Marriage, Assortative Mating, and Wealth Inequality

Because wealth is almost always measured at the household level, so the marriage market determines the initial distribution of wealth across newly formed families — and because household returns to wealth, which govern how that wealth then grows, are themselves shaped by who marries whom and by who manages the money. The paper notes that after people marry “their individual wealth is merged and owned jointly through the life of the marriage,” and that surveys and administrative databases “rarely, if ever, attempt to measure individually-owned assets.” Despite this, “the debate surrounding trends in wealth inequality and wealth concentration has rarely focused on the marriage market and the extent of assortative mating.” Two directions of effect are laid out. Marriage per se is an equalising force: pooling resources and sharing investment responsibilities “reduce the weight of the tails, lowering inequality in both wealth and returns to wealth among married households compared to single individuals,” so societies with falling marriage rates or more time spent single will — ceteris paribus — see more household wealth inequality. But the composition of marriages matters too, because “people arrive at marriage with very heterogeneous asset levels” and because household wealth then evolves at a rate set by the household’s return.

A two-point example showing that assortative mating and the post-marriage wealth-management rule jointly determine how much return heterogeneity survives the transition from a population of singles to a population of couples. Suppose returns take only two values, 1% and 5%, each with probability one half for both men and women; the average return among singles is 3% with a standard deviation of 2%, and heterogeneity measured by the squared coefficient of variation Ψ is 0.44. Under random mating, Ψ among couples falls to 0.22 when spouses share management equally and to 0.18 when the higher-return spouse takes over. Under perfect positive assortative mating, Ψ stays at 0.44 under either rule. The reading offered: “Random assortative mating tends to reduce return heterogeneity among couples (a simple application of the law of large numbers), but less so when household wealth management responsibilities are shared equally,” because specialisation “reduces the incidence of the lower tail of the distribution of returns, raising the mean faster than it lowers the variance.” Perfect assortative mating “reproduces among couples the same extent of return heterogeneity observed among singles, regardless of who manages household wealth.”

Linked Norwegian administrative registries covering the whole resident population from 1967 to 2015, with tax records on income, assets and liabilities from 1993, a shareholder registry and private-business balance sheets from 2004, and unique identifiers linking spouses to each other and to their parents. The paper lists four advantages. Coverage of “all individuals in the population who are subject to income and wealth tax, including people at the very top of the wealth distribution,” which allows assortative mating to be examined separately across wealth groups — something the Spearman index alone cannot do — and which means the data are free from recall bias and from attrition other than mortality and emigration. Third-party reporting by employers, banks and financial intermediaries “without any top- or bottom-coding,” avoiding the measurement error and confidentiality censoring of surveys. Parent–child links, which let the authors test own versus parental wealth as the sorting variable. And a long panel dimension, “crucial to compute reliable pre- and post-marriage average returns to wealth.” The sample comprises all couples that marry, or cohabit with one or more children, between 2005 and 2014, including first and subsequent marriages. The authors also note an unusual institutional feature: between 2001 and 2010, Norwegian tax records were effectively public and anonymously searchable online, so “one could collect information about a potential spouse’s income and assets (and, in principle, their dynamics) at a click of a mouse.”

As the return on total assets (ROA): capital income plus accrued capital gains on financial and real assets, minus the cost of debt, divided by beginning-of-period gross wealth adjusted for within-year net flows following Dietz (1968). The Dietz adjustment assumes net flows occur on average at mid-year and “avoids overstatement (understatement) of returns due to active saving (dissaving) decisions made during the year.” The authors prefer the return on gross assets over the return on net worth for two stated reasons: “the sign of the return depends only on the sign of the yield (and not on that of net worth),” avoiding assigning positive returns to indebted households whose debt costs exceed asset income; and “a non-negligible fraction of households have zero or close to zero net worth, which makes the return to net worth undefined or implausibly large.” All return measures are net of inflation using the 2011 CPI and gross of taxes and subsidies. A parallel measure for liquid financial wealth is also reported. Pre-marriage returns are averaged over the available years excluding the four years leading up to marriage, to avoid contamination from cohabitation.

Substantial and highly non-uniform: an overall rank-rank slope of 0.2, but negative at the bottom, 0.31 above the 20th percentile and 0.73 in the top decile. The Spearman rank correlation, pooling all years, is 0.19. The heat map of wealth-ventile pairings shows “more mass along the main diagonal than elsewhere,” but with non-linearities: “people at the top of the distribution are more likely to marry people at the bottom than people in the middle of the distribution.” In the rank-rank plot, random matching would give a horizontal line at the median and perfect assortative mating a 45-degree line. At the very bottom quintile, where net worth is negative, “matching is, if anything, negative (people with debt do not match with people with debt).” The corresponding rank-rank slope for parental wealth is 0.12 — “still positive, but much more attenuated” — and more stable across the wealth scale, which the authors attribute to debt being much less of an issue for the older parental generation.

