Housing Windfalls and Intergenerational Transfers in China1
Albert Park, HKUST
Maria Porter, Michigan State University
December 2013
Abstract
In this paper, we study the impact of housing reform and the rapid development of the housing
market in China on intergenerational transfers and elderly well-being. During the 1990s, the
Chinese government gave property rights to many urban residents who had been allocated
housing by their employers. These unexpected windfalls were substantial in size, and grew with
the rapid increase in housing prices over time, significantly impacting the asset holdings and
wealth of affected urban residents. We examine the effect of these exogenous changes in wealth
on inter-vivos transfers and intergenerational transfer motives.
1
Maria Porter acknowledges financial support from the Oxford Martin School and the Oxford
Institute of Population Ageing. Emails: albertpark@ust.hk, maria.porter@economics.ox.ac.uk.
1
1. Introduction
A large literature starting with Becker (1974, 1981) describes and seeks to explain the
motivations for intergenerational transfers. Understanding the responsiveness of private support
to the needs of others has important implications for the design of public social assistance and
social insurance programs. For this reason, it is of particular interest to know the extent to which
private transfers are motivated by altruism. Transfers also may be motivated by exchange
motives, either as part of mutual insurance or credit arrangements or in exchange for prior or
future benefits (Bernheim, Shleifer, and Summers, 1985). For example, children might provide
transfers to their parents in exchange for prior parental investments in their education, assistance
in child care, or to influence future bequests.
Previous studies of the motivations for intergenerational transfers have focused on testing
how differences in the income of givers or recipients impact intergenerational transfers. The
studies posit that altruism predicts that transfers are inversely related to recipients’ income levels
while exchange motives predict that transfers increase with recipients’ income (Cox 1987). It is
also possible for both motivations to exist at the same time, with different motives more
prominent at different income levels. Unfortunately, these income-based tests suffer from
conceptual, measurement, and identification problems. First, tests based on how recent transfers
respond to contemporaneous income levels cannot distinguish between altruism and mutual
insurance motivations, which both predict that current transfers increase with lower current
income levels. Recent income may be low due to temporary negative income shocks, rather than
reflecting lower permanent income. Second, income measurements are noisily measured, which
can lead to attenuation bias. Third, if transfer recipients are elderly, their annual incomes may not
well reflect their wealth or permanent income if they have stopped working. Finally, incomes
2
may reflect labor supply decisions of household members that respond to transfers, leading to
simultaneity bias. The empirical evidence from such empirical studies shows mixed results.2
Previous authors have addressed some of these problems, for instance by testing how
transfers respond to predicted incomes that are purged of transitory components and hold
constant working hours. Studies using US data find that the probability of children receiving
transfers from parents is positively associated with the children’s predicted age-adjusted
earnings. In China, using predicted household incomes assuming full-time work by all household
members Cai, Giles, and Meng (2006) find that transfers are altruistic for those below the
poverty line, and Kazianga (2006) finds evidence of altruism in Burkina Faso when modeling
permanent income as a function of livestock, farm, and other assets (Kazianga, 2006). However,
this approach remains problematic for retired individuals whose wealth reflects income earned in
the past when wage determination may have been quite different than in recent years. Moreover,
exercises based on cross-sectional comparisons often still are unable to deal with omitted
characteristics of recipients, such as an attractive personality, that may be correlated both with
earnings’ ability and permanent income and the desire of others to provide transfers to them.
If one tries to avoid the identification problems associated with unobserved heterogeneity
by examining how transfers respond to changes in income, then the problem arises that the
income changes may be anticipated or temporary, so that transfers may reflect insurance motives
and liquidity constraints.
Studies that test the responsiveness of transfers to recipients’ income that find evidence of altruism include
McGarry and Schoeni (1995) and Schoeni (1997) in the US; Wolff, Spilerman, and Attias-Donfut (2007) in France;
Raut and Tran (2005) in Indonesia; Cox and Jiminez (1992) and Cox, Eser, and Jiminez (1998) in Peru. Those
finding evidence of exchange motives include Cox (1987) and Cox and Rank (1992) in the US. Cox, Jansen and
Jimenez (2004) find that in the Philippines, transfers are altruistic at low income levels, while exchange motives
dominate at higher income levels.
2
3
In this paper, we follow a unique approach to examine how a permanent change in one’s
wealth influences transfers received from non-resident adult children. We identify the effect of a
permanent wealth increase on intergenerational transfers by instrumenting household wealth with
the market windfall many employees of work units received when they purchased housing from
their employers. In urban China, housing reform resulted in rapid appreciation of undersupplied
housing in the private sector. Prior to the housing reform, danwei employers allotted housing to
their employees. During the period of housing reform, since this housing was allocated before a
housing market existed; significant random variation resulted in the size of housing windfalls
associated with the market value later placed on specific locations. In this paper, we use this
variation in the appreciation of housing market values to identify the impact of an exogenous
increase in household wealth on intergenerational transfers.
Those who purchased subsidized housing from their work units received significant
windfalls, resulting in an increase in household wealth and permanent income. If transfers from
children to parents are motivated by altruism, we would expect children to reduce the amount
transferred to their parents.
On the other hand, if children increase the amount they transfer to parents when parents
receive this housing windfall, then their transfer motives are more likely to be strategic rather
than altruistic. Children may provide more to their parents in the hopes that parents will bequest
their housing to them, or in exchange for living with them in better housing than they could
afford at market rates. While children generally earn more than their parents, the housing
windfall may also shift bargaining power within the family from adult children to elderly
parents. They may be able to extract more time and care from their children as a result of this
increased bargaining power.
4
In addition, wealth or asset holdings may be a better measure of elderly wellbeing than
wage income, particularly in this context. In developing countries, income is hard to measure.
Usually, expenditure proxies for income. In China, retired elderly receive little income. Some
might receive pensions, but amounts are usually low. So they primarily rely on their wealth. In
addition, wealth may reflect the occurrence of negative life events. For example, in the U.S.,
death and divorce slow down asset accumulation. Often assets are actually drawn down during
such events. In the U.S., the evolution of assets is also very strongly related to health (Poterba,
Venti, and Wise 2010). In urban China, the ability to draw down assets in times of need would
be more difficult, since housing comprises nearly 80% of household wealth on average, and is an
illiquid asset. Children might help their parents when they are in need so that they do not sell
housing that would appreciate in value, allowing them to inherit this valuable asset or make use
of it by living with parents.
Other studies set in China find evidence of multiple transfer motives. As noted earlier,
Cai, Giles, and Meng (2006) find that transfers by children to parents are altruistic when parents’
household income per capita is below the poverty line. Secondi (1997) finds that child care
services may be one of the main services the elderly provide their adult children in exchange for
money, and that altruism alone cannot explain transfers from children to elderly parents. In
addition, Li, Rosenzweig and Zhang (2010) use twin data in China to exploit a "natural
experiment" resulting from parents having been forced to send away one of their children for
rustification during the Cultural Revolution, and use this experimental variation in parental time
with children to identify its effect on transfers to children later in life, and whether transfers are
motivated by altruism or guilt. The authors find that while transfers are motivated by altruism,
5
guilt also plays an important role, as children who had been sent away for a longer period of time
receive more transfers from parents, even when they have higher earnings.
The rest of the paper is organized as follows. In section 2, we present a theoretical model
that allows for both altruism and exchange motives in order to derive predictions and clarify
interpretation of the empirical specification. We provide background information on China’s
housing reforms and housing market development in section 3; introduce the data from the China
Health and Retirement Longitudinal Study and the empirical methodology in section 4; present
the results in section 5, and conclude with section 6.
2. Model
Consider the following model of intergenerational transfers that allows for both altruism
and exchange motivations. In contrast to previous models designed to capture decisions of
parents in the US and other developed countries on how much to transfer to children, our model
makes the child the principal, which is consistent with the fact that unlike in the West in China
transfers on average flow from children to parents.
We model transfers from one child to one parent. The child maximizes his or her own
utility defined as a function of three arguments:
𝑈 = 𝑈(𝑋𝑐 , 𝐵, 𝑉(𝑋𝑝 , 𝑇, 𝐵)).
(1)
Utility is a function of the child’s own consumption Xc, bequests received from parents B, and
the utility of the parent (defined by V), which itself is a function of parental consumption Xp,
transfers from children T, and bequests. Transfers may be valued for emotional reasons that are
independent of the consumption it facilitates. The inclusion of bequests in the parent’s utility
function allows for the possibility that parents may value leaving large bequests to children
6
which is independent of the negative utility associated with foregone consumption. This could
reflect a concern for the future welfare of their children, reflecting reciprocal altruism.3 Modeling
B separately in the utility functions of parents and children enables us to distinguish exchange
from altruism motives.
We assume that consumption of parents and children are defined as follows:
𝑋𝑝 = 𝑊 0 + 𝑌𝑝 + 𝑇 − 𝐵
(2)
𝑋𝑐 = 𝑌𝑐 − 𝑇.
(3)
Here, W0 is parental initial wealth, and Yp and Yc are the incomes of parent and child,
respectively. We assume that bequests can be influenced by transfers to parents, so B is a
function of T. Below we define the parent’s optimal bequest explicitly.
Finally, there is a participation constraint for the parent. Parental utility must be greater
than or equal to the utility of the parent absent the relationship with their children:
𝑉(𝑋𝑝 , 𝑇, 𝐵) ≥ 𝑉0 (𝑊 0 + 𝑌𝑝 , 0,0).
