Measuring income and wealth at the top using administrative and survey data.

Bricker, Jesse ; Henriques, Alice ; Krimmel, Jacob 等

II.C. Gross Capitalization for Fixed-Interest Assets

Much of the difference between our preferred estimates and the capitalized income top shares can be reconciled by trivial changes to the data, meaning whether or not to calibrate to the FA aggregates or whether to count the top 1 percent versus the top 1.3 percent of families. The remaining difference in top wealth shares is more about trends than levels, given that both the top 1 percent and top 0.1 percent wealth shares are rising more rapidly in the gross capitalization estimates, relative to even our constrained SCF-based estimates. It turns out that the gross capitalization implied rate of return on just one asset class (fixed income) is responsible for all the differential growth in wealth concentration at the very top. That is, when (more realistic) alternative rates of return are used in gross capitalization, the growth at the very top looks very much like the SCF-based top wealth share growth rates shown above.

The analysis of the biased gross capitalization factor begins with the actual (SCF) versus derived (gross capitalization) portfolios of the top 0.1 percent, as shown in figure 8. Assets of the top wealth holders are broken down into four broad categories: housing, pensions, equity plus business, and fixed income. The shares of the first three are very similar, and the share of fixed-interest assets is also similar through 2001 or so. Indeed, all the differential growth in wealth of the top 0.1 percent occurs in the fixed-interest asset class, consisting mostly of bonds, certificates of deposit, call accounts, money market accounts, and other savings instruments. As of 2013, the gross capitalization approach implied that nearly half the assets owned by the top 0.1 percent were in the fixed-interest class.

Is this dramatic shift in portfolio composition plausible, or just an artifact of the gross capitalization approach implemented by Saez and Zucman (2016)? To answer this, we consider the implied gross capitalization factor underlying these estimates, and compare it with the implied capitalization factors if one instead uses a market rate of interest or an alternative based on estate tax filings. The result of these comparisons is shown in figure 9. The current low-interest-rate environment has led to increases in capitalization factors based on 10-year Treasury yields, the Moody's Aaa bond yield, or the ratio of prior year interest income to estate tax fixed-interest assets, any of which may be on the high end of plausible values. However, the implied gross capitalization factor solved for using the ratio of FA assets to administrative tax data interest income is much higher, and has clearly reached implausible levels. (47) Based on this estimate, for every $1 in observed interest income, gross capitalization is currently generating nearly $100 in wealth. (48)

[FIGURE 8 OMITTED]

[FIGURE 9 OMITTED]

Figure 10 makes the point clearly that there is basically no remaining unexplained difference in top 0.1 percent wealth shares when the constrained SCF is compared with gross capitalization when even a slightly more reasonable rate of return is used. Lowering the implied capitalization factor at the top to be consistent with (the still very conservative) 10-year Treasury rate, the top 0.1 percent wealth share lies almost completely within the confidence interval for the constrained SCF estimates. The reestimated top 0.1 percent wealth share under the alternative gross capitalization parameters falls to just under 19 percent in the most recent period, which is still well above our preferred estimate of about 15 percent, but these differences are completely explained by the other constraints imposed above.

What is driving the implausible capitalization factors in the Saez and Zucman (2016) estimates? Our discussion of data and methods in section I indicates that a few things can go awry when using the ratio of the estimated FA asset value to measured income flows. The FA asset totals include holdings by nonprofits, while the taxable income flow does not, so the gross capitalization factor is biased up. The household sector of the FA tries to separate out direct holdings from pension and other tax-preferred asset holdings, but any misclassification toward direct holdings will also bias up the numerator of the gross capitalization ratio. The household sector of the FA is also a residual claimant on asset holdings, so any sectoral misallocation of a given asset holding toward households will introduce bias. It is also likely that in the current low-interest environment, the much lower interest earnings on checking and savings deposits are going unmeasured in the tax data, and to the extent that these are more relevant for families outside the top 1 percent, their share of fixed-interest assets is being allocated to the top wealth families that have (quantitatively observable) interest. Ultimately, however, given the available data, we cannot point to any one explanation with certainty. (49)

[FIGURE 10 OMITTED]

III. Top Income Shares in Administrative and Survey Data

Income concentration and wealth concentration are both contentious issues, and many see the two measures as strongly correlated. Everyone seems to know that the rich are getting richer, whether we categorize them as rich by their income or their wealth. In some ways income concentration is a more straightforward measure, because we can look directly at administrative data to gauge how the top income shares are evolving over time, rather than (as in gross capitalization for wealth shares) requiring additional assumptions about the relationship between income and the value of the assets that are generating this income. However, in another sense, the concept of income itself has changed in fairly dramatic ways during the period when top income shares have been rising, and we will show that these conceptual changes are having a first-order impact on estimated top shares.

