Population Aging, Consumption Expenditure Budget Allocation and Sectoral Growth*

Rui MaoAssistant Professor

School of Management, Zhejiang University&

Jianwei XuAssistant Professor

School of Economics and Business Administration, Beijing Normal University

Abstract This paper assesses the heterogeneous effects of population aging on domestic consumption at both age- and product-levels. With China’s household survey data, we break down each expenditure component of a household to its constituent members, and find consistent and robust age profiles of personal consumption-expenditure budget allocations. Young people spend larger proportions on food, and education, culture and recreation services. Conversely, the middle-aged spend significantly more on clothing, and transportation and communication. And the elderly consume substantially more food, and health care and medical services. After distinguishing age effects from period and cohort effects, and considering other contributory factors such as social-economic characteristics, we find demographic structures as the key driver of distinct budget allocations. We incorporate the estimated results with the population forecast data to predict the evolution of consumption patterns, and find the demographic change as a key source to structural change in the future on the demand side. Industries of food, household facilities, articles and services, health care and medical services, and residence may benefit from population aging, because their shares in the total consumption tend to rise. In contrast, industries of clothing, transportation and communication, education, culture and recreation services may lose.

Keywords: Personal budget allocation, population aging, structural change, age effects, domestic consumption.JEL Classification Numbers: D12, D91, E21, J11, L16, O11, R22.

*We thank Rafael Gomez, Binkai Chen, Xuejin Zuo, Chuliang Luo, Junsen Zhang, Xuheng Zang, John Piggott, Hong Mi, and participants at the International Symposium on Demographic Dividend and Social-Economic Development, CEPAR-Zhejiang University Workshop in Population Ageing, and Workshop of Shanghai Academy of Social Sciences for constructive comments.

1 Introduction

Population aging can have substantial impacts on domestic consumption. While a large amount of empirical studies tried to estimate the relationship between age and consumption, they typically stay at highly aggregated levels when characterizing both age and consumption categories. First, these studies implicitly assume that each consumption component responds homogeneously to population aging to the same degree. However, Pasinetti (1981) pointed out that “the proportion of income spent by each consumer on any specific commodity may be very different from one commodity to another”. Second, most studies measure the degree of population aging by the share of people in extremely broad age groups. A typical index frequently used is the old dependency ratio, which includes everyone above 64 into a group. As a result, these studies tend to conceal more detailed effects of population aging on consumption. In this paper, we use a large survey data of Chinese households to specifically evaluate disaggregated effects of population aging on domestic consumption at both product- and age-levels.

In order to examine how demographic impacts differ across products, we give a specific account of personal consumption budget allocations over expenditure components. Most existing literature usually only considers the total consumption expenditure, which alone cannot capture different situations that each product faces when it confronts an aging population. Although some studies examined the relationship between the consumption expenditure of a specific product category and individual characteristics such as income, few of them took the overall pattern of budget allocations into account. To the best of our knowledge, Foot and Gomez (2006) is the only paper that specifically studied the heterogeneous impacts of population aging across products. Using household expenditure data of the UK, they found that the country’s demographic change tended to benefit industries such as medical services and impair industries such as education. Consequently, population aging can not only influence the total domestic consumption, but can also change the relative sectoral growth. We try to improve their study by examining budget allocations at the more detailed individual level instead of the household level.

The individual-level data is also helpful to more accurately characterize the age profiles of consumption expenditures. In order to do so, we break down each household’s spending to its constituent members according to their age. This also differentiates our paper from most other literatures on demography and consumption, which rely on extremely broad age groups. (Horioka and Wan, 2006) The use of age groups assumes that consumption behaviors are the same within each group. Therefore, it can lead to ambiguous or conflicting results. Our study also differs from the large literature that attempts to link household expenditures to the age of household head, which can “flatten” the age profiles because this method obscures the distinct age structure within households according to Deaton and Paxson (2000).

Our paper directly splits age groups into each age, and thus allows for complete age profiles of consumption expenditures. We break down household expenditures to obtain age profiles by introducing age dummies in this paper as Mankiw and Weil (1989). They in particular assumed that people had a fixed demand for housing at each age, so the household’s total demand was simply the sum of all its members. And they obtained best linear predictions of these fixed demand by regressing the value of a family’s house on age dummies of its members. Poterba (2001) resorted to similar methods to characterize age profiles of asset holdings. However, to our best knowledge, this method has not yet been applied to studies on consumption expenditures.

Using the urban household survey data in 18 Chinese provinces from 2002 to 2009, our paper provides direct answers to the two biases that the previous literature has overlooked. Firstly, people within each age group tend to exhibit distinct consumption behaviors, verifying that the old dependency ratio cannot fully characterize the aging population. In particular, total consumption expenditures decrease until one is 15 years old. They start to increase up to the age of 45, and then begin to fall again. We also find starkly different consumption budget allocations for people at each age. In general, young people have large consumption shares of food and education, culture and recreation services. The middle-aged spend substantially more on clothing, and transportation and communication. And the old have significant consumption shares of food, and health care and medical services. In addition, adults also have greater shares of household facilities, articles and services, and residence, than the rest of the people. These age profiles of personal budget allocation were consistent and stable over time from 2002 to 2009.

We then examine whether the age profiles were mainly driven by age or other contributory factors that change along with age. We particularly examine social and economic characteristics, such as income, wealth and education, which differ for people at each age and can influence their consumption behaviors. We find that age was the main determinant of how spending components evolve over one’s lifetime in most product markets. However, for some certain products, social-economic characteristics could systematically alter the evolving patterns.

We also give special attentions to distinguishing age effects from period and cohort effects, which respectively reflect the influence of macro environment where people were observed and the influence of people’s past experiences. In order to avoid the problem of multi-collinearity, we assess the relative importance of period and cohort effects as Heathcote et al. (2004) before accounting for them. Our result confirms that the age profiles were mainly driven by age effects. This lends further support to the importance of demographic structures in influencing sectoral expansions and contractions.

Finally, we take the estimated age profiles of personal budget allocation to China’s population forecast data, and predict the changes in different domestic markets along with population aging. We first exclude other effects and assume all social-economic conditions remain unchanged from the period of 2002-2009. We find that up to 2030, the shares of food, household facilities, articles and services, health care and medical services, and residence in total domestic consumption will increase. In contrast, shares of clothing, transportation and communication, education, culture and recreation services, and miscellaneous goods and services will decrease. We then allow income to grow together with population aging, and find that most consumption shares will still change in the same direction, but will be much smoother. One exception is the share of food, which tends to fall in line with the Engel’s Law.

This paper also sheds light on the relationship between demography and structural change. Structural change was often thought of as a result of growth differentials in sectoral productivities (e.g. Ngai and Pissarides, 2007; Acemoglu and Guerrieri, 2008; Mao and Yao, 2012) or the difference of income elasticity across goods (e.g. Echevarria, 1997; Kongsamut et al., 2001). However, few studies noticed that personal consumption budget allocations are different for people at each age. Therefore, as demographic structure of any country changes, it could have potential impacts on structural change when the growth of each sector exhibits distinct responses to consumption demand change.

The rest of the paper is organized as follows. In Section 2, we break down the household total

consumption and expenditures on different products to its members, and estimate age profiles of personal budget allocation. We then, in Section 3, provide two robustness checks on the consistency of our results. In Section 4, we take social-economic characteristics into account and disentangle age effects from period and cohort effects. In Section 5, we use population forecast data to estimate the effects of population aging on future consumption budget allocation change.. Finally, Section 6 concludes the paper.

