Does the exchange motive influence intergenerational transfers? Evidence from Ireland

Abstract

This article examines the exchange motive in intergenerational monetary transfers. The exchange motive is in operation if parents make transfers to their children in exchange for services. The analysis incorporates data on current inter vivos transfers and the self-assessed probability of making future transfers via bequests. The focus is on the correlation between child-provided help and transfers from parents to adult children. Cross-sectionally, small transfers (between EUR 250 and EUR 5,000) and child-provided help are positively related. Endogeneity arising from omitted variables and reverse causality are addressed by using a lagged value of child-provided help. Further, in an analysis of first-differenced data, the association is statistically significant at the 10% level of confidence. The correlation is not significant in the case of large (above EUR 5,000) transfers, or the self-assessed probability of making future transfers via bequests. These findings suggest that small inter vivos transfers may be driven by exchange motives, although the findings are also consistent with altruism.

1 Introduction

This article examines the presence of the exchange motive in intergenerational monetary transfers. The exchange motive is in operation if parents make transfers to their children in exchange for services. The main alternative explanation for intergenerational transfers is altruistic behaviour, meaning that parents derive utility from their children’s utility. The analysis aims to explore the existence of the exchange motive, by examining the link between child-provided services and transfers from parents to children. Cox and Rank (1992) suggest this approach but it has rarely been implemented in empirical work. In contrast, a more often-explored way to identify transfer motives is to estimate the effect of child income on the value of transfers. The difficulties with the latter approach are that the exchange motive can only be identified if the demand for child-provided services is own-price inelastic, and that it requires longitudinal data on the value of transfers and child income. For these reasons, this analysis adopts the former approach. This work expands on previous literature by using longitudinal data from two survey waves, and by examining whether the relationship between child-provided care and transfers differs depending on the size of the transfer.

The data for this analysis are obtained from the first two waves of The Irish Longitudinal Study on Ageing (TILDA), focusing on a sample of parent households who have non-resident adult children. Both kinds of monetary intergenerational transfers are modelled: inter vivos transfers (while parents are alive) and future transfers via bequests (after parents’ death). Data on the two transfer types are reported differently in the data: while the inter vivos transfer measure available in TILDA measures the transfers that parents are making to their children between the survey waves, the bequest data consists of respondents’ self-assessed probabilities of leaving a bequest at the end of life. In the analyses that follow, inter vivos transfers and bequests are modelled separately. The regressor of interest is the practical help that children provide to their parents with household chores and paperwork.

Transfers from parents to their adult children are common even in developed countries with public income redistribution and public care provision. Kotlikoff and Summers (1981), Modigliani (1988), Gale and Scholz (1994), Piketty (2011) and Ohlsson et al. (2014) examine the magnitude of intergenerational transfers in the US and in Europe. They generally find that a large proportion of people’s wealth is passed on from one generation to the next, rather than accumulated over the individual’s life-cycle.

Existing empirical evidence suggests that sizeable proportions of older households in the US and Europe receive informal care from their children—see Cox and Rank (1992), Norton and Van Houtven (2006) and Alessie et al. (2014). Examining the latest wave of the sample of TILDA data used in this analysis, 47% of parent households have made inter vivos transfers to their non-resident adult children over the past 2 years with a total unconditional mean value of just over EUR 4,000. Of the parent households, 90% expect to leave (a non-zero) inheritance and two-thirds expect to leave an inheritance worth EUR 150,000 or more. Nearly 40% of parent households received help from their children with household chores or paperwork over the past 2 years, with an unconditional mean of 6 h per month.

In multivariate cross-sectional analysis, a positive and significant relationship between help and the probability of inter vivos transfers is found. The strength of the relationship is inversely related to the size of the transfer: it is only statistically significant in the case of small (between EUR 250 and EUR 5,000) transfers whereas it is not significant for larger transfers. Evidence of a causal relationship is necessary for identifying the exchange motive and differentiating it from two-way altruism (Cox and Rank 1992). The cross-sectional results are robust to the inclusion of a measure of the emotional closeness between the children and parents, proxy measures of the parents’ personality, and a lagged value of child-provided help. Because of the availability of only two waves of data, the scope of longitudinal analysis is restricted by the limited variability in the relevant measures over time. Though limited, the longitudinal data analysis enables the incorporation of fixed effects that account for unobserved time-invariant heterogeneity: factors such as preferences and the type of relationship between the parents and their children are likely to be correlated with within-family flows of both monetary transfers and services. In fixed effects estimates, even with limited variation available for identification of effects, the correlation between help provision and the probability of a transfer remains positive and statistically significant at the 10% level.

