The impact of food and economic crises on diet and nutrition

Transcription

1 The impat of food and eonomi rises on diet and nutrition Ralitza Dimova a,e Institute of Development Poliy and Management, University of Manhester Ira N. ang b,e,f,g Department of Eonomis, Rutgers University Monnet B.P. bakou University of Coody-Abidjan Daniel Hoffman d Department of Nutritional Sienes, Rutgers University dhoffman@aesop.rutgers.edu a Institute of Development Poliy and Management, Shool of Environment and Development, The University of Manhester, PO Box 88, Manhester, M60 1QD, UK b Department of Eonomis, 75 Hamilton Street, Rutgers University, New Brunswik NJ USA Université de Coody-Abidjan, Département d'eonomie, BP V 34, Abidjan 01 Côte d Ivoire d Department of Nutritional Sienes, Rutgers University, New Brunswik NJ USA e IZA, Shaumburg-Lippe-Str. 5-9, D Bonn ermany f CReAM, Department of Eonomis, University College London, Drayton House, 30 ordon Street, London WC1H 0AX United Kingdom g OEI, Institute for East European Studies, Landshuter Str. 4, Regensburg ermany Corresponding author. Ira N. ang Department of Eonomis, 75 Hamilton Street, Rutgers University, New Brunswik NJ USA phone: fax: gang@eonomis.rutgers.edu This work was supported by the Leverhulme Trust F/00 275/L. [1]

2 The impat of food and eonomi rises on diet and nutrition ABSTRACT The impat of skyroketing food pries and falling inomes on diet and nutrition is of major poliy and soial onern. We rigorously explore the dietary and nutritional impliations of prie and inome shoks in a before, during and after manner. We find not only a tendeny of households to realloate their onsumption baskets during a risis, but also a dramati hange in the inome and prie elastiities of demand for both food and nutrients. Our results hallenge the impliit assumption of relatively low and invariable prie and inome elastiities of demand for food and nutrition in the existing literature and have potentially important poliy impliations. Keywords: Crisis; Diet; Flutuation; Health; Nutrition JEL Codes: E320, I120, P230, P240, P360 [2]

3 1. Introdution One of the most high profile areas of eonomi researh is the impat of hanging pries and inomes on diet and nutrition, whih is not surprising, given the well established fat that both the alorie intake and the omposition of the diet have important impliations for health, mortality, fertility and eonomi produtivity. While researh in this area has traditionally foused on hunger alleviation and degenerative health in the proess of industrialisation and development, oasional food rises and poverty hikes as part of strutural reforms throughout the 1980s and 1990s turned the attention of experts towards the potentially detrimental effet of dramatially flutuating food pries and inomes on diet and nutrition. The global food and finanial rises in the post-2008 period sparked renewed interest in this topi. A large number of poliy researh papers, typially written under the auspies of the Food and Agriultural Organization (FAO) of the United Nations and the World Bank, have tried to identify the net losers from the global food risis, a term popularly used to desribe the rising staple and other food pries in 2008 and later 1. Welfare related onerns have been further aggravated by the severe inome shok in the aftermath of the global finanial risis, aounting for an estimated 73 million additional people living below $1.25 a day and 91 million additional people living under $2 a day within the two years subsequent to the 2008 global finanial meltdown (Chen and Ravallion, 2009: Of speial onern is the immediate effet of the rises on nutrition and health (Development Researh roup, 2008). While the nutritional onsequenes are bound to be largest for the poor 1 See, for instane, the Poliy Researh Working Paper series and similar FAO publiations suh as Zezza et al (2009). [3]

4 in Less Developed Countries, no eonomy is immune to the ombined effet of food prie inflation and delining inomes. Although rigorous aademi researh has attempted to keep up with the poliy onern by providing estimates of nutritional and other onsequenes of prie and inome shoks, it has failed to reah unambiguous onlusions. The prolifi 1990s literature on the impat of eonomi rises on nutrition and health has failed to agree on whether relatively poorer or relatively more affluent households are among the largest vitims of eonomi shoks (Stillman, 2006; Frankenberg et al, 1999). At the same time, the rapidly expanding post-2008 literature on the effet of rising staple food pries is dominated by (i) simulations based on Computable eneral Equilibrium Models, a methodology, ritiized for its blak-box type of analysis, and (ii) studies trying to identify the impat of the prie inrease of a partiular food item, e.g. rie or wheat, on the welfare of net buyers of that partiular food item. While providing a rigorous piture of the number and poverty levels of those affeted by a prie shok, onditional on everything else in the world remaining the same, these studies fail to aount for the possibility that a prie shok may indue onsumers to realloate their food baskets towards heaper food items, an ation with ex ante unlear nutritional onsequenes. A serious onstraint to truly omprehensive researh on the impliations of prie and general eonomi shoks on onsumption and nutrition is the sarity of rih enough data allowing the researher to explore the shok impliations in a before and after manner. To the best of our knowledge, only Eker and Quaim (2010) take seriously into aount the possibility that a prie shok on speifi food groups ould indue realloation of onsumption and nutrition. However, their analysis is based on a single ross-setion of data and orresponding simulations that assume unhanging food and nutrient elastiities, despite the possibility that [4]

5 severe prie and/or eonomi shoks may indue a strutural break and hene ompletely hange onsumer behaviour (Aker et al, 2011). The broader empirial literature, fousing on prie and nutrient elastiities of onsumption and nutrition, is typially based on single ross-setions of data, often from fairly uneventful periods of time. The general finding is that food onsumption, espeially of staple foods, is generally inome and/or prie inelasti (Berhman and Deolalikar, 1987; Diagana et al, 1999). While some studies find deleterious nutritional impliations of prie inelasti demand for food in the fae of a severe prie shok (Diagana et al, 1999), others find inome elastiities of nutrients to be smaller than the orresponding food elastiities, exemplifying greater willingness of households to ompromise on tastes than nutritional value (Behrman and Deolalikar, 1987) and thus explaining the absene of visible nutritional and health impliations of even severe eonomi rises suh as that of 1998 in Russia (Stillman and Thomas, 2008). The purpose of this paper is to fill some of the gaps outlined above and attempt to explain some of the existing ontroversies in the literature by using the natural experiment of Bulgaria in the mid-1990s-to early 2000s. The ontext and timing are ideal for exploring the impliations of severe inome and prie shoks, as well as the role of hanging relative food pries on onsumption and nutrition. Almost nowhere was the shok of strutural reform and risis as severe as in Bulgaria during the 1990s. The dissolution of the Counil for Mutual Eonomi Assistane (CMEA), the war in former Yugoslavia, and poliy stalemates, all led to a greater drop in output and higher inflation than in the majority of the Central and East European (CEE) ountries, ulminating in the risis of While the risis in Russia led to a 40% inrease in inflation from 20% to 60% between 1996 and 1997, inflation in Bulgaria inreased by 827% from the already high base of 122.9%. Lifetime savings were lost. [5]

