The Intergenerational Impact of China's New Rural Pension Scheme

Abstract

Old-age pensions have become an important component of systems aimed at protecting the elderly against poverty. The expansion of pension schemes can partly be attributed to important demographic changes that have pressured welfare institutions to respond to the demands of a growing elderly population. Bolivia's Renta Dignidad, India's Indira Gandhi National Old Age Pension Scheme, Mexico's 70 y Mas, and South Africa's Old Age Pension are notable examples of this new wave of old-age pension schemes in the global South. Only a few studies have examined the effects of old-age pension schemes on household income (Bertrand et al. 2003), assets and investment decisions (Landerretche and Martinez 2013), and labor market outcomes (Galiani, Gertler, and Bando 2014; Ardington et al., 2009), including those among adults living with pensioners (de Oliviera et al. 2017; Ardington et al. 2009). There is growing evidence that old-age pensions can influence retirement incentives and intertemporal behavior in labor supply and savings decisions, and can also affect wealth distribution across generations (Castañeda, Díaz-Giménez, and Ríos-Rull 2003; De Nardi and Yang 2014; Saez and Zucman 2016; Cowell et al. 2016). In Scandinavian countries, the effect of welfare regimes in smoothing inequality is often considered to be an important underlying mechanism that explains the increasing trend in intergenerational mobility in those countries (Jansson 2014; Landersø and Heckman 2017; Modalsli 2017). Among welfare policies, old-age pensions are strongly associated with intergenerational transmission of wealth. The study described here examines whether, and the extent to which, China's New Rural Pension Scheme (NRPS) has affected the intergenerational transmission of wealth in a country where the population has aged rapidly. We used a nationally representative panel survey, the China Health and Retirement Longitudinal Study (CHARLS), which collected data in 2011 and 2013, to answer these questions. The literature provides some plausible mechanisms underpinning the causal relationship between pension schemes and wealth transmission across generations. Feldstein (1974) and Gale (1998) suggested a substitution effect between pension savings and wealth of family members in a standard life-cycle model. Factors that can influence that rate of substitution include illiquidity of pension wealth and borrowing constraints, precautionary savings, and the functioning of insurance markets (Hubbard, Skinner, and Zeldes 1995), differential returns to pensions and other financial savings (Attanasio and Brugiavini 2003), and the value of benefits (Gustman and Steinmeier 2015). Other studies have argued for a jointly determined retirement age and savings effect in which pension wealth does not substitute for other forms of wealth but adds to total wealth, leading to higher savings before retirement in the presense of a pension (Gustman and Steinmeier 2001). High savings rates—even with public pensions, particularly in fast-growing economies like China—have been attributed, under an intertemporal decision framework, to habitual preferences (Attanasio and Weber 2010). Anticipating pension benefits can also change individual retirement incentives and intertemporal behavior in labor supply and savings, leading households to enjoy different wealth levels at retirement (Blundell, French, and Tetlow 2016; Chetty et al. 2014; De Nardi and Yang 2014; Saez and Zucman 2016). Intergenerational mobility models introducing bequest motives have predicted that in fast-growing economies in which the older generation suffered from very poor living conditions—as in the case of China—older individuals tend to save and bequeath considerable wealth to their children (Attanasio et al. 2016). In fact, empirical studies find both “crowding in” (Brandt and Deindl 2013) and “crowding out” (Rowlingson, Joseph, and Overton 2017) effects of public support on parental private financial support to adult children (i.e., downstream transfers). Social security policies affecting parental income or wealth can also crowd out children's transfers to parents (i.e., upstream transfers), as reported in the cases of China (Cai, Giles, and Meng 2006), Taiwan (Gerardi and Tsai 2014), and Mexico (Amuedo-Dorantes and Juarez 2015). However, earlier studies showed that children often transfer more time and/or money to parents in exchange for greater parental income or a greater wealth transfer (Bernheim, Shleifer, and Summers 1985; Cox and Rank 1992; Altonji, Hayashi, and Kotlikoff 1997). This can lead to “crowding in” of children's transfers to parents at the higher level of parental pension incomes, as shown in Chen et al. (2017). Intergenerational mobility models also predict that wealth inequality can persist into the next generation by bequests (De Nardi 2004) and parental investment in children's education and health (Becker et