Who Bequeaths, Who Rules: How the Right to Bequeath Affects Resource Allocation Within Household

Transcription

1 Who Bequeaths, Who Rules: How the Right to Bequeath Affects Resource Allocation Within Household Li Han and Xinzheng Shi July 2018 Abstract We examine the influence of changes in status-bequest rules on bargaining outcomes and intergenerational investments in urban China, where a reform in 1998 ended women s monopoly on the transmission of residency permits (hukou) to children. We find that the allocation of resources within pre-existing couples in which both spouses are local hukou holders responded to this reform in favor of men, at the cost of femalefavored consumption and investment in children. Moreover, the adjustment in intergenerational investment is conditioned on the different effects of the reform on the future marriage market by gender. This response was stronger in settings in which local men have better prospects of marrying migrants after divorce. Keywords: hukou; bequest; resource allocation within household; urban China. Li Han (lihan@ust.hk) is affiliated with the Hong Kong University of Science and Technology. Xinzheng Shi (corresponding author; shixzh@sem.tsinghua.edu.cn) is affiliated with Tsinghua University. We gratefully acknowledge the China National Bureau of Statistics for providing the Urban Household Survey data. We thank Andrew Foster and two anonymous referees for their insightful comments. We also thank seminar participants at Lingnan University, the Hong Kong University of Science and Technology, Peking University, Renmin University, and Shanghai University of Finance and Economics for helpful comments. Xinzheng Shi acknowledges financial support from the National Natural Science Foundation of China (Project ID ). All remaining errors are our own. 1

2 1 Introduction Forms of family resource allocation, such as consumption, investment in education, and saving for old age, constitute the micro-foundations of an economy. Economists have long acknowledged that husbands and wives do not always reach a consensus, and that their interactions in part determine family resource allocation. There is abundant evidence that spouses with greater economic power play a greater role in intrahousehold decision-making. As has often been argued, individuals with greater economic power have direct control over more resources and thus enjoy greater bargaining power within marriage. A corollary of this argument is that women can obtain a status within marriage equal to that of their husbands by gaining control of more economic resources. However, this mechanism is not as self-evident as it is often portrayed. Consider Becker s (1973, 1974) famous account of marriage: Nothing distinguishes married households more from single households... than the presence, even indirectly, of children. If children (or prospective children) provide a foundation for marriage, individuals capable of passing on more resources or income-enhancing traits to their children are likely to be more sought after in the remarriage market if divorce is a credible threat. To keep those individuals from leaving the marriage, their spouses are likely to transfer more resources to them. The availability of outside options provides those individuals with more bargaining chips within a marriage. In other words, the more one can give to one s children, the more one can get from one s partner. 1 Because the transferability of these resources/traits is likely to be affected by public policies or legal restrictions, it is important to explore the extent to which the ability to transfer resources or other income-enhancing traits to one s children affects within-marriage bargaining outcomes and intergenerational investment. Empirically, it is often difficult to distinguish the ability to transfer resources to children from direct control over resources. Fortunately, China offers a useful setting for empirical 1 For example, using Indian data, Foster and Rosenzweig(2001) find that if literate mothers contribute to the production of sons s human capital, the higher future agricultural technology will induce a higher current demand for literate wives in the marriage market. 2

3 investigation, because changes in the government s rules concerning status bequests affect individuals ability to transfer resources to the next generation, but are not associated with direct control over resources or outside options in the labor market. We therefore focus on the formal right to bequeath status and examine the effects of changes in this right on intrahousehold allocation and intergenerational investment. Our study is conducted in urban China, where a clearly defined rule governs the right to bequeath a very important form of hereditary legal status: hukou (household registration, or the class system of residency permits). Similar to caste in India, hereditary hukou entrenches social strata, emphasizing social distinctions between those with different types of residency status. Without local hukou, employment opportunities, access to public services such as inexpensive education and healthcare, and sometimes even the right to purchase cars and houses in the local area, are greatly restricted. Therefore, migrant workers in cities who have no local hukou are considered inferior to locals in the marriage market. In particular, the more developed a city is, the higher the value of local hukou therein. Due to enormous regional disparities in China, local hukou holders in economically advanced cities are considered high-status individuals. Thus, it is not surprising that local spouses are preferred over migrants, all else being equal. For convenience, we define those marriages where both husband and wife hold local hukou as local-local marriages. Changes in hukou status are tightly controlled. Ordinary people cannot alter hukou status through relocation, marriage, etc. However, a policy change in 1998 significantly affected the right to bequeath hukou, creating a context of particular relevance to this study. Before 1998, hukou inheritance was matrilineal, i.e., children were granted their mother s hukou. The law was amended in August 1998 to grant men the same right as women to transmit hukou to their children. We investigate the effects of the 1998 policy change on intrahousehold resource allocation in local-local marriages, the most common form of marriage in urban China. The 1998 reform did not directly affect the labor market status of current couples. Thus, this setting is well suited to our study, as it enables us to eliminate the mechanism of direct control over resources and isolate the mechanism of potential transfer to children. Furthermore, it is plausible that this policy change differentially affected the divorce option 3

