Journal of Development Economics 111 (2014) 34–47

Contents lists available at ScienceDirect

Journal of Development Economics

j ourna l homepage: www.e lsev ie r .com/ locate /devec

Brothers, household financial markets and savings rate in China☆

Weina Zhou ⁎Department of Economics, Dalhousie University, 6214 University Avenue, Halifax, NS B3H 4R2, Canada

☆ I thank Thomas Lemieux and Siwan Anderson for thguidance. I am also grateful to co-editor Gerard Padro i MHiroyuki Kasahara, Xin Meng, Giovanni Gallipoli, Paul Btheir detailed suggestions.⁎ Corresponding author.

E-mail address: weina.zhou@dal.ca.1 The savings rate is defined as 1 − LivingExpenditure/

China Statistical Year Book.

http://dx.doi.org/10.1016/j.jdeveco.2014.07.0020304-3878/© 2014 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 12 January 2014Received in revised form 1 July 2014Accepted 17 July 2014Available online 24 July 2014

Keywords:Household savingsFamily structureFinancial market development

This study analyzes the effect of the number of brothers an individual has on that individual's household savingsrate under the current underdeveloped household financial market in urban China. I show that having anadditional brother reduces an individual's household savings rate by at least 5 percentage points. Brothers helphouseholds (1) by sharing risks, providing a source of informal borrowing, and (2) by sharing the cost ofsupporting parents. Sisters play a minor role in affecting a household's savings rate, mainly because of culturalnorms. The decline in the average number of brothers in households induced by population policies explainedat least one-third of the increased aggregate household savings rate in urban China.

© 2014 Elsevier B.V. All rights reserved.

1. Introduction

It is well documented that the corporate financial markets in Chinaremain underdeveloped despite China's impressive GDP growth inrecent decades (Allen et al., 2005; Ayyagari et al., 2010; Chen et al.,2011; Guariglia et al., 2011; Song et al., 2011). Private entrepreneursusually find it difficult to borrow from banks, relying largely insteadon a network of family members or relatives to serve as financialresources (Cai et al., 2013; Estrin and Prevezer, 2011). To date, however,little attention has been paid to household financial markets thoughthey are no less underdeveloped than the corporate financial markets.According to the 2009 China Family Panel Study, even in China's mostdeveloped regions, Beijing, Shanghai, and Guangdong, more than 80%of debtors in 2008 borrowed from family members or relatives whilefewer than 20% borrowed from financial institutions. Households aresuffering from underdeveloped financial markets; at the same time,they also encounter a large uncertainty. Health care reforms, pensionreforms, and rising income uncertainties impel households towardhigh savings rates (Chamon and Prasad, 2010; Chamon et al., 2010).The household savings rate rose from 16% in 1990 to 24% in 2005 inurban areas.1

eir advice, encouragement andiquel, two anonymous referees,eaudry and Ashok Kotwal for

DisposableIncome. Data source:

In developing countries where household financial markets areunderdeveloped, research has provided evidence as to how extendedfamily members help households experiencing negative economicshocks by sending them transfers and gifts (Fafchamps, 2011;Fafchamps and Quisumbing, 2011). But the extent to which theexistence of family members who can represent potential transfersaffects a household's savings rate and whether the gender of thefamily members matters has so far remained unknown.

This paper explores the consequences of a weak householdfinancial market by studying the effect of brothers, the most impor-tant members of a household in the extended family, on the house-hold savings rate in urban China. This is one of the first papers toestimate the sibling effect on a household's savings rate using micro-level data.

Although individuals rely largely on their brothers under the currentenvironment of increasing uncertainties and incomplete financialmarkets, population control policies such as the One-Child Policy(1979) made the situation even worse. In contrast to the generationborn during the baby boom period (1946–1978), where an individualhas on average more than three siblings, individuals in the One-ChildPolicy generation have fewer or even no siblings.2 In addition to suffer-ing from incomplete financial markets, the One-Child Policy generationalso suffers from the lack of a family-based safety net. A simple calcula-tion suggests that the decline in the average number of brothers can ex-plain at least one-third of the increased aggregate household savingsrate.

2 Although the overall fertility rate was high during the baby boom period, it was lowduring the Chinese famine period (1959–1961).

35W. Zhou / Journal of Development Economics 111 (2014) 34–47

In estimating the effect of the number of brothers an individual hason that individual's household savings rate, endogeneity problemsarise: the number of brothers an individual has could potentially becorrelated with that individual's unobserved characteristics such asparents' preferred number of children. This paper has found thatconditional on an individual's number of siblings, the gender of thesiblings can be considered as a random assignment by nature forurban residents born during the baby boom (1946–1978). The genderassignments of siblings by nature help us identify the effect of havinga brother instead of a sister (a relative effect).

The identification strategy relies on the assumption that condi-tional on the number of siblings, the gender of the siblings is onlydetermined by nature. It is well known that in recent decadesChina has had a growing missing female problem (Anderson andRay, 2010; Qian, 2008). However, I find it unlikely that, amongurban residents born during the baby boom (1946–1978), parentswould have been able to control the gender of their children for agiven family size.3 The main reason is that ultrasound technologythat can identify gender before birth was only introduced in the1980s, after the baby boom. In addition, female infanticide wouldhave been difficult to carry out in urban areas, rendering it unlikelythat urban households would have risked criminal prosecution forson preference.

I find that having a brother instead of a sister reduces thathousehold's savings rate by at least 5 percentage points. If, likebrothers, sisters should also affect the household savings rate, the es-timated relative effect would be a lower bound. That is, the absoluteeffect (having one more brother rather than not) would be largerthan the relative effect (having a brother instead of a sister). The sta-tistical evidence for the baby boom generation reveals that sistershave almost no effect on a household's savings rate. Therefore, the es-timated relative effect of a brother is likely to be the same as the abso-lute effect. The lack of a sisters effect on the savings rate may resultfrom the relatively weak connections in Chinese culture between fe-male and male siblings and between parents and daughters. Thissaid, interestingly, as the number of siblings declines among theyoung generation, given the change in family planning policy, sisterscome to affect the savings rate in the way that brothers do. Wherethere are too few brothers, young households may use sisters as asubstitute for brothers.

I show that brothers can reduce a household's savings rate throughtwo channels: (1) sharing risks and extending borrowing limits, and(2) sharing the cost of supporting parents. In order to examine the effectof risk sharing/extending borrowing limits, this paper tests the effectthat brothers have on households with different levels of (a) wageuncertainties, (b) health risks, (c) regional financial development, and(d) income or asset levels. The estimation results are consistent withthe risk-sharing/extending-borrowing-limits hypothesis: householdsthat encounter larger wage uncertainties, have higher health risks, livein a financially less developed province, have lower incomes or havefewer assets, have a larger brothers effect. The robust and consistentresults suggest a strong risk-sharing/extending-borrowing-limits effectof having brothers.

In Chinese culture, the expectation is that male childrenwill supporttheir parents (Banerjee et al., 2013).4 A householdwith several brotherswould need to save less for parents' risks, in particular for medicalexpenditure risks, as these are largely shared among the brothers. Totest the parent-supporting aspect, I utilize information on whether

3 The baby boom was induced by family planning policies introduced in the 1950s andcarried on until the early 1970s.

4 This is the main reason why we observe a large increase in the male–female genderratio of newborns after the One-Child Policy.

parents are deceased. Once the parents have passed away, brothers nolonger play a role in sharing parents' risks. The difference in the numberof parents still living helps to identify the parent-supporting effect ofbrothers.

Recent papers have emphasized that changes in the demographicstructure could affect household savings rates given the effect ofthe intergenerational support. Ge et al. (2012) explore the regionalvariation in One-Child Policy fines to examine the effect of changingdemographics on household savings rates. Choukhmane et al. (2013)estimate an OLG model incorporating endogenous fertility, intergener-ational transfers, and human capital accumulation; they find thatchanges in the demographics explain more than one-third of the risein the aggregate savings rate. Banerjee et al. (2013) suggest thatthe partial equilibrium model could overstate the effect of changingdemographics on the savings rate. Wei and Zhang (2011) suggestthat the rising gender ratio induced parents to save more for theirmale children, helping them secure a better outcome in the marriagemarket.

This is one of the first papers to emphasize that in addition to theintergenerational support effect, the risk-sharing effect among brotherscould also explain why changes in demographics could raise the aggre-gate household savings rate. Furthermore, the role of risk sharing/extending the borrowing limits among family members could varygreatly depending on the gender of a family member. This paperdiscovers a new dimension to gender difference in China.

In further comparison to existing literature in which most papersexamine increases in savings at a particular point in time, such as reduc-tion in theprovision of social services, housing reform, or having a singlechild, this paper also explains why in China savings rates continue torise. First, the informal risk-sharing networks continue to shrinkbecause future generations will have even fewer or no cousins giventhe One-Child Policy. Second, because households with fewer brothersrequire greater savings to support elderly parents, these householdsmay not have sufficient savings for their own retirement. As such,they may continue to consume relatively little even after the parentshave passed away.

This paper also helps explainwhymixed evidence prevails regardingwhether the decreasing dependency ratio could explain the risingsavings rate. Modigliani and Cao (2004) use long-term national-leveldata and find that the decrease in both the young and old populationcontributes to the rising savings rate in China. On the other hand,Horioka andWan (2007) usemore recent data and find that the changein the dependency ratio does not adequately explain the increasingsavings rate. By emphasizing that individuals of prime age could saveless because they havemore brothers, this paper helps solve the puzzle.The recent younger generation contributes to the high savings ratebecause it does not have siblings.