Because bequest data were not collected in the first wave of TILDA, the analysis on the self-assessed probability of making future transfers via bequests is restricted to Wave 2 data. A cross-sectional examination of the probability of future bequests reveals no relationship with help provided by children. As the relationship between help and bequests is insignificant in cross-sectional analysis, an endogeneity-corrected estimate of the relationship is likely to be even less significant. The analysis of the bequest data has its limitations because parents who report a high probability of leaving a bequest may simply be risk averse, uncertain about their individual mortality risk, or a homeowner (with illiquid wealth)—therefore expecting to reach the end of life with positive wealth, rather than explicitly wishing to transfer that wealth to their children.

These findings support the hypotheses of McGarry (1999) and Bernheim and Severinov (2003), who suggest that inter vivos transfers are better suited to exchange, and therefore are likely not to be influenced by the same motives as bequests. The finding that the exchange motive only drives small inter vivos transfers is also consistent with this prediction: if transfers are made in small quantities, they can be made more frequently which in turn makes the enforcement of the contract between the parent and the child easier. Another possible explanation for the finding that child-provided help is found to only be associated with small transfers (but not with large transfers or the likelihood of future bequests) is the time disconnect between reports of the help and the transfers. In the first wave of data analysed in this paper, help and small inter vivos transfers are recorded over the preceding two years, whereas large inter vivos transfers are recorded over the preceding decade. The probability of leaving a bequest, by definition, relates to transfers made at the end of life. If transfers in the form of bequests are driven by exchange motives, the identification of such activity would possibly require data about the interactions between parents and children spanning decades.

Intergenerational transfers play a role in saving behaviour and decisions about investment in human and physical capital. They affect the distribution of wealth within families and have an impact on the equality of opportunities between individuals. From a public policy point of view, understanding the motives behind transfer behaviour is relevant for predicting the likely consequences of changes to public provision of care, taxation of estates and gifts and public income redistribution. Depending on what motivates intergenerational transfers, changes to public income distribution may either crowd out or reinforce private flows of intergenerational monetary transfers (Cox 1987; Cox and Rank 1992).

The remainder of this article is structured as follows: Section 2 summarises the existing theoretical and empirical literatures. Section 3 describes the data and provides summary statistics. Section 4 presents the empirical findings, while Section 5 offers concluding remarks.

2 Existing literature

2.1 Transfer motives

The main theories suggested as motivations for transfers from parents to children are altruism (Barro 1974; Becker 1974, 1981), exchange (Bernheim et al. 1985; Cox 1987), and egoistic giving or “warm glow” (Arrow 1975; Andreoni 1989, 1990). An altruistic parent’s utility is a function of the child’s utility. The parent makes transfers to the child as long as: (i) the parent’s income is high enough relative to the child’s, and (ii) the parent gives sufficient weight to the child’s utility in their own utility function. Transfers are compensatory, meaning that parents make transfers to children with relatively low incomes. The main prediction arising from the altruism model is the idea of income pooling within families. If a non-altruistic child’s income is reduced and an altruistic parent’s income is increased by the same amount via public income redistribution, a private transfer in the opposite direction cancels out the public transfer. Redistributive neutrality follows from altruistic behaviour: redistributing income between generations has no impact on consumption as long as transfers between parents and children are possible.

The exchange motive is in operation if parents make inter vivos transfers (or promises of future bequests) to children in exchange for services. The service can take on many forms, ranging from formal (such as caregiving) to casual (e.g., companionship). The parent derives utility from receiving the service, whereas providing the service is costly for the child. A transfer takes place if both benefit from the transaction. If transfers are motivated by exchange, the redistributive neutrality result does not hold: a private transfer does not necessarily cancel out a public one, and may even amplify it. The details of this prediction are discussed below.

If transfers are driven by egoistic (or “warm glow”) motives, transfers are made for the pure joy of giving, and the person making the transfer derives utility from the very fact that a transfer takes place. In this case, transfers are not made in order to increase the income of the recipient, or to receive something in exchange. Income pooling within families and redistributive neutrality do not necessarily hold in this case (Andreoni 1989).

Existing evidence suggest that inter vivos transfers and bequest may be driven by different motives. The fact that bequests are usually divided equally among children—while inter vivos transfers are not—has been noted in many studies (Menchik 1988; McGarry 1999; Bernheim and Severinov 2003; Norton and Taylor 2005). These observed differences in behaviour have led to the development of models that allow for the two types of transfers to be influenced by different motives. Norton and Van Houtven (2006) hypothesise that inter vivos transfers are preferred over bequests in households with an exchange motive. A parent can adjust the flow of inter vivos transfers, making the enforceability of the contract between parent and child easier than in the case of (once-off) bequests. Transferring wealth via bequests is also characterised by uncertainties arising from unexpected (medical or other) expenses and the uncertainty about length of life. Additionally, inter vivos transfers are potentially more convenient for exchange purposes because they can be hidden from the siblings of the recipient more easily than bequests—perhaps to prevent conflict between the children or to avoid the children perceiving unequal affection from the parents.