al. 2018).1 One may expect that old-age pensions create windfalls for retirees, which would beget more wealth for the next generation through the above two channels. In particular, when a pension serves as an income transfer to a financially constrained household, pension entitlements could lead to improvements in children's well-being (Duflo 2000; Gutierrez, Juarez, and Rubli 2016) and to increases in their education, which in turn can contribute to intergenerational welfare improvement (Mu and Du 2015). This study contributes to the literature in a number of ways. First, we examine the distributional effects of parental pension status on filial wealth accumulation and its association with parental wealth. The existing literature has studied outcomes of pensions mainly in cross-sectional settings, including recipients’ consumption (Zheng and Zhong 2016), private savings (Feng, He, and Sato 2011), labor supply (Galiani, Gertler, and Bando 2016), living arrangements (Hamoudi and Thomas 2014), and their extended family members’ labor mobility decisions (Chen 2016). While the impact of pensions can extend to the next generation, as discussed earlier, to our knowledge no empirical studies have explored the individual-level intergenerational dependence or mobility of wealth that is induced by pensions. To identify the causal effect of the pension, we adopt a fuzzy regression discontinuity (RD) design, which exploits not only the age eligibility threshold but also the exogenous variation in the rollout of the pension scheme. This differs from a standard RD, in that we utilize information on both treated compliers and never-participants for comparison, as it is possible that nonprogram communities could have benefited by the reassignment of the program (Ravallion 2007). Taking never-participants into account yields larger treatment effects. Second, we address two different forms of heterogeneity that are likely to affect the impact of the pension. One is related to heterogeneous behavioral responses to pensions, depending on observed characteristics. For example, Engen, Gale, and Scholz (1996) have shown that saving incentives induced by pension schemes can raise private savings when households finance contributions by reductions in consumption, increases in labor supply, or tax cuts. However, private savings cannot rise if households use existing assets to contribute to a pension scheme. This creates variations in wealth outcomes, depending on pensioners’ existing economic endowments. Household data from the United Kingdom and Italy suggest that the substitution effect between pension and nonpension wealth is particularly high for workers between ages 35 and 45, as their liquidity is often more constrained than that of older cohorts (Attanasio and Brugiavini 2003). From the perspective of family members who have parents participating in or receiving pensions, their transfers to parents are related nonlinearly to parental or household income in both developed and developing countries (Cox, Hansen, and Jimenez 2004). This situation seems to be driven by children's motivations (Chen et al. 2017) and economic conditions, notably income volatility (Albarran and Attanasio 2003). The other form of heterogeneity is related to unobserved characteristics such as individuals’ time preferences and perceived value of pension benefits, which can influence the timing of their benefit claims and life-cycle wealth accumulation (Gustman and Steinmeier 2015). To address not only the heterogeneous effects of pensions on adult children's wealth across the entire distribution, but also the endogeneity problem arising from self-selection into program treatment—conditional upon observed and unobserved characteristics at individual and community levels—we apply instrumental quantile regressions to our fuzzy RD design. We also compute our estimates separately, depending on the age of eligibility. The empirical findings shed new light on recent welfare policy in China. Our individual-level analysis provides the “net” policy impact of the NRPS after taking individual behavioral responses into account. The estimated impact of the program on intergenerational mobility helps us understand the effects of social protection policies in China and adds to the limited knowledge base for middle- and low-income countries that are currently considering introducing or expanding old-age pension schemes. At the end of 2000, China had 88.1 million people aged 65 and older—7 percent of the country's total population.2 This placed China in the United Nations’ “aging society” category. At an average annual growth rate of 4 percent, the elderly population would have continued to grow in size, to 222 million by the end of 2015, representing 16 percent of China's total population. This proportion is twice the 2015 world average of 8 percent (OECD 2015). The United Nations has projected that by 2050, people aged 60 and older will make up more than 30 percent of China's total