I start by comparing the KFE and CRE models. The CRE models utilized in this study take into account the correlation among income, number of children, and marital status and the unobserved household specific attributes. Appendix B contains the regression tables from all covariates of our baseline specification, except the year effects. We can see that both the KFE and the CRE models produce nearly identical results for
10 This is different from separating the data into nine subsamples along the unconditional distribution of
every variable. The baseline specification includes household wealth as the dependent variable, an age polynomial, income, married, divorced, children, extended family wealth, year dummies, and interaction terms for income and extended family wealth. Following the logic outlined in section 1.5, I interpret these results as an informal Hausman test, i.e., evidence in favor or using the correlated random effects model. The CRE model should thus provide more precise estimates and greater flexibility in model specification, which I utilize by including years of schooling and a dummy indicator of race in the full specification. All subsequent models use this full specification.
Figure 1.6 presents the first result of my analysis: the marginal effects of extended family wealth on household wealth accumulation across the entire conditional distribution. The horizontal axis represents deciles. The vertical axis represents the marginal effect of a one-dollar increase in the sum of extended family wealth on own-wealth accumulation, holding everything else constant. The solid purple line represents the marginal effect of extended family wealth on own-household wealth accumulation for Black households across the wealth distribution, and the solid blue line represents the same effect of non- Black households. The 95% confidence intervals for each group are shaded in their respective color. A useful way to interpret these results is to consider a hypothetical household with attributes such that, conditional on the model, we would expect it to lie at exactly the ith decile of the household wealth distribution. To the extent that our model can take the covariates of our sample households and accurately predict in which decile of the household wealth distribution they belong, we can think of these results as the marginal effects of extended family wealth on own-household wealth for the households located at each decile of the wealth distribution.
Figure 1.6. CRE Extended Family Wealth Full Specification
Two trends stand out in Figure 1.6. First, there are no statistically significant differences between the effects on Black and non-Black households. We see this because the confidence band for the Black household effect encompasses the non-Black household effect. Thus, we cannot reject the null that both effects are equal. Second, we see that hypothetical households, Black and non-Black, which we would expect to be located at the .8 or .9 deciles wealth distribution, given the model, receive statistically and economically significant benefits from the total accumulated wealth of their extended family network. Households at the ninth decile tend to, on average, accumulate an extra $92 worth of assets whenever one of their relatives accumulates an extra $1,000 of wealth. The effect is smaller for households in the .8 decile, and negligibly small for households located in any other decile. This effect is nontrivial, especially for households located at the .9 decile of the wealth distribution that on average hold several hundred thousand dollars’ worth of assets
and belong to extended family networks with over a million dollars’ worth of assets. This evidence indicates that belonging to the upper echelons of society correlate with a rising tide dynamic that lifts all members of the extended family network at the same time. However, the lack of a racial difference is surprising and warrants further attention, to which this study now turns. The graphs of the marginal effects of all covariates from this model can be found in Appendix B.
Due to the inclusion of a race dummy as well as race interaction terms, the interpretation of these results is complicated. The complication arises from the fact that the race dummy and interaction terms create two separate conditional distributions of household wealth, one for Black households and one from non-Black households. For example, consider a household with the following attributes: age 40, income $100,000, married, two children, and an extended family network with 2 million dollars in net worth. To keep the calculations simple, I ignore the household specific intercepts and use a - $100,000 constant. My estimated quantile regression function for the ninth decile predicts that this household would hold approximately $435,000 worth of assets if it was not a Black household, but only $125,000 worth of assets if it was a Black household. If one uncle accumulated an extra $100,000 worth of assets from a good investment, the non-Black household would accumulate an extra $10,000 dollars, bringing its wealth up to $445,000. The Black household, on the other hand, would accumulate only an extra $7,000, bringing its wealth up to $132,00011. This difference in marginal effects, $10,000 versus $7,000, is relatively small, but the point is that when we consider the marginal effects of additional wealth accumulation using my model, we are really looking at how the differential
11 I round the coefficients and ignore statistical significant in these calculations in order to keep the example
marginal effects impact two conditional distributions: households at the ninth decile of the non-Black conditional distribution, and households at the ninth decile of the Black conditional distribution. This point is important because the counterfactual of my regressions compares two otherwise identical households, but one is Black and the other is non-Black. Recall that in the unconditional ninth decile of the wealth distribution shown in Figure 1.5, non-Black households hold $469,000 worth of assets and Black households tend to hold $101,000 worth of assets. Although this example is contrived to produce illustrative results, it demonstrates the predictive usefulness of my model. Lastly, since Black households at the ninth decile tend to have about the same wealth as non-Black households at the sixth decile, it makes sense to compare these two groups rather than Black households at the ninth decile to non-Black households at the ninth decile.
