken production (e.g., Fiala, 2008;Gerber et al., 2013;Steinfeld et al., 2006).
Finally, the cereal equivalent measure used in this study does not consider varying feed requirements for animal production depending on technology (e.g., feed mix and efficiencies), production systems (Robinson, Thornton, Franceschini, Kruska, Chiozza, Notenbaert and You, 2011) and local resource availabilities. Differentiating CE coefficients by regions and by production systems would be a potential subject of future research. Further, our analytical technique in the empirical section does not take into account distortions from agricultural incentives, food price policy, or farm gate pricing differences related to product self-sufficiency.
It would be desirable to do this in future work, but we believe that the impacts on the aggregate measures that we consider are likely to be less than protection rates might suggest. Countries with high average rates of agricultural protection typically provide high rates of protection on traditional staple foods such as rice. But political economy pressures generally keep the rate of protection on feedgrains quite low and hence keep the prices of domestically-produced livestock products like pork and poultry low relative to staple grains. This structure of protection results in an incentive for consumers to increase their consumption of livestock products, thus accelerating the dietary transition that is the focus of this study. The increase in China's food consumption at the national level is attributable to both population growth and diet upgrading. Figure 3b shows that China's population increased from 1.0 billion in 1980 to 1.4 billion in 2009 and that its population growth is expected to taper off gradually. Figure 3c decomposes the change in CE consumption into the components of population growth 17 and of diet changes since 1980. The figure shows that about one-third of the increase in food consumption is attributable to China's population growth, and the remaining two-thirds can be explained by the change in diets. As China's population growth is slowing and its total population is projected to peak around the year 2025 (at a level about 2.3 percent higher than in 2014) (FAOSTAT), the primary driver of food consumption increases is likely to be change in diet, and therefore change in per capita consumption.

Income-Consumption Relationship: Regression Analyses
While we can observe the rapid growth in China's consumption of food over the period since 1980, this gives us little insight into the way this growth is likely to play out in the future. To gain some insights into this, we turned to econometric analysis using a large sample of countries.
This allows us to view a much larger range of real incomes and to obtain a better idea of the extent to which the growth of food consumption in cereal equivalents begins to decelerate. This relationship includes most importantly the effects of income growth on the demand for basic food staples and for foods with relatively high income elasticities but also other influences on demand such as changes in the rate of assistance to agriculture with income growth (Anderson, 1995).
We estimate the CE consumption-income relationship using the functional form used in Rask (2004, 2011). Specifically, where y is CE consumption per capita and x is PPP GDP per capita in 2005 constant prices. As ƒ' ˃0, ƒ''˂0, this functional form captures the observed pattern of the change in CE consumption, which rises more rapidly at early stages of development and tapers off at higher levels of income.
It implies that, as incomes continue to increase, consumption asymptotically approaches a limit  Figure 4 along with actual CE consumptions for the sample countries, including data for China in red. The second column in Table 2 reports an alternative regression using all the available data points for each year during the period 1980-2009. In this regression, the standard errors are adjusted for within-country correlation (clustering). The results turn out to be similar between the two regressions.  (2) shows the results using all the available data points for each year during the period 1980-2009. The standard errors in parentheses are based on heteroskedasticity-consistent estimates of the variance-covariance matrix and corrected for within country correlation (clustering).
The graph presented in Figure 4 shows a concave relationship between food consumption in cereal equivalents and real income levels Rask, 2004, 2011 shows the income-consumption pairs for a large range of countries, making clear that there is considerable variation around this broad trend. This is to be expected with such a simple measure, given the differences in food consumption patterns among nations, some of which may be due to inherent cultural features, with others perhaps related to habit formation patterns of the type analyzed by Atkin (2013), or to food price differences related to policy and product selfsufficiency.
Based on the FAO statistics that we use, China's consumption pattern is close to the average demand pattern for our global sample. Not surprisingly, consumption levels are particularly high in countries such as Australia and Brazil, where beef and sheepmeats are produced largely using pasture whose price is determined-given the available technology-by the prices of these beef and sheep products, rather than by arbitrage between pasture and grain.
Japan is a negative outlier, probably because of the importance of seafood in its historical animal-product diet with the price of fish determined-at least until recently-by costs of catching from the wild, rather than by the costs associated with fish-farming. Only recently has arbitrage between wild-caught and farm-raised seafood justified the use of our approach to aggregation on the assumption that fish can be produced using cereal products.