(4)
The model avoids time subscripts, reflecting our preference that the consumption, income, and
transfer variables be interpreted as total outcomes spanning the life of the parent-child
relationship. Such an interpretation is reasonable if children or parents have access to credit
markets, in which case the optimal timing of income and transfers is irrelevant and consumption
is perfectly smoothed over time (Cox, 1990). It is readily apparent that W 0 and Yp operate
identically in the model. One can think of W0 as reflecting the wealth accumulated by parents
before children become adults, earn income, and are in a position to make decisions affecting
Parent’s could also feel pride in amassing large amounts of wealth which they prefer to leave to children, or in
creating a legacy after their death through the wealth of their children. This formula makes the simplifying
assumption that parents do not care about the consumption of their children; this could easily be added without
affecting the main results.
3
7
their parents. Define W = W0 + Yp as permanent income of the parent during the life of the
intergenerational relationship.
To make the model more tractable, we assume that the utility functions are linearly
separable. An important parameter α is introduced to capture the degree of altruism, with α=1
reflecting perfect altruism, meaning that children care as much about their parents as about
themselves. We also substitute in for the definitions of Xc and Xp to arrive at the following:
𝑈 = 𝑢(𝑌𝑐 − 𝑇) + 𝑓(𝐵 ∗ (𝑇, 𝑊)) + 𝛼[𝑣(𝑊 + 𝑇 − 𝐵 ∗ (𝑇, 𝑊)) + 𝑔(𝐵 ∗ (𝑇, 𝑊), 𝑇)].
(5)
Here, B* denotes the optimal bequest decided upon by parents, which is the solution to the
parent’s utility maximization problem defined as follows:
max 𝑣(𝑊 + 𝑇 − 𝐵) + 𝑔(𝐵, 𝑇).
𝐵
(6)
Note that the utility from bequests is likely to be positively related to the amount of transfers
received from children, if this is the case then the cross-partial with respect to B and T is
𝜕𝑔2
positive, or 𝑔𝐵𝑇 = 𝜕𝐵𝜕𝑇 > 0. If T were not included as an argument in g, then transfers would
not affect bequests any differently than the parent’s own income.
We focus on the case in which the participation constraint of the parent is non-binding. If
the constraint binds, the child does not wish to make the parent better off despite their altruistic
feelings. This is most likely to occur when the parent’s lifetime income is much greater than that
of the child, which is relatively uncommon in a rapidly growing country such as China.
Thus, we can simply take the first order condition (FOC) for equation (5) with respect to T. This
yields:
−𝑢′ + 𝑓 ′ 𝐵 𝑇 + 𝛼(𝑣 ′ (1 − 𝐵 𝑇 ) + 𝑔𝐵 𝐵 𝑇 + 𝑔𝑇 ) = 0.
8
(7)
We then totally differentiate the FOC with respect to W and T to derive the comparative
static result:
𝑇
(𝑢′′ + 𝑓 ′ 𝐵 𝑇𝑇 + 𝑓 ′′𝐵 + 𝛼(𝑣 ′′ (1 − 𝐵 𝑇 )2 + (𝑔𝐵 − 𝑣 ′ )𝐵 𝑇𝑇 + 𝑔𝐵𝐵 𝐵 𝑇 + 𝑔𝑇𝑇 )) 𝑑 +
(𝑓 ′ 𝐵 𝑇𝑊 + 𝛼(𝑣 ′′ (1 − 𝐵 𝑇 )(1 − 𝐵 𝑊 ) + (𝑔𝐵 − 𝑣 ′ )𝐵 𝑇𝑊 ))𝑑𝑊 = 0
(8)
Rearranging terms, and noting that The first order condition for optimal bequests requires that
gB – v’ = 0, we derive the following expression for the effect of parental wealth on transfers:
𝑑𝑇
𝑓 ′ 𝐵𝑇𝑊 +𝛼𝑣 ′′ (1−𝐵𝑇 )(1−𝐵𝑊 )
= − 𝑢′′ +𝑓′𝐵𝑇𝑇 +𝑓′′𝐵𝑇 +𝛼(𝑣′′ (1−𝐵𝑇 )2 +𝑔𝐵𝐵 𝐵𝑇 +𝑔𝑇𝑇 )
𝑑𝑊
(9)
The denominator of this expression is unambiguously negative given decreasing marginal utility
in all arguments, which means that the sign of dT/dW is the same as the sign of the numerator.
The first term in the numerator is positive because parental wealth increases the influence of
transfers on bequests. This is intuitive since there is more wealth for the parent to bequeath, and
it can be easily derived from the optimal bequest problem solved by parents. It captures the
exchange motive of children to increase transfers in order to receive larger bequests when more
is at stake. The second effect is negative as long as transfers increase parental consumption (1BT>0), reflecting the altruism motive.4 With greater wealth, the marginal utility gain to parents
from transfers falls, which reduces transfers if children are altruistic (α>0). This leads to an
overall ambiguous prediction on the effect of parental wealth on transfers.
From (9) it is also possible to predict how the relative importance of altruism and
exchange motives varies with the permanent income, or wealth, of the recipient. At low levels of
wealth, the marginal utility of consumption is high so that greater wealth will mostly be
consumed. This implies that the impact of wealth on the responsiveness of bequests to
4
If transfers increase bequests by more than the amount of transfers (B T>1), then more transfers reduce parental
consumption, in which case there is an unambiguous prediction that greater parental wealth will increase transfers to
parents.
9
transfers(𝐵 𝑇𝑊 ) will be smaller, reducing the first term of the numerator in (5). At lower
consumption (and wealth) levels, if the utility function is sufficiently concave (v’’’>0)5 then v’’
will be more negative, and the impact of both transfers and wealth on bequests is likely to be
reduced. For these reasons, the second term in the numerator of (9) is also likely to be more
negative at lower wealth levels. Thus, altruism is more likely to dominate at lower wealth levels
while exchange motives are likely to dominate at higher wealth levels.
3. Housing Reform in China
Initial housing sales gave owners the right to use their property and pass it on through
inheritance, but not the right to sell. Initial experiments began between 1979 and 1981, with the
sale of new housing to urban residents at construction costs. In the early 1980s, all housing was
purchased from work units.
In Gansu, this was true until the 1990s, and a private market for housing did not seem to
really develop until 1998 (see Figure 1b). However, throughout much of the country, from 1982
to 1985, there were experiments with subsidized sales of new housing. The buyer paid one-third
of the cost, while the remainder was subsidized equally by the employer and the city
government. Such costs included the cost of land acquisition as well as the provision of local
public facilities.
Experiments with comprehensive housing reform began between 1986 and 1988. At this
time, public sector rents were raised; a housing subsidy for all public sector employees was
introduced; and sales of public sector housing were promoted. In November 1991, a conference
resolution “On Comprehensive Reform of the Urban Housing System,” resulted in large scale
5
This is the necessary condition for utility functions to exhibit prudence, which provides a motive for precautionary
saving.
10
sale of public housing at very low prices, particularly to current occupiers. But in 1993, the
government suspended this housing reform because of concern about very low prices of sales.
In 1994, a significant wave of housing reforms was put into place as a result of “The
Decision on Deepening the Urban Housing Reform.” Under this system, employers and
employees contributed to house savings funds. Rents were increased, especially for new tenants.
However, low-income households received lower rents for old or unsellable public housing.
Three price mechanisms were put into place for public housing sales: (1) high-income families
paid market prices and received full property rights, including the right to resell on the open
market; (2) low- and middle-income families paid below market prices, which covered all costs,
including land acquisition, construction, and local public facilities; (3) for those who could not
afford these costs, a transitional pricing scheme was offered, taking into account both cost and
affordability. The latter two types of subsidized purchases only gave home-owners use rights. In
order to resell, the owner would have to repay the housing subsidy that had been originally
provided. During this time, few cities had any secondary housing market.
In July 1998, a new policy was developed in order to stop the material distribution of
public housing to employees and to introduce a cash subsidy for housing issued by public sector
employers. Specific policies varied across regions and details were determined by local bodies.
In general however, affordable housing was built with government support for low- and middleincome families, which constituted 70-80 percent of urban residents). High-income families (1015 percent urban residents) purchased higher quality housing in the market. Poor families were
given subsidized rental housing by their employers or the city government. But local authorities
and state enterprises often had insufficient resources to pay such subsidies.6
6
For additional details on the history of housing reform policies in China, please see Wang and Murie (2000).
11
In both Gansu and Zhejiang, we can see the major housing reform policies in 1991 and
1996, when the number of housing purchases increased considerably from previous years (see
Figures 1a and 1b). Work unit housing purchases began to decline in 1999 when housing reform
policies were for the most part coming to an end. While there was virtually no private housing
market in Gansu until 1998, in Zhejiang, the number of houses purchased in the private market
seems to be closely related to housing reforms. For example, market-bought housing purchases
peaked in 1991, one year before work unit purchases peaked during the first major wave of
reforms. Market-bought housing also declined when work-unit housing purchases peaked. In
addition, the number of housing purchases in the private market rose significantly in 1998, two
years after the peak in work-unit purchases in 1996, during the second wave of major reforms.
These fluctuations suggest that as work-unit housing became increasingly available for purchase,
the private market for housing developed rapidly in response.
By 1999, few houses could be purchased from employers. Private housing markets began
to dominate the scene, with rapidly rising prices because of insufficient supply. Shing-Yi Wang
(2010) examines the equilibrium price effects of the privatization of housing assets that were
allocated by the state. Findings suggest that the removal of price distortions raised housing
demand and equilibrium prices. Using variation in the timing of housing reform across cities,
Iyer, Meng, and Qian (2010) find that housing reform increases the proportion of households
living in private housing and lowers the proportion in public rental housing. Wang and Murie
(2000) and Wang (2000) examine differences across different social groups and land-use zones
(central vs. peripheral), focusing on Beijing.