2.2 Identifying transfer motives

The way of identifying between altruistic and exchange motives in intergenerational transfer behaviour that has been most widely used in previous research has been to estimate the effect of child income on transfers.¹ Laferrère and Wolff (2006) review earlier empirical work in this area, while more recent studies include Hochguertel and Ohlsson (2009) and Nordblom and Ohlsson (2011). Differentiation between altruism and exchange can potentially be made when examining the effect of the child’s income (Y_k) on the value of transfers, T_value (the price of the service multiplied by quantity of the service traded). In the case of altruism, an increase in Y_k has a negative effect on T_value. In the case of the exchange motive influencing transfers, the direction of the effect of Y_k on T_value depends on the own-price elasticity of demand for child services: the effect is positive if the demand for child services is own-price inelastic. Conversely, the effect is negative if the demand for child services is own-price elastic—see Cox (1987) for a discussion. Therefore, estimating the effect of Y_k on T_value is informative of transfer motives only in the case of finding a positive coefficient, in which case it serves as evidence of the presence of the exchange motive.² Whether demand for child services is own-price elastic or inelastic depends on the availability of substitutes for the services, which in turn is likely to be related to the degree of formality of the service.³

A more direct way of testing for the presence of the exchange motive is to estimate the effect of child-provided services (S) on transfers (Cox and Rank 1992). In the case of exchange, the effect of S on the probability of a transfer taking place (T_prob) is positive.⁴ This identification strategy is pursued in this paper because the former approach requires assumptions—that are difficult to validate—about the own-price elasticity of child services. The identification strategy chosen here does not require similar assumptions to be imposed. In addition, the data set used in this analysis (described in detail below) lends itself to this analysis due to the rich longitudinal data available on child-provided services and transfers, whereas available information about child income is limited to proxy measures such as labour market participation, education and home ownership.

Very few existing studies have examined the relationship between services and transfers. The existing literature can be divided into two main types: studies that examine the determinants of service supply from the child’s point of view (Bernheim et al. 1985; Perozek 1998; Alessie et al. 2014) and studies that examine the effect of child-provided services on transfers (Cox and Rank 1992; McGarry and Schoeni 1997; Norton and Van Houtven 2006; Horioka et al. 2018). The four papers in the latter group are the most relevant for this work, and they are discussed in more detail below.

Cox and Rank (1992) carry out an analysis of the effect of child-provided care and contact between parents and children on inter vivos transfers using cross-sectional data from the National Survey of Families and Households (NSFH). They estimate models of both the probability of a transfer and the value of those transfers. They find patterns that are more consistent with exchange than with altruism: the child’s income has a negative effect on the probability of receiving a transfer but a positive effect on the value of those transfers. They estimate the effect of child-provided help on the probability of transfer to be positive and statistically significant. They find the effect of help to be insignificant in determining the transfer amount, suggesting that the demand for child help is own-price inelastic. When it comes to the effect of contact on transfer probability, they find a positive and statistically significant coefficient while the effect on the value of transfer is insignificant. The shortcomings of the analysis arise from data limitations: cross-sectional analysis potentially suffers from bias caused by the omission of unobservables that are potentially correlated with both caregiving and transfers. Also, the transfers in NSFH are measured over 5 years, whereas caregiving is only recorded over one year.

Although McGarry and Schoeni (1997) primarily focus on the effect of child income on transfers, they also include a dummy variable of a child providing help with (I)ADLs⁵ to parents in a cross-sectional fixed effects transfer amount regression. The fixed effects are at the family level. Therefore, identification relies on between-family variance. They find that the estimated coefficient is negative and statistically insignificant for both the dummy variable and a continuous measure of the hours of help provided. They do not, however, estimate models of the probability of a transfer taking place.

Norton and Van Houtven (2006) is the only existing study that exploits panel data—from the 1993 and 1995 waves of the Asset and Health Dynamics Study (AHEAD)—to investigate the effect of informal care provision on the likelihood of a child receiving inter vivos transfers from parents. They find significant effects of care provision on the likelihood of receiving a transfer (compared to a sibling who does not provide care) by estimating logit models with and without household fixed effects (using within-household variation). Their findings are robust to specifications accounting for the possible endogeneity of informal care using lagged values of informal care and instrumental variables. Norton and Van Houtven (2006) also examine the likelihood of a parent household planning to divide their bequests equally among their children. They find that the effect of informal care is not statistically significant.

