- Email: [email protected]
Income inequality, consumption, and social-status seeking
Journal of Comparative Economics 39 (2011) 191–204
Contents lists available at ScienceDirect
Journal of Comparative Economics journal homepage: www.elsevier.com/locate/jce
Income inequality, consumption, and social-status seeking Ye Jin, Hongbin Li, Binzhen Wu ⇑ School of Economics and Management, Tsinghua University, Beijing, 100084, PR China
a r t i c l e
i n f o
Article history: Received 18 May 2010 Revised 17 December 2010 Available online 4 January 2011 JEL classification: D12 D91 E21 P46 Keywords: Income inequality Social status Consumption and savings Status seeking Education investment
a b s t r a c t Jin, Ye, Li, Hongbin, and Wu, Binzhen—Income inequality, consumption, and social-status seeking Using the Chinese Urban Household Survey data between 1997 and 2006, we find that income inequality has a negative (positive) effect on household consumption net of education expenditures (savings) even after we control for household income. We argue that people save to improve their social status when social status is associated with pecuniary and non-pecuniary benefits. Rising income inequality can strengthen the incentives of status-seeking savings by increasing the benefit of improving status, and by enlarging the wealth level required for status upgrading. We also find that the negative effect of income inequality on consumption is stronger for poorer and younger people and that income inequality stimulates more education investment, which are consistent with the statusseeking hypothesis. Journal of Comparative Economics xxx (xx) (2011) xxx–xxx. School of Economics and Management, Tsinghua University, Beijing, 100084, PR China. Ó 2010 Association for Comparative Economic Studies Published by Elsevier Inc. All rights reserved.
1. Introduction The Chinese consumption rate has been low and declining in the last two decades. According to the Chinese Urban Household Survey data, the average propensity to consume (net of education expenditures) of urban residents declined from 81.6% in 1997 to 74.7% in 2006.1 Several explanations have been proposed for this phenomenon, including demographic changes (Kraay, 2000; Modigliani and Cao, 2004), high income growth and habit (Horioka and Wan, 2007), precautionary savings (Chamon and Prasad, 2010; Kuijs, 2006; Meng, 2003), change in the return rate of investment (Wen, 2009), and increasing sex ratio (Wei and Zhang, 2009). In this paper, we explore another potential cause of declining consumption: rising inequality. The classical theory on the relationship between income inequality and the consumption rate emphasizes that if the propensity to consume decreases with income, then at the macro level, rising inequality could reduce the consumption rate. This ‘‘macro’’ reason for the inequality–consumption link is mechanically caused by the non-linear association between consumption and income at the micro, or household, level. If only the macro-mechanism is at work, then inequality should not affect consumption behavior at the household level once household income is well controlled for. Empirical results drawn on the Chinese Urban Household Survey data for 1997–2006 show that income inequality within a reference group (i.e., families in the same province and same age group) has a negative effect on the household ⇑ Corresponding author. E-mail address: [email protected] (B. Wu). The average propensity to consume (APC) is defined as living expense net of education expenses divided by disposable income. The change in the APC here is similar to that in Chamon and Prasad (2010). 1
0147-5967/$ - see front matter Ó 2010 Association for Comparative Economic Studies Published by Elsevier Inc. All rights reserved. doi:10.1016/j.jce.2010.12.004
192
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
consumption rate (net of education expenditures) after we control for household income. This implies the existence of mechanisms at work other than the ‘‘macro’’ mechanism. We explore several potential explanations for the observed inequality– consumption link, including rising housing prices, high income risks, high return to education, poor social security, and rising sex ratios. However, none of these can account for the observed inequality–consumption link. We then propose and explore another potential explanation for the observed inequality–consumption link at the household level based on people’s desire to improve their social status. More specifically, people care about their social status, which is associated with pecuniary and non-pecuniary benefits. Social status often depends on a family’s rank in the wealth distribution or indicators such as educational attainment, which is closely associated with wealth when the credit market is imperfect. As a result, in order to ascend in the status hierarchy or keep the social status in the ‘‘Rat Race’’, families try to accumulate wealth by increasing savings. When income inequality increases, the benefit gap between the high-status and low-status groups widens, which in turn strengthens the incentives of status-seeking savings. In contrast to the macromechanism, the status-seeking hypothesis implies that given household income, rising income inequality can still discourage (stimulate) household consumption (savings). Furthermore, rising income inequality also raises the entry wealth level for the high-status group, which means that more savings are needed for one to enter the high-status group. The status-seeking theory also leads to further testable hypotheses. First, income inequality has a greater effect on consumption and savings for poorer and younger people because they have stronger incentives for moving up. Second, rising inequality should have a strong positive effect on families’ investment in education, which is an important indicator of social status. Third, inequality should have a negative effect on conspicuous (or unnecessary) consumption, and should have no impact on basic (or subsistence) consumption. Additional empirical analyses confirm these hypotheses. Finally, our results are robust to different measures of inequality. The rapid increase in income inequality in China suggests that rising inequality is a very important reason for the low household consumption rate (net of education expenditures) in the country. According to World Bank (2005), China’s Gini coefficient has risen from 0.33 to 0.47 in two decades.2 In our sample, the Gini coefficient in urban areas rose from 0.23 in 1997 to 0.29 in 2006, which implies a decline of the average propensity to consume (APC) by 2.08 percentage points, about 30.1% of the decline of the APC during the period.3 To the best of our knowledge, we are the first to find micro-evidence that rising inequality could be a reason for the declining consumption. Existing studies emphasize the link as ‘‘a macro-phenomena’’ and do not consider the micromechanism, that is, the effect of inequality on an individual household’s consumption or saving behavior given one’s income.4 Alesina and Perotti (1996) propose a micro-mechanism of how income inequality could increase social tensions, thus increasing the risk of investment and reducing private saving rate. However, their empirical evidence is based on cross-country data, which are still at the macro-level. We are also among the first to show the importance of status seeking in increasing savings. The recent literature on economic growth demonstrates that the status-seeking saving motive can be beneficial for economic growth (Corneo and Jeanne, 2001; Futagamia and Shibatab, 1998; Gong and Zou, 2001; Pham, 2005), but there is little empirical micro-evidence to confirm the importance of the status-seeking motive. The rest of the paper is organized as follows. The second section introduces our empirical framework. Section 3 describes the data and presents some descriptive statistics. Section 4 presents the empirical results and tests several potential explanations for the observed inequality–consumption link. Section 5 discusses how the status-seeking motive can lead to a direct effect of income inequality on household consumption and provide further tests for the status-seeking hypothesis. Section 6 presents results of robustness checks using other measures of inequality and consumption. Section 7 concludes. 2. Empirical models To identify the effect of income inequality on consumption (saving) behavior, we estimate the following empirical model:
lnðCÞ ¼ a þ b lnðYÞ þ c Gini þ d X þ e;
ð1Þ
where C is household consumption net of education expenditures, Y is household disposable income,5 Gini is the Gini coefficient, and X includes all other covariates. The reason why we focus on consumption net of education expenditures is that education expenditures are usually considered an investment rather than consumption.6 Throughout the paper, consumption means all consumption expenses net of educational expenditures, unless otherwise specified. 2 Moreover, there is a substantial variation across regions and time. For instance, our calculation illustrates that the Gini coefficient among urban residents in Beijing rose from 0.19 in 1997 to 0.25 in 2006, whereas the measurement in Zhejiang province changed from 0.23 to 0.32 in the same period. 3 Our baseline estimation in column 6 of Table 1 indicates that when the Gini coefficient increases by 0.1, the APC declines by 3.46 percentage points. Therefore, a 0.06 increase of the Gini leads to 2.076 (=0.06 3.46) percentage points decrease in the APC, which accounts for 30.09 (=2.706/(81.6 74.7)) percent of the total decline of the APC between 1997 and 2006. 4 For example, Musgrove (1980), Menchik and David (1983), Stoker (1986) and Dynan et al. (2004) suggest that due to the differences in the concavity of utility function, precautionary saving, or bequest motive, the saving rate of high-income families is higher than that of low-income families. Smith (2001) uses cross-country data to confirm that to the extent of credit market imperfection, income inequality has a robust positive effect on aggregate private saving rates. However, there are also some studies that find no effect of income distribution on the aggregate saving rate (Schmidt-Hebbel and Serven, 2000). 5 Disposable income includes wages, capital gains, and transfers, excluding the social security contribution and income tax. 6 We will also estimate the effect of inequality on education expenses, as well as total consumption that includes education expenditures.
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
193
Fig. 1. Average propensity to consume and income inequality.
In Eq. (1), c measures the effect of income inequality on consumption. It can also be seen as the effect of inequality on the average propensity to consume (or consumption rate), since log(consumption rate) is equal to the difference between log(consumption) and log(income). For expositional simplicity, we hereafter use consumption and consumption rate interchangeably in the empirical analysis. b is the income elasticity of consumption. If the link between income inequality and consumption works only through consumers’ heterogeneity in the propensity to consume, then c should be zero, particularly when we allow b to vary with income. Note that if b < 1, the average propensity to consume declines with income, which is consistent with the literature. For the measure of inequality, we focus on the Gini coefficient, and use other inequality measures, including the income ratio of the richest 10% (25%) to the poorest 10% (25%) or the Theil index as robustness tests.7 Moreover, we use the Gini coefficient among peers or the reference group rather than the inequality of the whole society because people usually compare themselves with their peers (Coleman, 1990; Li and Zhu, 2006).8 Specifically, we use inequality within the reference province-age (plus and minus 5 years) group. For example, for families with a 31-year-old head, the reference group includes families with a head between the ages of 26 and 36 in the same province.9 All inequality indices are based on income per equivalent person, considering family size or scale. The results presented below use the equivalence scale, where one of the adults in the household has weight 1, the other adults have weight 0.7, and children’s weight is 0.5. Applying different family equivalent scales delivers similar estimates (see Atkinson et al. (1995) for definitions of family scales). Control variables in X include the head’s age, family equivalent scale, average income of the reference province-age group, provincial fixed effects, age-group (five groups: under 35, 35–44, 45–54, 55–64, and 65 or older) fixed effects,10 and year fixed effects. Finally, we control for the interactions between province dummies and year and the interactions between agegroup dummies and year to allow for a specific linear trend for each province and each age group. Thus, the identification of our model relies on the differences in the non-linear temporal changes in income inequality across age groups and provinces.
3. Data and descriptive statistics The data come from the annual Urban Household Survey (UHS) conducted by the National Bureau of Statistics in China. The UHS covers all provinces in China and uses a probabilistic sampling and stratified multistage method to select households. It is a rotating panel in which one-third of the sample is replaced each year, and the full sample is changed every 3 years. The sampled households are asked to keep detailed records of their incomes and expenditures every day. The survey yields demographic and income information for every member of the family, but unfortunately it has no information on assets. We have access to data gathered from 1997 to 2006 on the nine Chinese provinces of Beijing, Liaoning, Zhejiang, Anhui, Hubei, Guangdong, Sichuan, Shaanxi, and Gansu, which represent different regions and economic conditions. The mean values and the trends of the most important variables are comparable between our sample and the national sample. 7 The Gini coefficient reflects the average dispersion of income. The 90/10 income ratio, 75/25 income ratio, and the Theil index are more sensitive to the difference in the tails. 8 Bakshi and Chen (1996) introduce the reference group’s average wealth in their theoretical model, and Clark and Oswald (1996) provide empirical evidence for the importance of the reference level of income against which an individual compares himself or herself. 9 We use province instead of smaller regional units to ensure that there are sufficient observations to calculate inequality measures for each group. 10 Results are not sensitive to different ways of classifying age groups. The main reason why we focus on this age category is to make sure that each age group has enough observations.
194
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Fig. 2. The Gini coefficient across provinces and age groups.
Our analysis is based on the consumption of the whole family because there is no information on each member’s expenditure. We drop families whose heads are under the age of 25 or above the age of 75 because we cannot find enough families in these age groups to obtain reliable estimates of inequality measures. Finally, we exclude outliers with an annual disposable income of less than 100 RMB, the 10 families with the highest income (annual disposable income of more than 500,000 RMB), families whose living expenditure is either five times larger than their income or two times larger than the income but larger than 200,000 RMB, and families whose family size is larger than 10.11 Our final sample includes 102,971 families in total. The simple time-series trend in Fig. 1 suggests that the household consumption rate declined, whereas inequality rose in our sample period. The average propensity to consume (APC) net of education expenditures dropped sharply from 82% in 1997 to 75% in 2006, whereas the within-reference-group inequality rose dramatically from 0.23 to 0.29. The within-province inequality exhibits a trend similar to that of the within-reference-group inequality. Fig. 2 suggests that there are large variations of the Gini coefficient across provinces and age groups, a pattern that bodes well for estimations.