population (UNRISD 2016), again substantially higher than the projected world average of 18 percent (OECD 2015). China's rural population has long been excluded from social protection. This was especially true for the rural elderly population, which had not been entitled to participate in any pension scheme until 2009, when the government piloted the NRPS. Even though the government began to introduce the rural minimum living standard guarantee scheme (rural Dibao) in 2001, the coverage rate increased only slightly, from 0.4 percent in 2001 to 8.6 percent in 2015; moreover, this scheme was not targeted at the elderly, but rather at the poor and vulnerable.3 The 2010 census showed that 46 percent of those past retirement age still relied on their own labor, while roughly 44 percent were supported by their families (see Table 1). The Chinese government piloted the NRPS in about 10 percent of counties in 2009, aiming to cover all rural adults by 2020. The rural population aged 16–59 who are not covered by other pension schemes and are not in school are eligible to join the program at their places of household registration and on a voluntary basis. Participants have to contribute for at least 15 years to receive the benefits. At the time of program implementation, the rural elderly were entitled to receive benefits without making any contributions, as long as all of their eligible adult children had joined the NRPS. These terms encouraged most rural adults to join the scheme. Individual annual contributions initially ranged between 100 and 500 yuan (about US$15 and US$75 according to the average market exchange rate in 2016), but because the provincial governments can set higher contribution rates than those provided by the central government, the maximum annual contribution approached 3,600 yuan (about US$542) in 2016. The benefits comprise (1) total savings in the private account before turning 60 (including individual contributions, government subsidies and accrued interest and (2) the pension payout from the central government, which was 55 yuan initially in 2009 and increased to 115 yuan in 2016.4 Studies have reported that the NRPS has improved consumption among beneficiaries (Zheng and Zhong 2016), especially for elders with only one child (Liu et al. 2015). Cheng et al. (2018b) report that the pension improved recipients’ nutrient intake, access to health care, use of inpatient services, and leisure time and reduced their reliance on their adult children, especially sons. Ding (2017) also found improved life satisfaction as a result of the pension, whereas Eggleston, Sun, and Zhan (2016) showed that it facilitated adult children's migration and off-farm work. We are not aware of previous studies examining the welfare effects of participation (rather than receipt) for those still having to contribute for decades, but only on determinants of their participation decisions. Those with fewer sons were more likely to join (Ebenstein and Leung 2010), while younger rural residents joined less often and at lower contribution rates (Lei, Zhang, and Zhao 2013). Rates of uptake also varied by age, value of durable assets, health status, and local enforcement of the program and by the size of the government payout (Zhao et al. 2016). These factors were found to threaten the long-term sustainability of the scheme (Bairoliya et al. 2017). CHARLS is a nationally representative panel dataset collected by the School of National Development at Peking University in 2011 and 2013. Populations aged 45 or older were interviewed using a stratified sampling framework, with units selected with probability proportional to size. The baseline survey included 17,708 individuals aged 45 or older, out of 10,257 households in 450 communities in 150 counties from 28 provinces. The follow-up survey covered 18,605 individuals out of 10,803 households in the same communities. Our sample included individuals who were interviewed in both waves and who resided in rural communities (as defined by the National Bureau of Statistics of China), and thus were eligible to receive the pension. They had at least one biological child (aged 18 or older) with valid information and reported information on pensions in the 2011 wave, which allowed us to disentangle the dynamic impact of the pension scheme. The selected individuals were defined as the parental generation. Each individual came from one household. These individuals were paired with their adult children, whom we refer to here as the filial generation. Table 2 cross-tabulates the samples by parental age. There were 1,990 parents aged 45–60 and 3,390 parents older than 60. There were 4,733 parent-child pairs for the former group and 12,876 intergenerational pairs for the latter. The NRPS had been introduced to about 43 percent of the sampled communities by 2011, slightly lower than the national coverage (60 percent) in the same year,5 but this proportion increased to 53 percent in 2013. In 2013, about 90 