Brothers, Household Financial Market, and China’s Savings Rate by Weina Zhou Dalhousie University Working Paper No. 2013-02 August 2013

DEPARTMENT OF ECONOMICS DALHOUSIE UNIVERSITY 6214 University Avenue PO Box 15000 Halifax, Nova Scotia, CANADA B3H 4R2

Brothers, Household Financial Market and China’s Savings Rate Weina Zhou∗ August 1, 2013

Abstract This paper suggests that China’s household financial market is underdeveloped. Having more brothers of a household reduces that household’s savings rate in urban China because brothers help each other by (1) sharing risks, providing a source of informal borrowing and (2) sharing the cost of supporting parents. The point estimates suggest that one additional brother reduces the savings rate by five percentage points. This feature helps account for 28% of the increased aggregate savings rate in China between 1990 and 2005, a period that saw a large decrease in the average number of brothers of households. In the estimation, I explore the fact that the number of brothers is random, conditional on the number of siblings for individuals born during the baby boom(1945-1978) and hold urban area residence cards. This paper suggests that sisters play a minor role in affecting a household’s savings rate in China, mainly because of cultural norms. This paper recommends that the Chinese government should consider developing the household financial market as soon as possible in response to changing demographics.

JEL Classification: D12, D81, E21, J12, O16, O17 Keywords: Household Savings, Family Structure, Financial Market Development

∗

Department of Economics, Dalhousie University, 6214 University Avenue, Halifax, NS, B3H 4R2, Canada. Email: [email protected]

1

1

Introduction

Despite China’s impressive growth in recent decades, the household financial market in China remains relatively underdeveloped. Households still rely mainly on family members and relatives for their borrowing and financing needs against shocks. According to the 2009 China Family Panel Study (see data appendix for details), more than 80% of debtors borrowed from family members or relatives while less than 20% borrowed from financial institutions.1 Households also encounter large uncertainties. Health care reforms and rising income uncertainties cause households to have large savings rates (Chamon and Prasad 2010; Chamon et al. 2010). When the development of the financial market does not keep pace with increasing demand for borrowing money and self-insuring, and when households encounter negative shocks, brothers become the first obvious source from which to borrow money. This paper demonstrates that having more brothers of households could reduce a given household’s savings rate. I provide two main channels through which having more brothers would reduce a household’s savings rate. First, brothers reduce their household’s savings rate by sharing risks and extending borrowing limits. Second, having more brothers could reduce a household’s savings rate by sharing the cost of supporting parents. In Chinese culture, the expectation is that parents will be supported by their male children (Banerjee et al. 2010).2 A household with several brothers would need to save less because the cost of supporting the parents is shared amongst the brothers. The key findings in this paper, that having more brothers of a household reduces that household’s savings rate, provides an explanation for the household savings rate puzzle in China. China’s household savings rate is growing surprisingly quickly. It rose from 0.16 in 1990 to 0.24 in 2005 in urban areas (China Statistical Year Book), where savings rate is defined as 1- Living Expenditure/Disposable Income. At the same time, the average number of brothers of prime age household declined due to population policies changes implemented decades ago. From the 1950s to the 1970s, China experienced a large baby boom, induced by China’s population expansion policies. People born in this period have three siblings on average. The population increased from 540 million in 1950 to 960 million in 1978. On the other hand, population contraction began in the 1970s due in large part to the One Child Policy (Scharping 2003, Ebenstein 2010), causing individuals to have fewer or even no siblings. Combining this population shift with the findings in this paper, that the number 1

This survey was conducted in Beijing, Shanghai, and Guangdong province − China’s most developed areas. In other relatively less developed areas, the proportion of households relying on family members could be even larger. 2 This is the main reason why we observe a large increase in male-female gender ratio of newborns after the “One Child Policy”.

2

of brothers is causally negatively correlated with a household’s savings rate, the decreased number of brothers explains at least 27.6% of the increased aggregate household savings rate between 1990 and 2005 in urban China. In estimating the effect the number of brothers has on household savings rate, difficulties arise with the endogeneity problem, that the number of brothers of a household could potentially be correlated with that household’s unobserved characteristics. This paper found that in urban areas during the baby boom induced by policies introduced in the 1950s and 1960s, the gender ratio was well-balanced among urban area residents. Evidence suggests it was unlikely that the urban area residents would or could take son-preference into practice during the baby boom. Conditional on the number of siblings, the fraction of male siblings could be considered as a random assignment by nature for people born in urban areas during the baby boom. By utilizing this finding, this paper identifies the effect of having a brother instead of having a sister (a relative effect) by controlling a function of the number of siblings of a household in the regression. In the robustness check found in Section 6.2, I estimate the effect the number of brothers has on household savings rate by using an instrument variable strategy for the individuals born after the baby boom. The results are consistent with the hypothesis that having one more brother rather than not would reduce a household savings rate (an absolute effect). This paper suggests that having a brother instead of a sister would reduce household savings rate by 5 percentage points. Statistical evidence suggests that sisters could have only a minor effect on a household’s savings rate, as sisters usually play a minor role in supporting parents or sharing risks with other siblings. A possible reason for this is that the connections between female and male siblings and the connections between parents and daughters are relatively weak in Chinese culture. These weak connections also account for brothers-in-law (brothers of female heads of households) having a smaller effect on a household’s savings rate. If sisters do not affect a household savings rate, the estimated relative effect of a brother (i.e., having a brother instead of a sister) is the same as the absolute effect, (i.e., having one more brother rather than not). If sisters also behave (partially) like brothers and affect household savings rate, the estimated relative effect would be a lower bound. In order to test the existence of a risk sharing / extending borrowing limits effect, this paper tests the effect brothers have on households with different levels of (1) wage uncertainties, (2) bonus uncertainties, (3) health risks, (4) regional financial development and (5) income or asset level. The estimation results are consistent with the hypothesis: households who encounter large wage and bonus uncertainties, have high health risks, live in a financially less developed province, have a low income or low asset levels would have a larger brother effects. The robust and consistent results suggest a strong risk sharing/extending borrowing 3

limits effect of having brothers. To test for the parent-supporting aspect, this paper utilizes information on whether parents are deceased. Once parents have passed away, brothers no longer play a role in sharing the cost of supporting them. The difference in the number of parents still living helps identify the parent-supporting effect of brothers. In China’s literature on savings, recent papers have provided evidence that the change in demographic structure could affect household savings rates due to intergenerational support (Wei and Zhang 2011; Banerjee et al. 2010; Ge et al. 2012; Yang 2012). Banerjee et al. (2010) provide evidence of how households’ savings rates would be affected by the number of children and the gender of their children. Ge et al. (2012) provide an OLG model, which incorporates the effects of intergenerational support in a life cycle framework, and estimate both effects from a macro perspective. Wei and Zhang (2011) suggest that the rising gender ratio induced parents to save more for their male children to help them secure a better outcome in the marriage market. This paper contributes to the literature by suggesting that, in addition to intergenerational support, the change in demographic structure may affect household savings rates in another dimension. The fact that brothers share risks with each other in the current, underdeveloped household financial market also accounts for higher household savings rates in recent decades. This paper also helps to explain why there is mixed evidence regarding whether the decreasing dependency ratio could explain the rising savings rate. Typically, prime age individuals save more than the young and old, due to higher labour income. Modigliani and Cao (2004) using long term provincial level data suggest that the decrease of population in young and old, compared to those of prime age, contributes to the rising savings rate in China. On the other hand, Horioka and Wan (2007) use more recent data and find that the change in dependency ratio does not sufficiently explain the increased savings rate. This paper helps to solve the puzzle by emphasizing that individuals of prime age could save less because they have more brothers. The recent younger generation contributes to the high savings rate because they don’t have siblings. As far as I am aware, this paper is one of the few research projects that uses micro level data to estimate the siblings’ effect on a household’s savings rate, not only in Chinese savings literature, but also in the savings literature of other economies.3 To date, it is not clear how much the presence of a given family member could affect a household’s savings rate. 3

The risk sharing among family members in developing countries is relatively well established in literature. However, most work has focused on realized gift transfers and consumption smoothing (Rosenzweig 1988; Fafchamps and Cox 2007; Foster and Rosenzweig 2001).

