DEPARTMENT OF ECONOMICS YALE UNIVERSITY P.O. Box 208268 New Haven, CT 06520-8268 http://www.econ.yale.edu/ Economics Department Working Paper No. 80 Economic Growth Center Discussion Paper No. 986 Trade, Tastes and Nutrition in India David Atkin Yale University April 2010 This paper can be downloaded without charge from the Social Science Research Network Electronic Paper Collection: http://ssrn.com/abstract=1596827 Trade, Tastes and Nutrition in India* David Atkin** March 2010 Abstract This paper introduces habit formation into an otherwise standard model of international trade. Household tastes evolve over time to favor foods consumed as a child. The opening of trade causes preferred goods to rise in price, as these were relatively inexpensive in autarky. Neglecting the correlation between tastes and agro-climatic endowments overstates the short-run nutritional gains from agricultural trade liberalization and masks potential caloric losses for laborers. I examine the predictions of this model of trade with habit formation using household survey data from India, both by looking across Indian regions and by examining the consumption patterns of inter-state migrants. JEL Codes: O10, O12, Q17, F10 Keywords: International Trade, Habit Formation, India, Tastes, Nutrition *Thanks to Richard Chiburis, Angus Deaton, Dave Donaldson, Adrian de Froment, Scott Fulford, Penny Goldberg, Marco Gonzalez-Navarro, Gene Grossman, Gordon Hanson, Giovanni Maggi, Chris Paxson, Jesse Rothstein, Sam Schulhofer-Wohl and seminar participants at Princeton, Duke, UCSD, Kellogg M&S, Yale, MIT, LSE, Stanford, Chicago Booth, Brown, UCLA Anderson, CREI, Munich, ULB, Oxford, the 2009 SITE at Stanford, Columbia and the Federal Reserve Bank of Minneapolis. Financial aid from the Fellowship of Woodrow Wilson Scholars and the International Economics Section at Princeton is gratefully acknowledged. The Center for Economic Policy Studies at Princeton is thanked for purchasing the Indian Harvest Data. **Department of Economics, Yale University, 27 Hillhouse Avenue, New Haven, CT; email: [email protected]. 1 Introduction Theimpactsofagriculturaltradeonthedevelopingworldareofgreatinteresttoeconomists and policymakers alike. For example, the Doha round of global trade negotiations aims to both substantially liberalize agricultural trade and ensure that trade brings greater benefits to developing countries. At the same time, surging demand from emerging markets has pushed up the traded prices of many food crops. Even in food-exporting countries, policy- makershaveraisedconcernaboutthepotentiallyadverseeffectsofthesepricechangesoncer- tainconsumers. Inthispaper,Iexploreanewchannelwhichprovidesamorecompleteunder- standing of the nutritional impacts of agricultural trade on the poor through examining the roleoftastesininternationaltrade,asubjectthathasbeenneglectedintheliteraturetodate. Standard international trade theory typically focuses on the production side, rather than the consumption side, and assumes that preferences are identical across regions and inde- pendent of resource endowments.1 This paper explores the consequences of relaxing this assumption in a very natural way. I incorporate habit formation into an overlapping genera- tions model of trade and demonstrate that this leads to regional food tastes that favor crops relatively well-suited to local agro-climatic endowments. This correlation between tastes and endowments systematically reduces the short-run nutritional gains from trade compared to models without habit formation. Using household consumption data from regions within India, I provide two sets of empirical evidence that support the model’s predictions and I quantify the extent to which regional taste differences will alter the nutritional impact of trade if India were to liberalize its highly restricted internal agricultural markets. In particular, I define tastes for food as the component of the food budget share that cannot be explained by the vector of prices or total food expenditure in a demand system. Based on extensive evidence in the nutrition literature, surveyed in Birch (1999), I assume that adult tastes favor the foods consumed as a child and term this process habit formation. 1International trade textbooks mention that taste differences can be one source of comparative advan- tage, however the literature dealing with cross-country taste differences only explores the consequences of non-homothetic preferences, for example Hunter and Markusen (1987) or more recently Fieler (2009). 1 While large literatures explore the implications of habit formation for demand, asset pricing, monetary policy, growth and addiction, this paper is, as far as I am aware, the first explo- ration of how habit formation in consumption alters standard models of international trade. The first generation of adults, who value only calories and dietary variety, purchase large quantities of a region’s relatively abundant (comparative advantage) agricultural goods, as these foods are relatively cheap under autarky. Their children are fed these locally abun- dant foods, and develop particular tastes for them in adulthood. Over many generations, a home bias in household consumption emerges endogenously through habit formation. The same affinities for local foods emerge if recipes and preparation techniques improve faster for commonly consumed foods. Households will choose to purchase the familiar local foods that they know how to transform into high-quality meals. At the time of trade liberalization, the preferred foods rise in price in each region as these foods were relatively inexpensive in autarky, and trade equalizes prices