Life-cycle consumption of food in France: food expenditures and home production

1. Introduction

The ageing population is a major demographic change documented in Europe and elsewhere. According to the Global Health Observatory, World Health Organization, the life expectancy for children born in 2015 is 71.4 years globally and 82.4 years in France—almost two decades into retirement in the case of France. Whether the ageing population can maintain the quantity and quality of consumption post-retirement is, therefore, a timely question. The past evidence of a distinct drop of food expenditures after the retirement (Fernandez-Villaverde and Krueger, 2007; Moreau and Stancanelli, 2013; Aguila, Attanasio and Meghir, 2011; Attanasio et al., 1999; Gourinchas and Parker, 2002; Bernheim, Skinner and Weinberg, 2001) promptly highlights a sizeable and, apparently, growing problem. Moreover, by now the well-documented association between nutritional quality and some metabolic diseases (see the French dietary guidelines (2012), among others), compounded by a substantially higher healthcare cost for the elderly in France (Tenand, 2014), could make ageing a potential public health concern as well. In this paper, we focus on the former issue, namely, whether the French households can maintain the quantity of consumption through the life cycles.

Insofar as expenditures are identifiable with consumption, the abovementioned decline in expenditures could raise a legitimate concern whether the elderly population can maintain consumption, particularly post-retirement. However, the quantity consumed may remain intact or even increase in the face of declining expenditures if it is offset or more than offset by a decline in prices paid. Previous evidence shows that lower prices could be achieved through strategic shopping: use of coupons and store promotions, patronising more economically priced stores or store types, stockpiling, substituting towards inexpensive brands, buying in bulk, opting for lower quality substitutes, etc. (Aguiar and Hurst, 2005, 2007; Been, Rohwedder and Hurd, 2015; Nevo and Wong, 2015; Griffith, O’Connell and Smith, 2015). Lower prices for prepared foods could also be achieved through preparing foods at home rather than buying prepared foods, or engaging in what is known as home production (Becker, 1965). Previous research shows that, by engaging in home production, households could improve both the quantity and nutritional quality of consumption (Aguiar and Hurst, 2005, 2007; Griffith et al., 2009; Griffith, O’Connell and Smith, 2015).

In the strategic shopping and home production discussed above, the substitution of money and time is fundamental (Becker, 1965; Ghez and Becker, 1975; Stigler, 1961; Aguiar and Hurst, 2007; Griffith, O’Connell and Smith, 2015; Nevo and Wong, 2015). In other words, households engage in either practice to achieve lower prices, subject to the relative availability of either resource. The life-cycle consumption setting provides the appropriate backdrop against which we outline the varying availability of these two resources throughout the life cycles and the resulting consumption choices. Indeed, we show that households in the age group of 70–75 years old do increase time spent in home production and strategic shopping by 11 per cent and 71 per cent, respectively, relative to the reference group of 25–29 years old households, enjoying 4 per cent and 2 per cent lower prices as a result.

Although previous literature aptly captures the trade-off between time in strategic shopping and price savings in modelling life-cycle consumption, the trade-off between time and money in home production, or the value added or created in the process of the home production, captured by the difference of the higher cost of prepared foods and lower cost of market inputs¹, has not been factored in the home production models previously (Becker, 1965; Ghez and Becker, 1975; Aguiar and Hurst, 2007; Griffith, O’Connell and Smith, 2015)². In other words, the previous research focused only on the variation of shopping time/effort as the sole source of variations in price (Aguiar and Hurst, 2007; Nevo and Wong, 2015; Griffith, O’Connell and Smith, 2015). This study, to the best our knowledge, is the first to exploit variations in price due to both time spent in shopping and in home production as the source of variation in price. We use this total variation in price to impute life-cycle consumption produced off of food expenditure. We find that households, as they age, produce increasingly more consumption off each euro spent on food, culminating with 70–75-year-old households generating 17 per cent more consumption than households in late 30’s.

To develop a price index capable of reflecting variation due to strategic shopping we move beyond the narrow identification of a food at the most disaggregate or Universal Product Codes (UPC) level that would enable the capture of price variation of switching brands, buying in bulk, etc. of closely related foods. For example, instead of using the UPC for a 1-l coke as the identification of the food, we use a broader identification like ‘Soft Drinks’ that houses both cokes in different sized containers as well as soft drinks of different brands. To be able to capture price variation due to home production, we broaden the food categories to also include prepared foods (Z-goods) with their market inputs. In this approach, raw dry spaghetti would be in the same category as cooked fresh or frozen pasta-based dishes. We create the food categories by identifying the ingredients of each prepared food purchased and combining the prepared foods in the same group with their main ingredients. The homogeneity of this categorisation clearly plays an important role. Hence, we use three different levels of aggregation to construct the price indices to ensure the robustness of our findings (Table A1 in Appendix).