In order to gain historical perspective, Figure 5 shows the changes in CE consumption between the beginning of the sample period (the year 1980) and the end (the year 2009). This shows that most countries where per capita income grew substantially observed a sizeable increase in consumption levels, with the slope of the resulting ray being very broadly similar to that of the estimated equation in the range relevant to the income growth of that country.
Contrasting cases, such as the decline in observed consumption levels in Australia, appear to reflect structural shifts in demand away from meats such as beef and sheepmeats that rely on relatively inefficient conversion processes into more efficiently-produced livestock products like poultry (Martin and Porter, 1985). The drop of CE consumption for Hungary appears to reflect the transition from a high level diet, which was induced by low food prices and production subsidies under a centrally planned low food price system, to a food price level more consistent with market economies (Rask and Rask, 2004).  Figure 6 shows two results that are significant for our analysis. First, consumption of calories tends to level off much earlier and at a much lower level than consumption of cereal equivalents. Second, China's per capita consumption levels for both calories and cereal equivalents have been closely consistent with global trends.

Income, Productivity and Land Endowments
Agricultural output tends to rise as real income rises (see Figure 9 below). The primary driving force for this relationship is the increase in productivity that contributes to increases in national incomes. This relationship is, however, influenced by several other factors, including: (i) the shift in demand away from staple foods, as discussed in the demand section, that influences the prices of non-traded or incompletely-traded foods and hence the incentives for their production, (ii) differentials in the rates of productivity growth between agriculture and other sectors (Martin and Mitra, 2001), and (iii) Rybczynski effects when high (or low) rates of capital accumulation change factor endowments (Martin and Warr, 1993) or when land use changes alter agricultural land endowments. Since the growth rate of the agricultural sector is almost invariably slower than that of the economy as a whole for most countries, we would generally expect a given percentage change in GDP to result in a less-than-proportional increase in agricultural output.
In order to see if the data support a positive association between income and agricultural productivity, Figures 7a and 7b plot the relationship between GDP and proxies for land and labor productivity. The income-productivity pairs for China are shown in red. Figure 7a shows the relationship between GDP PPP per capita and the average cereal yield per hectare as a proxy of land productivity. Figure 7a confirms that income level and land productivity are positively associated and that China achieved high land productivity relative to its current level of income, almost approaching the productivity level reached by high performing Organization for Economic Cooperation and Development (OECD) countries. This impressive achievement is likely to reflect a number of factors, for instance, a high degree of fertilizer use, expansion of irrigated land, 18 widespread use of multiple-cropping and the introduction of new seed varieties and other technology improvements (Huang and Rozelle, 1996). Figure 7b plots the relationship between real GDP per capita and agricultural value added per worker as an indicator of labor productivity. Not surprisingly, higher income is generally associated with higher agricultural labor productivity. However, in contrast to China's high land productivity, its labor productivity is found to be very low given its level of development.
China's low labor productivity may possibly be attributable to small farm size (0.6 hectare on average), land fragmentation (Jia and Petrick, 2013), and to the labor-intensive nature of family based farming. In addition, several scholars report that the labor productivity gap between farm and non-farm sectors remains high in China (e.g., Fan, Zhang and Robinson, 2003;Kujis and Wang, 2006). Labor is likely to move out of agriculture and this shift is an inherent part of the

Relating Income, Land Endowment and Production: Regression Analyses
We use a regression model to explain agricultural output using land endowment and GDP per capita. The particular specification that we use is: where z is CE production per capita, x is PPP GDP per capita in 2005 constant prices, l is hectares of land equivalent per capita. B 0 is intended to capture a subsistence level of agricultural production, assuming that people produce some food from local resources even when their per capita GDP levels are very close to zero. As the purpose of the exercise is to evaluate the demand and supply for "food", CE production per capita reflects "net" production; it is calculated subtracting from (gross) production (FAO, 2001) the use of agricultural output for feed, seed, food manufacture use, other uses and waste. Thus, the difference between food consumption and (net) food production reflects imports, exports and changes in stock.
