Consumption Volatility, Marketization, and Expenditure in Emerging Market Economies

Transcription

1 Consumption Volatility, Marketization, and Expenditure in Emerging Market Economies Daniel L. Hicks Department of Economics University of Oklahoma 5/25/2012 Abstract: Consumption and income in emerging market economies are significantly more volatile than in developed economies. A volatility decomposition using Mexico s ENIGH, a nationally representative household-level income and expenditure survey from 1984 to 2006, suggests that a significant portion of fluctuations result from the large expenditure share devoted to goods and services with a low cost of substitution between market and home inputs such as food provision. During shocks, expenditure is found to be more volatile than direct measures of household consumption. I show that this result holds both at the aggregate level over the business cycle and, using pseudo-panel analysis, at the household level over unemployment spells. Using expenditure as our measure of consumption over the business cycle creates marketization bias, generating excessive levels of aggregate volatility and overstating differences in volatility across countries. Keywords: marketization; household production; consumption; expenditure; bias; volatility; JEL CODES: E01, E2, E3, O11, O47, O54 Acknowledgments: I am especially grateful to Chang-Tai Hsieh, Charles Jones, and Edward Miguel for their encouragement and suggestions. I am indebted to two anonymous referees for valuable comments. I would also like to thank Pierre-Olivier Gourinchas, Kevin Grier, Joan Hamory Hicks, Gordon Hanson, Ethan Ligon, Richard Lyons, Alex Rothenberg, John Tang, and numerous seminar participants for helpful comments and discussions. Tilsa Ore provided valuable research assistance. I am grateful to the Instituto Nacional De Estadistica Geographica e Informatica for compiling the ENIGH survey data and for fielding questions on its use.

2 1. Introduction and Motivation Recent research in the US has documented that expenditure is a poor proxy for consumption when the opportunity cost of engaging with the market changes over the lifecycle (Aguiar and Hurst, 2005; 2007; 2009) and over the business cycle (Aguiar et al., 2011). This study confirms that the distinction between actual consumption and measured expenditure extends to Mexican households and demonstrates that substitution between home and market inputs leads to overstated aggregate volatility when consumption is measured by expenditure. 1 Furthermore, I provide evidence that this bias applies to a sizeable portion of the Mexican economy and is systematically larger among developing economies, confounding cross-country business cycle analysis. The idea that marketization can lead to overestimation of the rate of economic growth is not new. Deaton (2005) describes how, during periods of expansion, households shift resources from informal to formal market activities, leading to a greater share of economic behavior in the formal sector and generating positive growth bias in GDP statistics. I extend this line of reasoning to explain overestimation of aggregate volatility. I show that while Mexico has grown and become increasingly marketized over the past two decades, during economic contractions such as the Peso Crisis, households realized substantial declines in expenditure in patterns consistent with substitution away from the market sector and towards household time input in the consumption process. These reallocations imply that measured expenditure understates actual consumption of commodities during crises. Together, marketization and de-marketization shifts combine to exaggerate overall consumption volatility when measured by expenditure. Furthermore, marketization bias should be a larger problem for measuring consumption in emerging market economies because the typical expenditure bundle is more heavily skewed towards commodities for which household time is a viable alternative to market inputs in the consumption process, such as for food preparation. 2 For example, if households in both economies behaved in a similar fashion, the corresponding level of marketization bias in the US would be around one-third of the size of that in Mexico simply because of differences in the expenditure share devoted to food. Broadly, food expenditure for a typical emerging market household comprises two to three times the 1 Mexico is an ideal setting to examine fluctuations, enduring measured consumption volatility two and half times greater than the developed country average over the last two decades (as measured by the standard deviation of quarterly consumption expenditure). Similarly, Mexico s quarterly GDP over the past 20 years has a standard deviation roughly two times larger than the OECD average. 2 It is also plausible that the elasticity of substitution between home and market production would be highest in these economies where a range of consumption activities are not yet fully marketized. Similarly, among poor economies characterized by subsistence farming, a lack of opportunities to engage with the market may limit the scope of marketization changes. A growing set of estimates of this elasticity exist (see below).