4

This paper contributes to the literature by providing empirical suggestion that the number of brothers, which may represent the amount of potential transfers or quasi credit available to households, affects the households’ savings rate. Furthermore, this paper suggests that the role of risk sharing / extending the borrowing limits among family members could vary greatly depending on the gender of a family member. (To my limited knowledge, I am not aware any other paper has provided such suggestion.) This finding may not be limited to China; other countries where the household financial market is underdeveloped and male siblings have stronger family ties compared to female siblings could have such gender differences in risk-sharing among family members. The paper proceeds as follows: Section 2 introduces the background of the household financial market, population policies, and the current phenomenon of the savings rate in China. Section 3 introduces the identification strategies and presents the estimation results. Sections 4 and 5 explore the dimension of the brother-supporting-parents effect and risk sharing / extending borrowing limits effects, respectively. Section 6 provides additional analysis of differences between brothers and brothers-in-law, the estimation results for the individuals born after the One Child Policy, and how much of the savings rate puzzle can be explained by the brother effect. Section 7 concludes the paper. Note that the data sources are described in Appendix A.

2

Background

2.1

Financial Market

It is a well-known fact that the corporate financial market in China is underdeveloped (Ayyagari et al. 2010; Guariglia et al. 2011; Chen et al. 2011; Song et al. 2011; Allen et al. 2005). To date, little attention has been paid to the household financial market even though the degree of development in the household financial market is no better than that of the corporate financial market (Liao et al. 2010; Yao et al. 2011). Despite that the real interest rate on domestic bank deposits has often been negative (Gordon and Li 2003; Lardy 2012), Gan (2012) uses the China Household Finance Survey 2011, which suggests that two major financial assets for households are bank deposits (58%), and cash holdings(18%). The rate of consumer loans issued by all financial institutions in China was nearly zero in 1997(Chamon and Prasad 2010). Although it reached 2.2 trillion RMB at the end of 2005, mortgage loans amounted to about 80% of the total loans.4 Households also encounter large uncertainties. Medical reforms, pension reforms and 4

The other major loans were auto loans and large durable goods loans.

5

rising income uncertainties cause households to save more due to precautionary motives(Chamon and Prasad 2010; Chamon et al. 2010). How do households in China finance themselves when they encounter negative shocks? I use two different data sets to investigate how Chinese households finance themselves in the current underdeveloped household financial market. The first data set comes from the Chinese Household Income Project (See Data appendix). The CHIP 2002 urban area survey asked, “If your household suddenly encounters difficulty and needs 10,000 RMB immediately, where or to whom would you turn first?”5 I report the results in Figure 1. More than 60% of the individuals chose “family members and relatives” while less than 3% of the individuals chose “financial institutions.” Note that 10,000 RMB is approximately 1,600 USD, which is more than half of the median household’s yearly income in the 2002 CHIP data. The result strongly suggests that family members and relatives are a household’s primary borrowing source. It also suggests that the amount of potential transfer or quasi-credit available among family members is very large. The China Family Panel Study 2009 asked households if they actually borrowed money in 2008, if so the sources they borrowed from, and the reason they borrowed. In total, 14% of the survey respondents had borrowed money in 2008. As was the case in the report using the 2002 CHIP data, “Relatives and friends” was overwhelmingly the dominant borrowing resource for households. Conditional on borrowing money, 82.3% of the households borrowed from relatives and friends while less than 20% of the borrowers borrowed from banks.6 Note that this survey was conducted in Beijing, Shanghai, and Guangdong province − China’s most financially developed areas. In other relatively less developed areas, the proportion of households relying on family members could potentially be even larger. The reasons for borrowing also varied from relatives to banks.7 In the CFPS data, housing was the main reason for borrowing from financial institutions; it accounts for 85%. In contrast, reasons for borrowing from relatives and friends had a wide range that was fairly distributed among “education”(18%), “medical treatment”(20%), “housing”(22%), “living expense”(15%), and “other”(26%). It is worth noting that the housing loan market is quite developed in China, perhaps due 5

There were nine answers to choose from: (1) family members and relatives, (2) friend, (3) other individuals, (4) work unit, (5) bank and credit union, (6) other financial institutions, (7) need no help, (8) anywhere I can borrow, (9) other. I aggregated (5) and (6) together and named this category “financial institutes”; I aggregated (3) (4) (8) and (9) together as “other.” 6 Households had the following options to choose from in the survey: (1) banks (including credit unions), (2) relatives and/or friends, (3) loan from a private institute, (4) other. Only 2% of households had borrowed from (3) or (4). 7 Because only a few observations are available, I report the number of observations for each category instead of using proportion.

6

to the government’s enforcement of housing reforms, which encourages individuals to buy houses.8 Since the primary reason for people borrowing money from banks is housing, and mortgages are not considered an unexpected expense, relatives and friends become the only source for borrowing when a household encounters unexpected shocks.

2.2

Supporting Parents

A large amount of literature indicates that children, especially male children, in China support their parents in old age (Lee and Xiao 1998; Yu et al. 1990; Banerjee et al. 2010; Ge et al. 2012). Children support their elderly parents by transferring money to them for living costs, paying their medical expenses, and living with them in order to provide care. The 2008 China Health and Retirement Longitudinal Study (CHARLS) data provides detailed information about how children support their parents. Compared with female children, male children are more likely to live with their parents and make monetary transfers to them for medical and other reasons. Whereas only 4% of female children live with their parents, 28% of male children do. The proportion of male children making regular transfers to their parents is twice that of female children, and conditional on positive transfers, male children make twice the amount compared to female children. In regards to medical expenses, male children also make twice the transfer amounts that female children make.9 Public health care has become one of the major social issues in China. In 1978, out-ofpocket health spending was 20% of total health spending in China. In 2002, out-of-pocket health spending was 60% of the total health expenditure (Yip and Hsiao 2008). Children need to save for their parents for future expenditures, especially for the possibility of higher medical expenses resulting from health-care reform.

2.3

Population Expansion and the One Child Policy

China experienced a huge baby boom after 1949 and before the One Child Policy was introduced in 1978. Family planning has featured in Chinese policy makers’ discussions since the founding of the People’s Republic in 1949. The population policies in China can be divided into three main stages: population expansion (1949-1972), voluntary birth control (1972-1978), and the One Child Policy (1979-current). After the founding of the People’s Republic of China in 1949, policy makers promoted fertility. In 1949, Chairman Mao Zedong claimed,“It is a very good thing that China has a big 8

Before the housing reforms, employees’ housing had been provided by their employer, State-owned firms or Collectively-owned firms, at very low cost. 9 Samples are restricted to individuals who are above 60 years old in 2008. Sample is not restricted based on residence card status(Huko).