Column 1 in Table 3 reports a cross-section regression result using a 5-year average of CE production and GDP for the period 2005-2009. Removing some outliers on the production side results in a sample of 140 countries. 19 In parallel with the CE consumption side (Table 2), Column 2 in Table 3 shows the results using the data points available in our sample for the period 1980-2009 (regression (2)). The parameter estimates turn out to be reasonably similar between the two regressions. 19 It is noted that our results are somewhat sensitive to the exclusion of the outliers on the production side. If, for instance, we run regressions without excluding outliers, the results turn out to be: B o = 0.27**(0.11), B 1 = 8.9 × 10 -4 (1.5 × 10 -3 ), B 2 = 0.77***(0.16) and B 3 = 0.33***(0.036) (R 2 = 0.56, Observations = 154) for regression (1); and : B o = 0.22**(0.10), B 1 = 1.7 × 10 -3 (2.1 × 10 -3 ), B 2 = 0.71***(0.12) and B 3 = 0.31***(0.030) (R 2 = 0.57, Observations = 4100) for regression (2). However, qualitative results remain essentially unchanged with or without outliers. Based on the parameter values reported in column 1 in Table 3, Figure 9 shows the estimated relationship between income levels and cereal equivalent production. To allow comparison in two dimensions, the CE production schedule for each country is adjusted so that it has the same land endowment as China (0.21 hectare per capita as an average of the period 1980-2009 per person). 20 The estimated CE production curve is visually close to linear: it rises in line with income, although less rapidly than income because of the secular decline in agriculture's share of national income. From Figure 9, it appears that China has been an out-performer in terms of output. Agricultural output, which is slightly below the consumption level, is  Huang, Hu and Rozelle, 2002). For instance, a study by the IFPRI (2012) documents that more than one-third of the increase in global public agricultural R&D spending between 2000 and 2008 was attributable to China. 21 However, there have been some concerns expressed about measurement problems with China's livestock production, an issue that is discussed in more detail in the Appendix.

Supply and Demand Balance for Food
Figures 10a-c compare how the historical patterns of CE production and consumption differ depending on land endowments, translating the differences in endowments into differences in the income response curves based on the parameter results reported in Table 3 (regression (2)).  Figure 10a demonstrates the differing growth patterns of CE production and consumption. At early stages of development, demand for food tends to grow faster than production, widening the gap between supply and demand. As incomes grow, the growth of consumption will slow down relative to growth of production and the gap will begin to close. Figure 10a demonstrates that, at the onset of the reform, China's CE food production and consumption grew together, albeit from a very low level, at a much faster rate than the global trend, most likely reflecting the impacts of institutional reforms (Rozelle and Swinnen, 2004   where the growth of CE consumption has stabilized, although production growth seems to have been below what might have been expected. While productivity growth contributed to output growth in the United States, a shift of resources out of agriculture may have offset output growth, making the United States a relatively stable leading exporter. 22 In Brazil, production growth has been substantially greater than might have been expected-a factor that appears to be increasing Brazil's exports. Figure 10c reports estimated CE production lines evaluated at the Republic of Korea's and Japan's land endowment levels, 0.05 hectare and 0.04 hectare per person respectively, along with those two countries' actual CE consumption and production points for the period 1980-2009. In contrast to Figure 10b, their estimated CE production lines are below the CE consumption line throughout the period, showing that countries with scarce land endowments tend to be food importers throughout all income levels. In Japan and the Republic of Korea, both demand and supply growth appear to be relatively slow, with the slow growth rate of supply relative to overall income growth contributing to strong net import demand ( Figure   10c).
In order to gain insight into how China's consumption and production gap is likely to evolve in the future, we conduct some simulations with hypothetical scenarios. In scenario 1, we start with Figure 10a in which China's actual CE consumption and production as well as estimated CE consumption and production (adjusted to reflect China's average land endowment of 0.21 hectare) trends are shown. Then, we assume that the small gap between China's food consumption and the production from the model is due to factors-such as acquired tastes-that are likely to be time-invariant, and treat the residual from current levels as sustained and so shift the curve accordingly. In the same way, we assume that China's outperformance on the production side is sustained, and correspondingly shift the supply curve to remove this residual.
The resulting diagram is shown in Figure 11a. This figure shows that the slope of CE consumption and that of CE production evaluated at China's land endowment are comparable at around $16,350 PPP GDP. Thus, the gap between China's supply and demand for food may continue to grow slightly as China's income per capita rises from its current level to around $16,350. Above that level, it seems likely, according to this scenario, that the growth of consumption will slow down relative to production growth and the gap begins to decline.