In Figures 2a and 2b, we can see the rapid response of the market to the availability of
work-unit housing on the private market, as the 2008 market value of housing purchased from
12
work units increased considerably. In Zhejiang, the market value of housing purchased from
work units during the two major reform periods in 1992 and 1996 was relatively high. Gansu
being a much poorer province, has seen lower housing windfalls. But all of the windfalls in
Gansu have come from work-unit purchases. In Zheijiang, the average current market value of
housing purchased from the market increased substantially in 1996. This may have to do with the
resale of housing obtained during the earlier period of housing reform in the early 1990s. These
differences in market values also reflect the fact that the quality of housing purchased from work
unit was considerably higher than privately owned housing.
The degree to which households gained from the housing reform of the 1990s can also be
seen vividly when the 2008 market values are compared to the average prices for housing at the
time of purchase. In Figures 3a, we can see that work-unit purchases in Zhejiang were sold at
considerably lower prices than housing sold in the private market. In the early 1990s, prices were
extremely low and close to zero. In the second wave of housing form beginning in 1996, prices
for work-unit housing rose. This may reflect differences in the quality of housing made available,
or in the fact that by 1993, the Chinese government became concerned about the extremely low
prices at which danwei housing had been sold. Nonetheless, the purchase price for work-unit
bought housing was around half the price of market-bought housing. In addition, market prices
for housing rose considerably after the second wave of housing reforms in the mid to late 1990s.
In comparison, in Gansu where most houses purchased were purchased from work units, there
was not much change in purchase prices (see Figure 3b). Both work unit and market prices
remained extremely low. In comparing these figures with Figures 2a and 2b, it is clear that
households have benefited considerably from a rise in housing prices in recent years. In
Zhejiang, a private market for housing seems to have developed as a result. This is an important
13
point, since it indicates that housing assets were not completely illiquid assets. While in more
developed countries, housing assets tend to be relatively illiquid, it seems there were
considerable gains to selling a home that had been purchased from one’s work unit and may have
been located in a choice location of the city. The primary constraint in selling one’s home may
be due to a lack of affordable housing to find in its place. Further research is needed to
understand the extent to which such constraints have played a role. In terms of our analysis, such
constraints would underestimate our results, since we may overestimating windfall gains to
household wealth.
4. Data and Empirical Analysis
CHARLS Survey Data
We use data from a pilot study for the China Health and Retirement Longitudinal Survey
(CHARLS), which was conducted in 2008. Surveyed households were restricted to those with
members ages 45 and above. Much of the questionnaire was based on survey questions from the
Health and Retirement Study in the U.S. (HRS), the English Longitudinal Study of Ageing
(ELSA), and the Survey of Health, Ageing, and Retirement in Europe (SHARE). Two provinces
were chosen for the pilot study. Zhejiang is a relatively wealthy and highly urbanized coastal
province. In contrast, Gansu is the poorest province in China. It is located in the interior of the
country, and is populated by many non-Han minorities. In the two provinces, 48 communities or
villages were surveyed, belonging to 16 counties or districts. The overall sample was comprised
of 2,685 individuals living in 1,570 households.
In this study, since only official urban residents were affected by the housing reform, we
restricted our sample to those households in which the respondent or spouse held an urban hukou
14
or official urban registration, and the household was in an officially designated urban area. An
urban area was defined as an area with a designated community office rather than a village
office. Housing reforms were largely implemented by community offices, since work units or
danwei employers were located in these areas. This restricted our potential sample to 258
respondents. We restricted our sample of children of these respondents to those who were the
biological children of either the respondent or spouse, resulting in 588 children. Of these
children, 575 of them were adults (age 16 and above). The sample was further restricted to nonresident adult children, resulting in a sample of 475 children of 196 respondents.
The present value of all currently owned housing was summed to determine total
household wealth. Respondents were asked specifically about their housing assets and how they
were impacted by housing reform policies. The main respondent was asked detailed questions
regarding purchase of the house of residence. In addition, the main respondent and spouse were
asked about any housing they had purchased from a work unit. Questions included: who
purchased the house, who owned the house at the time of the survey, the year of purchase, the
purchase price, the market price at the time of purchase, and the current market price (at the time
of the survey). The latter price was available so long as the housing unit was the respondent’s
residence. If the unit had been sold prior to the time of the survey, then the current market price
was replaced with the price at which the housing unit had been sold. If the unit had not been
sold, and it was not listed as a residence (as was the case for 28 male and 7 female respondents),
then these values were imputed using the local community average market value for urban
housing bought from a danwei. For one respondent, there was no other such housing in the local
community; the county average market value for urban housing bought from a danwei was used
in this case. For one additional respondent, there was no such housing in the county, so the
15
current market value was imputed from the province average market value for urban housing
bought from a danwei.
Of the 475 children in our sample, the present market value of all housing currently
owned by their parents was available for 468 children, and the year housing was purchased by
their parents was provided for 457 children. The latter was needed to deflate the nominal value
of housing purchase prices to real terms. Thus, there were 7 children for whom we were missing
information on the present market value of their parents’ household wealth, resulting in a sample
size of 450 children. Five of these children were born to two respondents in Gansu whose current
residence was valued significantly higher than that of any other respondents. These extreme
outliers were excluded from our sample, as they more than doubled the average current market
value of a residence in urban Gansu. In addition, there were 16 respondents in urban Zhejiang
whose household wealth was more than five times greater than the average housing value for the
province. These respondents were also excluded from our sample.
In addition to questions regarding housing, respondents and spouses were asked detailed
questions regarding the value of household assets such as: a car, bike, refrigerator, washing
machine, TV, computer, and mobile phone. They were also asked about household and
individually-held financial assets such as cash, savings, stocks, and funds. In order to compute
total household wealth, we summed up all of these assets and deducted the household's
outstanding and unpaid loans and mortgage payments, and added it to our measure of household
wealth. Available data on this amalgamated measure of total household wealth resulted in a
sample of 395 observations.
This measure of total household wealth was instrumented by the total housing windfall,
which is the sum of: the difference between the current market price of all currently owned
16
housing and the original purchase price in real terms; and the difference between the price at
which housing was sold and the original purchase price in real terms. Thus, while differences in
household wealth may reflect some differences in lifetime earnings, the housing windfall is
independent of such differences because it varies depending on the price at which the housing
was originally purchased, which was exogenously determined by state policy.
For three current residences owned by the respondent, the original purchase price was
missing. As these residences had originally been purchased from a danwei, these prices were
imputed by the local community average purchase price in real terms for urban housing bought
from a danwei. Similarly, missing purchase price data for any danwei-bought housing that was
not a current residence was imputed in the same manner for 27 male respondents and 7 female
respondents. For one male respondent, there were no such prices in the local community, so the
county average purchase price for urban housing bought from a danwei was used instead.
Finally, in addition to these questions on asset-holdings, respondents were asked a
number of detailed questions about financial transfers. The first question in this section of the
survey was quite detailed, in order to remind the respondent of all of the different possible
sources of help they may have received or provided to others in the past year: “Did you or your
spouse receive more than 100 Yuan in financial help last year from any others? By financial help
we mean giving money; helping pay bills; covering specific costs, such as those for medical care
or insurance, schooling, and down payment for a home or rent; and providing non-monetary
goods.” The respondent was then asked: “What kind of help did you receive: regular or nonregular allowance? What is the total value of regular financial help you received? How much
non-regular financial help, such as an allowance, a living stipend, or money for school expenses,
did you receive in the past year?” Respondents also noted any transfers received during special
17
occasions such as festivals, holidays, birthdays, or marriages. The focus of our analysis was on
all of these received transfers combined, net of transfers provided to adult non-resident children.
We dropped six outliers on the variable for net transfers, which were above the 98th percentile for
this measure. With these outliers, average non-negative net transfers were 1307 RMB and
without these outliers, the average was 1048 RMB.
The resulting regression sample consisted of 389 respondents. For six of these children,
we did not know whether or not they were living in the same city as their parents. Of the
remaining 383 children, 343 children gave at least as much to their parents as they received from
them (non-negative net transfers from children), and 224 children gave more to their parents than
the amount they received from them (positive net transfers from children). Summary statistics on
all variables included in the analysis for our resulting sample are presented in Table 1. Means
and standard errors were estimated using household sampling weights. Nominal values were
deflated by the urban province-specific Brandt-Holtz deflator where possible (years 1986-2004).
For other years, nominal values were deflated by the urban province-specific consumer price
index (CPI) provided by the China National Bureau of Statistics in annual Statistical Yearbooks,
where the 2008 Zhejiang CPI was set to 100.
Linear Regression Analysis
We are interested in measuring the extent to which the value of total household assets
influences transfers received from non-resident adult children. To begin with, we estimate the
following OLS regression:
Transfersij = b0 + b1(Household Wealth)i + b2Xij + b3Xi + µc + εij
(10)
where Xij are child level controls, Xi are characteristics of the main respondent and spouse, and µc
are community fixed effects. Child characteristics include: age, birth order, gender, highest
18
education level attained, whether the child is currently working, whether the child is married,
whether the child lives in the same community or neighborhood as the parent, the number of
adult siblings, the average education level of adult siblings, and the average age of adult siblings.
The latter two variables are interacted with whether the child has any adult siblings.
Characteristics of the main respondent include: marital status, number of non-adult children,
household size, whether the CHARLS respondent lives with an adult child, age of household
head,7 age of household head squared, whether the household head was employed in the public
sector, dummies for the highest education level of the household head, and dummies for five
year intervals of the year of the first housing purchase interacted with whether a housing
purchase was made.
Estimates of b1 are likely to be biased since households owning a greater amount of
wealth may also behave differently than poorer households when it comes to decisions regarding
intergenerational transfers. For example, wealthier individuals may have better educated or
higher ability children who provide them with greater financial assistance, or alternatively, who
require less financial assistance from parents. In addition, reverse causality may be at play here.
Respondents who receive more transfers from children on a regular basis may save this source of
income and would have larger holdings of financial assets as a result.