The paper proceeds as follows. Section 2 introduces the back-ground to household financial markets, population policies, and thecurrent savings rate in China. Section 3 introduces the identificationstrategies and presents the estimation results. Section 4 exploresthe reason why a household with more brothers could reduce thesavings rate. Section 5 provides a robustness check for the identifica-tion strategy. Section 6 shows how much of the savings rate puzzlecan be explained by the brothers effect. Section 7 concludes thepaper. The Data Appendix provides information on all the data usedin this paper.

2. Background

2.1. Financial markets and household borrowing resources

It is a well-known fact that the corporate financial market in Chinais underdeveloped; private entrepreneurs have to rely largely onsuch networks as family members or relatives for financial resources(Allen et al., 2005; Ayyagari et al., 2010; Chen et al., 2011; Guariglia

Fig. 1. Sources for borrowing money in urban China: Self-reports of borrowing resource ifone encounters a negative shock (percentage of respondents). Note: The above results arecalculated by the author based on a question in the Chinese Household Income Project2002 Urban Sample: “If your household encountered an abrupt difficulty and needed10,000 RMB immediately, who (where) would you turn to first?” Sample size: 6790.

36 W. Zhou / Journal of Development Economics 111 (2014) 34–47

et al., 2011; Song et al., 2011). To date, little attention has been paid tothe household financial market even though this market is no less un-derdeveloped than the corporate financial market (Coeurdacier et al.,2013; Yao et al., 2011).

Despite the fact that the real interest rate on domestic bank de-posits has often been negative (Gordon and Li, 2003; Lardy, 2012),by using the China Household Finance Survey 2011, Gan (2012) sug-gests that the two main financial assets for households are bankdeposits (58%), and cash holdings (18%). The rate of consumer loansissued by all financial institutions in China was nearly zero in 1997(Chamon and Prasad, 2010). And though it reached 2.2 trillion RMBat the end of 2005, mortgage loans amounted to about 80% of thetotal loans.5

Households can also encounter significant uncertainties. Medicalreforms, pension reforms, and rising income uncertainties inducehouseholds to save more because of the precautionary motive(Chamon and Prasad, 2010; Chamon et al., 2010). How do house-holds in China finance themselves when they encounter negativeshocks?

I use two different data sets to investigate how Chinese householdsfinance themselves in the current underdeveloped household financialmarkets. The first data set comes from the Chinese Household IncomeProject (CHIP, see Data Appendix A). The CHIP 2002 urban area surveyasked, “If your household suddenly encountered difficulty and needed10,000 RMB immediately, where or to whom would you turn first?”6

I report the results in Fig. 1. More than 60% of the individuals chose“family members and relatives,”while fewer than 3% of the individualschose “financial institutions.” It is very clear that family membersand relatives are a household's primary borrowing source. The resultsalso suggest that the potential transfer or quasi-credit amount availableamong family members could also be very large. Note that 10,000 RMBis approximately 1600 USD, which is more than half of the medianhousehold's yearly income in the 2002 CHIP data.

The China Family Panel Study (CFPS) 2009 asked households if theyactually borrowed money in 2008 and if so, the sources they borrowedfrom and the reason for borrowing. In total, 14% of the surveyrespondents had borrowedmoney in 2008. Aswas the case in the reportusing the 2002 CHIP data, “Relatives and friends” was overwhelminglythe dominant borrowing resource for households. Conditional onborrowing money, 82.3% of the households borrowed from relativesand friends, while fewer than 20% of the borrowers borrowed frombanks.7 Note that this survey was conducted in Beijing, Shanghai, andGuangdong provinces China's most financially developed areas. Inother less developed areas, the proportion of households relying onfamily members could potentially be even larger.

It is worth noting that the housing loanmarket is quite developed inChina, perhaps because of the government's enforcement of housingreforms, which encourages individuals to buy houses. As the primaryreason for people borrowing money from banks is housing, andmortgages are not considered to be an unexpected expense, relativesand friends become the only source of borrowing when a householdencounters unexpected shocks.

5 The other major loan categories were auto loans and large durable goods loans.6 Therewere nine answers to choose from: (1) familymembers and relatives, (2) friend,

(3) other individuals, (4) work unit, (5) bank and credit union, (6) other financial institu-tions, (7) need no help, (8) anywhere I can borrow, and (9) other. I aggregated (5) and(6) together under the category “financial institutions,” and I aggregated (3) (4) (8) and(9) together as “other.”

7 Households had the following options to choose from in the survey: (1) banks (includ-ing credit unions), (2) relatives and/or friends, (3) loan from a private institution, and(4) other. Only 2% of households took recourse to a private institution or other (3) or (4).

2.2. Facts: household savings rate by number of brothers and sisters

I use the China General Social Survey (CGSS) 2006 to constructhousehold savings rate data. The CGSS 2006 data contain the totalincome, basic living expenditure, medical expenditure, and educationexpenditure information of individuals' households. Savings are calcu-lated as the household total disposable income minus the sum ofthese three household expenditures. The savings rate is defined assavings divided by household total disposable income.8 AppendixTable 1 shows detailed descriptive statistics of disposable incomes andexpenditures. The average savings rate in 2006 was 26% for urbanresidents, which is the main sample used in this paper. It is only 1percentage point higher than the 25% savings rate computed byusing the data in China Statistical Year book for urban households in2006.9

Fig. 2 presents the age profile of the household savings rate byindividuals' number of brothers and sisters. In the upper panel ofFig. 2, I divide the individuals into two groups: individuals with zeroor one brother, and individuals with two or more brothers. The figureclearly shows that individuals with zero or one brother have a highersavings rate than individuals with two or more brothers, for all agegroups. There is a strong negative correlation between the number ofbrothers and the household savings rate.

By contrast, the savings rate is quite similar regardless of anindividual's number of sisters (the lower panel of Fig. 2), in particularif that individual is over 35 years of age. It is interesting to note, howev-er, that the pattern of the savings rate by the number of sisters forthe young generations is close to that of the number of brothers: anindividual with fewer sisters is associated with a higher household sav-ings rate. As the number of siblings declines with the change in familyplanning policy, young households may use sisters as a substitute forbrothers where there is a shortage of brothers. Sisters might play thesame role as brothers and affect the household savings rate.

8 I compute income tax based on the Individual Income Tax Lawof the People's Republicof China introduced in 2005. In an earlier version of this paper, I used income instead ofdisposable income. The estimation results using disposable income are almost identicalto the results using (non-tax-deducted) income.

9 When we compute the savings rate by using the data in China Statistical Year Book,the household savings rate is defined as 1 − Expenditure/Income, where expenditure isper capita household living expenditure, and income is per capita household disposableincome.

Fig. 2. Age profile household savings rate by number of brothers and sisters. Data source:China General Social Survey 2006. Sample size: 6886.

Fig. 3. Age profile household savings rate by number of brothers and sisters— householdswith no living parents. Data source: China General Social Survey 2006. Sample size: 1732.

37W. Zhou / Journal of Development Economics 111 (2014) 34–47

To avoid the potential concerns of the siblings-supporting-parentseffect, Fig. 3 repeats the same exercise by using individuals withdeceased parents. The figure only presents savings rates for individualsover 40, because there are very few individuals below this age whoseparents are deceased. The figure suggests that even for individualswith no living parents, the number of brothers still has a strongnegativecorrelation with the household savings rate. For the number of sisters,the correlation with the savings rate is not clear (the lower panel ofFig. 3).

Fig. 4. Number of brothers and sisters by individuals' birth year. Data source: ChinaGeneral Social Survey 2006. Sample of urban area residents are used. Sample size: 3235.

2.3. Changes in demographics and China's savings rate puzzle

The number of an individual's siblings has changed dramatically inrecent decades. Fig. 4 presents individuals' number of brothers andsisters by those individuals' year of birth for the CGSS 2006 data. Thefigure shows that individuals born in the 1950s and 1960s have onaverage more than three siblings (with 1.5 brothers and 1.5 sisters)whereas individuals born during the 1970s and 1980s have fewer oreven no siblings.

The change in individuals' number of siblings is induced by thechange in family planning policies in China. The population policies inChina can be divided into three main stages: population expansion(1949–1972), voluntary birth control (1972–1978), and the One-ChildPolicy (1979–current).

After the founding of the People's Republic of China in 1949, theChinese government introduced policies to encourage births. ChairmanMao Zedong's famous saying “themore people, the strongerwe are” re-mains a well-known phrase in China, even among the current genera-tion. This large population growth was slowed by the second stage of

38 W. Zhou / Journal of Development Economics 111 (2014) 34–47

family planning policies implemented in 1972. During this stage, thegovernment deferred to the slogan “later, spaced, and few”: “later” forlater marriage, “spaced” for spaced birth, and “few” for fewer children.The policy emphasized birth spacing but placed no cap on the totalnumber of children; the population control policy at this stage beingvoluntary, no punishment was meted out for violations. Adoption ofbirth control methods was a decision left to the couples themselves.As a result of these policies, China's population almost doubled injust 30 years, increasing from 540 million in 1949 to 960 million in1978.

The famous One-Child Policy stage represents the third stage offamily planning policies. This policy was introduced in 1978 andapplied to babies born in 1979. Families in urban areas were allowedonly one child each. Families who violated the policy were requiredto pay monetary penalties and could be denied bonuses at theirworkplaces.