6 Poverty inidene inreased by 77% (Sahn et al, 2002), while the fall in food onsumption exeeded that of the majority of the CEE ountries (Elsner and Hartmann, 1998). There is evidene that onsumption of ommerially produed bread and meat must have been the most affeted, as both inome deline and agriultural setor problems made the prodution of grains and livestok espeially problemati and ontributed to signifiant inrease in their pries (Ivanova et al, 2006). Aording to OECD (1997) the broader eonomi risis in Bulgaria in the mid-1990s was aompanied by a bread risis. The prie of a standard white bread loaf inreased most dramatially by 34% from levs in 1995 to 545 levs in There was a spillover of the grain risis into a general food risis, exemplified for instane by a 25% inrease in the prie of a kg of pork from levs in 1995 to 5325 levs in 1997 and an 18% inrease in the prie of a kg of beef from levs in 1995 to 2988 levs in 1997 (National Statistial Offie of Bulgaria, 2000). With the use of uniquely timed data - from before the risis of 1996, through the risis of , until the risis-free year we explore the reation of household onsumption and nutrition to hanging food pries and inomes. To fully understand the welfare impliations of the prie and inome shoks, we estimate the prie and inome elastiities of food and nutrition for households belonging to the poorest, middle inome and rihest perentiles in the inome distribution. We find not only a tendeny of households to realloate their onsumption baskets during a risis, but also a dramati hange in the inome and prie elastiities of demand for both food and nutrients. Our results hallenge the impliit assumption of relatively low and invariable prie and inome elastiities of demand for food and nutrition. In setion 2 we provide a desription of the general eonomi bakground of this study on Bulgaria and present some onsumption and nutrient statistis. In setion 3 we outline the [6]

7 methodologies used to estimate prie and inome elastiities of food and nutrients. In setion 4 we disuss our results on inome and prie elastiities of key food groups, while in setion 5 we disuss the orresponding results on nutrient prie and inome elastiities. Setion 6 onludes. 2. The story of onsumption and nutrition in Bulgaria The main data soures for our analysis are the Living Standards Measurement Surveys (LSMS) for 1995, 1997 and 2001, provided by the World Bank 2. The surveys provide detailed information on monthly food onsumption and expenditures, total expenditures and inomes, demographi and other harateristis of interest from approximately 2500 randomly seleted households in eah of the three ross-setions 3. We supplement these data with data on the nutrient omposition of all food groups onsumed, olleted by the National Centre of Publi Health Protetion in Bulgaria 4. A preliminary analysis of the data shows that average monthly real inomes delined dramatially from levs to levs between 1995 and 1997 and then grew bak to approximately their original levels by There is evidene that the dramati hanges in inomes may have influened food expenditures signifiantly (Ivanova et al, 2006), and we would like to explore this possibility as a first step in our desriptive analysis. Table 1 highlights the perentage hanges in the food baskets of households belonging to different segments of the 1995 inome distribution between 1995 and 1997 and between 1997 and While a survey for 2003 is also available, it differs signifiantly from the three earlier surveys making omparisons aross the four ross-setions diffiult. The Bulgarian eonomy stabilized signifiantly after 2000 and we do not expet major hanges to have taken plae between 2001 and 2003 in the phenomena and indiators in whih we are interested. 3 Speifially, the surveys inlude information on 2468 households in 1995, 2323 households in 1997 and 2633 households in We thank Ludmila Ivanova and Plamen Dimitrov for making these data available to us. [7]

8 To keep our terms of referene broadly the same over time, we follow a proedure similar to that used by the LSMS team in onstruting omparable poverty lines over time. Speifially, we alloate households into inome perentiles in We then adjust the referene inome of households for inflation and define the perentile distribution of households in 1997 and 2001 aordingly. For example, let the 10 th perentile in 1995 inlude households whose inomes lie between 0 and X levs. In defining the 10 th perentile in 1997, we adjust X for inflation and inlude in the 10 th perentile of the 1997 distribution households whose inomes lie between 0 and X/CPI levs. Hene, while for 1995 we are dealing with the atual perentiles as stated, for 1997 and 2001 the division between the reported perentiles in fat refers to the 1995 boundaries for these perentiles, appropriately adjusted for inflation. In this way we are looking at absolute as opposed to relative welfare measures and their real hanges over time. Using real thresholds onstruted in this manner enables us to examine from survey to survey the variation in onsumption of the set of households with the same real expenditures. These are not the same households in eah quintile over time, so we are not traing how the risis affets the positioning of households in the inome distribution. Rather we are looking at the how households with the same expenditures hange their onsumption of food items and nutritional intake in response to prie and inome hanges. Perhaps the most striking observation from this table is the signifiantly larger proportion of bread and starhes in the food basket of the poorer perentiles and the signifiantly larger proportion of meat in the food basket of the rihest perentiles throughout the period. During the risis, the proportion of bread in the food basket of all groups of onsumers inreased, while the proportion of meat dereased slightly for the poorest perentiles and went up signifiantly for the [8]