3 services. 4 The home production literature provides existing theory concerning consumer behavior. In average budget share of their developed country counterparts, as would be expected by Engel's Law. Thus, marketization bias should systematically exaggerate the relative volatility of less developed economies. Because food expenditure can be a significant source of overall volatility (it alone accounts for close to 30% of total consumption variance in Mexico), this pattern may help explain the high level of consumption volatility in emerging markets relative to developed economies. 3 To estimate the impact of marketization bias on the measurement of fluctuations in Mexico, I decompose macroeconomic volatility using household survey data over the period 1984 to The bulk of consumption volatility can be attributed to changes in expenditure on non-durables and services. Non-durable volatility is primarily driven by fluctuations in food expenditure, nearly half of which is due to changes in expenditure on food consumed away from home. Volatility in expenditure on services, which accounts for over half of measured volatility, meanwhile appears to be driven in part by large changes in demand for a broad range of services with scope for household time input in the consumption process such as domestic service, childcare, and personal care models based on Becker (1965), commodities are composed of both market expenditures and time inputs. Individuals can substitute between these two components in the process of consumption. 5 Aguiar et al. (2012) shows that holding constant the marginal utility of wealth, when wages are growing, agents will shift toward market expenditures, and conversely when wage shocks lower the opportunity cost of time, agents will reallocate time to non-market production in the consumption process. 6 Adjusting between expenditure and time in consumption can be accomplished in a number of ways. For instance, within a specific service, such as childcare, households can substitute greater amounts of time in place of expenditure. They can do the same for a specific good such as a tortilla by purchasing the necessary inputs at different stages of processing, choosing between corn grain, corn meal, and fully prepared tortillas. Alternatively, households may simply substitute across 3 There are a number of intriguing differences between fluctuations in emerging markets and those in developed economies, including higher volatility of macroeconomic aggregates such as investment, consumption, and output in emerging economies (Backus and Kehoe, 1992; Lane, 2003; Agénor et al., 2000). Indeed, recent work has shown consumption is actually more volatile than income in emerging markets (Kose et al., 2003; Neumeyer and Perri, 2004; Aguiar and Gopinath, 2007; Raffo, 2008). 4 A significant portion is also attributable to swings in the estimated rental value of owner occupied housing. 5 This reallocation can be undertaken at either at the intensive margin through reduced market hours or at the extensive margin through lower participation in the labor force. 6 Aguiar et al. (2012) makes clear that the issue is complicated by household's ability to substitute across time as well. Formally, what is required for households to devote more time to producing commodities during recessions and less time during booms is for the elasticity of substitution between market inputs and time to be greater than the intertemporal elasticity of substitution. For this reason, the analysis here focuses on commodities for which the possibility of substitution of time for expenditure is large, such as food expenditure.

4 different but related commodities which vary in the level of market expenditure relative to time input required (i.e. choosing between dining out or preparing a meal at home). The evidence presented in this paper addresses several aspects of this theory. First, I document that over the past few decades, as wages and market employment opportunities have increased in Mexico, household consumption has become increasingly marketized. This shift is evident in long-term expenditure patterns, with households reallocating purchases away from goods and services requiring high levels of time input in favor of highly marketized substitutes, behavior consistent with an increase in the opportunity cost of their time. 7 Second, I show that households undertake this substitution even within the same product, as higher income households purchase more highly marketized versions of specific commodities. Furthermore, the shock of the Peso Crisis, which can be thought of as reducing the availability of well paying market jobs, temporarily reversed the marketization process (likely by lowering the opportunity cost to households of providing a greater extent of own time in the consumption process). In this instance, households reverted to purchasing less processed products such as corn meal in lieu of tortillas, as well as substituted across similar commodities with very different levels of market and own time inputs such as between restaurant meals and grocery store purchases. Finally, to explore differences between expenditure and direct measures of consumption, I show that nutritional intake is significantly less volatile than food expenditure, and provide descriptive evidence that the allocation of time spent on a range of household activities is consistent with time substitution in the consumption process. A set of macroeconomic models which incorporate home production have modeled substitution in the allocation of household resources during economic fluctuations (Benhabib et al., 1991; Greenwood and Herkowitz, 1991; Greenwood et al., 1993; and Baxter and Jermann 1999). These models assume a reasonable degree of elasticity in substitution between home and market production. Estimates of this value exist, with the most relevant being Aguiar et al. (2011) which finds a value of 2.5 in the context of business cycle fluctuations in the US. 8 Given the sheer size 7 The marketization process is akin to the process of Smithian growth where during expansions individuals become more highly specialized and more heavily engaged with the market. 8 A number of estimates for this elasticity of substitution exist including Rupert et al., (1995), McGrattan, Rogerson and Wright (1997) and Aguiar and Hurst (2007). See Aguiar et al., (2012a) for a summary of the literature.