While the average number of siblings has been decreasing in re-cent decades, the household savings rate has been increasing. Fig. 5shows individuals' average number of brothers between 1980 and2005 as well as the trend in the household savings rate. The house-hold savings rate increased dramatically during this period. It pre-sents one of the largest puzzles in China's savings literature,attracting much attention among researchers and policy makerswho now earnestly ask: why has the savings rate in China increasedsubstantially in recent decades. The figure suggests that the declinein the average number of brothers may be one of the solutions tothis puzzle.10

3. The impact of the number of brothers on households' savings rate

3.1. Identification

Let us first consider the following equation:

SavingRatei ¼ αBroi þ Xiγ þ ϵi ð1Þ

The definition of the savings rate is given in Section 2.2. Broi isthe number of brothers of an individual. Xi is a set of the individual'scharacteristics and the individual's household characteristics. α, thecoefficient on Broi, is the parameter we are interested in. Broi could becorrelated with unobserved family characteristics, such as parents'economic conditions or their preferred number of children, whichmay be correlated with the individual's household savings. Thus, αcannot be consistently estimated through Eq. (1).

In order to identify the effect of brothers on the savings rate, Iconsider the following case. If individuals' parents cannot manipulatethe gender of individuals' siblings, then given the number of siblings,the gender of siblings is only determined by nature. The number ofbrothers is not correlated with any unobserved characteristics for agiven number of siblings.

If, given the number of siblings, having a brother instead of a sister israndomly assigned by nature, then the effect of having a brother insteadof a sister on the savings rate can be interpreted as a randomly assignedtreatment. α can be consistently estimated through Eq. (2). SeeAppendix B for proof. Keep in mind that the interpretation of α isdifferent in Eq. (2) from that in Eq. (1), as α in Eq. (2) represents the

10 In addition to demographics, there are many other reasons for a high savings rate; forexample, housing reform, health care reform, pension reform, and rising income uncer-tainty (see Yang et al. (2010) for a detailed survey). The large increases in the average per-sonal income could also explain the higher savings rate. In addition, the higher averageincome level coexists with high income inequality, whichmay also contribute to the risingnational savings rate as rich households usually save a larger proportion of their incomethan do poor households.

effect on the savings rate of having a brother instead of a sister, for agiven number of siblings.

SavingRatei ¼ αBroi þ δ Sibið Þ þ Xiγ þ ϵi ð2Þ

The identification strategy compares the savings rate of individualswith different numbers of brothers but with the same number ofsiblings. The upper panel of Fig. 6 presents this variation. The figuresuggests that for each sibling group, having more brothers is associatedwith a lower savings rate. As the savings rate is defined as savingsdivided by income, one may be concerned that the negative correlationbetween the number of brothers and the savings rate (conditional onthe number of siblings) could be driven by the income correlation. Thelower panel of Fig. 6 suggests this is not a concern, as there is no clearpattern of how income is correlated with the number of brothersgiven the number of siblings.

The assumption that, conditional on the number of siblings,the number of brothers is a random assignment requires that nopredetermined family characteristics affect the assignment of thegender of the siblings (the only thing that can determine the genderof the siblings is nature). Several papers in the “missing female” litera-ture indicate that Chinese households have a son preference and thatthe sex ratio of newborns became significantly distorted following theintroduction of the One-Child Policy (1979) (Arnold and Liu, 1986;Wei and Zhang, 2011), with parents wanting to ensure they had a son.The main reason for the son preference is that male children financiallysupport their aging parents. Thus, using sex-selective abortion or femaleinfanticide, a practice sometimes found in rural areas, parents “chose”the gender of their children.

The evidence I found suggests that by restricting the sample to urbanresidents and to those born before theOne-Child Policy (1979) and afterWorld War II (1945), the gender of individuals' siblings is exogenouslyassigned. In the rest of the paper, I call this sample the restricted sample.

Gender distortion in the restricted sample is absent for severalreasons. First, the ultrasound technology required for sex-selectiveabortions was only introduced in the 1980s; households before the1980s had no reliable method for performing sex-selective abortions.Second, female infanticide occurred mainly in rural areas where babieswere delivered at home. In urban areas, babieswere usually delivered inhospitals, making it unlikely that urban households would risk criminalprosecution for son preference. Keep in mind that people born close to

Fig. 5. Number of brothers and household savings rate in urban areas. Note: The numberof siblings is restricted to individuals aged20–60. Death rates are used in order to computethe number of siblings in early years. Saving rate is defined as 1-living expenditure/disposable income. Saving rate and death rate data source: China Statistical Yearbook.Siblings data source: China General Social Survey 2006.

Fig. 6. Source of variation: average household savings rate by number of brothers fora given number of siblings. Note: China General Social Survey 2006 is used. Sample isrestricted to urban area residents born between 1946 to 1978.

39W. Zhou / Journal of Development Economics 111 (2014) 34–47

1979 are unlikely to have siblings born after 1979 because of the One-Child Policy. Third, Chairman Mao largely enforced gender equality inChina before he passed away in 1976 (Li, 2000). “Women hold half ofthe sky” is his famous slogan to enforce gender equality. In urbanareas, females enjoyed as many job opportunities as males. The greaterdegree of gender equality in general made parents in urban areas lesslikely to exhibit the same degree of son preference as before.

Two sets of statistical tests examine whether the gender of childrenis exogenously assigned in the restricted sample. Table 1 reports theproportion of male siblings given the number of siblings. The naturalgender ratio is 106 males per 100 females (Jacobsen et al., 1999). Thisimplies that the natural proportion of male siblings is 51.5%. If parentspractice son preference, this proportion would be significantly greaterthan 51.5%. The statistics computed in Table 1 show that the proportionof males is close to the natural level regardless of individuals' number ofsiblings in the restricted sample.

Table 2 provides a test of the random assignment of the number ofbrothers conditional on the number of siblings. In column 1, wherethe number of siblings is not controlled for, the number of brothers is

Table 1Fraction of male siblings by total number of siblings.

Number of siblings Obs Fraction of male 95% Conf. interval

1 398 0.51 [0.48, 0.54]2 569 0.51 [0.48, 0.53]3 590 0.50 [0.48, 0.52]4 or more 850 0.49 [0.47, 0.50]

Note: China General Social Survey 2006 is used. Sample is restricted to urban arearesidents born between 1946 to 1978.

significantly correlated with the mother's years of education. TheWald test suggests that all of the family characteristics are jointly signif-icant. In contrast, once the number of siblings is controlled for in column2, no parental characteristic is significantly correlated with the numberof brothers, and the Wald test suggests that they are not jointly signifi-cant. The results in Table 2 provide strong evidence that conditional onthe number of siblings, the number of brothers is random among urbanresidents born between 1946 and 1978.

One may have concerns that parents might be practicing a sonpreference by adopting a stopping rule; that is, they continue havingchildren until they have reached the desired number of boys. This isalso unlikely to happen in the restricted sample. An easy way to seewhether parents had adopted a stopping rule is to check the gender ofthe last child. If parents had adopted a stopping rule, we would likelyobserve that the youngest child is amale. Recall that the natural propor-tion ofmales is 51.5%. For urban residents born between 1946 and 1978,the proportion of males as the youngest child of parents is 51.7% in theCGSS data and 50.4% in the CULS data (see Data Appendix A), neitherbeing significantly different from the natural proportion of males.11

One might also want to know the effect of sisters on a household'ssavings rate. Ideally, we want to include the number of sisters in theregression to estimate the impact of the number of sisters on the savingsrate. Such an estimate is not feasible, however, because of the problemof collinearity (we cannot add both the number of brothers and sistersand siblings into one regression). As we control for the number ofsiblings, α measures the difference between the effect of brothers andthat of sisters. The coefficient on the number of siblings represents theeffect of sisters with bias induced by endogeneity. See Appendix C forthe proof.

Although the true effect of sisters could not be estimated, from Fig. 2,it is more likely that sisters have no effect on a household's savings rate.If this is the case,α, the estimated relative effect of brothers compared tothat of sisters, also equals the absolute effect of brothers. If sisters wereto play a similar role to brothers in risk sharing and supporting parents,the estimated brothers effect would be a lower bound of the absoluteeffect of brothers (see Appendix C).

3.2. Results: the impact of the number of brothers on household savings rate

The estimation results of Eq. (2) are presented in Table 3. Error termsare clustered at county level. Column 1 uses nonurban-residents data.The rest of the columns use urban-residents data because of the identi-fication strategy discussed in Section 3.1. Both columns 1 and 2 controlfor the number of siblings, years of education, gender, age, age squared,household income, and marital status.

In both urban and rural areas, we observe a negative effect of thenumber of brothers on the household savings rate. The coefficient onthe number of brothers for the sample of urban residents is −0.047and statistically significant at the 1% level. This means that having onebrother instead of one sister would, on average, reduce the savingsrate by 4.7 percentage points. Interestingly, the magnitude of thebrothers effect is larger in urban areas than it is in rural areas. The esti-mation results in the first two columns may suggest that urban house-holds rely more on their brothers than do their rural counterparts.Rural areas usually have less developed financial markets and experi-ence higher risks associatedwithfluctuations in agricultural production.However, compared with urban households, rural households canusually share riskswith villagemembers in addition to their own family

11 The CGSS data does not provide the exact birth order of respondents' siblings as it onlylists the number of younger brothers and sisters and older brothers and sisters. For thisreason, I check the gender of a respondent conditional on the individual being the youn-gest child in the family. The CULS 2001 data (see Data Appendix A for details) providesthe birth order of siblings. I restrict the sample to urban residents born between 1946and 1978. The sample size of the CULS data is 5351.

Table 2Test of random assignment of the number of brothers conditional on the number ofsiblings.