2 Age Profiles of Personal Budget Allocation

2.1. BackgroundOur paper is set against the backdrop of China. The country is a relevant case for studies on

demographic effects on domestic consumption for two reasons. First of all, with the natural population growth rate plunging from 33.33‰ in 1963 to below 5‰ in 2011, China’s demographic structure has been aging rapidly. The second national census in 1964 shows that people below 15 accounted for 40.4% of the total population. The share declined to 22.9% in the fifth census in 2000, and further fell to 16.6% in the sixth census in 2005. In contrast, the share of people above 64 rose from 7% in 2000 to 8.9% in 2005. Thus, the world’s most populous country has joined the club of aging populations, whose members are often advanced economies, typically European nations and Japan. Moreover, China’s demographic transition was noticeably faster than historical observations of other countries, because the great deceleration of its population growth was by and large a result of the family-planning policy. In fact, the death rate in China has almost remained flat around 7‰ since the mid-1960s. But the birth rate dropped dramatically from 33.43‰ in 1970 to 11.9‰ in 2010 after the policy was implemented. As a result, China inevitably foresees a quickly aging population in the future.

Secondly, China also exhibit heavy reliance on domestic consumption for economic growth at the same time. The average contribution of domestic consumption to economic growth was 40.7% from 2005 to 2008. It grew by almost ten percentage points to 50.1% in the period of 2009-2012. With the shrinkage of labor supply that comes along with its aging population and the contraction of the international market ensuing from the global financial crisis, both investment and export face strong headwinds. Consequently, China’s reliance on domestic consumption is likely to continue and even increase.

However, the empirical studies on how demographic change affects China’s consumption usually led to mixed results. For example, using provincial panel data of China, Wang et al. (2004) found that consumption ratio was negatively correlated with youth dependency ratio and positively correlated with old dependency ratio. In contrast, Li et al. (2008) found no effects of old dependency ratio, while Horioka and Wan (2006) even found a positive effect of youth dependency ratio, with provincial panel data from China too. This paper tends to add new insight into the literature by incorporating more detailed disaggregate evidence.

2.1 DataThe National Bureau of Statistics has been carrying out household bookkeeping surveys in

China’s urban regions since 1956. They were suspended during 1966 and 1979 when the Bureau was closed upon the onset of the Cultural Revolution, and then resumed in 1980 with the sample size increasing over the years. Our data is a subsample of this dataset that covers 18 provinces,

autonomous regions, or municipalities from 2002 to 2009.1 These areas were scattered around the country. Their average household income levels differed starkly, which reflects the diverse regional development stages in China. For each urban region included in the sample, usually a city or a county, households were randomly drawn among those who had stayed there for more than half a year, regardless of which hukou, i.e. rural or urban, they held. Each household could be observed for several successive years, but any household that had entered the sample three times must be replaced by a new one. Altogether, our data has taken 155,905 different households over this eight year period, with 294,422 household observations. This means on average, a household entered our sample 1.89 times. Each household needed to fill a personal sheet that contains individual information of each household member, e.g. age, gender, and educational attainment, and a household sheet which contains information of its income and expenditures.

Figure 1 compares the age distribution of our sample (UHS, i.e. Urban Household Survey, represented by the solid line) with the total national urban population (represented by the dashed line). It manifests that the former is skewed to the right of the latter. That is, older people were more likely to enter our sample. A possible reason is that to be observed, a household must remain in that region for more than half a year. Young people are usually more mobile than the old, so our sample tended to miss them. However, this right-skewedness will not necessarily bias empirical results, thanks to the large size of our data. Instead, since there were significantly fewer people above 60 years old in the national distribution, the right-skewedness may even help to correct inaccurate estimations for that population.

Figure 1. Age distributions of the UHS sample and the national data (2002-2009).Note: Authors calculated the age distribution of the total national urban population

with data of Hu et al. (2010).

Households in the sample were asked to keep daily records of total spending and its eight components: (1) food; (2) clothing; (3) household facilities, articles and services; (4) health care and medical services; (5) transportation and communication; (6) education, culture and recreation services; (7) residence; (8) miscellaneous goods and services. For each household, expenditure information was aggregated over a year and then reported to the Bureau in January of the next year. Table 1 below summarizes household consumption expenditures during these years. An

1 They are: Anhui, Beijing, Chongqing, Gansu, Guangdong, Heilongjiang, Henan, Hubei, Jiangsu, Jiangxi, Liaoning, Shandong, Shaanxi, Shanghai, Shanxi, Sichuan, Yunnan, and Zhejiang.

apparent feature of this summary is significant variations among households in expenditure levels. Except for total consumption and food expenditures, standard deviations for other products all exceeded their means. Variations in transportation and communication expenditures were the most noticeable. The standard deviation was more than three times the mean. In contrast, variations in food expenditures were the smallest, because the demand for food does not change much with a person’s social-economic conditions. Food is also the subsistence goods, so all households in our sample had positive food expenditures.

Table 1. Statistical description of household consumption expenditures (in current yuan).2

Items No. Obs. Mean Std. Min. Max.Total consumption 294422 25570 22554 149 617001Food 294422 9422 5986 91 245035Clothing 294422 2661 2789 0 78586Household facilities, articles and services 294422 1546 3209 0 154244Health care and medical services 294422 1898 3999 0 198478Transportation and communication 294422 3167 10214 0 474750Education, culture and recreation services 294422 3450 5331 0 208767Residence 294422 2513 5885 0 280084Miscellaneous goods and services 294422 913 1872 0 119514

Before employing a more rigorous statistical method, we can get a rough idea of age profiles of personal budget allocation from the data. We consider three groups of households with regards to the average age of their members: (1) the average age of family members was 30 or below; (2) the average age was between 31 and 60; (3) the average age was above 60. For each household, we calculate its per capita expenditures on the eight product categories by assuming members split the household’s total expenditures evenly. We can then compare average per capita expenditures among the three household groups. Figure 2 shows that the share of food expenditures increases with the average age of household members. As do the share of household facilities, articles and services and that of residence. The proportion of health care and medical services also increases. Of particular interest is its spike in the last group where the average age of household members was above 60. In contrast, shares of clothing, transportation and communication, and education, culture and recreation services monotonically decline with the average age of household members. A caveat is that Figure 2 only exhibits the budget allocation for “representative agents” of families, which ignores heterogeneities among household members. To precisely examine age profiles of personal budget allocation, we will decompose household expenditures to constituent members according to their age later.

2 China’s average inflation rate was 2.2% during 2002 and 2009. It is thus more reasonable to think of these expenditures in real terms. However, provincial level inflation rates of different products are not available, making it difficult to compare nominal spending across provinces and over time. In addition, since we are more interested in budget allocation, i.e. shares of expenditures on different products, instead of specific monetary levels, the nominal expenditures satisfy the need.

Figure 2. Budget allocations in different household groups.