across regions. Con- sumersspendalargeportionoftheirincomesonthesefavoredfoodsandarereluctanttosub- stitute out of them and into less familiar imports, which reduces the consumption gains from trade compared to a model without habit formation. Only decades after trade liberalization canconsumersrealizethefullcaloricgainsfromtrade, asfoodtastesgraduallyadapt, eventu- allyresultinginevenlargercaloricgainsthanamodelwithouthabitformationwouldpredict. Theaggregateproductiongainsfromtradealsoshrinkwithhabitformationbecausehabits reduce the autarky relative price differences between regions.2 As in a standard trade model, these reduced production gains will be distributed unevenly. If labor is mobile and combined withcrop-specificlandtoproduceafood, thespecificfactorsmodelimpliesthatlabor’snom- inal wage gains from trade will be strictly smaller than the price rise in the locally abundant food. Therefore, although there are still aggregate caloric gains from trade in this model— albeitsmallerthanthegainsinastandardtrademodel—tradecanspellshort-runnutritional losses for landless laborers. This is because habit formation results in this group spending a 2Habit formation has allowed each region to exploit some of the gains from specialization in previous generations, thereby reducing the production gains possible at the moment of trade liberalization. 2 large portion of their budget on the local staple whose price rises more than their income.3 The recent surge in world food prices provides supportive evidence of this link between tastes, trade and nutrition. In March 2006, Argentina banned all exports of beef in order to lower the domestic price for beef-loving Argentinians. In 2008, Vietnam, Cambodia and Egypt banned rice exports; Russia, Kazakhstan and Ukraine banned wheat exports; while Zambia and Malawi banned maize exports. In all of these cases, important agricultural exports are the preferred calorie sources of low-income households. Trade-induced price rises increased hunger among poor consumers, and governments explicitly tried to help this group by restricting trade and bringing prices of favored foods back towards autarky levels. In contrast, a standard trade model without habit formation would recommend transferring incomefromexporters,andallowingpoorconsumerstosubstituteintorelativelycheapfoods. I test my theory more directly by using detailed household survey data from India. India containsmanyagro-climaticzonesandextremelyvarieddiets. Atthesametime, Indiamain- tains extensive internal food-trade barriers in addition to a poor transport infrastructure. These barriers include tariffs at state borders, numerous licensing requirements for traders and physical movement restrictions. Despite much publicized economic reforms in the early 1990’s, agriculturecontinuestobesubjecttoenormousstateinterventioninthenameoffood self-sufficiency and agricultural markets are not integrated. Accordingly, my empirical work treats Indian regions like many small partially closed economies, and provides an excellent opportunity to test the autarky predictions of my model. In order to investigate the relationship between tastes and local resource endowments across India, I estimate the tastes defined as above with the further restriction that the price and income effects are common across India. I regress household demand for agricultural products on a set of regional dummies and a flexible set of common price and expenditure terms. The regional component of the unexplained variation in budget shares spent on each 3The model may not be relevant to households on the edge of starvation, who will presumably maximize caloric intake. However, the average rural Indian household consumes 300 calories less than the recommendeddailyintakeof2,400caloriesperday(2004/5NSSdata). As2,400caloriescouldbepurchased for forty percent of per-capita daily food expenditure yet half the children in rural India are underweight (2006 NFHS data), many malnourished households seem not to be maximizing nutrition alone. 3 food then provides my main taste measure. It is impossible to observe the impacts of trade liberalization in societies both with and without habit formation. Therefore, I use rural household survey data from 77 agro-climatic regions within India to provide empirical evidence for the mechanisms in my model that reduce the caloric gains from trade on the consumption side. In the first stage, I show that regions have stronger tastes for the foods that their agro-climatic endowments are relatively well-suited to producing, and that these foods are inexpensive compared to other regions. To highlight the role of habits, I confirm that the ordering of tastes within a region responds to relative price changes in the past. Therefore, if India were to liberalize internal trade, each region’s more favored foods will be expected to rise in relative price as regional prices con- verge to a uniform national price. In the second stage, I verify that these expected relative price rises in more favored foods will negatively impact nutrition by showing that between 1987 and 2005, caloric intake declined more in regions where (non trade-induced) price rises were more concentrated in locally favored foods, controlling for changes in food expenditure. I confirm my findings from looking across Indian regions using a second approach. Inter- state migrants mimic small economies opening to trade, since migrants bring their labor endowment and preferences but face a new set of prices. Therefore, moving should affect mi- grants similarly to how my model predicts trade liberalization affects consumers. I show that inter-state migrants in