Dependent variable

Brothers Brothers

Siblings 0.486⁎⁎⁎

(0.012)Mother education − .046⁎⁎⁎

(0.009)− .006(0.006)

Father education − .008(0.009)

0.006(0.006)

Mother communist party 0.048(0.137)

0.024(0.096)

Father communist party 0.031(0.066)

− .047(0.05)

Mother company type 0.046(0.142)

0.033(0.094)

Father company type − .051(0.095)

− .012(0.063)

Mother occupation skill level − .026(0.074)

− .013(0.055)

Father occupation skill level − .084(0.062)

0.009(0.045)

Mother occupation dummies Yes YesFather occupation dummies Yes YesObs. 2594 2594Wald statistics 11.02⁎⁎⁎ 1.44

Note: China General Social Survey 2006 is used. Sample is restricted to urban arearesidents born between 1946 and 1978. The Wald test examines the joint significance ofall the regressors in column 1. In column 2, number of siblings is not included in theWald test; all other regressors are included. Standard errors are clustered at countylevel. *** p b 0.01, ** p b 0.05, * p b 0.1.

40 W. Zhou / Journal of Development Economics 111 (2014) 34–47

members and relatives. The larger brothers effect in urban areas may bea result of the relative scarcity of sources for risk sharing aside familyand relatives. Keep in mind that the coefficient on brothers in therural sample may be biased because of the potential female infanticideproblem in rural areas.

Column3 adds a large set of demographic and characteristic controlsthat could potentially affect a household's savings rate: family size,parents-living-together dummy, Communist Party membership status,father's and mother's education, and a send-down dummy.12 Chamonand Prasad (2010) indicate that increases in children's educationexpenses and housing reform caused households to save more. Forthis reason, column 4 adds the number of children and children's agegroup dummies in order to control for the potential education expenseeffect. Column 5 adds households' housing characteristics: a dummyvariable indicateswhether thehousehold owns thehouse, themortgagevalue (if the house is owned), and the value of the house owned.13 Notethat by controlling for these housing variables, I also control for thehousehold asset accumulation information because housing is themost important vehicle of household asset accumulation (Wei andZhang, 2011). Column 6 uses a set of sibling dummies instead of thenumber of siblings. This relaxes the specification of the functionalform of δ(Sibi) in Eq. (2).

In columns 2 to 6 of Table 3, the coefficient on the number ofbrothers is very stable at around −0.047. The fact that the coefficienton brothers is fairly constant also provides evidence that the numberof brothers is unlikely to be correlated with family characteristics oncewe have controlled for siblings. If the number of brotherswas correlatedwith any of the related individual and family characteristics used in the

12 Send-down was a program during the Chinese Cultural Revolution (1967–1977) inwhich the government forcibly sent adolescents from urban areas to rural areas to carryout hard manual labor. Zhou (2013) found that this event had a large impact on thesend-down youths' income and ability to withstand hard work.13 The size of themortgage is calculated as the percentage of the housing property that isstill unpaidmultiplied by the housing value. Own housing is defined as a house owned bya family member.

regressions, then the coefficient on the number of brothers should havechanged considerably in columns 2 to 6.

One may have a concern about possible gender differences in thebrothers effect. Brothers on the husband's side might contribute differ-ently compared to brothers on thewife's side. I look for possible genderdifferences by introducing the variable “Brothers of Female Respondents”into the regression. This variable is generated by interacting Brotherswith the Female respondent indicator dummy. The interaction variableBrothers of Female Respondents captures the brothers effect on fe-males relative to males (the total brothers effect for females is thesum of the coefficients on the main Brothers variable and Brothersof Female Respondents). The coefficient of this variable is 0.02. How-ever, the standard error is relatively large; the coefficient is not sta-tistically different from zero. I conclude that the brothers effect onfemales is either equal to, or slightly smaller than, the brothers effecton males.

The population policy in 1972 changed from encouraging fertility topromoting voluntary birth control. The number of siblings of individualsdeclined gradually for people born between 1972 and 1979 (Fig. 4). Inorder to avoid the potential effect of this policy change, column 8drops individuals born after 1971. Doing this also allows us to estimatea relatively consistent sample of individuals with a similar number ofsiblings. Column 9 drops individuals close to retirement age. In allthese columns, brothers have a strong negative effect on the householdsavings rate.

One point of caution in the interpretation of the results is that thebrothers effect on individuals born after the One-Child Policy couldpotentially be different from the estimated effect of brothers whichuses the variation in individuals born during the baby boom. After theOne-Child Policy, individuals who have brothers could be differentfrom those who do not. They are more likely to be in the minority, forexample, or they are likely to have had parents who paid the One-Child Policy fine following the birth of an additional boy. (They areprobably more traditional and skeptical of financial savings, relyingmore readily on the family network. This could potentially affect thebrothers effect in younger generations in other ways.)

4. Why brothers reduce the savings rate: risk sharing/extendingborrowing limits and supporting parents

In this section, I propose that brothers reduce the savings ratethrough two channels: (1) sharing their own risks and extending bor-rowing limits, and (2) sharing the risks of their parents. A theoreticalframework is provided in the online appendix to support thearguments.

4.1. A. Individual-level income uncertainties and health risks

I use the degree of uncertainty that individuals encounter to test therisk-sharing/extending-borrowing-limits effect. If brothers play roles insharing risks/extending borrowing limits, those individuals with largerrisks or uncertainties will have a larger brothers effect. Householdswith large risks have a greater need to self-insure, so whether theyhave brothers (withwhom they can share risks) will affect their savingsrate considerably. By contrast, for those households with fewer risks,the presence of brothers might not matter so much; therefore, theyare likely to have a small brothers effect.

I use individual income uncertainties and health risks asmeasures ofthe degree of risk. The income uncertainty measures come from thequestions in the survey, “Is your basic monthly wage stable?” Anindividual can choose among three possible responses: “very unstable,”“a little unstable,” “stable.” I define individualswho responded that theirwage is “very unstable” as the high-risk group. The survey also asks,“How do you feel about the condition of your health?” The answersare “very satisfied,” “satisfied,” “not satisfied,” and “very unsatisfied.”Based on the answers, I define individuals who responded that they are

15 The insurance premium is the sum of the private sector and public sector premia.16 The number of consumer loans was almost zero in 1997 when the Chinese financialmarket was in its infancy. The direct effect of foreign banks on the financial market isreflected in the way that foreign banks offer more services and financial products to con-sumers in the market. The indirect effect is the spillover effect. Foreign banks bring toChina experience and knowledge accumulated in well-developed markets abroad. LocalChinese banks can enjoy a spillover effect by observing the foreign banks' modes of oper-ating in the market and by recruiting employees who have accumulated expertise from

Table 3The impact of number of brothers on household savings rate.

Dependent variable: savings rate

Born 1946–1978 (Age 28–60) Born before 1972(age 35–60)

Born after 1955(age 28–50)

Rural (2) (3) (4) (5) (6) (7) (8) (9)

Brothers − .033⁎ − .047⁎⁎⁎ − .047⁎⁎⁎ − .046⁎⁎⁎ − .046⁎⁎⁎ − .046⁎⁎⁎ − .055⁎⁎ − .062⁎⁎ − .076⁎⁎⁎

(0.018) (0.017) (0.016) (0.016) (0.016) (0.016) (0.022) (0.026) (0.026)Siblings 0.017 0.011 0.014 0.016 0.017

(0.012) (0.01) (0.01) (0.01) (0.01)Brothers of female respondents 0.021 0.022 0.021

(0.024) (0.028) (0.029)Years of education − .004 0.009⁎ 0.01⁎⁎ 0.012⁎⁎ 0.012⁎⁎ 0.012⁎⁎ 0.013⁎⁎ 0.015⁎⁎ 0.013⁎⁎

(0.006) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.007) (0.005)Household income 1.935⁎⁎⁎ 0.328⁎⁎⁎ 0.335⁎⁎⁎ 0.33⁎⁎⁎ 0.333⁎⁎⁎ 0.328⁎⁎⁎ 0.328⁎⁎⁎ 0.315⁎⁎⁎ 0.376⁎⁎⁎

(0.371) (0.093) (0.09) (0.091) (0.092) (0.091) (0.092) (0.094) (0.12)Basic controls Yes Yes Yes Yes Yes Yes Yes Yes YesDetailed backgrounds Yes Yes Yes Yes Yes Yes YesChildren Yes Yes Yes Yes Yes YesHousing Yes Yes Yes Yes YesSibling dummies Yes Yes Yes YesObs. 2364 2581 2540 2540 2503 2503 2503 2068 1816R2 0.179 0.175 0.18 0.227 0.225 0.229 0.23 0.21 0.27

Note: China General Social Survey 2006 is used. Sample is restricted to individuals born between 1946 and 1978. Column 1 uses non-urban residents data. Column 2 to column 11 useurban area residents data. Standard errors are clustered at county level. *** p b 0.01, ** p b 0.05, * p b 0.1.Other variables included:1. Basic controls: female, age, age squared, marital status, years of education, household income and city dummies.2. Detailed backgrounds: mother education, father education, number of people in households, communist party membership and send-down dummy.3. Children information: number of children, children age group dummies: 0–6, 6–18 or 18 and above.4. Housing information: housing dummy, value of mortgage and value of housing.

41W. Zhou / Journal of Development Economics 111 (2014) 34–47

“not satisfied” and “very unsatisfied” with their health as the high-riskgroup. A bad health condition, or unstable wage implies that individualsencounter greater risks.