It is also worth noting that due to data limitations, we only have access to expenditures of urban households. However, rural and urban households can have distinct budget allocations. The China Statistical Yearbooks show that on average, 44% of household expenditures were on food in the rural sector from 2002 to 2009, while the number was only 37% in the urban sector. Rural households also had a larger expenditure share on residence (17% vs. 10%) and a slightly larger share on healthcare (8% vs. 7%) than their urban counterparts. In contrast, urban households significantly allocated more spending to education and entertainment (14% vs. 9%), clothing (10% vs. 6%), and transportation and communication (12% vs. 9%). Since the UHS data only surveyed urban areas, we are unable to capture these rural-urban differences. Nonetheless, we believe that for the whole nation, the budget allocation pattern of urban households still dominates, because 75% of China’s domestic consumption was attributed to the urban sector.3

2.2 MethodologyThe most detailed consumption data is collected at the household level. Therefore, how to

decompose it to individuals at each age becomes a challenge. Mankiw and Weil (1989) proposed a method to infer personal housing demand from the total value of houses in which their families live. In particular, they introduced dummy variables to characterize the age composition of each family. They then regressed the total value of the family’s house on these dummies to get the best linear predictors of each individual’s housing demand. This method can be immediately applied to decomposing household expenditures. However, it is worth noting that although expenditures are broken down from the household level to individuals according to their age, the inferred personal expenditures are not identical to age effects. Age profiles of consumption behaviors can be subject to other contributory factors, such as wealth and cohort effects suggested by Deaton and Paxson (2000). We will distinguish the age effects on personal budget allocation in Section 4. However, in this section, estimates that conflate all these factors still satisfy the need of decomposition.

Let k be the total consumption or a spending component, then any household’s consumption Ek can be written as the sum of all its members’ expenditures:

3 In 2010, per capita consumption was 13,471 yuan in urban China and 4,382 yuan in rural China. Urban residents accounted for 49.7% of China’s total population in that year. So the consumption of urban households was about 75% of China’s total domestic consumption.

11\* MERGEFORMAT ()

where Ejk is the spending of its jth member, and N is total number of people within the family.One potential problem of Equation 1 is that some products, such as housing facilities, can be

shared among household members and exhibit the feature of public goods. Therefore, the total spending of a household may be smaller than the sum of its members’ expenditures. However, due to the one-child policy, a typical urban household in China usually includes three members, with two parents and a child, which can be seen from Table 2. As long as the family structure is stable, the effect of sharing can be roughly taken as constant. Therefore, it is still reasonable to compare the budget allocation of individuals at different ages.

Table 2. Household sizes in different years.Year No. Obs. Mean Std. Min. Max.2002 30299 2.82 0.90 1 92003 33716 2.84 0.86 1 92004 35673 2.82 0.84 1 82005 39011 2.80 0.87 1 82006 39055 2.76 0.85 1 82007 40820 2.73 0.87 1 112008 38944 2.76 0.98 1 112009 37480 2.72 0.93 1 8

Let us assume that the expenditure of the jth member in any family, Ejk, is a function of age:

22\*

MERGEFORMAT ()Here, DUMMYtjk’s are a series of dummy variables. If the jth member is of age t, then DUMMYtjk

equals 1 and other dummies equal 0. αtk is the expected spending by a person of age t. It measures the expenditure of a “representative agent” of that age, with other social-economic statuses all at average levels for their age group. Friedman (1957) pointed out that people of the same age may not spend in the same amount if they have different incomes, assets, social statuses, or preferences. In Equation 2, ejk measures the expenditure gap between the jth member and the representative agent of their age due to these differences. Since age does not predict how large this gap is, ejk is orthogonal to all dummy variables. Replacing Ejk with Equation 2, we can rewrite Equation 1 as:

33\*

MERGEFORMAT ()Regressing Equation 3 then gives the best linear predictors of αtk’s.

Note that the largest t is set at 80 in Equation 2. So α80k is actually the average spending of people aged 80 or above. In other words, we assume the expected expenditure level does not

change after a person passes 80 years old. The age of 80 is chosen for two reasons. Firstly, according to the sixth national census, the life expectancy of urban males was 78.7 and that of urban females was 82.9 in 2010. The average life expectancy of an urban resident was about 80. Secondly, less than 1% of the people in our sample were had passed 80 years of age. Introducing more dummy variables for very old will make these parameters inaccurately estimated.

2.3 Total consumption and savings ratiosWe first look at age profiles of total consumption expenditures. Let Ek be each household’s

total consumption and values of DUMMYtjk’s be set according to its members’ ages in Equation 3. Parameter αtk then characterizes the total expenditure by the representative agent of age t. We pool the data from 2002 to 2009 together and run the OLS regression to estimate these parameters. We then plot the representative agents’ expenditure in the order of their ages and get the following age profiles.

2000

4000

6000

8000

1000

012

000

Tota

l con

sum

ptio

n

0 20 40 60 80Age

Figure 3. Age profiles of total consumption expenditures (in current yuan).

Figure 3 shows that total consumption expenditures and age evolve along an S-shaped trajectory. In particular, infants have high annual consumption levels about 8,000. However, when they grow older, their consumption levels dramatically drop. A typical 14-year old consumes the least, only slightly above 2,000. This age is usually when the nine years of compulsory education are over. If a person decides to continue education, they must pay more tuition fees for high schools, or vocational schools. This may lead their total consumption to rise. If the individual decides to work instead, their consumption may also increase due to more living expenditures, because schools usually provide subsidies for certain products such as food, transportation, and sometimes even housing. This explains why total consumption expenditures rise dramatically after the age of 14. The expenditure level peaks at about 13,000 when a person reaches the age of 44, which is the period in which they have both the money and the need to consume. Total consumption expenditures start to fall again afterwards, and eventually return below 8,000.

The traditional neo-classical model typically predicts that the young and the old have low consumption levels. They believe that when the total population has a large share of either group, the national domestic consumption tends to fall. Figure 3 favors this argument, but does not follow it exactly. In fact, we do see that in general, the young and the old spend less than the middle-

aged, but their spending does not remain low at all times. Particularly, infants have significantly more spending than the rest of the young, and the young-old between 60 and 70 spend significantly more than the rest of the old. Therefore, if an economy has lots of infants or young-olds, its domestic consumption will inevitably fall, but may not immediately show a decline. Due to this heterogeneity among the young and the old, to draw conclusions on domestic consumption simply from youth and old dependency ratios could be biased.

It is also worth noting that the S-shaped trajectory in Figure 3 does not imply that the LCPIH fails. Although the LCPIH expects a person to smooth her consumption over time, it in fact does not require her to remain at a constant consumption level. Friedman (1957) noted that preferences, habits, and social statuses may change a person’s spending in different ages by altering their utility function. In particular, in order to meet their social statuses and life styles, the middle-aged may consume more than the young and the old. As long as discounted marginal utilities are equalized over her life, the LCPIH still holds.

It is not difficult to derive age profiles of savings ratio from the previous results. The total consumption by the representative agent of age t is αtk. The income of each age, denoted by It , could be calculated directly from our sample. The savings rate is thus (It−αtk)/It. The results for people aged more than 23 are depicted in Figure 4. Before the age of 23, savings ratios were negative. This means that these agents borrow money to finance their consumption expenditures, which corresponds with the LCPIH. After the threshold of 23, the savings ratio rises dramatically until the age of 35, and then it moves onto a falling trend. In other words, a typical Chinese urban resident saves the greatest share of their income in their mid-30s before retirement. Of course, this is not when an individual’s earnings peak. The income of a 45-year-old was actually on average about 1,000 yuan more than that of a 35-year-old. Nevertheless, Figure 3 demonstrates that a 35 year old consumption level increases at an even faster pace up to their mid-40s. In other words, a 45-year-old typically consumes 1,500 yuan more than a typical 35-year old. Therefore, the higher consumption need of people in their 40s renders the decline in savings ratio in the period that income attains its peak.