In this section we present our preferred estimates of the top income shares, and, as with the top wealth shares, we show how these preferred estimates compare with and contrast to both the published SCF and the administrative tax-based estimates. Our preferred top income share estimate is constructed by starting with SCF income measures, then adding components of NIPA personal income that are not measured in the SCF. The preferred measure shows slower growth in income concentration than the estimates by Piketty and Saez (2003), based on administrative tax data; but unlike the top wealth shares, our preferred top income shares are also (modestly) lower and have been rising more slowly than published SCF estimates. We investigate the source of divergence in top income growth rates and levels by once again constraining the SCF to conceptually match the administrative tax-based estimates. Using this approach, we are able to confirm that the differentials in income concentration are not (at least on a first approximation) attributable to a lack of population coverage at the very top or to survey underreporting in the SCF.

III.A. Preferred Estimates of the Top Income Shares

In all the estimates discussed here, the top income shares in the United States are high and have been increasing over time. The top panel of figure 2 shows the estimated share of income received by the top 1 percent for the period 1988-2012 based on three different measures, and the bottom panel of figure 2 shows the same for the top 0.1 percent income shares. In general, the estimated top income shares based on administrative tax data from Piketty and Saez (2003) are higher and have been rising more rapidly than the top income shares in published SCF estimates, and are also higher than those based on our preferred measure.

The differences between the various estimated top income shares are, as with wealth shares, first-order. For 2012, our preferred estimate of the top 1 percent income share is just under 18 percent, while the administrative tax-based estimate is nearly 23 percent. The gap is proportionally larger for the top 0.1 percent, and both gaps have been increasing over time, though, as with wealth, much of the increase in the top 1 percent income share can be accounted for by the top 0.1 percent income share. That is, the substantial income gains are occurring within the top 1 percent and not just for the 1 percent as a whole.

Our preferred measure for top income shares begins with the published SCF Bulletin concept and estimates. As with top wealth shares, the first adjustment on the income side is needed because the Forbes 400 is excluded from the SCF sample. Although the Forbes 400 account for about 3 percent of total household sector net worth, the relationship between income and wealth is such that the Forbes 400 account for a much smaller fraction of income, and thus adding them generally increases the average incomes of the top groups by a more modest amount. (50) Thus, the estimated shares of income received by the top income groups are pushed up, but the effects are much more muted than for the top wealth shares.

The more substantial adjustments are to the SCF income concepts, and involve adding the in-kind transfers included in NIPA PI but not measured in the SCF survey. In particular, we add the value of employer-provided health insurance; the value of in-kind government transfers such as SNAP; and the value of Medicaid, Medicare, and other government health care programs. Together, these incomes amounted to about 7 percent of NIPA PI in 1988, but had roughly doubled as a share of PI by 2012. This increasing share of total PI interacts with the casual observation that these forms of income are much less concentrated than the measured incomes, and this pulls down the preferred top shares every year, but disproportionally more in recent years. (51) This is seen most clearly in the gaps between the published SCF income measure and our preferred measure; the modest but rising Forbes 400 income share is pulling the two together, but the addition of in-kind incomes is larger and, on net, pushing the two apart.

III.B. Reconciling the Income Concentration Estimates

We approach the reconciliation of the income shares from the same basic starting point as we used for wealth shares. If the SCF sampling strategy does a good job capturing the top end of the income distribution and SCF respondents do a good job reporting their incomes, what is causing the substantial divergence between the estimated top income shares in the SCF-based preferred and administrative tax-based measures? Again, we constrain the SCF to be conceptually and empirically similar to the tax-based measures, and we show that most of the divergence is eliminated. In particular, when we measure the top income shares after constraining the SCF income concept to match the tax-based concept and we adjust the number of families in the top fractile to be consistent with the tax unit counts, most of the level differences are eliminated, or are at least brought within the range of SCF statistical confidence.