Horioka et al. (2018) focus on expected bequests, examining data from a Japanese household survey. They find that adult children are more likely to live with or near their parents and/or to provide services to them if they expect to receive a transfer (either an inter vivos transfer or a bequest) from the parents, providing evidence of the existence of the exchange motive. The expectation about future transfers is measured on the child (recipient) side, and no distinction can be made between the two types of transfers.

3 Data

The data used in this analysis come from the first two waves of The Irish Longitudinal Study on Ageing (TILDA) collected in 2009–2011 (Wave 1) and in 2012–2013 (Wave 2). TILDA provides information on the health, lifestyles and socio-economic characteristics of a nationally representative sample of Irish people aged 50 and over and their spouses. The first wave of TILDA data contains information on 8,504 individuals living in 6,279 households. Each participant underwent a face-to-face computer-assisted personal interview (CAPI) in their home, was given a self-completion questionnaire and was invited to a health assessment. The overall response rate of the first wave was 62%. The second wave had an overall response rate of 86%. See Barrett et al. (2011) for a detailed description of the data, including sampling and the construction of survey weights (which are accounted for in this analysis). TILDA contains information about the parents, children, inter vivos transfer amounts, self-assessed probability of leaving a bequest, and information about help and care provided within families.

The selection of the sub-sample of TILDA data used in this analysis is described in Table 1. The sample is restricted to families with children, all of whom are aged 18 or older and no longer live with the parents. This sample selection criterion is common in the literature—see Bernheim et al. (1985) Norton and Van Houtven (2006), McGarry (1999) and Alessie et al. (2014). Exclusion of families where children are financially dependent on the parents is necessary because co-residence is a type of transfer, the value of which is difficult to estimate. Further, families whose children are in education are excluded because parents are likely to under-report transfers in the form of payments associated with education (such as rent or tuition fees). The sample is restricted to households with both a financial and a family respondent present (so that information about transfers is recorded), to households where both spouses take part in the study (in the case of married or cohabiting parents) and households with non-missing data for the analysis variables. ⁶The resulting sample size is 1,035 families (499 two-parent and 536 one-parent families) with a total number of 3,602 children (an average of 3.48 per family). 72% of the one-parent households are widows.

Table 1 Selection of the analysis sample

As data about the self-assessed probability of leaving a bequest were only collected in Wave 2, the panel analysis is limited to examining inter vivos transfers. When it comes to the larger (EUR 5,000 and over) inter vivos transfers, Wave 1 records transfers over the preceding 10 years, whereas Wave 2 data covers the 2 years between the waves. Due to this time inconsistency, the panel analysis is restricted to the smaller (EUR 250 to EUR 5,000) inter vivos transfers for which the questions in both waves were identical.

3.1 Inter vivos transfers

In TILDA, the question regarding the large (EUR 5,000 and over) transfers is worded as follows:

Not counting any shared housing or shared food, in the last two years, have you given financial help or gifts, including help with education, of EUR 5,000 or more to any child (or grandchild)? ⁷

The question was explained further: “By financial help we mean giving money, helping pay bills, or covering specific types of costs such as those for medical care or insurance, schooling, down payment for a home, rent, etc. The financial help can be considered support, a gift or a loan.”

If the parents had made transfers, a follow-up question about the total value of the transfers was asked.

Information about smaller (EUR 250 to EUR 5,000) transfers was recorded by asking:

I would now like to ask about financial assistance to your children apart from any large lump sums that you mentioned in the previous question. During the last 2 years, did you (or your spouse/partner) give financial or in-kind support totalling EUR 250 or more to any of your children and/or grandchildren (or their spouse/partner)?

Again, a question about the total value of the transfers was asked if the parents had made transfers. Descriptive statistics of inter vivos transfers are presented in Table 2. In both waves, nearly 40% of parent households report having made small transfers to their children over the preceding two years. The mean total value of the transfers declines between waves. The larger transfers are less common, with 29% of families reported having made them in the 10 years preceding Wave 1, and 14% in the two years preceding Wave 2.

Table 2 Inter vivos transfer amounts from parents to children during past 2 years

3.2 Subjective probability of leaving a bequest

In Wave 2 of TILDA, respondents were asked about their expected probability of leaving a bequest:

What are the chances that you will leave any inheritance?

It was explained to the respondents that the question covers properties and other valuables. If the given answer was greater than zero, follow-up questions were asked about the probabilities of leaving an inheritance totalling EUR 50,000 or more (and EUR 150,000 or more).

The bequest data are summarised in Table 3. The vast majority of parent households report a non-zero probability of leaving a bequest. The average reported percentage probability of leaving a bequest is 86. As expected, the (conditional and unconditional) mean probability of leaving a bequest of decreases with larger bequest values. The high probabilities of leaving large bequests are reasonable considering that 88% of the parent households were homeowners in Wave 1 (see Table 5) and as few people draw on housing equity in retirement, many bequeath residential housing. As a reflection of this, the home ownership rate increases with age in cross-sectional analysis of either wave of data.