9 rihest perentiles. After the risis, onsumption patterns shifted bak towards their original positions, but never returned to their pre-risis levels. Average alori intake dereased signifiantly during the risis for all inome perentiles and started reovering afterwards, though never returning to pre-risis levels (Ivanova et al., 2006). Furthermore, the nutrient omposition of the diet shifted with a lower intake of fats and a higher intake of proteins and arbohydrates during the risis (Figure 1) 5. Table 1: Changes in the budget share of key food groups, Variable th perentile Bread 0.15 (0.10) 0.26 (0.13) 0.22 (0.12) Starhes 0.15 (0.08) 0.14 (0.10) 0.16 (0.08) Meat 0.21 (0.11) 0.19 (0.12) 0.18 (0.11) Fruit-vegetables 0.18 (0.10) 0.14 (0.10) 0.14 (0.10) Oil-fat 0.05 (0.03) 0.04 (0.03) 0.06 (0.04) Dairies 0.22 (0.11) 0.18 (0.12) 0.19 (0.11) Sweets 0.04 (0.03) 0.05 (0.06) 0.04 (0.04) th perentile Bread 0.10 (0.07) 0.16 (0.08) 0.16 (0.09) Starhes 0.12 (0.06) 0.11 (0.06) 0.13 (0.06) Meat 0.24 (0.11) 0.28 (0.13) 0.23 (0.11) Fruit-vegetables 0.22 (0.12) 0.20 (0.13) 0.18 (0.10) Oil-fat 0.05 (0.05) 0.04 (0.03) 0.05 (0.03) Dairies 0.20 (0.12) 0.16 (0.09) 0.19 (0.10) Sweets 0.06 (0.04) 0.06 (0.05) 0.06 (0.04) 90 th perentile Bread 0.06 (0.08) 0.08 (0.05) 0.11 (0.10) 5 iven the limited spae and the large amount of information, we must arefully hoose whih piees to report. In this diagram and subsequent empirial analysis of nutrition we fous only on maronutrients. Note that our mironutrient information indiates that, as expeted, the proportion of all mironutrients in the diet - niain, iron, alium, thiamine and riboflavin - dereased during the risis and did not return to its original level by [9]

10 Starhes 0.08 (0.04) 0.08 (0.05) 0.11 (0.09) Meat 0.32 (0.12) 0.41 (0.15) 0.28 (0.15) Fruit-vegetables 0.26 (0.11) 0.21 (0.11) 0.19 (0.10) Oil-fat 0.04 (0.05) 0.04 (0.02) 0.04 (0.03) Dairies 0.17 (0.10) 0.15 (0.05) 0.18 (0.10) Sweets 0.08 (0.05) 0.05 (0.04) 0.08 (0.07) Soure: Own alulations based on the LSMS data set. Notes: shares (standard errors). See the text for a desription of perentiles determination. Briefly, for 1995 we use atual perentiles, for 1997 and 2001 the division between the reported perentiles refers to the 1995 boundaries for these perentiles, appropriately adjusted for inflation. iven the omplexity of the eonomi situation during the fous period, it is diffiult to attribute hanges in onsumption and nutrition to one partiular fator. Thus, the hanges ould have been driven by either the redution of purhasing power alone, or hanging relative pries of key food items or hange in the responsiveness of households to these inomes and pries. From a poliy making perspetive, it is instrutive to disentangle the impliations of all of these influenes. Figure 1: Changes in nutrient onsumption share of proteins share of fats share of arbohydrates Soure: Own alulations based on the LSMS and data on the nutrient omposition of all food groups onsumed, olleted by the National Centre of Publi Health Protetion in Bulgaria. Notes: The figure highlights the total monthly alories averaged aross perentiles defined on the basis of per adult equivalent expenditures See text and Table 1 for a desription of perentiles determination. [10]

11 Unfortunately, we do not have aess to reliable pries at either the household or regional level. We must extrat prie related information from the available evidene on unit values total expenditures divided by total quantities of food items - a problem as unit values reflet the quality hoies that households make as well as the pries that they fae (Prais and Houtakker, 1955). We disuss and rigorously address this issue in our empirial analysis. However, as a first attempt at making prie related sense of the information available, it is useful to look at the hanges in unit values of key food groups. The information on unit values reported in Table 2 is onsistent with observed onsumption patterns and provides some tentative explanation of these patterns that goes beyond that of shifting real inomes over time. In partiular, we see that in eah of the years, the unit Table 2: Average unit values of key food groups, Variable th perentile Bread 0.62 (2.57) 0.86 (0.44) 0.63 (0.15) Starhes 1.33 (0.41) 1.13 (3.13) 0.89 (0.29) Meat 5.03 (1.08) 5.30 (2.81) 3.69 (0.97) Fruit-vegetables 1.17 (0.46) 1.02 (0.61) 1.08 (0.61) Oil-fat 1.98 (0.41) 1.38 (1.01) 1.60 (0.31) Dairies 1.60 (1.45) 1.73 (1.69) 1.52 (0.74) Sweets 1.34 (0.56) 1.29 (0.80) 1.14 (0.94) th perentile Bread 0.62 (2.57) 0.89 (0.59) 0.67 (0.60) Starhes 1.29 (0.36) 0.99 (0.47) 0.92 (0.28) Meat 5.66 (1.25) 6.11 (2.37) 4.23 (1.18) Fruit-vegetables 1.28 (0.48) 1.21 (0.68) 1.03 (0.46) Oil-fat 2.31 (1.19) 1.71 (0.93) 1.71 (0.52) Dairies 1.78 (2.64) 1.63 (0.94) 1.86 (3.07) Sweets 1.30 (0.71) 1.40 (0.95) 1.14 (0.62) 90 th perentile Bread 0.71 (2.44) 0.87 (0.15) 0.75 (0.83) [11]