The effects of constraining the SCF-based preferred top income share estimates to be conceptually and empirically equivalent to the administrative tax-based estimates are shown in the top panel of figure 11 for the top I percent, and in the bottom panel of figure 11 for the top 0.1 percent. The first adjustment, which involves moving from the "Preferred" line to the "Market income, families" line, is based on restricting the SCF income concept to match what is available in the tax data (see table 2). This basically involves removing cash transfers--most notably Social Security benefits, but also other cash transfers--from the SCF income concept. Because these forms of income are disproportionately received by families in the bottom 99 percent by income, removing these forms shifts the concentration numbers up. And because these forms are becoming increasingly important, their effects have been larger in recent years. The quantitative effect of moving from the SCF Bulletin income measure to the more restrictive market income measure is to move the income concentration estimates further away from the preferred income measure, and for the same reasons.

[FIGURE 11 OMITTED]

The second reconciliation, as with the wealth shares, also uses the constrained market income concept, and further involves redefining how many families the top fractiles represent. Again, there are 30 percent more tax units than families in 2012, and thus the top 1 percent on a tax unit basis represents about 1.6 million families instead of the 1.2 million families in the top 1 percent using the SCF and preferred distributional measures. Adding the extra 400,000 families to the top 1 percent, and the extra 40,000 families to the top 0.1 percent, increases the top share estimates in a predictable and sizable way, the lines labeled "Market income, tax units." The remaining differences between the top income shares in the constrained SCF and administrative tax data are mostly about volatility, and not levels per se. Further, the width of the confidence intervals shows how income variability and sampling interact, especially at the very top, to generate a wide confidence interval for estimated top shares. (52) Indeed, the point estimates for the constrained SCF top 1 percent income shares are actually above the administrative tax-based estimates, and are basically the same for the top 0.1 percent.

III.C. Even More Comprehensive Incomes?

The steps taken to reconcile our preferred top income shares with the administrative tax-based estimates are suggestive of a broader question. What else is missing from an even more comprehensive income measure, and what might be the result of incorporating these other missing pieces into the analysis of top income shares? Figure 12 reinforces the fact that the more comprehensive income measures in our preferred top income shares diverge from the narrow administrative tax-based measures and the SCF Bulletin measure, and that even our preferred measure is not complete. Even though the three income measures in the micro data all include something the PI measure does not--realized capital gains--even our most comprehensive income estimate is still less than the NIPA total.

[FIGURE 12 OMITTED]

The remaining divergence between NIPA PI and our preferred income measure involves a mix of imputations, known and unknown underreporting, and unreconciled conceptual discrepancies. It might be feasible in principle to produce distributional estimates for incomes, such as imputed rent on owner-occupied housing or the value of in-kind financial services, using a data set like the SCF. One could also imagine rescaling the SCF-reported incomes in categories for known underreporting for, say, a nroprietor's income, but this underreporting is also known to have a distributional component (small proprietors are worse when it comes to underreporting) that would need to be considered. Some adjustments of tax basis versus economic profit and rent have also been incorporated into the NIPA, and one would need to work through them in order to align the comprehensive PI measure. Although these various adjustments might affect the estimated top shares, it is not clear in what direction. What is clear is that further adjustments such as these should be done very carefully, and that simply scaling the available data to match the aggregates could bias the final answer.

IV. Concluding Remarks

Rising top wealth and income shares are often cited as a call to action by those who believe that government can and should do more about inequality vis-a-vis taxation, spending, regulation, and other market interventions. Rising inequality raises obvious normative concerns, and there is a growing belief that recent macroeconomic instability and slow growth may be additional symptoms of the same underlying phenomenon. (53) Economists disagree about the fundamental causes of rising inequality, as some argue that the trends are associated with free market prices adjusting to equate supply and demand, while at the other extreme some argue that the influence wielded by those who are already wealthy improves their market shares by changing the rules of the game. (54)

The preferred estimates for the top wealth and income shares presented here reflect what we think can be gleaned from the best available data sources, including administrative tax data, the SCF, and macro aggregates. The estimates agree with the widely held view that inequality, at least as reflected in the top wealth and income shares, has been rising in recent decades. However, the levels and trends in our preferred top share estimates are more muted than those in recent studies that are based directly on administrative income tax data (Piketty and Saez 2003; Saez and Zucman 2016), but the levels and trends for the top wealth shares are a bit larger than the estimates based on estate tax data (Kopczuk and Saez 2004).

Although the SCF makes it possible to inform and improve on direct estimates of the top wealth and income shares derived from administrative tax data, the survey is still far from capturing comprehensive wealth and income measures. The SCF adds some government transfers to the tax-oriented income measures, but it still misses employer-provided benefits, government in-kind (especially health care) transfers, and other forms of income that are both substantial and growing over time. There are also direct analogs in shortcomings in the wealth measures; for example, the value of most families' key retirement asset--Social Security--is not measured as part of household net worth. (55) The effect of these omissions is important for understanding the top shares, and even more important when looking at inequality across the entire distribution.