For these reasons, we employ an instrumental variables approach, using estimates of the
housing market windfall as instruments for household wealth. The following two-stage leastsquares regressions are estimated using equations (11) and (12):
(Household Wealth)i = b0 + b1Hi + b2Xij + b3Xi + µc + εij
(11)
Transfersij = b0 + b1(Estimated Household Wealth)i + b2Xij+ b3Xi + µc + εij
(12)
7
By household head, we mean the husband if the respondent is married and living with a spouse, and the respondent
otherwise (whether male or female).
19
where Hi is our instrument for household wealth. The instrumental variable used here is the
housing market windfall, which is the difference between the current housing market value and
the price at which the house was originally purchased if it is still currently owned, combined
with the difference between the price at which any work unit housing was sold and its purchase
price.
There are a number of potential endogeneity issues here. Housing with higher markups
over time is likely to be better than average quality. Since housing wealth is included in our
measure of household wealth, the source of identification in instrumenting with housing windfall
is the housing purchase price. Another potential bias is that the type of person who is allocated
this type of housing and can afford to purchase it (perhaps even with a subsidy) may be different
from the average person. Some may choose to work in a work unit in order to receive subsidized
housing in a good location. For this reason, we control for characteristics of the household head,
such as education and whether he/she worked in the public sector. We also estimate regressions
where the housing windfall is the dependent variable and the unit of observation is the CHARLS
respondent or parent and transfer recipient. These estimates indicate that a significant
determinant of the windfall is the year of the initial housing purchase (see Table 4). In particular,
housing purchased in the early 1990s has a higher windfall. This reflects the relatively low price
for housing purchases during this early period of the reform. In addition, the number of adult
children is a statistically significant predictor of housing windfalls. We include it as a control
variable in all regressions. Relatedly, one might consider that living arrangement decisions of
children may be impacted directly by these changes in the housing market, which would also
impact intergenerational transfers. In fact, we find that living arrangement decisions of adult
children are not impacted by household wealth when instrumented by the housing windfall (see
20
Table 5). We also examine whether a child’s characteristics are affected by household wealth
when instrumented by the housing market windfall. We find that while education is not
influenced by parent’s wealth, a child’s age is negatively related to wealth and positively related
to the probability that an adult non-resident child is working. These characteristics are included
among the covariates in all regressions.
Non-Linearities and the Conditional Least Squares Model
In order to determine transfer motives, recent studies in developing countries examining
the relationship between transfers and recipient pre-transfer income have emphasized the nonlinearity of the relationship. While linear estimations do not necessarily point towards a strong
relationship between recipient income and transfers, authors do find evidence of a non-linear
relationship between recipient income and transfers (Cox et al. 2004, Cai et al. 2006).
We also find evidence of such non-linearity in our case. Using our main regression
sample for non-negative transfers, and controlling for the other covariates in equation (12), we
plot net transfer receipts as a function of recipient wealth instrumented by housing windfalls to
transfer recipients. More specifically, we regress net transfer receipts on Xij, Xi, and µc, and plot
the residuals from this regression on the y-axis. We then estimate our first stage equation (11) to
calculate predicted household wealth. We regress this predicted wealth on our covariates Xij, Xi,
and µc, and plot the residuals from this regression on the x-axis. We graph a lowess estimator for
the relationship between these residuals with a 0.99 bandwidth. Figure 4 displays the resulting Ushaped relationship between net transfer receipts and wealth levels. For relatively low wealth
levels, predicted wealth and transfer receipts are negatively related, pointing towards altruistic
motives. But as recipients’ predicted wealth increases, transfers flatten and then begin to
21
increase.8 Most of our observations are clustered around the first half on the initial U-shaped
curve. Nonetheless, the curvature follows the model proposed by Cox et al. 2004 (see their
Figure 1), where here we replace recipient income with recipient wealth. For relatively low
wealth levels, transfer receipts decline with wealth, as children provide less support to parents
who are less in need of it. These children are altruistically motivated to provide financial support
to their parents. But at a certain wealth level, transfer motives switch to an exchange motive,
where children provide more transfers as wealth levels increase. For these parents, children may
provide financial transfers in exchange for services such as care of grandchildren (Secondi
1997). As parent wealth increases, parents may want more reimbursement for their time until
parents become so wealthy that they want to devote less of their time to grandchild care, and
therefore receive fewer transfers from children.
In order to examine this potential relationship further, we follow the approach used by
Cox et al. 2004 and Cai et al. 2004. The latter authors showed that this method yields similar
results to those derived from a partial linear model (described in greater detail by Yatchew 1998,
2003). As the latter method requires a considerably large sample size, we implement the
conditional least squares model. In this model,
Transfersij = b0 + b1min(Wi, K) + b2max(0, Wi-K) + b3Xij + b4Xi + µc + εij
(13)
where Wi is recipient household wealth and K is the point at which transfer motives switch from
altruistic to exchange motives. If transfers are altruistically motivated, then the derivative
between transfers and wealth would be negative, since children would provide fewer transfers to
parents who are in less need of them. The other variables are defined as before. Similarly to Cai
8
The positive slope at higher wealth levels is identified from relatively few observations; it disappears if the largest
16 wealth residual observations are excluded. However, the kink in the downward slope remains (steep decline
followed by lesser decline) .
22
et al. (2006), we use net transfers received as the main dependent variable, and gross transfers
received as a robustness check (the latter results are available upon request from the authors).
As household wealth is potentially endogenous, we instrument this variable with housing
windfall (Hi). Thus, we estimate the following set of regressions using a two-step instrumental
variables tobit estimator:
Wi = b0 + b1Hi + b2Xij + b3Xi + µc + εij
(14)
Transfersij = b0 + b1 est. min(Wi, K) + b2 est. max(0, Wi-K) + b3Xij + b4Xi + µc + εij
(15)
In this model, K is the value of instrumented household wealth at which point transfer motives
change from altruism to exchange.
IV. Empirical Results
Relationship between Household Wealth and Net Transfer Receipts
Table 2 summarizes results from OLS and Tobit estimations where the net transfer
amount received from adult non-resident children is the dependent variable and household
wealth is the main independent variable. Where wealth is entered linearly into the regression, or
in those regressions where a squared wealth term is added, coefficient estimates on household
wealth are not statistically significant and are quite small.
However, in the Conditional Least Squares Model (CLM), coefficient estimates on
wealth below the kink point are negative, very high, and statistically significant. Also, the
estimated kink point is at the sixth or second percentile, which indicates that altruistic
motivations would only be relevant here for the very poorest households. Such estimates on
household wealth may be overestimating the effect of wealth on transfers if the poorest
households have children who are also very poor. If this is the case, then children may respond
23
very strongly to a marginal increase in parental wealth because they are less willing and able to
provide additional resources to parents whose need for support declines at the margin.
In all of these regressions, we also include controls for sibling characteristics to examine
the extent to which competition among siblings might also play a role in determining transfer
motives. Those who have an adult sibling provide significantly less transfers, indicating that
perhaps siblings share in the responsibility of providing support to their parents. On the other
hand, there is also some evidence of possible competition among siblings, as transfers are
positively related to the average education level attained by adult siblings. This positive
relationship may also be explained by the fact that those with better educated siblings may
themselves earn more as a result, and are therefore more willing and able to provide financial
support to their parents. Delineating between these two hypotheses would be an interesting
research question for future work in this area.
Adult siblings may play another role in determining financial transfers provided to
parents. If an adult sibling lives with the parent, then a non-resident child may provide fewer
transfers to the parent. Since money is generally fungible among household residents, at least
some of the support provided by a non-resident child to a parent may be shared with co-residing
children. In addition, this child may provide additional income to the household, so that parents
would be in less need of support from their non-resident children. While we have shown that
adult children are not more likely to live with parents who have received greater windfalls, the
presence of an adult child in the household reduces transfers from non-resident children,
particularly for the sample of non-negative net transfers. Coefficient estimates on this variable
are strongly negative and statistically significant for non-negative net transfers.
Estimating the effect of household wealth on net transfer receipts
24
By instrumenting for wealth with windfalls arising from housing reform, we can identify
the effects of an increase in wealth on transfers (see Table 3). In the first stage of the two-stage
least-squares regressions, coefficient estimates on housing windfall are statistically significant
and very close to one, indicating that the housing windfall is a strong predictor of household
wealth.
In the second stage of the two-stage least squares results, where household wealth is
instrumented by housing windfall and enters linearly into regressions on transfer receipts,
coefficient estimates are negative, but relatively small in magnitude and statistically
insignificant. However, when a squared term on predicted wealth is added to the regression,
coefficient estimates are higher, and for the subsample of positive net transfers, estimates are
statistically significant at the 1% level. An increase of 10,000 RMB in household wealth results
in a reduction in non-negative net transfers of approximately 17 RMB and a reduction in positive
net transfers of roughly 27 RMB. With average household wealth at 250,000 RMB and net
transfer receipts at 1031 RMB including zeros and1580 excluding zeros, these results imply that
a 4% average increase in household wealth reduces transfer receipts on average by roughly 1.7%.
Next, we apply the conditional least squares model with instrumental variables as
outlined above to examine whether the linear relationship between predicted household wealth
and net transfer receipts may be underestimating the degree to which transfer motives are
altruistic. Transfers at higher wealth levels may not be as greatly affected by marginal increases
in wealth, with altruism being strongest for lower levels of wealth. Results of these regressions
indicate there is a non-linear relationship between instrumented household wealth and net
transfer receipts. These non-linear estimates also indicate that linear regressions underestimate
the relationship between transfers and household wealth for those parents with wealth levels at or
25
below the 32nd or 43rd percentile (a predicted household wealth level of around 100,000-146,000
RMB). Beyond this kink point, the relationship between wealth and transfers becomes relatively
flat and coefficient estimates are not statistically significant. Thus, household wealth and transfer
receipts are more strongly negatively related at lower wealth levels, with altruism playing a
stronger role at these lower levels. An increase in household wealth of 10,000 RMB (4%
increase on average) results in a decline of 46 to 57 RMB in transfers, a decline of roughly 4.5%
or 3.6% on average, with the lower estimates resulting from the smaller subsample of positive
net transfers.