In Eq. (3),α1 represents the difference in the brothers effect betweenindividuals with high risks and those without, where HighRiski equals 1if individual i belongs to a high-risk group, and 0 otherwise.14

SavingRatei ¼ α0Broi þ α1Broi � HighRiski þ δ Sibið Þ þ Xiγ þ ϵi ð3Þ

The regression results are presented in columns 1 to 2 of Table 4.The results strongly support the risk-sharing hypothesis: householdswith a large income or health risk have a larger brothers effect whereashouseholds with a small income or health risks have a small brotherseffect.

4.2. B. Regional financial development

I further test for the brothers effect of risk sharing/extending bor-rowing limits by exploring the regional variations in financial develop-ment. If the incomplete state of the financial market makes householdmembers rely on their brothers, we should observe that households infinancially developed regions have a smaller brothers effect than dohouseholds in regions where the financial market is underdeveloped.This is because formal credit market information is relatively widelyavailable in financially developed areas. In addition, households havemore alternatives through which to borrow or lend funds in suchareas. Therefore, households face a lower cost of accessing the financialmarket, and they can use the instruments available in financial markets

14 One may want to add Siblings × High Risk to Eq. (3) to ensure the randomness of theeffect of a brother's characteristics on siblings. However, Siblings × High Risk is highly cor-relatedwith Brothers × High Risk when both Brothers and Siblings are included in the re-gression. Although adding Siblings × High Risk can further ensure the randomness ofbrothers, doing this leaves very limited variation to identify the risk-sharing effect ofbrothers. On the other hand, as long as Brothers is a random variation conditional on Sib-lings, the contribution of ensuring randomness by including an interaction termof Siblingswith a risk indicator may be limited.

to insure themselves. These households have less need to rely onbrothers to borrow money or to share risks. In financially underdevel-oped regions, the brothers effect should be large because householdshave no other alternative for acquiring insurance or borrowing money.

I use the provincial-level insurance density and the number offoreign banks per capita in 2005 to measure regional financial develop-ment. See Appendix Table 2 for the statistics of these two variables.Insurance density is provincial-level insurance premiums per capita.15

Insurance density is used to capture overall development in the insur-ance market. The number of foreign banks per capita has direct andindirect effects on local financial development.16

SavingRatei ¼ β0Broi þ β1Broi � FinancialDevelopmenti þ δ Sibið Þþ Xiγ þ ϵi ð4Þ

Eq. (4) is estimated. Note that the city dummies are included in allregressions in this paper in order to control for the regional fixed effect.For this reason, the provincial-level financial development indicators

foreign banks. We observe that the number of foreign banks in each province is deter-mined primarily by government policies rather than by local consumers' demand for fi-nancial instruments. The Chinese government first allowed foreign banks to establishbranches in four cities in Guangdong and Fujian provinces. Only foreign currency busi-nesses were allowed to operate at that time. The next city to acquire permission to haveforeign banks was Shanghai in 1990. In 1992, the government granted the permission toan additional seven cities located in Liaoning Shandong, Jiansu, ZheJiang, Fujian, andGuodong provinces, and Tianjin municipality. In 1996, foreign banks were allowed to en-gage in business using Chinese currency in Shanghai. Later, this policywas extended to theprovinces around Shanghai.

Table 4Brother's sharing risks/extending borrowing limits effect.

Dependent variable: savings rate

(1) (2) (3) (4)

Stability of income

Brothers − .044⁎

(0.026)Brothers × high risk − .088⁎⁎

(0.045)

Personal health

Brothers − .047⁎⁎

(0.022)Brothers × high risk − .057⁎⁎⁎

(0.021)

Regional development

Brothers − .063⁎⁎

(0.029)− .057⁎⁎

(0.028)Brothers × insurance density 0.003⁎⁎⁎

(0.0008)Brothers × # of foreign bankper capita

0.032⁎⁎⁎

(0.009)

Obs. 1408 2503 2503 2503R2 0.309 0.231 0.228 0.229

Note: Sample is restricted to urban area residents born between 1946 and 1978. Standarderrors in columns 1–3 are clustered at county level; Standard errors in columns 3 and 4 areclustered at province level. *** p b 0.01, ** p b 0.05, * p b 0.1.Other variables included:1. Basic controls: siblings, female, age, age squared, marital status, years of education,household income, brothers of female respondents and city dummies.2. Detailed backgrounds: mother education, father education, number of people inhouseholds, communist party membership and send-down dummy.3. Children information: number of children, children age group dummies: 0–6, 6–18 or18 and above.4. Housing information: housing dummy, value of mortgage and value of housing.5. Columns 4 and 5 also include number of brothers × provincial level growth regionalproduct.

42 W. Zhou / Journal of Development Economics 111 (2014) 34–47

are not included in Eq. (4) because of collinearity with the citydummies.17 Regional financial development is usually correlated withregional GDP growth. In order to avoid the potential concern that thebrothers' effect is driven by economic growth instead of financial devel-opment, an interaction term of the number of brothers and regionalGDP growth is also included to control for the potential economicgrowth effect. The error term is clustered at province level to controlfor the random shocks correlated within the province.18

The results in columns 3 and 4 of Table 4 show that the brotherseffect is indeed larger in financially less developed regions. For example,in a province with the smallest insurance density (density= 1), havingan additional brother reduces the savings rate by 6 percentage points(−0.063 + 0.003); in a province with the average insurance density(density = 7), having an additional brother reduces the savings rateby only 4.2 percentage points (−0.063 + 7 × 0.003).

4.3. C. Supporting parents

Health care has become one of the major social issues in China inrecent years. The rising private burden of health care is one of themain explanations of the high savings rate in China, in particular thehigh savings rate among the elderly (Chamon and Prasad, 2010).19

17 Cities dummies (in total 50 cities) absorb all the variation at the city and province (alower level of regional aggregation) level.18 The significance level remains unchanged if I cluster the error term at county level.19 In 1978, out-of-pocket health spending represented 20% of total health spending inChina. In 2002, out-of-pocket health spending represented 60% of total health spending(Yip and Hsiao, 2008).

In Chinese culture, parents are supported primarily by their malechildren (Banerjee et al., 2013; Ge et al., 2012; Lee and Xiao, 1998;Yu et al., 1990). According to the CHARLS 2008 data, the conditionalmean of transfers from male children to parents for medical expensesis almost twice the amount of that from female children: 2964 frommale children and only 1508 from female children.20 A householdwith fewer brothers would need to save more for parents' risks, in par-ticular, risks associated with medical expenditure and shared mainlyamong the brothers.

The absence of a brother can also increase savings because of the fearthat transfers to parents may need to be reduced when an individual ishit by a negative shock. The burden of providing a transfer to parentsgrows even larger following a negative shock, and hardship would beimposed on parents if transfers were reduced. Not having a brothereliminates the risk sharing of transfers to parents, causing the house-hold to save more not only to protect its own consumption but also toprotect that of the parents.

If children do save for their parents, then once the parents havepassed away, a household need no longer save for its parents. I utilizethis idea to identify the size of the brothers effect associated withsupporting parents: I add (1) the number of individual's-parents-deceased term and (2) an interaction term between the number of(individual's) brothers and the number of (individual's) deceasedparents. If parents are deceased, brothers will no longer be playinga role in sharing the risks of parents; therefore, the higher thenumber of parents who have passed away, the smaller the size of thebrothers effect. In Eq. (5), we would expect δ2 to have the oppositesign to δ1.

SavingRatei ¼ δ1Broi þ δ2Bro� ParentDeceasedi þ δ3ParentDeceasediþ δ Sibið Þ þ Xiγ þ ϵi

ð5Þ

Table 5 reports the estimation results. First note that δ3 is significantlynegative. This suggests that households do save for their parents: once aparent has passed away, a household saves less. Second, the brotherseffect becomes smaller if the parents have passed away: δ2 has theopposite sign to δ1. When both parents have passed away, having onebrother reduces the savings rate by 0.032 (0.024 × 2 − 0.08), andwhen no parents have passed away (the brother-parents-deceasedinteraction term also equals zero), the size of the brothers effect reachesits maximum value, |−.08|.

Column 2 uses the One Parent Deceased and Two Parents Deceaseddummies instead of the number of parents deceased variable. Theestimation results reveal that the supporting-parents effect is quitelinear in the number of parents: the coefficient on the Two ParentsDeceased interaction term (0.049) is almost twice that of the coefficienton the One Parent Deceased interaction term (0.017). Similarly, linearityis observed between the Parents Deceased dummies (the noninteractionterms).

Note that two other variables are also added to Eq. (5): whether anindividual has male children and whether a parent (of an individual)is living with that individual. The presence of male children reducesthe household savings rate. This is consistent with the theory thatmale children carry out the duty of supporting parents. Bearing inmind that the financial support of the three generations is suggestedhere: individuals share the cost of supporting parents with theirmale siblings. At the same time, households also expect their ownmale children to support them and therefore reduce their currentsavings rate. Second, the Parents Living Together dummy has a negativesign. Households who live with their parents usually pay a large

20 The sample is restricted to parents aged 60 years and over in 2008.

Table 6The brothers effect in different income groups and asset groups.