In order to further explore the underlying trends in the data, I first separate the sample by generation, running separate regressions on each sample by capturing possible cohort differences in wealth accumulation behavior. I then decompose wealth into financial wealth, nonfinancial wealth, and debt in order to focus on financial wealth accumulation. I focus on financial wealth because it is the biggest difference between the portfolios of Black and non-Black households. Following Chiteji and Hamilton (2002), I define nonfinancial wealth as the net value of housing, real estate, and vehicles. Financial wealth is defined as the net value of stocks, bonds, other financial instruments, cash accounts, and businesses. These are assets that directly generate or provide economic resources, whereas nonfinancial assets provide a stream of consumption (Chiteji & Hamilton, 2002; Oliver & Shapiro, 1997). Household wealth is decomposed in this manner to test the hypothesis that financial wealth accumulation follows different dynamics than nonfinancial wealth
accumulation. Figure 1.7 shows the marginal effects of changes in total extended family wealth on household-financial wealth accumulation across the financial wealth distribution for second-generation households. The associated graphs of all covariates can be found in Appendix B.
In Figure 1.7, we see that Black and non-Black financial wealth accumulation dynamics diverge among second-generation households. The purple line is the marginal effect of one additional dollar of extended family wealth on own-household wealth for second-generation Black households. Non-Black households in the .8 and .9 deciles tend to experience statistically and economically significant gains in financial wealth whenever their extended family accumulates additional wealth. The marginal effect of an additional dollar of network wealth accumulation on financial wealth accumulation for second- generation Black households is negligible and insignificant across all deciles. Importantly, ninth decile black households behave similarly to 6th decile non-Black households: both groups reap little to no benefit from the addition wealth accumulation of their extended family networks. As I noted earlier, it makes sense to compare ninth decile blacks to sixth decile non-Blacks because they are in similar wealth positions. This similarity is indicative of a strong threshold effect: only high wealth households benefit from the additional wealth accumulation of their extended family networks and race may not be an important factor, on its own. On top of this threshold effect, however, there may also be unique racial obstacles preventing Black households from accumulating wealth, further exacerbating racial wealth inequality.
Third-generation households tend to have much weaker ties to their extended family network and experience economically and statistically insignificant extended
Figure 1.7. CRE Second-Generation Financial Wealth – Full Specification
family wealth effects on both financial and nonfinancial wealth accumulation. Third- generation regression figures are located in Appendix B. Since I am controlling for age and age squared, I interpret these generational differences as cohort effects. Those households whose heads were born in the 1980s and 1990s tend to experience much weaker wealth linkages with their extended family than second-generation households, born in the 1960s and 70s.
Figure 1.8 respecifies my baseline model, breaking up second-generation extended family network wealth into familial categories: total sibling wealth and total offspring wealth. Offspring wealth has a large and significant effect on second-generation own- household wealth from the fourth decile to the ninth decile. Sibling wealth has a smaller, but still economically and statistically significant effect on the top two deciles. This result contributes to the empirical story of household wealth accumulation.
Figure 1.8 Second-Generation Marginal Effects of Sibling Wealth and Offspring Wealth 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Second generation: Marginal effect of sibling wealth
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
In a similar vein, I look at how sibling wealth, cousin wealth, and parent wealth contribute to the wealth accumulation of third-generation households. Uncle wealth is omitted due to collinearity. Figure 1.9 shows that sibling wealth is the most important of the extended family wealth components for third-generation households. Whereas cousin wealth has a moderate effect at the ninth decile, and parent wealth has no statistically significant effect, sibling wealth has a moderately sized, steady, and significant effect on own-household accumulation from the fifth decile to the ninth decile.
Initially, it seemed like there were no differences between the wealth accumulation dynamics of Black and non-Black households, but looking at financial wealth and generational differences, new patterns appear. Second-generation Black households do not accumulate additional financial wealth when their extended family becomes wealthier. Even the wealthiest Black households do not benefit, but wealthy non-Blacks benefit strongly. Third-generation households have economically small network effects when it comes to financial wealth, less than one fifth the size of the second-generation effects, with no appreciable racial difference. Finally, second-generation household wealth is connected to the wealth of their offspring, whereas third-generation households are connected to the wealth accumulation of their siblings.