These estimates also indicate that non-instrumented estimates significantly overestimate
the effect of household wealth on transfer receipts. The poorest respondents may have poor
children who are less able to provide support to their parents, particularly when the parents
receive a marginal increase in wealth. In contrast, an exogenous increase in household wealth
indicates that parents do indeed receive fewer transfers from their children, particularly for those
with predicted wealth levels below the 32nd percentile. These results point towards evidence of
altruistic motivations for transfers from children for the lowest wealth households.
In addition, as in prior regressions, these results point to some potentially interesting
dynamics in the role of sibling characteristics in determining net transfer receipts. As before,
those who have an adult sibling provide significantly fewer transfers to parents, though
coefficient estimates are only statistically significant for the sample of non-negative net transfer
receipts. When zero transfers are excluded from the sample, coefficient estimates are imprecise.
Similarly, the average education level of adult siblings is positively related to net transfers,
though again, estimates are statistically significant for only the sample of non-negative net
transfers. As previously mentioned, these results may point towards some sharing of the
26
responsibility for caring for elderly parents, and also perhaps some competition among siblings
for currying favor with parents. But on the other hand, those with more educated siblings may be
wealthier as a result, and therefore may be better able to provide financial support to their
parents. Understanding these dynamics across siblings is an important question for future
research.
Finally, we include a dummy variable for living with an adult child in these regressions.
Coefficient estimates on this variable are strongly negative and statistically significant for nonnegative net transfers. Thus, the presence of an adult child in the household may deter some nonresident children from providing financial transfers to their parents.
Determinants of Household Wealth and Housing Windfall
In order to examine potential sources of selection bias in our instrument and instrumented
variable, we estimate regressions on household wealth and the housing windfall on a number of
covariates, with the unit of observation being the parent who receives the transfers and
potentially received a windfall in housing assets. The results summarized in Table 4 show that
one statistically significant predictor of household wealth is the number of adult children of the
parent. A respondent with one additional adult child has a greater wealth level of roughly 94,000
RMB, and received an additional windfall amount of roughly 184,000 RMB. However, the
number of adult children does not have a significant impact on wealth or windfalls for our
smaller sample of positive net transfers. Coefficient estimates are relatively low in magnitude
and statistically insignificant. For all other characteristics of the parent, coefficient estimates on
the windfall are relatively low in magnitude.
One significant factor in determining housing windfalls is the timing of the housing
purchase. Purchases made in the 1990s, particularly between 1990 and 1994, resulted in
27
significantly higher windfalls than purchases made in other years. In particular, a housing
purchase made between 1990 and 1994 resulted in an increase in the housing windfall of 180,000
to 390,000 RMB. With an average housing windfall of 109,000 RMB for our sample, this
implies that housing purchased in the early 1990s resulted in a windfall that was 65% greater
than the average windfall, and over 250% higher than average for the smaller sample of positive
net transfer recipients. This indicates the importance of the timing of housing purchases, as
government policies varied sale prices over the years of the reforms. As the timing is so
important, we control for the year of housing purchase in all our regressions.
Living arrangements and characteristics of adult children
While the empirical results above indicate that transfer motives are altruistic in this
context, the relationship between transfers and household wealth may be confounded by the fact
that parents with higher household wealth live in nicer and larger accommodations which may be
more attractive places for adult children to live in with their parents. The first column of Table 5
shows results of IV probit estimates of instrumented household wealth on the likelihood a child
lives with the parents. While previous regressions included only non-resident adult children,
these estimates also include co-resident adult children. Results indicate that children are not more
likely to live with parents whose household wealth is greater.
Regression results also indicate that household wealth as predicted by one’s housing
windfall does not impact a child’s educational attainment. To the extent that the highest
education level achieved is a reasonable proxy for income, these regressions indicate that
respondents with higher household wealth do not generally have better off children. However,
parent wealth does increase the likelihood that a child is currently working. When the dependent
variable is birth order, coefficient estimates on household wealth are also not statistically
28
significant, and magnitudes are relatively low. However, parents’ wealth is negatively related to
a child’s age. As wealthier parents have younger children who are also more likely to be
working, we control for these characteristics among others in all our regressions.
Sensitivity Analysis
The above results have shown the importance of examining the non-linear relationship
between recipient wealth and net transfer receipts. These results have also shown that children
are likely to be altruistically motivated in making transfers to their parents, particularly when
recipients have wealth levels below the 32nd percentile. In this section, we examine how robust
these findings are to variations in our model specification and sample.
In the above results, we had eliminated the top 2% in net transfer receipts as these were
outliers in our sample. In Appendix Table A1, we summarize the results of regression estimates
when these outliers are included in the sample. Coefficient estimates on total assets remain
negative for all regressions, and results of the second stage of CLM with IV are also robust to
including these outliers. While coefficient estimates on wealth levels below the kink point are
higher in magnitude for this sample, average net transfers are also considerably higher when
these outliers are included. For the sample of non-negative net transfers, the coefficient estimate
is statistically significant at the 10% level, and the wealth level at which point transfer motives
switch is similar to the previous sample (33rd percentile). For the smaller sample of positive net
transfers, the wealth level for a switch in transfer motives remains close to the 34th percentile,
and the coefficient estimate is statistically significant at the 1% level. Finally, coefficient
estimates on sibling variables are fairly similar to those in our main regressions, though
magnitudes differ somewhat in some cases.
29
In order to examine the role of adult siblings further, we estimate our regressions on the
subsample of children who have adult siblings. Our results are robust to using this subsample.
When a squared wealth term is included in the regression, coefficient estimates on total wealth
are negative and statistically significant at the 5% and 1% levels. For the second stage of CLM
with IV, coefficient estimates on wealth levels below the kink point are very similar to those in
our main regressions and they are statistically significant at the 1% level. The wealth level at
which point transfer motives switch from being altruistic to an exchange motive is at the 83rd
percentile for non-negative net transfers, and the 69th percentile for positive net transfers. Thus,
the kink point is considerably different when the sample is restricted to those with adult siblings
and zero net transfers are included in the sample. Finally, the number of adult siblings and their
average education level remain positively related to net transfer receipts. Thus, while sibling
competition may be at play here, these regressions primarily point towards strong evidence of
altruistic motivations behind transfers to parents when children have adult siblings.
Yet, the presence of an adult child living with the parent seems to have a strongly
negative effect on transfers from non-resident children. We examine this issue further by
restricting our sample to those children whose parents do not live with any adult children (see
Table A3). For this smaller sample, altruistic motivations are more pronounced than in prior
regressions, with considerably higher magnitudes in our coefficient estimates, which are
statistically significant at the 1% level. In addition, the wealth level at which point transfer
motives switch from altruism to exchange is much higher than in previous regressions. Thus, for
a majority of the sample of CHARLS respondents who do not live with an adult child, nonresident children are altruistically motivated in providing transfers to their parents. This further
points to the importance of other household members in determining transfers. Non-resident
30
children may be less altruistically motivated in providing support to parents who may share
transfers with co-resident siblings. Alternatively, parents may be less reliant on non-resident
children because of the support they receive from co-residing children.
Since co-residing adult children are an important factor in determining financial transfers
to parents, more broadly, proximity to parents may also be important here. In the second set of
regressions in Table A3, we restrict our sample to those children who live in the same city as
their parents. If parents received a significant housing windfall, housing costs in the city they live
in may be relatively high. For non-resident children living in the same city, such high housing
costs may counteract altruistic motivations for providing transfers. In fact, results of CLM with
IV are robust to this subsample. Coefficient estimates on wealth levels below the kink point are
similar in magnitude to those of prior regressions, and they are statistically significant at the 1%
level. In addition, coefficient estimates for wealth levels above the kink point are positive and
statistically significant. For this sub-sample, the switch in transfer motives is most apparent.
However, transfers are altruistically motivated for the majority of the subsample. Non-resident
children may be more altruistically linked to parents when they live closer to them.
We also examine this issue by including an interaction term between recipient wealth and
a dummy variable for the child living in the same city as the parent. Here, the wealth kink point
is quite high, at around the 95th percentile. In addition, coefficients on wealth for the wider
sample of non-negative net transfers are much smaller in magnitude and are not statistically
significant. There does not seem to be much of a relationship between non-negative net transfers
and wealth when it is interacted with a dummy for living in the same city as parents. But for our
subsample of positive net transfers, coefficient estimates are very similar in magnitude to those
in our main regressions, and they are statistically significant. These differences may reflect
31
differences between those children who are net givers to parents and those who do not provide
any support.
Finally, as previously mentioned, we imputed a number of missing values on market
values and purchase prices for housing. Our results are robust to excluding these imputed values.
Coefficient estimates on the wealth level below the kink point are statistically significant at the
1% level. For our wider sample of non-negative net transfers, the coefficient on wealth below the
kink point is higher in magnitude than in our main regressions, and for the smaller sample of
positive net transfers, the coefficient is smaller in magnitude. For the sample of non-negative net
transfers, the wealth kink point is fairly similar to that of previous regressions, but for the smaller
sample of positive net transfers, it is much higher (84th percentile). Thus, there is some
sensitivity in our estimates to these variations and it is important to keep this in mind when we
consider how strong of a role altruism plays; the extent to which altruism is important may vary
somewhat for different populations within our sample.