7

population. Even if China’s population multiplies many times, she is fully capable of finding a solution; the solution is production” (Tien 1964). The Chinese government introduced many policies to encourage fertility. For example, in 1952, the government published a regulation to restrict sterilization and abortions (Banerjee et al. 2010). The policy allowed a female to have an abortion only if the female was over 35 or had already had six or more children. Chairman Mao Zedong’s famous words “the more people, the stronger we are” is still a well-known phrase in China, even for the current generation. This large population growth was slowed by the second stage of family planning policies implemented in 1972. During this stage, the government carried the slogan “later, spaced, and few.” “Later” for later marriage, “spaced” for spaced birth, and “few” for fewer children. The policy did not place a cap on the total number of children; rather, it emphasized birth spacing. However, the population control policy at this stage was voluntary and no punishment was meted out for violations. The final decision to adopt birth control methods was left to the couples themselves. Due to these population policies, China’s population almost doubled in just 30 years: it increased from 540 million in 1949 to 960 million in 1978. The third stage of family planning policies is the famous One Child Policy stage. This policy was introduced in 1978 and applied to the babies born in 1979. In urban areas, each family was allowed only one child. In rural areas, a second child was allowed if the first child was not male. Additional children resulted in large fines; families violating the policy were required to pay monetary penalties and could be denied bonuses at their workplaces. These family birth-planning policies resulted in a massive change in the number of siblings for people born during the early and later stages. The policies introduced in the first stage caused people born in that period to have on average three or more siblings while people born after the One Child Policy generally have no siblings. Figure 2 presents the number of brothers and sisters by respondents’ year of birth. The median number of siblings is 3 for individuals born between 1946 and 1978 (the mean is 2.8). There is no significant difference between the number of male siblings and the number of female siblings. The average number of both male and female siblings is 1.4.

2.4

Household Savings Rate

Using data from the China Statistical Year Book, I compute the savings rate of urban households. The savings rate national account is defined as 1 − AverageExpenditure/AverageIncome, where expenditure is households per capita living

8

expenditure, and income is households per capita disposable income.10 Figure 3 plots the savings rate from the early 1980s to 2005. The savings rate increases over time in the 1990s, increasing particularly quickly as of the late 1990s. Figure 3 also plots the average number of brothers for households with heads between the ages of 20 and 60 by using CGSS data. Since the change in the number of brothers is mainly driven by the change in the new cohort, the average number of brothers decreases slowly during the 1980s and falls quickly by the late 1990s. The figure suggests a negative correlation between the trends of household savings rates and the average number of brothers. The 2006 CGSS urban area data contains the total income, living expenditure, medical expenditure, and education expenditure information of households. Savings are constructed as total disposable income minus sum of these three expenditures (see Appendix Table 1 for summary statistics of these variables). Savings rate is defined as savings divided by the disposable income.11 Appendix Table 1 shows detailed descriptive statistics of disposable incomes and expenditures. The average savings rate in 2006 was 0.26 for households in urban areas.12 This number is only 1.4 percentage points higher than the savings rate computed by using the data in China Statistical Year book for urban households (0.246). Figure 4 shows the age profile household savings rate by the number of brothers and sisters. I divide the households into two groups: households with zero or one brother/sister and households with more than one brother/sister. The upper panel of Figure 4 suggests that households with zero or one brother have a higher savings rate compared to households with more than one brother, at any point of age. There is a strong negative correlation between the number of brothers and the household savings rate at any point of age. However, no clear pattern is observed for the savings rate and the number of sisters (the lower panel of Figure 4).

10

Before 1996, the China Statistical Year Books do not distinguish between disposable and actual income. However, the difference between the two incomes was only 0.1% in 1996. Arguably, there is almost no difference between disposable and actual income before 1996. 11 The income taxes are computed by author based on the Individual Income Tax Law of the People’s Republic of China adopted in 2005. In an early version of this paper, author used income instead of the disposable income. The estimation results which use disposable income are very similar to the results using (non-tax deducted) income: the direction and the significance levels of the coefficients are all the same, with almost no noticeable change in the size of coefficients. The estimation results using (non-tax deducted) income will be provided upon request. 12 Due to the identification strategy explained in the following section, the sample is restricted to those respondents who have urban area residency cards with the ages of 28 and 60.

9

3

The Impact of the Number of Brothers on Households’ Savings Rate

3.1

Identification

In order to identify the effect of brothers on the savings rate, let us first consider the following equation: SavingRatei = αbroi + Xi γ + i (1) The definition of savings rate is given in the section 2.4. broi is the number of brothers of the household head. It could be either brothers of a male head or brothers of a female head (brothers-in-law).13 . Here I do not distinguish the difference in brothers and brothers-in-law. Section 6.1 distinguishes this difference by using a subsample which has both brothers and brothers-in-law information. Xi is a set of household and household head’s characteristics. α, the coefficient on broi , is the parameter we are interested in. broi could be correlated with unobserved family characteristics, which may have impact on households savings rate. Thus α cannot be consistently estimated through equation 1. I consider the following control function approach to solve the endogeneity problem. If conditional on number of siblings the fraction of male siblings is determined by nature, i.e., parents could not manipulate the gender of the siblings, then conditional on the number of siblings the number of brothers is randomly assigned by nature. Therefore, conditional on number of siblings, number of brothers is not correlated with any unobserved characteristics; α can be consistently estimated. Mathematically, under assumption 1, α can be consistently estimated through equation 2. See Appendix B for proof. SavingRatei = αbroi + δ(sibi ) + Xi γ + i

(2)

Assumption 1: i ⊥broi |sibi . δ(sibi ) is an arbitrary function of sibi . The function of the number of siblings could be any form. Here, I use the number of siblings as a simplest function of the number of siblings. The results are robust if I relax the functional form of the number of siblings. The assumption that conditional on number of siblings number of brothers is a random assignment requires that there are no predetermined family characteristics would affect the 13

According to the survey design and sample restriction, either a female head or male head would be chosen to answer the survey. About 52% of the respondents in the restricted sample are female head.

10

assignment of the gender of the siblings (the only thing that can determine the gender of the siblings is nature). Several papers in the “missing female” literature indicate that Chinese households have a son preference and that the sex ratio became largely distorted after the One Child Policy was introduced in 1979 ( Wei and Zhang 2011; Arnold and Zhaoxiang 1986). Parents “chose” the gender of their children by practicing sex-selective abortion; sometimes female infanticide was also found in rural areas. I found that by restricting the sample to individuals holding an urban area residency card (Huko), and born before the One Child Policy(1979) and after World War II (1945), the evidence suggests that the gender of individuals’ siblings are exogenously assigned. There are several reasons for this. First, the ultrasound technology required for sex-selective abortions was only introduced in the 1980s; households before the 1980s had no reliable method for performing sex-selective abortions. Second, before 1979, people who held an urban area residence card had almost no financial incentives for having sons instead of daughters as the social welfare system before 1979 (in Mao’s era) for urban residence was very good. The main economic reason of son preference is that male children provide financial support to parents when parents get old. However, before the end of the 1970s, housing was almost free for retired urban residents, as were medical services (known as “Lao Bao” in Chinese). Parents having babies during that time were unlikely to practice son preference for their post-retirement support.14 Third, female infanticide was found most in rural areas where households delivered babies at home. In urban areas, babies were usually delivered in the hospital. It was unlikely that urban households would risk criminal prosecution for son preference. Note that people born close to 1979 are unlikely to have siblings born after 1979, because of the One Child Policy. The estimation results are also robust if I restrict the data to individuals born before 1972 (the year that the second stage of family planning policy was introduced). Two sets of the statistical evidence suggest that the gender of children is exogenously assigned in the restricted sample. Table 1 reports the proportion of male siblings given a number of siblings. The natural gender ratio is 106 male per 100 female (Jacobsen et al. 1999). This implies that the natural fraction of male siblings is 51.5. We are worried the fraction may be above 51.5 due to son preference in China. The statistics computed in Table 1 show that the proportion of males is around the natural outcome, regardless of respondents’ number of siblings. Table 2 provides a test of random assignment of the number of brothers conditional on the number of siblings. In column 1, where the number of siblings is not controlled, the 14

The social welfare system changed considerably after the 1980s. Health care reform and housing reform largely increased individuals’ health care expenditure and housing expenditure.