Dependent variable: savings rate

All No livingparents

All No livingparents

(1) (2) (3) (4)

Low incomeBrothers − .122⁎⁎⁎

(0.027)− .054(0.034)

Brothers × # of parents deceased 0.039⁎⁎

(0.017)

High incomeBrothers 0.014

(0.016)0.007(0.038)

Brothers × # of Parents deceased 0.016(0.014)

Low assetBrothers − .090⁎⁎⁎

(0.026)− .053(0.043)

Brothers × # of Parents deceased 0.028(0.018)

High assetBrothers − .056⁎⁎⁎

(0.022)− .047(0.033)

Brothers × # of Parents deceased 0.019(0.016)

Obs. 2491 663 2312 615R2 0.313 0.239 0.238 0.21

Note: Sample is restricted to urban area residents born between 1946 and 1978. Standarderrors are clustered at county level. *** p b 0.01, ** p b 0.05, * p b 0.1. The brothers effectin high income group is calculated from the interaction term, high income groupdummy × brothers. The brother's supporting parents effect in high income group is calcu-lated from a triple interaction term: high income group dummy × brothers × number ofparents deceased.Other variables included:1. Basic controls: siblings, female, age, age squared, marital status, years of education,household income and city dummies.2. Detailed backgrounds: mother education, father education, number of people inhouseholds, communist party membership and send-down dummy.3. Children information: number of children, children age group dummies: 0–6, 6–18 or18 and above.4. Housing information: housing dummy, value of mortgage and value of housing.5. Presence of male children.6. Column 1 and 3 also controls number of parents deceased and parents living togetherdummy.

Table 5The impact of number of brothers on household savings rates — the effect of supportingparents.

Dependent variable: savings rate

Brothers − .080⁎⁎⁎

(0.025)− .078⁎⁎⁎

(0.025)Brothers × # of parents deceased 0.024⁎

(0.014)Brothers × one parent deceased 0.017

(0.028)Brothers × two parent deceased 0.049⁎

(0.028)# of Parents deceased − .065⁎⁎

(0.028)One parent deceased − .071

(0.046)Two parents deceased − .131⁎⁎

(0.058)Male children presence − .044⁎

(0.024)− .047⁎⁎

(0.024)Parents live together − .107⁎⁎⁎

(0.038)− .106⁎⁎⁎

(0.038)Obs. 2503 2503R2 0.232 0.240

Note: Sample is restricted to urban area residents born between 1946 and 1978. Standarderrors are clustered at county level. *** p b 0.01, ** p b 0.05, * p b 0.1.Other variables included:1. Basic controls: siblings, female, age, age squared, marital status, years of education,household income, brothers of female respondents and city dummies.2. Detailed backgrounds: mother education, father education, number of people inhouseholds, communist party membership and send-down dummy.3. Children information: number of children, children age group dummies: 0–6, 6–18 or18 and above.4. Housing information: housing dummy, value of mortgage and value of housing.

43W. Zhou / Journal of Development Economics 111 (2014) 34–47

portion of their parents' living expenses. Thus, the household saveless.

One concern is that having few or no brothers could implythat individuals can benefit from not sharing their bequests, such astheir parents' housing.21 Therefore, individuals need to save less ifthey have few brothers (in anticipation of future income flows). Ifthis is the case, the estimated negative brothers effect is a lowerbound; that is, without concern over bequests being shared mainlyamong male siblings, having more brothers would reduce savingseven more.

Transfers to parents are usually not included in household consump-tion and are therefore treated as part of savings based on the definitionof savings rate in this paper. Consequently, the estimation resultthat brothers reduce savings rate (the supporting parents channel) isnot limited to risk sharing but may also occur because respondents(currently) need to transfer less to parents.22

4.4. D. Brothers effect in different income and asset groups

Low-incomehouseholds usually have smaller emergency fundswithwhich to protect themselves from risks. In addition, it is common inChina, and probably in most other financially underdeveloped coun-tries, for banks to lend money only to households with stable jobs andhigh income. This is consistent with the literature that supports theidea that low-income households in developing countries are usuallyborrowing constrained and have difficulty accessing the formal creditmarket (Morduch, 1995). Households with low incomes or few assets

21 This holds particularly true since the privatization of public housing in the 1990s.22 According to China Health and Nutrition Survey 2006, money or gifts a householdtransferred to the parents (non-household members) accounts for 1.03% of householdnet income. The statistics is conditional on household male head aged between 28 and60. Parents include both the mother and father of the husband or wife in the household.Sample size 3073.

may have to rely mainly on their brothers; therefore, these householdswould have a large brothers effect.23

I divide households into low- and high-incomegroups depending onwhether the household income is below or above the median of theoverall income distribution of the sample. The household income levelsare used to approximate the degree of demand for brothers becauseof extending borrowing limits or risk sharing. Column 1 of Table 6reports the brothers effect for each income group. The results revealthat the brothers effect is driven mainly by the low-income group. Thecoefficients of both the number of brothers and its interaction termwith the number of parents deceased are much larger in the low-income groups compared with the previous results (column 1 ofTable 5). In contrast, both of these coefficients are not statisticallydifferent from zero in the high-income group. In column 2, I furtherrestrict samples to individuals with no living parents to exclude thebrothers-supporting-parents effect. Although the standard errors of

23 The 2002 CHIP data suggest that high-income households might have accumulatedenough emergency savings to insure themselves against a shock: 28% of the top-incometertile households stated that they had adequate savings in their bank to finance an emer-gency comparedwith only 8% in the low-income tertile. These data relate to theCHIP2002question “If your household suddenly encountered difficulty and needed 10,000 RMB im-mediately, where or to whomwould you turn first?”

Table 7Robustness check: son preference.

Dependent variable: savings rate

Without son, With son,daughterpreference

daughterpreference

(1) (2)

Basic results

Brothers − .089⁎⁎⁎

(0.027)− .087⁎⁎⁎

(0.026)Son preference 0.057

(0.039)Girl preference 0.057

(0.054)Obs. 927 927

Supporting parents

Brothers − .109⁎⁎⁎

(0.04)− .108⁎⁎⁎

(0.039)Brothers × # of Parents deceased 0.027

(0.017)0.028⁎

(0.017)Obs. 927 927

Individual wage risks

Brothers − .005(0.029)

− .005(0.028)

Brothers × high risk − .086(0.066)

− .086(0.067)

Obs. 511 511

Regional financial development

Brothers − .074⁎

(0.043)− .072⁎

(0.043)Brothers × # of Foreign bank per capita 0.022⁎

(0.013)0.021⁎

(0.013)Obs. 927 927

Income heterogeneity

Brothers × low income dummy − .236⁎⁎⁎

(0.051)− .236⁎⁎⁎

(0.051)Brothers × high income dummy − .029

(0.037)− .029(0.037)

Obs. 927 927

Note: The Family Survey of the China General Social Survey 2006 is used. Sample isrestricted to urban area residents born between 1946 and 1978. Wage unstable equalsone if a respondent characterized his/her wage is very unstable or unstable; 0 otherwise.Wage stable equals one if a respondent characterized his/her wage is stable. Standarderrors are clustered at county level. Standard errors are clustered at county level.*** p b 0.01, ** p b 0.05, * p b 0.1.Other variables included:1. Basic controls: siblings, female, age, age squared, marital status, years of education,household income and city dummies.2. Detailed backgrounds: mother education, father education, number of people inhouseholds, communist party membership and send-down dummy.3. Children information: number of children, children age group dummies: 0–6, 6–18 or18 and above.4. Housing information: housing dummy, value of mortgage and value of housing.

44 W. Zhou / Journal of Development Economics 111 (2014) 34–47

the coefficients are large because of the small sample size, the results areconsistent with what we expected: the low-income group has a muchstronger effect of brothers compared with the high-income group.

I further confirm the heterogeneity of the brothers effect by dividinghouseholds by their assets instead of by their incomes (columns 3and 4).24 Similar to columns 1 and 2, the brothers effect is larger inthe low-asset group than it is in the high-asset group, which confirmsthe strong risk-sharing/extending-borrowing-limits effect ofbrothers.

5. Robustness check: son preference

The identification strategy in this paper relies on parents with a sonpreference not acting on it by selecting the gender of their children. Inthis section, I test the extent to which, if any, the subjective preferencefor a son creates bias in our results by controlling an indicator of sonpreference.

The indicator comes from the question in the Family Survey ofCGSS 2006: “If you are only allowed to have one child, do you prefer aboy or a girl.” A respondent can choose “Boy,” “Girl,” or “Both boyand girl are the same for me.” (The Family Survey of CGSS 2006 is asubset of the China General Social Survey.) The proportion of individualchoices in each category is 20%, 12%, and 67%, respectively, in therestricted sample. Based on the answer to this question, I generated ason-preference indicator and a daughter-preference indicator, wherethe indicator equals one if an individual chooses a specific gender. Thegender preference question is only asked in the Family Survey of CGSS2006, which is a relatively small sample. One limitation of this indicatoris that the son preference refers to individuals, not to individuals'parents. However, the literature has shown that the gender preferenceis largely transmitted fromparents to childrenwithin a family (Escricheet al., 2004).

Table 7 reports the estimation results for the Family Survey sample.Column 1 does not control for gender preferences, while column 2controls for gender preferences. The coefficient of brothers in column1 is very close to the coefficient in column 2. The coefficient ofboth gender preference indicators is not statistically significant(the top panel of column 2). Interestingly, the estimated coefficientof son preference is the same as the coefficient of daughter preference.In the rest of the table, I repeat the same strategy in the estimationof the different channels of the brothers effect. The estimation resultsare nearly identical with or without controlling for son preferenceand daughter preference. These estimation results suggest that thebrothers effect on the savings rate is unlikely to be affected by the sonpreference.