5 difference of the informal sector and the comparatively weaker market institutions in developing countries, it would seem plausible this value is higher in less developed countries. 9 Existing studies have documented declines in the consumption of basic necessities during large crises, arguing that individuals lack sufficient resources and mechanisms for coping with these shocks. For instance, in their study of the Indonesian Financial Crisis, Frankenberg et al. (1999) document a dramatic decline in both education and health spending. McKenzie (2006) finds a similar decline in both food and health spending in Mexico during the Peso crisis. 10 Reductions in human capital investments imply possible long-term costs to short-run volatility and accurately quantifying the impact of crises is essential for policy-making. 11 This paper suggests that expenditure based measures of consumption during shocks overstate the extent to which final consumption of many commodities declines, and reliance on these expenditure metrics mask other important costs faced by the household such as large time reallocations within the household. 12 The remainder of this paper is organized as follows. Section 2 describes the sources of nutrition, income, and expenditure data and presents some basic summary statistics. Section 3 documents changes in the level of marketization during periods of growth and over. Section 4 examines the importance of marketization bias in driving overall consumption volatility in Mexico. Section 5 compares nutritional measures of food consumption with those for food expenditure, revealing that in the presence of home production, measures based on expenditure overstate volatility relative to direct measures. Section 6 examines time use data to show that time allocation of Mexican households is consistent with household substitution between home and market goods across numerous sectors of the economy. Section 7 concludes. 2. Data Description 2.1 The ENIGH Household Income and Expenditure Survey 9 Cardoso-Lecourtois (2002) estimates that the elasticity of substitution between home and market goods in Mexico is significantly higher than in the United States. The inclusion of these values for substitution between home and market goods into an RBC model suggests that home production can explain over two-thirds of observed consumption volatility in Mexico. 10 McKenzie shows that households coped with the shock primarily by changing the composition of consumption, increasing the share of spending on food to an even greater degree than would be predicted by Engel s Law. Divergence between the path of actual consumption and consumption expenditure may help explain observed changes in food expenditure which are larger than those predicted by theory. 11 Morduch (1995) argues that consumption and income smoothing are substitutes, and that efforts to estimates of the cost of consumption smoothing will be understated if risk aversion drives individuals to smooth income as well. Consumption smoothing via income smoothing might entail lower risk taking in production, employment, and investment decisions, all of which can reduce longer term income and profitability. 12 In this regard, this research relates to the existing literature which examines differences between household surveys and national accounts in the measurement of economic activity (Ravallion, 2003; Deaton, 2005).

6 Mexico s Encuesta Nacional de Ingresos y Gastos de los Hogares (ENIGH) is a nationally representative survey of income and expenditure carried out by the Instituto Nacional de Estadística Geografía e Informática (INEGI). The ENIGH surveys are highly detailed household-level questionnaires that record both monetary and non-monetary transactions for over 500 distinct goods and services. Households also report transfers and detailed income information by source. Additionally, information is collected on a range of individual characteristics such as age, occupation, employment status, hours worked, and educational attainment, as well as detailed records on characteristics of the residence and on household asset ownership. Comparable ENIGH surveys have been carried out at irregular intervals for the past two decades. In total the surveys provide eleven rich cross-sectional observations: 1984, 1989, 1992, 1994, 1996, 1998, 2000, 2002, 2004, 2005 and With the exception of the 1984 survey which includes just over 4,700 households, the surveys range in coverage from 10,500 to 23,805 households. There is no panel component, and enumeration occurs during the third quarter of each survey year, making the surveys comparable but restricting ability to explore seasonal variations or track individual households over time. The surveys also lack precise geographic information, limiting the scope for regional comparisons within Mexico. The 1994 and 1996 ENIGH surveys bracket a severe economic shock in Mexico, the Peso Crisis. The latest in a series of exchange rate shocks for Mexico, the aftermath included a large decline in stock market valuation, a more than 5% percentage point rise in the unemployment rate, and a rapid run up in prices. In addition, households experienced a 23% decline in consumption expenditure over the crisis resulting from both falling income and rising prices. Inflation over the period was enormous, averaging 2.5% a month yielding an 86.5% increase in prices in just two years. Prices for food more than doubled. Since the crisis period, Mexico s central bank performance has improved, with inflation moderating in recent years. In calculating consumption and income, I correct for the introduction of the new peso in 1993 and deflate the surveys using the National CPI from September of each survey year. 13 Looking more broadly over the ENIGH survey period, one would characterize households in Mexico as having experienced rather rapid economic growth and a number of large shocks. The high level of both consumption and income volatility can clearly been seen in Figure 1, with 13 This practice follows McKenzie (2006), although he employs the basic needs CPI which he argues more accurately captures welfare changes over the crisis period because prices increases disproportionately affected the consumption basket of low-income households. I compare aggregate changes in the data to those in the national accounts over a longer period and employ the National CPI which seems likely to provide greater comparability in this context. Much of the analysis employs expenditure shares for which the results are largely insensitive to the choice of deflator.