Table 3 Expected bequest probabilities (in %)

3.3 Help provided by children

In both waves of the survey, TILDA respondents were asked:

In the last 2 years, have your children or grandchildren spent at least 1 hour a week helping you with things like:

• 1. Practical household help, e.g., with home repairs, gardening, transportation, shopping, household chores?

• 2. Help with paperwork, such as filling out forms, settling financial or legal matters?

The respondents were also advised: This refers only to help received from children outside the household i.e., help received from co-resident children is to be excluded.

If the household had received help, a follow-up question was asked about the total number of monthly hours of help.

Table 4 presents summary statistics of these data. On average, a lower share of households received help in Wave 2 than in Wave 1; however, the unconditional mean of total monthly hours increased slightly between waves, reflecting the increase in conditional hours.

Table 4 Help provided by children to parents during past 2 years

The TILDA questionnaire also contains information about help that the respondents receive with (I)ADLs. Although approximately 10% of the households have a member who receives ADLs and/or IADL related help, the caregiver is most often the spouse. Less than 4% of the households in the sample receive (I)ADL help from their children.

3.4 Control variables

Table 5 summarises the explanatory variables across the two waves of data. At the mean, 49% of a family’s children are female, the average child is 38 years old in Wave 1 and has 1.4 children. The number of children that the respondents’ children have was not recorded in Wave 2, and therefore Wave 1 figures are also presented in the Wave 2 column. Home ownership of children was recorded in Wave 1, with 75% being homeowners. Some children acquire education between the waves. As families with children in full-time education are excluded from the analysis, any children who have obtained education have done so part-time or between the survey waves.

Table 5 Means and standard deviations of control variables

The geographical location of children is relatively constant across time, with slight decreases in shares of children living in the same neighbourhood as their parents and increases in children living abroad. As expected, the marriage, divorce, separation and widowhood rates among the children increase, whereas fewer children are single or cohabiting in Wave 2. The recession experienced in Ireland around this time is evident when examining the children’s labour market status: fewer are employed full-time while more are self-employed or out of the workforce. The share of unemployed decreases slightly across waves, possibly reflecting exits from the workforce and emigration.

Although current level of child income is not recorded, changes to labour market status indicate shocks to current income. Permanent income—proxied for by education and home ownership—may be preferred to current income because the latter is more likely to suffer from reverse causality if children adjust their labour supply because of transfers. Arrondel and Masson (2001) discuss the use of proxy measures for child income. Estimating transfer models using a sub-sample with child income data and proxy data for permanent income, they find no significant difference in coefficient estimates.

Examining the characteristics of the parents, the average age of the spouses (if married or cohabiting) is 68 years in Wave 1. The mean annual income of the household is EUR 30,020 in Wave 1 and EUR 25,260 in Wave 2, reflecting the reductions in incomes as people retire. Perhaps surprisingly, individuals report their overall health levels to be better in Wave 2 than in Wave 1. As expected, fewer household heads are employed in Wave 2 and compared to Wave 1, and a larger share report being retired in Wave 2. Nearly half of the parent households are married or cohabiting, whereas just over a third are widows.

4 Multivariate analysis

The unit of analysis is the family because the data on child-provided help are aggregated at the family level. Both average parental and average child-level variables are included in the models.

The analysis follows a hurdle specification where the probability of transfer (T_prob) and the value of the transfers (T_value) are modelled separately, similarly to Cox and Rank (1992) and Alessie et al. (2014). The hurdle model approach is chosen because the prediction of the exchange model that the effect of S may be positive on T_prob but negative on T_value. T_prob is modelled using a probit specification, whereas T_value is modelled using linear regression, conditional on a transfer taking place (T_value > 0).

A Tobit model could be used to account for zeros in a distribution, but it would be unsuitable because it estimates a single set of coefficients for both the T_prob and the T_value models, therefore assuming that the effect of S is of the same sign in both (Greene 2003).

After the models are estimated using data from the two waves separately, the analysis is extended to account for possible biases in cross-sectional analysis. The additional specifications consist of cross-sectional models with added covariates and fixed effect conditional logit models.

4.1 Cross-sectional probit models

The probit models fit the probability that parents make (or plan to make) monetary transfers to their children:

$$T_{prob\,i} = f\left( {\alpha _0 + \alpha _1S_i + \alpha _2{\mathbf{Y}}_{\mathbf{k}i} + \alpha _3{\mathbf{K}}_i + \alpha _4Y_{pi} + \alpha _5{\mathbf{P}}_i + a_i} \right)$$

(1)

where: f (.) is the standard normal cumulative distribution function. T_{prob i} is a probability of either:

• parents of family i making inter vivos transfers (to any child)

• parents of family i reporting a positive probability of leaving a bequest.