The results of scenario 11a depend on an assumption that China's agricultural resource endowment remains the same. However, as China's economy develops and urbanization proceeds, it is likely that China would experience some loss in its cultivated land both in terms of quantity and quality (Deang et al., 2013;World Bank and DRC, 2014). Figure 11b shows the result of a scenario in which we assume that China's agricultural land areas and land bioproductivity (agricultural production potential) continues to decline at the rate found by Deng et al. (2013), 0.47 percent and 1.68 percent over the period 2000-2008, respectively (scenario 2). 23 Specifically, it is assumed that China loses its "effective" land at 0.27 percent annually (combination of land area and bio-productivity loss), given projected population growth and economic growth rates taken from Huang et al. (2013). In this scenario, as the CE production at each GDP level is evaluated at the projected level of China's land endowment, the estimated CE production line in Figure 11b becomes flatter than that in Figure 11a. As a result, the changes in China's CE consumption and production turn out to be comparable when China's income reaches around $18,500 in PPP terms per person. Until China reaches that point, the gap between supply and demand increases more under scenario 2 than under scenario 1. The results of these scenarios highlight that sustainable land management appears to be a key determinant to ensure reasonable supply and demand balance for food in the future.
These are, of course, only hypothetical scenarios. If, for instance, China's demand for food were to stabilize at a lower level than is assumed in these scenarios, then the gap between supply and demand might start to close at an earlier time. On the supply side, if China were to reduce its investments in agricultural productivity, or if climate change were to decrease its productivity, or if China were to lose agricultural land at a faster rate than that under scenario 2, the gap might increase further. Nevertheless, China is in a very different situation from a country such as Japan or the Republic of Korea, where the much smaller land endowments almost ensure that continuing large net imports of food will be required.
One important caveat of this study is that the data used for the analyses rely on FAOSTAT data which in turn are based on official statistics. In particular, several scholars point out that the livestock data for China are flawed (Fuller, Hayes and Smith, 2000;Ma, Huang and Rozelle, 2004). Although we are aware of this potential bias in FAOSTAT data, it turns out to be impossible for us to replace FAOSTAT data with revised data due to the lack of comparable data for other countries. We therefore deal with this issue by conducting a sensitivity analysis in the Appendix.

Some Policy Challenges
China's ongoing shift in demand into more affluent, and particularly animal-based, foods is likely to impose substantial pressure on agricultural resources and environment. In particular, animal production puts pressure on land both directly 24 and indirectly through feedstuff production. Cropland area tends to decrease as it competes for space with urban and industrial uses. Since income growth is associated with both dietary upgrading and economic activities which generate income, economic development is likely to intensify the competition for land. In addition, the quality of China's cultivated land is reportedly deteriorating due to soil degradation, pollution and desertification (Chen, 2007;OECD and FAO, 2012;Ye and Ranst, 2009). Climate change adds another challenge, as the overall impact of climate change on agricultural productivity is likely to be negative (Ju, van der Velde, Lin, Xiong and Li, 2013).
Given increasingly tight resource constraints, the evolution of China's net import demand for food depends heavily on its productivity growth in agriculture. This is especially so, as China 24 About 20 percent of the world's pastures and rangelands are degraded to some extent mainly through overgrazing. In China, the shift of production towards a large-scale grain-based industrial system appears to be leading to nutrient overload of soils and water pollution in some geographically concentrated areas (Steinfeld et al., 2006). 25 continues to develop from an upper-middle to high income country. 25 The experiences of high income countries reveal that their agricultural output growth tends to rely increasingly on TFP growth rather than input growth (Fuglie, 2012). 26 The increase in productivity would also generally have the desirable effect of increasing the incomes of farmers. In particular, it has very powerful poverty-reducing impacts because so many of the poor-especially in China-live in rural areas and depend on agriculture for their livelihoods (Christiaensen, Demery and Kuhl,2011;Christiaensen, 2012).
Continued investment in R&D is likely to be a key factor for sustained growth of agricultural output in China. However, as China's land productivity has already attained a high level, there is a possibility that further intensification of the use of cropland would result in diminishing returns and environmental degradation (Brown, 1994;Wirsenius, 2010).
Technological developments beyond the focus on yield growth, in particular, to explore sustainable management of natural resources and to address environmental concerns, seem likely to become increasingly important. In contrast to China's high land productivity, the labor productivity of farmers in China remains low given China's development level and relative to other sectors of its economy. The shift of labor out of agriculture is likely to continue, imposing challenges on China's current labor-intensive, family based agricultural production system.