Nonetheless, our general finding that transfer motives are primarily altruistic is robust to
numerous modifications to our model specification and sample. It is apparent that the
relationship between transfers and recipient wealth is non-linear. Particularly for the bottom third
of our sample in terms of recipient wealth, transfers from non-resident children appear to be
altruistically motivated.
6. Conclusion
We have shown that housing reform policies have significantly raised housing market
values. Employees of work units were allocated housing by their employers, were then given the
right to buy this housing at subsidized rates, and benefited considerably as private housing
32
markets developed and housing prices rose considerably. We have shown that such windfalls
have had a considerable impact on household wealth, and that family members take this into
consideration when providing financial help to windfall recipients. Those who benefited from
such windfalls receive considerably less financial help from children, indicating that transfers
from children are likely to be altruistically motivated.
33
References
Altonji, Hayashi, and Kotlikoff. 1997. “Parental Altruism and Inter Vivos Transfers: Theory and
Evidence,” Journal of Political Economy 105(6): 1121-66.
Becker, Gary S. 1974. “A Theory of Social Interactions.” Journal of Political Economy 82:
1063-94.
Becker, Gary S. 1981. A Treatise on the Family. Cambridge: Harvard University Press.
Bernheim, Shleifer and Summers. 1985. “The Strategic Bequest Motive.” Journal of Political
Economy 93(6): 1045-1076.
Cai, F., J. Giles, and X. Meng. 2006. "How Well Do Children Insure Parents Against Low
Retirement Income? An Analysis Using Survey Data from Urban China." Journal of
Public Economics 90(12): 2229-2255.
Cox. 1987. “Motives for Private Income Transfers.” Journal of Political Economy 95(3): 508546.
Cox. 1990. “Intergenerational transfers and liquidity constraints.” Quarterly Journal of
Economics 105(1): 187-217.
Cox, Eser, and Jimenez. 1998. “Motives for Private Transfers over the Life Cycle: An Analytical
Framework and Evidence for Peru.” Journal of Development Economics 55: 57-80.
Cox, Jansen, and Jimenez. 2004. “How responsive are transfers to income? Evidence from a
laissez-faire economy.” Journal of Public Economics 88(9-10): 2193-2219.
Cox and Jappelli. 1990. “Credit rationing and private transfers: evidence from survey data.”
Review of Economics and Statistics 72(3): 445-454.
34
Cox and Jimenez. 1992. “Social Security and Private Transfers in Developing Countries: The
Case of Peru.” World Bank Economic Review 6(1): 155-169.
Cox, Jimenez, and Okrasa. 1997. “Family Safety Nets and Economic Transition: A Study of
Worker Households in Poland.” Review of Income and Wealth 43(2): 191-209.
Cox and Rank. 1992. “Inter-vivos Transfers and Intergenerational Exchange.” Review of
Economics and Statistics 74(2): 305-314.
Guiso and Jappelli. 1991. “Intergenerational transfers and capital market imperfections:
Evidence from a cross-section of Italian households.” European Economic Review 35(1):
103-120.
Hoddinott, J. 1992. “Rotten Kids or Manipulative Parents: Are Children Old Age Security in
Western Kenya?” Economic Development and Cultural Change 40: 545-566.
Iyer, Meng, and Qian. “Estimating the value of individual property rights: evidence from China’s
urban housing reforms.” Working Paper.
Jensen, R.T. 2004. "Do Private Transfers Displace the Benefits of Public Transfers? Evidence
from South Africa." Journal of Public Economics 88: 89-112.
Kazianga, H. 2006. “Motives for household private transfers in Burkina Faso.” Journal of
Development Economics 79(1): 73-117.
La Ferrara, Eliana. 2007. “Descent rules and strategic transfers. Evidence from matrilineal
groups in Ghana.” Journal of Development Economics 83: 280-301.
McGarry. 1999. “Inter vivos transfers and intended bequests.” Journal of Public Economics 73:
321-351.
McGarry and Schoeni. 1995. "Transfer behavior: Measurement, and the redistribution of
resources within the family." Journal of Human Resources 30: S184—S226.
35
Raut, Lakshmi K. and Lien H. Tran. 2005. “Parental human capital investment and old-age
transfers from children: Is it a loan contract or reciprocity for Indonesian families?”
Journal of Development Economics 77: 389-414.
Schoeni, Robert. 1997. “Private Interhousehold Transfers of Money and Time: New Empirical
Evidence.” Review of Income and Wealth 43(4): 423-448.
Secondi, G. 1997. “Private Monetary Transfers in Rural China: Are Families Altruistic?”
Journal of Development Studies 33(4): 487-511.
Sloan, Zhang, and Wang. 2002. “Upstream intergenerational transfers.” Southern Economic
Journal 69(2): 363-380.
Wang, Shing-Yi. “State Misallocation and Housing Prices: Theory and Evidence from China.”
New York University Unpublished Manuscript.
Wang, Ya Ping. 2000. “Housing reform and its impacts on the urban poor in China.” Housing
Studies 15(6): 845-864.
Wang, Ya Ping and Alan Murie. 2000. “Social and spatial implications of housing reform in
China.” International Journal of Urban and Regional Research 24(2): 397-417.
Wolff, Spilerman, and Attias-Donfut. 2007. “Transfers from migrants to their children: evidence
that altruism and cultural factors matter.” Review of Income and Wealth 53(4): 619-644.
Yatchew. 1998. “Non-parametric regression techniques in economics.” Journal of Economic
Literature 36: 669-721.
Yatchew. 2003. Semiparametric Regression for the Applied Econometrician. Cambridge
University Press: Cambridge.
36
Figure 1a. Number of Housing Purchases in
urban Zhejiang
30
Number of Houses Purchased
Number of houses bought in the private market
25
Number of houses bought from danwei
20
15
10
5
0
1980 to 1984
1985 to 1989
1990 to 1994
1995 to 1999
2000 to 2004
2004 to 2008
Year of Purchase
Figure 1b. Number of Housing Purchases in
urban Gansu
30
Number of Houses Purchased
Number of houses bought in the private market
25
Number of houses bought from danwei
20
15
10
5
0
1980 to 1984
1985 to 1989
1990 to 1994
1995 to 1999
Year of Purchase
37
2000 to 2004
2004 to 2008
Average Current Market Value per Square Meter (1000
RMB per square meter)
Figure 2a. Current Market Value of Housing in
urban Zhejiang
0.07
0.06
Current Market Value of Privately Bought Housing
Current Market Value of danwei bought housing
0.05
0.04
0.03
0.02
0.01
0
1980 to 1984 1985 to 1989 1990 to 1994 1995 to 1999 2000 to 2004 2004 to 2008
Year of Purchase
All values are deflated by the urban province-specificCPI, where the 2008 Zhejiang
CPI=100. The Brandt-Holtz delfator is used for 1986-2004. For all other years, the urban
province-specific CPI is from the China Statistical Yearbooks.
38
Average Current Market Value per Square Meter (1000 RMB
per square meter)
Figure 2b. Current Market Value of Housing in
urban Gansu
0.014
0.012
Current Market Value of Privately Bought Housing
Current Market Value of danwei bought housing
0.01
0.008
0.006
0.004
0.002
0
1980 to 1984 1985 to 1989 1990 to 1994 1995 to 1999 2000 to 2004 2004 to 2008
Year of Purchase
All values are deflated by the urban province-specificCPI, where the 2008 Zhejiang
CPI=100. The Brandt-Holtz delfator is used for 1986-2004. For all other years, the urban
province-specific CPI is from the China Statistical Yearbooks.
39
Average Purchase Price per Square Meter (1000 RMB per
square meter)
Figure 3a. Purchase Prices of Housing in urban
Zhejiang
0.07
Purchase Price in Private Market
0.06
Purchase Price of danwei bought housing
0.05
0.04
0.03
0.02
0.01
0
1980 to 1984
1985 to 1989
1990 to 1994 1995 to 1999
Year of Purchase
2000 to 2004
2004 to 2008
All values are deflated by the urban province-specificCPI, where the 2008 Zhejiang CPI=100.
The Brandt-Holtz delfator is used for 1986-2004. For all other years, the urban provincespecific CPI is from the China Statistical Yearbooks.
40
Average Purchase Price per Square Meter (1000 RMB per
square meter)
Figure 3b. Purchase Prices of Housing in urban
Gansu
0.014
Purchase Price in Private Market
Purchase Price of danwei bought housing
0.012
0.01
0.008
0.006
0.004
0.002
0
1980 to 1984
1985 to 1989
1990 to 1994 1995 to 1999
Year of Purchase
2000 to 2004
2004 to 2008
All values are deflated by the urban province-specificCPI, where the 2008 Zhejiang CPI=100.
The Brandt-Holtz delfator is used for 1986-2004. For all other years, the urban provincespecific CPI is from the China Statistical Yearbooks.
41
800
-200
0
200
400
600
Figure 4a. Non-Negative Net Transfers and Recipient Wealth
-50
0
Recipient Wealth Residuals
50
Note: 3 out of 343 Observations taken out (wealth residuals greater than 100 taken out).