While it has been previously demonstrated that the use of demographic variables and time effects would improve the explanatory power of the intertemporal consumption models, the past literature on consumer expenditure typically abstracts from time, region and cohort effects (Attanasio et al., 1999; Fernandez-Villaverde and Krueger, 2007; Boissinot, 2007). Attanasio et al. (1999), for example, demonstrate that the inclusion of simple demographic information helps reconcile the theory of permanent income by Hall (1978) and the empirical evidence. Boissinot (2007) argues that changes in the household size account for half the expenditure hump in France. Fernandez-Villaverde and Krueger (2007) argue that it is key to control for the cohort effects as well since, in the face of increasing real wages and aforementioned life expectancy, a 35-year-old individual in 2014 would have higher discounted lifetime earnings and, therefore, face different consumption possibilities and make different consumption choices than a 35-year-old individual in 1960, all else equal. We exploit the panel nature of our data to fully explore the time, region and cohort effects, in addition to the age effects, on both the life-cycle expenditures and implied consumption.

Finally, we follow and build on the literature by providing a profile of the life-cycle food at home consumption in France. While previous work addressed the intra-household allocation of time to work and non-work related activities (Bourguingnon and Chiuri, 2005; Chiappori, Fortin and Lacroix, 2002; Rapport, Sofer and Solaz, 2011) or the life-cycle evolution of food expenditures (Moreau and Stancanelli, 2013; Boissinot, 2007) in the country, to the best of our knowledge, no previous effort of empirical analysis of the life-cycle evolution of food consumption level in France exists.

The rest of the paper is organised as follows: In Section 2, we set up the conceptual framework upon which we build our model. Section 3 describes the data we use for this analysis. In Sections 4 and 5 we display the life-cycle patterns of time spent in shopping and home production and the associated returns. In Section 6, we discuss the price estimation strategy. The life-cycle consumption model along with the estimation results are discussed in Sections 7 and 8. The concluding remarks and recommendations appear in Section 9.

2. Conceptual set-up

We model the trade-off between time and money in a constrained expenditure minimisation set-up. At each time period, households spend time and money to maintain a certain level of consumption. The trade-off between these two components reflects households’ ability to achieve lower prices for food market inputs, holding consumption fixed. Following Aguiar and Hurst (2007) and Griffith, O’Connell and Smith (2015), we also condition the price upon a set of factors not directly related to time—|$N$|⁠. The elements of the vector |$N$| include the average number of items purchased per trip—|$n$|⁠, the quantity of a composite market good—|$Q$| and the fraction of items purchased in bulk—|$a$|⁠. The first two reflect the effects of bundling or lower per-item search and shopping time. As such, it would be expected to have a positive effect on the price paid. The fraction of items purchased in bulk reflects the quantity discount or the bulking effect that, all else equal, is expected to be associated with lower prices. We entertain an alternative measure—household size, in Table A5, in Annex.

where |$Q$| is the quantity of the composite market good; |$P$| is the price paid; |$s$| and |$h$| are the time spent in shopping and home production, respectively; |$N$| includes the number of items purchased |$n$|⁠, the quantity of the composite market good—|$Q$| and the fraction of items purchased in bulk—|$a$|⁠; |$\mu $| represents the cost of time, and |$c$| is a parameter representing the desired level of consumption³. Regular concavity conditions of the production function |$f\left( {h,Q} \right)$| are implied. Note that to allow the abovementioned trade-off between the market expenditures and time, we condition the price |$P$| on time |$s$| and |$h$|⁠, referring back to the discussion above.

Conditions (3) and (4) are the optimality conditions for the shopping and home production time. The former equates the benefits of extra shopping to the cost of extra shopping. The latter equates the returns to home production to the cost savings due to purchasing more lower-priced ingredient foods and the value created at home production. As mentioned above, all else equal, we would expect that as time becomes a more abundant resource and the opportunity cost of time declines with ageing, the households could spend some of that time in shopping and home production.