7 slowdowns evident from , and , although the 1995 Crisis is clearly the most dramatic. Movements in consumption exhibit a high degree of co-movement with those in income. In contrast to evidence from national accounts figures, consumption swings in the household data are smaller than those in income. Several demographic and socioeconomic changes occur over the 22 year survey period. The majority of households are headed by males, although this fraction declines steadily over time. Mean age of household head rises from 44.5 to 47. Household heads tend to be literate and to have increasing educational attainment. Questions on literacy were dropped from surveys in later years, but for the years during which the information was collected, it was consistently above 85% and rising. The majority of households are nuclear, with a clear downward trend in household size over time, falling from roughly 5 to 4 over the period. These trends are consistent with the maturation of a still relatively young population. The extent to which declines in household size affect household economies of scale and thus influence the cost of providing home production relative to purchasing market consumption is likely a valuable avenue of future research. 2.2 USDA Food and Nutrient Database for Dietary Studies ENIGH households were asked to keep consumption logs for one week detailing daily food transactions, including both prices paid and quantities purchased. These logs cover expenditure on over 200 distinct categories of food and drink purchased by Mexican households. I create a database of household-level nutrient availability by linking the quantity of each food and drink purchased to nutritional information from the USDA s Food and Nutrient Database for Dietary Studies 3.0 (USDA, 2008). There are a number of limitations of this constructed database that should be noted. First, for the subset of meals consumed outside the home, only total expenditures (and neither quantities nor specific foods consumed) are recorded in the ENIGH consumption logs. In order to estimate caloric and nutrient values for these purchases, I follow the practice of Subramanian and Deaton (1996) of computing the average price per calorie (in this case the average price per nutrient) from the current household food consumption bundle of food purchased for the home and apply a 50% premium since restaurant meals are likely to be more expensive than those at home. Expenditure is then deflated by these nutrient prices to compute estimated calories and nutrients consumed outside the home. While useful, this estimation technique is likely to overstate the volatility of calories relative to consumption expenditure, as households are likely to consume cheaper meals out during

8 shocks and splurge on more expensive restaurants during times of plenty. Fortunately the direction of this bias would serve to make expenditure smoother than nutrition and thus suggest any finding to the contrary would be understated. 14 Similarly, this analysis will be unable to address issues of intra-household food distribution. 15 Consumption logs only provide information on food and drink spending for the household as a whole, revealing nothing about which household members actually consume those specific expenditures. Nevertheless, we can recognize the role of life-cycle dietary needs and employ adultequivalent household figures (in which household members under age 16 are treated as 0.5 adults). Another limitation of the database constructed here is that food quality can only be observed across categories, and while food categories are highly disaggregated, this likely misses out on changes in food quality. For instance, food logs distinguish between many different cuts of steak, but not between purchases of lean or regular ground beef. To some extent some of this substitution will occur on the price margin, and be captured by the fact that households get a larger quantity for or purchase a cheaper food substitute when times are tough. Section 5 provides evidence that households substitute across goods with various levels of market processing, but it is likely that actual household substitution is more complex than revealed by the ENIGH data. A final concern is the issue of wastage. Measures derived from merging the ENIGH food expenditure logs and the USDA database capture caloric and nutrient availability to the household, not actual intake. One factor driving a wedge between the two is the amount of food wastage which occurs. While wastage is generally regarded as a more important issue in developed economies, it is likely to increase with income. Further, wastage is liable to increase during good times and decrease during hard times, meaning that the true path of caloric intake should be smoother than the observed path of caloric availability. Therefore, this issue could result in overstated volatility in calories in the subsequent analysis. Once again, the direction of this bias works in my favor, and will only attenuate any finding that nutrition is smoother than expenditure in Mexican households. 3. Growth, Volatility, and Marketization Bias in Mexico and Beyond This section examines whether aggregated measures of consumption expenditure are consistent with the marketization hypothesis and compares the case of Mexico to other economies. 14 On the other hand, the presence of countercyclical mark-ups in restaurants may lead this approach to understate volatility because during recessions the mark-up would actually increase and the amount of nutrients per expenditure would then fall (although, a similar offsetting countercyclical mark-up may exist in grocery stores as well). Additionally, it is likely that the average dinner party size and the availability of restaurants also vary over the cycle making too much reliance on this information problematic. 15 It is however possible to address intra-household time allocation, which will be examined in Section 6.