percent of parents lived in treated communities, while 76 percent of parents living in treated communities joined or received payments from the NRPS. This is equivalent to an overall enrollment rate (number of participants and recipients divided by the number of surveyed parents aged 45 or above) of 69.4 percent. This is higher than the identical enrollment rate from another nationally representative household survey, the China Family Panel Study (CFPS), which found enrollment rates of 51.3 percent in 2012 and 60.6 percent in 2014. Another survey in five provinces, which was conducted by the Chinese Academy of Agricultural Sciences in 2012 using a similar multilevel sampling procedure as the CHARLS and the CFPS, showed that 74 percent sampled individuals aged 16–92 were enrolled in the NRPS in 2011 (Chang et al. 2014). These survey-based rates were relatively closer to the national rate of 63.8 percent in 2010, as expected,6 but are much lower than the national rate in 2013, when the NRPS was rapidly expanded.7 The CHARLS survey provided sampling probabilities reflecting their sampling procedures and non-response rates during interview. We used the inverse sampling probabilities as the weights in all regressions, in order to correct for possible sampling biases during our sample selection. Given that roughly one parent aged below 60 had two children and one aged above 60 had three children, we clustered standard errors at parental level. Tables 3 and 4 compare various aspects of life between generations by parental age. As the key indicator, wealth was defined as net worth—that is, the sum of housing assets, fixed assets (including the value of productive and household business assets), the value of consumer durables, financial assets (such as savings, equity, and loans), and other forms of assets (such as jewelry), net of all debts. Pension wealth and human capital were not included, following Cowell et al. (2016) and Saez and Zucman (2016). For assets jointly owned by household members, we split the value equally by dividing it by household size. For assets—typically housing, fixed assets, and financial assets—we divided the value by the individual household members’ share.8 Debts were calculated analogously for each household member. Finally, we constructed estimates of individual net wealth for both generations. Table A1 in Appendix A presents descriptive statistics on the variables of interest. The parental generation earned less income than the filial generation, due in part to age effects and to the amount of human capital held by the filial generation. Adult children owned more wealth as a result of the higher value of housing assets. Only 6 percent and 16.5 percent of parents and 3–6 percent of adult children reported negative net wealth. This fraction is smaller than in Sweden (26 percent) (Black et al. 2015), and similar to Germany (9 percent) and the United States (14 percent) (Cowell et al. 2016). Within each generation, the older the age, the less income and wealth owned, which is consistent with an age effect. Nearly two-thirds of parents preferred to live with adult children (see Table 3), and this preference was much stronger among parents without a spouse (71–78 percent). The rate of actual coresidence with children was lower than the stated preferences, but it was still higher than one-third in the two elderly cohorts (see Table 3). A total of 90 percent of participants (and 72 percent of recipients) paid for the NRPS by themselves rather than relying on their children (see Figure 1). More than 70 percent of parents expected financial support from children in old age, as shown in Table 3, regardless of their age compared with the eligibility age of 60 years old, while only 13–15 percent considered pension income as their main source of financial support. Distribution of means of old-age support in the parental generation, by age-group SOURCE: Authors’ calculations, based on the CHARLS 2013. Table 3 shows that the average annual payment to the NRPS in 2013 was 171 yuan (approximately US$28) among 45–59 year old participants and 177 yuan among participants 60 and older, similar to the national average of 177 yuan.9 Indeed, 79 percent of parents affiliated with the NRPS and aged between 45 and 59 (and 75 percent of those aged 60 or older) paid the lowest contribution rate of 100 yuan (approximately US$16); 75 percent of parents who had already received the monthly benefits in 2013 reported receiving the lowest level of payout, 55 yuan (approximately US$9). Transfers within the extended family are extensive as well as intensive. Roughly one-quarter of parents made transfers to children over the previous year. The amount of these transfers was substantial, representing 71 percent of parental net income among those younger than 60 (about 3,287 yuan, out of 4,625 yuan net income), and mainly in cash. Those aged 60 and older transferred less (54 percent of net income, or roughly 867 yuan, out of 1,603 yuan