The reconciliations made here cannot be extended back in time before the development of the modern SCF household survey, but the specific issues raised draw attention to how changes in government policies and market practices are affecting the measurement of top shares over time. In particular, although the administrative tax data make it possible to show that the top share families are getting increasingly large slices of a particular pie, the pie's overall size being measured in these data is shrinking relative to more economically meaningful concepts of wealth and income. The increasingly unmeasured part of the pie is not disappearing, but it is evolving. It may be difficult or even impossible to allocate the missing pieces in the very long historical series; thus, any very long-term trends should also be viewed with an eye toward the conceptual divergence being driven by evolving government policy and economic institutions.

Building on the theme of conceptual measurement, the reconciliation of top shares presented here speaks directly to the underlying impetus for--and possible approaches to--public policy toward wealth and income distribution. The failure to properly measure the effects of government policies and market practices that disproportionately benefit families in the middle and bottom of the wealth or income distribution leads directly to an overstatement of the top wealth and income shares. Policies and practices such as social insurance and government investment in human capital generate real benefits, and the debate is thus properly focused on the distribution of these benefits. If we measure only the costs of such policies and practices, without measuring the benefits, it becomes more difficult to make the case in favor of such policies in debates.

ACKNOWLEDGMENTS We would like to thank our colleagues on the Survey of Consumer Finances project who made this research possible: Lisa Dettling, Sebastian Devlin-Foltz, Joanne Hsu, Kevin B. Moore, Sarah Pack, Jeffrey P. Thompson, and Richard Windle. For input and comments on this and earlier versions of this paper, we also thank our editor, James Stock; and our discussants, Katharine Abraham and Wojciech Kopczuk; as well as Mariacristina De Nardi, Diana Hancock, Arthur Kennickell, Jose-Victor Rios-Rull, Emmanuel Saez, Gabriel Zucman, and the seminar participants at the Brookings Panel on Economic Activity, the Federal Reserve Board, the Bank of England, the Bank of Spain, and the Household Finance and Consumption Network's meeting at the European Central Bank. Jesse Bricker thanks Olympia Bover and the Bank of Spain for hospitality at the early stages of this work. Finally, we are grateful to Michael Parisi for providing unpublished tabulations of Statistics of Income data, and to Barry W. Johnson and the Statistics of Income staff for contributions to the Survey of Consumer Finances sample design. The analysis and conclusions set forth in this paper are those of the authors alone, and do not indicate concurrence by other members of the research staff or the Board of Governors of the Federal Reserve System.

References

Armour, Philip, Richard V. Burkhauser, and Jeff Larrimore. 2014. "Levels and Trends in U.S. Income and Its Distribution: A Crosswalk from Market Income towards a Comprehensive Haig-Simons Income Approach." Southern Economic Journal 81, no. 2: 271-93.

Atkinson, Anthony B., Thomas Piketty, and Emmanuel Saez. 2011. "Top Incomes in the Long Run of History." Journal of Economic Literature 49, no. 1: 3-71.

Attanasio, Orazio, Erik Hurst, and Luigi Pistaferri. 2015. "The Evolution of Income, Consumption, and Leisure Inequality in the United States, 1980-2010." In Improving the Measurement of Consumer Expenditures, edited by Christopher D. Carroll, Thomas Crossley, and John Sabelhaus. University of Chicago Press.

Auerbach, Alan J., and Kevin Hassett. 2015. "Capital Taxation in the Twenty-First Century." American Economic Review 105, no. 5: 38-42.

Bordo, Michael D., and Christopher M. Meissner. 2012. "Does Inequality Lead to a Financial Crisis?" Working Paper no. 17896. Cambridge, Mass.: National Bureau of Economic Research.

Bricker, Jesse, Lisa J. Dettling, Alice Henriques, Joanne W. Hsu, Kevin B. Moore, John Sabelhaus, Jeffrey Thompson, and Richard A. Windle. 2014. "Changes in U.S. Family Finances from 2010 to 2013: Evidence from the Survey of Consumer Finances." Federal Resen'e Bulletin 100, no. 4.

Bricker, Jesse, Alice Henriques, Jacob Krimmel, and John Sabelhaus. 2015. "Measuring Income and Wealth at the Top Using Administrative and Survey Data." Finance and Economics Discussion Series, no. 2015-030. Washington: Board of Governors of the Federal Reserve System.