1020

3040

50S

avin

gs ra

tio, %

20 40 60 80Age

Figure 4. Age profiles of savings ratio.

However, contrary to the prediction of the LCPIH, the savings ratio continues to rise after a person retires, until the age of 80 is reached. At the age of 79, the savings ratio peaks at 55%. This

“rising tail” is a mirror image of falling consumption levels. Figure 3 shows that consumption dropped dramatically after retirement. In contrast, our sample shows that the average income only declined gradually. As a result, the savings ratio for the elderly increases. This phenomenon is also confirmed by Chamon and Prasad (2010) (Figure 5). The literature provides some explanations for it. For example, the bequest motive was emphasized by Lydall (1955), Mirer (1979), and Menchik and David (1983). They believed that the old save for their offspring. Conversely, Hurd (1990) believed that they save for themselves. He argued that since the old do not work, they are more risk averse than the young. In other words, it is not appropriate to assume that they have the same utility functions and compare their savings ratios directly. Similarly, Nardi et al. (2010) noted another place that uncertainties could come from, i.e. unpredictable physical conditions. They argued that the old make precautionary savings for medical expenditures. Our finding is open to all of these explanations. Although, we must also note that the savings ratio suddenly plummeted once a person reaches 80, the “last day of life” imposed for the purposes of this study. Of course, in reality savings ratio never changes so abruptly. Such a sharp change is present in Figure 4 only because we considered all people aged 80 or above as a single group. However, this big turnaround does imply that savings ratios of the oldest old are in fact low in urban China, consistent with what the LCPIH predicts. Because previous studies did not break savings ratios down to individuals and simply associated them with the age of the household’s head, they might have overlooked the change of picture in the very end. Given that the savings ratio eventually drops, we doubt that the Chinese old are purely saving altruistically. Otherwise, their savings ratio shall remain high at all times. The sudden decline evident in Figure 4 suggests that other factors, such as precautionary motives, may better explain why the elderly save in China. In fact, if people completely save for precautionary motives, the savings ratio will drop to zero when the age of 80, i.e. the end of life as we supposed, is reached.

2.4 Personal budget allocationWe can directly apply the same specification of Equation 3 to obtaining estimates of each

spending component for the representative agent at each age and consequently, examine how their budget allocations differ. In particular, let Ek be any household’s expenditures on the kth product out of the eight categories that the UHS classifies. Then, αtk measures the expenditure by the t-year old representative agent on product k. In order to get the pattern of their budget allocations, we divide αtk by an agent’s total consumption Σk (αtk), which represents the share of their spending on product k. Figure 5 shows how this pattern changes with age.

Figure 5. Age profiles of personal budget allocation.

The first thing to notice from Figure 5 is that people have starkly different budget allocations at each age. The share of food expenditures in total spending follows a U-shaped path. Both the young (especially infants) and the old allocate a big share of consumption to food. The share of clothing changes almost in an opposite manner. People in the middle age groups have larger shares and those on the two ends have comparatively smaller shares. Nevertheless, even compared with children, the old still allocate a significantly smaller part of consumption to clothing. This general pattern is very close to what the share of transportation and communication exhibits. To be specific, the share of transportation and communication remains small before one is 20. Thereafter, it grows quickly, and remains high throughout middle-age. But for the old, this share apparently shrinks. It is worth noting that both clothing and transportation and communication are related with a person’s social activities. Consequently, the old, who are less socially active in China, have smaller shares in both categories.

Since both kids and the old tend to suffer from weak physical conditions, they substantially consume more health and medical services than other people. In particular, infants have an especially large share of medical expenditure, exceeding 20%. The share then falls as they grow up to the age of five. Afterwards, it remains small throughout their childhood and adolescence, and starts to grow when they step into their 20s. In adulthood, this share remains stable until they reach the age of 50. It then expands quickly as they grow older, and finally exceeds 20% again in their late 70s.

While children and adolescents have tiny consumption shares of health and medical services, they have huge shares of education, culture and recreation services. The 16-year-olds have the largest share which exceeds 60%. So undoubtedly, education, culture and recreation services are the most important product that the young consume. Education expenditure becomes extremely large in two periods. The first period is when kids go to kindergartens around the age of four to six. The second period is when they go to high schools, vocational schools, and universities between 15 and 22. Both periods are beyond the coverage of China’s nine-year compulsory education. It is noteworthy that in China, schools may provide free or subsidized products such as food and medical services. This partly explains a falling consumption share of food whilst the share of education, culture and recreation services peaks during those two periods, in addition to

the tiny share of health care and medical services. It also sheds light on the low total spending levels throughout childhood and adolescence evident in Figure 3. In order to examine how pure educational expenditures evolved, we are going to further break down this product category in the next section of robustness checks.

The consumption share of household facilities, articles and services is related to the share of residence. First, infants have larger shares in both items than the rest of the young. This is sensible because as newborns, they increase the number of family members. They thus make additional claims on household facilities and residence. Both shares remain low for the most of the time before one is 20 years old. However, there is a rise in the share of residence for children in primary schools. This could be because they start to require more space for living and studying during that period. Afterwards, both shares begin to rise in one’s early 20s and then remain stable. This implies that for the increase in housing-related expenditures, the most important drivers are those who step into adulthood and form new families. In contrast, these expenditures do not tend to change much along with population aging. Mankiw and Weil (1989) discovered that housing demand rose during one’s 20s and 30s, and then started to fall gradually. However, because we are looking at consumption shares instead of monetary levels, we find that in relative sense, the housing demand stabilizes earlier, at about the age of 25, and then remains roughly unchanged. The rising housing-related expenditures in one’s 20s that we found also echo with the argument of Wei et al. (2012) that rising housing prices in urban China was partly due to marriage motives.

Finally, the consumption share of miscellaneous goods and services remains almost constant over one’s life-time.

To sum up, people have heterogeneous budget allocations across products at each age. Generally speaking, the young have larger consumption shares of food, and education, culture and recreation services. The middle-aged have greater consumption shares of clothing, and transportation and communication. And the old consume substantially more food, and health care and medical services. In addition, all adults have significant consumption shares of household facilities, articles and services, and residence, compared with people of other ages. These results imply that demographic shifts not only affect the total domestic consumption, but also change its allocation over product categories.

3 Robustness Checks

3.1 Consistency of budget allocations over timeIn Section 2, parameters αtk’s are pooled OLS estimates. However, the rapid economic growth

in China implies that a t-year-old in 2002 could be entirely different from a t-year-old in 2009. Figure 6 demonstrations that for anyone above 20, their average income level in 2009 was significantly higher than that in 2002. The peak of the income-age curve in 2009, corresponding to 26,459 yuan for a typical 45-year-old, is 2.3 times the peak of the curve in 2002, corresponding to 11,748 yuan earned by a typical 41-year-old. In addition to income growth, people of the same age may also differ in other aspects such as asset holdings, education levels, and preferences during the period of 2002-2009. While specific expenditure levels might be changed by these factors, the particular question that we are predominately concerned with is: Are the age profiles of personal budget allocation that we found in Figure 5 consistent over time? This issue is especially relevant to us because without a stable pattern, we will not be able to infer the effects of China’s changing

demography on the allocation of its domestic consumption.