3. Data

We use grocery purchase data from Kantar WorldPanel (henceforth Kantar) from a nationally representative panel of participants that scan the bar codes of the grocery purchases year-round, from 1998 to 2016. The survey provides a rich set of information categorised in three areas: products, purchases and household/panelist demographics. The information on products covers the producer and the brand, a product identification number akin to UPC, as well as a complete physical description of products (colour, flavour, package size, etc.). The area concerning purchases provides information on the expenditures and the quantities purchased for each food item, the shopping venue (the retailer name, typically), the type of the store (hypermarket, supermarket, specialty store, etc.) and shopping trip date, complete with trip identification numbers. Also, an annual survey collects individual- and household-level demographic information concerning the age, year of birth and gender, area of residence, household income, household size, number of children, etc.

We applied a two-tier censoring to the data. First, households with reported purchases for at least 10 calendar months per year were retained to eliminate the seasonality effects and ensure the quality of the reported data. Second, we restricted our sample to households with the panelists aged between 25 and 75 years old only. The upper limit on age (less than 75) is imposed to ensure that the preferences, rather than possible health issues, driving the food choices. The lower limit is imposed to retain only the households with more or less established purchase habits. The resulting sample is an unbalanced panel of 44,418 households with 204,743 household-year observations. In our sample, on average, 89.9 per cent of the panelists are women, approximately 34 per cent of households have at least one child, and average household size is 2.72 persons. The mean age is 48.5 and the average length of the stay in the panel is 4.6 years.

To account for the time, age and cohort effects, we identify 10 five-year cohorts and 10 five-year age groups⁴. Using the geographic area or market indicators, we defined the interaction of 12 regions (official geographical units in France) and four levels for the urbanisation level of the residential areas—Paris metropolitan area, urban areas with more than 200,000 population, and urban areas less than 200,000, to generate 36 market indicator variables.

Despite the detailed information concerning individual purchases, our data contain no information on the duration of shopping trips or the time spent in strategic shopping, in general. To construct an estimate of time in shopping we considered several alternative measures including the number of total shopping trips; the number of trips to a large-format store (supermarkets and hypermarkets with the area larger than 250 sq.m), as they are fewer and sparsely distributed and would require more time to reach, all else equal; the ratio of the trips to large stores per unit of time, per quantity of food or per number of foods purchased, etc. While all of these are logically appealing as time measures (Aguiar and Hurst, 2007; Nevo and Wong, 2015; Griffith, O’Connell and Smith, 2015), they also have serious shortcomings with unpredictable effects. For example, trips to supermarkets are fewer but larger quantities and varieties are purchased, making it less indicative of the effort made at obtaining a food item. Similarly, ratios such as the proportion of trips to large stores fail to distinguish one trip out of two from two trips out of four, the latter of which would be associated with more time, all else equal.

In the absence of information related to the time needed for preparations of a food item, we use products’ physical descriptions to identify ingredients to be used as a proxy to time spent in home production. The ingredient foods are identified as foods that satisfy two conditions: (i) they need thermal processing (raw beef) to be edible and (ii) they are composed of only one ingredient (head of lettuce). Note that using one or the other condition would under- or overestimate home production, respectively. If using condition (i) only, we will miss all foods that normally require time to prepare without cooking—washing, cutting and mixing a green salad, for example. If using condition (ii) only, we would prescribe all foods we eat by themselves, e.g. cheese, as ingredients which would overestimate home production. Finally, to allow comparison across households that purchase different quantities of market foods, we express ingredients as the fraction of all foods purchased. In our sample, on average, 28.37 per cent of all foods purchased were ingredients.

Finally, considering that all continuous variables are bounded from below by zero the distributions of these variables would be skewed. In the face of naturally occurring zero-valued dependent variables, we opted for the inverse hyperbolic sine (IHS) transformation—a logarithmic-like transformation that allows us to keep zero-valued observations and interpret coefficients as elasticities (Bellemare, Barrett and Just, 2013). Under the IHS transformation, each variable, |${y^*} = ln \left( {y + {{\left( {{y^2} + \theta} \right)}^{{\raise0.7ex\hbox{$1$} \!\mathord{\left/ {\vphantom {1 2}}\right.} \!\lower0.7ex\hbox{$2$}}}}} \right)$|⁠, where |$\theta = 1$| is set customarily.

4. Shopping and home production patterns over life cycle

In the preceding discussion, we motivated the use of strategies that entail a substitution of money and time to smooth life-cycle consumption. In this section we document the life-cycle change of the degree households engage in these strategies, namely, the time spent in shopping and home production.