12 Starhes 1.40 (0.38) 0.95 (0.27) 0.99 (0.49) Meat 6.42 (1.38) 6.78 (1.51) 4.68 (1.72) Fruit-vegetables 1.53 (0.45) 1.19 (0.59) 1.17 (0.60) Oil-fat 2.56 (0.89) 2.24 (1.16) 1.82 (0.62) Dairies 2.10 (2.62) 1.98 (1.13) 2.33 (3.42) Sweets 1.44 (0.70) 1.10 (0.67) 1.26 (0.82) Soure: Own alulations based on the LSMS data set. Notes: The values are expressed in real 2001 terms. The numbers in brakets are standard deviations. See text and Table 1 for a desription of perentiles determination. values of meat signifiantly exeed the unit values of staple foods, whih is onsistent with the apparent greater ability of the riher strata of the population to afford meat ompared to those belonging to the poorer perentiles 6. In addition, the signifiant inrease in the unit value of bread between 1995 and 1997 and the orresponding rise of the share of bread in the food basket of all inome perentiles possibly indiates low elastiity of bread Bulgaria s main staple food to prie hanges. iven that meat and staple foods (bread and starhes) are the main items in Bulgaria s food basket and that some of the most notieable results in both our desriptive and subsequent empirial analysis are related to these food items, we will fous on them in the desription of our elastiity estimates. The full set of empirial results is available in our full tehnial appendix, whih is available upon request. 3. Eonometri methodology 3.1. Estimation of inome and prie elastiities of food intake The main shortoming of our data is the absene of information on pries and hene the need to infer responses of households to prie hanges on the basis of information on unit values. For instane, we are likely to observe higher unit values for households whose basket onsists of 6 Note that this information is onsistent with information on aggregate yearly pries of food items, provided by the National Statistial Institute of Bulgaria. [12]

13 higher quality items. Unlike the market prie, over whih an individual household does not have any ontrol, the unit value represents a hoie variable, whih is under the ontrol of households. If we are to therefore infer prie elastiities on the basis of unit value data, our results are likely to be tarnished by a simultaneity bias: households hoose both the quantity and the quality of a good and better off households would tend to buy higher quality goods, whose unit value is positively related to total finanial outlays. To overome this problem and produe unbiased estimates of prie elastiities, we implement the Crawford et al. (2003) model, whose main advantage over alternative models is that it allows us to exploit the expliit links between unit values and pries in a way that is onsistent with the Almost Ideal Demand System (AIDS) approah. In keeping with the rest of the literature, foods are organized in m groups (bread, starhes, meat, et.). Under the assumptions of separability of preferenes and homogeneity, we define the following relationship: V h V Q, [1] where V is the unit value for group, Q is the orresponding quantity index and homogeneous prie index (e.g. a Paashe prie index), onstruted assuming a onstant struture of relative pries within group. Taking a double logarithm of [1] and given a funtional form for the budget shares w, we therefore need to estimate a onsistent system: X ln V ln ln h X,, and [2] w X,, [3] [13]

14 where X is total expenditures, and is a vetor of group prie levels (the omission of indiates that these parameters refer to all groups). To make the estimation omputationally tratable, a speial funtional form for h is adopted suh that ln V a b lnq ln. [4] As for the funtional form of the demand funtion,, the model uses the approximate Almost Ideal Demand (AID) model with a loglinear approximation of the log index prie (LA/AID). While the full AID speifiation or its quadrati extension would be preferable, the non-linear form would not be tratable by the within-luster estimation adopted in this method. We attempt to extrat at least some of the information that non-linear inome speifiation would provide by estimating prie and inome elastiities for households belonging to different perentiles of total expenditures. Assuming fixed pries for households loated within a luster, the demand funtion for group by household h is: w h h h h 0 Z α H ln H ln x u, [5] H where h x is deflated expenditure, h x h C ln X ln P h ln X H ln ln, H H P is a luster prie index with suitably hosen weights, H is the prie of group H in luster. Equation (5) an be re-written as: w h ln X h h h 0 Z α H ln H, [6] H u [14]

15 where H. Vetor Z h inludes soio-demographi harateristis and other H H onditioning variables. Following the same logi, the unit value equation beomes: h h h h ln V a0 Z a ln b ln Q v [7] The estimation proeeds under the restriting assumption of independene among observations, as households are grouped by luster and hene by onstrution ommon fators affet the demand for ommodities within the luster. However, under Lewbel s (1993, 1996) assumption of stohasti independene between relative good pries that are allowed to vary aross lusters and the luster prie index, this luster effet an be shown to be innouous (Crawford et al., 2003). The estimation proeeds in three stages. In the first stage, we ompute the within-luster estimates, whih allow the anelling of the unobserved prie effets and retrieving the estimated vetors αˆ and â, and the estimated salars ˆ and bˆ. h h h h w w Z α ln X ln X u u Z, [8] h h h h and ln V ln V Z a b ln Q ln Q v v Z. [9] The seond stage onsists of estimating the prie oeffiients H using between-luster information, as the fixed nature of the within luster prie effets has already been used in the first stage. At this stage, we impose the standard homogeneity restrition from demand H theory, 0 (whih implies also an adding-up restrition). Vetor λ is subjet to positive H linear homogeneity of the prie index restritions 0 and 1. Sine this is not suffiient to identify the parameters of interest, λ is arbitrarily set equal to w, the vetor of H H [15]

16 [16] average budget shares. The estimation of ˆ (the prie effets in the budget equation for group ) also assumes homosedastiity of the variane of ' ', h h v u and takes into aount the measurement errors in the unit values. The resulting relationship is: C vu C v n n ˆ ˆ ˆ ˆ ' ' λ Ω Ω λ Ω v, [10] where, n is the size of eah luster H H H u X w ln ln ˆ ˆ 0 α Z ' 1 1, m, with m v a Q b V 1, ln ln ˆ ˆ ln 0 a Z, v v v Ω Ω Ω Ω v ˆ ˆ ˆ ˆ 1 ˆ u u u n u V where eah term of Ωˆ is obtained from the first stage residuals. The variane of the prie oeffiients (without imposing symmetry) is obtained by bootstrapping. In the third stage, we impose the symmetry, H H, by minimum distane estimation. By using the effiieny arguments of Kodde et al (1990, theorem 5), we minimise only over rather than over and. Prie elastiities are omputed for household belonging to the 10 th, th and 90 th expenditure perentiles using the formula, H H H H w w e 1 ~ ~.; where w ~ and H w ~