India do carry their food tastes with them, consuming food bundles less similar to those consumed in their destination state and more similar to those consumed in their origin state. Migrant households consume fewer calories for a given level of food expenditure, because they continue to buy favored products from their origin state that are now relatively expensive. This effect dissipates with time, disappearing only four generations after migration. Finally, mirroring the effects of temporal price variation within regions, I find that for the 484 observed migration routes, the caloric intake from a given level of food expenditure declines more where the relative price rises faced by a migrant are more concen- trated in that migrant’s preferred origin-state foods. These results even hold when I restrict 4 attention to households in which the household head’s wife moved for marriage, and compare households where the wife moved inter-state to those where the wife moved intra-state. With these two sets of evidence in place, I quantify the caloric impact of Indian agri- cultural trade liberalization, and the error associated with ignoring the correlation between tastes and endowments that comes about through habit formation. If India were to liberalize its internal agricultural trade (when food prices converge to the Indian median price), trade will have to generate income gains of at least 3.3 percent for the average Indian household to maintain their pre-liberalization caloric intake. No such income gains would be neces- sary if tastes were identical across India. Poorer regions, which consume larger shares of the local staple foods predicted to rise in price, will require even larger income gains to maintain caloric intake. Household incomes are likely to increase with liberalization through increased specialization in food production.4 However, the 3.3 percent increase required to avoid caloric losses is larger than existing estimates of the static income gains for typical rural households from other agricultural trade liberalization scenarios.5 Inthispaper, Ifocusonthenutritionalimpactsoftrade, measuredthroughcaloricintake,6 although my theoretical results apply to welfare in the isomorphic model where preparation techniques improve with past consumption. In few countries is malnutrition a more impor- tant issue than in India, which has a higher prevalence of undernutrition than Sub-Saharan Africa (Deaton and Dreze, 2008). Food consumption itself has regularly been used as a poverty measure. For example, poverty lines for countries as diverse as the US and India initially derive from the amount of money required to meet basic caloric needs. There are several reasons why economists should be directly concerned about poor nu- trition. Sen (1999) has argued that improving the health of the poor and increasing their capabilities should be a goal of development in itself. Low caloric intake directly reduces 4For poor Indians, these gains will primarily come through nominal wage gains or falls in the average food price, since most rural households derive the majority of their income from labor. 5The most direct comparison is China, where reductions in caloric intake over the reform period (Du et al., 2006) were accompanied by the dismantling of extensive barriers to internal agricultural trade. 6Caloric intake is not the same as nutrition, with vitamin and protein intake also important. However, as there is no single agreed-upon nutritional metric, in this paper I focus only on calories. 5 productivity by reducing energy levels, health capital and the ability of the immune system to fight off infectious disease. These effects exert externalities on other members of society. Malnourished populations allow contagious diseases to spread more readily, and Fogel (1994) has argued that improved nutrition and its synergies with technological advance can account for much of the economic growth seen in the West since the Industrial Revolution. Policymakers often cite explicitly paternalistic concerns. Many of the gains from proper nutrition come through good health later in life, which uninformed consumers may un- dervalue. Barker (1992) and others have demonstrated the substantial scarring effects of nutritional shortfalls at young ages on productivity, earnings and health in adulthood. Ac- cordingly, even the short-term nutritional declines that can occur during an episode of trade liberalization are of serious concern, because an entire generation malnourished as children will continue to suffer irremediable consequences for the rest of their lives. In section 2, I provide a diagrammatic discussion of the theory, with the formal proofs relegated to appendix C. Section 3 introduces the data and my taste estimates. In section 4, I investigate variations in tastes, prices, endowments and caloric intake across 77 regions of India. Section 5 uses data on inter-state migrants within India to confirm the regional re- sults. Section 6 discusses India’s internal trade restrictions and predicts the caloric impact of relaxing these restrictions. Finally, section 7 concludes and discusses the policy implications. 