4. Estimation results This section reports regression results. All regressions are Ordinary Least Square estimations. Standard errors are robust to heteroskedasticity and clustered at the province-age level. We focus on the results using the Gini coefficient as the measure for income inequality. Estimates based on other measures (reported in Section 6) show similar patterns.
4.1. Income inequality and consumption Table 1 reports the effect of income inequality on household consumption net of education expenditures. We control for family income, age of household’s head, family size, average income of the reference group, and the province, age-group, and year fixed effects in these regressions. We also allow for the linear time trend of the average consumption to differ across provinces and age groups. The first column shows that when income increases by 1%, consumption (net of education expenditures) rises by 0.756%, which implies that the average propensity to consume declines by 0.244%. 11 In 2002, the questionnaire was substantially changed. A greater number of questions were included, and the sample size was expanded from 21,000 to 56,000. These adjustments may present inconsistencies in the caliber corresponding to the same variables between the data from 2002 to 2006 and the data from 1997 to 2001. Accordingly, our estimation is based on the data from 1997 to 2006. We use the data of 2002–2006 as a robust check, and the results are similar.
195
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204 Table 1 OLS regressions estimating the effect of income inequality on household consumption. Dependent variables:
Log(consumption net of education expenditures) (2)
(3)
(4)
(5)
(6)
0.756a (0.009) 0.001c (0.001) 0.048a (0.006) 0.049 (0.037)
0.387a (0.121) 0.756a (0.009) 0.001c (0.001) 0.048a (0.006) 0.063 (0.039)
0.413a (0.124) 0.729a (0.012) 0.001c (0.001) 0.059a (0.008) 0.089b (0.041) 0.033a (0.006) 0.042a (0.011)
0.424a (0.125) 0.718a (0.014) 0.001c (0.001) 0.060a (0.008) 0.090b (0.041) 0.064 (0.171) 0.154 (0.233) 0.010 (0.018) 0.020 (0.023)
0.346a (0.109) 0.252a (0.019) 0.001c (0.001) 0.044a (0.006) 0.108a (0.034) 0.466a (0.163) 0.829a (0.192) 0.051a (0.017) 0.087a (0.020)
102,971 0.70
102,971 0.70
102,971 0.70
102,971 0.70
0.443a (0.156) 0.718a (0.014) 0.001c (0.001) 0.059a (0.007) 0.088c (0.048) 0.064 (0.171) 0.153 (0.234) 0.010 (0.018) 0.019 (0.024) Y Y 102,971 0.70
Gini Log(income) Age Equivalent family size Group average income Middle class Rich Middle class log(income) Rich log(income) Province dummies year dummies Age-group dummies year dummies Observations Adjusted R2
APC
(1)
102,961 0.10
Note: We control for the year, province, and age-group fixed effects, interactions between provincial dummies and year, and interactions between agegroup dummies and year in all regressions. APC is the average propensity to consume (net of education expenditures). Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
The second column shows that after we control for disposable family income, the Gini coefficient has a significant negative effect on household consumption. More specifically, when the Gini coefficient rises by 0.1, household consumption drops by 3.87%. The effect of inequality on consumption still holds when we allow income to have non-linear effects. Specifically, in column 3, we include a ‘‘middle-class’’ dummy and a ‘‘rich’’ dummy, which are the middle 1/3 and top 1/3 in the income distribution of the reference group, respectively, and in column 4 both income-group dummies and their interactions with log(income) are controlled for. The estimated effect of income inequality becomes stronger as we allow for more flexible non-linear income effects on consumption. Column 4 (our preferred specification) indicates that when the Gini coefficient increases by 0.1, consumption or the average propensity to consume declines by 4.24%. When we control for the annual time trend for each province and each age group in column 5, the effect of inequality becomes even stronger. Results are comparable when we use the average propensity to consume instead of log consumption. In the last column of Table 1, we present the estimate using the average propensity to consume (APC), defined as the ratio of consumption net of education expenditures to disposable income, as the dependent variable. When the Gini coefficient increases by 0.1, APC declines by 3.46 percentage points, which is comparable with estimates using log consumption as the dependent variable (columns 1–5).12
4.2. Potential explanations There are several potential explanations for the negative effect of income inequality on consumption net of education expenditures. The first probable explanation concerns the housing market in China. Income inequality among members of the reference group may be positively correlated with provincial housing prices, and rising housing prices may reduce household consumption and stimulate savings for housing. To test whether housing prices can account for the correlation between inequality and consumption, we augment model (1) by including in it the variable ‘‘provincial housing price index.’’ The first column of Table 2 shows that after controlling for the housing price index, the coefficient of the Gini does not change much from those reported in Table 1.13 Moreover, housing price itself does not have a significant effect on consumption. The second explanation is about the correlation between inequality and downward income risks. Generally, higher income risks result in more precautionary savings and less consumption. Therefore, rising inequality can have a negative effect 12 13
Theoretically, using log consumption or log(APC) (=log consumption log income) as dependent variable should yield the same coefficient on Gini. The housing price index is only available after 1999.
196
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Table 2 OLS regressions examining potential explanations for the association between consumption and inequality. Dependent variable: log(consumption net of education expenditures)
Gini Log(income) Log(provincial housing prices)
(1)
(2)
(3)
(4)
(5)
(6)
0.436a (0.130) 0.717a (0.014) 0.014 (0.028)
0.453a (0.152) 0.713a (0.013)
0.453a (0.125) 0.718a (0.014)
0.461a (0.138) 0.708a (0.013)
0.423a (0.125) 0.718a (0.014)
0.465a (0.154) 0.706a (0.012) 0.058 (0.035) 0.011 (0.035) 0.936b (0.462) 0.194a (0.039) 0.309a (0.070) 0.224 (0.270) 63,372 0.69
Downward income risks
0.024 (0.038) 0.778b (0.353)
Education return
0.216a (0.039) 0.269a (0.076)
Social security beneficiary rate Social security contribution rate Sex ratio Observations Adjusted R2
92,284 0.69
63,372 0.69
102,971 0.70
76,304 0.70
0.044 (0.185) 102,971 0.70
Note: In columns 2, 4, and 6, we use the data between 2002 and 2006 because the relevant information is not available before 2002. Education return is estimated based on the classical Mincer regression. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level.
on consumption due to the increase in income risks. To test this assertion, we exploit the fact that the Urban Household Survey is a rotating panel and measure the downward income risk by calculating the proportion of families in a province-age group that experienced a decrease in income (for families that are in the sample 2 years in a row).14 We find that the correlation between income inequality and downward income risk is 0.31, but only 30% of the families experienced a decrease in income, and less than 15% of the families had a decrease of more than 10%. Moreover, after we add the downward income risk variable in the baseline model (column 2), the negative effect of the Gini is even larger, indicating that the income risk is not the reason for the negative effect of income inequality on consumption. Similar outcome appears when we measure the income risk by the proportion of families that experienced a decrease in income by more than 10% in the reference group. The third explanation is related to the rate of return to education. Regions with higher inequality are also likely to have higher return to education, and people tend to invest more on education and less on other types of consumption. The third column shows estimates controlling for the education returns estimated by Mincer regressions. Again, there is little change in both the magnitude and the significance level of the effect of inequality on consumption. The return to education is positively correlated with the consumption net of educational expenditure.15 The following explanation concerns the phenomenon that an unequal society often does not have a good social safety net and that the negative correlation between inequality and consumption could be due to the precautionary-savings motive. We use two city-level measures reflecting the coverage rate of the social security system to proxy the development of the social safety net: the beneficiary rate and the contribution rate. The beneficiary rate is defined as the ratio of the number of current beneficiaries of the social security program to the total number of people over age 60 and who have retired. The contribution rate is the ratio of the number of contributors to the social security program to the number of workers under 60 and who have not retired. The second measure is only available after 2002 in our dataset. The result in column 4 shows little change in the response of consumption to income inequality. The coverage rate of social security in both measures stimulates more consumption as expected. The next plausible explanation is related to the imbalanced sex ratio in China. Wei and Zhang (2009) argue that in the marriage market, competition resulting from the rising sex-ratio imbalance accounts for about half of the actual increase in the household savings rate during the period of 1990–2007. If the sex ratio is correlated with our inequality measures, our estimation of the effect of inequality without controlling for the sex ratio will be biased. In column 5 of Table 2, we include the sex-ratio variable used by Wei and Zhang (2009), which is the sex ratio for the age cohorts 7–21. Controlling for the sex ratio does not change our main results. Sex ratio itself has no significant effect on consumption in our sample. Applying the sex ratio for the age cohorts 16–25 based on the 1990 census delivers similar results.
14 The UHS had no consistent identifier for each family until 2002. Thus, this part of the analysis focuses on the survey pertaining to the 2002–2006 period. We construct a two-year panel with 17,000 families. 15 The education return is estimated for each province in each year by a classical Mincer regression, in which we regress log wage on the year of schooling, working experience, square of experience, and gender.
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
197
In the last column of Table 2, we include all the variables discussed above. The negative effect of income inequality on consumption remains almost unchanged. Therefore, these explanations are not likely to be the main reasons for the reduction in consumption in response to the rising inequality.
5. Status-seeking and household consumption behavior In this section, we propose another potential explanation for the negative effect of income inequality on consumption net of education expenditures: Chinese save for status-seeking motives. We first discuss the literature on status-seeking and describe a potential theory. The status-seeking theory can generate some additional testable hypotheses, which will then be tested empirically.