17 represent the budget shares of group and group H respetively. Total expenditure elastiities are also omputed using the formula, e 1 w~. We follow losely the speifiation proposed by Crawford et al (2003). Sine the results from the first two stages are not used in our disussion, we keep the variable desription and results from these two stages along with the full set of inome and prie elastiities only in the full tehnial appendix to be available eletronially in a working paper version of the paper, but publishable upon the disretion of referees and editors Estimation of inome and prie elastiities of nutrients To derive nutrient elastiities, we apply Huang s (1996) method, whih uses elastiities from standard demand analysis to estimate elastiities of hanges in the nutritional ontent of onsumer diets. On the basis of the demand struture for food and the bundle of orresponding nutrient attributes it is possible to derive the implied relationship between nutrient availability and hanges in food pries and inomes. Huang s approah provides information on how to derive the formula from an underlying demand model. Let a ki be the quantity of the th k nutrient obtained from a unit of the th food group. The total quantity of that nutrient,, obtained from various food groups an be expressed as: k a Q k k. [11] Equation (11) represents the onsumption tehnology in the sense of Lanaster (1966). It is straightforward to show that: d k k H e a Q d e a Q dx X DkH H H k k d dx X H H k H H [17] k k, [12]

18 where D e a Q is a prie elastiity measure apturing the effet of the kh H k k th H food th group prie on the availability of the k nutrient; k e a k Q k is an inome (or total expenditure) elastiity measure relating the effet of a hange in inome on the availability of that nutrient. In other words, the measurement of D kh represents the weighted average of all ' th own- and ross-prie elastiities e H s in response to a hange in the H prie, with eah weight expressed as the share of eah food group s ontribution to the k th nutrient a k Q ' k s. Similarly, k represents the weighted average of all inome elastiities with eah weight expressed as the share of eah food s ontribution to the k th nutrient. The matrix of nutrient elastiities is thus obtained as the produt of nutrient shares of food groups S, and food demand elastiities: e ' s, N S E. [13] As before, we report the full set of maronutrient and mironutrient elastiities in Tables TA12- TA26 of our full tehnial appendix. In the text, we only fous on a seletion of maronutrient elastiities. However, the mironutrient elastiity pattern is onsistent with our story line. 4. Eonometri results 4.1. Highlights from the first stages of our empirial analysis The set of variables used in our empirial analysis is desribed in table TA1 of the appendix. In Tables TA2-TA4, we report the budget share estimates, while in Tables TA5-TA7 of the appendix we report the unit value regressions from the first stage of the empirial analysis, desribed in setion 3. Sine these first stage results are of only marginal importane for our [18]

19 main line of argument, here we fous on only a few tehnial aspets, whih are relevant for the robustness of our key elastiity estimates. To begin with, we selet seven ategories of (food) groups for our demand estimations. The hoie is onstrained by the need to have both quantities and expenditures available. iven that we have no reason to believe that the availability of quantity information is diretly related to the struture of preferenes, there is also no reason to believe that these preferenes are separable in the orresponding partition of goods. To address the potential problem of nonseparability of preferenes in the partition of modelled and non-modelled goods, we ondition the budget shares of the inluded goods on the expenditures on the exluded goods and durable good ownership (Browning and Meghir, 1991; Crawford et al, 2003). Browning and Meghir (1991) argue that this is an eonomial way of relaxing separability and still maintaining the fous on our goods of interest. Homogeneity is imposed by expressing onditioning good expenditures with respet to non-modelled food expenditures (i.e., we divided expenditure of eah onditioning good by the expenditure of non-modelled food). To address the problem of zero onditioning expenditure, we inlude dummy variables, indiating zero spending on onditioning goods. This allowed us to keep households with zero onditioning expenditure in our sample (Crawford et al, 2003). The null hypotheses of non-separability would be rejeted if the onditioning goods play no role in the demand equations. Our results in Tables TA2-TA4 indiate that in most ases this is indeed the ase: most of the onditioning expenditures do not affet the budget shares of our seven modelled groups. This is partiularly true in the ase of lothing, shoes and expenditures on furniture, although some exeptions our at the 5% level of signifiane and there are hanges in relevane of the signifiant variables over the years. [19]

20 A seond important issue to onsider is the potential endogeneity of some of our regressors in the budget share equations. This is partiularly the ase with the logarithm of total expenditures and durable ownership. To orret for potential endogeneity, we one again followed Crawford et al (2003) in implementing a within-luster 2SLS estimator, using the set of instruments proposed by these authors. Speifially, we used the logarithm of household total inome as an instrument for the logarithm of household total expenditure and quantity of eah group, while durable ownership and the onditioning expenditures were instrumented with the use of their luster means. The exlusion restritions (of instruments) did not pass the Sargan s test; and the Durbin-Wu-Hausman test of endogeneity failed to onfirm endogeneity of any of the onsidered variables; hene, the reported figures are from within-luster OLS estimation. In Tables TA2-TA4, we see that during all years, total expenditures have a negative impat on the shares onsumed of bread, starhes, fats and oils and dairy produts and a positive impat on the shares onsumed of meat, fruit and vegetables and sweets. These results are onsistent with our desriptive statistis on the greater proportions of the latter types of food items in the baskets of riher households. These results onfirm the existene of Engel urves in Bulgaria. The rest of our results are onsistent with any onventional assumptions. When the unit values are regressed on the orresponding food quantities and other ontrols (Tables TA5-TA7) the quantity variable is signifiant, supporting our approah to estimation 7. 7 Reently it has been argued (e.g. MKelvey, 2011) that the lak of prie data at the individual ommodity level presents an insurmountable diffiulty for onsumption analysis and that even when perfet prie data is available, there is still a bias. Our fous is on hanging elastiities aross the years (rather than a perfet elastiity estimate for eah year). The approah we follow provides approximately onsistent results under mild versions of traditional assumptions suh as separability. [20]