To substantiate our claim that there exists a link between our time measure in (6) and the actual shopping time, we compare life-cycle trajectory of our measure to that of shopping time (in minutes) from the Time Use and Decision-making within Couples Surveys (Enquête Emploi du Temps et Décisions dans les Couples or EDT) that provide detailed accounts of time use in France in 1999 and 2009. The graph appears in Panel a, Figure 1, depicts the parameters from regression of time measures (in IHS form)—shopping time measure in (6) to on dummy variables for the panelists’ age groups. The corresponding parameter estimates appear in Table A2, in Appendix.

Fig. 1.

Life-cycle shopping and home production time.

Open in new tabDownload slide

Notes: The data come from Kantar WorldPanel and EDT. The graphs are created from random effects regressions of time measures (in IHS form)—shopping time measure in (6) to (in Panel a) and fraction of ingredients (in Panel b) on dummy variables for the panelists’ age groups. The vertical axis measures the log deviation from the reference group of 25–29-year-olds.

Despite the fact that the two databases have different time spans, the graph clearly indicates that households engage in shopping more extensively as they age. Specifically, the similarity of the trajectories of shopping time from EDT and of our measure until the early 60’s is apparent. Post 60’s, while our measure increases monotonically through life cycles, the EDT measure plateaus during 60’s and drops afterwards. There could be several factors contributing to this. The nationally representative samples in both surveys and our age cut-off of 75 years and below, as we discussed in the Data section, possibly subdue self-selection bias, if any, due to mobility or other age/health related conditions. In this light, it is probably more likely that this drop is due to methodological factors. For example, the shopping time in EDT includes all, not just grocery shopping, and ageing could have a dampening effect on non-food shopping (clothes, etc.).

There is also a perceptible difference between the scales of the two curves. Our measure of time could possibly overestimate the time in shopping by not taking into account the effects of shopping efficiency or experience and combining or bundling shopping trips. Nevertheless, the near-parallel trends of the curves in the graph testify to the validity of our measure of time spent in shopping.

To measure time spent in home production, we utilise the information on the degree of value added for each product purchased as an indicator of commitment to spend time to add value at home (Panel b, Figure 1). For this analysis, we use a binary indicator whether a food is an ingredient or not and use the fraction thereof to proxy the time spent in home production. To validate our proposed proxy of time spent in home production, we perform the same exercise as with the shopping time above. The resulting graphs are depicted in Panel b, Figure 1. Despite the differences in scales, the trajectories from both data sources are remarkably similar. Relative to the reference age groups of 29 and younger, the share of ingredients in Kantar indicates 15.2 per cent growth at the age of 70–75 compared to almost 80 per cent growth in home production time in EDT (Columns III and IV). The large gap in the scales is likely attributable to the binary interpretation of the continuous degree of value added. Nonetheless, attenuated as it is, the close resemblance between the two series is apparent.

In sum, while the factors discussed above and others, undoubtedly, unaccounted for in this analysis help shed light on the gap between these trajectories, the near-parallel trend demonstrates the indelible link between them.

5. Returns to shopping and home production

The price index in (8) measures the variation in price, holding the purchase basket fixed. To be able to exploit the price savings due to substitution between money and time, the choice of the purchase basket |$i$| in (8) is very important. Defined too narrowly, say at the UPC level, the index in (8) will not be responsive to some shopping strategies that require substitution with close substitutes with different UPCs (e.g. the same product in a different-sized package), or purchase of ingredients for production at home. Defined too broadly, the price index in (8) is susceptible to capturing variation in price of unrelated foods that would result in counterintuitive outcomes. In other words, to capture the variation in price of prepared foods and their unprepared ingredients, the product basket needs to be more homogeneous to avoid comparison of some expensive unprepared foods (e.g. seafood) and cheaper prepared foods (e.g. potatoes).

As mentioned in the Data section, Kantar provides product information beyond the UPC—product description (e.g. gratin dauphinois) and recipe (e.g. gratin) that could be used to define baskets in (8). These levels of aggregation, however, are still too low as they do not allow substitutions due to shopping strategies or home production. To accommodate the former, we define the baskets to include common substitutions across different products within the same food group, across different qualities (national and generic brands), package size (quantity discounts), etc. To accommodate the latter, we classify each prepared food to a broader food group that identifies with its main ingredient. As a result five definitions of baskets are available: (i) food name as it appears in Kantar (e.g. gratin dauphinois); (ii) recipe category (gratin); (iii) a 5-digit code (INSEE, 2017) of the main ingredient (e.g. potatoes); (iv) a 4-digit code (INSEE, 2017) of the main ingredient (e.g. potatoes and starchy tubers); and (v) a 3-digit code (INSEE, 2017) of the main ingredient (vegetables; Table A1, Appendix). The baskets (iii)–(v) are better suited for the purpose of this research and are considered in the following section.