income). The 60 and older cohort received more transfers from children, however—83 percent of them received such transfers, compared with 56 percent of their counterparts younger than 60. As a result, 78 percent of parents older than 60 had positive net transfers from adult children, with an average amount that tripled their annual net income. In comparison, 14 percent of those aged below 60 made net transfers to their adult children rather than receiving transfers from them (Table 3). Of those making net transfers to their adult children, the average amount of net transfers was 1,380 yuan per year, which was about 31 percent of the children's average annual net income (4,414 yuan in Table 4). Moreover, about one-third of the parental generation younger than 60 also helped their parents. The average amount of this transfer (365 yuan) was equivalent to 7.9 percent of their annual net income. The incidence and amount of this upstream transfer was much smaller for the cohort aged 60 and above, which is predictable, given low survival rates of their parents. For the filial generation, increasingly more adult children make upstream transfers to their parents when their parents turned 60 or older, from an average of 38 percent to 59 percent (Table 4). We also note that 34 percent of adult children with parents under 60 and 56 percent of those with 60 and older were “net givers” to parents (see bottom row of Table 4). Table 5 presents data on the the intergenerational transitions of wealth. For both cohorts, the percentages along the diagonal line are higher than those off it, indicating intergenerational dependence of wealth. Among the filial generation whose parents were below 60 years old (Table 5, top panel), 72 percent of the richest quintile successfully maintained their parental position in the wealth distribution, as opposed to 58 percent in the first quantile, indicating stronger dependence among the rich than the poor. Of the filial generation in this cohort, 57 percent had the same wealth position as their parents; 19 percent increased their ranking relative to their parents, while 24.2 percent ranked lower than their parents. By contrast, the poorest quintile showed stronger intergenerational dependence (90 percent) than the richest quintile (79 percent) for those whose parents were 60 and older (Table 5, lower panel). Of the filial generation in this cohort, 74 percent had the same wealth rank as their parents; only 15 percent moved to higher wealth positions, while 11.2 percent shifted downwards. Comparing both panels of Table 5, it appears that wealth dependence is stronger among the older members of the filial generation than among the younger ones, while the latter experience more mobility, both upward and downward. Table 6 compares parental and filial observed characteristics between participants (recipients) and nonparticipants (nonrecipients) in the NRPS. Among those older than 60, recipients are younger and less educated than nonrecipients. Both participants and recipients are more likely to be female and to earn less, while recipients’ children earned more than nonrecipients’ children. This is consistent with the requirements of the NRPS—those aged 60 and older can receive payments without making a contribution, so long as all of their adult children join. Participants younger than 60 earned significantly less than their nonparticipating counterparts. There were no wealth differences in the cohort younger than 60 years old, but recipients older than 60 and their adult children had less wealth than nonrecipients and their adult children. These observations are consistent with the nonrandom, pro-poor targeting mechanisms of the pilot phase of the NRPS. In general terms, we observed no significant differences in parental health or preferential or actual living arrangements. Nor did parents’ expected age of retirement or the transfers from or to children vary by their NRPS status. The details of our modeling and identification strategy are given in Appendix B. In general, we estimate separately the causal effects of parental NRPS on filial wealth and intergenerational dependence for those whose parents had not reached 60 years of age or who were 60 and older. The rationale was that anticipating and contributing to the pension might bring different incentives relative to those having been eligible to receive pension payments. This might be particularly the case in China, because of the strict criteria for those older than 60 to receive payments without having made contributions during their working ages. Table 7 reports the results from the two-step least squares (2SLS) and the quantile RD estimators. The 2SLS uses community treatment status () and other as the excluded instruments to parental compliance with the NRPS. These performed well in terms of overidentification tests and other goodness-of-fit tests for both cohorts. The local average treatment effect (LATE) estimators (see Imbens and Angrist 1994 and