Some scholars point out that farm size in China is too small to reap economies of scale necessary for domestic production to satisfy domestic demand (Otsuka, 2013). As China's comparative advantage has been shifting from the farm to the non-farm sector and the 25 According to the World Bank's income classification, China moved up from "low" to "lower middle income" status in 1996 and further advanced to "upper middle income" status in 2010 (http://data.worldbank.org/about/country-classifications). 26 Decomposing agricultural output growth into contributions from inputs and TFP for the period 1960period -2009period , Fuglie (2012 shows that developed countries as a whole have relied increasingly on TFP growth to keep output from falling. Total agricultural inputs for developed countries as a group have been declining since the 1980s (Fuglie, 2012,  opportunity cost of farm labor is rising, farm sizes need to be expanded, because the substitution of large machines for labor requires scale economies (Otsuka, 2013;Yamauchi, 2014).
Promoting scale economies and mechanization is likely to involve a number of changes, including development of land rental markets (Deininger and Jin, 2005;Zhang, Qingguo and Xu, 2004), institutional development such as machine rentals (Takahashi and Otsuka, 2009) as well as removal of restrictions on labor movement from farm to non-farm sectors (Fan et al., 2003).
About one-third of the world's farmers are still in China (Christiaensen, 2012). How to promote farm size expansion while ensuring the well-being of smallholders is an important challenge which requires future research. For instance, using farm panel data from Indonesia, Yamauchi (2014) finds that while relatively large farmers tended to increase the scale of operation by substituting machines for labor and by renting more land, such a dynamic change was not observed among relatively small holders.
In principle, a gap between the demand and the supply for food could be diminished by a protection policy that raises the price of all agricultural products relative to non-agricultural products. However, this would have the undesirable effect of reducing the access of some poor people to the staple foods that they need. In any event, as shown in Figures 12a and 12b, it appears that per capita demand for key food staples, such as rice and wheat, is now declining quite sharply in both urban and rural settings. This suggests that it is unlikely that the gap between food demand and supply in China will manifest itself as large net imports of these staples. Given this, there does not seem to be a self-sufficiency argument for increased protection of these staple foods.
Protection for the feedstuffs demanded by China's rapidly growing livestock production sectors could reduce demand for these products, but would do so at the risk of hindering the development of a modern livestock sector. Given China's land constraints, such a policy, if pursued strongly, could create a demand for imports of staple products by taking land out of staple crops and potentially creating self-sufficiency concerns despite declining consumption of these products.

Conclusions
This paper explored the evolution of China's demand and supply for food using resource based cereal equivalent measures of the type proposed by Yotopoulos (1985) and extended by Rask (2004, 2011). We note that, while demand for food calories has probably come close to its peak level in China, the ongoing shift in demand into high-protein, and particularly animal based, foods induced by income growth, is likely to require a considerably greater agricultural effort than would continuation of past demand patterns. Using the experience of a wide range of countries, we find that China's demand pattern-and the growth of that demand in terms of cereal equivalents-is broadly similar to the international average.
On the supply side, we find that output of agricultural products in terms of their cereal equivalents depends strongly on both the growth of income and on the country's endowment of agricultural land. Countries with much larger land endowments per person than China-that is countries such as Brazil and the United States -tend to be exporters over a wide range of income levels. By contrast, economies with much more limited land endowments than China'seconomies such as Japan and the Republic of Korea-tend to become net food importers at a relatively low income level and to remain net importers. China is a relatively land scarce country with a per capita land endowment measured at about one-half of the world average. It appears that China-probably because of the high productivity of much of its agricultural land and its heavy investments in agricultural research and development-has produced much more food than would be expected given its income level and land endowment.
The analyses of income-consumption-production dynamics suggest that China's current income level appears to fall in the range where consumption growth outstrips production growth, widening its supply and demand gap for food. If China's past outperformance can be maintained, then it seems likely that, although China's net imports of food will rise from current levels for a while, the gap will begin to decline as China's population growth and dietary transition slow down. In the meantime, the quantity of agricultural resources that will be needed to feed the Chinese population is likely to continue to increase. In particular, our simulation exercises show that the evolution of the supply and demand gap for food depends on the changes in China's agricultural land availability both in terms of quantity and quality.