21 4.2. Inome and prie elastiities of food groups The onsumption of speifi food items is shaped by what is happening to both relative pries and inomes. The more prie and inome elasti a food item, the greater the impat of prie and inome hanges on quantities onsumed. As we disussed above, in our ontext the eonomi risis sharply lowered inomes between 1995 and 1997 at a time of rapidly hanging absolute and relative pries, with some returning to the pre-risis levels by To grasp the impat of the risis on diet we need to examine the hanging prie and inome elastiities over the ourse of Bulgaria s eonomi transition. In this setion we report the prie and inome elastiity of bread, starhes and meat onsumption, alulated at the real expenditure levels of the 10 th, th and 90 th perentiles of per adult equivalent expenditures of the population in eah of the available years, where, as indiated earlier, perentiles are fixed in 1995 real terms. Figure 2: Inome elastiities inome elastiities, 10 th % inome elastiities, th % inome elastiities, 90 th % Soure: Own alulations based on the LSMS data set. Notes: The figure highlights the total monthly alories averaged aross perentiles defined on the basis of per adult equivalent expenditures See text and Table 1 for a desription of perentiles determination. For the sake of visual larity, the unbiased inome elastiities for eah of the key food groups in the sample are presented in Figure 2, for eah of the years and inome perentiles of [21]

22 interest. The atual elastiities and their standard errors are reported in Table TA8 at the end of the text, while estimates for other quintiles are available upon request. We see that during all years and aross all inome perentiles, meat was a luxury good (inome elastiity of demand exeeding one), while bread and starhes were neessities (inome elastiity of demand between zero and one). However, during the risis, the positive elastiity of meat inreased signifiantly in the ase of the 10 th perentile, inreased only slightly in the ase of the middle perentile and remained almost unhanged for the 90 th perentile. At the same time, the inome elastiity of bread dereased aross all inome perentiles and dereased most dramatially for the 90 th perentile, for whih bread beame an inferior good in These results are onsistent with our observations on hanges in the broad onsumption patterns aross the inome perentiles. Meat is a luxury good; this is onsistent with our observation from Table 1 that lower inome households during the eonomi risis redue their share of household expenditures on meat. Likewise, the share of bread in the household expenditures for all inome levels falls, as we expet, given that bread shows up in our estimates as a normal-to-inferior good. However, the inrease by households in the higher inome perentiles of their onsumption of meat a luxury good in the fae of falling inomes must be influened to a larger extent by either hanging relative pries or different responsiveness to pries. This is not obvious when looking at the unit values in Table 2, but beomes muh learer using our estimates. The own prie and ross-prie elastiities of the key food groups over time and aross inome perentiles are summarized in Figure 3 and highlighted in Table TA9 (one again, results for the rest of the per adult equivalent expenditure quintiles are available upon request). We observe that the own prie elastiities of eah of the food groups inreased dramatially over [22]

23 time (they beome more elasti). We also observe that the substitutability (i.e. the positive rossprie elastiity) between staple foods and meat inreased signifiantly during the risis. The onsumer behaviour of those belonging to the higher inome perentiles was haraterised by greater own-prie elastiity of staple foods and lower own-prie elastiity of meat. The lower prie elastiity of meat in the basket of the better off households provides a trustworthy explanation of their ability to sustain and even inrease the onsumption of meat during the risis, when the onsumption of meat by the poor went down. Figure 3: Seleted prie elastiities bread, 10 th % bread, th % bread, 90 th % meat, 10 th perentile meat, th perentile meat, 90 th perentile [23]

24 starhes, 10 th % starhes, th % starhes, 90 th % Soure: Own alulations based on the LSMS data set. Notes: The figure highlights the total monthly alories averaged aross perentiles defined on the basis of per adult equivalent expenditures See text and Table 1 for a desription of perentiles determination. 5. Seleted inome and prie elastiities of nutrients The preeding analysis indiated that the food omposition of the Bulgarian diet hanged signifiantly during the risis. The hanges differed aross inome perentiles and were driven by a omplex interplay of hanging real inomes and relative pries, as well as hanging responses to these inomes and pries. Despite the differenes in the hanging food omposition aross inome perentiles, different groups of households experiened similar hanges in nutrient intakes, whih were marked by an inrease in the onsumption of protein and arbohydrates and a derease in the onsumption of fats aross inome groups. Sine these hanges may be indiative of hanging responsiveness of nutrients to pries and inomes, we address this possibility in the next and last step of our analysis. Nutrient elastiity is given by the hange in nutrients suh as arbohydrates, proteins and fats as food pries hange by 1 perent. The prie and inome elastiities of nutrients are alulated as in Huang (1996) using the nutrient omponents of different food groups to onvert the estimated prie and inome elastiities into respetive nutrient elastiities, and thus provide us with the elastiities of hanges in the nutritional ontent of onsumer diets. As before we highlight and disuss the inome and prie elastiity of a seleted set of key nutrients and report [24]

25 the full set of elastiities in the full tehnial appendix. To save spae and given that our main purpose is to explore the dimension and possible hange in nutritional responses to hanging inomes and pries rather than analyse in detail eah effet, here we only fous on maronutrients. However, an interested reader an also explore in the full tehnial appendix (available upon request) the mironutrient estimates, whih are onsistent with our general patterns and message of the paper. The inome elastiities of maronutirents are highlighted in Figure 4 and reported in Table TA10. These elastiities hanged signifiantly during the risis, when the elastiity of fat inreased and the elastiity of other maronutrients dereased substantially. The elastiity of protein, arbohydrates and alories dereased the most in the ase of the rihest perentiles, undoubtedly due to the better ability of households belonging to this group to afford preserving their nutrient status. These inome elastiities provide a onvining explanation of the pattern of nutrient hanges that we observed in Figure 1. Figure 4: Inome elastiity of nutrients Soure: Own alulations based on the LSMS and data on the nutrient omposition of all food groups onsumed, olleted by the National Centre of Publi Health Protetion in Bulgaria. Notes: The figure highlights the total monthly alories averaged aross perentiles defined on the basis of per adult equivalent expenditures See text and Table 1 for a desription of perentiles determination. [25]

