1Marital sorting by socioeconomic background is a well-known phenomenon, but its underlying mechanisms are complex. Michel Bozon and Wilfried Rault (2012) compared the places where people meet their first sexual partner and their first life partner, and showed that for some – notably women and highly educated persons of both sexes – the choice of these places involves certain elements of strategy. In this article, Nicolas Fremeaux examines the mechanisms of marital sorting from a new and original angle, looking at the dimensions of inheritance and labour income. Aggregating data from several French wealth surveys, he measures the partners’ wealth in terms of inheritance on the one hand, and labour income on the other, estimating the value of these assets accumulated over a lifetime. He shows that the degree of marital sorting is influenced by both dimensions, but that the two forms of wealth are not substitutable.
2In many nineteenth century novels, the role of inheritance [1] in marital choices appears to be decisive. For example, Eugène de Rastignac in Father Goriot by Balzac or Bel Ami in the eponymous novel by Maupassant are characters with no parental wealth who try to move up the social ladder by marrying rich heiresses. Today, people probably marry more frequently for love, and family has less power than during the nineteenth century. But has the importance of inheritance in marital choices disappeared? With the return of wealth and inheritance in rich countries (Piketty, 2011; Piketty and Zucman, 2014) the question of marriage is key to understanding the dynamics of inequality. Marital sorting allows us to explore this macroeconomic evolution from a microeconomic perspective. Until now, the existing literature has mostly looked at assortative mating from the viewpoint of labour income, social status or education. This article investigates, for the first time, the importance of inherited wealth in the choice of spouse. Moreover, the return of inherited wealth raises the issue of the source of wealth in marital choices. We begin by asking whether human capital can compensate for a lack of parental wealth. Our data enable us to assess the likelihood, for a Rastignac of today, of marrying an heiress. The aim of our second research question is to propose a bidimensional analysis of the marriage market and to explore the relationship between acquired and inherited traits.
3In this article, we use data from the French wealth surveys conducted by the French National Institute for Statistics and Economic Studies (INSEE) in 1992, 1998, 2004 and 2010, [2] avoiding the two main drawbacks of existing studies. First, in the French wealth survey, available information on inheritance concerns wealth inherited in the past (i.e. already received by individuals). Yet expectations of inheritance (from both the husband’s and the wife’s sides) also matter in marriage decisions. We therefore use information about parental wealth (type of goods owned by the parents) to estimate respondents’ expected inheritance and hence to consider the total inherited wealth someone will receive rather than simply the observed estate declared at the time of the survey. This unique information allows for more detailed analysis of the role of inherited wealth in marital choices. Second, the use of current labour income (reported at the time of the survey) generates endogeneity issues since the division of labour within the couple depends on decisions taken after union formation. Moreover, income at a given moment in time may not reflect an individual’s potential income because of lifecycle biases and transitory shocks. We provide estimates of permanent income to resolve this issue.
4This article measures the extent of marital sorting by total inherited wealth and its new findings are important for several reasons. First of all, our conclusions are key to understanding the dynamics of inequality. Indeed, this research is grounded upon recent findings on long-term trends in inheritance. Piketty (2011) indicates that the annual flow of inheritance, which represented around 20-25% of national income between 1820 and 1910 in France but less than 5% in 1950, had returned to 15% by 2010. Whether this situation is important for the dynamics of inequality over time depends, among other things, on marital decisions – do heirs marry heiresses? – as family plays a decisive role in the transmission of capital, be it human, social or material. Atkinson (1975) provides an illustration of this. Suppose that all households have two children (one boy and one girl) and that all wealth is held by just 5% of households. In the extreme case where the rich marry the rich, the degree of wealth concentration will be extreme. In other words, class marriage, where wife and husband come from families with the same level of wealth, is equivalent to a situation where all property is inherited by sons, and these sons marry their sister.
5This article also ties in with the literature on marital decisions. Becker’s (1973) seminal work has inspired a vast literature about the economics of marriage. Among other topics, Gary Becker initiated a debate about the substitutability/complementarity of spouses’ characteristics. He argues that the household division of labour resulting from the comparative advantages of market and non-market productivities leads to negative assortative mating with respect to labour income. Lam (1988), on the other hand, invokes the desire to maximize a household’s common goods as a source of complementarity between spouses. Several empirical studies have sought to contribute to this debate. Most concentrate on income (Smith, 1979; Becker, 1981; Nakosteen et al., 2004; Schwartz, 2010; Lise and Seitz, 2011) or education (Schwartz and Mare, 2005; Chiappori et al., 2012a; on French data: Forsé and Chauvel, 1995; Goux and Maurin, 2003; Vanderschelden, 2006), and find positive assortative mating based on income and educational levels. Studies based on French data suggest that the extent of marital sorting based on education in France is close to the average for developed countries, but no data are available for income, so comparison is not possible.
6While spouses’ parental traits have important effects on the intergenerational transmission of socioeconomic status, studies in this area are much scarcer. Some research focuses on social origins (Burgess and Wallin, 1943; Kalmijn, 1991; Uunk et al., 1996; on French data: Girard, 1964; Bozon, 1991; Desrosières, 1978; Thélot, 1982), but while easier to measure, occupational status imperfectly measures parents’ financial resources. Charles et al. (2013) opt for a measure of total parental wealth. They do not observe inherited wealth per se, but only the wealth owned by the parents. Yet inheritance taxation and family structure (number of heirs) also need to be considered when determining inheritance on the basis of total parental wealth. The only articles that directly study the role of inheritance in marital choices focus either on the role of dowries in the bargaining process between spouses (Fafchamps and Quisumbing, 2005; Zhang and Chan, 1999) or on the rationales of dowries (Anderson, 2003; Botticini and Siow, 2003).
7The main contribution of our article is to use direct estimates of inherited wealth instead of proxies of parental wealth so as to provide a much more precise analysis of the implications of marital sorting on inequality. Moreover, this article explores a new dimension of assortative mating by focusing on permanent labour income. In comparison with education, marital sorting based on income offers a more accurate measure of financial resources. Above all, using income instead of education allows us to compare spousal sources of wealth directly by investigating the equivalence between inheritance and labour income.
8Our empirical strategy is divided into two parts. First, we estimate the extent of assortative mating for each source of wealth (inheritance and labour income) by relying on correlations and risk ratios. For each type of statistical test, we estimate a net effect by controlling for potential differences in observable characteristics between inheritors and income earners (age and level of education). We then test the robustness of these results by using different definitions of labour income or inheritance and by focusing on specific sub-samples.