As China progresses from an upper-middle to a high income country, some loss of agricultural land is likely and a large shift of labor out of agriculture is inevitable. The experiences of developed countries reveal that they increasingly rely on agricultural productivity growth in sustaining their agricultural output while their input growth rates tend to decline (Fuglie, 2007;. In conclusion, we suggest that continued agricultural productivity growth through further investment in R&D, and through expansion in farm size and increased mechanization, as well as sustainable management of agricultural resources, appear to be critical for raising farm incomes and increasing food supplies to ensure that it is primarily China that will feed China in the 21 st century. Notes: Total for rice, wheat, maize and soybeans. Self-sufficiency is measured by dividing (gross) production by (gross) domestic consumption excluding stock changes. Gross production and consumption include uses for non-food purposes such as feed use.  1960/1961 1963/1964 1966/1967 1969/1970 1972/1973 1975/1976 1978/1979 1981/1982 1984/1985 1987/1988 1990/1991 1993/1994 1996/1997 1999/2000 2002/2003 2005/2006 2008     C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a C hin a F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce F ra nce   Table 2. The fitted calorie consumption line (left-hand side) is based on the parameter values obtained using the same functional form as the CE consumption estimate (equation (1)). During the period 2005-2009, the world average per capita cereal consumption was 0.403 kilogram per day (.147 ton per year) which contained 1,296 kcal per day on average. Thus, one ton of CE per capita per year is equivalent to about 8800 kcal per capita per day (1,296/.147≈8800). 1 0 0 0 0 2 0 0 0 0 3 0 0 00 4 0 0 00 5 0 0 00 R e a l G D P (P P P ) p e r c ap ita in 2 0 0 5 in t. $ 1 9 8 0 -2 0 0 9 C h ina C alorie C ons um pti on F itted C alorie C ons um ption C h ina C E C onsum p tion F itted C E C ons um ption     1980-1984 1985-1989 1990-1994 1995-1999 2000-2004 2005-2009 1980-2009 (2011), hectares of agricultural land per capita is computed as a summation of arable land, land in permanent crops, and one-third of land in permanent pasture. *1 The data for EU reflect the averages of member countries. Thus, the evolution of EU data reflects partly the composition of member countries.   1 0 0 00 2 0 0 0 0 3 0 0 0 0 4 00 00 5 0 0 0 0 G D P p e r c a p ita , P P P 2 00 5 in t. $ C h in a C E C on sum ption C h in a C E P rodu ction C E P rodu ction ad ju sted a t C hin a La nd Le vel C E P rod uction , la nd = .2 1 ha (C hin a)

Appendix: Some Sensitivity Analysis: CE Production and CE Consumption Using Alternative Livestock Data
Several scholars have argued that the Food Balance Sheets (FBSs) data for China from FAOSTAT, which are based on official Chinese statistics, are biased. In particular, livestock production is widely believed to be over reported introducing potentially serious bias into food supply and demand balance estimates (e.g., Fuller, Hayes and Smith, 2000;Ma, Huang and Rozelle, 2004). Ma et al. (2004) provide revised series on production and consumption for livestock products for China over the period 1980 to 1999. Their results on livestock production are lower than the official numbers up to 1999, although they still show rapid growth with, for instance, a tripling of pork production over the 1980 to 1999 period, rather than an increase of three and a half times. To our knowledge, after 1999, revised livestock production and consumption data are available only for 2000, 2010 and 2012 from the China Agricultural Policy Simulation Model (CAPSiM) .
To see how sensitive our results might be to the use of alternative data, we repeat scenario 1 (Figure 11a), replacing 2009 livestock data (which is the latest year available in FBSs) with the 2010 livestock production and consumption data from the CAPSiM model . The results suggest that China's CE production point is close to the global trend, rather than substantially above the trend, while its consumption point turns out to be below the global trend. 27 Following the assumptions in scenario 1, we shift both CE production and consumption curves so that they cross China's alternative CE production and consumption points. The shifted lines suggest that China's current income level lies in the range where consumption growth outpaces production growth, but that the gap will likely start to decrease as China's consumption growth rate declines. However, evaluating the bias of FBSs data is not within the scope of this study and further studies are required to address the data quality issues.

Figure A CE Production and Consumption
With Alternative Livestock data 27 Since the two datasets are not directly comparable, the difference in the results may be influenced by the differences in characteristics of the datasets. For instance, Kearney (2010) suggests that the FBSs data generally tend to overestimate consumption since they are referring to "available" rather than "actual" food consumption. Chi na C E produc tion (al terna tiv e data) S hifted CE production S hifted CE cons ummption