Apppendix B for a fuller discussion) of parental compliance with the NRPS on filial wealth differ between the 2SLS and quantile specifications. For those whose parents are younger than 60 years old, the negative LATE on filial wealth accumulation in the short term seems to be driven by those in the bottom quintile of the filial wealth distribution; in contrast, for those whose parents have reached the eligible age of the NRPS, the positive LATE in the short term is driven by those at the top of the filial wealth distribution. Differences also exist in other key variables, like parental wealth, for the cohort younger than 60 years old and its interaction with the NRPS status for the cohort being at least 60 years old. These reflect differences in behavior between those whose parents still contribute and those whose parents have received the pension. Moreover, given that we cannot reject the null hypotheses of homoscedasticity and the absence of intracluster correlation, quantile regressions are more appropriate specifications than the standard linear 2SLS. Table 7 shows the intergenerational dependence of wealth regardless of parental age. This dependence appears throughout the distribution of filial wealth and is particularly strong at the top end. We also reestimated the models in Table 7 by replacing the binary compliance variable with the logarithmic premium and payouts in two years for the two age-cohorts, respectively. Table A2 in Appendix A presents the results. Overall, the results indicate that parental compliance in 2013 appears to weaken intergenerational dependence of wealth among those at the top wealth decile by helping filial wealth accumulation, although this effect is not significant. Given the wealth-differentiated LATEs of the NRPS, we focus in the next sections on the quintile results and examine some potential underlying mechanisms, particularly for those at the bottom of the wealth distribution. Figure 2 shows the direct impact of parental NRPS status on filial wealth against the entire distribution of filial wealth—i.e., and —and the total impact after taking parental wealth into account—i.e., and for parental compliance in 2011 and 2013, respectively, as derived in Equation 13 in Appendix B. For those having less wealth than the median, parents’ receipt of the pension did not account for filial wealth accumulation (the black dashed lines in Figures 2a and 2b). Impact of the NRPS on filial wealth (by parental age) NOTES: The grey and black dashed lines depict direct treatment effects of parental NRPS on filial wealth for 2011 and 2013, respectively. The grey and black solid lines depict total treatment effects of parental NRPS on filial wealth, including both the direct and indirect effects conditional on parental wealth for 2011 and 2013, respectively. SOURCE: Authors’ calculations, based on the CHARLS 2011 and 2013. The negative impact becomes much larger for those in the bottom quintile, when parental wealth is taken into account (the black solid line in Figures 2a and 2b). At first sight, this finding seems at odds with the general argument that old-age pensions can function as an insurance mechanism that protects the poor. We investigated this issue more closely by separating the two age-cohorts. For those whose parents are younger than 60 years old, there are two likely reasons for the negative effect of the pension scheme on filial wealth. One is that the NRPS places additional economic burdens on the poor. Only 38.5 percent of the filial generation in the bottom wealth quintile had participating parents in the NRPS. The most frequently reported reason for nonparticipation in treated communities was “unaffordability.” Among the filials in the bottom wealth quintile who had participating parents, the parental average annual premium was 146 yuan, equivalent to 5.9 percent of parental annual net income. All participating parents chose the method of annual payment and had to pay for 9.7 years, on average, before starting to receive the benefit. In total, 9.8 percent of participating parents relied on their children to pay the premiums. In this case, the annual premium payment for filials, including him/her and at least one parent, would constitute 7.8 percent of their net income (≈146 × 2/3,742 yuan). Thus, it is not surprising to see that individuals in the bottom wealth quintile having participating parents experienced negative “net” transfer—i.e., the transfer they received from parents, minus the transfer they gave to parents. Their parental participation in the NRPS aggravated this negative net transfer, at least in the short term, by 42.8 percent, because it increased the filial total transfer to parents by 44 percent (significant at a 10 percent level), from 537 yuan per annum to 773 yuan. Nevertheless, it is worth noting that, when regressing the net transfer on parental NRPS under an RD specification (i.e., the IV specification as Column 1 of Table 7 without interaction terms), we obtain