9 Specifically, do we observe increased marketization during periods of growth and demarketization during shocks? Changes in the pattern of income and consumption among ENIGH households over the period are shown in Table 1 and are suggestive of a growing economy becoming increasingly marketized over time. Column 1 reports that consumption expenditure nearly doubles over this period, with households experiencing an average annual growth rate of 2.2% for real consumption expenditure and of 2.3% for real income. Changes in the source of income are also consistent with increased marketization as net compensation of employees, a measure of market wages and salaries, rose as a share of total income from 64% to 71% (not shown). As Mexican households become more affluent, the composition of the typical consumption bundle shifts from nondurable goods to services (columns 3-5). This shift is in large part due to declines in the expenditure share on food, which falls from 42% to 27% of the average household budget over the sample period. Market expenditure on many services traditionally considered to have a high elasticity of substitution between home and market production such as domestic service, personal care, infant care, and meals consumed outside the home increased over the period, both in terms of total consumption and in terms of expenditure share (as evidenced by the faster average annual growth rate for these consumption categories than for overall consumption). This trend is consistent with increasing marketization as the Mexican grew, where as market wages rise, the tradeoff between home and market inputs in the consumption process shifts more in favor of market goods. As a robustness check on the data quality of the 1984 and 2006 survey waves, these trends are evident when comparing the average expenditure share from the first 2 to 3 survey rounds and final 2 to3 survey rounds as well. Evidence concerning regression from the market can be seen by examining changes in expenditure over the Peso crisis (Table 1, column 6). From 1994 to 1996 expenditure on these same activities falls much more rapidly than overall spending. For example, expenditure on food out falls by 31% and expenditure on infant care by 69%, while overall expenditure declines only 23%. Some of these were offset by increases in expenditure share of substitutes with different levels of marketization. For example, while from spending on domestic services declines nearly 40%, the budget share devoted to purchasing household cleaning and care products increased, suggesting that households may be substituting their own labor for what was previously a service purchased in the market. These trends are consistent with a story of marketization during expansion and de-marketization during contraction.

10 To what extent does marketization bias extend across countries? Column 7 of Table 1 presents expenditure shares for a number of goods and services categories in the United States for 2006, and comparison with Mexico in 2006 (column 4) reveals stark differences. The expenditure share on food among Mexican households exceeds that of the US by more than a factor of two. This finding has broad implications for studies of cross-country differences in measured consumption expenditure. To the extent that household consumption is much smoother than expenditure, measuring food expenditure rather than consumption overstates actual consumption volatility for over 30% of the Mexican economy, a much larger share than for the US. This result is not limited to just Mexico and the US. Figure 2 presents food's share of consumption expenditure by income across a range of developed and less-developed countries. Food expenditure shares are declining in income in a systematic manner and the clear downward slope illustrates Engel s Law. Research by Regmi et al. (2001) has shown that lower income economies spend significantly more of their budgets on food (~47%) than both middle income (29%) and high income economies (13%). 16 If households in less developed countries are more likely to substitute own time for market time in the production of food in response to changing wages, then we might expect to observe larger expenditure responses to shocks and thus higher income elasticity of food expenditure in this settings. This is indeed the case, as low income economies additionally have much higher income elasticity for food expenditure (0.73) compared to middle (0.58) and high income economies (0.29). Most emerging market economies fall into the middle income category and have income elasticities roughly five to six times that of the U.S. (Seale et al., 2003). 4. Sources of Consumption Expenditure Volatility among Mexican Households The previous section presented evidence that the process of marketization is procyclical, rising during periods of growth and declining during downturns. It also suggested that on average, a sizeable share of the Mexican economy is exposed to marketization bias. To more rigorously assess how much of overall volatility is attributable to specific commodities such as food expenditure, total swings have to be decomposed into component fluctuations. This is because aggregate volatility depends not only on the expenditure share of components of consumption, but also on the extent to which these components move with the economic cycle. One simple way to decompose volatility is to think of aggregate fluctuations as a weighted sum of the fluctuations in the individual 16 Here, low income economies are defined to have <15%, middle income 15-50%, and high income > 50% of U.S. GDP.