The life-cycle behaviour of the corresponding price indices is depicted in Figure 2. The trajectories of all three price indices are similar, particularly those of the schemes (iv) and (v). The scheme (iii), being the one with the lowest aggregation level among the three, perhaps proves the point of allowing the substitution relevant to the purpose of this research. It should be mentioned here that the aggregation level also reflects the homogeneity of the nutritional quality of foods in each food group and, possibly, the quality and quantity trade-off. Regardless, all three indices fall steadily until the early 40’s, in the era of early family life stage, early careers and lower wages, suggesting prudently lower prices. In the middle age, as wage/wealth builds up, as does the demand for time in more advanced careers and possibly raising family, the price rises until the peak in the late 50’s and drops afterwards, once the reduced income and time demands for raising a family take hold. In fact, the households in their 70’s pay prices 1.11–1.82 per cent lower than those at the peak at the age of 50’s. Aguiar and Hurst (2007) show near-identical life-cycle trail of price.

Fig. 2.

Price indices through life cycles.

Open in new tabDownload slide

Notes: The data source is Kantar WorldPanel. The graphs are created from by random effects regressions of the price index (in IHS form) on dummy variables for the panelists’ age groups, as well as the gender, time cohort and region dummies. The vertical axis measures the log deviation from the reference group of 25–29-year-olds.

In the light that the results are remarkably similar in all scenarios, for the further analysis we adopt aggregation scheme (iv). The parameter estimates of the regressions plotted in Figure 2 are reported in Table A4.

5.1. Returns to sopping and home production

The empirical evidence of returns to shopping and home production is depicted in Figure 3. In Panel a, we plotted the returns to shopping time, in deciles. The graph clearly shows monotonically decreasing price as the shopping intensity increases. Specifically, households shopping most intensively (the last decile) pay, on average, 1.54 per cent lower prices (Table A3), demonstrating a statistically significant cost savings to be had through sheer increase of shopping time.

Fig. 3.

Returns to shopping and home production by deciles.

Open in new tabDownload slide

Notes: The data source is Kantar WorldPanel. The graphs are created from regressing price (in IHS form) on dummy variables representing the deciles of the shopping time measure in (6) (Panel a) and fraction of ingredients (Panel b). The vertical axis measures the log deviation from the reference group of 25–29-year-olds.

The trajectory of the price decline as the intensity of home production increases (Panel b) is similar to the price decline in Panel a, albeit on a slightly larger scale. In other words, households that purchase ingredient foods most intensively (decile 10) pay 3.39 per cent lower prices compared to the first decile. In summary, unmistakable decline in prices due to increased shopping and home production time is apparent. The parameter estimates of the regressions depicted in Figure 3 appear in Table A3.

6. The price estimation

6.1. Endogeneity issues

For the parameters in (10) to be consistently estimated, the independent variable is required to be uncorrelated with the error term. It is, however, plausible that the shopping (⁠|$s$|⁠) and home production (⁠|$h$|⁠) times could be systematically affected by environmental-, household- and individual-level preferences not accounted for in (10). Indeed, the fixed and variable components of the shopping time estimate in (6) could be related to the supply and demand-side factors, respectively. Insofar as the former reflects the availability of the retail outlets in the market areas, the failure to account for other characteristics of the area could give rise to omitted variable issues. For the same token, we cannot take into account factors that could affect the variable time of shopping, such as shopping efficiency expressed by the prior knowledge of the store layout, or clever bundling to avoid backtracking, etc. Additionally, our data do not provide information concerning the actual distance between residences and shopping venues, opening it up to measurement errors. Similar cases could be made for the estimate of the time spent in home production. For example, unaccounted factors, such as the availability and distribution of venues that are substitutes to home production—restaurants, etc., in the proximity of households would clearly bear upon the decision to prepare meals at home or eat out (see, for example, Currie et al., 2010, among others). The latter decision could also be conditioned whether the household has the requisite skill set to prepare meals.

To mitigate the potential issues that are fixed over time, we include market and household fixed effects in the estimations. To capture the abovementioned time-variant exogenous change in time allocations to home production and shopping, we turn to instrumental variable estimations. For identification, we use the number of children and the total time availability in a household.