26 The pattern of prie elastiities of nutrients, highlighted in Figure 5 and Table TA11 is also onsistent with the rest of our desriptive statistis and empirial results. We see that, over time, the staple food prie elastiity of all maronutrients inreased signifiantly, while the meat prie elastiity of alories and fats went down between 1995 and This long-term pattern is onsistent with the logi of nutritional transition, haraterised by a permanent shift out of staple foods and arbohydrates into meat and the related proteins and fats (Popkin, 1993). However, the Figure 5: Seleted prie elastiities of nutrients alories, 10 th perentile alories, th perentile alories, 90 th perentile proteins, 10 th perentile proteins, th perentile proteins, 90th perentile arbohydrates, 10 th perentile arbohydrates, th perentile arbohydrates, 90 th perentile fats, 90 th perentile fats, 90 th perentile fats, 90 th perentile [26]

27 Soure: Own alulations based on the LSMS and data on the nutrient omposition of all food groups onsumed, olleted by the National Centre of Publi Health Protetion in Bulgaria. Notes: The figure highlights the total monthly alories averaged aross perentiles defined on the basis of per adult equivalent expenditures See text and Table 1 for a desription of perentiles determination. hange of diretion of the meat prie elastiity of arbohydrates during the risis highlights the tendeny of households to shift out of fats/proteins into arbohydrates in the fae of dramatially inreasing meat pries (and vie versa) in periods of eonomi shoks. 6. Conlusion One of the most hallenging researh areas of eonomi and nutrition siene researh is the ability of individuals and households to smooth their onsumption and nutritional stream during food prie and general eonomi shoks. While potential nutritional and health impliations of the ombined shok of the reent global food and finanial rises have been among the main onerns of politiians and journalists, little rigorous researh has attempted to unravel their full omplexity. Although a few nutrition siene studies witness major hanges in nutritional behaviour during rises, hanges that have potentially important epidemiologial onsequenes (Ivanova et al., 2006), supporters of the permanent inome hypothesis postulate an ability of individuals and households to smooth their nutrient stream even during rises (Stillman and Thomas, 2008). Moreover, the eonomis literature tends to report relatively low food and nutrient elastiities, as well as evidene, highlighting greater willingness of households to ompromise on tastes than nutritional value over short enough periods of time (Behrman and Deolalikar, 1987). [27]

28 Using data olleted with fortuitous timing before, during and after a major marofinanial and food risis in Bulgaria we explored the dietary impat of not only falling real inomes in the ontext of hyperinflation and risis, but also of hanging relative food pries and the hanging responsiveness of different groups of people to these inomes and pries over six years of fundamental strutural reforms of the eonomy. Our results highlight large and dramatially hanging food and nutrient elastiities, whih hallenge the pereption of low and relatively stable prie and inome elastiities of food and nutrition. In the speifi ase of Bulgaria we find a trend onsistent with the logi of nutritional transition, expressed in a long term tendeny to substitute staple foods for meat, whih is reversed during the risis, possibly on aount of the inability of impoverished households to afford luxury goods suh as meat. Our analysis has several potential limitations related to the data used. While a rigorous eonometri methodology helps us overome the problem of absene of reliable prie data, this approah restrits our ability to fous on detailed food items, as opposed to broad food groups. In partiular, due to the need of dividing food expenditures by the orresponding food quantities to obtain unit value observations; we obtain missing values eah time a household does not onsume a partiular food item. To avoid this problem, we group items into seven broad food groups, unfortunately preventing us from getting potentially interesting information on the possible reshuffling of household onsumption aross narrow food ategories. However, our desriptive statistis, e.g. Figure 1, show that the onsumption of nutrients, not onstrained by food groupings, indeed hanged over time. Despite these shortomings, whih plague the majority of the eonomis literature on nutrition, our paper is a signifiant ontribution to the both the aademi literature and related poliy debate. Our most important finding is that of dramati hanges in prie and inome [28]

29 elastiities of both food groups and nutrients and tendeny of households to realloate their onsumption streams during eonomi upheavals. Although this analysis is not based on a panel survey and hene does not trak the same households over time, we do trak over time households belonging to the same inome strata, whih is what matters from a poliy point of view. We highlight limitations of both studies that assume stable elastiities and base poliy advie on simulations that use household behaviour during a speifi past period of time as a point of departure and/or studies that onsider the welfare impliations of a prie shok fousing on one onsumption good at a time and ignoring the possibility of omplex realloations of the onsumer basket. [29]

30 Referenes Aker, J., Blok, S., Ramahandran, V. and Timmer, C. West Afrian Experiene with the World Rie Crisis, , 2011, Center for lobal Development Working Paper 242. Berhman, J. and Deolalikar, A. Will developing ountry nutrition improve with inome? A ase study for rural South India. Journal of Politial Eonomy, 1987; 95; Browning, M.J. and Meghir, C.D.H. The effet of male and female labour supply on ommodity demands. Eonometria 59, , Crawford, I., Laisney, F. and Preston, I. Estimation of household demand system with theoretially ompatible Engel urves and unit value speifiation. Journal of Eonometris, 2003; 114; Deaton, A. The Analysis of Household Surveys. A Miroeonomi Approah to Development Poliy. The International Bank for Reonstrution and Development, Development Researh roup. Lessons from World Bank researh on finanial rises, Poliy Researh Working Paper 4779, The World Bank, Diagana, B., Akindès, F., Savadogo, K., Reardon, T. and Staatz, J. Effets of CFA fran devaluation on urban food onsumption in West Afria: overview and ross-ountry omparisons. Food Poliy, 1999; 24; Eker, O. and Qaim, M. Analysing nutritional impats of poliies. An Empirial study for Malawi. IFPRI Disussion Paper 01017, Elsner, K. and Hartmann, M. Convergene of food onsumption patterns between Eastern and Western Europe. Institute of Agriultural Development in Central and Eastern Europe, Frankenberg, E., Thomas, D. and Beegle, K. The real ost of Indonesia s eonomi [30]