α₀= constant term; S_i= a variable that equals 1 if family i’s children provide help to parents, and 0 otherwise; Y_ki= a vector of proxy measures for family i’s children’s income; K_i = a vector of family i’s children’s characteristics; Y_pi = total parental income of family i;P_i = a vector of family i’s parental characteristics; a_i = residual term of family i.

The main coefficient of interest is α₁ which is expected to be positive in the case of exchange. The expected sign of α₂ depends on the particular element of the vector: education and house ownership (proxies for permanent income) are expected to have negative effects on T_prob, whereas a child currently not working is expected to have a positive effect. α₄ is expected to be positive.

The estimated marginal effects for inter vivos transfer models are presented in Table 6; small transfers in Models 1 and 2, and large transfers in Models 3 and 4. The estimates suggest that the probability of making small inter vivos transfers has a statistically significant positive association with the children of the family providing help to the parents. The estimated value of α₁ for small transfers using Wave 1 data is 0.11: a family where at least one child provides help to the parents is 11 percentage points more likely to make small inter vivos transfers to their children, compared to a family where no children provide help. The estimate for α₁ using Wave 2 data is 0.09, also statistically significant at the 1 per cent level. These findings are consistent with exchange but also with two-way altruism.

Table 6 Probit models of T_prob for inter vivos transfers, marginal effects at mean

Instead of using a dummy indicator of help as the main explanatory variable, alternative specifications presented in Table 7 use a continuous variable measuring the total daily hours of help that the children provide. The relationship is less clear than when a binary variable of help is used, possibly due to measurement error in the continuous variable. It is likely that in surveys, retrospectively asked questions regarding detailed information (such as continuous hours measure of help received) suffer from more measurement error than simple binary indicator variables (Bound et al. 2001).

Table 7 Probit models of T_prob using help hours, marginal effects at mean

The relationship between the probability of a family making large transfers and child-provided help is not statistically significant (see Models 3 and 4). One needs to bear in mind the time disconnect in Wave 1 data between large transfers (10 years prior) and child-provided help (2 years prior), when examining the findings of Model 3.

This finding is likely to be due to the difference in the suitability of different types of transfers for exchange: the smaller the transfer is in value, the more frequent the transfers can be, which in turn increases the enforceability of the contract between the parent and the child. As an indication of the frequency of small vs. large inter vivos transfers, Table 2 shows that in Wave 2, 39% of parents report having made small transfers in the preceding 2 years, whereas only 14% had made large transfers in the preceding 2 years.

The estimates of α₂ (effect of child income measures) are statistically insignificant in most cases; however, the models in which the coefficient estimate is statistically significant, the signs mostly confirm prior expectations. There is a significant negative association between the share of children with university degrees and transfer probability (Model 2). An interesting finding in Model 3 is the significant positive effect of the share of children who are homeowners in Wave 1. This effect is likely to be linked to the parents assisting the children with property purchases. Considering that large inter vivos transfers were recorder over the preceding 10 years in Wave 1 when the average age of children was 38 years, many of the children’s property purchases are likely to have taken place within this time period. Unexpectedly, the share of children not working has a negative association with probability of small transfers in Wave 1. The effect of parental income on transfer probability is estimated to be positive, as expected. Also, the proxy measures of parental permanent income (home ownership, education) have positive effects on transfer probability, possibly reflecting lower perceived risk of future income fluctuation.

Some of the control variable marginal effect estimates warrant discussion. Children’s age is negatively associated with small transfers, but positively associated with large transfers. This finding may indicate credit constraints experienced by younger children, leading parents to make small transfers to assist with day-to-day finances. Larger transfers (partly indicating property purchases and transfers of family businesses) are more likely to be made to older children. The share of children living in the same county as the parents is positively associated with the likelihood of parents making small transfers (but insignificant when it comes to large transfers). Parents who perceive their health to be very good or excellent are more likely to make transfers to their children, perhaps due to lower expected medical expenses in the future.