Brown, Meta, Andrew Haughwout, Donghoon Lee, and Wilbert van der Klaauw. 2013. "Do We Know What We Owe? A Comparison of Borrower- and Lender-Reported Consumer Debt." Staff Report no. 523. Federal Reserve Bank of New York.

Burkhauser, Richard V., Shuaizhang Feng, Stephen P. Jenkins, and Jeff Larrimore. 2012. "Recent Trends in Top Income Shares in the United States: Reconciling Estimates from March CPS and IRS Tax Return Data." Review of Economics and Statistics 94, no. 2: 371-88.

Burkhauser, Richard V., Jeff Larrimore, and Kosali I. Simon. 2012. "A 'Second Opinion' on the Economic Health of the American Middle Class." National Tax Journal 65, no. 1: 7-32.

Congressional Budget Office. 2014. "The Distribution of Household Income and Federal Taxes, 2011." Washington, https://www.cbo.gov/publication/49440

Debacker, Jason, Bradley Heim, Vasia Panousi, Shanthi Ramnath, and Ivan Vidangos. 2013. "Rising Inequality: Transitory or Persistent? New Evidence from a Panel of U.S. Tax Returns." Brookings Papers on Economic Activity, Spring: 67-122.

Dettling, Lisa J., Sebastian J. Devlin-Foltz, Jacob Krimmel, Sarah J. Pack, and Jeffrey P. Thompson. 2015. "Comparing Micro and Macro Sources for Household

Accounts in the United States: Evidence from the Survey of Consumer Finances." Finance and Economics Discussion Series, no. 2015-086. Washington: Board of Governors of the Federal Reserve System.

Devlin-Foltz, Sebastian, Alice M. Henriques, and John Sabelhaus. 2016. "Is the U.S. Retirement System Contributing to Rising Wealth Inequality?" Russell Sage Foundation Journal of the Social Sciences, forthcoming.

Dungan, Adrian. 2015. "Individual Income Tax Shares, 2012." Statistics of Income Bulletin (Internal Revenue Service, Statistics of Income Division), Spring.

Fagereng, Andreas, Luigi Guiso, Davide Malacrino, and Luigi Pistaferri. 2016. "Heterogeneity in Returns to Wealth and the Measurement of Wealth Inequality." American Economic Review 106, no. 5: 651-55.

Feldstein, Martin. 2015. "The Uncounted Trillions in the Inequality Debate." Wall Street Journal December 13.

Greenwood, Daphne. 1983. "An Estimation of U.S. Family Wealth and Its Distribution from Microdata, 1973." Review of Income and Wealth 29, no. 1: 23-44.

Guvenen, Fatih, Greg Kaplan, and Jae Song. 2014. "How Risky Are Recessions for Top EarnersT American Economic Review 104, no. 5: 148-53.

Harwood, John. 2015. "10 Questions with Bemie Sanders." CNBC, May 26. www. cnbc.com/2015/05/26/10-questions-with-bemie-sanders.html

Henriques, Alice M. 2013. "Are Homeowners in Denial about Their House Values? Comparing Owner Perceptions with Transaction-Based Indexes." Finance and Economics Discussion Series, no. 2013-79. Washington: Board of Governors of the Federal Reserve System.

Henriques, Alice M., and Joanne W. Hsu. 2014. "Analysis of Wealth Using Micro- and Macrodata: A Comparison of the Survey of Consumer Finances and Flow of Funds Accounts." In Measuring Economic Sustainability and Progress, edited by Dale W. Jorgenson, J. Steven Landefeld, and Paul Schreyer. University of Chicago Press.

Johnson, Barry W., and Kevin Moore. 2008. "Differences in Income Estimates Derived from Survey and Tax Data." In Proceedings of the Joint Statistical Meetings, Survey Research Methods Section. Alexandria: American Statistical Association.

Jones, Charles I. 2015. "Pareto and Piketty: The Macroeconomics of Top Income and Wealth Inequality." Journal of Economic Perspectives 29, no. 1: 29-46.

Kaplan, Steven N., and Joshua Rauh. 2010. "Wall Street and Main Street: What Contributes to the Rise in the Highest Incomes?" Review of Financial Studies 23, no. 3: 1004-50.

--. 2013. "It's the Market: The Broad-Based Rise in the Return to Top Talent." Journal of Economic Perspectives 27, no. 3: 35-56.

Kennickell. Arthur B., and R. Louise Woodburn. 1999. "Consistent Weight Design for the 1989, 1992, and 1995 SCFs, and the Distribution of Wealth." Review of Income and Wealth 45, no. 2: 193-215.