4 for men and women. After local husbands have been granted the right to bequeath their hukou to their children, they will become more popular in the remarriage market and their utility will be higher than before in the case of marrying a migrant. In contrast, local women will face a tighter remarriage market when the competition for a local spouse is fiercer. If intrahousehold allocation responds to changes in the divorce option, we would expect to observe that local wives increase marital transfers to their local husbands to keep them from leaving the marriage. In other words, we would expect to observe the increase of malefavored consumption and the decrease of female-favored consumption. If women care more about children than men, we would also expect investment in children to decrease. Furthermore, to the extent that the reform has similar effects on the future marriage market for the next generation, local-local couples are expected to adjust their investment in sons and daughters differently. For daughters, the adjustment mainly arises from two indirect channels: the income effect and the competition effect. The income effect suggests that investment in daughters will decrease because of the worsening position of daughters in future marriages. Lower bargaining power may translate into a lower share of marital surplus and hence discourage pre-marital investment from parents. However, the competition effect suggests that investment in daughters will increase, because women will face fiercer competition from migrant women in the marriage market. Therefore, parents have stronger incentives to invest in daughters so that they can gain a competitive margin in the future marriage market. If the competition effect is greater/weaker than the income effect, the overall investment in daughters may increase/decrease. For sons, these two indirect effects work in the opposite direction: the income effect suggests an increase of investment in sons, while the competition effect suggests a decrease of investment in sons. Aside from these two effects, the reform will affect investment in sons through a direct channel, which we call the cross-marriage channel. Sons expected benefits from marrying a migrant are increased by the reform, because their offspring can have local hukou. As cross-hukou marriages become more acceptable to local men, it weakens their incentive for premarital investment to compete for local spouses. Given that this cross-marriage effect only applies to local men instead of local women, we would expect that the decrease in investment would be larger for sons than 4

5 for daughters. We explore how resource allocation and intergenerational investment within local-local marriages respond to this reform using a difference-in-differences (DID) strategy. We exploit the fact that policy intensity differs between cities, because local men s prospects of remarrying migrants, which are proxied by the density of female migrants, differ between cities. We thus compare changes in the allocation of resources in marriages in cities with higher densities of female migrants with that in cities with lower densities of female migrants. The first dimension of difference in this DID model is the 1998 policy change, which improves the divorce option for local men in the remarriage market. The second difference is the variation between cities in the density of female migrants who are potential migrant brides (instrumented by the 10-year lagged density of female migrants in each city). The policy change is expected to have caused greater improvement in the divorce option for local urban men in cities where more potential migrant brides are available. 2 Our data are mainly drawn from China s Urban Household Survey (UHS). The survey provides not only individual-level data but also detailed information on household expenditures. Our outcome variables include gender-specific consumption, including expenditures on both female-favored goods, namely women s clothes and cosmetics, and male-favored goods, namely men s clothes, cigarettes, and alcohol. Our outcome variables also include intergenerational investments in children, including expenditures on children s clothes and education. Our findings support the hypothesis that the 1998 change in China s hukou inheritance rules reduced the bargaining power of local women within households. We find that for cities with a one standard deviation higher proportion of female migrants, the policy change significantly reduced not only expenditure on women-related consumption (reducing expenditures 2 Because childrens hukou status depended on the mothers hukou before the reform, local men did not prefer to marry migrant women. After the reform, the childrens hukou status can follow that of their fathers, and the local men became more willing to marry migrant women. In other words, the reform increased the local mens options for marriage. The more migrant women in the local city, the larger the local men s outside option set. 5

6 on women s clothes by 8.4% and cosmetics by 10.7%), but also children-related consumption (reducing expenditures on children s clothes by 17.3% and children s education by 9.9%). In contrast, for cities with a one standard deviation higher proportion of female migrants, the same change led to 11.9%, 10.8%, and 9.1% increases in expenditures on men s clothes, cigarettes, and alcohol, respectively. We further explore the difference in the adjustment of investment in children by child s gender. We find that the decrease in children-related expenditures mainly arose from the decrease in investment in sons. The reform impact on investment in daughters is positive and statistically significant. This result is consistent with the explanation that the competition effect dominates the income effect, and that the cross-marriage effect for sons drives down investment in sons. We then investigate whether the mechanism for policy effects on intrahousehold bargaining is indeed the improvement in local men s divorce option. Two key elements underlie arguments for the mechanism: first, divorce is a credible threat; second, people care about the hukou status of children. We conduct two tests to corroborate these points. We first show that divorce rates are higher in cities with higher female migrant densities, which suggests that divorce is a more credible threat in such places. Then, we construct a test based on differences between groups of couples in terms of the likelihood of having another child. We compare the effects of the policy on younger and older couples. The older individuals are, the less likely they are to have another child, even after divorce. We find that the policy effect is stronger for young couples (below age 40). This finding further corroborates our hypothesis that the potential transfer of status to children significantly affects intrahousehold allocation. One underpinning for our empirical strategy is that the density of female migrants is sensible for local men. To explore this argument, we further show that the policy effect is stronger if local men are employed in industries with higher female migrant densities within cities. This finding suggests that the intrahousehold allocation of resources is more responsive to the 1998 policy change if husbands have higher chances of meeting female migrants, which lends more support to our empirical strategy. 6