I – Theoretical framework
1 – Determinants of marital sorting
9Two main mechanisms can be invoked to explain the similarity of spousal traits. The limited social diversity at school or in neighbourhoods conditions the universe of potential partners. The example of school is key to apprehending the role of social background in marital choices. Evidence from the United States, (Lauman et al., 1994) indicates that almost one-fourth of married couples met in school, and the share is around 15% in France (Bozon and Rault, 2013). [3] Holmlund (2008) studied the impact of an educational reform in Sweden which increased the minimum school leaving age and abolished streaming. The author argues that this reform has significantly increased social diversity and has reduced assortative mating based on social background. There is no such direct evidence in France, but the importance of socioeconomic background in accessing elite schools suggests that this channel is likely to be decisive. Arrondel and Grange (1993) studied the strategies implemented by certain families in order to preserve or increase the social rank of the dynasty. “Rallies” [**] are a visible manifestation of such behaviour. Pinçon and Pinçon-Charlot (2000) note that “The rally almost always reaches its goal: to make sure that young people do not ruin a brilliant future […] by a bad marriage which would break up the aristocratic or bourgeois dynasty. There is no free competition in the aristocratic marriage market’’. Inherited wealth is a first-order condition for entering this specific marriage market. Even if such matrimonial strategies are probably limited to the very top of the social hierarchy, the partner’s socioeconomic background is an important consideration at all levels of society. [4] Bourdieu (1979) also explains that the social prestige and symbolic power of inheritance may also influence marital strategies.
10Socioeconomic background also affects individual preferences. Recent studies (Kimball et al., 2009; Dohmen et al., 2012; Arrondel and Frémeaux, 2014) provide evidence of positive assortative mating based on attitudes to risk and time. Besides, if preferences are partly influenced by the composition of wealth, and hence by the importance of inheritance, then marital sorting may be partly explained by the similarity of spouses’ preferences. Arrondel and Masson (2007) suggest that heirs are more altruistic and are more likely, in turn, to leave a bequest to their own children. The importance of sharing similar dynastic preferences, and a similar conception of family may thus partly explain our results. The effects of these mechanisms may differ according to the composition of wealth.
2 – Substitutability between inheritance and labour income
11The monetary inputs of a household can come from two different sources: inheritance or labour income. The composition of wealth, and more specifically the share of inherited wealth, varies across households. The main difference between these two inputs lies in their nature. Inheritance is, by definition, an inherited input while labour income is acquired. This difference determines the mechanisms of marital sorting.
12The bi-dimensionality of the analysis allows us to address the issue of substitutability between the two sources of wealth. There are two main categories of reasons why inheritance and labour income may not be substitutes. The first is linked to the differences between income earners and inheritors. Among observable traits, age and education are obvious candidates. Unobservable traits, like social prestige for instance, which can be correlated to inheritance, may also affect marital decisions. Furthermore, the source of wealth, and socioeconomic background more generally, influence people’s preferences. For instance, if inheritors share the same dynastic preferences, then heirs and heiresses may be more likely to meet and to marry.
13The second type of explanation relates to the differences between labour income and inheritance per se. First, the exogenous nature of inheritance contrasts with the endogenous nature of labour income, which depends on a decision to enter the labour market. As a consequence, observing a couple just after union formation or several years later does not produce the same results. Second, labour income and inheritance differ with respect to the timing of receipt and the degree of uncertainty. Inheritance is usually received in one or two transfers and depends on the volatility of asset prices, while labour income is spread over the whole life and is affected by career choices and macroeconomic factors. The difference arises not from the degree of uncertainty (which is difficult to measure and compare) but rather from its nature. Last, matrimonial property regimes have an influence because they govern the transmission of wealth between spouses and children after death or divorce. In France, the default legal regime is the “community of acquisitions’’ regime, whereby each spouse remains the sole owner of his or her inherited assets and of assets acquired before the marriage (so-called “separate assets’’), but the returns to these assets are considered community property, along with other income flows including labour income. In the event of divorce, the community assets are shared equally between husband and wife but each spouse keeps his or her separate assets. In other words, inheritance is protected in case of divorce while labour income can be “captured’’ via the redistribution of common assets. So an equal level of wealth may lead to an unequal redistribution of assets among spouses because the composition of their respective wealth is not the same. For this reason, the source of wealth affects the choice of matrimonial regime (Frémeaux and Leturcq, 2014) and, by extension, preferences in the choice of a partner.
II – Data
1 – The Actifs Financiers and Patrimoine French wealth surveys
14Since 1986, the French National Institute for Statistics and Economic Studies (INSEE) has conducted a national survey on wealth every six years. In this article, we use a pool of the last four available waves: 1992, 1998, 2004 and 2010. We consider all couples, married or not. In all, our sample contains 27,723 observations: 7,050 in 1992, 6,708 in 1998, 5,793 in 2004 and 8,172 in 2010. [5]
15Annual labour income is defined as the sum of wages, mixed income, pensions and unemployment benefits. This variable is collected at the individual level and self-reported in 1992 and 1998. In 2004 and 2010, the information about labour income is matched with fiscal data.
16Data on gifts and bequests are collected at the individual level and provide information about all wealth transmissions received up to the survey date. The value of each transmission is self-reported. For each transmission, we know the nature of the transmission (bequest or inter vivos gift), the type of assets transmitted, the identity of both legatee and successor, and the year of the gift/bequest.
2 – Estimating total inheritance and permanent income
Expected inheritance
17Data on the inheritances and parental wealth of both partners shed light on an important dimension of marital sorting. The information about inheritance is partial, however. We know the characteristics of observed wealth transmissions (i.e. already received) but not about future transmissions; [6] less than 30% of our sample reported positive inherited wealth at the time of the survey (Appendix Table B.3). There are two reasons for this low percentage: respondents had not inherited either because their parents were not rich enough to leave a bequest, or because they were still alive. Yet if we ignore these expected bequests or gifts, we assume that people do not consider their own (or their spousal) future wealth when they make their marriage decisions. [7] This assumption is unrealistic; we must also take account of parental wealth that has not yet been transmitted. This correction allows us to consider total inherited wealth (received and expected) and to go farther than existing studies on dowries or parental wealth. In this section we describe the basic aspects of these imputations.
18Our analysis proceeds in two stages. First, on the basis of information about respondents’ parents (Are they still alive? Do they own any assets? Have they experienced severe financial difficulties? etc.) we identify people who are likely to receive a bequest. We then impute the expected inheritance based on information about the parents’ wealth. The amount of their wealth is not known, but we know what types of assets they hold (real estate, land, equities, bonds, life insurance contracts, professional assets). We also take account of the parents’ occupational status and the respondent’s number of siblings (to determine the number of direct heirs). Appendix A.2 presents the technical aspects of the methodology.