050

0010

000

1500

020

000

2500

0A

vera

ge in

com

e

0 20 40 60 80Age

2002 2009

Figure 6. Age profiles of average income levels (in current yuan).

Therefore, we use cross-sectional data from 2002 and from 2009 to re-estimate Equation 3. Figure 7 compares the evolution of the consumption share of each product in these two years. For all the eight products, the solid line, which represents the consumption share in 2002, moved closely with the dashed line, which represents the share in 2009. To be specific, the share of food and the share of health care and medical services were U-shaped in both years. The share of clothing was inversely U-shaped, declining from the 40s in both 2002 and 2009. The share of education, culture and recreation services had twin-peaks before the 20s, with one in the pre-school period and another in the late-10s, and a steep fall in the early-20s present in both years. These comparisons suggest that although people of the same age had changed in many ways throughout the sample period, the consumption shares that they allocated to different products remained relatively stable. In other words, the age profiles of personal budget allocation were consistent, regardless of changing social and economic conditions.

Figure 7. Consumption shares of different products in 2002 and 2009.Note: Consumption shares in 2002 are represented by solid lines and correspond to

the left axes. Shares in 2009 are represented by dashed lines and correspond to the right axes.

3.2 The consumption share of educationIn the eight product categories we have considered so far, education, culture and recreation

services were taken as a bundle. It is therefore unclear whether the twin-peaks before the 20s and the sharp decline afterwards that we found in Figure 5 were driven by education expenditures. In addition, as we mentioned in Section 4, Chinese schools often provide subsidized food or housing

to students. Therefore, some parts of educational expenditure reported by households may in fact be spent on other products. If the age profiles of education, culture and recreation services that we obtained do not agree with those of pure educational expenses, then our inferences about the effect of demographic shifts on educational consumption tend to be biased.

Figure 8. The consumption share of education for people equal or below 25 (in current yuan).

Figure 8 considers three types of educational consumption. The first is defined in the broadest sense, which includes not only all education-related expenses, but also expenditures on culture and recreation services. This is what we have been considered so far. The second excludes cultural and recreational spending, but includes expenditures not directly associated with educational purposes such as food and boarding fees. The third definition is the narrowest and thus may be the best to represent “pure” educational expenditures. This category only includes tuition fees, textbooks, software and other educational materials, and out-of-school training and tutoring fees. We let Ek be these three types of household educational consumption and estimate their evolutions respectively using Equation 3. However, in order to focus on educational expenses, we also restrict ourselves to the population equal to or below 25 years old. In other words, we only keep 26 dummies, i.e. DUMMY0 to DUMMY25.4

Figure 8 demonstrates that the educational consumption under all these different definitions evolved similarly. In particular, educational expenditures always rise in the pre-school period, when a person is below five. They remain low and flat throughout the nine years of compulsory education, typically from seven to 15. They quickly climb up when one steps into high school, and then fall even more sharply in one’s early-20s. Additionally two peaks are always present in these curves, one at the age of five and another at the age of 19. Therefore, in China, pre-school children and high school students generally have the heaviest financial burdens for education. The on average low spending by primary-school and middle-school students also lend support to the view that the country’s compulsory education system effectively alleviated household’s education expenditures.

4 Identifying Age Effects on Budget Allocation

4 That is, we assume people above 25 generally have negligible educational expenditures. In fact, our sample confirms that the number of people who were in school but above 25 was small. But even if we do not impose this assumption on the upper bound of age, we still find almost identical patterns.

Figure 6 illustrates that people at each age have different average incomes. In fact, they could also differ in other social and economic conditions including assets, preferences and status, which, according to Friedman (1957), are all potential determinants of consumption behaviors. Are the age profiles of personal budget allocation that we found in Figure 5 purely driven by age, or do they simply reflect the effects of other variables that change along with age? In other words, as we mentioned in Section 4, how much of αtk’s can be explained by age effects? To answer these questions, we must examine other factors aside from age.

We particularly consider three groups of factors. The first group includes all social-economic characteristics of individuals, such as income, assets and educational attainment. Household’s total income and its income squared are used to capture the possibility that the relationship between income and consumption is not linear, if the propensity to consume changes with income levels. As to the effects of assets and educational attainment on consumption, we can simply control household total asset values and the average of its members’ years of schooling. A problem arises in that we do not have sufficient data concerning household asset values. As an alternative measure of this effect, we follow Yang and Chen (2009) and control the number of durable goods, such as motorcycles, automobiles, refrigerators, washing machines and TV sets, owned by a household.

The other two factors that could affect consumption behaviors are period and cohort effects. The period effect tends to be important if individuals are subject to great impacts from the macro environment that they live in. The cohort effect tends to be important if people’s current behaviors are related to their past experiences (especially those when they grew up), due to habit formation for example. However, it is theoretically impossible to include age effect, period effect and cohort effect at the same time. To explain why, consider the following linear relationship:

44\* MERGEFORMAT ()

In Equation 4, x(a, t, t − a) is the remaining expenditure of any a-year-old in period t, when all the social-economic effects that we discussed in the paragraph above are excluded. g1(a) is the age effect (a stands for age), i.e. a fixed effect that measures the average consumption of a-year-olds in any period. g2(t) is the period effect (t stands for period), i.e. a fixed effect that measures the average consumption of all agents in period t. Lastly, g3(t − a) is the cohort effect (t − a stands for the year of birth), i.e. the fixed effect that measures the average consumption of the cohort who were born in t − a. Equation 4 thus says we cannot introduce all these three fixed effects in a single specification, because any person’s age, period or cohort can be immediately inferred if we know the other two variables.

Several methods have been raised in the literature to avoid this challenge of multi-collinearity. For example, Holford (1983, 1991) specified additional constraints to restrict some age, period or cohort effects to zero and reduce the degree of freedom, so other fixed effects could be properly estimated. But Glenn (2003, 2005) and Keyes et al. (2010) criticized this method, because they found that the results are very sensitive to the additional constraints imposed and it is usually very difficult to justify these constraints, either theoretically or, empirically. In contrast, we are going to follow the method of Heathcote et al. (2004) which does not require these assumptions.

In particular, rather than introducing these three sets of fixed effects together, we are going to evaluate their relative importance to each other beforehand. Age effects are the main effects with

which we are concerned. Consequently we must decide whether period or cohort effects are more important. In order to do so, we notice that the difference of x(a, t, t − a) between any two people can be broken down into three components. The first component is the difference between any two people of the same age but observed in different years. On average, this within-age difference between period t and t + 1 is:

55\* MERGEFORMAT ()

The second component is the difference between any two people of the same cohort, i.e. born in the same year but observed in different years. On average, this within-cohort difference between period t and t + 1 is:

66\* MERGEFORMAT ()

where c = t − a stands for any specific cohort. The third component is the difference between any two people of different ages but observed in the same year. On average, this within-period difference of year t is:

77\* MERGEFORMAT ()

Note that ∆g2(t) is the common component of Equations 5 and 6. Therefore, if period effects are

more important, and must be highly correlated. Nevertheless even if period effects

do not exist, and could still be highly correlated when and are

correlated. To conclude, Heathcote et al. (2004) suggested two criteria to determine if period

effects are important and cohort effects are negligible: (1) and are highly correlated;

(2) is not constant and is different from zero. We introduce household social-

economic characteristics as mentioned before into Equation 3 to re-estimate αtk’s, which now correspond to x(a, t, t − a)’s. We then calculate within-age, within-cohort and within-period consumption differences according to Equations 5, 6 and 7 respectively. Results in Table 3 imply that both criteria are satisfied for all eight product markets. Therefore, according to the suggestion of Heathcote et al. (2004), we only need to take period effects into account.