7 Our study makes several contributions. First, most researchers emphasize gender bias in the right to inherit ancestral property, names, or status. 3 Although the right to inherit undoubtedly affects individuals well-being, another defining feature of any inheritance system is the right to bequeath, i.e., the right to pass on one s status or property to the next generation. The right to bequeath is not always consistent with the right to inherit, and has received much less attention from researchers. In examining the right to bequeath, we provide a new perspective on the effects of inheritance regulations on gender relations. To the best of our knowledge, our study is the first to emphasize the role of status-bequest rules in intrahousehold bargaining. The rules governing property bequests may have a similar influence. Second, we disentangle the two mechanisms by which individuals who possess more resources gain a better standing in their households: direct control over resources or the potential transfer of resources to children. Distinguishing these two mechanisms helps us understand how policies that aim at increasing intergenerational mobility can affect equality within the current generation. A more general implication of our study is that a decrease in intergenerational mobility helps secure the within-marriage bargaining position of individuals who can pass resource-enhancing traits to the next generation, and vice versa. Although some previous studies (e.g., Anderson, 2003; Li and Wu, 2011) point out the importance of children for marital outcomes, they do not treat transfers to children as a distinct mechanism in determining intrahousehold resource allocation. Third, our results support the collective framework of household decisions. In particular, we provide evidence that extra-household environmental factors matter for the allocation of resources within the household (Browning et al., 1994). Finally, our study also speaks to the literature on premarital investment. We provide evidence that parents condition investment in children on the expected future marriage market. The remainder of this paper is organized as follows. Section 2 reviews the relevant 3 For example, Deininger et al. (2013) find that strengthening women s inheritance rights improves their bargaining position within their families and increases their financial independence. Roy (2015) finds that strengthening mothers inheritance rights increases their daughters level of schooling. Jayachandran (2015) provides a more detailed survey. 7

8 literature, and background information is provided in Section 3. Section 4 presents the theoretical framework before the data are described in Section 5. Section 6 introduces the empirical strategy. Section 7 reports the main results and the heterogeneous analysis. In Section 8, we explore the mechanisms through which the policy effects arise. Finally, Section 9 concludes the paper. 2 Relevant Literature Our paper contributes to the literature on the distribution of bargaining power within households. As pointed out in Browning and Chiappori (1998), within a collective utility framework of household decisions, household demands are sensitive not only to the allocation of resources within the household, but also, more generally, to any environmental variable that may influence the decision process (or distribution factor, as defined in Browning et al. (1994)). The empirical research in this vein can be largely classified into two categories. Research in the first category focuses on how the within-couple allocation of expenditures is affected by changes in gender differences with respect to various types of economic values, such as income and asset ownership. Duflo (2003) finds that pensions received by women in South Africa have a significant positive influence on the anthropometric status of children, especially girls, whereas pensions received by men have no similar effects. Anderson and Eswaran (2009) find that relative contribution to earned income plays a particularly important role in empowering women in Bangladesh, and Luke and Munshi (2011) report similar results in India. Wang (2014) finds that transferring household property rights to men increases household expenditure on certain goods favored by men and time spent by women on chores, whereas transferring property rights to women decreases household expenditure on certain male-favored goods. Studies of the influence of gendered inheritance rights on the intrahousehold allocation of resources also belong to this category (e.g., Deininger et al., 2013; Roy, 2015). Our paper falls into the second category of research on within-household bargaining power, which explores the effect of extra-household environmental factors or distribution 8