5.1. Status-seeking savings According to Weiss and Fershtman (1998), social status is the rank of an individual or a group of individuals in a given society; the rank rests on commonly agreed-upon criteria such as wealth, education, origin, and occupation. In addition, social status is often associated with a particular group and is shared by all members of the group, regardless of their individual characteristics. Sociologists have long emphasized that individuals care about social status and that human behaviors often reflect motivations to improve personal rank in the given hierarchy, which is no less important than pecuniary rewards such as consumption.16 Early justifications focus on psychological returns of higher status. Cole et al. (1992) and Corneo and Jeanne (1999) emphasize the advantages of the high status in accessing non-market resources and show that people care about their status in equilibrium even if social status does not directly enter the utility function. These researchers highlight some social rewards for attaining higher social status. First, there is a favorable club effect. Many social activities or opportunities, such as marriage and invitation to a party or club, occur only within a group of people possessing the same social status. Second, the benefits of status improvement are related to the fact that people with high status may have privileges in the rationing of non-market goods. Furthermore, people with high status can gain trust, courtesy, and approval, and build up leadership more easily. Due to these rewards, individuals try to improve their social status through investments in assets (e.g., physical, human, and social capital), group affiliation, and an appropriate choice of actions, among which the status-seeking savings motive has been the focus in the literature (Weiss and Fershtman, 1998). Cole et al. (1992) prove that when wealth is observable, the existence of non-market decisions can naturally yield a wealth-is-status equilibrium, where social status is determined by the rank in the wealth distribution.17 The literature review by Weiss and Fershtman (1998) also states that ‘‘to the extent that the wealthy care (or are taught to care) more about status, an equilibrium emerges in which status groups are stratified by wealth’’. As a result, status seeking becomes an important motivation for wealth accumulation. People are more frugal in consumption and save more than in the case where there is no status seeking. The idea that social status depends on the ranking in the wealth distribution is widely applied in the literature on social status, particularly in the studies on how the social-status motive affects economic growth and stock market investment. For example, Bakshi and Chen (1996) study the stock-market price and emphasize that ‘‘investors accumulate wealth not only for the sake of consumption but also for wealth-induced social status. Max M. Weber refers to this desire for wealth as the spirit of capitalism’’. The literature on social-status motive and economic growth also highlights the mechanism that socialstatus seeking can stimulate more savings and thus promote economic growth (Corneo and Jeanne, 2001; Futagamia and Shibatab, 1998; Gong and Zou, 2001; Pham, 2005). Rising income inequality can intensify the status-seeking motive through several channels. First, increases in income inequality usually make entering a high-status club more attractive because differences in (financial and non-financial) resources between the high- and low-status groups widen. This corresponds to the effect of ‘‘status prize’’ in Corneo and Jeanne (1999). Second, higher inequality implies a larger wealth gap between the high- and low-status groups, which means that low-status families need to save more to join the high-status clubs.18 These channels lead to a positive effect of income inequality on household savings. However, when the dispersion of wealth becomes so large that the marginal status gain achieved by accumulating additional wealth declines or it becomes infeasible for the low-status to move up, it is possible that rising income inequality weakens the status-seeking savings incentives (Corneo and Jeanne, 2001).19 To summarize, rising income inequality on average has a positive (negative) effect on savings (consumption) unless the increase in inequality induces many families to switch from seeking status to giving up. 16
See Weiss and Fershtman (1998) for a review of the literature. They show that this equilibrium can be more stable than the aristocratic equilibrium (or, say, the status-inherited equilibrium) because the latter is not subgame perfect. Social status also depends on other indicators such as education, occupation, and relative income (Fershtman et al., 1996; Neumark and Postlewaite, 1998). 18 In addition, dispersion in income distribution may increase stratification in society and intensify the competition for social status. This raises the marginal returns of savings in social competition, in accordance with the effect of ‘‘increasing segmentation’’ in Corneo and Jeanne (1999). 19 Corneo and Jeanne (2001) show that the dispersion of initial wealth distribution reduces status-seeking savings. 17
198
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Table 3 OLS regressions estimating the effect of inequality on consumption across income groups. Dependent variable: log(consumption net of education expenditures)
Gini Log(income) Log(income) Gini
Whole sample (1)
Whole sample (2)
Poor (3)
Middle class (4)
Rich (5)
3.431b (1.674) 0.626a (0.050) 0.299c (0.164)
0.713a (0.174) 0.713a (0.014)
0.557a (0.200) 0.727a (0.017)
0.465a (0.164) 0.723a (0.032)
0.166 (0.211) 0.710a (0.015)
0.322b (0.150) 0.508b (0.213) 0.112 (0.166) 0.212 (0.221) 102,971 0.70
33,598 0.64
34,278 0.55
35,095 0.55
Middle class Gini Rich Gini Middle class Rich Observations Adjusted R2
0.037 (0.163) 0.082 (0.217) 102,971 0.70
Note: The poor, middle, and rich are the top, middle, and bottom one-third of the income distribution, respectively. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
The theory of status seeking will lead to some further testable hypotheses, including the heterogeneous effect of inequality with income and age and the effect of inequality on other instruments that can help improve social status, on conspicuous consumption, and on subsistence consumption. We will explore and empirically test these hypotheses next. 5.2. Heterogeneous effects across income groups The status-seeking motives imply that the effect of income inequality on consumption net of education expenditures can vary across income groups. To illustrate this, suppose there are three income-clubs: the rich, the middle-class, and the poor. The rich have no incentive to save for the purpose of joining a club because their wealth is already high enough to be in the highest club. In contrast, the poor and the middle class have the incentive to accumulate wealth to climb up to the next social ladder (Long and Shimomura, 2004).20 To examine how the effect of inequality varies with income, we estimate the following model:
lnðCÞ ¼ a þ b lnðYÞ þ c1 Gini þ c2 Gini lnðYÞ þ d X þ e:
ð2Þ
Results reported in the first column in Table 3 show that the negative effect of inequality on consumption indeed declines with income. When income increases by 1%, the effect of the Gini coefficient on consumption declines by about 0.3 percentage points. Next, we allow the effect of the Gini coefficient to vary nonlinearly with income. Particularly, we again consider three income groups: the poor for the bottom one-third of the income distribution within the reference group, the middle class for the middle one-third, and the rich for the top one-third, and estimate the following model:
lnðCÞ ¼ a þ b lnðYÞ þ c1 Gini þ c2 middle þ c3 Gini middle þ c4 rich þ c5 Gini rich þ d X þ e:
ð3Þ
The variable ‘‘middle’’ is a dummy variable that equals 1 if the family belongs to the middle class, ‘‘rich’’ is a dummy variable that equals 1 if the family is rich, and the (omitted) reference group is the poor. The second column of Table 3 shows that the coefficient of the interaction between the middle class and Gini is significantly positive. For this group, the effect of inequality is 0.391, which is much smaller in magnitude than that for the poor (0.713). The interaction term of rich Gini is also positive and significantly different from zero, with an estimated coefficient of 0.508, suggesting that the effect of inequality for the rich is even smaller (only 0.205). Regressions based on the three sub-samples (i.e., the poor, the middle class, and the rich) exhibit a similar pattern (columns 3–5), revealing that the effect of inequality on consumption for the rich is not significantly different from zero. The status-seeking theory implies the possibility of giving up: when inequality is too large, the poor may not be able to save enough to enter the next club, and thus they simply give up. Our empirical findings indicate that the poor in urban China do not give up. One possible reason is that poor households in urban China may not be poor enough to give up, because most of the poor in China live in rural areas, which are not in our sample. 20 When the size of the highest-status club is fixed, even the richest families have the incentive to accumulate wealth to defend their status in the ‘‘rat-race.’’ However, their incentives are still weaker than those attributable to families who need to save more both to upgrade and to prevent downgrading at the same time.
199
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204 Table 4 OLS regressions estimating the effect of inequality on consumption across age groups. Dependent variable: log(consumption net of education expenditures)
Gini Log(income) Age Gini
Whole sample (1)
Whole sample (2)
Young (head age<55) (3)
Old (head age P 55) (4)
1.072b (0.421) 0.846a (0.027) 0.011 (0.008)
0.291 (0.183) 0.682a (0.020)
0.652a (0.126) 0.714a (0.016)
0.317c (0.178) 0.721a (0.024)
0.370c (0.202) 0.375a (0.126) 102,971 0.70
78,493 0.71
24,478 0.66
Young Gini Young dummy Observations Adjusted R2
102,971 0.70
Note: Young means household head is younger than 55, and old means household head is 55 or older. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), the age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
5.3. Young versus old The status-seeking motive is also likely to vary with age. The benefit of attaining better status obviously lasts longer for the young than for the old. Therefore, when rising income inequality intensifies social rivalry, younger households tend to restrain more consumption to compete with their peers.21 Regression results reported in Table 4 show that the effect of inequality on consumption net of education expenditures declines with age. In the first column of Table 4, we allow the effect of inequality to vary linearly with age by including an interaction term of age Gini. Although the positive sign of this interaction term indicates that the negative effect of inequality becomes weaker as the head of the family becomes older, it is not statistically significant. In the second column, we allow the effect of inequality to vary with age nonlinearly, and we find that the effect of inequality on consumption is stronger for the young (household head younger than 55) than for the old (55 or older), and this difference between the young and old is statistically significant. Regressions based on the two sub-samples (young versus old) show a similar pattern (columns 3 and 4): for the young, income inequality has a negative effect on consumption, whereas the effect is much weaker for the old. 5.4. Status-seeking education investment Aside from accumulating wealth through savings, another way of improving status is investing in education. Education is considered not only an indicator of social status itself but is also correlated with higher income and wealth. Thus, we hypothesize that rising income inequality may encourage households to increase education investment.22 We consider only families with children between age 3 and 30 because there is generally very little education investment for adults. Results reported in Table 5 show that rising income inequality has a positive effect on education investment. We have different model specifications in columns 1–4, and all show that the Gini coefficient has a strong positive effect on household education expenditures. Our preferred specification, which is column 1 that allows the effect of income to be flexible, suggests that when the Gini coefficient goes up by 0.1, education expenditures increase by 50.6%. Unlike the results for consumption, the effect of income inequality on education expenditures does not differ significantly across different income groups (columns 5). One explanation for this result is that the human capital of the parents is more difficult to pass onto the next generation than traditional forms of wealth; thus, the rich also have a strong incentive to invest in children’s education when inequality increases. 5.5. Status-showing consumption People may try to show their status in the form of conspicuous consumption, that is, consumption of particularly visible and expensive, although not necessarily very useful, goods. In this case, higher inequality means more (conspicuous) consumption and less savings. However, as conspicuous consumption cannot allow one to enter a social club whose member21
Consumption of the old also responds to inequality because of altruism or the bequest motive. The same argument does not fit the health expenditures, as basic health is a necessity for people. Empirical results (not reported) show that inequality has no effect on health consumption. 22
200
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Table 5 OLS regressions estimating the effect of inequality on households’ education expenditures. Dependent variable: log(education expenditures)
Gini Log(income) Middle class Rich Middle class log(income) Rich log(income)
(1)
(2)
(3)
(4)
(5)
5.055c (2.578) 0.979a (0.067) 0.095 (0.851) 2.646b (1.000) 0.010 (0.087) 0.254b (0.098)
4.881c (2.574) 0.909a (0.044)
4.831c (2.588) 0.859a (0.067) 0.068 (0.042) 0.075 (0.075)
7.656b (3.239) 0.978a (0.065) 0.013 (0.865) 2.613b (0.989) 0.001 (0.088) 0.250b (0.097)
10.572 (7.798)
10.531 (7.798)
10.542 (7.803)
5.297c (2.673) 0.984a (0.067) 0.145 (0.851) 2.566b (0.996) 0.023 (0.092) 0.217b (0.086) 0.282 (0.803) 1.072 (1.154) 10.570 (7.794)
Middle class Gini Rich Gini Education return Province dummies year dummies Age-group dummies year dummies Observations Adjusted R2
73,951 0.22
73,951 0.22
73,951 0.22
Y Y 73,951 0.23
73,951 0.22
Note: The poor, middle, and rich are the top, middle, and bottom one-third of the income distribution, respectively. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
Table 6 OLS regressions estimating the effect of inequality on household conspicuous consumption. Dependent variable: log(conspicuous consumption) First measure
Gini Log(income)
(2)
(3)
(4)
0.780b (0.376) 1.328a (0.065)
0.030 (0.464) 1.293a (0.071)
1.012b (0.400) 1.354a (0.065)
1.250c (0.649) 2.404a (0.760)
2.536a (0.622) 1.300a (0.054) 2.054a (0.491) 2.982a (0.620) 0.951c (0.531) 2.047a (0.546)
1.186c (0.634) 2.101a (0.729)
102,971 0.44
102,971 0.45
1.260b (0.624) 2.375a (0.721) 1.217c (0.636) 0.427 (0.487) 102,971 0.45
Middle class Gini Rich Gini Middle class Rich
Second measure
(1)
Young Gini Young Dummy Observations Adjusted R2
102,971 0.46
Note: The first measure of conspicuous consumption includes expenditure on jewelry, watches, and appearance; the second one includes additional car expenditures. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
201
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204 Table 7 OLS regressions estimating the effect of inequality on household subsistence consumption. Dependent variables:
Log(cereals expenditure) (1)
Log(food expenditure) (2)
Log(cereals expenditure) (3)
Log(cereals expenditure) (4)
Gini
0.007 (0.367) 0.033c (0.017)
0.137 (0.150) 0.502a (0.013)
0.062 (0.260) 0.081a (0.020)
0.164 (0.164) 0.372 (0.244)
0.377b (0.149) 0.660a (0.232)
0.257 (0.393) 0.003 (0.016) 0.266 (0.200) 0.408 (0.271) 0.083 (0.063) 0.178b (0.087)
76,413 0.17
103,086 0.59
76,413 0.17
Log(income) Middle class Gini Rich Gini Middle class Rich Young Gini Young Dummy Observations Adjusted R2
0.139 (0.161) 0.363 (0.270) 0.203 (0.435) 0.055 (0.218) 76,413 0.17
Note: Food expenditure includes cereals, cakes, fish, beverages, and restaurant orders. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
ship is determined by the level of wealth, poor people may not have the incentive to do so (Corneo and Jeanne, 1998). Ultimately, whether conspicuous consumption increases with inequality remains an empirical question. In this paper, we have two measures for conspicuous consumption: the first measure includes expenditure on jewelry, watches, and appearance (e.g., clothes, hairdressing, cosmetics, beauty, and bath), and the second measure includes additional automobile expenditures. Regression results reported in Table 6 suggest that conspicuous consumption is also negatively affected by income inequality. The magnitude is actually stronger than that for other types of consumption. The variations of the effect of inequality across income and age groups are also similar to those for the overall consumption. This is consistent with our expectation: poorer and younger individuals are less likely to engage in such conspicuous expenditure because doing so cannot really gain them access to the wealthy club. It actually lowers their chances of getting into the wealthy club.23 5.6. Subsistence consumption We argue that income inequality has a negative effect on household consumption because of the status-seeking motive underlying the act of saving, but this motive should not affect the basic or subsistence consumption for human needs. If we find a similar effect of income inequality on basic consumption, then the status-seeking theory is not well grounded. As expected, regression results reported in the first column of Table 7 show that income inequality has no effect on subsistence consumption (consumption of cereal). When we switch to non-basic food consumption as a dependent variable, including cakes, fish, beverages, and restaurant orders (column 2), the negative effect of inequality on food consumption becomes stronger although still not significant. Columns 3 and 4 show that inequality has no effect on consumption for all income and age groups. In summary, our empirical findings align well with the prediction based on the status-seeking story, and other potential theories cannot explain the negative effect of inequality on consumption. In results not reported, we find that the variations of the effects of inequality across income and age groups are robust after we control for the provincial housing price, downward income risks, education returns, social security coverage, and/or sex imbalance. The results for education investment, conspicuous consumption, and basic consumption also survive. 6. Robustness tests In this section, we check the robustness of the results using other measures of inequality, including the 90 percentile/10 percentile income ratio, the 75 percentile/25 percentile ratio, and the Theil index, which are more responsive to the distribution at the tails. Results reported in Table 8 show a similar pattern of the effect of inequality on consumption net of education expenditures. For example, when the 90/10 (75/25) income ratio increases by 1, non-educational consumption 23 This is consistent with the findings in other studies. For example, Wei and Zhang (2009) suggest that for marriage, most people believe wealth accumulation is more likely to provide long-term security than conspicuous consumption does.