Table 3. Within-age, within-cohort and within-period differences and their correlations.Transportation and communication

Year ρ( , )

2003 -5.34 -7.18 1.852004 -21.99 -10.60 10.452005 -35.35 -36.57 -7.18

0.872006 -20.26 13.68 35.752007 37.16 53.06 22.902008 -22.56 -12.20 2.40

2009 9.68 2.26 -11.02Household facilities, articles and services

Year ρ( , )

2003 24.22 14.22 -8.642004 4.48 -1.93 -7.072005 -34.14 -33.37 3.01

0.992006 15.95 -0.23 -17.552007 3.65 -6.83 -7.472008 -26.11 -33.24 -10.762009 23.71 10.50 -11.54

Education, culture and recreation services

Year ρ( , )

2003 27.49 27.46 1.022004 -4.78 1.15 2.502005 40.01 34.62 -6.01

0.992006 -18.83 -8.57 10.592007 20.02 36.49 13.942008 -84.68 -87.37 -6.532009 91.23 89.23 0.75

Clothing

Year ρ( , )

2003 7.65 7.46 -0.222004 -7.35 -7.02 -0.452005 -15.65 -15.21 -0.22

1.002006 -20.81 -18.24 2.552007 -24.52 -17.50 4.922008 46.63 48.02 0.322009 -34.19 -34.66 1.14

Health care and medical services

Year ρ( , )

年份 -11.66 -32.69 -19.702003 71.05 33.28 -38.942004 58.13 37.05 -15.04

0.962005 8.75 -20.54 -30.662006 30.08 -5.03 -31.022007 -10.08 -30.47 -19.732008 22.90 -17.33 -43.78

Residence

Year ρ( , )

年份 -15.94 -14.28 2.172003 35.82 28.87 -7.342004 9.05 17.49 9.81

0.982005 -29.89 -38.32 -11.112006 19.57 20.47 4.182007 28.05 29.54 1.662008 -10.50 -11.01 -2.49

Miscellaneous goods and services

Year ρ( , )

2003 5.30 3.09 0.002004 -23.84 -17.27 4.942005 -3.85 0.22 4.97

0.962006 -11.53 -13.36 0.462007 -14.72 -8.25 6.082008 -23.59 -18.38 2.052009 -0.23 0.92 1.23

Food

Year ρ( , )

2003 65.87 48.68 -17.212004 178.04 170.17 -8.362005 -85.93 -88.44 -1.28

0.992006 112.60 64.17 -50.312007 122.52 105.76 -12.312008 -1.58 -23.32 -22.152009 194.16 169.23 -21.02

Note: ρ is the correlation coefficient.

After accounting for social-economic characteristics and period effects, Equation 3 can be rewritten as:

88\*

MERGEFORMAT ()where SOCECONτ are household social-economic characteristics and γτ’s are period fixed effects of year τ. α'ik’s in Equation 8 thus measure the pure age effects on consumption expenditures. Figure 9 shows that in most product markets, age effects move closely with the age profiles of consumption which we estimated in Equation 3. It implies that although age is not the only determinant of consumption expenditures, it is the key driver of their age profiles. Therefore, the demographic transition in China could have important impacts on the allocation of its domestic

consumption.

Figure 9. Age effects vs. age profiles of expenditures (in current yuan).Note: Age effects are estimated by Equation 8. They are represented by solid lines

and correspond to the right axes. Age profiles of expenditures are estimated by Equation 3. They are represented by dashed lines and correspond to the left axes.

There are two additional facts which are worth noting according to Figure 9. First, the age effects on clothing expenditure do not feature a thin left tail for people below 30, which is evident in the age profiles. It implies that clothing spending could be highly responsive to social-economic status, particularly income and wealth. Therefore, when we take into account that young people tend to have lower income and wealth levels, the actual age effects do not change much before retirement. Second, the age effects of expenditures on household facilities, articles and services and those of residence expenditures are relatively stable throughout one’s lifetime except the period of childhood. This means the jump that we see in the age profiles in the 20s could be driven by their rise in social-economic status.

5 Effects of Population Aging on the Allocation of Domestic Consumption

In the preceding sections, we have discovered and confirmed a stable pattern of age profiles

of personal budget allocation which is predominately driven by age effects. In the meantime, the existing trend of an aging population is certain to continue and increase in scale. According to the prediction by Hu et al. (2010), the peak median population age will reach 60 in 2050, far exceeding recent years’ average of 45. Therefore, we can use our study to predict how future aging trend will affect China’s consumption.

In the first part of this section, we predict the effect of aging on future consumption alone. To be more precise, the question we are concerned with is: what are the effects of an aging population on the allocation of China’s domestic consumption if all the other factors remain unchanged? That is, if we still consider the representative agents of the period 2002-2009, but, society contains a greater proportion of elder agents, how will the share of each product sector change? In the second part of this section, we evaluate the accuracy of the prediction using a method of counterfactual analysis, which shows that demography is one of the key factors in driving consumption changes. To ease the inaccuracy, the income effect is additionally included in our prediction.

5.1 Effects of population aging aloneChina faces the problem of a rapidly aging population over the next few decades, according

to the forecast by Hu et al. (2010). The forecast predicted the population for each age from 0-100 in China’s rural, urban and migrant sectors based on the fifth national census in 2000. In particular, they used China’s official aggregate-level rural and urban population data from 2002 to 2009 to calibrate the fertility model, the mortality model and the migration model. They then predicted the number of people at each age from 2002 to 2050 based on the census data using these models, ensuring that their prediction aggregated up to the official data for the corresponding year from 2002 to 2009. They considered several scenarios. First, they allowed the total fertility rate to take three different values, 1.4, 1.5 and 1.6, throughout the prediction period. Second, they allowed two urbanization outlooks, i.e. to continue at the current pace and to stop permanently. That is, there were six cases altogether. In this section, we only focus on one of them, which we consider closest to reality, i.e. the total fertility rate equals 1.5 and urbanization continues. Figure10 compares the age distribution of China’s urban population in 2011 and 2050 in this case. The apparent rightward shift of its peak manifests China’s daunting future of an aging population.

0.0

05.0

1.0

15.0

2.0

25

0 20 40 60 80 100Age

2011 2050

Figure 10. The age distribution in urban China, 2011 vs. 2050.Note: The data source is Hu et al. (2010).

When we consider the effects of population aging alone, we let αtk be a typical t-year-old’s spending on product k in any year τ, where αtk’s are still estimated according to Equation 3 for the period of 2002-2009. However, the number of t-year-olds in year τ, which we denote as Ntτ, comes from the population forecast data. Therefore, the total expenditure on product k by China’s urban residents in year τ can be written as:

99\* MERGEFORMAT ()

We first analyze the household total consumption expenditures on all products. Equation 9 thus gives the sum of total expenditures by all urban residents. We then divide Ckτ by the number of urban residents in year τ to get that year’s per capita expenditures.