Kopczuk, Wojciech. 2015a. "Recent Evolution of Income and Wealth Inequality: Comments on Piketty's Capital in the Twenty-First Century Tax Law Review 68, no. 3: 545-62.

--. 2015b. "What Do We Know about the Evolution of Top Wealth Shares in the United States?" Journal of Economic Perspectives 29, no. 1: 47-66.

Kopczuk, Wojciech, and Emmanuel Saez. 2004. "Top Wealth Shares in the United States, 1916-2000: Evidence from Estate Tax Returns." National Tax Journal 57, no. 2: 445-87.

Kuznets, Simon. 1953. Shares of Upper Income Groups in Income and Savings. New York: National Bureau of Economic Research.

Mankiw, N. Gregory. 2015. "Yes, r > g. So What?" American Economic Review 105, no. 5: 43-47.

McKeever, Brice S. 2015. "The Nonprofit Sector in Brief 2015: Public Charities, Giving, and Volunteering." Washington: Urban Institute, Center on Nonprofits and Philanthropy.

O'Muircheartaigh, Colm, Stephanie Eckman, and Charlene Weiss. 2002. "Traditional and Enhanced Field Listing for Probability Sampling." In Proceedings of the Joint Statistical Meetings, Survey Research Methods Section. Alexandria: American Statistical Association.

Pareto, Vilfredo. 1896. Cours d'Economie Politique. Geneva: Droz.

Parker, Jonathan A., and Annette Vissing-Jorgensen. 2010. "The Increase in Income Cyclicality of High-Income Households and Its Relation to the Rise in Top Income Shares." Brookings Papers on Economic Activity, Fall: 1-55.

Piketty, Thomas. 2014. Capital in the Twenty-First Century. Belknap Press.

--. 2015. "About Capital in the Twenty-First Century." American Economic Review 105, no. 5: 48-53.

Piketty, Thomas, and Emmanuel Saez. 2003. "Income Inequality in the United States, 1913-1998." Quarterly Journal of Economics 118, no. 1: 1-39.

Rajan, Raghuram G. 2010. Fault Lines: How Hidden Fractures Still Threaten the World Economy. Princeton University Press.

Sabelhaus, John, David Johnson, Stephen Ash, David Swanson, Thesia I. Gamer, John Greenlees, and Steve Henderson. 2015. "Is the Consumer Expenditure Survey Representative by Income?" In Improving the Measurement of Consumer Expenditures, edited by Christopher D. Carroll, Thomas Crossley, and John Sabelhaus. University of Chicago Press.

Saez, Emmanuel, and Gabriel Zucman. 2016. "Wealth Inequality in the United States since 1913: Evidence from Capitalized Income Tax Data." Quarterly Journal of Economics 131, no. 2: 519-78.

Smeeding, Timothy M., and Jeffrey P. Thompson. 2011. "Recent Trends in Income Inequality: Labor, Wealth and More Complete Measures of Income." In Who Loses in the Downturn? Economic Crisis, Employment and Income Distribution, edited by Herwig Immervoll, Andreas Peichl, and Konstantinos Tatsiramos. Bingley, U.K.: Emerald Group.

Stiglitz, Joseph E. 2012. The Price of Inequality: How Today's Divided Society Endangers Our Future. New York: W. W. Norton.

Vermeulen, Philip. 2014. "How Fat Is the Top Tail of the Wealth Distribution? Working Paper no. 1692. European Central Bank.

Weil, David N. 2015. "Capital and Wealth in the Twenty-First Century." American Economic Review 105, no. 5: 34-37.

Wolfers, Justin. 2015. "The Gains from the Economic Recovery Are Still Limited to the Top One Percent." New York Times, January 28.

Yellen, Janet L. 2014. "Perspectives on Inequality and Opportunity from the Survey of Consumer Finances." Speech given at the Conference on Economic Opportunity and Inequality, Federal Reserve Bank of Boston, Boston, October 17.

Comments and Discussion

COMMENT BY

KATHARINE G. ABRAHAM Evidence that income and wealth have become increasingly concentrated in recent years--more specifically, evidence that a growing share of income and wealth is controlled by households in the top 1 percent or top 0.1 percent of these distributions--has attracted enormous scholarly and popular attention. Although there is general agreement that both income and wealth have indeed become more concentrated, different data sources tell somewhat different stories about the magnitude of these changes. Most notably, estimates of the share of wealth controlled by households at the very top of the distribution based on income tax records (Saez and Zucman 2016) show much larger recent growth than estimates based on data from the Survey of Consumer Finances (SCF).