31 risis: preliminary findings from the Indonesia Family Life Survey, Rand Labor and Population Program Working Paper Series 95-04, Huang, K. Nutrient elastiities in a omplete demand system. Amerian Journal of Agriultural Eonomis, 1996; 78; Ivanova, L., Dimitrov, P., Ovharova, D., Dellava, J., Hoffman, DJ. Eonomi transition and household food onsumption: a study of Bulgaria from 1985 to Eonomis and Human Biology, 2006; 4; Kodde, D., Palm, F. and Pfann,. Asymptoti least-squares estimation effiieny onsiderations and appliations. Journal of Applied Eonometris, 1990; 5; Lewbel, A. Stohasti Hiksian aggregation with an appliation to grouping goods without separable utility. Annales d Eonomie et de Statistique, 1993; 29; Lewbel, A. Aggregation without separability: a generalized omposite ommodity theorem. Amerian Eonomi Review, 1996; 86; MKelvey. Prie, unit value and quality demanded. Journal of Development Eonomis, 2011, 95, Popkin, B. Nutritional patterns and transitions. Population and Development Review, 1993; 19; Prais, S.J and H.S. Houtakker, The Analysis of Family Budgets. Cambridge University Press, Cambrige, Sahn, DE., Younger, SD. And Myerhoefer, C. Rural poverty in Bulgaria: harateristis and trends. Mimeo, College of Human Eology, Cornell University Ithaa, Stillman, S. Health and nutrition in East Europe and the former Soviet Union during a deade of transition. A review of the literature. Eonomis and Human Biology 5, , [31]

32 Stillman, S., and Thomas, D. Nutritional status during an eonomi risis: evidene from Russia. Eonomi Journal, 2008; 118; Zezza, A., Davis, B., Azzarri, C., Covarrubias, K., Tasiotti, L. and Anriquez,. The impat of rising food pries on the poor., Paper presented at the International Assoiation of Agriultural Eonomists, Beijing, [32]

33 TA1: Definition of variables and desription of goods APPENDIX Variables Definition Mother language of head Mother language of the head; 1 if Bulgarian, 0 otherwise Age of head Age of the head in years Age of head 2 /100 Age of the head square divided by 100 Male head Sex of he head: 1 if Male, 0 otherwise No shool/elementary eduation of head No studies, day-are, elementary or preshool of the head: 1 if yes; 0 otherwise Seondary / middle general eduation of head Middle shool or general seondary eduation of the head: 1 if yes; 0 otherwise Tehnial /voational eduation of head Tehnial or voational seondary eduation, or other oupation-speifi eduation after seondary of the head, inlude ollege (e.g. nurses, polie): 1 if yes; 0 otherwise University of head University eduation of the head: 1 if yes; 0 otherwise Married head Marital situation of the head: 1 if married, 0 otherwise Urban Residene loation; 1 if urban, 0 otherwise Household size Total number of household members Owner-oupier Owner oupies the house: 1 if yes, 0 otherwise Spae per person Area of the dwelling in sqm/ divided by total number of persons oupying the dwelling Car or motoryle Have a ar or motoryle: 1 if yes; 0 otherwise Freezer Have a freezer: 1 if yes; 0 otherwise Automati washing mahine Have an automati washing mahine : 1 if yes; 0 otherwise Total number of leisure durables Total number of leisure durables (olour TV, video reorder, paraboli antenna, stereo, radio, personal omputer) ln(total expenditures) log total expenditures of food ln(tobao) log expenditures of tobao ( igarettes and tobao) ln(hygiene) log expenditures of hygiene produts and servie and personal produts (toilet soap, luxury toilette soap, shampoo, onditioner, shampoo and onditioner, hand ream, hydrating lotion, fae ream, leansing ream, deodorant, tooth paste, hair ut, hygieni servies, purhased wash soaps, value of made soaps, washing powder, bleah, dishwashing soap, other washers, other leaners, hild are-baby sitting) ln(energy) log expenditures of energy ( distrit heating, eletriity, gas, oal, oil, wood, other energy soures) ln(transport and ommuniation) log expenditures of transport and ommuniation (gas and oil, ar servie, maintenane, taxi, tram and buses, trains-outside ity, mail servie, telephone) ln(rereation) log expenditures of rereation (ultural ativities, books, newspapers, stationery, membership fees, pet food and expenses) ln(housing) log expenditures of housing (water and rent) ln(loths and shoes) log expenditures of loths and shoes (textile, loths, and shoes) ln(furniture) log expenditures of furniture (kithen equipment, home repairs, furniture, bedding, sheets, others) ln(health) log expenditures of health ( dentist, dotor, hospital/sanatorium, mediines, mediations, optial equipment, osmetis, others) No tobao No expenditures of igarettes and tobao: 1 if no expenditures, 0 otherwise No hygiene No expenditures of hygiene and personal produts: 1 if no expenditures, 0 otherwise No energy No expenditures of energy: 1 if no expenditures, 0 otherwise No transport and ommuniation No expenditures of transport and ommuniation: 1 if no expenditures, 0 otherwise No rereation No expenditures of rereation: 1 if no expenditures, 0 otherwise No housing No expenditures of housing: 1 if no expenditures, 0 otherwise No loths and shoes No expenditures of loths and shoes: 1 if no expenditures, 0 otherwise No furniture No expenditures of furniture: 1 if no expenditures, 0 otherwise No health No expenditures of health: 1 if no expenditures, 0 otherwise Share bread Share of expenditures of bread Share starhes Share of expenditures of starhes (maize flour, wheat flour, pasta, rie, beans, potatoes, arrots, lentils, sweet peas) Share vegetables and fruits Share of expenditures of vegetables and fruits (tomatoes, eggplants, onions, squash vegetables, leafy vegetables, peppers, abbage, uumbers, oranges, apples, pears, bananas, nuts, grapes, watermelon, melon, strawberries, herries, anned fruits, and anned vegetables) Share meat Share of expenditures of meat (veal and beef, pork, lamb, hiken/birds, sausages/sala, baon, anned meat, ground meat) Share fats and oils Share of expenditures of fats and oils (butter, margarine, lard, olive oil, vegetable oil,) Share dairy Share of expenditures of dairy (fresh milk, white heese, yellow heese, yogurt, powder milk, eggs) Share sweets Share of expenditures of sweets (sugar, jam, honey ) ln(quantity) log quantity (of eah food) Other foods Fresh fish, frozen fish, anned fish, ondiments and spies (salt, spies, offee, tea, others), drinks (water, wine, beer, Bulgarian liquor, hard liquors, other drinks), prepared food (not at home) [33]