2 Theoretical Background 2.1 Modelling Habit Formation in Food Consumption I start by describing consumer preferences in an overlapping generations model that fea- tures habit formation in food tastes. I then justify my assumptions about the form of the utility function by reviewing the literature on the development of food preferences. Finally, I analyze this simple economy in autarky. Identicalindividualsinasmallregionlivefortwoperiods, childhoodandadulthood. Inthe second period, individuals obtain factors of production, spend their full income from these factors and have a single child. Adults in generation t choose their relative consumption of 6 thetwogoodsintheeconomy, rice, r, andwheat, w, bothmeasuredincalories. Thechildand the parent share the parent’s preferred consumption bundle, and form a single household.7 I model household demand as follows. The budget share spent on rice is a function h (.) r of relative prices, p /p , total (food) expenditure, food, and a rice taste shifter, tastes :8 r w r p rt bshare = tastes +h ( ,food ). (1) rt rt r t p wt In the next period the child grows up and the bundle that he or she consumed as a child in- fluenceshisorheradultpreferences. Iwillcallthishabitformation, withanadultdeveloping tastes for the foods of which he or she consumed relatively more as a child.9 Specifically, r ∂tastes t−1 rt tastes = g( ), with > 0. rt w ∂(rt−1) t−1 wt−1 Therefore, the adult’s utility function depends on both past and present consumption, as is common in the habit formation literature starting with Stone (1956) and including Becker and Murphy (1988). Additionally, I assume that parents are myopic, and so do not antici- pate the effect of their consumption choices on the future tastes of their child. Accordingly, household demand in period t does not depend on expected future prices or incomes.10 Strong and enduring taste patterns characterize food consumption. Ample evidence in the psychologyandnutritionliteraturesindicatesthatcertainfoodpreferencesforminchildhood. Children have a predisposition to fear new foods, which is only overcome through repeated opportunities to consume a food (Birch, 1999). The literature hypothesizes that this re- sponse serves a protective function, so that foragers can learn what foods are safe to eat. This is common across omnivores, and has been shown in controlled experiments among 7Specifically, parents gain equal utility from their own and their child’s consumption. Parents are not altruistic, or are uninformed about the importance of childhood nutrition, as they choose their child’s diet based solely on their own preferences. 8The budget share spent on rice increases with the tastes for rice and decreases with the relative rice price. The Cobb-Douglas utility function generates household demands of this type, with h (.) r equal to zero. The main results from my paper carry through with a more general demand specification, bsharert =ehr(ppwrtt,foodt,tastesrt), as long as ∂bsharert/∂tastesrt >0. 9Habit formation may occur more quickly than this, which can be accommodated by including several stages of adult life, with past consumption influencing current preferences. For example, preferences for eating raw fish in sushi have developed rapidly in the West over a single generation. 10Even if this assumption is violated, parents will only partially adjust their child’s diet if there is uncertainty over the prevailing relative prices when their child reaches adulthood, or if there are costs to preparing two separate meals for both themselves and their child. 7 both humans and rats. More directly, a mother’s diet during pregnancy and lactation affects her child’s preferences for flavors and foods in later life (Mennella et al., 2001). Social factors also play an important role in forming preferences. There is abundant ev- idence, again from controlled experiments involving humans and other mammals, that the young are more likely to accept new or disliked foods if they observe their mothers or other role models consuming them (Birch, 1999). This effect works through two channels; role models both induce children to try a food for the first time which overcomes the initial neophobia, and provide a pleasant context in which the food is eaten. As an example of the latter channel, second generation US immigrants from India may enjoy eating curry more than their non-Indian peers in part because of the positive association that has formed between consuming Indian food and dining at home with family members. Crucially for my assumptions about habit formation, preferences gained in childhood per- sist in the available longitudinal data. Data from the Minnesota Heart Health Program (Kelder, 1994) show that food preference rankings remain unchanged over 6 years.11 There- fore, this extensive set of evidence supports my assumption that food preferences are posi- tively related to the consumption patterns of the previous generation. In order to model the production side of the economy, I use a specific factors model that matches the realities of food production well. A region has a fixed endowment of laborers, L, and two additional factors: land suitable for rice cultivation, T , and land suitable for wheat r cultivation, T . Both rice and wheat are produced with constant returns to scale technology w using labor and the single specific land factor, with diminishing returns to an increase in any one factor.12 To grow wheat, a farmer requires well-drained soil, low humidity and moderate temperatures. Rice grows most easily in coastal plains, lowland deltas and tidal plains where paddies are submerged in water. These agro-climatic conditions are fixed over time and can explain why the arid plains of Rajasthan produce mainly wheat and West Bengal in the 11All of the sample faced the same relative prices, yet there were substantial and persistent variations in food rankings which can not be explained by the common price effects. 12All of the main results carry through to the Heckscher-Ohlin model where there are only two factors, T and T , which are required to produce both goods, but rice is relatively more intensive in factor T . r w r 8