202
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Table 8 OLS regressions estimating the effect of inequality on household consumption: different measures of income inequality. Dependent variable: log(consumption net of education expenditures)
Inequality Log(income)
90/10 percentile income ratio
75/25 percentile income ratio
Theil index
90 income percentile–10 income percentile
(1)
(2)
(3)
(4)
(5)
(6)
0.016a (0.005) 0.718a (0.014)
0.006 (0.007) 0.682a (0.020)
0.039a (0.013) 0.719a (0.014)
0.354a (0.090) 0.718a (0.014)
0.053a (0.014) 0.719a (0.014)
0.068 (0.171) 0.153 (0.233)
0.031a (0.008) 0.712a (0.014) 0.016b (0.008) 0.026b (0.010) 0.096 (0.161) 0.179 (0.215)
0.066 (0.171) 0.149 (0.232)
0.062 (0.171) 0.150 (0.233)
0.067 (0.171) 0.153 (0.233)
102,971 0.70
102,971 0.70
0.089 (0.170) 0.167 (0.223) 0.022a (0.008) 0.400a (0.120) 102,971 0.70
102,971 0.70
102,971 0.70
102,971 0.70
Middle class Gini Rich Gini Middle class Rich Young Gini Young Dummy Observations Adjusted R2
Note: We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. Indicate statistical significance at the 10% level. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level.
declines by 1.6% (3.9%), and the effect is stronger for the poor and young. In the last column of Table 8, we use the absolute income difference between the 90th percentile and 10th percentile as the inequality measure; the results are similar.24 Finally, we examine the effect of inequality on total consumption, including educational expenditures. As previous results indicate that income inequality has opposite effects on education expenditures and other consumption, whether the effect on total consumption is positive or negative remains an empirical question. The first column in Table 9 shows that although income inequality has a negative coefficient, it is not statistically significant. However, it becomes significant at the 5% level when we allow each province-age group to have a different linear time trend (column 2) and the magnitude is large. Furthermore, column 3 shows that inequality has a smaller effect on the consumption of the middle class and the rich and that the differences are significant. These results are not surprising given that the education expenditures are only 8.7% of the total consumption in our sample. The last two columns show that when the Gini coefficient increases by 0.1, the APC declines by 2.72 to 2.94 percentage points.
7. Conclusions The rapid increase in income inequality in China and the substantial variation in inequality across regions and groups provide us a good opportunity to examine how the rising inequality affects consumption. Using the Chinese Urban Household Survey data, we find that income has a negative effect on household consumption net of education expenditures after we control for the income effect itself. The estimate indicates that rising inequality can explain a large part of the declining non-educational consumption as a percentage of income between 1997 and 2006. In our sample, the urban Gini coefficient rose from 0.23 in 1997 to 0.29 in 2006 on average, and the estimates imply a decline in the average propensity to consume (APC) by 2.54% or by 2.08 percentage points, which can explain 30.1% of the decline of urban household APC excluding education expenditures during this period. The declining propensity to consume with income cannot explain the remaining correlation between inequality and consumption after we control for income, thus calling for a new explanation. We propose and test several potential stories including rising housing prices, high income risks, high return to education, poor social security and rising sex ratios. However, none of these can account for the observed inequality–consumption link. Finally, we propose an account based on the status-seeking motive. We argue that people save to improve their social status when social status is associated with pecuniary and non-pecuniary benefits. Rising income inequality can strengthen the incentives of status-seeking savings by increasing the benefit of improving status, and by enlarging the wealth level required for status upgrading. The status-seeking theory also implies that the negative effect of income inequality on consumption is stronger for poorer and younger people. Moreover, rising inequality should have a strong positive effect on families’ investment in education, 24
Using the income indifference between the 75th and 25th percentile gives similar estimates.
203
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204 Table 9 OLS regressions estimating the effect of inequality on household total consumption. Dependent variables
Gini Log(income) Middle class Rich
Log(total consumption) (2)
(3)
(4)
(5)
0.139 (0.123) 0.718a (0.015) 0.157 (0.184) 0.217 (0.245)
0.280b (0.130) 0.717a (0.015) 0.150 (0.185) 0.213 (0.246)
0.390b (0.171) 0.714a (0.015) 0.199 (0.177) 0.269 (0.235) 0.294c (0.152) 0.426c (0.228)
0.272c (0.139) 0.427a (0.095) 0.544 (0.661) 1.634a (0.542)
0.294b (0.132) 0.428a (0.095) 0.541 (0.662) 1.631a (0.544)
102,971 0.69
Y 102,971 0.69
102,971 0.69
102,971 0.08
Y 102,971 0.08
Middle class Gini Rich Gini Province-age-group dummies year Observations Adjusted R2
APC (including education expenditures)
(1)
Note: We control for the year, province, and age-group fixed effects, the average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. APC1 is the average propensity to consume (including education). Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
which is an important indicator of social status. Additional empirical analyses confirm both hypotheses. Additionally, we find that inequality has a negative effect on conspicuous consumption, and confirm that it has no impact on basic (or subsistence) consumption. Finally, our results are robust to different measures of inequality. Acknowledgments We would like to thank Daniel Berkowitz, two anonymous referees, Chongen Bai, conference and seminar participants for their valuable suggestions. Any remaining errors are ours alone. Hongbin Li acknowledges supports from the National Natural Science Foundation of China (Project ID: 71025004). Binzhen Wu acknowledges supports from the National Natural Science Foundation of China (Project ID: 70903042). We also acknowledge supports from the China Data Center of Tsinghua University and Xiaolong Chen for data supports.
Appendix A Table A1. Table A1 Descriptive statistics of the variables.
Gini coefficient
Disposable income Consumption (excluding education expenditures) APC (excluding education expenditures) APC (including education expenditures) Education expenditures Education (years) Family equivalent scale Family size Age
Year
Observations
Mean
Standard deviation
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Total
5392 5382 5378 5355 5378 12,878 14,379 15,593 16,677 16,822 103,234
0.232 0.243 0.244 0.270 0.271 0.274 0.286 0.289 0.288 0.289 0.277
0.033 0.033 0.027 0.034 0.031 0.035 0.041 0.038 0.037 0.039 0.041
104,665 104,550 104,541 104,551 104,665 104,665 104,665 104,665 104,665
28,178 21,858 0.77 0.84 1629 11.58 2.28 3.00 48.31
21,101 17,099 0.31 0.55 3094 2.99 0.49 0.77 11.32
204
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
References Alesina, Alberto F., Perotti, Roberto, 1996. Income distribution, political instability, and investment. European Economic Review 40, 1203–1228. Atkinson, Anthony B., Rainwater, Lee, Smeeding, Timothy M., 1995. Income Distribution in OECD Countries. OECD Social Policy Studies, No. 18, Paris. Bakshi, Gurdip S., Chen, Zhiwu, 1996. The spirit of capitalism and stock-market prices. American Economic Review 86, 133–157. Clark, Andrew E., Oswald, Andrew J., 1996. Satisfaction and comparison income. Journal of Public Economics 61, 359–381. Chamon, Marcos D., Prasad, Eswar S., 2010. Why are saving rates of urban households in China rising? American Economic Journal: Macroeconomics, American Economic Association 2 (1), 93–130. Cole, Harold L., Mailath, George J., Postlewaite, Andrew, 1992. Social norms, savings behavior, and growth. The Journal of Political Economy 100, 1092–1125. Coleman, James S., 1990. Foundations of Social Theory. Harvard University Press, Cambridge, MA. Corneo, Giacomo, Jeanne, Olivier, 1998. Social organization, status and savings behavior. Journal of Public Economics 70, 37–51. Corneo, Giacomo, Jeanne, Olivier, 1999. Social organization in an endogenous growth model. International Economic Review 70, 37–51. Corneo, Giacomo, Jeanne, Olivier, 2001. Status, the distribution of wealth, and growth. The Scandinavian Journal of Economics 103, 283–293. Dynan, Karen E., Skinner, Jonathan, Zeldes, Stephen P., 2004. Do the rich save more? Journal of Political Economy 112, 397–444. Fershtman, Chaim, Murphy, Kevin M., Weiss, Yoram, 1996. Social status, education, and growth. The Journal of Political Economy 104, 108–132. Futagamia, Koichi, Shibatab, Akihisa, 1998. Keeping one step ahead of the joneses: status, the distribution of wealth, and long run growth. Journal of Economic Behavior & Organization 36, 109–126. Gong, Liutang, Zou, Hengfu, 2001. Money, social status, and capital accumulation in a cash-in-advance model. Journal of Money Credit and Banking 33, 284– 293. Horioka, Charles Y., Junmin, Wan, 2007. The determinants of household saving in china: a dynamic panel analysis of provincial data. Journal of Money, Credit, and Banking. Kraay, Aart, 2000. Household saving in china. World Bank Economic Review 14, 545–557. Kuijs, Louis, 2006. How will China’s saving-investment balance evolve? World Bank Policy Research Working Paper #3958. Li, Hongbin, Zhu, Yi, 2006. Income, income inequality, and health: evidence from China. Journal of Comparative Economics 34, 668–693. Long, Ngo V., Shimomura, Koji, 2004. Relative wealth, status-seeking, and catching-up. Journal of Economic Behavior & Organization 53, 529–542. Menchik, Paul L., David, Martin, 1983. Income distribution, lifetime savings, and bequests. The American Economic Review 73, 672–690. Meng, Xin, 2003. Unemployment, consumption smoothing, and precautionary saving in urban China. Journal of Comparative Economics 31 (3), 465–485. Modigliani, Franco, Cao, Shi L., 2004. The Chinese saving puzzle and the life-cycle hypothesis. Journal of Economic Literature 42, 145–170. Musgrove, Philip, 1980. Income distribution and the aggregate consumption function. The Journal of Political Economy 88, 504–525. Neumark, David, Postlewaite, Andrew, 1998. Relative income concerns and the rise in married women’s employment. Journal of Public Economics 70, 157– 183. Pham, Thi Kim Cuong, 2005. Economic growth and status-seeking through personal wealth. European Journal of Political Economy 21, 407–427. Schmidt-Hebbel, Klaus, Serven, Luis, 2000. Does income inequality raise aggregate saving? Journal of Development Economics 61, 417–446. Smith, Douglas, 2001. International evidence on how income inequality and credit market imperfections affect private saving rates. Journal of Development Economics 64, 103–127. Stoker, Thomas M., 1986. Simple tests of distributional effects on macroeconomic equations. The Journal of Political Economy 94, 763–795. Wei, Shangjin, Zhang, Xiaobo, 2009. The Competitive Saving Motive: Evidence from Rising Sex Ratios and Savings Rates in China. NBER Working Paper No. w15093. Weiss, Yoram, Fershtman, Chaim, 1998. social status and economic performance: a survey. European Economic Review 42, 801–820. Wen, Yi., 2009. Saving and Growth under Borrowing Constraints Explaining the ‘High Saving Rate’ Puzzle. Working Papers 2009-045, Federal Reserve Bank of St. Louis. World Bank, 2005. World Development Report 2006: Equity and Development. Washington, DC: World Bank.