6. How the decline in the number of brothers in households couldexplain the savings rate puzzle

Data from the China Statistical Year Book indicate that the averagesavings rate in urban areas increased from 16% in 1990 to 24% in 2005,

24 Aside from indicating income, assets are also an indicator of the demand for brothers,for potentially two reasons. First, a householdwith sufficient assets would be less likely toborrowmoney from brothers because it can finance consumption using its own emergen-cy funds following shocks. Second, assets, especially housing assets, improve a household'sability to access the formal financial market because assets could act as collateral for bor-rowing money from banks. Most bank loans in China require collateral, and the only ac-ceptable collateral for most banks is buildings or land (Gregory and Tenev, 2001;Ayyagari et al., 2010). Only 4% of commercial loans are secured by movable assets inChina. The value of housing assets is generated by subtracting mortgage balances (unpaidamount) from thehousing values owned by a household. Ideally, total assets value is a bet-ter indicator than housing assets value. As CGSS does not provide total asset data, I usehousing value instead. This caveat is unlikely to cause problems because the rankof house-hold total assets and the rank of housing assets are highly correlated. Using the 2002 CHIPdata, I generate the three-level (low,medium, high) housing value asset rank and total as-set rank. These two ranks are highly correlated: the correlation coefficient is 0.77 and sig-nificant at the 1% level.

5. Number of parents deceased, parents living together dummy and presence of malechildren.

where the average savings rate is defined as “average saving/averagedisposable income.” In this section, I calculate, holding everything elseconstant, to what extent the change in the number of brothers canexplain the change in the savings rate. I also assume that sisters haveno effect on the savings rate.

From the estimation results of the previous sections, we know thatthe brothers effect depends on the number of living parents and theiraverage incomes. Thus, I divide households into six groups: two incomegroups times three age groups. The two income groups are low andhigh; they are equally divided over the income distribution. The three

45W. Zhou / Journal of Development Economics 111 (2014) 34–47

age groups are ages 22–39, 40–49, and 50–60. The changes in thesavings rate in each group depend on the average income, the numberof parents deceased, the number of brothers and the estimated brotherseffect in that group. The total change in average savings is the sumof thechange in savings in each group weighted by each group's density.Mathematically, it can be described in the following way.

Δaveragesaving ¼XA

XI

IncI;AdbroIncI;A þ DPI;A � dbroDPI;A

� �ΔbroI;A f I;Að Þ

ð6Þ

A denotes the age group, and I denotes the income group. Inc is theaverage income. DP is the number of parents deceased. ∇bro denotesthe change in the number of brothers between 1990 and 2005. f(⋅) isthe density of each group. dbroInc is the estimated brothers effect. dbroDPis the estimated brother-supporting-parents effect. The statistics ofthese variables based on the CGSS data are presented in AppendixTable 3. Note that only the statistics of the low-income groups are pre-sented because the savings rate of the high-income group is not affectedby the number of brothers. Themarriage rate is also used in the calcula-tion in order to take into account the change in the number of brothersof both the husband and wife of a household.

The simple calculation suggests that from 1990 to 2005 in urbanChina, declines in the number of brothers in households explained 38%of the increase in the aggregate savings rate. Note that the estimatedexplained increase would be larger if sisters affected the householdsavings rate in the same way as brothers did.25

7. Conclusion and discussion

In this paper, I found that an individual having one brother morereduces that individual's household savings rate by 5 percentage pointsin urban China. Brothers reduce the savings rate because they share therisks/extend borrowing limits, and share the cost of supporting the par-ents. The change in the number of brothers of households explained38% of the increase in the household savings rate.

It is interesting to note that although China is the world'ssecond largest economy, household financial markets are still under-developed even in urban areas. The Chinese governmentmight considerdeveloping household financial markets as soon as possible. Thebaby boom generation can rely on their siblings to finance themselves.They face few hurdles while household financial markets are under-developed. However, the current and future younger generationshave few siblings because of the One-Child Policy. They lack a family-based safety net and they carry the huge burden of supporting theirparents. Developing household financial markets is a necessary andurgent task.

This paper is one of the first papers to estimate the number ofsiblings effect on the household savings rate. The results may not belimited to China only. It would be interesting to see whether other

25 There might be a concern that income transferred to a parents' household may becounted as income of parents' household. If double counting occurs in this manner, thatis, the same amount of income (transfer) is counted in both the income of the householdand the income of the parents' household, it could falsely increase a country's aggregateincome and lead to an over-calculation of the aggregate savings rate. If there is such a dou-ble counting problem, it could bias the estimation of the impact of the change in the num-ber of brothers on aggregate savings. The bias, however,would be limited if parents shouldreceive the same transfer amount (in total) regardless of the number ofmale children theyhave. If total transfers received by parents are fixed, transfers would be canceled outwhenwe calculate the difference between the situations where we have more brothers andthose where we have fewer brothers. Note that even if the change in aggregate savingsis not affected by this double counting, whenwe calculate the change in aggregate savingsrate, that is, the change in aggregate savings out of the total aggregate income, the ratewould be under-calculated if the aggregate income is over-calculated. On the other hand,the double counting problem is unlikely to affect themicro level estimation as I restrict thesample to respondents aged below 60 (they are unlikely to receive transfers from theirchildren).

countries, such as India and other East Asian countries, where house-holds share risks with their siblings and children to support their par-ents financially, have a similar sibling effect on the savings rate. Inaddition, if these countries have cultures similar to that of China, thatis to say, male siblings have stronger family ties compared with femalesiblings, wemay also observe gender differences in the siblings effect onthe savings rate.

Appendix A. Data

The primary data source for this study is the China General SocialSurvey (CGSS) 2006 urban areas sample. It is an individual-levelcross-sectional dataset. The data was collected jointly by theSociology Department of People's University of China and the HongKong University of Science and Technology Survey Research Center.It covered 24 provinces and 4 municipalities. Only three autono-mous provinces were not included in the survey: Tibet, Qinghai,and Ninxia. The survey was conducted based on a probabilistic sam-ple. The stratification design was based on the 2000 populationcensus.

According to the CGSS documentation, the survey only asked onerandomly selected household member, between the ages of 19 and70, to answer all the questions. I dropped all students from the CGSSsample. Among urban area residents who are born between 1946 and1978 (aged 28–60 in 2006), 91% of respondents are married and 7.3%of respondents are living alone. In the following cases, respondentsmay not be counted as the household head: a respondent is livingwith a sibling or with working parents under 60 years old. Fortunately,only 1.4% of respondents are living with siblings. This suggests thatbrothers are most likely to be the members of extended families ofrespondents. Furthermore, only 0.3% of the respondents live with aworking parent under the age of 60. The estimation results remainessentially unchanged when these 1.7% of respondents are excludedfrom the sample. A situation in which a respondent lives with anuncle or an aunt might also be of concern. Unfortunately, the CGSSquestionnaire does not provide the choice of aunt or uncle whenquerying how other household members are related to respondents.This might be due to the fact that, in urban areas, it is quite rare for anindividual to live with an uncle or aunt. Furthermore, since only 0.3%of respondents live with working parents under 60 years old, it isunlikely that any significant number of respondents live with an uncleor aunt who is working and under 60 years old. The basic summarystatistics for all variables used in the regression are presented inAppendix Table 2.

Other supplementary data sets are used in this paper: China FamilyPanel Study (CFPS), Chinese Household Income Project (CHIP) urbanarea sample, Chinese Health and Retirement Longitudinal Study(CHARLS) and China Urban Labor Survey (CULS). CFPS was conductedby the Peking University Institute of Social Science survey in Beijing,Shanghai, and Guangdong province. This study was also based on aprobabilistic sample and stratified design. It is currently available forthe 2008 and 2009 series. CHIP was conducted under the auspices ofthe Chinese Academy of Social Science. The sampling frame is a subsam-ple of the official household survey conducted by theNational Bureau ofStatistics (NBS). The 2002 CHIP survey is used in this study. CHARLSwasconducted by the National School of Development (China Center forEconomic Research) at Peking University. Currently, only the 2008survey is available. The provincial-level data was primarily collectedfrom the China Statistical Year Book published by the NBS. Theprovincial-level financial development data was collected from theAlmanac of China's Finance and Banking. The China Urban Labor Survey(CULS) was administered from November 2001 to January 2002 in fivelarge Chinese cities: Shanghai, Shenyang,Wuhan, Xian, and Fuzhou. Thesurvey was administered by the Institute for Population Studies at theChinese Academy of Social Sciences (CASS-IPS), in collaboration withlocal offices of the NSB in each of the five cities.

Appendix Table 1Household expenditure and total income.

Age group Basic livingexpenditure

Educationexpenditure

Medicalexpenditure

Disposableincome

25–30 12,219 541 660 33,85930–35 11,950 1324 824 30,97635–40 11,790 2203 1119 30,04040–45 10,745 3830 885 24,95345–50 12,144 4218 1109 26,26150–55 11,429 2267 1539 24,89155–60 12,043 882 1755 27,360

Note: Chinese RMB is presented in the table. Exchange rate in 2006: 1 US Dollar = 7.97RMB. ChinaGeneral Social Survey 2006 is used. Sample is restricted to urban area residentsborn between 1946 and 1978.

Appendix Table 2Summary statistics.

Variable Obs Mean Std.dev.