9 factors, i.e., variables that can affect intrahousehold decision-making without influencing individual preferences or joint consumption (Browning and Chiappori, 1998). Commonly used examples of such factors include the sex ratio and divorce legislation. Chiappori, Fortin, and Lacroix (2002) use a structural model to investigate the effects of various distribution factors on within-marriage transfers. Rangel (2006) finds that the extension of alimony rights in Brazil reduced time spent on housework by female members of cohabiting couples, and also reduced their labor supply. Edlund, Liu, and Liu (2013) find that the importation of foreign brides to Taiwan has increased domestic brides fertility. Rasul (2008) shows that if couples bargain without commitment, the threat point in marital bargaining and the distribution of bargaining power determines the influence of each spouse s fertility preference on fertility outcomes. Sun and Zhao (2016) find that the reduction in divorce costs under China s new divorce law has resulted in a higher divorce rate and fewer sex-selective abortions. Lafortune et al. (2017) analyze the impact of a reform granting alimony rights to cohabiting couples in Canada and find that the reform benefits women in existing couples. However, for couples not yet formed, they find that the reform generates offsetting intrahousehold transfers and then lowers the benefits for women. Majlesi (2016) explores a novel direction, i.e., how outside options in the labor market affect women s decision-making power within households. In this paper, we show that the rule governing the right to bequeath status is an important distribution factor. The novelty of the policy experiment in our study is that household resources are not changed. The reform only affects the divorce option of existing same-status couples in the remarriage market and thereby affects within-couple resource allocation. Our finding also speaks to the literature on premarital investment. Theories show that human capital investment may respond not only to labor market returns but also to marriage market conditions (e.g., Chiappori, Iyigun, and Weiss, 2009; Browning, Chiappori, and Weiss, 2014). Lafortune (2013) theoretically outlines and empirically tests how pre-marital investment in men and women responds differently to changes in the sex ratio. Our result supports her findings. 9

10 3 Hukou and Marriages in China The hukou system has been strictly enforced to control internal migration in China since the late 1950s (Cheng and Selden, 1994). This system identifies each person as a resident of a specific area and determines the allocation of public services and economic resources. Those without local hukou have no access to local public services such as inexpensive education and healthcare, and limited job opportunities in the formal sector. Their right to purchase houses and cars locally may also be restricted. Given the enormous economic disparity between Chinese regions, individuals from economically prosperous cities with local hukou are effectively high-status citizens. Due to its entrenchment of social strata, the hukou system is often regarded as the Chinese version of a caste system. Hukou is inherited from one s parents. Ordinary individuals are rarely permitted to change their hukou status. To limit migration through marriage (typically the marriage of migrant women to higher-status men), the government imposed the additional requirement that only mothers can pass on their hukou to their children. The transmission of hukou status was matrilineal until August 1998, when the national government relaxed its restriction on hukou inheritance to allow both men and women to pass on hukou to their children. Under the new policy, children born after August 1998 can inherit hukou from either parent. Minors born before this policy change whose hukou is not in the same place as their fathers can request a change of status. The government has committed itself to solving this problem gradually in response to individual requests. 4 However, the process of changing one s hukou is usually tedious and costly. Changes in hukou status through marriage or relocation remain tightly controlled. Migrants without local hukou are also at a disadvantage in the marriage market. The rate of inter-hukou marriage is low. Previous studies have shown that only 4% to 6% of local residents who married between 1980 and 2000 had spouses who were born in another province. It is perhaps not coincidental that unmarried local urban men aged between 30 and 39 significantly outnumbered unmarried local urban women. Most local urban men 4 Source: State Council Order (1998) No

11 would rather undertake a prolonged search than marry a migrant woman. This tendency began to shift after the 1998 policy change. Han, Li, and Zhao (2015) show that, among couples marrying for the first time, the likelihood of the marriage involving a local man and a migrant woman increased by nearly one half within the two years after the 1998 change in the hukou-inheritance rule. Although the overall rate of inter-hukou marriage remains modest, this sharp short-term increase illustrates that migrant women are close substitutes for at least some local women in the marriage market if their prospective children do not have to inherit their migrant hukou status. Similarly, migrant women are more likely to be potential brides for divorced local urban men, as now they do not need to worry about the hukou status of their children with migrants. Moreover, as the one-child policy allowed remarried couples to have one more child if one party was childless and the other party had no more than one child or had never married before, 5 the divorce option of most husbands in our sample was improved by the policy change in Theoretical Framework 4.1 Basic Setup We utilize the framework developed by Browning and Chiappori (1998) to illustrate the channel through which the 1998 reform affects the intrahousehold allocation of resources of local-local households. For simplicity, we denote the husband s utility within marriage by 5 The regulations regarding remarried couples vary somewhat across provinces. In provinces where the policy was most strictly implemented, if one party had never married before entering the marriage and the other party had no more than one child, they were allowed to have one more child. In other provinces, as long as the number of children from previous marriages was no more than one, remarried couples were allowed to have one more child. Table 1 in the Appendix summarizes the relevant policies as they existed in our sample provinces in % of households in our sample had no child and 71% had only one child. 43% of the female migrants in our sample had no child and 35% had only one child. 11