Contents lists available at ScienceDirect
Journal of Comparative Economics journal homepage: www.elsevier.com/locate/jce
Income inequality, consumption, and social-status seeking Ye Jin, Hongbin Li, Binzhen Wu ⇑ School of Economics and Management, Tsinghua University, Beijing, 100084, PR China
a r t i c l e
i n f o
Article history: Received 18 May 2010 Revised 17 December 2010 Available online 4 January 2011 JEL classification: D12 D91 E21 P46 Keywords: Income inequality Social status Consumption and savings Status seeking Education investment
a b s t r a c t Jin, Ye, Li, Hongbin, and Wu, Binzhen—Income inequality, consumption, and social-status seeking Using the Chinese Urban Household Survey data between 1997 and 2006, we find that income inequality has a negative (positive) effect on household consumption net of education expenditures (savings) even after we control for household income. We argue that people save to improve their social status when social status is associated with pecuniary and non-pecuniary benefits. Rising income inequality can strengthen the incentives of status-seeking savings by increasing the benefit of improving status, and by enlarging the wealth level required for status upgrading. We also find that the negative effect of income inequality on consumption is stronger for poorer and younger people and that income inequality stimulates more education investment, which are consistent with the statusseeking hypothesis. Journal of Comparative Economics xxx (xx) (2011) xxx–xxx. School of Economics and Management, Tsinghua University, Beijing, 100084, PR China. Ó 2010 Association for Comparative Economic Studies Published by Elsevier Inc. All rights reserved.
1. Introduction The Chinese consumption rate has been low and declining in the last two decades. According to the Chinese Urban Household Survey data, the average propensity to consume (net of education expenditures) of urban residents declined from 81.6% in 1997 to 74.7% in 2006.1 Several explanations have been proposed for this phenomenon, including demographic changes (Kraay, 2000; Modigliani and Cao, 2004), high income growth and habit (Horioka and Wan, 2007), precautionary savings (Chamon and Prasad, 2010; Kuijs, 2006; Meng, 2003), change in the return rate of investment (Wen, 2009), and increasing sex ratio (Wei and Zhang, 2009). In this paper, we explore another potential cause of declining consumption: rising inequality. The classical theory on the relationship between income inequality and the consumption rate emphasizes that if the propensity to consume decreases with income, then at the macro level, rising inequality could reduce the consumption rate. This ‘‘macro’’ reason for the inequality–consumption link is mechanically caused by the non-linear association between consumption and income at the micro, or household, level. If only the macro-mechanism is at work, then inequality should not affect consumption behavior at the household level once household income is well controlled for. Empirical results drawn on the Chinese Urban Household Survey data for 1997–2006 show that income inequality within a reference group (i.e., families in the same province and same age group) has a negative effect on the household ⇑ Corresponding author. E-mail address: [email protected] (B. Wu). The average propensity to consume (APC) is defined as living expense net of education expenses divided by disposable income. The change in the APC here is similar to that in Chamon and Prasad (2010). 1
0147-5967/$ - see front matter Ó 2010 Association for Comparative Economic Studies Published by Elsevier Inc. All rights reserved. doi:10.1016/j.jce.2010.12.004
192
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
consumption rate (net of education expenditures) after we control for household income. This implies the existence of mechanisms at work other than the ‘‘macro’’ mechanism. We explore several potential explanations for the observed inequality– consumption link, including rising housing prices, high income risks, high return to education, poor social security, and rising sex ratios. However, none of these can account for the observed inequality–consumption link. We then propose and explore another potential explanation for the observed inequality–consumption link at the household level based on people’s desire to improve their social status. More specifically, people care about their social status, which is associated with pecuniary and non-pecuniary benefits. Social status often depends on a family’s rank in the wealth distribution or indicators such as educational attainment, which is closely associated with wealth when the credit market is imperfect. As a result, in order to ascend in the status hierarchy or keep the social status in the ‘‘Rat Race’’, families try to accumulate wealth by increasing savings. When income inequality increases, the benefit gap between the high-status and low-status groups widens, which in turn strengthens the incentives of status-seeking savings. In contrast to the macromechanism, the status-seeking hypothesis implies that given household income, rising income inequality can still discourage (stimulate) household consumption (savings). Furthermore, rising income inequality also raises the entry wealth level for the high-status group, which means that more savings are needed for one to enter the high-status group. The status-seeking theory also leads to further testable hypotheses. First, income inequality has a greater effect on consumption and savings for poorer and younger people because they have stronger incentives for moving up. Second, rising inequality should have a strong positive effect on families’ investment in education, which is an important indicator of social status. Third, inequality should have a negative effect on conspicuous (or unnecessary) consumption, and should have no impact on basic (or subsistence) consumption. Additional empirical analyses confirm these hypotheses. Finally, our results are robust to different measures of inequality. The rapid increase in income inequality in China suggests that rising inequality is a very important reason for the low household consumption rate (net of education expenditures) in the country. According to World Bank (2005), China’s Gini coefficient has risen from 0.33 to 0.47 in two decades.2 In our sample, the Gini coefficient in urban areas rose from 0.23 in 1997 to 0.29 in 2006, which implies a decline of the average propensity to consume (APC) by 2.08 percentage points, about 30.1% of the decline of the APC during the period.3 To the best of our knowledge, we are the first to find micro-evidence that rising inequality could be a reason for the declining consumption. Existing studies emphasize the link as ‘‘a macro-phenomena’’ and do not consider the micromechanism, that is, the effect of inequality on an individual household’s consumption or saving behavior given one’s income.4 Alesina and Perotti (1996) propose a micro-mechanism of how income inequality could increase social tensions, thus increasing the risk of investment and reducing private saving rate. However, their empirical evidence is based on cross-country data, which are still at the macro-level. We are also among the first to show the importance of status seeking in increasing savings. The recent literature on economic growth demonstrates that the status-seeking saving motive can be beneficial for economic growth (Corneo and Jeanne, 2001; Futagamia and Shibatab, 1998; Gong and Zou, 2001; Pham, 2005), but there is little empirical micro-evidence to confirm the importance of the status-seeking motive. The rest of the paper is organized as follows. The second section introduces our empirical framework. Section 3 describes the data and presents some descriptive statistics. Section 4 presents the empirical results and tests several potential explanations for the observed inequality–consumption link. Section 5 discusses how the status-seeking motive can lead to a direct effect of income inequality on household consumption and provide further tests for the status-seeking hypothesis. Section 6 presents results of robustness checks using other measures of inequality and consumption. Section 7 concludes. 2. Empirical models To identify the effect of income inequality on consumption (saving) behavior, we estimate the following empirical model:
lnðCÞ ¼ a þ b lnðYÞ þ c Gini þ d X þ e;
ð1Þ
where C is household consumption net of education expenditures, Y is household disposable income,5 Gini is the Gini coefficient, and X includes all other covariates. The reason why we focus on consumption net of education expenditures is that education expenditures are usually considered an investment rather than consumption.6 Throughout the paper, consumption means all consumption expenses net of educational expenditures, unless otherwise specified. 2 Moreover, there is a substantial variation across regions and time. For instance, our calculation illustrates that the Gini coefficient among urban residents in Beijing rose from 0.19 in 1997 to 0.25 in 2006, whereas the measurement in Zhejiang province changed from 0.23 to 0.32 in the same period. 3 Our baseline estimation in column 6 of Table 1 indicates that when the Gini coefficient increases by 0.1, the APC declines by 3.46 percentage points. Therefore, a 0.06 increase of the Gini leads to 2.076 (=0.06 3.46) percentage points decrease in the APC, which accounts for 30.09 (=2.706/(81.6 74.7)) percent of the total decline of the APC between 1997 and 2006. 4 For example, Musgrove (1980), Menchik and David (1983), Stoker (1986) and Dynan et al. (2004) suggest that due to the differences in the concavity of utility function, precautionary saving, or bequest motive, the saving rate of high-income families is higher than that of low-income families. Smith (2001) uses cross-country data to confirm that to the extent of credit market imperfection, income inequality has a robust positive effect on aggregate private saving rates. However, there are also some studies that find no effect of income distribution on the aggregate saving rate (Schmidt-Hebbel and Serven, 2000). 5 Disposable income includes wages, capital gains, and transfers, excluding the social security contribution and income tax. 6 We will also estimate the effect of inequality on education expenses, as well as total consumption that includes education expenditures.
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
193
Fig. 1. Average propensity to consume and income inequality.
In Eq. (1), c measures the effect of income inequality on consumption. It can also be seen as the effect of inequality on the average propensity to consume (or consumption rate), since log(consumption rate) is equal to the difference between log(consumption) and log(income). For expositional simplicity, we hereafter use consumption and consumption rate interchangeably in the empirical analysis. b is the income elasticity of consumption. If the link between income inequality and consumption works only through consumers’ heterogeneity in the propensity to consume, then c should be zero, particularly when we allow b to vary with income. Note that if b < 1, the average propensity to consume declines with income, which is consistent with the literature. For the measure of inequality, we focus on the Gini coefficient, and use other inequality measures, including the income ratio of the richest 10% (25%) to the poorest 10% (25%) or the Theil index as robustness tests.7 Moreover, we use the Gini coefficient among peers or the reference group rather than the inequality of the whole society because people usually compare themselves with their peers (Coleman, 1990; Li and Zhu, 2006).8 Specifically, we use inequality within the reference province-age (plus and minus 5 years) group. For example, for families with a 31-year-old head, the reference group includes families with a head between the ages of 26 and 36 in the same province.9 All inequality indices are based on income per equivalent person, considering family size or scale. The results presented below use the equivalence scale, where one of the adults in the household has weight 1, the other adults have weight 0.7, and children’s weight is 0.5. Applying different family equivalent scales delivers similar estimates (see Atkinson et al. (1995) for definitions of family scales). Control variables in X include the head’s age, family equivalent scale, average income of the reference province-age group, provincial fixed effects, age-group (five groups: under 35, 35–44, 45–54, 55–64, and 65 or older) fixed effects,10 and year fixed effects. Finally, we control for the interactions between province dummies and year and the interactions between agegroup dummies and year to allow for a specific linear trend for each province and each age group. Thus, the identification of our model relies on the differences in the non-linear temporal changes in income inequality across age groups and provinces.
3. Data and descriptive statistics The data come from the annual Urban Household Survey (UHS) conducted by the National Bureau of Statistics in China. The UHS covers all provinces in China and uses a probabilistic sampling and stratified multistage method to select households. It is a rotating panel in which one-third of the sample is replaced each year, and the full sample is changed every 3 years. The sampled households are asked to keep detailed records of their incomes and expenditures every day. The survey yields demographic and income information for every member of the family, but unfortunately it has no information on assets. We have access to data gathered from 1997 to 2006 on the nine Chinese provinces of Beijing, Liaoning, Zhejiang, Anhui, Hubei, Guangdong, Sichuan, Shaanxi, and Gansu, which represent different regions and economic conditions. The mean values and the trends of the most important variables are comparable between our sample and the national sample. 7 The Gini coefficient reflects the average dispersion of income. The 90/10 income ratio, 75/25 income ratio, and the Theil index are more sensitive to the difference in the tails. 8 Bakshi and Chen (1996) introduce the reference group’s average wealth in their theoretical model, and Clark and Oswald (1996) provide empirical evidence for the importance of the reference level of income against which an individual compares himself or herself. 9 We use province instead of smaller regional units to ensure that there are sufficient observations to calculate inequality measures for each group. 10 Results are not sensitive to different ways of classifying age groups. The main reason why we focus on this age category is to make sure that each age group has enough observations.
194
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Fig. 2. The Gini coefficient across provinces and age groups.
Our analysis is based on the consumption of the whole family because there is no information on each member’s expenditure. We drop families whose heads are under the age of 25 or above the age of 75 because we cannot find enough families in these age groups to obtain reliable estimates of inequality measures. Finally, we exclude outliers with an annual disposable income of less than 100 RMB, the 10 families with the highest income (annual disposable income of more than 500,000 RMB), families whose living expenditure is either five times larger than their income or two times larger than the income but larger than 200,000 RMB, and families whose family size is larger than 10.11 Our final sample includes 102,971 families in total. The simple time-series trend in Fig. 1 suggests that the household consumption rate declined, whereas inequality rose in our sample period. The average propensity to consume (APC) net of education expenditures dropped sharply from 82% in 1997 to 75% in 2006, whereas the within-reference-group inequality rose dramatically from 0.23 to 0.29. The within-province inequality exhibits a trend similar to that of the within-reference-group inequality. Fig. 2 suggests that there are large variations of the Gini coefficient across provinces and age groups, a pattern that bodes well for estimations.