11 components of consumption spending. Taking as weights their contribution to overall consumption growth between periods, the contribution of component i to an overall change in consumption over τ years can then be written as: CAAPC i C C i, t Agg, t C C i, t Agg, t C C Agg, t Agg, t 1 Figure 3 graphically displays changes in consumption between ENIGH survey rounds, broken out by broad expenditure category using this method. The first point of interest is the behavior of durable goods. Not only are durable goods purchases the first to be abandoned in a slowdown as is the case from , and , but it also apparent that consumers are merely delaying these purchases, as indicated by the explosive growth from and Cecchetti et al. (2006) argues that durable goods are frequently cited as an important factor driving consumption volatility in developed nations and McKenzie's (2006) analysis found that Mexican consumers responded to the crisis by postponing purchases of durables. 18 Given the high purchase price of many durables, liquidity constraints (which are more severe in less developed economies) may be a factor driving volatility in durables. 19 If Mexican consumers are merely postponing durable expenditures and not eliminating these purchases over the course of income shocks, this could drive high consumption volatility by depressing consumption during recessions and inflating consumption during expansions. Nevertheless, throughout the survey period, durables only comprise 6-7% of total consumption expenditure. Thus, even though they are individually much more volatile than non-durables and services, they still have only a small impact on overall changes. Instead, Figure 3 suggests the bulk of changes in consumption expenditure can be attributed to swings in food and service expenditures. This is even apparent during the Peso Crisis, where despite a rise in food s share of consumption, food expenditure, services and other non-durables comprise the vast majority of the expenditure decline Where CAAPC approximates the discrete contribution of component i to the average annual percentage change of aggregate consumption between survey waves. For instance, if the average annual growth rate of aggregate consumption between 1998 and 2000 was 4% and one-half of this increase could be attributed to growth in services expenditure, then the CAAPC of services would be 2 percentage points. 18 McKenzie (2006) provides a rigorous evaluation of the consumer response to the Peso Crisis and shows that Mexican consumers alter the timing and composition of their consumption behavior. I find that by the time of the 1994 ENIGH surveys, the economy had already begun to slow. This can be seen in Figure 3, where in 1994, prior to the full onset of the crisis, households begin postponing purchases of durables, especially big ticket items such as automobiles. Given the timing of the 1994 survey, some of the decrease in consumption could be attributable to consumer uncertainty surrounding the future of the economy or to the tightening of credit markets. 19 Similarly, sudden stops in foreign investment may externally impose credit crunches.

12 While the above disaggregation intuitively captures period by period fluctuations, it misses out on which components of the economy can create shocks through covariance. A second method of variance decomposition would be to analyze the individual contribution of a specific sector to aggregate volatility over a longer period using fundamentals from portfolio theory, where individual assets each contribute to an overall portfolio's volatility. Koenig and Ball (2007) show that a random variable I C which is the weighted sum of individual components C i, where the weights α i i i C i are constant, has covariance Cov(C,C i ) = C, C i C Ci, and overall volatility given by I C i C, Ci C. Assuming expenditure shares are relatively stable, we can calculate how much i i of aggregate volatility is attributable to each individual component of consumption by looking at the product of three factors: individual expenditure shares, correlation with aggregate consumption, and individual volatility. 20 Table 2 presents decompositions of consumption volatility from using each of the two methods detailed above. Both exercises paint a similar picture of the driving forces behind volatility in consumption expenditure. The results suggest that expenditure on food and alcohol alone accounts for more than a quarter of total consumption volatility, suggesting that for Mexico a sizeable portion of consumption volatility is directly attributable to expenditure for which the household can easily substitute between own and market inputs during the consumption process. 21 Assuming that this phenomenon extends beyond food preparation and into other activities with high potential for non-market substitution in the consumption process, then this fraction rises to roughly 1/3 rd once we include a small subset of additional services for which there is likely a large scope for substitution of home labor for market labor such as child care, personal care, and domestic service. 5. Food Expenditure and Consumption The volatility decomposition of the previous section suggests that any examination of consumption volatility should focus on the role of non-durables and services in driving fluctuations. This section undertakes several exercises which analyze the volatility of food consumption, the 20 Expenditure shares are in fact slowly changing over time, so as an approximation I employ the average expenditure share over the entire period. 21 Similarly, separate results (not reported) calculating variance and incrementally removing food expenditure from total consumption results in a decrease in variance of 25%.

13 largest component of non-durable purchases. First, households are shown to substitute between own time inputs and market production in the consumption of a specific commodity with identifiable levels of marketization, tortillas. Additionally, households appear to revert to versions requiring more time input during the Peso Crisis. Next, expenditure volatility across households is explored rigorously using pseudo-panel regression analysis to examine changes in consumption expenditure over a particular household income shock - unemployment spells. Finally, although nutritional intake is shown to vary with household resources, nutritional measures of food consumption are found to be significantly smoother than food expenditure both at the aggregate level over the business cycle and at the level of the household during unemployment. First, I examine substitution between home and market inputs in food provision as income rises. Perhaps the clearest way to examine household substitution of own time for market inputs within a specific commodity is to identify categories of food expenditure for which there is clear differentiation in the level of time input required and ideally a sizeable degree of quality homogeneity. Fortunately, the richness of the ENIGH surveys allows for such an exercise. Tortillas, particularly those made of corn, are a staple of the Mexican diet and can be purchased in various stages of finishing. To produce tortillas from scratch involves a rather intensive time input from the household. They must purchase corn grains, soak them in a lime-alkaline solution, rinse them, sometimes hull the kernels, grind the grains produce corn meal or dough, and then shape and cook them to finally produce the tortillas. 22 Alternatively households can purchase cornmeal flour or dough to cook directly, or purchase tortillas outright. Each increase in market based preparation reflects fewer household time inputs but also increasing cost. Over the entire sample, 81% of households reported purchasing some corn tortillas in the last week, and 12% reported positive expenditure on corn grains and 12% reported positive expenditure on corn meal or dough. Table 3 presents the results of semi-log regressions for the various production stages of corn and wheat tortillas, and meals outside the home. Specifically, I estimate a semi-log regression of the following form: E it ln( YPerm, it ) X where the dependent variable is an indicator taking on the value 1 when household i undertakes positive expenditure in category E during year t. Household income, Y has both a permanent and a transitory component. Following Aguiar and Hurst (2005), I instrument for current income with it it (1) 22 Processing of corn kernels through the use of the lime-alkaline solution to produce corn meal is known as Nixtamalization.