11

number of brothers is significantly correlated with several variables such as the mother’s education and the father’s Huko status (residence card status). The Wald test suggests that all the family backgrounds are jointly significant. However, once the number of siblings is controlled (column 2), parents characteristics are no longer significant, and the Wald test suggests that they are not jointly significant. I repeat the same test for the fraction of male siblings (column 3), and it yields similar results. Results in Table 2 provide strong suggestive evidence that, conditional on the number of siblings, the number of brothers is random. For this identification strategy, variation derives from the differences between having a brother instead of a sister. The panel A of Figure 5 presents the average savings rate by the number of brothers for a given number of siblings. The table suggests that for any given number of siblings, having more brothers is associated with a lower savings rate. Since savings rate is defined as savings divided by income, we want to make sure that the negative correlation between number of brothers and savings rate is not driven by the income correlation. Panel B of Figure 5 suggests that this is not a concern as there is not a clear pattern of how income is correlated with number of brothers, given a number of siblings. One might also want to know the effect of sisters on a household’s savings rate. Ideally, we want to include the number of sisters in the regression to estimate the impact of the number of sisters on the savings rate. However, such an estimate is not feasible due to the problem of collinearity (we cannot add both the number of brothers and sisters and siblings in one regression). As we control the number of siblings, α gives the difference between the effect of brothers and that of sisters. The coefficient on the number of siblings represents the effect of sisters with bias induced by endogeneity; see Appendix C for proof. Although the true effect of sisters could not be estimated, from Figure 4, it is more likely that sisters have zero effect on a household’s savings rate. If this is the case, the estimated relative effect of brothers compared to that of sisters, α, also equals the absolute effect of brothers. If sisters behave like brothers, also playing a role with other siblings in the interaction over risk-sharing and supporting parents, the estimated brother effect here would be a lower bound of the absolute effect of brothers (Appendix C). Because the missing female problem could still occur in rural areas, one might be concerned that people migrating from rural areas to urban areas could potentially bias the results. I restrict the sample to individuals who have urban area residency cards. This rules out migrants, since urban area residency cards are only issued to city area permanent residents.15 One might also have concern that some parents with a strong son preference would 15

The urban area residency card system was introduced during the Mao era and was used to prevent permanent migration because the social security system and welfare provided by the government were, and continue to be, different from urban to rural areas.

12

adopt a practice of having babies until they achieve a desirable number of boys. However, given the population expansion policies introduced during the baby boom, it is unlikely that parents would be able to take it into practice.

3.2

Results: the Impact of the Number of Brothers on the Savings Rate

I restrict the sample to the household heads holding an urban area resident card and born between 1946-1978 because of the identification strategy discussed above. The basic summary statistics for all variables used in the regression are presented in the Appendix Table 2. The main regression results are presented in Table 3. The baseline specification, which controls for the number of siblings, years of education, gender, age, age squared, household income, and marital status is reported in column 1. City dummies are added into all regressions as well in order to control the regional differences in the savings rate. The coefficient on the number of brothers is -0.048 and statistically significant at the 1% level. This means that having one brother instead of a sister would, on average, reduce the savings rate by 4.8 percentage points. Columns 2, 3, and 4 add further controls that could potentially affect a household’s savings rate. Column 2 adds a large set of demographic and characteristic controls: family size, parents living together dummy, Communist Party membership status, father’s and mother’s education, and a send-down dummy.16 Chamon and Prasad (2010) indicate that increases in children’s education expenses and housing reform induced households to save more. Column 3 adds the number of children and children’s age group dummies in order to control the potential education expense effect. Column 4 adds the household’s housing characteristics: a dummy variable indicates whether the household owns the house, the mortgage value (if the house is owned), and the value of houses a household may own.17 Note that by controlling these housing variables, I also control the household asset accumulation information since housing is the most important vehicle of household asset accumulation ( Wei and Zhang 2011). Column 5 uses a set of sibling dummies instead of the number of siblings. This relaxes the specification of the function form of δ(sibi ). Throughout column 1 to column 5 of Table 3, the coefficient on the number of brothers is relatively stable; it stays around -0.05. The fact that the coefficient on brothers is fairly 16

Send-down was a program during the Chinese Cultural Revolution (1967-1977) in which the government forced adolescents in urban areas to go to rural areas to do hard manual labor. Zhou (2013) found that this event had a large impact on the send-down youth’s income and ability to withstand hard work. 17 The value of the mortgage amount is calculated by the percentage of the housing property that is still unpaid multiplied by the housing value.

13

constant also gives us evidence that the number of brothers is unlikely to be correlated with family characteristics once we have controlled for siblings. If the number of brothers were correlated with any of the related individual and family characteristics used in the regressions, then the coefficient on the number of brothers should have changed considerably. The population policy switched from encouraging fertility to voluntary birth control in 1972. The number of siblings of individuals gradually declined for people born between 1972 to 1979 (Figure 2). In order to avoid the potential effect of this policy change, column 6 drops individuals born after 1971. Doing this also allows us to estimate a relatively consistent sample of individuals with similar amount of siblings. Column 7 drops individuals close to retirement age. The last column focuses solely on individuals age between 35 and 50. Throughout these columns, brothers have a strong negative effect on household savings rate. I denote all the controls included in column 4 of Table 3 as ‘full controls’. In the remaining sections, I include the ‘full controls’ in all the regressions.

4

Why Brothers Reduce the Savings Rate: Supporting Parents

The point estimates in the last section suggests that on average one brother reduces the savings rate by about 5 percentage points (Column 1 to 5 of Table 4). In this section and the following, I will explore the reasons why a brother would reduce the savings rate. If children do save for their parents, then once parents have passed away, a household need no longer save for its parents. In order to identify the quantity of the brother effect due to supporting parents, I add (1) the number of (household head’s) parents deceased term and (2) an interaction term between the number of (household head’s) brothers and the number of (household head’s) deceased parents. The idea is that, if parents are deceased, brothers would no longer be playing a role in sharing the cost of supporting them; therefore, the higher the number of parents who have passed away, the smaller the size of the brothers effect. Table 4 reports the estimation results of the following equation. SavingRatei = α1 broi + α2 bro × parent deceasedi + α3 parent deceasedi

(3)