19The final share of inheritors (who have received or will receive a bequest) represents around 55-60% of our sample, a proportion comparable with fiscal data (Arrondel and Masson, 2008). [8] However, we may slightly underestimate the proportion of inheritors, notably at the bottom of the distribution, since low-value bequests are not fully captured by survey data. Appendix B presents the distributions of labour income and inheritance as well as general descriptive statistics.
Permanent labour income
20Using income at the time of the survey raises two issues since the household division of labour, and hence income, depends on decisions taken after union formation. Moreover, current income may be affected by transitory shocks and depends on an individual’s position in the life cycle. This problem also arises when estimating intergenerational mobility. Estimates based on annual income lead to under-estimation of the intergenerational correlation. Solon (1992) shows that estimates based on current income averaged over several years (generally 3 to 5 years) are higher and more accurate than those using a single year of income. This finding could logically be applied to assortative mating. In absence of panel data (and hence of incomes averaged over several years), the alternative solution is to consider spouses’ permanent income. Beyond the life cycle issue, comparing current and permanent income allows us to investigate the role of labour income. Specifically, income can be used as a resource for the household, but also as a signal about the value of the spouse (Chiappori et al., 2012a). Estimating permanent income thus avoids the main pitfalls of existing studies (i.e. measurement errors linked to the observation of a single year of income) and enables us to analyse household specialization.
21Following Lollivier and Verger (1999), individual permanent income is estimated from current income. We decompose individual permanent income as a function of two elements: the first is age, to take account of both life cycle position and (quasi) permanent and exogenous traits (education, [9] socioeconomic background, job sector), and the second is a structural element to capture changes in standard of living. However, this estimate of permanent income cannot be imputed on the whole sample. Specifically, self-employed people are excluded because of the high volatility of their labour income. Second, current income must be strictly positive in order to approximate permanent income. As a consequence, we estimate permanent income on a final sample of 17,384 couples. These corrections affect the sample size and composition. Taking couples whose current labour income is positive for both spouses obliges us to focus exclusively on couples who were dual earners at the time of the survey, so the results must be interpreted with caution. [10] Nevertheless, for most of our estimates, the difference between samples is not significant. Appendix A.3 presents the technical aspects of our imputation.
III – Empirical analysis
22Our empirical strategy is divided into two parts. First, we estimate the extent of marital sorting by spousal inherited wealth and labour income. Second, we complete our analysis by studying the substitutability between these sources of wealth. Similar econometric specifications are implemented. For both research questions, we analyse levels (log correlations) and positions (relative positions in the distribution). Appendix C presents the robustness tests and the complementary results.
1 – Assortative mating
Correlations
23Before analysing our estimates of inherited wealth, we begin with a study of marital sorting using two categories of parental wealth. Using parental wealth allows us to provide evidence with non-estimated variables. We divide the population into two categories: no parental wealth and positive parental wealth. [11] Table 1A presents the contingency table for our whole sample. It provides evidence of positive assortative mating by parental wealth. Specifically, 73% of men reporting positive parental wealth are in a couple with a woman in the same wealth range, and 27% with a woman whose parents have no wealth. The proportions are 77.5% and 22.5% for women. In Table 1B, we repeat the same exercise with inheritance by dividing our population into non-inheritors (no wealth) and inheritors (positive wealth). The extent of marital sorting is similar. These findings confirm the robustness of our measure of inherited wealth.
Sorting by parental wealth
Sorting by parental wealth
Sorting by inherited wealth
Sorting by inherited wealth
Interpretation: (Table 1A) 72.6% of men with positive parental wealth are married to a woman in the same situation. The percentage is 77.5% for women.Note: The percentages for men are read horizontally, those for women (in italics) are read vertically. Chi-squared: Pr = < 0.001.
24These binary variables are of limited use, however, as we do not consider the value of wealth. The indicator we now use to measure the similarity of endowments between spouses is the partial correlation. More specifically, we regress the logarithm [12] of either labour income or inheritance on covariates (for each spouse, separately) and then estimate the correlation between error terms. This standard econometric test allows us to compare our estimates with the existing literature. The covariates vary depending on the specification. First, we only include the age of spouses to control for potential life-cycle bias. Then, we control for each spouse’s human capital (measured by the number of years of education) in order to estimate the share of assortative mating explained beyond any sorting in terms of education. [13] For each specification, we add a time fixed effect to control for potential differences between surveys. Last but not least, like most studies of assortative mating, we face a standard selection issue because our sample is a stock of households who report being in a couple at the time of the survey. Our estimates may therefore be biased if, for example, the couples with less similar traits are more likely to separate than the average. To address this problem, we take the date of couple formation into account and we run our tests on a sub-sample of recent couples who met less than 10 years before the survey. [14] Moreover, focusing on recent couples allows us to observe them when they are less specialized, and not very different from when they first met. However, by doing so we may also capture a change in the patterns of assortative mating among young couples. Unfortunately, without panel data we cannot disentangle the selection and cohort effects.
25Table 2 presents the estimate of marital sorting by inheritance, current labour income and permanent labour income. In Panel A, we provide estimates for inheritance and current labour income for the whole sample, while Panel B gives the correlations for all variables of the sub-sample of couples for whose permanent income can be estimated. Panel A shows that correlations are positive for both sources of wealth, but stronger for inheritance than for labour income (0.25 versus 0.12). Controlling for spousal education (column 2) explains a limited share of sorting (20% for inheritance, 10% for current income). Panel B confirms that using current labour income leads to an underestimation of marital sorting. The correlation for permanent income equals 0.46, while sorting based on current income is around 0.21. The correlation for inheritance decreases slightly on this specific sample. [15] For the sake of comparison, we also control for educational attainment. The effect of education is higher on this sample, explaining almost 60% of the relationship. For permanent income, the share is 70%. While this result comes as no surprise (since permanent income is estimated as a function of educational attainment), this specification demonstrates that the effect of educational assortative mating is very strong when we consider dual-earner couples. The comparison between spouses’ current and permanent labour income reveals the extent of measurement error caused by life-cycle biases. Using one-year current income leads to a substantial underestimation of the relationship between spousal earnings. This comparison also suggests that human capital can be used as a signal on the marriage market.
Assortative mating – Partial correlation estimates of marital sorting
Assortative mating – Partial correlation estimates of marital sorting
Note: Panel A includes all couples in the 1992-2010 waves of the Patrimoine surveys. Panel B includes all couples for whom we can estimate permanent income. Columns 3 and 4 include all couples who met less than 10 years before the survey. The coefficients represent the partial correlations between the logarithms of the amounts.Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.