14 indicators for household head education and occupation, and a full set of interactions between the two, in order to elicit the permanent component of income variation. X is a vector of controls including gender of household head, number of young and old age dependents, number of men and women, adult equivalent household size, as well as an indicator for urban households and for each survey year. Regressions are run for the period because food, occupation, and education categories are consistent over this period (and not so for the 1984, 1989, 2005 and 2006 survey rounds). The sample is restricted to households with heads aged 35 to 50 who work full time (at least 40 hours per week), out of concern that the inclusion of young, elderly, or underemployed heads may yield non-standard elasticities owing to differences in preferences and demands over the lifecycle. Semi-elasticities estimated using equation (1) are reported in column of Table 3. The results are significant and consistent with the expected sign. Poorer households are more likely to consume raw corn grain, and wheat flour as well as slightly more likely to buy corn meal and dough. Higher income is associated with fewer purchases of these products and more expenses on finished corn and wheat tortillas. Specifically, a doubling in household permanent income is associated with a 12 percentage point decline in the likelihood of purchasing corn grain, a 4.1 percentage point decline in the likelihood of corn meal purchases, and a 7.6 percentage point increase in the likelihood of buying corn tortillas. As the bottom panel of Table 3 shows, these patterns extend beyond tortilla production to the marketization of food expenditure more broadly. A doubling in household income increases the probability that a household purchases meals away from the home by 16.5 percentage points and the probability that they spend a positive amount on prepared foods (such as frozen pizza or readymade meals) by 8.2 percentage points. Together, all of this indicates that households with higher income appear to substitute towards market labor in the place of their own labor, for example shifting from home consumption to consumption of meals outside the home. If households are substituting own time for market time in the production of tortillas, then in response to a shock to labor market opportunities such as presented by the Peso Crisis, we should observe a reallocation in market expenditures towards less processed products (those requiring a greater degree of household time input in the consumption process). Column 5 of Table 3 presents changes in expenditure over the crisis period from Household consumption expenditure is consistent with this behavior. Expenditure on corn grain, corn meal and dough, and wheat flour, actually rises significantly over the crisis period. At the same time, expenditure on prepared corn

15 and flour tortillas which have larger market inputs and on Nixtamal services (an activity which can be replaced with household time input) declines. As would be expected, reductions in expenditure on other highly marketized foods such as meals out and prepared foods are even larger. Reallocations in expenditure between goods and services with a low level of marketization and highly marketized commodities are also evident over the whole period. Recall, Table 1 shows that as household income in Mexico grew over , the share of food expenditure devoted to meals out increased and the share of food for home consumption decreased (column 5). Figure 4 shows the dynamics of this relationship by focusing on the decomposition of changes in food expenditure between survey waves. The procyclical nature of fluctuations in expenditure on meals out is evident. One can explore expenditure volatility across households more rigorously using regression analysis to examine changes in food consumption expenditure over a particular type of household income shock - unemployment spells. To do this, I perform an exercise analogous to that of Aguiar and Hurst (2005), who focus on changes in food consumption after retirement and unemployment in the US. Also, because the ENIGH data lacks a panel component and individuals may select into unemployment, I analyze such changes using a pseudo-panel, cohort based analysis. The biannual nature of the cross-sectional ENIGH surveys during the 1990s and early 2000s presents a natural setting to employ cohort analysis. I construct cohorts based on year of birth and educational attainment of household head, and utilize survey data collected during the period , when questions classifying educational attainment remained relatively consistent over time. The large sample size of the ENIGH allows for 5-year year of birth cohorts across 5 educational groups generating 40 cohorts per survey year with an average size of 277 households per cohort. 23 I assess the impact of household level unemployment shocks on expenditure by estimating the following regression equation: ln( C ) unemp it it X it J where, as before, i indexes households and t indexes survey years. The definition of household consumption expenditure (C) varies across different specifications to include total expenditure, food expenditure, and expenditure on food at home or food outside of the house separately. The independent variable of interest is the unemployment term (defined in terms of the status of the household head). X is a vector of socioeconomic and demographic controls including the gender, i it (2) 23 Results presented are insensitive to variations in the cohort definition such as variation in inclusion of particular sets of years of birth, or redefining cohorts as year of birth only, or year of birth / education / gender cohorts.