+ α4 sibi + Xi γ + i The estimation results in column 1 suggest that, first households do save for their parents. If one parent passed away, the household savings rate would decrease by 0.081 (the coefficient 14

of α3 ). Second, the higher the number of parents who have passed away, the smaller the brothers effect. α2 , the coefficient of the interaction term of brothers and parents deceased term, has the opposite sign to α1 , the coefficient of number of brothers. When the number of parents passed away equals to zero (the brother-parents-deceased interaction term also equals zero), the size of the brothers effect reaches its largest value, | − .08|; when both parents had passed away, having one brother reduced savings rate by 0.028 (0.026 × 2 − 0.8). Column 2 uses the one parent and two parents deceased dummy instead of the number of parents deceased to investigate the nonlinearity in brothers sharing the cost of supporting parents effect. The estimation results reveal that the effect is linear: the coefficient on the two-parents-deceased interaction term (0.052) is almost twice that of the coefficient on the one-parent-deceased interaction term (0.019). Similarly, linearity is observed between the parents deceased dummies (the non-interaction terms). Two additional results are also worthy of our attention. First, the presence of male children reduces the household savings rate. This is consistent with the theory that male children carry out the duty of supporting parents. Note what the financial support of the three generations have suggested here: households share the cost of supporting parents with their male siblings. At the same time, households also expecting their own male children would support them. Second, the parents living together dummy is controlled and it has a negative sign. Households who live with their parents usually pay a large portion of their parents’ living expenses.18 Thus if a household lives with the parents, that household’s living expenditure increases due to the parents; the household saves less. Note that there is difference in supporting wife-side parents and husband-side parents. I investigate this difference in detail in Section 6.1.

5

Why Brothers Reduce the Savings Rate: Risk Sharing / Extending Borrowing Limits

In this section, I propose that brothers reduce the savings rate through the dimension of risk-sharing and extending borrowing limits. Figure 6 plots the age profile savings rate for households with no living parents. I restrict the sample to households with ages over 45 because there are very few households with no living parents below this age. The figure 18

According to the CLUS data, if a senior is living with his or her child, the senior only pays 58% of his/her own living expenses; 38% of the living expenses are paid by the child who lives with them. However, seniors not living with a child pay 88% of their own expenses; the remainder is shared by those children not living with the parents and other family members.

15

suggests that without the concern of supporting parents, brothers still have a strong negative correlation with the household savings rate. For sisters, the correlation with the savings rate is still not so clear. I test the risk sharing / extending borrowing limits by exploring (A) individual level income uncertainties and health risks, (B) regional financial development and (C) income and asset level. A theoretical framework is provided in the online appendix to support the arguments. A. Individual Level Income Uncertainties and Health Risks In this subsection, I use the degree of uncertainties that individuals encounter to test the risk-sharing / extending borrowing limits effect. The idea is that if brothers play roles in sharing risks / extending borrowing limits, those individuals with larger uncertainties would have a larger brother effect. Households with large uncertainties have more needs to self-insure, so whether they have brothers (with whom they can share risks) would affect their savings rate considerably. By contrast, for those households with fewer uncertainties, the presence of brothers might not matter so much; therefore they are likely to have a small brother effect. I use individual income uncertainties and health risks as measures of the degree of uncertainties. The income uncertainties measures come from the questions in the survey, “Is your basic monthly wage stable?” and “Is your monthly bonus stable?” A respondent can choose among three possible responses: very unstable, a little unstable, stable. The survey also asks, ”How do you feel about the condition of your health?” The answers are “very satisfied,” “satisfied,”“not satisfied,” and “very unsatisfied.” Based on the answers, I evaluate the individual’s health condition as “very good,” “good,” “bad,” and “very bad”. A bad health condition, unstable wages, or bonus imply that individuals face large uncertainties. The regression results are presented in column 1-3 of Table 5. The regression results strongly support the hypothesis that brothers play a role in sharing risks and extending borrowing limits: households with large income uncertainties or health risks have a greater brother effect; households with small income uncertainties or health risks have a small brother effect. B. Regional Financial Development In this subsection, I test the brother effect of risk sharing/extending borrowing limits by exploring the regional variations in financial development. If the incomplete state of the financial market makes households rely on their brothers, we should observe that households in financially developed regions have a smaller brother effect than households in regions where the financial market is underdeveloped. This is because in financially developed areas, formal credit market information is relatively widely available; households have more alternatives 16

through which to borrow or lend funds. Therefore, households face a lower cost for accessing the financial market and they can use the instruments available in financial markets to insure themselves. These households have less need to rely on brothers to borrow money or share risks. In financially underdeveloped regions, the brother effect should be large, because households have no other alternatives for acquiring insurance or borrowing money. I use two indicators for provincial-level financial development: insurance density and number of foreign banks per capita in 2005. See Appendix Table 2 for the statistics of these two variables. Insurance density is the ratio of regional insurance premiums to regional population − the unit is 100 RMB per capita.19 These two variables are used to capture overall development in the insurance market. The number of foreign banks has direct and indirect effects on local financial development.20 Note that the city dummies are included in all the regressions in order to control for the regional fixed effect. For this reason, only the interaction terms of financial development indicators and number of brothers are included in the regression, while financial development indicators themselves are not included due to collinearity with city dummies. An interaction term of the number of brothers and growth regional product is also included to control for the potential economic growth effect. Columns 4 and 5 of Table 5 present the results of these three regressions. The sign of coefficients of both the number of brothers and its interaction terms in both regressions are consistent with the risk-sharing/extending borrowing limits hypothesis. For example, the regression results in the 4th column show that, in a province with the smallest insurance density (density=1), having an additional brother reduces savings rate by 9.1 percentage points (-0.093+0.002); in a province with the largest insurance density (density=32), having an additional brother reduces savings rate by only 2.9 percentage points (-0.093+32×0.002). C. Brothers Effect in Different Income and Asset Groups 19

The insurance premium is a sum of both private sector and public sector premium. The number of consumer loans was almost zero in 1997 when the Chinese financial market was in its infancy. The direct effect of foreign banks on the financial market is reflected in the way foreign banks offer more services and financial products to consumers in the market. The indirect effect is the spill-over effect. Foreign banks bring to China experience and knowledge accumulated in well-developed markets abroad. Local Chinese banks can enjoy a spillover effect by observing the foreign banks’ ways of operating in the market and recruiting employees who have accumulated expertise from foreign banks. Note that the number of foreign banks in each province is mostly determined by government policies rather than by local consumers’ demand for financial instruments. The Chinese government first allowed foreign banks to establish branches in four cities in Guangdong and Fujian provinces. Only foreign currency businesses were allowed at that time. The next city to acquire permission was Shanghai in 1990. In 1992, the government granted permission to an additional seven cities located in Liaoning Shandong, Jiansu, ZheJiang, Fujian, and Guodong provinces, and Tianjin municipality. In 1996, foreign banks were allowed to engage in business using Chinese currency in Shanghai. Later, this policy was extended to the provinces around Shanghai. 20