26Columns 3 and 4 of Table 2 provide results for a sub-sample of recent couples. The correlation coefficients are not substantially different. This leads to two conclusions. First, the timing of bequests does not seem to matter. The share of actual inheritors (reporting positive wealth at the time of the survey) among these recent (and therefore young) couples is lower than the average but it does not significantly modify the extent of marital sorting. This result suggests that assortative mating relies more on socioeconomic background than on inherited wealth per se when spouses meet. Second, the lower sorting on permanent income in these couples may be explained in two ways, through either a selection effect or a cohort effect. However, in the absence of panel data and given the lack of consensus in France on changes in marital sorting based on education, [16] it is impossible to properly disentangle between these competing theories.
27Supplementary econometric tests presented in Appendix Table C.3 demonstrate that these results are robust. More specifically, we run these regressions with observed inheritance only and then on couples for whom both spouses are inheritors (to compare our estimates with those of Charles et al., 2013). This test confirms that the estimate is not biased by the estimates of permanent income and total inheritance. We then add extra control variables in order to consider potential differences between inheritors and income earners that might affect marital sorting (number of children, timing of inheritance, type of matrimonial property regime, nationality, year of couple formation, and previous experience of divorce). All these extra controls explain no more than 15% of the overall marital sorting by inheritance. This indicates that most marital sorting is explained by unobservable characteristics, probably linked to the socialization process and to preferences.
Risk ratios of marital sorting
Risk ratios of marital sorting
Interpretation: The coefficient 4.29 (at the bottom of the first column) means that women in the top 5% of the inheritance distribution are 4.29 times more likely to be in a couple with a man in the top 5% of the inheritance distribution than women in the bottom 95% of the inheritance distribution.Note: Columns 1, 2, 4 and 5 include all couples in the 1992-2010 waves of the Patrimoine surveys. Columns 3 and 6 include all couples for whom we can estimate permanent income. Robust standard errors in parentheses.
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.
28Did the patterns of assortative mating change between 1992 and 2010? The extent of marital sorting by inherited wealth does not present any clear trend, for either the whole sample or for the recent couples. For labour income, the pattern depends on the definition. We observe a decrease for current labour income but an increase for the permanent estimate. When focusing on the recent couples only, the extent of marital sorting is quite stable, in line with trends revealed by existing studies for the United Kingdom (Lise and Seitz, 2011) or the United States (Schwartz, 2010). The correlation between spousal earnings in these countries increased above all between 1970 and 1990 due to the rise in female labour force participation over that period.
29While comparison with existing studies on assortative mating is complex, our estimates for labour income are close to those of Lise and Seitz (2011), and Schwartz (2010). For inheritance, the work of Charles et al. (2013) is most similar to our own. The authors estimate a correlation between spouses’ parental wealth of around 0.4, after controlling for race and age. The transition from parental wealth to inheritance depends on inheritance taxation and the number of siblings (and hence the number of heirs). [17] Even though these issues are not addressed by the authors, we can conclude that the estimates for France and the United States are reasonably close. It is difficult to make comparisons with other French studies, which focus mainly on education and social status. Nonetheless, parallels can be drawn between similarity of inheritance and parents’ social status, with authors such as Bozon (1991) highlighting a strong similarity of socioeconomic background.
Risk ratios
30Correlations provide a summary linear measure of the relationship between two variables. However, sorting of couples at the tails of the distribution must also be studied specifically, given that the distribution is skewed. We therefore use risk ratios to compare marital decisions depending on individuals’ positions in the distribution. We divide the population into two parts: first, people below and above the median (P50), then the top 10% versus the bottom 90% (P90), and finally the top 5% versus the bottom 95% (P95). Then, for people below and above each threshold we estimate the probability of the event “living with a partner who is above the threshold’’ (probit model). Finally, we compute the ratio of these probabilities (Equation 1).
31Mathematically, we have:
33where Y is a dummy equal to 1 if the man/woman belongs to the top quantile and X is a dummy equal to 1 if the man/woman belongs to the same top quantile. A ratio equal to 1 implies random sorting (i.e. the probabilities of each category are equal).
34We consider the position of the male partner as a dependent variable and we compare the probabilities of the event for women. [18] Like for the partial correlations, we control for age first (columns 1 to 3) and then for age and spousal education.
35Table 3 presents the results. We provide evidence of non-linearity for both inheritance and labour income. Specifically, a women above the median in the distribution of inheritance is 1.6 times more likely than a woman below the median to be in a couple with a man above the median. In other words, for inheritance, this ratio is 60% higher than the incidence we would expect relative to random sorting. So the absence of parental wealth is a disadvantage for anyone wishing to form a couple with an inheritor. [19] These ratios are 1.3 for current income and 1.9 for permanent income. Furthermore, it is very likely for a non-inheritor to be with a top inheritor. For a woman, belonging to the top 10% of the inheritance distribution multiplies by 3.7 the chance of living with a male partner in the top 10% compared with a situation of random sorting. This ratio is similar for current labour income and is even larger for permanent income. Controlling for education only explains 10% of marital sorting based on inherited wealth, versus 47% for current income and 60% for permanent income. This estimate completes our analysis of marital sorting and indicates that the similarity of spousal traits grows with their rank in the distribution. The greater the difference between a man’s and a woman’s inherited wealth, the more unlikely they are to form a couple. Similar patterns exist for labour income but they are mostly explained by educational sorting.
36Even if comparison with the late nineteenth century is impossible, this article demonstrates that inheritance still matters in marital choices. The difficult question is, of course, “Why?’’. Our estimates, like those of Charles et al. (2013), indicate that controlling for spousal education accounts for only a small share of the overall marital sorting based on inherited wealth, and the introduction of supplementary control variables barely increases the explained share of sorting. As a consequence, in order to interpret these estimates we need to explore other channels, such as the socialization process and the role of preferences.
2 – The substitutability between inheritance and labour income
37The first part of this article has shown that marital sorting based on inheritance is substantial, especially at the top of the distribution. It would be interesting to know whether labour income can compensate for a lack of parental wealth. In other words, is there any substitutability between these two sources of wealth, and hence between human capital and inheritance? This question is crucial but complex. In real life, matching is multidimensional since spouses consider several economic and non-economic traits during the matching process. Therefore, considering only one trait simplifies the analysis of matching, but ignores the effects of other traits. Chiappori et al. (2012b) provide a bidimensional analysis, with a study of the trade-off between beauty and wages or education. In this article, we compare the roles of inherited and acquired traits in the choice of spouse. This also sheds light on the mechanisms behind marital sorting and helps us to better understand individual preferences.