4. Estimation results This section reports regression results. All regressions are Ordinary Least Square estimations. Standard errors are robust to heteroskedasticity and clustered at the province-age level. We focus on the results using the Gini coefficient as the measure for income inequality. Estimates based on other measures (reported in Section 6) show similar patterns.
4.1. Income inequality and consumption Table 1 reports the effect of income inequality on household consumption net of education expenditures. We control for family income, age of household’s head, family size, average income of the reference group, and the province, age-group, and year fixed effects in these regressions. We also allow for the linear time trend of the average consumption to differ across provinces and age groups. The first column shows that when income increases by 1%, consumption (net of education expenditures) rises by 0.756%, which implies that the average propensity to consume declines by 0.244%. 11 In 2002, the questionnaire was substantially changed. A greater number of questions were included, and the sample size was expanded from 21,000 to 56,000. These adjustments may present inconsistencies in the caliber corresponding to the same variables between the data from 2002 to 2006 and the data from 1997 to 2001. Accordingly, our estimation is based on the data from 1997 to 2006. We use the data of 2002–2006 as a robust check, and the results are similar.
195
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204 Table 1 OLS regressions estimating the effect of income inequality on household consumption. Dependent variables:
Log(consumption net of education expenditures) (2)
(3)
(4)
(5)
(6)
0.756a (0.009) 0.001c (0.001) 0.048a (0.006) 0.049 (0.037)
0.387a (0.121) 0.756a (0.009) 0.001c (0.001) 0.048a (0.006) 0.063 (0.039)
0.413a (0.124) 0.729a (0.012) 0.001c (0.001) 0.059a (0.008) 0.089b (0.041) 0.033a (0.006) 0.042a (0.011)
0.424a (0.125) 0.718a (0.014) 0.001c (0.001) 0.060a (0.008) 0.090b (0.041) 0.064 (0.171) 0.154 (0.233) 0.010 (0.018) 0.020 (0.023)
0.346a (0.109) 0.252a (0.019) 0.001c (0.001) 0.044a (0.006) 0.108a (0.034) 0.466a (0.163) 0.829a (0.192) 0.051a (0.017) 0.087a (0.020)
102,971 0.70
102,971 0.70
102,971 0.70
102,971 0.70
0.443a (0.156) 0.718a (0.014) 0.001c (0.001) 0.059a (0.007) 0.088c (0.048) 0.064 (0.171) 0.153 (0.234) 0.010 (0.018) 0.019 (0.024) Y Y 102,971 0.70
Gini Log(income) Age Equivalent family size Group average income Middle class Rich Middle class log(income) Rich log(income) Province dummies year dummies Age-group dummies year dummies Observations Adjusted R2
APC
(1)
102,961 0.10
Note: We control for the year, province, and age-group fixed effects, interactions between provincial dummies and year, and interactions between agegroup dummies and year in all regressions. APC is the average propensity to consume (net of education expenditures). Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
The second column shows that after we control for disposable family income, the Gini coefficient has a significant negative effect on household consumption. More specifically, when the Gini coefficient rises by 0.1, household consumption drops by 3.87%. The effect of inequality on consumption still holds when we allow income to have non-linear effects. Specifically, in column 3, we include a ‘‘middle-class’’ dummy and a ‘‘rich’’ dummy, which are the middle 1/3 and top 1/3 in the income distribution of the reference group, respectively, and in column 4 both income-group dummies and their interactions with log(income) are controlled for. The estimated effect of income inequality becomes stronger as we allow for more flexible non-linear income effects on consumption. Column 4 (our preferred specification) indicates that when the Gini coefficient increases by 0.1, consumption or the average propensity to consume declines by 4.24%. When we control for the annual time trend for each province and each age group in column 5, the effect of inequality becomes even stronger. Results are comparable when we use the average propensity to consume instead of log consumption. In the last column of Table 1, we present the estimate using the average propensity to consume (APC), defined as the ratio of consumption net of education expenditures to disposable income, as the dependent variable. When the Gini coefficient increases by 0.1, APC declines by 3.46 percentage points, which is comparable with estimates using log consumption as the dependent variable (columns 1–5).12
4.2. Potential explanations There are several potential explanations for the negative effect of income inequality on consumption net of education expenditures. The first probable explanation concerns the housing market in China. Income inequality among members of the reference group may be positively correlated with provincial housing prices, and rising housing prices may reduce household consumption and stimulate savings for housing. To test whether housing prices can account for the correlation between inequality and consumption, we augment model (1) by including in it the variable ‘‘provincial housing price index.’’ The first column of Table 2 shows that after controlling for the housing price index, the coefficient of the Gini does not change much from those reported in Table 1.13 Moreover, housing price itself does not have a significant effect on consumption. The second explanation is about the correlation between inequality and downward income risks. Generally, higher income risks result in more precautionary savings and less consumption. Therefore, rising inequality can have a negative effect 12 13
Theoretically, using log consumption or log(APC) (=log consumption log income) as dependent variable should yield the same coefficient on Gini. The housing price index is only available after 1999.
196
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Table 2 OLS regressions examining potential explanations for the association between consumption and inequality. Dependent variable: log(consumption net of education expenditures)
Gini Log(income) Log(provincial housing prices)
(1)
(2)
(3)
(4)
(5)
(6)
0.436a (0.130) 0.717a (0.014) 0.014 (0.028)
0.453a (0.152) 0.713a (0.013)
0.453a (0.125) 0.718a (0.014)
0.461a (0.138) 0.708a (0.013)
0.423a (0.125) 0.718a (0.014)
0.465a (0.154) 0.706a (0.012) 0.058 (0.035) 0.011 (0.035) 0.936b (0.462) 0.194a (0.039) 0.309a (0.070) 0.224 (0.270) 63,372 0.69
Downward income risks
0.024 (0.038) 0.778b (0.353)
Education return
0.216a (0.039) 0.269a (0.076)
Social security beneficiary rate Social security contribution rate Sex ratio Observations Adjusted R2
92,284 0.69
63,372 0.69
102,971 0.70
76,304 0.70
0.044 (0.185) 102,971 0.70
Note: In columns 2, 4, and 6, we use the data between 2002 and 2006 because the relevant information is not available before 2002. Education return is estimated based on the classical Mincer regression. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level.
on consumption due to the increase in income risks. To test this assertion, we exploit the fact that the Urban Household Survey is a rotating panel and measure the downward income risk by calculating the proportion of families in a province-age group that experienced a decrease in income (for families that are in the sample 2 years in a row).14 We find that the correlation between income inequality and downward income risk is 0.31, but only 30% of the families experienced a decrease in income, and less than 15% of the families had a decrease of more than 10%. Moreover, after we add the downward income risk variable in the baseline model (column 2), the negative effect of the Gini is even larger, indicating that the income risk is not the reason for the negative effect of income inequality on consumption. Similar outcome appears when we measure the income risk by the proportion of families that experienced a decrease in income by more than 10% in the reference group. The third explanation is related to the rate of return to education. Regions with higher inequality are also likely to have higher return to education, and people tend to invest more on education and less on other types of consumption. The third column shows estimates controlling for the education returns estimated by Mincer regressions. Again, there is little change in both the magnitude and the significance level of the effect of inequality on consumption. The return to education is positively correlated with the consumption net of educational expenditure.15 The following explanation concerns the phenomenon that an unequal society often does not have a good social safety net and that the negative correlation between inequality and consumption could be due to the precautionary-savings motive. We use two city-level measures reflecting the coverage rate of the social security system to proxy the development of the social safety net: the beneficiary rate and the contribution rate. The beneficiary rate is defined as the ratio of the number of current beneficiaries of the social security program to the total number of people over age 60 and who have retired. The contribution rate is the ratio of the number of contributors to the social security program to the number of workers under 60 and who have not retired. The second measure is only available after 2002 in our dataset. The result in column 4 shows little change in the response of consumption to income inequality. The coverage rate of social security in both measures stimulates more consumption as expected. The next plausible explanation is related to the imbalanced sex ratio in China. Wei and Zhang (2009) argue that in the marriage market, competition resulting from the rising sex-ratio imbalance accounts for about half of the actual increase in the household savings rate during the period of 1990–2007. If the sex ratio is correlated with our inequality measures, our estimation of the effect of inequality without controlling for the sex ratio will be biased. In column 5 of Table 2, we include the sex-ratio variable used by Wei and Zhang (2009), which is the sex ratio for the age cohorts 7–21. Controlling for the sex ratio does not change our main results. Sex ratio itself has no significant effect on consumption in our sample. Applying the sex ratio for the age cohorts 16–25 based on the 1990 census delivers similar results.
14 The UHS had no consistent identifier for each family until 2002. Thus, this part of the analysis focuses on the survey pertaining to the 2002–2006 period. We construct a two-year panel with 17,000 families. 15 The education return is estimated for each province in each year by a classical Mincer regression, in which we regress log wage on the year of schooling, working experience, square of experience, and gender.
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
197
In the last column of Table 2, we include all the variables discussed above. The negative effect of income inequality on consumption remains almost unchanged. Therefore, these explanations are not likely to be the main reasons for the reduction in consumption in response to the rising inequality.
5. Status-seeking and household consumption behavior In this section, we propose another potential explanation for the negative effect of income inequality on consumption net of education expenditures: Chinese save for status-seeking motives. We first discuss the literature on status-seeking and describe a potential theory. The status-seeking theory can generate some additional testable hypotheses, which will then be tested empirically.