17

The statistics in section 2 tell us that, in China, where formal financial markets are incomplete, households use family members and relatives as their primary borrowing source. Statistics also reveal that, in comparison with high-income households, low-income households rely on relatives more. Low-income households usually have less buffer stock with which to protect themselves from risk. Even for housing and education expenditures (expected expenditures), more low-income households rely on relatives to finance their needs. This is consistent with the literature that low-income households in developing countries are usually borrowing-constrained and have difficulty accessing the formal credit market (Morduch 1995). There could exist a large heterogeneity of the brothers effect among income and asset groups. Households with low incomes or low assets could potentially have a large brother effect because they are more likely to borrow and less likely to be able to access the financial markets. Therefore, they would rely more on brothers, if brothers play a role in sharing risks and extending borrowing limits. The 2002 CHIP data suggests that high-income households might have accumulated enough buffer stock savings to ensure themselves against a shock: 28% of the top-income tertile households answered that they had enough deposits in their bank to finance an emergency compared to only 8% in the low-income tertile.21 Although low-income households are more likely to borrow, they are less likely to borrow from financial institutions. It is common in China, and probably in most other financially underdeveloped countries, for banks to tend to lend money only to households with stable jobs and high pay. I use household income levels to approximate the degree of demand for brothers due to extending borrowing limits or risk-sharing. I divide households into three income groups: low, medium and high, depending on whether the household income is below 1/3, between 1/3 and 2/3, or above 2/3 of the whole income distribution in the sample. I interact the number of brothers with these income group dummies. Column 1 and 2 of Table 6 calculate the brother effect for each income group. The size of the brother effect is largest in the low-income group; the effect becomes smaller in the medium-income group, and disappears in the highest-income group. The brother’s supporting-parents effect is also ranked as the same order (column 1). I further exclude the supporting-parents’ effect in column 2 by restricting to households with no living parents. The magnitudes of the brother effects in all three income groups become smaller while the order of the coefficients does not change.22 In column 3, I further confirm the heterogeneity

21

The statistics come from the CHIP 2002 question “If your household suddenly encounters difficulty and needs 10,000 RMB immediately, where or to whom would you turn first?” 22 Note that the average age of households with no living parents in the sample is 51. The relatively large coefficient in column 2 could be affected by the age/cohort effect.

18

by dividing households by their assets instead of incomes (column 3).23 The large heterogeneity of the brothers effect among income and asset groups in Table 6 confirmed a strong risk-sharing/extending borrowing limits effect of brothers.

6

Additional Analysis

6.1

Brother versus brother-in-law

In this section, I compare the brothers effect within households by using a subsample in the CGSS 2006 data which contains information of the number of both brothers and brothers-in-law. The regression results are presented in Table 7. In each regression of Table 7, the number of brothers, siblings, brothers-in-law, siblings-in-law, and all the baseline controls are added. Column 1 suggests that brothers-in-law have a smaller effect on household’s savings rate compared to brothers. Column 2 explores the channel of supporting parents. The tiny and insignificant coefficient of the interaction term of the number of brothers-in-law indicates a zero effect of supporting parents-in-law. Column 3 investigates the brother effect in different income groups. In the lowest income group, a brother-in-law also reduces the savings rate, though the size is about two-thirds that of the effect of a brother. This suggests that brothers-in-law play a role in sharing risks / extending borrowing limits. However the effect is smaller than the brother effect.

6.2

Brothers Effect of Individuals Born after the One Child policy

In the main sample, I use individuals who were born before the One Child Policy to identify the brother effect. In this section I provide a robustness check for the main sample to show that, for people born after the One Child Policy, there still exists a strong brothers 23

Besides income, assets are also an indicator of the demand for brothers, for potentially two reasons. First, a household with sufficient assets would be less likely to borrow money from brothers because it can accommodate consumption by way of its own buffer stocks when encountering shocks. Second, assets, especially housing assets, also represent a household’s ability to access the formal financial market since assets could act as collateral when borrowing money from banks. Most bank loans in China require collateral, and the only acceptable collateral in most banks are buildings or land (Gregory & Tenev 2001; Ayyagari et.al., Cousin 2006). Only 4% of commercial loans are secured by movable assets in China. The value of housing assets is generated by subtracting mortgage (unpaid amount) from all housing value owned by a household. Ideally, total asset value could be a better indicator than housing asset level. Since CGSS does not provide total asset information, I use housing value instead. This caveat is unlikely to cause problems since the rank of household total assets and rank of housing assets are highly correlated. Using the 2002 CHIP data, I generate the three level (low, medium, high) housing value asset rank and total asset rank. These two ranks are highly correlated: the correlation coefficient is 0.77 and significant at 1% level.

19

effect. Due to the missing female problem that started to prevail after the One Child Policy, the control function approach is no longer valid for this sample (Ebenstein 2010). I use the One Child Policy Fines for unauthorized birth as an instrument for the number of brothers a household has in urban areas. The One Child Policy is enforced by a set of financial disincentives for excess fertility (Scharping 2003; Ebenstein 2010). The disincentives include denial of housing allowance and fines. Fines are calculated as a percentage of an individual’s annual income for a certain amount of years. The fines are set by local governments. Regional and temporal differences exist. The data was collected by Scharping (2003). Like Ebenstein (2010), I calculated the present value of fines which yields a single amount. The typical fine requires both parents to pay 10 percent of their annual income for 14 years, which equals to a 1.23 multiple of one year’s income of both parents.24 I control for province fixed effects and time fixed effects upon using the instrument variable method. The identified variation comes from the differences in fines in each year within a province. Due to the instrument I use, the identified effect is a local average treatment effect in the sense that it identifies the effect of compilers, i.e., households’ parents who would have one more child if fines or bonuses are low, but not otherwise. For this group the impact of the number of brothers might be large because they are people who very much like children and have strong family ties. In the estimation, the sample includes individuals born between 1979 and 1984 (22 to 27 years old in my sample) and who hold a city area residency card. We should bear in mind that the IV estimation results might not be very precise due to the small sample size (355 observations). Column 1 of Table 8 presents the result of the first stage. Fines significantly reduce the number of brothers a households has. Changing fines from zero to one year of annual income reduces on average 0.779 number of brothers of households and it is statistically significant at a 1% level. The second-stage estimation results are reported in the rest of the columns. I also provide the Anderson-Rubin 95% confidence intervals for key variables, in order to take into account the weak IV problem. The IV estimation results are consistent with the findings from the before-the-One ChildPolicy-sample: having one more brother instead of a sister of a household reduces that household savings rate (column 2). The results are also consistent with the previous findings that low-income or wage-unstable households have a larger brother effect compared to highincome or wage-stable households, which implies brothers play roles in sharing risks and

24

A 2% of annual discount rate is applied to calculate the present value of fines.

20

extending the borrowing limit.25 The estimation results could not detect the supportingparents effect (column 3). There are too few households with deceased parents, due to their young age (22-28). Note that Anderson-Rubin confidence intervals, which are robust in the presence of weak instruments, are provided in square brackets for all the coefficients with significant estimation results (Statistics 1949). The IV estimation results are consistent with the findings from the before-the-One ChildPolicy-sample. It suggests a strong brother effect for individuals born after the One Child Policy.

6.3

How the Decline in the Number of Brothers of Households Could Explain the Savings Rate Puzzle

Data from the China Statistical Year Book indicates that the average savings rate in urban areas in 1990 was 16 percentage points, where average savings rate is defined as “average saving/average disposable income”. This number becomes 24 in 2005. In this section, I calculate how much the change in the number of brothers can explain the change in the savings rate. From the estimation results of the previous sections, we know that the brother effect depends on the number of living parents and their average incomes. Thus, I divide households into nine groups based on their incomes and ages (three income groups times three age groups). The three income groups are low, medium, and high; they are equally divided from the income distribution. The three age groups are between the ages of 22-39, 40-49, and 50-60. The brother effect in each group depends on the average income, number of parents deceased, number of brothers, and the estimated brother effect in that group. The total change of average savings is the sum of change of savings in each group weighted by each group’s density. Mathematically it can be described in the following way.