38To tackle this issue we follow the empirical strategy that we used to measure the extent of marital sorting. Table 4 presents the cross-correlations of spouses’ labour income and inheritance. We use the same controls as in Table 2 and the same sub-sample of recent couples. Panel A describes the correlations between spouses’ current labour income and inheritance. A complete absence of substitutability would give correlations equal to 0. The observed correlations are positive and significant, but low (around 0.05). The comparison between estimates of marital sorting indicates that labour income is only partially substitutable for inheritance. Moreover, when education is taken into account, more than two thirds of the relationship is explained (column 2). Substitutability is greater among recent couples, but we still note the same effect of education. Panel B indicates that permanent income and inheritance are more closely related. Once again, education explains most of the effect, reflecting both the effect of education on income and the importance of marital sorting based on education. These findings must be interpreted with caution, however, because of the specificity of this sub-sample. This table shows that substitutability between acquired and inherited traits is low.
Partial correlation estimates of substitutability between inheritance and labour income
Partial correlation estimates of substitutability between inheritance and labour income
Note: Panel A includes all couples in the 1992-2010 waves of the Patrimoine surveys. Panel B includes all couples for whom we can estimate permanent income. Columns 3 and 4 include all couples who met less than 10 years before the survey.Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.
39As before, we focus on the top of the distribution. The econometric test is more complex than for risk ratios as we consider two dimensions. We implement the following specification (Equation 2):
41where the dependent and explanatory variables are dummies equal to 1 if the individual is in the top 10% of the inheritance distribution or the income distribution (depending on the specification). So, β1 (resp. β2) represents the marginal probability of a top inheritor (resp. a top income earner) of living with a partner in the top 10%. In other words, it measures the marginal effect of being in a top position in a given dimension. An equality between β1 and β2 would mean that inheritance and labour income are perfect substitutes in this part of the distribution. Xij is a vector of control variables in which we include spousal age and a time fixed effect. All the specifications are estimated by probit with robust standard errors.
42Table 5 presents the results. Inheritance and labour income are poor substitutes in marital choices even at the upper tail of the distribution. Top heirs are more likely to be in a couple with top heiresses than with top income earners. Specifically, being in the top 10% of the female inheritance distribution increases the probability of living with a top heir by 20%, while the marginal probability for a top income earner is only 6.7%. There is a clear separation between the sources of wealth because this result also holds for permanent income (column 2). In Panel B, we reverse the roles to study these effects for men. The results are similar, suggesting the absence of a gender difference in preferences. Therefore, our results suggest that labour income does not compensate for a lack of inherited wealth, including – and perhaps even more so – at the top of the distribution. [20]
Variations in the probability of having a spouse at the top of the income or inheritance distribution by ego’s position in the distribution
Variations in the probability of having a spouse at the top of the income or inheritance distribution by ego’s position in the distribution
Interpretation: 0.200 (first column) means that for a woman, belonging to the top 10% of inheritance distribution increases by 20% the probability of being in a couple with a top heir; 0.067 is the same probability for a woman belonging to the top 10% of permanent labour income distribution; the third coefficient is the difference.Note: Sample includes all couples for whom we can estimate permanent income. In panel A, the male partner’s traits are the dependent variable; in panel B, the female partner’s traits are the dependent variable. The coefficients are marginal effects from probit estimation (with robust standard errors). P-values in brackets.
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.
43Robustness tests (Appendix tables C.4 and C.5) confirm these results. In Appendix Table C.4 we present additional results. We replicate the bidimensional analyses of the model of Equation (2), this time looking at spouses below and above the median. Substitutability is higher than in the top 10% of the distribution, but remains low. A final important question concerns individuals with both inherited wealth and high labour income. In Appendix Table C5 we replicate the bidimensional test, excluding individuals who are in the top 10% of both distributions. The low observed substitutability disappears, since the probability for top male income earners of being in a couple with top female inheritors is not significantly different from 0.
44Again, we may wonder if any clear pattern emerges over our study period. The hypothesis of perfect substitutability is rejected for all waves of the Patrimoine survey, both for the entire population and for the top 10% of the distribution. However, our estimates indicate that substitutability follows a U-shaped curve over the period 1992-2010.
45Several assumptions are required (perfect capital market, a perfect marriage market and also assumptions about the time horizon of couples) to test the perfect substitutability between inheritance and labour income. In this article we indicate that inheritance and labour income are poor substitutes, but it is difficult to rule out the possibility that one of the required assumptions has been violated. However, this very new evidence is revelatory because it suggests a disparity of mechanisms behind marital sorting based on inheritance and labour income. Our conclusions open wide prospects for future research.
IV – Discussion and conclusion
46This article presents an original analysis of the extent of marital sorting based on inherited wealth and permanent income. Our use of the French Patrimoine survey allows us to overcome the main drawbacks of the existing literature. More specifically, our estimates of expected inheritance and permanent income greatly improve the analysis of marital sorting and provide new results.
47Our results reveal a strong similarity of spousal traits in terms of inheritance and labour income. Our preferred estimate indicates that the correlation of inheritance between spouses equals 0.25. Marital sorting is significantly stronger for inherited wealth than for labour income, especially at the top of the distribution. Neither the timing of inheritance nor the selection into or out of marriage affect our results.
48The second research question addressed in this article concerns people’s sensitivity to their spouse’s source of wealth. There is clear proof that labour income and inheritance are poor substitutes. Specifically, there is a partition between the two dimensions: heirs marry heiresses (or the reverse) and income earners marry income earners. The degree of marital sorting remains stable between 1992 and 2010, and substitutability increases slightly at the end of the study period.
49While education explains a large share of martial sorting in terms of labour income, it explains only 20% of the correlation for inheritance. We suggest two complementary ideas to analyse these results. First, the socialization process may explain why people from the same socioeconomic background (and hence with potentially similar levels of inheritance) have a greater opportunity for interaction and are more likely to share common tastes. Moreover, matrimonial strategies and social prestige attached to inheritance are likely to reinforce this mechanism. Second, inherited wealth and labour income differ in terms of their timing and of their treatment under different matrimonial property regimes. These differences may affect individual preferences and influence the choice of spouse.
50Thorough data testing has led to new findings on the question of marital sorting and has substantially improved the quality of analysis. Our results have major implications for the analysis of intergenerational inequality. Lefranc and Trannoy (2005) estimate that the elasticity of son’s with respect to father’s income [21] is around 0.4 (0.3 for father-daughter elasticity) in France, in an intermediate position between the United States and Nordic countries. Arrondel (2013) repeats the exercise for wealth and finds that elasticity equals 0.22. The Atkinson’s illustration used in the introduction of this article suggests that our findings are a step forward in the study of this relationship. The Patrimoine survey does not allow us to estimate precisely the role of assortative mating in the reproduction of inequalities. We can analyse this relationship here by considering two issues: the impact of the source of wealth and recent changes in the family.