12 U H and the wife s utility within marriage by U W : U H = α H ln(x H ) + (1 α H )lng U W = α W ln(x W ) + (1 α W )lng where x H and x W are the private consumption of the husband and the wife, respectively, and G is the consumption of household public goods such as expenditures on children s clothes and education. Note that α H and α W are the weights attached to private consumption by the husband and wife, respectively. If α H > α W, the husband cares more about his private consumption relative to public goods (expenditures on children, in our case) than the wife does. Without loss of generality, we assume that the wife in the family maximizes U W (x W, G) subject to: a) the husband achieving at least a given utility U H U H and b) the resource constraint x H + x W + pg Y ; where Y is family income and p is the price of G. The price of private consumption is normalized to one. It is equivalent to maximizing U W (x W, G) + µu H (x H, G) subject to constraint (b), where µ is the Lagrange multiplier associated with constraint (a). The optimal x H, x W, G can be written in terms of µ: x H = µα HY 1 + µ ; x W = α W Y 1 + µ ; pg = θ Y 1 + µ (1) where θ = [µ(1 α H ) + (1 α W )]. It is easy to see that as µ increases, x H increases while x W decreases; and G decreases if α H > α W, and vice versa. Note that µ = du W (x W,G ) du, H that is, µ is the wife s marginal disutility from an increase in U H, and µ increases with U H. 7 Thus, as the husband s divorce option U H the private consumption allocated to the husband x H improves, to keep him in the marriage, has to increase at the cost of the wife s private consumption x W. Expenditure on public goods, such as children s clothes or education, decreases with µ if the wife cares about the child more than the husband does (1 α W > 1 α H ). 7 Intuitively, to achieve a higher utility level of their husbands, wives need to give up their utility, and at the higher utility level of their husbands, wives need to give up more to improve husbands utility further. Therefore, µ increases with U H. 12

13 4.2 Impacts on Intrahousehold Allocation Now let us consider how the 1998 reform affects the allocation of resources within the household. It is worth noting that the 1998 reform directly changes neither family income nor the relative prices of consumption goods within the household. This reform only changes the post-divorce outside option for the husband, U H. We characterize U H in a simple form: U H = (1 ρ)ω L + ρ(ω M + νγ) where ω L and ω M represent the utility level that divorced men would achieve without the 1998 reform by marrying a local woman and a migrant woman, respectively; ρ denotes the likelihood of marrying a migrant woman; ν is the value attached to the child in a marriage with a migrant woman, ν > 0 ; γ is the hukou status of the child from the marriage with the migrant woman; thus, γ {0, 1}, and γ changes after the 1998 reform. This characterization of U H demonstrates that the 1998 reform improves urban men s divorce option because their child from a marriage with a migrant woman can inherit urban hukou status. Because µ increases with x W. We thus make the following predictions: U H, we expect that the reform will increase x H and decrease Prediction 1. After the 1998 reform, the private consumption of the husband in the local-local marriage will increase, while the private consumption of the wife will decrease. Note that a greater ρ is associated with a larger improvement in U H. That is, local men who are more likely to marry migrant women after divorce experience a greater improvement in the divorce option. Thus, we expect that their within-marriage consumption is more responsive to the reform. This prediction can be stated as follows. Prediction 2. Ceteris paribus, an urban man with a greater likelihood of marrying a migrant woman after divorce is more likely to respond to the 1998 reform. Household consumption allocated to this type of man is more likely to increase compared to those with a small likelihood of marrying migrants. Note that the likelihood of marrying migrants ρ tends to be positively correlated with the chances of meeting a migrant, which is affected by the density of migrants in the city. Our empirical strategy will be based on this prediction. 13

14 4.3 Impacts on Intergenerational Investment To the extent that the reform will also affect future marriage market conditions for children, forward-looking parents may adjust their investment in their children. Given that the reform affects two genders in different ways, we would expect the adjustment of parental investment in children to differ by the gender of the child. We extend our basic framework to incorporate this wrinkle and derive a testable hypothesis. We modify the basic setup to allow risky investment in children. Therefore the parent s utility function becomes U j = α j ln(x j )+(1 α j )E(lnG), j {H, W }. Now we assume that G is the child s future benefit, which depends on parental investment i. 8 The household is thus subject to x H + x W + i = Y. Without loss of generality, we consider that G is given by the sum of the child s future consumption c. The child s consumption c depends on investment in this child i and the investment of the child s future spouse. The child can marry a local spouse with investment level i, or the spouse can come from the pool of migrants. Marrying a migrant incurs a utility cost for each gender k, denoted by τ k, k {m, f}. The utility cost for males τ m contains the cost of having offspring without local hukou. As the reform allows fathers to pass their local hukou to their children, τ m will decrease, i.e., τ m for the future marriage market is expected to decrease when γ in Section 4.2, the hukou status of child of local men from marriage with migrants, is 1. Once married, individuals bargain and obtain a share of family output that may depend on marriage market conditions. Denote males and females consumption when married to locals as c ml and c fl, respectively, and if married to migrants as c mm and c fm. Assume that these consumption functions are increasing in both one s own investment level and spousal investment, are concave in one s own investment level, and display complementarity in investment levels. Our result in the last subsection implies that c in local-local marriages could be affected by the cost of marrying migrants τ m, τ f, c fl / τ m > 0, c ml / τ m < 8 For simplicity, we assume away parents financial transfer to children. It does not qualitatively change the result. 14