Min Max Unit

Savings rate 2635 0.260 0.545 −5.000 0.947Number of brothers 2635 1.432 1.188 0 8Number of siblings 2635 2.823 1.745 0 9Year of education 2581 10.244 3.057 1 22Age 2635 43.992 8.925 29 60Household yearlyincome

2635 0.285 0.332 0.009 6 100,000 RMB

Marital status 2635 0.908 0.289 0 1Female 2635 0.527 0.499 0 1Mother's years ofeducation

2609 4.746 3.620 1 17

Father's years ofeducation

2597 6.447 3.810 1 17

Family size 2635 2.867 1.039 1 9Communist party 2635 0.126 0.332 0 1Send-down 2635 0.113 0.317 0 1Number of parentsDeceased

2608 0.809 0.837 0 2

Presence of malechildren

2635 0.528 0.499 0 1

Parents live together 2635 0.161 0.368 0 1Children age b6 2635 0.097 0.296 0 1Children age 6–12 2635 0.169 0.374 0 1Children age 12–18 2635 0.179 0.384 0 1Children age N18 2635 0.116 0.320 0 1Number of children 2635 1.070 0.665 0 6No house 2635 0.306 0.461 0 1Value of mortgage 2616 0.013 0.130 0 4.9 100,000 RMBValue of other houses 2612 0.124 0.772 0 20 100,000 RMBMother's companyownership

2635 0.552 0.839 0 2

Father's companyownership

2635 1.099 0.941 0 2

Mother's skill level 2631 0.291 0.884 0 4Father's skill level 2632 0.857 1.457 0 4Father's communistparty

2635 0.139 0.346 0 1

Mother's communistparty

2635 0.031 0.174 0 1

Provincial level data:Number of thebranches of foreignbanks

28 0.360 0.761 0 3.093

Insurance density 28 7.171 9.157 1.030 32.930 100 RMB/Person

46 W. Zhou / Journal of Development Economics 111 (2014) 34–47

Appendix B. Proof of the identification strategy

I show that under the assumption that ϵi is conditional independentof the number of brothers given the number of siblings; that is,

ϵi⊥broijsibi:

α can be consistently estimated in the following equation. (For simplic-ity I ignore other controls.)

Yi ¼ αbroi þ δ sibið Þ þ ϵi ð7Þ

where δ(sibi) is a function of sibi.

Proof. Using the definition of conditional independence, we have

f ϵijsibi; broið Þ ¼ f ϵibroijsibið Þf broijsibið Þ

¼ f ϵijsibið Þ f broijsibið Þf broijsibið Þ

¼ f ϵijsibið Þ

where f(⋅) is the density function. Thus,

E ϵijsibi; broið Þ ¼Zϵiϵ f ϵijsibi; broið Þdϵi

¼Zϵiϵ f ϵijsibið Þdϵi

¼ E ϵijsibið Þ

Since E(ϵi|sibi) is a function of sibi, let

eδ sibið Þ ¼ E ϵijsibið Þ

where eδ sibið Þ is an unknown function of sibi.Assume

Yi ¼ αbroi þ βsibi þ ϵi

Since E(ϵi|sibi, broi) is not dependent on broi, we have

E Yijbroi; sibið Þ ¼ αbroi þ βsibi þ E ϵijsibi; broið Þ¼ αbroi þ βsibi þ eδ sibið Þ

Thus, α can be consistently estimated under Eq. (7), where δ sibið Þ ¼βsibi þ eδ sibið Þ . δ(sibi) is a control function, in order to consistentlyestimate α.

Appendix C. Relative effect of number of brothers

I show that if sisters have an effect on savings rate, we can still deter-mine the difference of the effect between brothers and sisters. Supposethat we are interested in estimating Eq. (8) (for simplicity I ignore othercontrols).

Yi ¼ αbbroi þ αssisi þ ϵi ð8Þ

where sisi is the number of sisters.αb and αs cannot be consistently estimated because broi and sisi are

correlated with the error term ϵi. For this reason, we use the controlfunction approach explained in the previous appendix by adding afunction of sibi into Eq. (8). We can have

E Yijbroi; sisi; sibið Þ ¼ αbbroi þ αssisi þ δ sibið Þ

Due to collinearity,αb andαs cannot be estimated together. However,

E Yijbroi; sibi; sisið Þ ¼ αbbroi þ αs sibi−broið Þ þ δ sibið Þ¼ αb−αsð Þbroi þ δ0 sibið Þ

where δ′(sibi) = δ(sibi) + αssibi. Thus we can still identify the effect ofbrothers relative to sisters which is αb − αs.

Appendix Table 3Statistics used for calculating increased savings rate due to decreased number of brothers.

Age group

22–39 40–49 50–60

dbroIncl;A −0.06 −0.21 −0.13

dbroDPl;A0.01 0.07 0.04

Average income (10,000 RMB) 1.19 1.12 1.16Distribution of each group 0.19 0.15 0.16Marriage rate 0.81 0.88 0.85Number of parents deceased 0.3 0.9 1.5Decreased number of brothers 0.7 0.3 0.4

Note: The statistics of low income households are presented.

47W. Zhou / Journal of Development Economics 111 (2014) 34–47

References

Allen, F.,Qian, J.,Qian, M., 2005. Law, finance, and economic growth in China. J. Financ.Econ. 77 (1), 57–116.

Anderson, S., Ray, D., 2010. Missing women: age and disease. Rev. Econ. Stud. 77 (4),1262–1300.

Arnold, F.,Liu, Z., 1986. Sex preference, fertility, and family planning in China. Popul. Dev.Rev. 12 (2), 221–246.

Ayyagari, M.,Demirguc-Kunt, A.,Maksimovic, V., 2010. Formal versus informal finance:evidence from China. Rev. Financ. Stud. 23 (8), 3048–3097.

Banerjee, A.,Meng, X.,Porzio, T.,Qian, N., 2013. Fertility change and household savings inChina: Evidence from a general equilibrium model and micro data. Working Paper.

Cai, H., Li, H., Park, A., Zhou, L.-A., 2013. Family ties and organizational design: evidencefrom Chinese private firms. Rev. Econ. Stat. 95, 850–867.

Chamon, M.D., Prasad, E.S., 2010. Why are saving rates of urban households in Chinarising? Am. Econ. J.: Macroecon. 2 (1), 93–130.

Chamon, M.,Liu, K.,Prasad, E., 2010. Income uncertainty and household savings in China.Working Paper.

Chen, Y.,Ma, Y., Tang, K., 2011. The Chinese financial system at the dawn of the 21stcentury: an overview. IEB Int. J. Finance 2, 2–41.

Choukhmane, T.,Coeurdacier, N.,Jin, K., 2013. The one-child policy and household savings.LSE Working Papers.

Coeurdacier, N.,Guibaud, S.,Jin, K., 2013. Credit constraints andgrowth in a global economy.LSE Working Papers.

Escriche, L., Olcina, G., Sanchez, R., 2004. Gender discrimination and intergenerationaltransmission of preferences. Oxf. Econ. Pap. 56 (3), 485–511.

Estrin, S., Prevezer, M., 2011. The role of informal institutions in corporate governance:Brazil, Russia, India, and China compared. Asia Pac. J. Manag. 28, 41–67.

Fafchamps, M., 2011. Risk sharing between households. Handb. Soc. Econ. 1, 1255–1279.Fafchamps, M.,Quisumbing, A.R., 2011. Household formation and marriage markets in

rural areas. Handbook of Development Economicspp. 3187–3247.Gan, L., 2012. Findings from the China household finance survey. Working Paper.Ge, S.,Yang, D.T.,Zhang, J., 2012. Population policies, demographic structural changes, and

the Chinese household saving puzzle. IZA Discussion Papers 7026.Gordon, R.H., Li, W., 2003. Government as a discriminating monopolist in the financial

market: the case of China. J. Public Econ. 87 (2), 283–312.Gregory, N.,Tenev, S., 2001. The Financing of Private Enterprise in China. Finance and De-

velopment 38 (1).Guariglia, A., Liu, X., Song, L., 2011. Internal finance and growth: microeconometric

evidence on Chinese firms. J. Dev. Econ. 96 (1), 79–94.Horioka, C.Y.,Wan, J., 2007. The determinants of household saving in China: a dynamic

panel analysis of provincial data. J. Money Credit Bank. 39 (8), 2077–2096.Jacobsen, R.,Moller, H.,Mouritsen, A., 1999. Natural variation in the human sex ratio. Hum.

Reprod. 14 (12), 3120–3125.Lardy, N., 2012. Interest rate liberalization and the international role of the RMB. Working

Paper.Lee, Y., Xiao, Z., 1998. Children's support for elderly parents in urban and rural China:

results from a national survey. J. Cross-Cult. Gerontol. 13, 39–62.Li, X., 2000. Where do we go during 50 years?— a review on Chinese women's liberation

and development process. Zhejiang Acad. J. 2.Modigliani, F., Cao, S.L., 2004. The Chinese saving puzzle and the life-cycle hypothesis. J.

Econ. Lit. 42 (1), 145–170.Morduch, J., 1995. Income smoothing and consumption smoothing. J. Econ. Perspect. 9

(3), 103–114.Qian, N., 2008. Missing women and the price of tea in China: the effect of sex-specific

earnings on sex imbalance. Q. J. Econ. 123 (3), 1251–1285.Song, Z., Storesletten, K., Zilibotti, F., 2011. Growing like China. Am. Econ. Rev. 101,

196–233.Wei, S.-J.,Zhang, X., 2011. The competitive saving motive: evidence from rising sex ratios

and savings rates in China. J. Polit. Econ. 119 (3), 511–564.Yang, D.T.,Zhang, J., Zhou, S., 2010. Why are saving rates so high in China?.Yao, R.,Wang, F.,Weagley, R.O., Liao, L., 2011. Household saving motives: comparing

American and Chinese consumers. Fam. Consum. Sci. Res. J. 40 (1), 28–44.Yip, W., Hsiao, W.C., 2008. The Chinese health system at a crossroads. Health Aff.

(Millwood) 27 (2), 460–468.Yu, L.C.,Yu, Y.,Mansfield, P.K., 1990. Gender and changes in support of parents in China:

implications for the one-child policy. Gend. Soc. 4 (1), 83–89.Zhou, W., 2013. How does a traumatic experience during youth affect later life? The long

term impact of the send-down program during the Chinese cultural revolution.Working Paper.