While the number of the children could potentially affect the time and money allocations through a household-size effect, it also has the characteristic constraint on the household availability of time that a sheer increase of household size lacks. It is in this latter component that we would like to tap in to net out the variation in time exogenous to our price model. We achieve this by accounting for the household size effect through the components in |${D_{jt}}$| in (10). For a second instrument we use a measure of households’ total time available to be allocated to shopping and home production. We construct this variable as an indicator of the number of household heads that are not employed (part- or full-time). This instrument has the appealing quality that it is unlikely to be associated with the error term in (10) as there is no reason to believe that the unemployed and retired households pay systematically different prices, unless through strategic shopping and/or home production. In our sample, 8.63 per cent of households have one and 17.81 per cent have two household heads unemployed or retired. The estimation results appear in Table 1.

Considering the previous evidence of the importance of the demographic controls, we estimated two price function specifications—with and without the demographic controls. The estimates appear in columns I and II in Table 1, respectively. The first-stage regression estimates are generally of the predicted sign and are significant at 1 per cent level or better, indicating relevance of the instruments. The significant F-statistics in the first stage testify to the strength of the excluded instruments.

Combined, the Kleibergen-Paap rk Wald F-statistic of the weak identification of the price function in the second stage rejects the null hypothesis that the model is weakly identified in both specifications. The IV parameter estimates for both h and s are consistent with the hypothesised sign. The elasticity of price with respect to time spent in home production −1.48 and −0.80 in the extended (with demographic controls) and the simple (without demographic controls) specifications, respectively. The elasticity of price with respect to time in shopping is −1.30 and −0.83, respectively. Although the differences of the elasticity estimates between the two specifications appear to be large and statistically significant, vindicating the findings in the previous literature, they are also remarkably consistent in the hypothesised sign, indicating an attenuation rather than direction-reverting effect.

The parameter estimate for the bundling control in the |$N$| vector is of the hypothesised sign, indicating a significant price increase associated by the increase in the products purchase per trip. A similar prediction for the composite quantity index was not substantiated as the effect is small and insignificant. Surprisingly, the effect of bulking is positive and significant when controlling for demographic variables, with the effect dissipating without the demographic controls.

Despite the differences in the estimates under the two specifications, all elasticities are not statistically different from 1. In other words, the estimates indicate that 1 per cent increase in time spent in home production or shopping will induce at least as much change in the general price paid, in the opposite direction, as hypothesised previously.

7. Life-cycle consumption model

We use the estimates from (14), combined with the parameter estimates |${\alpha _Q}$|⁠, |${\alpha _h}$| and |${\alpha _s}$| available from the estimation in (10), and provide the necessary parameters to impute the consumption level |$c$| from (11). However, as without making extra assumptions, the parameters |${\varphi _h}$|⁠, |${\varphi _Q}$| and |$\gamma $| cannot be identified, we restrict |$\gamma $| and |${\varphi _Q}$| to be equal to unity, or |$\hat \varphi = {{{\varphi _h}} \over {{\varphi _Q}}} = {e^{\rho {{\hat \beta }_0} - ln\left( {{\alpha _Q} + 1} \right)}}$|⁠. Imposing |$\gamma $| to be equal to unity also ensures that the technology has constant returns to scale, as mentioned above.

8. Life-cycle consumption results

This specification normalizes the produced consumption to reflect each household’s propensity to generate consumption from each euro expended. It should be noted that the proportional representation also accounts for the varying household size, allowing comparisons across households.

8.1. Home production

To trace the path of the life-cycle consumption, we fit the life cycle model with the implied consumption–expenditure ratio in (15) as the dependent variable. The parameter estimates appear in Table A6. In Figure 4, we plot the life-cycle path of the implied consumption-to-expenditures ratio.

Fig. 4.

The implied consumption-to-expenditure ratio over life cycle: Log deviation from 25–29-year olds.

Open in new tabDownload slide

Notes: Data source, Kantar WorldPanel. The set of the demographic variables include the gender and couple status of the panelist, income and car ownership. All regressions include year, cohort and market dummies. All continuous variables were IHS transformed.

As depicted in Figure 4, compared to the reference age group of 25–29-year olds, the life-cycle consumption–expenditure ratio declines until late 30’s, reaching its minimum at −2.16 per cent below the reference level. It picks up thereafter and steadily increases throughout life cycles to reach 14.5 per cent by the age of 70–75, relative to the age group of 25–29, or 16.64 per cent, relative to the age group of 35–39.