4averagesaving =

XX A

\I,A + DPI,A × broDP \I,A )4broI,A f (I, A) IncI,A (broInc

(4)

I

A denotes an age group and I denotes an income group. Inc is the average income. \ is the estimated brother effect. DP is the number of parents deceased. broDP \ is broInc the estimated brother supporting parents effect. ∇bro denotes the change in the number of brothers between 1990 and 2005. f (·) is the density of each group. The statistics of these 25

Due to the small sample size, I divided households into two income groups (low and high) instead of three. For the same reason, I also group “wage very unstable” and “wage unstable” into one group.

21

variables based on the CGSS data are presented in Appendix Table 3. I suggest that decline in the number of brothers of household explained 27.6% of the increase in aggregate savings rate from 1990 to 2005 by using equation 4 and making the following assumptions. Note the estimated explained portion could be larger if sisters also behave like brothers. 1. Sisters have zero effect on household savings rate 2. The supporting-parents effect is different across age groups but the same across income groups.26 3. The number of brothers is the same across income groups for a given age group. 4. For the married couple, the brother-in-law effect is half of the brother effect (Table 7).

7

Conclusion

In contrast with the fact that China has become the world’s second largest economy, the household financial market remains underdeveloped. Households still largely rely on family members to provide a source of informal borrowing in case of negative shocks. Elderly parents still rely on their children for financial support. This paper provides evidence that having more brothers reduces the household savings rate because brothers share the risks/ extend borrowing limits, and share the cost of supporting parents. One brother reduces the savings rate by 5 percentage points. Due to the population policies introduced by the government between the 1950s and 1970s, the younger generation has very few brothers compared to the middle-aged and older generations. The decrease in the number of brothers explains 28% of the increased household savings rate in urban China between 1990 and 2005. In terms of the policy implications, the Chinese government might consider developing the household financial market as soon as possible. The baby boom generation can rely on their siblings to finance themselves. They face few hurdles while the household financial market is underdeveloped. However, current and future young generation have almost no siblings due to the One Child Policy. They lack a family-based safety net and they carry the huge burden of supporting parents. Developing the household financial market will be a necessary and urgent task. 26

When I estimate the supporting-parents effect by both age and income groups, I find that the coefficient of the interaction term of the number of brothers and the number of parents deceased has a large standard error. Alternatively, I also estimate the supporting parents effect only by income groups. Looking at how much the increased savings rate is due to the decreased number of brothers, I find almost no difference between dividing the sample by income groups or by age groups for the supporting parents part.

22

Appendix A

Data

The main data used in this paper is from the China General Social Survey (CGSS) urban areas sample. The data was collected jointly by the Hong Kong University of Science and Technology Survey Research Center and the Sociology Department of People’s University of China. It covered 24 provinces and 4 municipalities. Only three autonomous provinces were not included in the survey: Tibet, Qinghai, and Ninxia. The survey was conducted based on a probabilistic sample and stratified design. CGSS is also a part of the East Asian Social Survey program. The 2006 CGSS asks detailed questions of household characteristics, including a household’s income, expenditure, and the age of household’s head, education, and other statuses. Three other supplementary data sets that are used in this paper: China Family Panel Study (CFPS), Chinese Household Income Project (CHIP) urban area sample, and Chinese Health and Retirement Longitudinal Study (CHARLS). CFPS was conducted by the Peking University Institute of Social Science survey in Beijing, Shanghai, and Guangdong province. This study too was based on a probabilistic sample and stratified design. It is currently available for the 2008 and 2009 series. CHIP was conducted under the auspices of the Chinese Academy of Social Science. The sampling frame is a subsample of the official household survey conducted by the National Bureau of Statistics (NBS). The 2002 CHIP survey is used in this paper. CHARLS was conducted by the National School of Development (China Center for Economic Research) at Peking University. Currently only the 2008 survey is available. The provincial level data was mainly collected from the China Statistical Year Book published by the National Bureau of Statistics of China. The provincial level financial development data was collected from the Almanac of China’s Finance and Banking.

B

Proof of the identification strategy

In this appendix, I show that under the assumption that i is conditional independence of number of brothers given number of siblings; that is, i ⊥broi |sibi . α can be consistently estimated in the following equation.(For simplicity I ignore other controls.) Yi = αbroi + δ(sibi ) + i (5) where δ(sibi ) is a function of sibi . Proof:

23

Use the definition of conditional independence, we have f (i broi |sibi ) f (broi |sibi ) f (i |sibi )f (broi |sibi ) = f (broi |sibi ) = f (i |sibi )

f (i |sibi , broi ) =

where f (·) is the density function. Thus, Z E(i |sibi , broi ) =

f (i |sibi , broi )di i

Z f (i |sibi )di

= i

= E(i |sibi )

Since E(i |sibi ) is a function of sibi , let ˜ δ(sib i ) = E(i |sibi ) ˜ where δ(sib i ) is an unknown function of sibi . Assume Yi = αbroi + βsibi + i Since E(i |sibi , broi ) is not depend on broi , we have E(Yi |broi , sibi ) = αbroi + βsibi + E(i |sibi , broi ) ˜ = αbroi + βsibi + δ(sib i)

˜ Thus, α can be consistently estimated under equation 5, where δ(sibi ) = βsibi + δ(sib i ). δ(sibi ) is a control function, in order to consistently estimate α.

C

Relative effect of number of brothers

In this section, I show that if sisters have an effect on savings rate, I can still have the difference of the effect between brothers and sisters. Suppose we are interested in estimating equation 6. (For simplicity I ignore other controls). Yi = αb broi + αs sisi + i 24

(6)

where sisi is the number of sisters. αb and αs cannot be consistently estimated because broi and sisi are correlated with the error term i . For this reason, we use the control function approach explained in appendix C by adding a function of sibi into equation 6. We can have E(Yi |broi , sisi , sibi ) = αb broi + αs sisi + δ(sibi ) Due to collinearity, αb and αs cannot be estimated together. However, E(Yi |broi , sibi , sisi ) = αb broi + αs (sibi − broi ) + δ(sibi ) = (αb − αs )broi + δ 0 (sibi )

where δ 0 (sibi ) = δ(sibi ) + αs sibi Thus we can still identify the effect of brothers relative to sisters which is αb − αs .