This very useful paper by Jesse Bricker, Alice Henriques, Jacob Krimmel, and John Sabelhaus does two main things. First, it carefully reconciles the competing estimates of growth in the concentration of income and wealth based on different data sources. It has persuaded me that the true growth in the concentration of wealth in recent years has been considerably smaller than suggested by the widely cited estimates reported by Emmanuel Saez and Gabriel Zucman (2016). Second, the paper offers a preferred set of estimates of the top 1 percent and top 0.1 percent shares of income and wealth that are neither the SCF estimates nor the tax data-based estimates. Mainly because I am skeptical about the choice of concepts for constructing the preferred estimates, I find this part of the paper less compelling. Absent from the paper is any discussion of who has lost as those at the top of the distributions of income and wealth have gained, a point to which I return below.

RECONCILING ESTIMATES OF INCOME AND WEALTH CONCENTRATION The paper looks at a great deal of evidence in the course of evaluating potential biases in alternative estimates of the top 1 percent and top 0.1 percent shares of income and wealth. In working through the discussion of this evidence and trying to keep it all straight, I found myself thinking that the total survey error framework familiar to my survey methodology colleagues might have provided a useful way to organize the information presented.

From a total survey error perspective, two types of things can go wrong with a survey estimate (Groves and others 2009). First, the surveyed units might do a poor job of representing the population of interest. This could be due to (i) problems with the coverage of the survey sampling frame (coverage error), (ii) the unavoidable variability that arises when information is collected only from a subset of the units of interest (sampling error), (iu) problems that arise because the units that respond differ in some important respect from those that do not (nonresponse error), or (iv) problems related to any postsurvey adjustments made to the survey weights (adjustment error). Second, the information collected from survey units might not capture what the survey designer actually hoped to learn about. This could be due to (i) the specified measure not corresponding adequately to the construct of interest (concept validity), (ii) problems with the accuracy of the responses provided in answer to the questions posed (measurement error), or (iii) problems with the processing of the responses obtained (processing error).

Representation issues. Although the total survey error framework--with its categorization of potential problems as being related to either representation or measurement--was originally developed for thinking about survey estimates, and though some of the specific issues it identifies are less applicable outside the survey context, it can easily be adapted to the evaluation of estimates based on administrative or other nonsurvey data. Within this framework, the fact that, by design, the SCF does not capture the 400 or so very wealthiest U.S. households can be classified as a representation issue. In the paper's preferred estimates, the income and wealth known to be missing because the top 400 families are excluded from the SCF sample is added back before the top 1 percent and top 0.1 percent shares are calculated.

Given the very low response rate in the SCF for wealthy households more generally, a more interesting question is whether, aside from the top 400 households, the SCF adequately represents the well-to-do. The paper makes a convincing case that households in the higher-wealth SCF strata that provide usable responses to the survey are quite similar to those that do not respond. The analysis reported in the paper gives me a good deal of confidence that, once the SCF data have been reweighted to account for differential nonresponse rates across strata, they should do a reasonably good job of representing all but the 400 wealthiest households.

The representation issue for the tax data lies at the opposite end of the distribution, with low-income households that do not file a tax return and therefore are missing from the data. Estimates based on tax data must be adjusted to account for the absence of these households, but seemingly adequate methods have been developed for doing this.

A related issue concerns the unit of accounting used to produce the SCF and tax data-based estimates. The SCF produces estimates for families, whereas estimates based on tax data pertain to tax-filing units. One new fact I learned from the paper is that high-income families are less likely than low-income families to file multiple tax returns, meaning that the top 1 percent (or top 0.1 percent) of families account for a materially smaller share of income and wealth than the corresponding share of tax-filing units. This discrepancy is fairly uniform across years, however, meaning that the choice of reporting unit does not explain the different trends in the tax data-based and SCF estimates of the concentration of wealth.

Measurement issues. The measurement side of the total survey error framework begins with the concepts that are adopted. For the purpose of measuring the share of income received by those at the top of the income distribution, the paper adopts as its benchmark the definition of personal income in the National Income and Product Accounts (NIPA). This consists of market income derived from current production plus current transfer receipts less current social insurance contributions. The paper's preferred wealth concept is a market measure, consisting of the value of household financial and nonfinancial assets plus rights to defined-benefit pensions, less the value of household liabilities.