16 age, and education level of the household head, as well as adult equivalent household size, numbers of men, women, young age dependents, and old age dependents in the household, an indicator for whether the household is rural, and survey year dummies. J represents the year of birth, education cohort fixed effects. Table 4, Panel A presents the results of these regressions. The ENIGH surveys contain rich information about individual employment status, and each column examines a different definition of household head unemployment. In column 1, an individual is counted as unemployed if they are part of the labor force, but lacked work for any number of reasons including temporary work loss due to strikes, supply shortages, necessary equipment repair, and sickness. One concern is that household may view a bout of unemployment, particularly one due to a strike or seasonal factors, as highly transitory. In column 2, the definition is restricted to household heads that report both lacking work and active job search, a situation where the household faces greater uncertainty. Finally, since many employment changes occur at the margin, column 3 examines changes in expenditure occurring when the household head works less than full time. The results in columns 1 and 2 of Panel A are strikingly similar. Unemployment of the household head is associated with significant reductions in expenditure. Total expenditure declines 17 to 23% when the household head becomes unemployed, and expenditure on food declines more than 13 to 14%. The reduction in food consumed at home is smaller than the reduction in food consumed out, although this difference is statistically significant only for column 3. As one would expect, results in column 3 where the dependent variable is an indicator for full-time employment are smaller in magnitude, but otherwise substantively similar to the results in columns 1 and 2. The results in Tables 3 and 4 (panel A) together suggest that households substitute own production for market production during difficult times. Unlike the tortillas exercise, I cannot distinguish the between substitution within goods and substitution across goods in this case, although both are likely occurring. For example, households can buffer against shocks through changes in dietary composition, such as switching consumption from meat to beans in order to maintain protein more cheaply. Still, if households do smooth consumption in this way, then expenditure would likely not accurately capture true household consumption, and although welfare would likely be impacted, the expenditure based measure could still be misleading. As previously mentioned, the Peso Crisis provides an excellent instance to examine extreme volatility in an emerging market setting. While households reduced overall expenditure on food by nearly 25% between 1994 and 1996, food s share in total expenditure actually increased (see Figure

17 5, left panel). However, households dramatically reduced expenditure on meals away from the home (see Figure 5, right panel) and shifted expenditure away from relatively expensive foods such as meals out, meat, and fruit towards foods generally considered staples such as cereals and dairy. A larger share of calories after the crisis was derived from cheap sources such as cereals, tubers, and vegetables (results not shown). 24 Were Mexican households able to smooth nutritional availability over the Peso Crisis? Figure 6 presents average annual changes in adult equivalent food expenditure, calories, protein and a broad nutrient index from 1992 to It is immediately apparent that over the crisis households on average were better able to smooth calories, protein, and nutrient availability over the aggregate shock of the crisis than they were able to smooth expenditure. Whereas expenditure on food declined nearly 9% per year over the crisis, calories declined by 0.6% and the nutrient index declined by 0.29%. In order to compare expenditure and nutrition volatility more rigorously in a regression context, it is useful to first estimate the income elasticity of the typical Mexican household's nutritional profile. This will inform us as to which nutrients we should expect to see change in availability to the household during expenditure shocks. To calculate this elasticity, I estimate an instrumental variables regression similar to equation (1), where this time the dependent variable is the log of household level availability of a given nutrient. 26 As in the semi-elasticity regressions estimated previously, current income is instrumented in order to identify the permanent component of household income, X includes the same set of controls, and the sample is again limited to household heads aged who work full time. 27 Results of these income elasticity regressions are presented in Table 5. Both calories and protein exhibit low income elasticities, with coefficients of 0.08 and 0.10 respectively, indicating that their consumption does not dramatically rise with income. Vitamins and minerals appear more responsive to income changes, especially Calcium, Vitamins A, B12, and E. Surprisingly, Vitamin E exhibits negative income elasticity, which likely results from poorer households purchasing greater quantities of certain goods that are relatively rich in the nutrient such as cooking oil and beans. Surprisingly, "bads" such as sugars, cholesterol and fat exhibit relatively low income elasticities. 24 Regmi et al. (2001) suggests high income elasticity can be seen among for many food categories in lower income countries, suggesting that households are very willing to substitute across food categories. 25 The nutrient index includes household availability of Calcium, Iron, and Vitamins A, B12, C and E. The index is constructed by demeaning each nutrient and then weighting each equally. 26 I estimate these regressions using nutrient availability of food consumed at home only. 27 Additionally, the log of total household caloric intake is included as a control in all regressions other than the one in which calories are the dependent variable.