51The existing studies about the role of marital sorting in inequality (Burtless, 1999; Schwartz, 2010) do not consider the source of wealth but only the household total wealth. We know that the extent of marital sorting is higher for inherited wealth and that inheritance is more unequally distributed than labour income, but to have a complete view of this issue, a third parameter needs to be considered. Piketty (2011) makes clear that inheritance flows have returned to their pre-World War I levels and that they are unlikely to decrease in the coming decades. As a consequence, inherited wealth relative to labour income will become increasingly important in marital choices. Given the poor substitutability between these sources of wealth, the possibility of marrying up without parental wealth, and hence of achieving social mobility, is likely to be more limited. While the direct impact of spousal inheritance on wealth inequality is difficult to estimate, our results suggest that the effect of marital sorting by inheritance is likely to become ever more substantial.
52In order to fully understand the implications of our results we must also consider changes in the family since the time of Balzac and Maupassant in the nineteenth century. At least two complementary recent changes are likely to make the consequences of assortative mating on inequality even more complex to study. First, divorce and remarriage are more frequent nowadays. This implies that people can have several partners during their life and that several wealth transfers between spouses can occur (alimonies, redistribution of common assets, etc.). So the current wealth that will be transmitted by an individual to his or her inheritors depends on his or her current spouse but also on past partners. Second, recent evidence indicates that there is an increasing tendency among French couples to separate their assets either by not marrying (equivalent, de facto, to a separation of assets) or by choosing separate property contracts when they marry. Frémeaux and Leturcq (2014) estimate that more than 15% of newly married couples chose this type of contract. Even if, in fine, all parental wealth is transmitted to children, these changes considerably affect the timing of these transmissions. [22] Timing matters because it affects not only the path of wealth accumulation but also more general economic decisions regarding labour supply, for instance (Holtz-Eakin et al., 1993; Joulfaian and Wilhem, 1994). These issues are beyond the scope of this article but changes in family structure must be taken into account when examining the role of marital sorting in intergenerational inequality.
53The distinction between inheritance and permanent income opens new scope for analysing the dynamics of inequality. With the increasing inequality of wealth in rich countries, in Europe especially, matrimonial strategies could play a more important role in intergenerational mobility.
Acknowledgements
I am grateful to Thomas Piketty, Luc Arrondel, Clément de Chaisemartin, Olivier Donni, Lucie Gadenne, Arnaud Lefranc, Marion Leturcq, Anne Solaz, Gabriel Zucman and to three anonymous referees. I also thank seminar and conference participants in San Francisco (PAA), Paris (AFSE, INED and PSE), Cergy-Pontoise (THEMA), Brest (JMA) and Malaga (EEA annual congress) for their comments. I would also like to thank the Centre Maurice Halbwachs and INSEE for providing the databases. I acknowledge financial support from Région Île-de-France and from Labex MMEDII (ANR11-LBX-0023-01).A.1 – Distribution and discounting of wealth transmissions
54Wealth transmissions are grouped into brackets by value in 1992, 2004 and 2010. For these waves, we estimate simulated residuals (Lollivier and Verger, 1989) to obtain continuous values. More specifically, we use a set of exogenous variables to estimate the real value declared by respondents, conditional on the bracket. These variables are: nature of transmission (bequest or gift), type of assets received (real estate, land, financial assets, cash, etc.) and the socioeconomic status of the respondents’ parents. We also add a year fixed effect to all econometric specifications. The comparison between surveys does not present any significant differences.
55Moreover, we discount the value of inheritance to take account of changes in the value of inherited wealth. Most gifts/bequests comprise real estate and moveable assets. We use a composite index (the same as in Piketty (2011)) that takes price variations into account for all types of assets. This index incorporates long-term changes in prices of consumption goods (CPI), real estate, equities and bonds. We compute a weighted average of these changes in order to build our index. Finally we use information about the year of transmission in order to discount each of them.
A.2 – Imputations: expected inheritance and missing values
56One of the main drawbacks of surveys on inheritance concerns the absence of information on expected inherited wealth. Even though future wealth is an important factor in individual decisions, this aspect of inheritance is ignored. Our method is divided into two steps: first we identify the potential inheritors and then we estimate an expected bequest based on observable characteristics.
57To identify the potential inheritors we use information about respondents’ parental characteristics. First, we only keep people who have at least one living parent. Then, we exclude people who experienced frequent periods of poverty during childhood. We also need to consider people who have already received wealth transmissions and whose parents are still alive. If people have received less than €15,000 (2010 euros) with two living parents or less than €7,500 with only one living parent then they are considered as potential inheritors. Choosing higher thresholds leads to a modest increase in the share of inheritors but it also adds some imprecision in the estimation. Last, respondents with parents who do not hold any assets are not considered as potential inheritors.
58We then apply a standard missing values procedure to estimate the expected inheritance. As explanatory variables we use: receipt of financial help from the respondent’s parents (dummy), receipt of inter vivos gift (dummy), type of assets held by the respondent’s parents (real estate, land, equities, bonds, life insurance contracts, professional assets), number of siblings and parents’ socioeconomic status. Finally, we add this expected wealth to the observed wealth (if any).
59The second type of imputation we implement concerns missing values. The proportion of missing values in the declaration of labour income and bequests amounts to 2-3% of all reported values. We use the observed self-declared values as well as observable characteristics to estimate these missing values. For labour income, we use age, education, parents’ socioeconomic status (occupational category), job sector and work experience. For inheritance, we use the respondent’s socioeconomic status, parents’ socioeconomic status and the nature of the transmitted assets (real estate, financial assets, life insurance, etc.).
A.3 – Estimation of permanent labour income
60The respondent’s current annual labour income may not be representative of his/her lifetime income because of life-cycle bias and transitory shocks. We correct for this potential issue by estimating permanent income from current income and other individual information, using the method developed by Lollivier and Verger (1999) whereby individual permanent income is decomposed as a function of three elements: the individual’s age, c(a) (reflecting the variations of income caused by age); a structural part, s(t) (reflecting overall changes in standard of living), and quasi-permanent characteristics Xi.