5.1. Status-seeking savings According to Weiss and Fershtman (1998), social status is the rank of an individual or a group of individuals in a given society; the rank rests on commonly agreed-upon criteria such as wealth, education, origin, and occupation. In addition, social status is often associated with a particular group and is shared by all members of the group, regardless of their individual characteristics. Sociologists have long emphasized that individuals care about social status and that human behaviors often reflect motivations to improve personal rank in the given hierarchy, which is no less important than pecuniary rewards such as consumption.16 Early justifications focus on psychological returns of higher status. Cole et al. (1992) and Corneo and Jeanne (1999) emphasize the advantages of the high status in accessing non-market resources and show that people care about their status in equilibrium even if social status does not directly enter the utility function. These researchers highlight some social rewards for attaining higher social status. First, there is a favorable club effect. Many social activities or opportunities, such as marriage and invitation to a party or club, occur only within a group of people possessing the same social status. Second, the benefits of status improvement are related to the fact that people with high status may have privileges in the rationing of non-market goods. Furthermore, people with high status can gain trust, courtesy, and approval, and build up leadership more easily. Due to these rewards, individuals try to improve their social status through investments in assets (e.g., physical, human, and social capital), group affiliation, and an appropriate choice of actions, among which the status-seeking savings motive has been the focus in the literature (Weiss and Fershtman, 1998). Cole et al. (1992) prove that when wealth is observable, the existence of non-market decisions can naturally yield a wealth-is-status equilibrium, where social status is determined by the rank in the wealth distribution.17 The literature review by Weiss and Fershtman (1998) also states that ‘‘to the extent that the wealthy care (or are taught to care) more about status, an equilibrium emerges in which status groups are stratified by wealth’’. As a result, status seeking becomes an important motivation for wealth accumulation. People are more frugal in consumption and save more than in the case where there is no status seeking. The idea that social status depends on the ranking in the wealth distribution is widely applied in the literature on social status, particularly in the studies on how the social-status motive affects economic growth and stock market investment. For example, Bakshi and Chen (1996) study the stock-market price and emphasize that ‘‘investors accumulate wealth not only for the sake of consumption but also for wealth-induced social status. Max M. Weber refers to this desire for wealth as the spirit of capitalism’’. The literature on social-status motive and economic growth also highlights the mechanism that socialstatus seeking can stimulate more savings and thus promote economic growth (Corneo and Jeanne, 2001; Futagamia and Shibatab, 1998; Gong and Zou, 2001; Pham, 2005). Rising income inequality can intensify the status-seeking motive through several channels. First, increases in income inequality usually make entering a high-status club more attractive because differences in (financial and non-financial) resources between the high- and low-status groups widen. This corresponds to the effect of ‘‘status prize’’ in Corneo and Jeanne (1999). Second, higher inequality implies a larger wealth gap between the high- and low-status groups, which means that low-status families need to save more to join the high-status clubs.18 These channels lead to a positive effect of income inequality on household savings. However, when the dispersion of wealth becomes so large that the marginal status gain achieved by accumulating additional wealth declines or it becomes infeasible for the low-status to move up, it is possible that rising income inequality weakens the status-seeking savings incentives (Corneo and Jeanne, 2001).19 To summarize, rising income inequality on average has a positive (negative) effect on savings (consumption) unless the increase in inequality induces many families to switch from seeking status to giving up. 16
See Weiss and Fershtman (1998) for a review of the literature. They show that this equilibrium can be more stable than the aristocratic equilibrium (or, say, the status-inherited equilibrium) because the latter is not subgame perfect. Social status also depends on other indicators such as education, occupation, and relative income (Fershtman et al., 1996; Neumark and Postlewaite, 1998). 18 In addition, dispersion in income distribution may increase stratification in society and intensify the competition for social status. This raises the marginal returns of savings in social competition, in accordance with the effect of ‘‘increasing segmentation’’ in Corneo and Jeanne (1999). 19 Corneo and Jeanne (2001) show that the dispersion of initial wealth distribution reduces status-seeking savings. 17
198
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Table 3 OLS regressions estimating the effect of inequality on consumption across income groups. Dependent variable: log(consumption net of education expenditures)
Gini Log(income) Log(income) Gini
Whole sample (1)
Whole sample (2)
Poor (3)
Middle class (4)
Rich (5)
3.431b (1.674) 0.626a (0.050) 0.299c (0.164)
0.713a (0.174) 0.713a (0.014)
0.557a (0.200) 0.727a (0.017)
0.465a (0.164) 0.723a (0.032)
0.166 (0.211) 0.710a (0.015)
0.322b (0.150) 0.508b (0.213) 0.112 (0.166) 0.212 (0.221) 102,971 0.70
33,598 0.64
34,278 0.55
35,095 0.55
Middle class Gini Rich Gini Middle class Rich Observations Adjusted R2
0.037 (0.163) 0.082 (0.217) 102,971 0.70
Note: The poor, middle, and rich are the top, middle, and bottom one-third of the income distribution, respectively. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
The theory of status seeking will lead to some further testable hypotheses, including the heterogeneous effect of inequality with income and age and the effect of inequality on other instruments that can help improve social status, on conspicuous consumption, and on subsistence consumption. We will explore and empirically test these hypotheses next. 5.2. Heterogeneous effects across income groups The status-seeking motives imply that the effect of income inequality on consumption net of education expenditures can vary across income groups. To illustrate this, suppose there are three income-clubs: the rich, the middle-class, and the poor. The rich have no incentive to save for the purpose of joining a club because their wealth is already high enough to be in the highest club. In contrast, the poor and the middle class have the incentive to accumulate wealth to climb up to the next social ladder (Long and Shimomura, 2004).20 To examine how the effect of inequality varies with income, we estimate the following model:
lnðCÞ ¼ a þ b lnðYÞ þ c1 Gini þ c2 Gini lnðYÞ þ d X þ e:
ð2Þ
Results reported in the first column in Table 3 show that the negative effect of inequality on consumption indeed declines with income. When income increases by 1%, the effect of the Gini coefficient on consumption declines by about 0.3 percentage points. Next, we allow the effect of the Gini coefficient to vary nonlinearly with income. Particularly, we again consider three income groups: the poor for the bottom one-third of the income distribution within the reference group, the middle class for the middle one-third, and the rich for the top one-third, and estimate the following model:
lnðCÞ ¼ a þ b lnðYÞ þ c1 Gini þ c2 middle þ c3 Gini middle þ c4 rich þ c5 Gini rich þ d X þ e:
ð3Þ
The variable ‘‘middle’’ is a dummy variable that equals 1 if the family belongs to the middle class, ‘‘rich’’ is a dummy variable that equals 1 if the family is rich, and the (omitted) reference group is the poor. The second column of Table 3 shows that the coefficient of the interaction between the middle class and Gini is significantly positive. For this group, the effect of inequality is 0.391, which is much smaller in magnitude than that for the poor (0.713). The interaction term of rich Gini is also positive and significantly different from zero, with an estimated coefficient of 0.508, suggesting that the effect of inequality for the rich is even smaller (only 0.205). Regressions based on the three sub-samples (i.e., the poor, the middle class, and the rich) exhibit a similar pattern (columns 3–5), revealing that the effect of inequality on consumption for the rich is not significantly different from zero. The status-seeking theory implies the possibility of giving up: when inequality is too large, the poor may not be able to save enough to enter the next club, and thus they simply give up. Our empirical findings indicate that the poor in urban China do not give up. One possible reason is that poor households in urban China may not be poor enough to give up, because most of the poor in China live in rural areas, which are not in our sample. 20 When the size of the highest-status club is fixed, even the richest families have the incentive to accumulate wealth to defend their status in the ‘‘rat-race.’’ However, their incentives are still weaker than those attributable to families who need to save more both to upgrade and to prevent downgrading at the same time.
199
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204 Table 4 OLS regressions estimating the effect of inequality on consumption across age groups. Dependent variable: log(consumption net of education expenditures)
Gini Log(income) Age Gini
Whole sample (1)
Whole sample (2)
Young (head age<55) (3)
Old (head age P 55) (4)
1.072b (0.421) 0.846a (0.027) 0.011 (0.008)
0.291 (0.183) 0.682a (0.020)
0.652a (0.126) 0.714a (0.016)
0.317c (0.178) 0.721a (0.024)
0.370c (0.202) 0.375a (0.126) 102,971 0.70
78,493 0.71
24,478 0.66
Young Gini Young dummy Observations Adjusted R2
102,971 0.70
Note: Young means household head is younger than 55, and old means household head is 55 or older. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), the age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
5.3. Young versus old The status-seeking motive is also likely to vary with age. The benefit of attaining better status obviously lasts longer for the young than for the old. Therefore, when rising income inequality intensifies social rivalry, younger households tend to restrain more consumption to compete with their peers.21 Regression results reported in Table 4 show that the effect of inequality on consumption net of education expenditures declines with age. In the first column of Table 4, we allow the effect of inequality to vary linearly with age by including an interaction term of age Gini. Although the positive sign of this interaction term indicates that the negative effect of inequality becomes weaker as the head of the family becomes older, it is not statistically significant. In the second column, we allow the effect of inequality to vary with age nonlinearly, and we find that the effect of inequality on consumption is stronger for the young (household head younger than 55) than for the old (55 or older), and this difference between the young and old is statistically significant. Regressions based on the two sub-samples (young versus old) show a similar pattern (columns 3 and 4): for the young, income inequality has a negative effect on consumption, whereas the effect is much weaker for the old. 5.4. Status-seeking education investment Aside from accumulating wealth through savings, another way of improving status is investing in education. Education is considered not only an indicator of social status itself but is also correlated with higher income and wealth. Thus, we hypothesize that rising income inequality may encourage households to increase education investment.22 We consider only families with children between age 3 and 30 because there is generally very little education investment for adults. Results reported in Table 5 show that rising income inequality has a positive effect on education investment. We have different model specifications in columns 1–4, and all show that the Gini coefficient has a strong positive effect on household education expenditures. Our preferred specification, which is column 1 that allows the effect of income to be flexible, suggests that when the Gini coefficient goes up by 0.1, education expenditures increase by 50.6%. Unlike the results for consumption, the effect of income inequality on education expenditures does not differ significantly across different income groups (columns 5). One explanation for this result is that the human capital of the parents is more difficult to pass onto the next generation than traditional forms of wealth; thus, the rich also have a strong incentive to invest in children’s education when inequality increases. 5.5. Status-showing consumption People may try to show their status in the form of conspicuous consumption, that is, consumption of particularly visible and expensive, although not necessarily very useful, goods. In this case, higher inequality means more (conspicuous) consumption and less savings. However, as conspicuous consumption cannot allow one to enter a social club whose member21
Consumption of the old also responds to inequality because of altruism or the bequest motive. The same argument does not fit the health expenditures, as basic health is a necessity for people. Empirical results (not reported) show that inequality has no effect on health consumption. 22
200
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Table 5 OLS regressions estimating the effect of inequality on households’ education expenditures. Dependent variable: log(education expenditures)
Gini Log(income) Middle class Rich Middle class log(income) Rich log(income)
(1)
(2)
(3)
(4)
(5)
5.055c (2.578) 0.979a (0.067) 0.095 (0.851) 2.646b (1.000) 0.010 (0.087) 0.254b (0.098)
4.881c (2.574) 0.909a (0.044)
4.831c (2.588) 0.859a (0.067) 0.068 (0.042) 0.075 (0.075)
7.656b (3.239) 0.978a (0.065) 0.013 (0.865) 2.613b (0.989) 0.001 (0.088) 0.250b (0.097)
10.572 (7.798)
10.531 (7.798)
10.542 (7.803)
5.297c (2.673) 0.984a (0.067) 0.145 (0.851) 2.566b (0.996) 0.023 (0.092) 0.217b (0.086) 0.282 (0.803) 1.072 (1.154) 10.570 (7.794)
Middle class Gini Rich Gini Education return Province dummies year dummies Age-group dummies year dummies Observations Adjusted R2
73,951 0.22
73,951 0.22
73,951 0.22
Y Y 73,951 0.23
73,951 0.22
Note: The poor, middle, and rich are the top, middle, and bottom one-third of the income distribution, respectively. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
Table 6 OLS regressions estimating the effect of inequality on household conspicuous consumption. Dependent variable: log(conspicuous consumption) First measure
Gini Log(income)
(2)
(3)
(4)
0.780b (0.376) 1.328a (0.065)
0.030 (0.464) 1.293a (0.071)
1.012b (0.400) 1.354a (0.065)
1.250c (0.649) 2.404a (0.760)
2.536a (0.622) 1.300a (0.054) 2.054a (0.491) 2.982a (0.620) 0.951c (0.531) 2.047a (0.546)
1.186c (0.634) 2.101a (0.729)
102,971 0.44
102,971 0.45
1.260b (0.624) 2.375a (0.721) 1.217c (0.636) 0.427 (0.487) 102,971 0.45
Middle class Gini Rich Gini Middle class Rich
Second measure
(1)
Young Gini Young Dummy Observations Adjusted R2
102,971 0.46
Note: The first measure of conspicuous consumption includes expenditure on jewelry, watches, and appearance; the second one includes additional car expenditures. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
201
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204 Table 7 OLS regressions estimating the effect of inequality on household subsistence consumption. Dependent variables:
Log(cereals expenditure) (1)
Log(food expenditure) (2)
Log(cereals expenditure) (3)
Log(cereals expenditure) (4)
Gini
0.007 (0.367) 0.033c (0.017)
0.137 (0.150) 0.502a (0.013)
0.062 (0.260) 0.081a (0.020)
0.164 (0.164) 0.372 (0.244)
0.377b (0.149) 0.660a (0.232)
0.257 (0.393) 0.003 (0.016) 0.266 (0.200) 0.408 (0.271) 0.083 (0.063) 0.178b (0.087)
76,413 0.17
103,086 0.59
76,413 0.17
Log(income) Middle class Gini Rich Gini Middle class Rich Young Gini Young Dummy Observations Adjusted R2
0.139 (0.161) 0.363 (0.270) 0.203 (0.435) 0.055 (0.218) 76,413 0.17
Note: Food expenditure includes cereals, cakes, fish, beverages, and restaurant orders. We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
ship is determined by the level of wealth, poor people may not have the incentive to do so (Corneo and Jeanne, 1998). Ultimately, whether conspicuous consumption increases with inequality remains an empirical question. In this paper, we have two measures for conspicuous consumption: the first measure includes expenditure on jewelry, watches, and appearance (e.g., clothes, hairdressing, cosmetics, beauty, and bath), and the second measure includes additional automobile expenditures. Regression results reported in Table 6 suggest that conspicuous consumption is also negatively affected by income inequality. The magnitude is actually stronger than that for other types of consumption. The variations of the effect of inequality across income and age groups are also similar to those for the overall consumption. This is consistent with our expectation: poorer and younger individuals are less likely to engage in such conspicuous expenditure because doing so cannot really gain them access to the wealthy club. It actually lowers their chances of getting into the wealthy club.23 5.6. Subsistence consumption We argue that income inequality has a negative effect on household consumption because of the status-seeking motive underlying the act of saving, but this motive should not affect the basic or subsistence consumption for human needs. If we find a similar effect of income inequality on basic consumption, then the status-seeking theory is not well grounded. As expected, regression results reported in the first column of Table 7 show that income inequality has no effect on subsistence consumption (consumption of cereal). When we switch to non-basic food consumption as a dependent variable, including cakes, fish, beverages, and restaurant orders (column 2), the negative effect of inequality on food consumption becomes stronger although still not significant. Columns 3 and 4 show that inequality has no effect on consumption for all income and age groups. In summary, our empirical findings align well with the prediction based on the status-seeking story, and other potential theories cannot explain the negative effect of inequality on consumption. In results not reported, we find that the variations of the effects of inequality across income and age groups are robust after we control for the provincial housing price, downward income risks, education returns, social security coverage, and/or sex imbalance. The results for education investment, conspicuous consumption, and basic consumption also survive. 6. Robustness tests In this section, we check the robustness of the results using other measures of inequality, including the 90 percentile/10 percentile income ratio, the 75 percentile/25 percentile ratio, and the Theil index, which are more responsive to the distribution at the tails. Results reported in Table 8 show a similar pattern of the effect of inequality on consumption net of education expenditures. For example, when the 90/10 (75/25) income ratio increases by 1, non-educational consumption 23 This is consistent with the findings in other studies. For example, Wei and Zhang (2009) suggest that for marriage, most people believe wealth accumulation is more likely to provide long-term security than conspicuous consumption does.