The estimates of probit models of the subjective probability of leaving a bequest are presented in Table 8. The estimated marginal effects associated with the help indicator are statistically insignificant in all of the bequest model specifications. One potential explanation for this finding may be that the time disconnect between the child-provided help and the (future) bequest can potentially be measured in decades. Future research with access to new data—later-life care provision and actual bequests, for example—may reveal more on the relationship between help (or other services provided by children) and bequests. The lack of variation in the outcome variable used in Model 1 may lead to imprecise estimates, as only 10% of the sample report a zero probability for leaving a bequest. This issue, however, is alleviated in Models 2 and 3, where the share of the analysis sample reporting zero probabilities for leaving a bequest are 83% and 67%, respectively (see also Table 3). The control variable marginal effects are largely of the same sign as in the inter vivos transfer models; however, the statistical significance is generally lower in the case of the children’s characteristics but stronger in the case of the parental characteristics. The low significance of children’s characteristics may be linked to the uncertainty associated with future events for the parents, such as unexpected medical expenses and the uncertainty about longevity. The marginal effect estimates of parental house ownership have very small standard errors. This is potentially linked to the finding that most parents report a very high probability of leaving a bequest, which may be an indication of the high rate of home ownership in Ireland.

Table 8 Probit models of T_prob for planned bequests, marginal effects at mean

Alternative specifications presented in Table 9 use a continuous variable measuring the total daily hours of help that the children provide. The estimated effect of help on the subjective probability of leaving a bequest is statistically insignificant.

Table 9 Probit models of T_prob for planned bequests (marginal effects at mean), using help hours

As discussed, transfers via inter vivos and bequest may be driven by separate motives. For the first time in the literature, this article examines inter vivos transfer size groups separately. The evidence of the exchange motive only being present in small inter vivos transfers is consistent with the theoretical predictions: large inter vivos transfers resemble bequests more than small transfers do. The larger the individual transfer, the more difficult it is to conceal from the recipient’s siblings and the fewer possibilities the parent has to adjust the transfer to reflect the quality or quantity of the child’s service.

It is possible that the bequest expectation questions are not interpreted by the respondents as planned (or desired) transfers to children at the end of life, but may also capture perceived uncertainties about future events as well as the uncertainty about length of life itself. Whereas bequests can be accidental due to these uncertainties, inter vivos transfers are always intentional. Therefore, inter vivos transfer data are be better suited to the examination of transfer motives.

4.2 Cross-sectional models of transfer value

This part of the analysis focuses on the determinants of the total value of the inter vivos transfer, estimated using an OLS specification:

$$T_{value\,i} = \beta _0 + \beta _1S_i + \beta _2{\mathbf{Y}}_{{\mathbf{k}}i} + \beta _3{\mathbf{K}}_i + \beta _4Y_{pi} + \beta _5{\mathbf{P}}_i + b_i\quad \forall T_{value\,i} > 0$$

(2)

where the explanatory variables are the same as in the probit model presented in Equation 1.

The results allow for conclusions to be drawn about the elasticity of demand for child help with respect to the implicit price of the service, Y_k: a negative β₁ indicates inelastic demand. Additionally, a positive β₂ can be interpreted as evidence of the exchange motive (whereas a negative estimate is consistent with both exchange and altruism).

The estimates are presented in Table 10. The estimate for β₁ is statistically insignificant in all of the specifications. ⁸ This result is consistent with the findings of Cox and Rank (1992) who, despite finding statistically insignificant estimates, interpret the finding to be consistent with the exchange model with own-price inelastic demand for child services.

Table 10 OLS models of T_value for inter vivos transfers, conditional on T_value > 0

Table 11 OLS models of T_value for inter vivos transfers conditional on T_value > 0, using help hours

An analysis of the continuous measures of the percentage probability of leaving a bequest was also carried out. The results were largely similar to those of the probit bequest models, with statistically insignificant coefficient estimates associated with the child help variable.

4.3 Issues of causality

The correlation between help and the probability of a transfer taking place is a necessary but not a sufficient condition for the exchange motive to exist: it is also consistent with two-way altruism (Cox and Rank 1992). For proof of the existence of the exchange motive, establishing a causal relationship is necessary. In order to estimate the causal effect of S on transfers, all endogeneity concerns would need to be addressed. As suggested by Norton and Van Houtven (2006) a lagged value of S goes some way towards alleviating endogeneity concerns. Models 5 and 6 of Table 12 use the lagged value of S, with the estimates largely unchanged. Models 1–4 of Table 12 also contains estimates of models which include usually unmeasured covariates that are potentially correlated with both help and transfers: the emotional closeness between the children and parents, and measures of parental personality. ⁹ The following section makes use of the panel structure of TILDA to further address endogeneity issues.

Table 12 Models of small inter vivos transfers (T_prob and T_vazlue) in Wave 2 (with additional covariates)

4.3.1 Fixed effects analysis

A fixed effects (FE) logit model is estimated by conditional maximum likelihood (Chamberlain 1980). The models can only be estimated for families for whom the transfer status changes between periods (n = 330). In the two-wave panel case, the FE logit is equivalent to a binary logit model with the change in the explanatory variables explaining the change in the outcome variable (Verbeek 2008). In this modelling framework, by construction, the effects of time-invariant variables cannot be estimated.