15 0. That is, the increase in the utility cost of marrying migrants would be favorable for females and harmful for males in local-local marriages. We also expect the effect of τ m on consumption in local-local marriages to be smaller for highly educated local individuals, i.e., 2 c ml / τ m i > 0, and 2 c fl / τ m i < 0. Assume one s probability of finding a match within the local marriage market is related (but not perfectly) to one s investment level. Let us denote by p k (i, i ) the probability that an individual of gender k with investment level i finds a local spouse with investment level i, and by p k (i) the probability that s/he finds any local spouse, where p k (i) = Σ i p k (i, i ). The probabilities of marrying locals p(i, i ) may be a decreasing function of the density of migrants. The expected utility from investing in their child Eu(G) is as follows: Eu(G k ) = p k (i)lnc kl + (1 p k (i))ln(c km τ k ) (2) It is worth noting that τ m only enters the consumption c fl for daughters, while τ m enters both c ml and c mm for sons. The change in the utility cost affects sons future benefits in two ways: the first is by indirectly changing their bargaining position in local-local marriages through changing their position in the marriage market (c ml ); the second is by directly changing their utility level from marrying migrant women (c mm ). In contrast, for daughters, τ m affects their utility only through the indirect channel the marriage market channel. Plugging equation (2) back into the objective function of the household, the choices of i, x H and x W must satisfy α H = µα H = θ Eu(Gk ) (3) x H x H i Solving the model, we can demonstrate the response of the investment in daughter i f and in son i m to the change in utility cost of local men marrying migrants τ m through different channels, according to the following equations: i m τ θσ m i (ã cml τ m i f τ θσ m i (a cfl τ + b[ m c ml i m c ml + b[ τ ]) m c fl i f c fl ]) (4) τ m + θ p(im ) i m 1 (c mm τ m ) 2 + θ(1 p(im )) cmm / i m (c mm τ m ) 2 (5) 15

16 where a, b, ã, and b are all positive. The derivation of these two equations is presented in Part A in the Online Appendix. Equation (4) illustrates that changes in τ m affect parental investment in daughters in two ways. The first is captured by the first element in the bracket of equation (4) (i.e., c fl τ m ), which we name the income effect. This effect is similar to the result from our baseline model. The more local males lose from marrying migrants (higher τ m ), the larger share of marital surplus local females can get from marriages with local males, i.e., cfl τ m > 0. This income effect encourages local parents to invest more in daughters in a marriage market more favorable to local females, because they can obtain more marital surplus. The second effect, which we define as the competition effect, is captured by the second element in the bracket of equation (4) (i.e., c fl i f c fl τ m c fl i f c fl τ m ). As shown in Part A in the Online Appendix, < 0. Intuitively, it means that with the increase of local males loss from marrying migrants (higher τ m ), the return rate on parental investment on daughters decreases since local women in marriages with local men face less competition from migrant women. Equation (5) shows three roles that τ m plays in affecting parental investment in sons. The first two, the income and competition effects, are similar to those for daughters but have the opposite signs. The income effect is captured by the first element in the bracket in equation (5) (i.e., cml τ m ). An increase in τ m weakens local males position in local-local marriages and induces a lower level of investment. The competition effect suggests that with a higher τ m, the return rate on parental investment increases because local males in the marriage face more competition from other local males in the marriage market. Furthermore, in addition to these two indirect effects, τ m affects sons expected benefits directly by affecting their marital gain from marriage with migrant women. This third effect, which we term the cross-marriage effect, is captured by the last two elements in equation (5). Because higher parental investment i m can increase local male s probability to marry with local women (i.e., p(i m ) i m c mm i m > 0) and their share of marital surplus in the marriage with migrant women (i.e., > 0), therefore, when local male s utility loss from marrying migrant women τ m is higher, parents will increase investment to offset this loss. 16

17 Given that the cross-marriage effect only applies to local males, we would expect that a policy change that reduces τ m would induce lower investment in sons relative to daughters. Whether investment in daughters will increase or decrease depends on whether or not the income effect dominates the competition effect. Now let us consider the impact of the 1998 reform on intergenerational investment. The reform affects two parameters in our framework: first by increasing fathers outside option U H and second by reducing τ m. From the last subsection, we know an increase in U H will result in lower investment levels in both boys and girls if wives care more for their children than their husbands. Combining this bargaining effect with the roles of τ m as we discuss above, we can make the following prediction: Prediction 3 : After the 1998 reform, parents in local-local marriages will reduce their investment in sons. In contrast, parental investment in daughters will not decrease as much as in sons. If the competition incentive is sufficiently strong, investment in daughters may even increase. 5 Data Our principal data are drawn from the UHS conducted by the National Bureau of Statistics (NBS) in China. The UHS only sampled households with local hukou in urban areas of all Chinese provinces before 2002, and therefore all migrants were excluded from the sample. The survey was expanded to include migrant households in urban areas after 2002 (Ge and Yang, 2014). The NBS uses a probabilistic and stratified multistage sampling methodology to select households. The sample is a rotating panel in which one-third of the sample is replaced each year, and hence the entire sample is changed every three years. Therefore, the data are essentially repeated cross-sections. The UHS not only contains demographic and income information for every member of the family, but also collects detailed information about household expenditure. 9 Unfortunately, it contains no information on assets. During sampling, households are asked to keep daily records of their income and expenditures, 9 Table 2 in the Appendix lists the categories. 17