It would appear that the presence of home production equipment/skills would be of key importance in production technology, in addition to household demographics. Unfortunately, we cannot recover this information from our data consistently for the entire period of coverage. The inclusion of the household fixed effects in the estimation will mitigate this issue, as it seems likely that the effects of household technology/skills would be relatively fixed over time.

The consumption-to-expenditure ratio without taking into account the demographic information follows a similar trend albeit slightly lower level at each age. It declines until the late 30’s, plateaus at −3.24 per cent and increases steadily thereafter, reaching 12.92 per cent increase by the age of 70–75—an increase of 16.16 per cent from the lowest point in the late 30’s.

While the 16–17 per cent increase in the consumption off the food expenditures appears impressive, it remains to be established whether this increase can offset the drop in expenditures. In their article, Moreau and Stancanelli (2013) maintain that upon the retirement of the male household heads the household income drops significantly, by approximately 2,300 euros. Boissinot (2007) demonstrates that the non-durable expenditures have a significant hump late 30’s to early 40’s, decreasing monotonically thereafter by as much as 60 per cent by the age of 80 years, equivalent to a drop of approximately 3,000 euros since the age of 60 years. While these estimates are for non-durables in general and may not be entirely characteristic to the food expenditures, it is still safe to maintain that the increase in the produced consumption may mitigate if not offset the declining food expenditures.

Both with and without demographic information specifications of our estimation model (14) yielded the elasticity of substitution between time and money—|$\sigma $| of 3.71 and 1.85, respectively. These estimates, particularly the estimate of the specification without demographics, are close to the estimates reported in the literature, ranging from 1.2 to 2.3 (Aguiar and Hurst, 2007). The larger estimate for our specification with demographics information is perhaps due to imprecise and not common set of variables used in estimations.

8.2. Life-cycle cost of time

where |${\hat \alpha _Q}$|⁠, |$ {\hat \alpha _h}$|⁠, |$\hat \varphi $| and |$\hat \rho $| are available from the estimates of (14) and (10).

The life-cycle evolution of the implied cost of time is reported in Table A6, Columns III and IV, in Appendix. We also plot it in Figure 5. In the model specification with demographic information the cost of time declines until late 50’s, plateaus for the next decade or so, and declines sharply afterwards. In the specification without demographics, while the trail picks up slightly after the reference age group, the increase is not statistically significant (Column IV) and eventually declines after late 30’s and generally follows the trajectory of the cost of time from the specification with demographics.

Fig. 5.

The implied cost of time over life cycle: Log deviation from 25–29-year olds.

Open in new tabDownload slide

Notes: Data source, Kantar WorldPanel. The set of the demographic variables include the gender and couple status of the panelist, income and car ownership. All regressions include year, cohort and market dummies. All continuous variables were IHS transformed. The vertical axis measures the log deviation from the reference group of 25–29-year-olds.

Both the trajectories in Figure 5 are consistent with expectations. In the years of family development and early career stages before mid-40’s, time is of limited availability and households are generally poorer than later into carriers. Towards the more advanced years, but before retirement, more time is freed up from the family development and households are considerably richer, with the dynamics indicating a steady but lower price of time than previously. The cost of time drops sharply after the retirement—mid to late 60’s in France, with more time resources now that work-related time is freed too and less financial resources. Aguiar and Hurst (2007) report a similar decline using US scanner data, with much steeper—33 per cent decline at the age of 65–74. Our estimates indicate that the cost of time for persons in the age group of 70–75 is approximately 10.43 per cent less than that of the peak group when controlling for the demographic profile. It is even lower without controlling for the demographic variables.

It would hardly be expected for the opportunity cost of time derived in our analysis and the wage rate to behave similarly. The opportunity cost of time in (15) allows a much more dynamic estimation of the cost of time than the wage rate. Perhaps the most influential of the reasons giving rise to this discrepancy is the consumers’ inability to change labour participation at the margin. Moreover, consumers who are not a part of labour force permanently or temporarily would appear to have zero cost of time. The repercussion of easing into retirement throughout the life cycles, rather than dropping at the time of retirement, as the drop in income would indicate, finds empirical support in the literature (Aguiar and Hurst, 2007).