25

References Allen, F., J. Qian, and M. Qian (2005, July). Law, Finance, and Economic Growth in China. Journal of Financial Economics 77 (1), 57–116. Arnold, F. and L. Zhaoxiang (1986). Fertility, Sex Preference, and Family Planning in China. Population and Development Review 12 (2), 221–246. Ayyagari, M., A. Demirguc-Kunt, and V. Maksimovic (2010, May). Formal Versus Informal Finance: Evidence from China. Review of Financial Studies 23 (8), 3048–3097. Banerjee, A., X. Meng, and N. Qian (2010). The Life Cycle Model and Household Savings: Micro Evidence from Urban China. Working Paper . Chamon, M., K. Liu, and E. Prasad (2010). Income Uncertainty and Household Savings in China. Working Paper . Chamon, M. D. and E. S. Prasad (2010, January). Why Are Saving Rates of Urban Households in China Rising? American Economic Journal: Macroeconomics 2 (1), 93–130. Chen, Y., Y. Ma, and K. Tang (2011). The Chinese Financial System at the Dawn of The 21st Century: An Overview. The IEB International Journal of Finance 2, 2–41. Ebenstein, A. (2010). The “Missing Girls” of China and the Unintended Consequences of the One Child Policy. The Journal OF Human Resource 45 (1), 87–115. Fafchamps, M. and D. Cox (2007). Extended Family and Kinship Networks: Economic Insights and Evolutionary Directions. Handbook of Development Economics 4 (07). Foster, A. D. and M. R. Rosenzweig (2001). Imperfect Commitment, Altruism, and the Family: Evidence from Transfer Behavior in Low- Income Rural Areas. The Review of Economics and Statistics 83 (3), 389–407. Gan, L. (2012). Findings from the China Household Finance Survey. Working Paper . Ge, S., D. T. Yang, and J. C. U. o. H. K. Zhang (2012). Population Control Policies and the Chinese Household Saving Puzzle: A Cohort Analysisy. Working Paper . Gordon, R. H. and W. Li (2003, February). Government as a Discriminating Monopolist in the Financial Market: The Case of China. Journal of Public Economics 87 (2), 283–312. Guariglia, A., X. Liu, and L. Song (2011, September). Internal Finance and Growth: Microeconometric Evidence on Chinese Firms. Journal of Development Economics 96 (1), 79–94. Horioka, C. Y. and J. Wan (2007, December). The Determinants of Household Saving in China: A Dynamic Panel Analysis of Provincial Data. Journal of Money, Credit and Banking 39 (8), 2077–2096.

26

Jacobsen, R., H. Moller, and A. Mouritsen (1999, December). Natural Variation in the human Sex Ratio. Human Reproduction (Oxford, England) 14 (12), 3120–5. Lardy, N. (2012). Interest Rate Liberalization and the International Role of the RMB. Working Paper . Lee, Y. and Z. Xiao (1998). Children’s Support for Elderly Parents in Urban and Rural China: Results from a National Survey. Journal of Cross-Cultural Gerontology 13, 39–62. Liao, L., N. Huang, and R. Yao (2010, July). Family Finances in Urban China: Evidence from a National Survey. Journal of Family and Economic Issues 31 (3), 259–279. Modigliani, F. and S. L. Cao (2004). The Chinese Saving Puzzle And Life-Cycle Hypothesis. Journal of Economic Literature 42 (1), 145–170. Morduch, J. (1995). Income Smoothing and Consumption Smoothing. Journal of Economic Perspectives 9 (3), 103–114. Rosenzweig, M. R. (1988). Risk, Implicit Contracts and the Family in Rural Areas of LowIncome Countries. The Economic Journal 98 (393), 1148–1170. Scharping, T. (2003). Birth Control in China 19492000. New York, Routledge Curzon. Song, Z., K. Storesletten, and F. Zilibotti (2011). Growing Like China. American Economics Review 101 (February), 196–233. Statistics, M. (1949). Ectimation of the Parameters of a Single Equation in a Complete System of Stochastic Equations. Annals of Mathematical Statistics 20 (1), 46–63. Tien, H. Y. (1964). The Demographic Significance of Organized Population Transfers in Communist China. Demography 1 (1), 220–226. Wei, S.-j. and X. Zhang (2011). The Competitive Saving Motive: Evidence from Rising Sex Ratios and Savings Rates in China. Journal of Political Economy 119 (3), 511–564. Yang, D. T. (2012, November). Aggregate Savings and External Imbalances in China. Journal of Economic Perspectives 26 (4), 125–146. Yao, R., F. Wang, R. O. Weagley, and L. Liao (2011, September). Household Saving Motives: Comparing American and Chinese Consumers. Family and Consumer Sciences Research Journal 40 (1), 28–44. Yip, W. and W. C. Hsiao (2008). The Chinese Health System at a Crossroads. Health affairs (Project Hope) 27 (2), 460–8. Yu, L. C., Y. Yu, P. K. Mansfield, S. Gender, and N. Mar (1990). Gender and Changes in Support of Parents in China: Implications for the One-Child Policy. Gender and Society 4 (1), 83–89. Zhou, W. (2013). 27

Figure 1: Sources for Borrowing Money: Self-Reports of Borrowing Resource if One Encounters a Negative Shock (Percentage of Respondents)

Note: The above results are calculated by the author based on a question in the Chinese Household Income Project 2002: ”If your household encountered an abrupt difficulty and needed 10,000 RMB immediately, who (where) would you turn to first?” Sample size: 6779.

28

Figure 2: Number of Brothers and Sisters by Individuals’ Birth Year

Data source: China General Social Survey 2006, 2008. Note: Sample is restricted to individuals who live in urban areas, have urban area residence cards.

29

Figure 3: Number of Brothers and Household Savings Rate in Urban Areas

Note: 1. The number of siblings are restricted to individuals aged 20-60. 2. Siblings data source: China General Social Survey 2006, 2008. Death rates are used in order to compute the number of siblings in early years. 3. Saving rate is defined as 1-living expenditure/disposable income. 4. Saving rate and death rate data source: China Statistical Yearbook.

30

Figure 4: Age Profile Household Savings Rate by Number of Brothers and Sisters

Note: 1. Data source : China General Social Survey 2006, 2008 2. Sample is restricted to persons who live in urban areas and have urban area residence cards.

31

Figure 5: Average Household Savings Rate and Average Household Income by Number of Siblings and Number of Brothers

Note: 1. Data source : China General Social Survey 2006, 2008 2. Sample is restricted to persons who live in urban areas and have urban area residence cards.

32

Figure 6: Age Profile Household Savings Rate by Number of Brothers and Sisters - Households with No Living Parents

Note: 1. Data source : China General Social Survey 2006, 2008 2. Sample is restricted to persons who live in urban areas and have urban area residence cards.

33

Table 1: Fraction of Male Siblings by Total Number of Siblings Number of Siblings 1 2 3 4 or more

Obs 572 846 756 1085

Fraction of Male 0.52 0.52 0.49 0.48

[95% Conf. Interval] 0.50 0.55 0.50 0.53 0.48 0.51 0.47 0.49

Note: 1. Data source : China General Social Survey 2006, 2008 2. Sample is restricted to persons who live in urban areas, have urban area residence cards and were born between 1946-1978.

34

Table 2: Test of Random Assignment of the Number of Brothers Conditional on the Number of Siblings Brothers Siblings Mother Education

∗∗∗

-0.041

Dependent Variable Brothers 0.484∗∗∗

Fraction -0.009∗∗

(0.012)

(0.005)

-0.008

-0.003

(0.009)

(0.006)

(0.003)

Father Education

-0.009

0.009

0.004

(0.009)

(0.006)

(0.003)

Mother Residence Card Status

0.048

0.011

-0.030

(0.111)

(0.082)

(0.03)

-0.076

0.007

Father Residence Card Status

-0.227

∗

(0.125)

(0.096)

(0.036)

Mother Company Type

-0.060

-0.041

-0.026

(0.059)

(0.044)

(0.02)

Father Company Type

0.073

-0.007

-0.005

(0.048)

(0.035)

(0.015)

Yes Yes Yes 2611 10.46∗∗∗

Yes Yes Yes 2611 1.37

Yes Yes Yes 2386 1.45

Parents Communist Party Parents OCC Skill Levels Parents OCC Dummies Obs. Wald statistics

Note: 1. The Wald test examines the joint significance of all the regressors in column 1. In column 2 and 3, number of siblings is not included in the Wald test; all other regressors are included. 2. Sample is restricted to persons who live in urban areas, have urban area residence cards and were born between 1946-1978. 3. Standard errors are clustered at province level. 4. *** p