42
Table 1. Summary Statistics
Variable
Obs Weight Mean Std. Dev. Min
Max
Net transfers received (RMB)
343
130 1031.21 1757.12
0.0 10794.9
Total household wealth (10,000 RMB)
343
130
25.84
42.53
-3.6
309.3
Housing windfall (10,000 RMB)
343
130
10.86
23.41
-4.1
200.0
Gender of child
343
130
1.50
0.50
1.0
2.0
Household head is married = 1
343
130
0.70
0.46
0.0
1.0
Household size
343
130
2.49
1.19
1.0
6.0
Child lives outside local area = 1
343
130
0.77
0.42
0.0
1.0
Housing purchase was made = 1
343
130
0.02
0.15
0.0
1.0
Year of first housing purchase
198
82 1994.45
7.56 1970.0 2008.0
Male head of household = 1
343
130
0.75
0.43
0.0
1.0
Age of household head
343
130
66.64
9.57
47.0
86.0
Education level of child
343
130
5.38
1.69
1.0
10.0
Child is married = 1
343
130
0.83
0.37
0.0
1.0
Child is currently working = 1
343
130
0.77
0.42
0.0
1.0
Age of child
343
130
38.34
10.05
16.0
77.0
Birth order of child
343
130
1.95
1.20
1.0
8.0
Number of adult siblings of child
343
130
2.06
1.55
0.0
7.0
Child has adult sibling(s) = 1
343
130
0.87
0.34
0.0
1.0
Mean education of adult siblings
343
130
4.48
2.20
0.0
9.0
Mean age of adult siblings
343
130
33.78
15.43
0.0
58.5
Number of non-adult children
343
130
0.02
0.15
0.0
1.0
Household head employed in public sector = 1 343
130
0.34
0.47
0.0
1.0
Education level of household head
343
130
3.90
2.32
1.0
12.0
Child lives in same city as parent = 1
343
130
0.74
0.44
0.0
1.0
Respondent lives with an adult child = 1
343
130
0.33
0.47
0.0
1.0
Sample used in main regression results (non-resident adult children, non-negative net transfers).
43
Table 2. Relationship between Household Wealth and Net Transfers Received from Children (Tobit and OLS)
Linear wealth term only
Total Assets (10,000 RMB)
Squared wealth term included
Non-Negative Net
Positive Net
Transfers (Tobit) Transfers (OLS)
3.202
3.763
(3.374)
(5.126)
Total Assets Squared
Non-Negative
Net Transfers
(Tobit)
1.28
(8.526)
0.007
(0.030)
Positive Net
Transfers
(OLS)
-5.682
(10.833)
0.038
(0.040)
Wealth Levels Below Kink Point
Conditional Least Squares Model
Non-Negative
Net Transfers
(Tobit)
Observations
R-squared or Pseduo R-squared
150.206
(144.558)
-4,981.140*
(2629.362)
466.092***
(152.449)
7.225
(59.625)
-1,307.010**
(586.617)
343
0.0382
17.004
(158.696)
-3316.372
(2771.378)
143.205
(146.998)
11.504
(65.260)
-827.778
(635.268)
224
0.4298
146.391
(145.460)
-4,960.787*
(2655.077)
463.972***
(154.528)
7.084
(59.823)
-1,311.775**
(594.419)
343
0.0382
0.342
(163.746)
-3272.240
(2818.450)
130.661
(151.727)
10.236
(65.969)
-809.423
(654.451)
224
0.435
-1,534.420**
(609.507)
3.529
(3.408)
149.345
(140.503)
-5,706.099**
(2639.266)
485.178***
(152.326)
18.718
(59.624)
-1,340.276**
(580.071)
343
0.0398
Total Wealth Kink Point
Percentile for Wealth Kink Point
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
0.0524
5.52%
Wealth Levels Above Kink Point
Number of adult siblings
Have adult sibling = 1
Avg education of adult siblings
Avg age of adult siblings
Parents lives with adult child = 1
Notes: *** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses, clustered by CHARLS respondent.
Sample exclues outliers on housing windfall, housing wealth, and net transfers.
Sample includes all non-resident adult children.
44
Positive Net
Transfers (OLS)
-13,095.065***
(2213.226)
6.856
(5.731)
-28.493
(165.367)
-4,126.210*
(2407.259)
175.437
(136.385)
27.491
(58.735)
-295.378
(527.437)
224
0.4789
0.0398
2.10%
Table 3. Effect of Household Wealth on Net Transfers Received from Children (2SLS)
First Stage of 2SLS
Windfall (10,000 RMB)
NonPositive
Negative Net
Net
Transfers
Transfers
1.220***
1.181***
(0.140)
(0.080)
Total Assets (10,000 RMB)
2nd Stage of 2SLS
NonNegative
Net
Positive Net
Transfers
Transfers
(IV T-Tobit)
(IV)
2nd Stage of 2SLS
NonNegative
Net
Positive Net
Transfers (IV Transfers
T-Tobit)
(IV)
-2.806
(3.090)
-17.375*
(10.236)
0.053
(0.034)
-4.445
(3.699)
Total Assets Squared
2nd Stage of CLM
NonNegative
Net
Positive Net
Transfers (IV Transfers
T-Tobit)
(IV)
-27.454***
(7.505)
0.090***
(0.029)
Wealth Levels Below Kink Point
Number of adult siblings
7.393***
(2.555)
Have adult sibling = 1
-32.857
(21.917)
Avg education of adult siblings 0.667
(1.988)
Avg age of adult siblings
0.577
(0.491)
Parents lives with adult child = 1 14.877
(10.034)
Observations
343
R-squared or Pseduo R-squared 0.8877
11.444***
(2.742)
-24.506
(30.326)
2.509
(2.366)
0.124
(0.773)
5.222
(10.737)
224
0.9426
224.851
(136.535)
-4,988.219*
(2636.004)
469.797***
(155.877)
5.704
(59.820)
-1,176.304**
(566.129)
343
0.0381
152.376
(148.152)
-3,100.86
(2776.119)
134.364
(150.801)
9.556
(64.726)
-717.397
(633.355)
224
0.4299
247.807*
(138.244)
-5,079.887*
(2612.729)
464.951***
(154.523)
7.857
(59.111)
-1,087.840*
(553.794)
343
0.0387
190.701
(157.023)
-3,333.22
(2775.759)
138.44
(150.472)
8.574
(64.562)
-625.538
(615.997)
224
0.4594
-45.584**
(20.292)
-1.814
(3.120)
318.878**
(149.363)
-5,537.062**
(2602.398)
517.103***
(151.400)
12.511
(59.307)
-1,020.683*
(544.716)
343
0.039
Total Wealth Kink Point
n/a
Percentile for Wealth Kink Point n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
9.96
32.36%
Wealth Levels Above Kink Point
-56.984***
(17.668)
-2.618
(3.776)
326.133**
(157.579)
-4,284.371
(2617.934)
214.301
(143.682)
21.237
(61.889)
-553.249
(569.049)
224
0.461
14.59
43.30%
Robust standard errors in parentheses, clustered by CHARLS respondent. *** p<0.01, ** p<0.05, * p<0.1 Sample excludes outliers on housing
windfall, housing wealth, and net transfers. Sample includes all non-resident adult children.
45
Table 4. Determinants of Household Wealth and Housing Windfall
Total Household Wealth
Housing Windfall
Non-Negative Positive Net Non-Negative Positive Net
Net Transfers
Transfers
Net Transfers
Transfers
Made housing puchase = 1
-13.572
8.808
1.588
13.946
(25.320)
(17.650)
(34.402)
(23.868)
Housing purchase in 1980-84
-16.847
9.251
-2.08
-2.264
(27.122)
(18.906)
(47.876)
(33.217)
Housing purchase in 1985-89
2.541
12.653
-2.169
11.056
(21.848)
(15.230)
(48.536)
(33.675)
Housing purchase in 1990-94
11.893
33.925***
18.052
39.151*
(16.039)
(11.180)
(29.494)
(20.463)
Housing purchase in 1995-99
2.741
16.544*
-2.331
10.276
(14.178)
(9.883)
(21.937)
(15.220)
Housing purchase in 2000-04
-12.377
10.579
-24.302
4.319
(18.839)
(13.132)
(27.960)
(19.399)
Housing purchase in 2005-09
5.547
19.623
26.829
24.753
(30.273)
(21.102)
(40.800)
(28.308)
Number of adult children
9.400**
2.67
18.358***
6.927
(4.003)
(2.790)
(6.441)
(4.469)
Number of adult sons
4.393
-1.526
-4.798
-6.767
(5.740)
(4.001)
(9.248)
(6.416)
Live with adult child = 1
20.975
9.269
10.935
11.208
(20.534)
(14.314)
(31.225)
(21.664)
Married Respondent
-9.188
2.749
48.745
5.504
(24.122)
(16.815)
(75.705)
(52.525)
Household Size
1.083
-0.521
0.204
-0.389
(5.965)
(4.158)
(9.698)
(6.729)
% children living outside area
-0.036
-0.053
-0.053
-0.198
(0.165)
(0.115)
(0.309)
(0.214)
Household head Is male = 1
18.73
-2.449
-34.241
-3.908
(25.030)
(17.448)
(72.707)
(50.445)
Age of household head
-4.972
0.689
2.432
6.64
(9.199)
(6.412)
(16.095)
(11.167)
Age of household head squared
0.024
-0.005
-0.039
-0.048
(0.067)
(0.046)
(0.117)
(0.081)
Household head works in public sector 9.292
12.063
22.99
18.368
(11.468)
(7.994)
(21.491)
(14.911)
Mean children's education
0.867
-2.401
3.665
-2.574
(4.870)
(3.395)
(8.357)
(5.798)
Percent married children
0.046
0.029
-0.056
0.026
(0.254)
(0.177)
(0.417)
(0.289)
Percent working children
0.17
0.126
0.101
0.334
(0.206)
(0.144)
(0.401)
(0.278)
Mean age of children
0.265
-0.75
1.329
-0.964
(1.357)
(0.946)
(2.472)
(1.715)
Observations
105
105
75
75
R-squared
0.6822
0.5014
0.7273
0.5646
Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Unit of observation is the CHARLS
respondent, or parent recipient. Additional controls include number of non-adult children,
household head education dummies and community fixed effects.