202
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
Table 8 OLS regressions estimating the effect of inequality on household consumption: different measures of income inequality. Dependent variable: log(consumption net of education expenditures)
Inequality Log(income)
90/10 percentile income ratio
75/25 percentile income ratio
Theil index
90 income percentile–10 income percentile
(1)
(2)
(3)
(4)
(5)
(6)
0.016a (0.005) 0.718a (0.014)
0.006 (0.007) 0.682a (0.020)
0.039a (0.013) 0.719a (0.014)
0.354a (0.090) 0.718a (0.014)
0.053a (0.014) 0.719a (0.014)
0.068 (0.171) 0.153 (0.233)
0.031a (0.008) 0.712a (0.014) 0.016b (0.008) 0.026b (0.010) 0.096 (0.161) 0.179 (0.215)
0.066 (0.171) 0.149 (0.232)
0.062 (0.171) 0.150 (0.233)
0.067 (0.171) 0.153 (0.233)
102,971 0.70
102,971 0.70
0.089 (0.170) 0.167 (0.223) 0.022a (0.008) 0.400a (0.120) 102,971 0.70
102,971 0.70
102,971 0.70
102,971 0.70
Middle class Gini Rich Gini Middle class Rich Young Gini Young Dummy Observations Adjusted R2
Note: We control for the year, province, and age-group fixed effects, average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. Standard errors are robust to heteroskedasticity and clustering at the province-age group. Indicate statistical significance at the 10% level. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level.
declines by 1.6% (3.9%), and the effect is stronger for the poor and young. In the last column of Table 8, we use the absolute income difference between the 90th percentile and 10th percentile as the inequality measure; the results are similar.24 Finally, we examine the effect of inequality on total consumption, including educational expenditures. As previous results indicate that income inequality has opposite effects on education expenditures and other consumption, whether the effect on total consumption is positive or negative remains an empirical question. The first column in Table 9 shows that although income inequality has a negative coefficient, it is not statistically significant. However, it becomes significant at the 5% level when we allow each province-age group to have a different linear time trend (column 2) and the magnitude is large. Furthermore, column 3 shows that inequality has a smaller effect on the consumption of the middle class and the rich and that the differences are significant. These results are not surprising given that the education expenditures are only 8.7% of the total consumption in our sample. The last two columns show that when the Gini coefficient increases by 0.1, the APC declines by 2.72 to 2.94 percentage points.
7. Conclusions The rapid increase in income inequality in China and the substantial variation in inequality across regions and groups provide us a good opportunity to examine how the rising inequality affects consumption. Using the Chinese Urban Household Survey data, we find that income has a negative effect on household consumption net of education expenditures after we control for the income effect itself. The estimate indicates that rising inequality can explain a large part of the declining non-educational consumption as a percentage of income between 1997 and 2006. In our sample, the urban Gini coefficient rose from 0.23 in 1997 to 0.29 in 2006 on average, and the estimates imply a decline in the average propensity to consume (APC) by 2.54% or by 2.08 percentage points, which can explain 30.1% of the decline of urban household APC excluding education expenditures during this period. The declining propensity to consume with income cannot explain the remaining correlation between inequality and consumption after we control for income, thus calling for a new explanation. We propose and test several potential stories including rising housing prices, high income risks, high return to education, poor social security and rising sex ratios. However, none of these can account for the observed inequality–consumption link. Finally, we propose an account based on the status-seeking motive. We argue that people save to improve their social status when social status is associated with pecuniary and non-pecuniary benefits. Rising income inequality can strengthen the incentives of status-seeking savings by increasing the benefit of improving status, and by enlarging the wealth level required for status upgrading. The status-seeking theory also implies that the negative effect of income inequality on consumption is stronger for poorer and younger people. Moreover, rising inequality should have a strong positive effect on families’ investment in education, 24
Using the income indifference between the 75th and 25th percentile gives similar estimates.
203
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204 Table 9 OLS regressions estimating the effect of inequality on household total consumption. Dependent variables
Gini Log(income) Middle class Rich
Log(total consumption) (2)
(3)
(4)
(5)
0.139 (0.123) 0.718a (0.015) 0.157 (0.184) 0.217 (0.245)
0.280b (0.130) 0.717a (0.015) 0.150 (0.185) 0.213 (0.246)
0.390b (0.171) 0.714a (0.015) 0.199 (0.177) 0.269 (0.235) 0.294c (0.152) 0.426c (0.228)
0.272c (0.139) 0.427a (0.095) 0.544 (0.661) 1.634a (0.542)
0.294b (0.132) 0.428a (0.095) 0.541 (0.662) 1.631a (0.544)
102,971 0.69
Y 102,971 0.69
102,971 0.69
102,971 0.08
Y 102,971 0.08
Middle class Gini Rich Gini Province-age-group dummies year Observations Adjusted R2
APC (including education expenditures)
(1)
Note: We control for the year, province, and age-group fixed effects, the average income of the reference group, two income-group dummies and their interactions with log(income), age of household head, and equivalent family size in all regressions. APC1 is the average propensity to consume (including education). Standard errors are robust to heteroskedasticity and clustering at the province-age group. a Indicate statistical significance at the 1% level. b Indicate statistical significance at the 5% level. c Indicate statistical significance at the 10% level.
which is an important indicator of social status. Additional empirical analyses confirm both hypotheses. Additionally, we find that inequality has a negative effect on conspicuous consumption, and confirm that it has no impact on basic (or subsistence) consumption. Finally, our results are robust to different measures of inequality. Acknowledgments We would like to thank Daniel Berkowitz, two anonymous referees, Chongen Bai, conference and seminar participants for their valuable suggestions. Any remaining errors are ours alone. Hongbin Li acknowledges supports from the National Natural Science Foundation of China (Project ID: 71025004). Binzhen Wu acknowledges supports from the National Natural Science Foundation of China (Project ID: 70903042). We also acknowledge supports from the China Data Center of Tsinghua University and Xiaolong Chen for data supports.
Appendix A Table A1. Table A1 Descriptive statistics of the variables.
Gini coefficient
Disposable income Consumption (excluding education expenditures) APC (excluding education expenditures) APC (including education expenditures) Education expenditures Education (years) Family equivalent scale Family size Age
Year
Observations
Mean
Standard deviation
1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Total
5392 5382 5378 5355 5378 12,878 14,379 15,593 16,677 16,822 103,234
0.232 0.243 0.244 0.270 0.271 0.274 0.286 0.289 0.288 0.289 0.277
0.033 0.033 0.027 0.034 0.031 0.035 0.041 0.038 0.037 0.039 0.041
104,665 104,550 104,541 104,551 104,665 104,665 104,665 104,665 104,665
28,178 21,858 0.77 0.84 1629 11.58 2.28 3.00 48.31
21,101 17,099 0.31 0.55 3094 2.99 0.49 0.77 11.32
204
Y. Jin et al. / Journal of Comparative Economics 39 (2011) 191–204
References Alesina, Alberto F., Perotti, Roberto, 1996. Income distribution, political instability, and investment. European Economic Review 40, 1203–1228. Atkinson, Anthony B., Rainwater, Lee, Smeeding, Timothy M., 1995. Income Distribution in OECD Countries. OECD Social Policy Studies, No. 18, Paris. Bakshi, Gurdip S., Chen, Zhiwu, 1996. The spirit of capitalism and stock-market prices. American Economic Review 86, 133–157. Clark, Andrew E., Oswald, Andrew J., 1996. Satisfaction and comparison income. Journal of Public Economics 61, 359–381. Chamon, Marcos D., Prasad, Eswar S., 2010. Why are saving rates of urban households in China rising? American Economic Journal: Macroeconomics, American Economic Association 2 (1), 93–130. Cole, Harold L., Mailath, George J., Postlewaite, Andrew, 1992. Social norms, savings behavior, and growth. The Journal of Political Economy 100, 1092–1125. Coleman, James S., 1990. Foundations of Social Theory. Harvard University Press, Cambridge, MA. Corneo, Giacomo, Jeanne, Olivier, 1998. Social organization, status and savings behavior. Journal of Public Economics 70, 37–51. Corneo, Giacomo, Jeanne, Olivier, 1999. Social organization in an endogenous growth model. International Economic Review 70, 37–51. Corneo, Giacomo, Jeanne, Olivier, 2001. Status, the distribution of wealth, and growth. The Scandinavian Journal of Economics 103, 283–293. Dynan, Karen E., Skinner, Jonathan, Zeldes, Stephen P., 2004. Do the rich save more? Journal of Political Economy 112, 397–444. Fershtman, Chaim, Murphy, Kevin M., Weiss, Yoram, 1996. Social status, education, and growth. The Journal of Political Economy 104, 108–132. Futagamia, Koichi, Shibatab, Akihisa, 1998. Keeping one step ahead of the joneses: status, the distribution of wealth, and long run growth. Journal of Economic Behavior & Organization 36, 109–126. Gong, Liutang, Zou, Hengfu, 2001. Money, social status, and capital accumulation in a cash-in-advance model. Journal of Money Credit and Banking 33, 284– 293. Horioka, Charles Y., Junmin, Wan, 2007. The determinants of household saving in china: a dynamic panel analysis of provincial data. Journal of Money, Credit, and Banking. Kraay, Aart, 2000. Household saving in china. World Bank Economic Review 14, 545–557. Kuijs, Louis, 2006. How will China’s saving-investment balance evolve? World Bank Policy Research Working Paper #3958. Li, Hongbin, Zhu, Yi, 2006. Income, income inequality, and health: evidence from China. Journal of Comparative Economics 34, 668–693. Long, Ngo V., Shimomura, Koji, 2004. Relative wealth, status-seeking, and catching-up. Journal of Economic Behavior & Organization 53, 529–542. Menchik, Paul L., David, Martin, 1983. Income distribution, lifetime savings, and bequests. The American Economic Review 73, 672–690. Meng, Xin, 2003. Unemployment, consumption smoothing, and precautionary saving in urban China. Journal of Comparative Economics 31 (3), 465–485. Modigliani, Franco, Cao, Shi L., 2004. The Chinese saving puzzle and the life-cycle hypothesis. Journal of Economic Literature 42, 145–170. Musgrove, Philip, 1980. Income distribution and the aggregate consumption function. The Journal of Political Economy 88, 504–525. Neumark, David, Postlewaite, Andrew, 1998. Relative income concerns and the rise in married women’s employment. Journal of Public Economics 70, 157– 183. Pham, Thi Kim Cuong, 2005. Economic growth and status-seeking through personal wealth. European Journal of Political Economy 21, 407–427. Schmidt-Hebbel, Klaus, Serven, Luis, 2000. Does income inequality raise aggregate saving? Journal of Development Economics 61, 417–446. Smith, Douglas, 2001. International evidence on how income inequality and credit market imperfections affect private saving rates. Journal of Development Economics 64, 103–127. Stoker, Thomas M., 1986. Simple tests of distributional effects on macroeconomic equations. The Journal of Political Economy 94, 763–795. Wei, Shangjin, Zhang, Xiaobo, 2009. The Competitive Saving Motive: Evidence from Rising Sex Ratios and Savings Rates in China. NBER Working Paper No. w15093. Weiss, Yoram, Fershtman, Chaim, 1998. social status and economic performance: a survey. European Economic Review 42, 801–820. Wen, Yi., 2009. Saving and Growth under Borrowing Constraints Explaining the ‘High Saving Rate’ Puzzle. Working Papers 2009-045, Federal Reserve Bank of St. Louis. World Bank, 2005. World Development Report 2006: Equity and Development. Washington, DC: World Bank.