Figure 11 shows how the figure evolves over time. Since the demographic prediction into the distant future can be inaccurate, for example, the total fertility rate may not be constant for such a long time, 2030 is set as the final year for this study. It is apparent that the per capita consumption expenditures will generally follow a U-shaped curve that peaks in 2026, if population aging is the only driver and other factors such as people’s social-economic conditions all remain unchanged from the period 2002-2009. The curve increases first, because the majority who were 25 to 35 years old in 2011 will turn 40 or 50 in 2026. Those people, according to Figure 3, have the largest consumption expenditures compared with agents in other age groups. Nevertheless, this curve eventually declines because more and more people get old and their consumption expenditures fall. As a result, the curve generally has a U-shaped pattern. It is worth noting that Figure 3 does not include other effects that could push the per capita consumption expenditures upward over time, such as increasing incomes and assets. With improving social-economic conditions, expenditure levels are in fact very likely to rise throughout the examination window. By excluding these effects, Figure 3 can focus more sharply on the impacts of demographic shifts.

1555

015

600

1565

015

700

Per

cap

ita c

onsu

mpt

ion

expe

nditu

res

(yua

n)

2013 2015 2017 2019 2021 2023 2025 2027 2029

Figure 11. Per capita consumption expenditures from 2013 to 2030 in urban China.

We then let k indicate each product successively and calculate product-level expenditures in urban China. To answer how the consumption share of each product changes over time, we first divide the sum of expenditures on each product Ckτ by the total expenditures on all products in that

year to get the consumption share, and then calculate the annualized growth rate for each share. As before, we restrict the prediction to the period of 2013-2030. Table 4 shows that the consumption share of food and that of health care and medical services will increase as the population ages. The reason is that as Figure 5 demonstrates, the old generally have larger consumption shares on these two items compared to people of other ages. During 2013-2030, the consumption share of health care and medical services will increase at an average rate of 1.12%, and that of food will increase at an average rate of 0.2%. Similarly, consumption shares of residence and household facilities, articles and services will increase along with an aging populace, but at slightly lower rates of 0.15% and 0.06% respectively. In contrast, products like clothing, transportation and communication, and education, culture and recreation services, are according to Figure 5 mostly consumed by the young. Hence, population aging will bring down these products’ shares of consumption. The consumption share of clothing will fall the most dramatically. Its average growth rate will be -0.59%. The consumption share of transportation and communication will fall slightly slower, with an average growth rate of -0.56%. Even the consumption share of education, culture and recreation services will only grow at an average rate of -0.33%. So Table 4 clearly demonstrates the significant change in the allocation of domestic consumption along with population aging.

Table 4. Annualized growth rates of consumption shares in urban China (%).

Year Food Cloth. Facilities Medical Trans.Educatio

nResidence Misc.

2013 0.10 -0.21 0.35 0.67 -0.21 -0.61 0.24 0.12 2014 0.08 -0.30 0.23 0.73 -0.29 -0.31 0.14 0.03 2015 0.12 -0.40 0.13 0.77 -0.30 -0.32 0.22 -0.13 2016 0.15 -0.44 0.01 0.76 -0.21 -0.25 0.09 -0.27 2017 0.16 -0.45 -0.03 0.92 -0.44 -0.24 0.22 -0.22 2018 0.25 -0.36 -0.08 1.10 -0.75 -0.14 -0.04 -0.25 2019 0.21 -0.51 -0.02 1.14 -0.49 -0.39 0.19 -0.30 2020 0.15 -0.63 0.18 1.22 -0.41 -0.33 0.17 -0.34 2021 0.24 -0.55 0.10 1.17 -0.73 -0.29 0.15 -0.29 2022 0.25 -0.64 0.06 1.10 -0.59 -0.31 0.11 -0.31 2023 0.27 -0.70 0.00 1.18 -0.69 -0.29 0.18 -0.36 2024 0.23 -0.73 0.08 1.22 -0.71 -0.29 0.23 -0.28 2025 0.24 -0.72 0.00 1.27 -0.66 -0.36 0.18 -0.19 2026 0.26 -0.79 0.05 1.43 -0.84 -0.28 0.13 -0.29 2027 0.24 -0.80 0.07 1.30 -0.66 -0.39 0.18 -0.24 2028 0.26 -0.81 0.04 1.42 -0.75 -0.40 0.09 -0.15 2029 0.26 -0.76 -0.08 1.44 -0.72 -0.44 0.13 -0.22 2030 0.19 -0.78 0.06 1.39 -0.59 -0.39 0.14 -0.29 Mean 0.20 -0.59 0.06 1.12 -0.56 -0.33 0.15 -0.22

It should be noted that the prediction made in this section could not be viewed as a forecast of consumption. After all, only the effect of population aging was considered. However, it illustrates

that demographic structures alone can have significant influences on domestic consumption and its allocation across sectors.

5.2 Other effectsThe pattern that Table 4 exhibits is only a result of population aging, with other factors such

as social-economic conditions all unchanged from the period of 2002-2009. But in China, the rapid growth and economic transition imply that these other factors will also be important. In fact, we can use the same method to compute consumption shares from 2002 to 2009 and compare them with the real shares calculated from our sample. The difference then captures the effects of other factors aside from population aging. Table 5 shows our results. Negative values indicate that the estimated consumption shares are smaller than real shares, and positive values indicate that the estimated consumption shares are bigger than real shares.

Table 5. Estimated consumption shares vs. real shares in urban China (%)Year Food Cloth. Facilities Medical Trans. Education Residence Misc.2002 -2.39 -0.53 -5.89 -0.13 22.98 -5.78 -3.87 8.74 2003 -1.14 3.18 -4.27 -6.05 13.75 -1.67 -6.15 6.14 2004 -2.15 6.75 2.98 -3.06 6.51 -4.22 -1.05 5.16 2005 1.45 -0.26 6.30 -4.79 -2.04 -2.72 1.55 0.19 2006 3.92 -3.54 5.15 2.18 -6.44 -2.74 -1.92 -1.23 2007 1.80 -4.80 1.76 3.70 -5.46 -0.55 3.90 -2.30 2008 -1.65 -2.39 -2.43 -0.01 2.12 7.58 -0.05 -4.37 2009 1.41 -4.64 -5.95 4.34 -6.29 8.28 1.83 -8.34

Table 5 implies that the estimated shares were generally close to the real data, which means that demography is the key driver of changes in consumption shares. Nevertheless, factors other than population aging also have systematic impacts on consumption shares. To be specific, in food expenditures, the estimates consumption shares were smaller than the real shares before 2005, but turned out to be generally larger afterwards. This means that the improving social-economic conditions might have reduced the consumption share of food, thus in reality it did not rise as much as the population aging would suggest. In contrast, the estimated consumption shares of transportation and communication were significantly lower than the real shares before 2005. This means social-economic conditions in those years could have restrained urban residents from expenditures related to this category. To summarize, Table 5 means that although only considering population aging as we did in Table 4 might give correct directions of consumption-share changes in the future, it could be biased given that other factors also have systematic impacts.

However, to completely predict evolutions of these factors is difficult, especially when some factors such as household assets, social status and preferences are hard to even measure. Our next step is to consider only one additional factor, i.e. income. In other words, we believe income is a relatively good indicator of the general level of social development. We assume there is a linear relationship between this indicator and expenditures. So Equation 3 becomes:

1010\*

MERGEFORMAT ()where inc is the household income. Then in any year τ, a t-year-old’s expenditures on product k is defined as:

1111\* MERGEFORMAT ()

where inctτ is the income of the t-year-old in year τ. Replacing the αtk’s in Equation 9 by ctkτ’s in Equation 11, the total expenditures by China’s urban residents on product k in year τ become:

1212\* MERGEFORMAT ()

Using Equation 12, we can re-calculate the annualized growth rates of consumption shares and compare them with what we obtained in Table 4. Table 6 below shows the result for the period 2013-2030 on average.