46
Table 5. Effect of Household Wealth on Child Characteristics
All Adult Children
Likelihood
child lives
with parent1
IV Probit
Household Wealth (10,000 RMB) -0.003
(0.004)
Observations
388
Wald Chi-squared or R-squared 308.6
x
Likelihood
child is
Age of Child x
working
2SLS
IV Probit
-0.010**
0.013*
(0.005)
(0.007)
431
325
0.9251
1323
Adult Non-Resident Children
Education of
Child
2SLS
0.000
(0.002)
343
0.5371
Age of Child
2SLS
-0.010*
(0.005)
343
0.9166
Birth Order of
Child
2SLS
-0.001
(0.001)
343
0.8085
Sample includes non-negative net transfers. Robust standard errors in parentheses, clustered by CHARLS respondent. ***
p<0.01, ** p<0.05, * p<0.1 First stage results are not shown here as they are qualitatively similar to the first stage results
when the dependent variable is transfers.
1
Dummy for parent living with adult child is excluded here. All other controls are the same as those in the main
regressions. For this reason, there is a smaller sample size than one might expect in the regressions on the likelihood that
the child lives with the parent; community fixed effects are included in these regressions.
47
Appendix Table A1. Effect of Household Wealth on Net Transfers Received from Children (Including Outliers)
Windfall (10,000 RMB)
First Stage of 2SLS
NonNegative Positive
Net
Net
Transfers Transfers
1.227*** 1.184***
(0.140)
(0.080)
Total Assets (10,000 RMB)
2nd Stage of 2SLS
NonPositive
Negative Net
Net
Transfers (IV Transfers
T-Tobit)
(IV)
2nd Stage of 2SLS
NonPositive
Negative Net
Net
Transfers (IV Transfers
T-Tobit)
(IV)
-9.665
(6.453)
-18.988
(15.612)
0.034
(0.046)
-7.607
(5.797)
Total Assets Squared
10.958***
(2.759)
-29.308
(29.127)
2.972
(2.325)
0.178
(0.818)
4.641
(10.856)
230
0.9408
541.155**
(242.944)
-10,013.562**
(4163.056)
761.231***
(265.642)
26.707
(87.499)
-2,018.268***
(770.482)
349
0.0365
571.843*
(318.886)
-8,354.36
(5075.329)
318.308
(301.241)
54.174
(122.824)
-1,409.120*
(737.118)
230
0.4283
556.494**
(247.381)
-10,151.706**
(4128.610)
761.048***
(265.431)
29.07
(86.572)
-1,958.773**
(767.603)
349
0.0366
593.543*
(333.398)
-8,792.972*
(5221.429)
336.946
(303.804)
57.151
(126.430)
-1,368.099*
(736.230)
230
0.4329
-57.776*
(29.868)
-8.38
(6.962)
650.250**
(260.331)
-10,765.211***
(4085.064)
821.588***
(257.673)
36.29
(85.739)
-1,824.287**
(760.926)
349
0.0369
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
10.53
32.95%
Wealth Levels Above Kink Point
Have adult sibling = 1
Avg education of adult siblings
Avg age of adult siblings
Parent lives with adult child = 1
Observations
R-squared or Pseduo R-squared
7.489***
(2.495)
-37.266*
(20.881)
0.809
(1.997)
0.632
(0.493)
14.873
(10.121)
349
0.8867
Total Wealth Kink Point
n/a
Percentile for Wealth Kink Point n/a
Non-Negative
Net Transfers Positive Net
(IV T-Tobit) Transfers (IV)
-23.796*
(13.300)
0.063
(0.045)
Wealth Levels Below Kink Point
Number of adult siblings
2nd Stage of CLM
-79.445***
(29.286)
-5.373
(6.630)
764.781**
(356.378)
-10,096.232**
(4936.051)
423.995
(292.575)
78.273
(122.640)
-1,193.487*
(685.082)
230
0.4432
11.78
33.91%
Robust standard errors in parentheses, clustered by CHARLS respondent. *** p<0.01, ** p<0.05, * p<0.1 Sample excludes outliers on housing
windfall and housing wealth. Sample includes all non-resident adult children.
48
Appendix Table A2. Effect of Household Wealth on Net Transfers Received from Children (Sample with adult siblings)
Windfall (10,000 RMB)
First Stage of 2SLS
NonNegative
Positive
Net
Net
Transfers Transfers
1.186***
1.136***
(0.130)
(0.073)
Total Assets (10,000 RMB)
2nd Stage of 2SLS
Positive
Non-Negative
Net
Net Transfers Transfers
(IV T-Tobit)
(IV)
2nd Stage of 2SLS
NonNegative
Positive
Net
Net
Transfers Transfers
(IV T-Tobit)
(IV)
0.138
(3.521)
-26.626**
(12.174)
0.099**
(0.045)
-1.106
(3.654)
Total Assets Squared
-36.984***
(6.121)
0.145***
(0.021)
Wealth Levels Below Kink Point
Wealth Levels Above Kink Point
Number of adult siblings
Avg education of adult siblings
Avg age of adult siblings
Parent lives with an adult child = 1
Observations
R-squared or Pseduo R-squared
7.354**
(2.835)
0.799
(2.003)
0.496
(0.469)
12.941
(10.159)
326
0.8961
10.091***
(3.308)
0.957
(2.236)
-0.838
(0.602)
4.171
(10.808)
215
0.9529
312.620*
(180.147)
446.562***
(155.930)
-15.158
(60.965)
-1,063.768**
(521.044)
326
0.0339
220.469
(208.800)
69.436
(154.418)
-23.626
(65.477)
-457.577
(572.220)
215
0.4407
372.910**
(179.908)
444.393***
(152.611)
-11.553
(59.303)
-923.086*
(516.981)
326
0.0358
2nd Stage of CLM
Positive
Non-Negative
Net
Net Transfers Transfers
(IV T-Tobit)
(IV)
358.038**
(174.271)
62.108
(146.698)
-12.637
(60.004)
-231.018
(512.134)
215
0.502
-47.473***
(17.966)
6.318
(3.948)
489.449**
(196.066)
488.729***
(151.275)
-12.717
(58.842)
-605.454
(487.879)
326
0.0367
-73.949***
(10.561)
5.141**
(2.520)
633.461***
(149.967)
174.512
(136.832)
-34.69
(59.197)
112.675
(435.153)
215
0.5180
Total Wealth Kink Point
n/a
n/a
n/a
n/a
n/a
n/a
38.9093
27.49
Percentile for Wealth Kink Point
n/a
n/a
n/a
n/a
n/a
n/a
82.82%
69.30%
Robust standard errors in parentheses, clustered by CHARLS respondent. *** p<0.01, ** p<0.05, * p<0.1. Sample excludes outliers on housing
windfall, housing wealth, and net transfers.
49
Appendix Table A3. Sensitivity Analysis: Conditional Least Squares Model
Non-Negative Net Transfers
Positive Net Transfers
T-Tobit
IV T-Tobit
OLS
2SLS
Sample includes only those children whose parents do not live with any adult children
Wealth Kink Point (10,000 RMB) 0.02
36.59
0.15
28.84
Wealth Kink Point Percentile
2.56%
81.18%
7.18%
74.59%
Total Wealth Below Kink
-12,641.022*** -91.462***
-9,201.793*** -88.457***
(15.535)
(2.834)
(2688.223)
(19.989)
Total Wealth Above Kink
-1.359
4.013***
-1.327
-0.085
(1.096)
(0.941)
(5.081)
(3.524)
Observations
255
255
181
181
R-squared or Pseduo R-squared 0.0546
0.0551
0.5245
0.5526
Sample includes only those children who live in the same city as parents
Wealth Kink Point (10,000 RMB) 0.016
37.77
201.12
27.11
Wealth Kink Point Percentile
1.83%
79.09%
97.33%
68.16%
Total Wealth Below Kink
-19,159.059*** -46.398***
3.705
-60.613***
(2710.663)
(13.497)
(4.893)
(10.691)
Total Wealth Above Kink
8.295**
10.393***
47.017***
8.441**
(3.525)
(3.904)
(12.783)
(3.340)
Observations
263
263
179
179
R-squared or Pseduo R-squared 0.0523
0.0485
0.5868
0.6091
Regressions include interaction between total wealth & living in same city as parents
Wealth Kink Point (10,000 RMB) 10.12
116.11
0.04
110.55
Wealth Kink Point Percentile
48.90%
95.34%
2.10%
95.54%
Total Wealth Below Kink
59.968
-15.868
-13,312.734*** -41.160***
(98.986)
(12.541)
(2722.885)
(12.059)
Total Wealth Above Kink
0.427
0.775
5.203
15.067**
(4.589)
(7.469)
(5.640)
(7.111)
Live in same city = 1
1,003.335*
-91.089
-442.005
-576.491
(510.725)
(353.582)
(366.134)
(476.948)
Interaction w/wealth below kink 5.157*
5.837
2.342
-5.076
(2.666)
(6.821)
(2.225)
(6.307)
Interaction w/wealth above kink -189.925**
-1.177
-151.818
17.539
(75.497)
(9.828)
(1443.952)
(11.593)
Observations
343
343
224
224
R-squared or Pseduo R-squared 0.0405
0.0391
0.4834
0.4678
No imputed values for total wealth or windfall
Wealth Kink Point (10,000 RMB) 0.040
6.58
0.04
43.57
Wealth Kink Point Percentile
3.75%
29.64%
2.19%
83.56%
Total Wealth Below Kink
-1,590.295*** -75.674***
-11,841.599*** -38.998***
(53.469)
(5.197)
(2193.608)
(9.186)
Total Wealth Above Kink
6.795***
1.616**
8.957*
7.596**
(0.660)
(0.661)
(5.279)
(3.469)
Observations
334
334
219
219
R-squared or Pseduo R-squared 0.0409
0.0408
0.5217
0.5143
Robust standard errors in parentheses, clustered by CHARLS respondent. *** p<0.01, ** p<0.05, * p<0.1
50