No. The elasticity falls when demographics are added but survives controls for education-pair and income-pair interactions at essentially the same magnitude. The unconditional rank-rank slope is 0.20 (s.e. 0.003). Adding age, years of schooling, whether the spouse holds an economics or business degree, county fixed effects, marriage order and year-of-marriage dummies lowers it to 0.11 (s.e. 0.003). Controlling for all possible interactions of the wife’s and husband’s education groups leaves the results “economically and statistically similar,” and the same holds when education interactions are replaced with husband–wife income-percentile interactions. The visual counterpart: among couples where both spouses have a college education or more, assortative mating on personal wealth persists — “in fact, even more than in the full sample” — and the same is true among couples both in the top income quartile.

On their own. Once own wealth is included, the parental-wealth coefficient “drops by a factor of 20 and loses significance.” Replicating Charles et al. (2013) by regressing the man’s parents’ wealth percentile on the woman’s gives a slope of 0.13, which falls to 0.046 (s.e. 0.003) with demographic controls. But in the regression of own wealth on own wealth that also includes both sets of parental wealth, the spouse’s parental wealth becomes insignificant, while the own-wealth elasticity is unaffected. The paper’s conclusion: “assortative mating on parental wealth emerges only because the latter tends to proxy for assortative mating on personal wealth.” The same regression yields an intergenerational wealth elasticity of 0.10, “in the ballpark of estimates in other papers, such as Boserup et al. (2014),” and because controlling for the man’s parents’ wealth leaves the assortative-mating coefficient unchanged, “intergenerational persistence in wealth and assortative mating on wealth are distinct phenomena.” On why young Norwegians have wealth to sort on at all: most leave the parental home for college, so “by the time they marry (in their early 30s) they have already saved (or been financially active) for about 10-15 years.”

Cohabitation, bequests, negative net worth, the choice of wealth measure, and the functional form. To rule out that pre-marriage wealth already reflects shared resources, the authors restrict to couples living in different counties four years before marriage — “by definition, not cohabiting” — and find results similar or “if anything suggesting even more assortative mating.” To rule out inherited wealth driving the own-wealth result, they restrict to individuals with both parents alive four years before marriage; results are unchanged, though the authors note “inter-vivos transfers could be a potential factor.” Restricting to non-negative net worth raises the measured sorting, consistent with the negative assortment at the bottom. Financial wealth “replicates closely that for net worth,” with a stronger intergenerational component “since parents do not usually transfer debt to their kids.” An extended net-worth measure including expected bequests (parental wealth divided by the number of siblings) gives a slightly smaller elasticity, 0.098 versus 0.109. A log–log specification on gross wealth gives an elasticity of 0.21, “implying that a 10% increase in the gross wealth of the woman is associated with a 2% increase in the average wealth of the matching partner.”

That sorting on pre-marriage returns to wealth is as strong as sorting on wealth itself — a Spearman rank correlation of 0.18 — and that it is not an artefact of sorting on wealth. The authors state plainly: “on this metric, sorting on returns is hence as strong as sorting on wealth,” and describe it as “a totally novel finding.” The rank-rank plot is “fairly linear… except for the very top, where men with extremely high realizations of wealth returns tend to experience a form of ‘mean reversion’,” marrying women much lower in the return distribution and vice versa. The main threat is scale dependence — a positive correlation between wealth and returns could make sorting on wealth look like sorting on returns. The visual check restricts to couples where both spouses are in the top quartile of the pre-marriage net-worth distribution, and assortative mating on returns “continues to hold (in fact, even more so than in the full sample).”

With progressively more granular controls for the couple’s joint position in the wealth distribution; the elasticity falls only from 0.14 to 0.13 and stays highly significant. The unconditional elasticity is 0.18. Adding age, years of schooling, economics/business degree, county fixed effects, marriage order and year-of-marriage dummies lowers it to 0.14 (s.e. 0.003). Adding all interactions between the woman’s and the man’s wealth deciles, and then 10,000 dummies for wealth-percentile interactions, gives 0.13 (s.e. 0.003). The paper spells out what this means: “even within a narrow wealth pairing (say, husband and wife both in the top percentile), the husband’s pre-marriage return to wealth is positively associated with the wife’s pre-marriage return.”

Controlling for the pre-marriage risky-asset share leaves the results “basically unchanged,” though the paper does not claim the two are unrelated. Risk tolerance is proxied by the share of financial wealth held in risky assets four years before marriage, grouped into six categories because “the distribution is highly skewed towards zero,” and entered as the 36 interaction dummies. As an alternative, the appendix tests sorting on individual Sharpe ratios computed from pre-marriage returns, with “qualitatively similar” results. Separately, the appendix documents that there is assortative mating on risk tolerance itself. The paper also notes that sorting on returns is not sorting on education: “we find similar levels of sorting on returns within narrowly defined educational groups.”