18 which are collected by the NBS every three months. 10 The data we can get access to are aggregated at the year level and collected in 48 cities in 9 Chinese provinces representing a variety of geographical locations and economic conditions, including Beijing, Liaoning, Zhejiang, Anhui, Hubei, Guangdong, Sichuan, Shaanxi, and Gansu. Our main measure of female migrant density is the share of female migrants between 20 and 45 years old in the same female age cohort in each city. This measure is constructed for the years 1990 and 2000 using a 1% sample of the 1990 population census and a 0.095% sample of the 2000 population census. Both waves of the population census ask respondents about their hukou location in the census year and their province of residence five years prior. A migrant is defined as a person whose hukou is not in the place of residence in the census year and who was not living in the local province five years before. As no other reliable data are available on the number of migrants in each city in each year, the population census is the most reliable source for computing the density of migrants. However, one disadvantage of using census data is that the census is conducted every 10 years, so we only have information on migrant density for the census years 1990 and We drop Shenzhen City from the sample because it had outlier level female migrant densities in both 1990 and 2000 (see Figure 1 in the Appendix). Our final sample therefore includes 47 cities. The plot of female migrant density in 2000 versus female migrant density in 1990 for these 47 cities is shown in Figure 2 in the Appendix, from which we can see that migrant densities vary among cities. We use data from 1997 and 1999 in the first instance. Data from 1995, 1996, 2000, and 2001 are used in robustness checks. We do not use the 1998 data because the hukou reform took place in August 1998 and therefore those data are a mixture of pre- and post-reform information. Because we investigate the impacts of the policy change on intra-household resource allocation, we therefore need information about both husbands and wives. We thus restrict our sample to households for which we have information on both the husband and wife. 11 In total, 29,889 households from six years are 10 For in-kind income, households only need to keep monthly records and NBS collects this information every year. 11 About 95% of the original households include information on both the husband and wife, meaning that 18

19 retained in the sample, all of which have local hukou according to the sampling design of the UHS. We also compile a city-level data set containing information on economic conditions. We obtain data on various city-level macroeconomic variables, including GDP per capita, GDP growth rate, share of GDP in the primary, secondary, and tertiary industries, and average wages, from the Chinese City Statistical Yearbook ( excluding 1998). Additionally, we use the UHS data to calculate the share of employees in state-owned enterprises (SOEs) in each city in each year. We also calculate city-level divorce rates using 1990 and 2000 census data. 12 Table 1 presents summary statistics for the variables in our analysis. All monetary values are adjusted using provincial consumer price indices. Panel A reports means and standard deviations for household-level variables. The average family size is 3.14 and the average number of children is 0.99, which suggests that most households in our sample are core families. Approximately 48% of families have sons. The average expenditure per capita is 5,771 yuan. As it is difficult to discern which gender is the main consumer of most goods, we focus on goods that are most likely to be gender-specific. Two of the categories of expenditure are female-favored goods: women s clothes and cosmetics. The share of expenditure on women s clothes is 0.037, and the share of expenditure on cosmetics is Male-favored goods are men s clothes, cigarettes, and alcohol. The share of expenditure on men s clothes is The shares of expenditures on cigarette and alcohol are and 0.014, respectively. We also examine variables measuring intergenerational investment, including expenditures on children s clothes and children s education. The average values of these two variables are and 0.050, respectively. 13 In addition, we examine genderneutral expenditures, namely overall food expenditure and expenditure on specific food items selection effects are not a particular concern. 12 There is a question in both censuses about marital status, and the divorce rate is defined as the ratio of divorced individuals to all individuals who have ever been married (including currently married, divorced, and widowed). 13 We do not investigate expenditures on children s health, because UHS only contains information on health expenditure at the household level. 19