61This estimate of permanent income cannot be imputed on the whole sample. Specifically, self-employed people are excluded because of the high volatility of their labour income. Second, we need to have a strictly positive current income. Appendix Table B.1 indicates that current income equals 0 for approximately 15-30% of women in each survey. As a consequence, we have a final sample of 17,384 couples, compared with 27,723 in the initial sample. These corrections affect both the size and composition of the sample. Taking couples for whom current labour income of both spouses is positive obliges us to focus exclusively on dual-earner couples at the time of the survey, and this selection is partially endogenous with marital choices. As a consequence, the results must be interpreted with caution, given the differences with respect to the initial sample.
62We run our model on 8 sub-populations defined by gender (2 categories) and initial education (4 categories). The division by gender is based on the fact that men and women do not have the same careers; women’s careers are much more frequently interrupted. Level of education affects wage profiles: the lower an individual’s educational level, the slower his or her wage growth. Focusing on education allows us to consider a permanent individual characteristic. Given that we have different generations in our sample, we do not consider level of qualification but rather the relative length of education (measured by the number of years of education) within each cohort. Finally, we introduce permanent and exogenous characteristics such as the social position of parents and the job sector in order to improve the specification. The final specification is the following:
64where yi (t,a) is the annual wage, Xi the permanent characteristics and vit an error term.
65The final step of this imputation consists in summing (from date of labour market entry to death) and discounting incomes. To do so we assume constant purchasing power via a discount rate equal to the real interest rate. This method allows us to consider career effects and also variations in purchasing power.
A.4 – Measurement errors
66As self-declared variables are used, we must check for measurement errors and their potential effects on our estimations. Beyond sampling errors, measurement errors consist in under- or over-reporting of labour income and inheritance. In this article, we are more concerned about the correlation of measurement errors between spouses than about measurement errors per se. Indeed, in the presence of uncorrelated classic measurement errors, our correlations are liable to be under-estimated. If measurement errors are correlated, the effect on measures of marital sorting are likely to be biased upwards or downwards depending on the direction of the correlation.
67The differing nature of the income variables across surveys allows us to test for the presence of this potential bias. In 2004 and 2010, the survey is matched with fiscal data, and in 1992 and 2004 incomes are self-declared. Appendix Table B1 indicates that fiscal incomes are higher than self-declared incomes. However, the correlations of labour income between spouses do not present significant differences, suggesting that measurement errors for labour income do not affect our estimates.
68For inheritance, such comparison across surveys is not possible since values are self-declared for all waves. Nevertheless, gender differences do exist. Specifically, Appendix Table B.2 suggests than men seem to overestimate their inheritance compared with women (See Appendix B). However, this declaration bias is not likely to bias our estimate of marital sorting since the whole male distribution seems to be affected. Again, the econometric tests implemented survey by survey do not present major differences, so the existence and the extent of potential bias are difficult to determine.
69This issue might affect our estimates only if there were a differential in measurement errors between labour income and inheritance. Again, the 2004 and 2010 waves provide an answer. Indeed, for these waves, the measurement error for labour income is limited to a sampling error. Here too, no difference is observed between surveys, so the extent of potential bias is likely to be limited.
B – Descriptive Statistics
70Appendix Tables B.1 and B.2 detail the distribution of labour income and inheritance from 1992 to 2010. We restrict the sample to men and women in a couple. For both dimensions, we describe the sample mean, the thresholds and averages by decile. We divide the top decile into three parts: P90-95, P95-99 and P99-100.
71Appendix Table B.1 presents the labour income distribution for the four waves. Annual labour income is the sum of wages, mixed income, pensions and unemployment benefits. It is self-declared in 1992 and 1998. In 2004 and 2010, survey data are matched with fiscal returns. The differences between the surveys reflect both changes in the labour income distribution in France (labour market feminization, change in top income shares) and the construction of the survey. Women have lower incomes than men, but women’s average income grows over time. The comparison with existing studies is limited because we only consider couples (and not the adult population) and labour income. Godechot (2012) focuses on individual wages only and finds that the wage share held by the top decile is around 26-27%. This suggests that our income distribution is relatively comparable.
72Appendix Table B.2 demonstrates that the inheritance distribution is more skewed. Almost half of the sample receives no inheritance, and the top 10% holds more than 60% of total inherited wealth. Fiscal data show that among the strictly positive estates, the top 10% of the largest bequests represent more than 50% of total bequests in 2000 (Arrondel and Masson, 2008). Life insurance and inter vivos gifts are, under certain circumstances, not taken into account in estate returns. The introduction of this type of asset in the self-declared values we use, as well as the difference in terms of samples (whole population versus couples) may explain the differences between these estimates and ours. An unexpected result needs to be considered. Under French law, bequests must be shared equally between children who inherit from their parents (and hence between the sexes). A gap exists, however, and arises mainly from self-declaration. The share of men who report positive bequests is higher (2-3 points higher than women) and men declare higher values than women (+€20,000 on average for the sub-sample of respondents with strictly positive bequests). Part of this gap can be explained by the age gap between men and women (Appendix Table B.3) but it also raises the question of the existence of a gender bias in the evaluation of inherited wealth.
73Appendix Table B.3 presents more general characteristics of our sample. The main information of this table concerns the share of inheritors. Between 1992 and 2010, the share of future inheritors becomes larger than the share of actual inheritors (i.e. declaring positive inherited wealth at the time of the survey). The increase in life expectancy mechanically delays the transmission of wealth between parents and children and reduces the share of observed inheritors. Nevertheless, the imputation of expected bequests leads to stability of the total share of inheritors.
Appendix Table B.1. Income distribution
Appendix Table B.1. Income distribution
Note: Income = wages + mixed income + pensions + unemployment benefits; self-declaration at the individual level for 1992 and 1998, matching with fiscal data in 2004 and 2010.Coverage: All couples in the 1992-2010 waves of the Patrimoine surveys.
C – Supplementary results
74The two dimensions studied in this article, namely inherited wealth and labour income, are strongly correlated, so it is important to study the link between these two traits. Appendix Tables C.1 and C.2 confirm the two main results of our study. First, the tables indicate that the matching process is not random for inherited wealth and labour income. Second, there is a separation between the two dimensions, since a top heir is more likely to be with a top heiress than with a top income earner (and vice-versa). People cumulating top positions in the two distributions are more likely to be in couple with top income earner.
75One may worry about a bias in the results caused by people who are at the top 10% of the distribution for both inheritance and labour income. In Appendix Table C.2 we exclude couples for whom at least one spouse is in this situation. Their weight is below 3% for men and 2% for women.
76The results suggest that most substitutability between inheritance and labour income is actually due to people who occupy top positions for both sources of wealth. Indeed, when these people are excluded, the marginal probabilities for people with opposite sources of wealth are close to 0. This table confirms our conclusions and indicates that human capital is not sufficient to compensate for a lack of parental wealth.