Table 6. Average annualized growth rates of consumption shares in urban China (%).

Food Cloth. Facilities Medical Trans.Educatio

nResidence Misc.

Population aging 0.20 -0.59 0.06 1.12 -0.56 -0.33 0.15 -0.22 + income growth -0.01 -0.57 0.22 1.08 -0.19 -0.21 0.13 0.01

In Table 6, the first row shows the average annualized growth rates, i.e. the last row of Table4 which we have previously obtained. The second row corresponds to our re-calculations based on Equation 12 when income growth is taken into account. Since the real income has been growing at 7% on average in China from 1979 to 2011, we assumed that this trend will continue and that for the entire population, income will grow at an annual rate of 7% from 2009 levels. This is a heroic assumption because the income growth rates could differ across age groups. Unfortunately, no historical income data of sufficient duration is available to calculate the trend in income growth for people at each age. Table 6 shows that when income growth is considered, all consumption shares except for food and miscellaneous goods and services will still change in the same direction. In particular, because Engel’s law says that richer people spend a smaller share on food, the consumption share of food will actually fall slightly along with population aging if people’s income keeps increasing. In contrast, because miscellaneous goods and services include gold and jewelries, cosmetics, and hotel charges which are income-elastic expenditures, their consumption share will actually rise slightly. In addition, it is also worth noting that for most products, changes in consumption shares become smoother when income growth is considered. The only exception is the share of household facilities, articles and services, whose average annual growth rate moves from 0.06% up to 0.22%. Finally, Table 6 highlights that whenever the effect of income growth is taken into account, the consumption share of health care and medical services will always rise dramatically in conjunction with population aging. However, the health care and medical system is largely provided by the government in China, and is already under pressures. The rising

consumption share thus indicates an even greater challenge to the government created by its aging population.

6 ConclusionsIn order to assess the heterogeneous effects of population aging on domestic consumption at

both age- and product-levels, this paper breaks down household consumption expenditures to constituent members using the urban household survey data in 18 Chinese provinces from 2002 to 2009. We find distinct age profiles of each consumption component. Young people tend to have larger consumption shares of food and education, culture and recreation services. The middle-aged spend substantially more on clothing, and transportation and communication. And the old have significant consumption shares of food, and health care and medical services. In addition, adults also have greater shares of household facilities, articles and services, and residence, than the rest of the people. These age profiles of personal budget allocation were consistent and stable over time.

This paper then tries to identify whether the age profiles were mainly driven by age or other contributory factors that change along with age. We differentiate two sorts of contributory factors. Firstly, we distinguish age effects from period and cohort effects. To solve the multi-collinearity problem, we assess the relative importance of period and cohort effects following Heathcote et al. (2004), and then confirm the main contribution of age effects. Secondly, we consider other social and economic characteristics such as income, wealth and education. We find that age effects were still the main determinant of age profiles of budget allocations when these factors are controlled.

Finally, we incorporate the estimated results into population forecast data and predict the evolution of China’s consumption patterns driven by its demographic change. We find that the shares of food, household facilities, articles and services, health care and medical services, and residence in total domestic consumption tend to rise along with population aging, while the shares of clothing, transportation and communication, education, culture and recreation services, and miscellaneous goods and services tend to fall.

ReferencesAcemoglu, D., & Guerrieri, V. (2008). Capital deepening and nonbalanced economic growth.

Journal of Political Economics, 116(3), 467-498.Chamon, M. D., & Prasad, E. S. (2010). Why are saving rates of urban households in China

rising? American Economic Journal: Macroeconomics, 2(1), 93-130.Deaton, A., & Paxson, C. (2000). Growth and saving among individuals and households. Review

of Economics and Statistics, 82(2), 212-225.Echevarria, C. (1997). Changes in sectoral composition associated with economic growth.

International Economic Review, 38(2), 431-452.Foot, D. K., & Gomez, R. (2006). Population aging and sectoral growth: the case of the U.K.,

2006-2016. Oxford Journal of Business and Economics, 5(1), 85-94.Friedman, M. (1957). A Theory of Consumption Function. Princeton, NJ: Princeton University

Press.Glenn, N. D. (2003). Distinguishing age, period, and cohort effects. In Mortimer, J. T., &

Shanahan, M. J. (Eds.), Handbook of the Life Course, 465-476. New York, NY: Kluwer

Academic.Glenn, N. D. (2005). Cohort Analysis (2nd ed.). Thousand Oaks, CA: Sage.Heathcote, J., Storesletten, K., & Violante, G. L. (2004). Two Views of Inequality over the Life-

Cycle. CEPR Working Paper No. 4728.Holford, T. R. (1983). The estimation of age, period and cohort effects for vital rates. Biometrics,

39(2), 311-324.Holford, T. R. (1991). Understanding the effects of age, period, and cohort on incidence and

mortality rates. Annual Reviews in Public Health, 12, 425-457.Horioka , C. Y., & Wan, J. (2006). The Determinants of Household Saving in China: A Dynamic

Panel Analysis of Provincial Data. NBER Working Papers No. 12723.Hu, Y., Cai, F., & Du, Y. (2010). Population changes in the twelfth policy period and the

anticipation of future population trends. In Cai, F. (Ed.), Green Book of Population and Labor, 44-65. Beijing: Social Sciences Academic Press.

Hurd, M. D. (1990). Research on the elderly: economic status, retirement, and consumption and saving. Journal of Economic Literature, 28(2),565-637.

Keyes, K. M., Utz, R. L., Robinson, W., & Li, G. (2010). What is a cohort effect? Comparison of three statistical methods for modeling cohort effects in obesity prevalence in the United States, 1971-2006. Social Science & Medicine, 70(7), 1100-1108.

Kongsamut, P., Rebelo, S., & Xie, D. (2001). Beyond balanced growth. Review of Economic Studies, 68(4), 869-882.

Lydall, H. (1955). The life cycle of income, saving, and asset ownership. Econometrica, 23(2), 131-150.

Mankiw, N. G., & Weil, D. N. (1989). The baby boom, the baby bust, and the housing market. Regional Science and Urban Economics, 19(2), 235-258.

Mao, R., & Yao, Y. (2012). Structural change in a small open economy: an application to South Korea. Pacific Economic Review, 17(1), 29-56.

Menchik, P. L., & David, R. (1983). Income distribution, lifetime savings, and bequests. American Economic Review, 73(4), 672-690.

Mirer, T. (1979). The wealth-age relation among the aged. American Economic Review, 69(3), 435-443.

Nardi, M. D., French, E., & Jones, J. B. (2010). Why do the elderly save? The role of medical expenses. Journal of Political Economy, 118(1), 39-75.

Ngai, L. R., & Pissarides, C. A. (2008). Structural change in a multi-sector model of growth. American Economic Review, 97(1), 429-443.

Pasinetti, L.L., 1981, Structural Change and Economic Growth, Cambridge University Press, Cambridge.

Poterba, J. M. (2001). Demographic structure and asset returns. Review of Economic and Statistics, 83(4), 565-584.

Yang, R., & Chen, B. (2009). Higher education reform, precautionary saving and consumer behavior. Journal of Economic Research (Jingji Yanjiu), 44(8), 113-124.