The cross-sectional logit model of the probability that a family makes small inter vivos transfers is defined as:

$$T_{prob\,it} = f\left( {\gamma _0 + \gamma _1S_{it} + \gamma _2{\mathbf{Y}}_{{\mathbf{k}}{it}} + \gamma _3{\mathbf{K}}_{it} + \gamma _4Y_{p\,it} + \gamma _5{\mathbf{P}}_{it} + c_{it} + d_i} \right)$$

(5)

where f (.) is the logistic cumulative distribution function. The error term is divided into two components: a time-varying error (c_it) and a time-invariant error (d_i). The FE estimation controls for d_i, consisting of time invariant unobservable factors such as personality traits and risk aversion.

Table 13 presents the estimates of the FE logit models. The statistical insignificance of Model 1 estimate may be expected in a FE model with only two time periods and little variation in the dummy dependent variable and the main explanatory dummy variable: for approximately 70% of the families, the help indicator does not vary across time.

Table 13 Fixed effects Logit (T_prob)

In order to use a fixed-effects model specification in which there is more variability over time in the measures of interest, Model 2 uses a continuous measure of the number of hours of child-provided help. The associated coefficient estimate is statistically significant at the 10% level of confidence. Given that there are only two waves of data available for this analysis, and the potential for measurement error in the continuous measure of help hours, the statistically significant effect of S on T_prob provides further evidence of the existence of the exchange motive when it comes to small inter vivos transfers.

Future analyses could draw from longer panels of data in the examination of within-family variation over time in child-provided help and inter vivos transfer behaviour. More longitudinal observations could provide the necessary variation in the prevalence of help provision and transfer behaviour for robust analysis of the causal effect of help on transfers.

5 Discussion and conclusion

Research into intergenerational transfer motives is relevant for a range of public policies, many of which are of particular importance in countries with ageing populations: provision of care to older people, public income redistribution between generations, and the taxation of inter vivos transfers and bequests.

If the exchange motive is driving monetary transfers within families, the implications for policy partly depend on the own-price elasticity of demand for these traded services—and therefore the availability of substitutes for them. An increase in public service provision decreases parental demand for child-provided services, therefore reducing the likelihood of a transfer from parents to children. However, the value of transfers may increase or decrease, depending on the price elasticity of the demand. Therefore, differentiating between formal and informal care is important.

The outcomes of changes to the taxation of inter vivos transfers and bequests depend on transfer motives. The relative tax treatment of inter vivos transfers and bequests may only affect the timing of wealth transfer between generations if intergenerational transfers are motivated by altruism. However, if transfers are driven by exchange, taxation of intergenerational transfers may reduce parental demand for child-provided services, therefore increasing pressures for public provision of care.

Depending on what motivates intergenerational transfers, changes to public income distribution between cohorts may either be counteracted or reinforced by private flows of monetary transfers (Cox 1987; Cox and Rank 1992). If the exchange motive drives intergenerational transfers, increased public care provision to older people may decrease caring provided by children and consequently decrease monetary transfers to children (Kohli and Künemund 2003). When it comes to bequests, if they are planned and used to compensate children for caring for their elderly parents, inheritance taxes may dis-incentivise within-family provision of these services (Jürges 2001). In the case of the exchange motive driving intergenerational transfers and child-provided services having few market substitutes, a public redistribution of income from the parent to the child generation is expected to increase the value of transfers from parents to children.

This paper examines the presence of the exchange motive in intergenerational monetary transfers, estimating the correlation between child-provided services and the probability of transfers being made within a family. The findings suggest a relationship between care-giving provided by adult children and the small inter vivos transfers that parents make to their children. The relationship between child-provided help and the probability of a family making large inter vivos transfers is not statistically significant. Also, there is no evidence of a correlation between help and the subjective probability of leaving a bequest in the future. The evidence of the exchange motive only being present in small inter vivos transfers is consistent with the predictions of McGarry (1999) and Bernheim and Severinov (2003): different types of financial transfers within families are likely to be driven by different motives. The findings of this research suggest that large inter vivos transfers resemble bequests, and provides some evidence of the exchange motive as one driver of small inter vivos transfers from parents to adult children.

A more definitive assessment of the role of the exchange motive in intergenerational transfers, and the distinction between exchange and two-way altruism, would require a richer data set than the one used in the present analysis. Future work could draw from longer panels of data with time variation in child-provided services and parental financial transfers. Alternatively, a data set containing an instrumental variable which only affects parental financial transfers through its effect on child-provided services could shed more light on the mechanisms in operation.

6 Funding

This study was funded by a Government of Ireland Postgraduate Scholarship.