20 (rice, pork, and vegetables). The share of expenditure on food is 0.488, and the shares of expenditure on rice, pork, and vegetables are 0.031, 0.053, and 0.050, respectively. Table 1 also shows that 72% of women and 81.2% of men were employed, while 73.9% of women and 81.9% of men participated in the labor force. 14 Panel B of Table 1 presents summary statistics of the city-level variables. The proportion of female migrants has followed an increasing trend since the early 1990s. The average proportion of migrant women in the cohort of females aged years is for the year 2000 but only for The average GDP per capita is 9,288 yuan, and the average annual growth rate of GDP is 15%. We can also observe that on average, primary industries contribute 19% of GDP, secondary industries contribute 45%, and tertiary industries contribute 36%. The city-level average wage is 8,037 yuan and 59.2% of employees work for SOEs. The divorce rate increased from in 1990 to in Empirical Strategy Our benchmark empirical model is a DID type model that uses both the 1998 policy change and differences in the density of female migrants aged across cities. The density of female migrants aged is used as a proxy for the intensity of the policy shock to local remarriage markets. The rationale behind this proxy is that the 1998 policy change is expected to have bigger effects on the remarriage market in cities where more migrant women are available. 15 We examine whether the changes in intrahousehold bargaining outcomes 14 The ratio of women or men employed is computed with respect to the total sample, and the ratio of women or men participating in the labor force is the ratio of women or men employed or without a job but looking for one to the total sample. 15 One concern might be that migrant women are not substitutable for local women. We thus tabulate the education of local women in our sample and the education of migrant women used to construct the treatment variable in our paper. For the local women, the percentages of women with primary school education or below, middle school, high school, and college or above are 13%, 36%, 37% and 13%, respectively. The same percentages for migrant women are 21%, 40%, 32% and 7%, respectively. Among both local and migrant women, the majority have primary or middle school education. It is feasible for local husbands to find 20

21 and intergenerational investments after the policy change differ across cities with different densities of female migrants. That is, we separately estimate the following equation for shares of expenditure on different goods. Y ict = α 0 + α 1 Mig density c,2000 P ost + γ 1 X ict + γ 2 M ct + δ c + λ t + ɛ ict (6) where Y ict denotes shares of expenditure on different types of good for household i in city c at year t; Mig density c,2000 is the density of female migrants aged 20 to 45 in city c, which is computed using the 2000 population census data; P ost is an indicator for the postreform period, which takes the value 1 for years after 1998; X ict is a vector of covariates including couples characteristics (husbands and wives age and years of schooling), family characteristics (the natural logarithm of total expenditure per capita, family size, and the age structure of family members). 16 Cities with different female migrant densities could have different macroeconomic cycles, which might also be correlated with household expenditure patterns. To control for these macroeconomic cycles, we include certain city-level macroeconomic variables (GDP per capita, GDP growth rate, shares of GDP in different industries, and average wage) in the regressions. In addition, an SOE reform starting in 1998 laid off many SOE employees. 17 If more SOE employees were laid off in cities with more female migrants, then the change in household expenditure patterns could be due to the change in the bargaining position between husband and wife because of the change in their relative economic status induced by the SOE reform, leading to bias in our estimates. Therefore, we control for the city-level share of SOE employees in the regression. All these city-level variables are included in M ct. We also include city fixed effects δ c and year fixed effects λ t. The coefficient α 1 thus captures how household outcomes changed after the policy change in cities with different densities of female migrants. We do not control for the post-reform dummy because the migrant women substitutable for their local wives. 16 The age structure of the family includes the proportion of male family members aged 0 6, 7 18, 19 60, and above 60, and the proportion of female family members aged 0 6, 7 18, and The proportion of female family members older than 60 is omitted to avoid collinearity. 17 See Wu (2003) for a detailed description of the SOE reform. 21

22 effect has been included in the year fixed effects. We also do not control for female migrant density because it is a city-level time-invariant variable and thus absorbed by the city fixed effects. Standard errors are calculated by clustering over the city level to deal with any heteroskedasticity problem. In the main analysis, we focus on gender-specific expenditures and intergenerational investments. The UHS data contain information on two categories of female-specific consumption women s clothes and cosmetics. Expenditures on women s clothes have been used in studies such as Lundberg, Pollak, and Wales (1997) and Browning et al. (1994). Although there is no direct empirical evidence, women are the main consumers of cosmetics products in China. 18 Male-favored consumption includes men s clothes, cigarettes, and alcohol. Similar to expenditures on women s clothes, expenditures on men s clothes have also been used by Lundberg, Pollak, and Wales (1997) and Browning et al. (1994). Wang (2014) shows that men consume much more cigarettes and alcohol than women and uses these two variables as male-specific consumption. 19 expenditures on children s clothes and children s education. 20 and 1999 (post-reform) are used in estimating Equation (6). We also examine intergenerational investments including Data for 1997 (pre-reform) The assumption underlying Equation (6) is that female migrant density in the year 2000, Mig density c,2000, should not be correlated with the error term. This assumption would be violated if the policy change induced a change in migration patterns in different cities. To address this concern, we use the city-level density of female migrants aged in 1990 as an IV for the density in Concerns remain as to whether the IV is correlated with the error term. One possible channel through which the IV might correlate with the error term is that expenditure in cities with a high migrant density in 1990 could have followed a different time trend from cities with low migrant densities had there been no policy change. Our estimates would 18 A China Daily article provides some description. See http : // 06/07/content htm 19 Headgear and footwear are potential gender-specific expenditures, but the UHS do not collect this information by gender. 20 Previous studies show that women tend to invest more in children than men (e.g., Duflo, 2003). 22