77In Table C.4 we replicate the bidimensional test with a modification of the categories. Instead of focusing on the top 10%, we look at the categories below and above the median. Our estimates show that substitutability is greater than among the top 10%, but remains low. Specifically, the probability for an income earner of being in couple with an inheritor is higher than that of being with a non-inheritor. This result indicates that the importance of inheritance grows with the position in the distribution.
Appendix Table C.1. Relationship between inheritance and permanent income – top 50%*,●,■,◊
Appendix Table C.1. Relationship between inheritance and permanent income – top 50%*,●,■,◊
* Individual in the top half of both the income and the inheritance distributions.● Individual in the top half of the income distribution.
■ Individual in the top half of the inheritance distribution.
◊ Individual in the bottom half of both distributions.
Appendix Table C.2. Relationship between inheritance and permanent income - top 10%*,●,■,◊
* Individual in the top half of both the income and the inheritance distributions.● Individual in the top half of the income distribution.
■ Individual in the top half of the inheritance distribution.
◊ Individual in the bottom half of both distributions.
Appendix Table C.3. Marital sorting – Robustness tests, correlation coefficients[1],[2],[3],[4]
Appendix Table C.3. Marital sorting – Robustness tests, correlation coefficients[1],[2],[3],[4]
[1] Only observed inheritance is considered (i.e. declared at the time of the survey).[2] Couples where both members have a positive total inheritance.
[3] Extra control variables are added: marital status (non-married, married with community regimes, married with separate regimes), number of children, timing of inheritance (dummy equal to 1 in case of imputation of expected inheritance) and nationality.
[4] Two controls are added: year of couple formation and a dummy equal to 1 if (at least) one spouse already experienced divorce.
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.
Appendix Table C.4. Substitutability – top 50% Probability of having a spouse in the top 50% for inheritance or permanent income, by presence in the top 50% for inheritance or permanent income
Appendix Table C.4. Substitutability – top 50% Probability of having a spouse in the top 50% for inheritance or permanent income, by presence in the top 50% for inheritance or permanent income
Note: Sample includes all couples for whom we can estimate permanent income. In panel A, the male partner’s traits are dependent variables; in panel B, the female partner’s traits are dependent variables. Coefficients are marginal effects from probit estimation (with robust standard errors). P-values in parentheses.Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.
Appendix Table C.5. Substitutability - Excluding individuals in the top 10% of both distributions
Appendix Table C.5. Substitutability - Excluding individuals in the top 10% of both distributions
Note: Sample includes all couples for whom we can estimate permanent income. In panel A, the male partner’s traits are dependent variables; in panel B, the female partner’s traits are dependent variables. Coefficients are marginal effects from probit estimation (with robust standard errors). P-values in parentheses.Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01.
Notes
-
[*]
Thema, Université de Cergy-Pontoise.
-
[**]
Social events organized by upper-class families to initiate a meeting between young people from good families.
-
[1]
Throughout this article, the words “inheritance’’, “inherited wealth’’ or “bequest’’ will refer to the sum of bequests and gifts received by an individual.
- [2]
-
[3]
The difference between the two studies arises from the broader definition of school in the study by Lauman et al., which also took account of peer groups formed at school.
-
[4]
Arrondel and Frémeaux (2014) note that 60% of couples believe that a similar socioeconomic background is important for union stability. Only 20% of couples express a similar view with respect to income.
-
[5]
Estimates are weighted to correct for sampling ratio effects. In addition, to ensure that each wave has the same weight (25%) in our final database, their weights are normalized.
-
[6]
See Appendix A.1 for the treatment of wealth transmissions.
-
[7]
Bel Ami, the main character of the eponymous novel by Maupassant, is a perfect illustration. At the end of the novel, Bel Ami attempts to seduce the daughter of his wealthy patron. For Bel Ami, her current absence of wealth does not matter, since he knows that she is the sole future heir of a large family fortune.
-
[8]
The comparison is imperfect, however, since fiscal data identify the share of individuals who bequeath wealth after death. To estimate the share of inheritors, fertility and inheritance taxation must be taken into account. This comparison – the only one available – is instructive nonetheless.
-
[9]
Education is taken into account indirectly. To refine our estimate, the sample is divided into four categories by level of initial education.
-
[10]
For income, Table 2 shows that the correlation increases from 0.12 to 0.20, as expected, since the zero-values are excluded. For inheritance, the correlation decreases slightly when dual-earner couples are considered.
-
[11]
Parental wealth is positive if the respondent’s parents own(ed) at least one type of asset among the following: real estate, land, equities, bonds, life insurance contracts, professional assets.
-
[12]
Like in Browning et al. (1994), we keep the zero-values and we consider them as equal to 0. With this transformation, the evolution is linear for values close to 0 and then logarithmic for higher values.
-
[13]
For all specifications, age and education are continuous variables. Education is defined as the number of years of schooling (initial education). It would also have been interesting to consider the geographical location of households (since couples are more likely to form between individuals living in close proximity), but the variables in the Patrimoine survey are too imprecise.
-
[14]
Tests on couples who met less than five years before the survey produce similar results.
-
[15]
Self-employed and inactive people, and those declaring no current labour income more generally, are excluded. This selection may exclude couples for whom inherited wealth is positively correlated.
-
[16]
While Goux and Maurin (2003) emphasize a growing similarity of spousal education among recent cohorts, Forsé and Chauvel (1995) find stability and Vanderschelden (2006) notes a slight decrease in marital sorting based on education. It is also possible that our estimate of permanent income is less accurate for younger couples.
-
[17]
A simple difference in the number of siblings of each spouse can lead to large inequalities in terms of actual inheritance despite equal parental wealth.
-
[18]
The reverse specification, with the position of the female partner as a dependent variable, provides similar results.
-
[19]
Actually, the categories “inheritors’’ and “non-inheritors’’ cannot be exactly compared using the median. If the categories in Table 1B are used, we obtain a ratio of 1.53, versus 1.3 when the median is considered.
-
[20]
The wealth of income earners (capitalized sum of labour income and inheritance) is greater than that of heirs. Assuming linearity, their probability of marrying a given spouse is therefore higher. However, this wealth effect only marginally modifies the results by increasing the marginal probabilities of rich male and female heirs.
-
[21]
Measure of the impact of father’s income variation on son’s income variation.
-
[22]
Under the separation of property matrimonial regime, assets are bequeathed to the children of the deceased, whereas under a community of property regime, the spouse is protected. The growing popularity of separation of property marriages will lead to earlier inheritance by the children (unless measures are taken to protect the spouse).