The authors acknowledge the limitation directly and test it. Ideally the pre-marriage average return “should capture the permanent component of the return and hence be based on a long panel to average out transitory deviations from the mean”; but the sample period “covers only 11 years of data,” and four years are dropped before marriage to avoid cohabitation contamination. Restricting to individuals with at least four years of pre-marriage data shrinks the sample to about a third, but “the findings remain remarkably similar.”

Post-marriage household returns load 79% on the higher pre-marriage return of the two spouses and 21% on the lower — so the low-return spouse still matters, which is what makes assortative mating on returns consequential. The specification regresses the post-marriage household return on the maximum and the minimum of the two spouses’ pre-marriage returns, with controls for the household’s initial risky share, both spouses’ pre-marriage net worth percentiles, age at marriage, year-of-marriage dummies and the same demographics as before. The weights are computed as q = β₁/(β₁+β₂) and 1−q = β₂/(β₁+β₂). The paper had set out the logic in advance: if management were always left to the spouse best equipped to do it, “there would be no incentive to sort also on returns to wealth,” because the low-return spouse would be irrelevant. Since β₂ ≠ 0, “assortative mating on returns would be justified by wealth preservation (or growth) strategies.” The appendix model of the allocation of wealth-management tasks, trading off efficiency against bargaining power, predicts q bounded between 1/2 and 1, varying with bargaining power, and closer to 1 for high-asset households.

No: “men receive a higher weight than would be warranted by their pre-marriage return alone.” Splitting by which spouse has the higher pre-marriage return, men’s weight is 1.05 when they are the higher-return spouse (versus the pooled 0.79) and 0.36 when they are the lower-return spouse (versus the pooled 0.21). “In couples where men have the highest return the wealth management is all in their hands. When women have the highest return their weight is 0.64, significantly below 1.” The gender difference in q is statistically significant when estimated on the full sample with an interaction term. The paper’s interpretation — offered as consistency rather than as a test — is that “this result is consistent with a male-biased gender norm that grants more power to men even when they have lower skills than women,” citing Ke (2021), Guiso and Zaccaria (2023) and Gu et al. (2021).

Among couples where both spouses come from the top decile of the wealth distribution at marriage, the weight on the low-return spouse “drops to zero so that wealth management is (statistically) all in the hands of the high-return spouse.” All other households reproduce the population pattern. The implication the paper draws: “this result implies that wealthier households earn on average higher returns to wealth, providing one rationale for the positive correlation between wealth and returns documented empirically by e.g. Fagereng et al. (2020) and Bach et al. (2020), and whose importance for wealth inequality dynamics is emphasized by Gabaix et al. (2016).” In other words, the intra-household allocation of wealth-management authority is offered as a micro-foundation for scale dependence, which in turn is one of the mechanisms through which persistent return heterogeneity generates extreme wealth concentration in the sense of Benhabib et al. (2019).

Both forms of assortative mating declined in Norway over the sample, alongside a decline in the Gini coefficient for wealth at marriage — but the paper describes the link as suggestive only. Computing the Spearman rank correlation year by year, assortative mating on wealth falls from around 0.24 to around 0.16, and on returns to wealth from 0.23 to 0.15. The Gini for wealth at marriage also declines. The authors are explicit about the strength of the inference: “The association between the indexes in the two panels, albeit crude, is suggestive of the potential importance of assortative mating for the time evolution of wealth inequality.” They add the broader point that “because mating patterns and wealth management arrangements are very likely to evolve over time (e.g. because social norms and relative gender skills change), time variation in assortative mating and wealth management allocation rules can be an independent and so far un-noticed causes of changes in wealth inequality over time.”

It provides moments, not a calibrated model, and lists four features its illustrative example ignores. The conclusion states the predictions “come from a simple example that ignores a number of realistic features.” First, the dynamics of marriage itself: assortative mating may occur over multiple dimensions, and marital dissolution matters because “since divorce destroys and partitions wealth, the degree of wealth concentration among stable couples is likely to be higher than in a population comprising single households due to family dissolution events.” Second, changes in wealth accumulation from inter vivos transfers or saving decisions. Third, the progressivity of the wealth tax, which “may distort optimal accumulation decisions.” Fourth, there is no idiosyncratic component to returns in the example. Properly accounting for these “would require the set up and the calibration of a full fledged life cycle model,” which the authors say “not only deserves but also requires a completely separate research effort as well as a much longer longitudinal dimension.” One data limitation is also flagged: offshore assets escaping the tax authority would understate concentration, though Alstadsæter et al. (2018) suggest accounting for hidden wealth raises the top 0.1% share by roughly one percentage point on average.