Personal Savings from Top Incomes and Household Wealth Accumulation in the United States

Abstract:

This article explores the determinants and distribution of household wealth. Looking at U.S. data since 1980, it finds convincing evidence that top incomes were saved at high rates and contributed to the steady increase in the household wealth–income ratio. First, I rule out counterclaims regarding the role of housing and real estate prices finding little evidence of their influence on the trends and magnitudes of household net worth relative to disposable income. With savings as the remaining explanation, I present an accounting decomposition formula that captures savings rates for any reference group using the dynamics of intergroup accumulation rates. This methodology is applied to data from national accounts, balance sheets, and income distribution statistics in order to compute saving rates for the top 1 percent of households in the U.S. income distribution. The estimates also support the idea that top income earners have outsaved other households, thereby capturing an increasing share of wealth.

Keywords:

• national accounts
• saving rates
• top incomes
• wealth

ACKNOWLEDGMENTS

The author wishes to thank Duncan Foley, Tom Michl, Mark Setterfield, and Lance Taylor for helpful comments. Suggestions by two anonymous referees were very useful.

Notes

Their capitalization technique is briefly summarized in later sections.

As part of a symposium on Piketty’s findings, see also the comments on Taylor—in particular by G.C. Harcourt—which appear in vol. 43, no. 3 of this journal.

See, for example, Rognlie (2015) and Chirinko (2008), who emphasize elasticities ranging at best between 0.4 and 0.6, especially in net rather than gross terms.

See the note Housing Services in the National Economic Accounts: Bureau of Economic Analysis, www.bea.gov/papers/pdf/RIPfactsheet.pdf.

Since assets are tradable on the market, the price of housing capital is equal to rent/rate of return.

For a discussion of these measurement issues, see Guidolin and La Jeunesse (2007) and Baker and Rosnick (2011).

The Distribution of Household Income and Federal Taxes (2010), www.cbo.gov/publication/44604/.

This calculation is done under the assumption that the income generated by housing output is equal parts imputed and actual payments to landlords. Further, I attribute a 50 percent cost of maintenance and depreciation to this revised computation. Thus, 75 percent of housing services are removed.

In the sense of skyscrapers and new constructions.

Since marginal pricing is used, labor earns its marginal product. If skills are scarce, then a skilled segment would earn greater income. There could in principle be other factors that endogenize this marginal premium such as wealth endowment at birth, network effects, or geography.

To deliver these estimates, Saez and Zucman use the capitalization technique, that is, using capital income to infer wealth, under the assumption that capitalization factors are the same for everyone. Suppose a person reports $10 worth of capital income on a bond with a 10 percent rate of return, then it can be inferred that their stock of wealth in that bond itself has a capitalized value of $100. By matching capital income reported on tax returns across households with the rate of return for every asset class, the authors arrive at estimates of wealth inequality.

Simple class-oriented models of wealth distribution tend to prove more accurate than the results of workhorse dynamic models with degrees of heterogeneity and explanatory inputs such as idiosyncratic risk. See the survey in De Nardi (2015).

Wealth levels and shares are computed by capitalizing income from assets with rates observed on the flow of funds.

Even synthetic estimates remain valid within limited social mobility frameworks as long as people marry within their class of income and wealth, for example, doctors marrying lawyers.

The section at the end of this study discusses the likelihood of such possibilities.

Among other differences, Saez and Zucman (2014) do their own application for the top fractiles of the wealth distribution.

There is also oversampling correlated with wealth portfolios. See the discussion in Kennickell (1999) and Bricker et al. (2015).

Saez and Zucman (2014) use this concept in their capitalization technique. By their calculations, returns equalize across asset classes even if there is noise at the individual level. Like the CBO study, they use administrative income tax data, which prove helpful for my data in terms of asset returns.

In the two-period case of Carroll’s model, wealth directly enters the utility function. The agent decides between consuming in the first period (c_t) and deriving utility from end of period wealth (w_t+1). End of period wealth enters the utility function in the Stone–Geary form while consumption is the standard Constant Relative Risk Aversion (CRRA) form. $$c_t=min[c_t^{eq},w_t]$$ imposes the condition that agents do not die in debt. If ρ > α then wealth is a luxury good and γ> 0, with each symbol respectively being the consumption, wealth, and Stone–Geary parameter.

Additional information

Notes on contributors

Rishabh Kumar

Rishabh Kumar is a Ph.D. candidate in the Department of Economics at the New School for Social Research, New York.