9. Concluding remarks

Over the next few decades, the share of the elderly population in France and worldwide will increase steadily. There is concern that upon retirement, ageing people cannot maintain the pre-retirement level of consumption, giving rise to nutrition and health deprivation, even food insecurity. Compounded by the increased healthcare costs of the elderly, this can quickly become a public health threat. In this research, we examine whether the observed decrease in food expenditures translates into lower food consumption in France. We find that households engage in strategic substitution of time and money through life cycles, particularly in more advanced age groups, to maintain consumption.

We demonstrate that elderly consumers with more time and less monetary resources spend more time in shopping and home production achieving lower food prices relative to consumers in younger age groups. Specifically, consumers in the age group of 70–75 increase time spent in home production and strategic shopping by 11 per cent and 71 per cent, respectively, relative to the reference group of 25–29-year-old households, enjoying 4 per cent and 2 per cent lower prices as a result. We estimate that, unlike the previous evidence of declining food expenditures towards more advanced age, the produced consumption steadily increases, reaching approximately 14.5 per cent more consumption off of food expenditures by the age of 70–75.

This evidence is consistent with our findings in the opportunity cost of time, which steadily declines throughout the life cycles, with approximately 10.5 per cent less than that of the peak group. The steady decline in the opportunity cost of time explains the steadily maintained level of life-cycle consumption more coherently than the hump-shaped life-cycle wages and their eventual drop at the retirement would suggest. Additionally, we estimate the elasticity of substitution between market goods and time to be 3.71 and 1.85, with and without demographic information, respectively. A little on the higher side when compared to other estimated in the literature, nonetheless this is an important finding in the light of recent coronavirus disease pandemic and rise of home production alternatives to the market products.

A cautionary note is in place however—our results pertain to home consumption only. Unfortunately, we do not have access to the data, most notably, to address this issue for the entire food consumption. Nevertheless, as is well documented previously (INSEE), the population steadily reduces food away from home consumption with ageing, which renders our findings appropriate and valid for informing policies aimed at mitigating the income shock at retirement and other nutrition and food policies aimed at improving consumption for the elderly population.

Future research utilising more complete food acquisition data, notably information on food away from home, would enable a more holistic approach and render a complete insight into the life-cycle food consumption. In addition, the ever-evolving foods and the food retail system redefine food search and home production practices continuously. The invasive digital food acquisitions—Click and Collect, Drive and home delivery, takeout, etc., are becoming more and more common for food acquisition. This not only changes the time and effort spent in search and home production but also puts increasingly more demand on information concerning nutritional, functional, cultural, geographical and other attributes of foods to make the leap from observable food purchase to unobservable (at least on the nationally representative and statistically significant level) food consumption viable and more rigorous. The availability of such information would open up possibilities to extend this line of research to investigating the nutritional quality of the life-cycle consumption, indelibly related to this line of research in food quantity life-cycle consumption.

Funding

The ALIMASSENS Collaborative Project is funded by the French National Research Agency (grant no. ANR-14-CE20-0003). The AgreenSkills fellowship programme, which has received funding from the European Union’s Seventh Framework Programme under grant agreement no. FP7-609398. Toward offering healthy food products better adapted to elderly people – AlimaSSenS, proposes to design innovative food products for elderly population living at home. AgreenSkills+, coordinated by INRA, the French National Institute for Agricultural Research in collaboration with its consortium partner, Agreenium-IAVFF, and the French Agricultural, Veterinary and Forestry Institute, was an international postdoctoral mobility programme designed to support inventive, promising postdoctoral young researchers with up to 10 years postdoctoral experience who wished to carry out challenging basic or targeted research projects in the fields of agriculture, food technology, nutrition and environmental sciences.

Footnotes

References

As discussed above, to be able to capture the returns to home production or savings in price due to purchasing less value-added or ingredient foods, it is important to classify the prepared foods into broader food groups judiciously to avoid unexpected results. For this reason, the prepared foods with typically more than one ingredient were assigned to broader groups⁶ in the order the foods appear in the food name. For example, a frozen dinner labelled ‘Chicken rice with mushrooms’ would be assigned to the food group ‘Meats’ because ‘chicken’ appears before ‘rice’ or ‘mushroom’. In case the food item name is not expressive of the ingredients, for example, ‘Paella’, a recipe was sought from the Individual and National Study on Food Consumption—INCA 2 (INCA 2, 2017, French Agency for Food, Environmental and Occupational Health and Safety) or the Internet, and the food was assigned to the groups of the main ingredient by weight in the prepared dish. For recipes found from the Internet, to minimize the error due to the source of recipes, we confined the source to two-recipe websites only⁷.