Neither the tax data-based income measure nor the SCF income measure corresponds exactly to the NIPA's personal income concept; nor do they correspond exactly to one another. All three income concepts include most market income, though there are some differences across them related to the treatment of capital gains and of retirement income. The more significant differences across the three income measures have to do with how transfers are treated. The concept underlying the tax data-based estimates completely excludes transfers. The SCF concept includes cash transfers from Social Security, Supplemental Security Income, Temporary Assistance for Needy Families, and other transfer programs. Because cash transfers are more important at the lower end of the income distribution, their inclusion lowers the estimated top 1 percent and top 0.1 percent shares of income in the SCF estimates relative to the tax data-based estimates, though the trends in the two sets of estimates are not too different. The even lower level of the authors' preferred top income share estimates is largely attributable to both cash and in-kind transfers being treated as income.

The concepts underlying the tax data-based and SCF estimates of wealth concentration are more similar, though the SCF estimates miss defined-benefit pension wealth and the tax data-based estimates miss nonmortgage debt. The key factor, however, for understanding why estimates of the top wealth shares based on the two sources have trended so differently turns out to be how the data--and especially the tax data--are processed. For the most part, the SCF data measure the value of asset holdings directly. In the tax data, wealth holdings are estimated as the capitalized value of observed income flows. The basic idea is that, given the amount of investment income of a particular type and information about the rate of return on assets of that type, the value of the underlying assets can be inferred. More specifically, the implied value of a household's holdings of fixed-income assets would be calculated as the income realized on these holdings times 1/r, where r is an appropriate interest rate. The ratio 1/r would be referred to as the capitalization factor for the fixed-income asset category.

In the analysis by Saez and Zucman (2016), the capitalization factor for inferring the value of fixed-income assets from the associated flow of interest payments is set equal to the ratio of the aggregate value of fixed-income assets, as recorded in the Financial Accounts of the United States, to aggregate interest income from tax data. In recent years, the capitalization factor implied by these calculations has grown to be very large--by 2012, each $1 in reported interest income translated into nearly $100 in implied asset holdings. The resulting increase in the implied value of the fixed-income assets held by the wealthiest taxpayers accounts for most of the growth in the top 1 percent and top 0.1 percent shares of overall wealth as estimated by Saez and Zucman (2016).

Although the approach just described is logical, the results are potentially very sensitive to errors in the calculated interest rate that determines the capitalization factor. Bricker, Henriques, Krimmel, and Sabelhaus show that applying capitalization factors based on any of a set of market interest rates produces much different answers than those obtained by Saez and Zucman (2016), with each $1 in interest income in 2012 (the most recent year for which Saez and Zucman report) translating into about $25 to $50, rather than nearly $100, in implied asset holdings.

One potential contributor to Saez and Zucman's (2016) capitalization factor having grown too large that was identified by Bricker and his colleagues is that, in the recent environment of very low interest rates, a significant share of interest-bearing accounts have not generated the minimum $10 in interest payments that triggers the issuance of Form 1099-INT. To the extent that interest payments for which no 1099-INT is generated are not reported on recipients' tax returns, total interest income will be understated, leading to a corresponding overstatement in the capitalization factor and the holdings of fixed-income assets by high-income taxpayers. Perhaps surprisingly, the back-of-the-envelope calculations done by Bricker and his colleagues (shown in the online appendix to their paper) suggest that this factor alone could have caused Saez and Zucman's (2016) capitalization factor to be overstated by as much as 25 percent.

Although the paper does not fully explain the reasons for the difference between Saez and Zucman's (2016) capitalization factors for fixed-income assets and the capitalization factors based on market interest rates, I am convinced that Saez and Zucman's (2016) calculations significantly overstate the recent growth in the top 1 percent and top 0.1 percent shares of market wealth. Rather than having grown dramatically from about 28 percent in 1992 to about 42 percent in 2013, as implied by calculations based on income tax data using the methods developed by Saez and Zucman (2016), the authors' preferred estimate is that the top 1 percent share of wealth has grown by considerably less, from about 27 percent in 1992 to about 33 percent in 2013. The effect of the different methods on the estimated growth in the top 0.1 percent share of market wealth is, if anything, more marked. I view this as the paper's most important finding.

CONSTRUCTING PREFERRED ESTIMATES OF INCOME AND WEALTH CONCENTRATION Let me turn now to the paper's second objective: the production of preferred estimates of the top 1 percent and top 0.1 percent shares of income and wealth that are neither the SCF estimates nor the tax data-based estimates. A number of decisions have been made regarding the production of these estimates, including decisions about both what should be measured and how the measures should be constructed. In deciding what should be measured, Bricker and his colleagues are guided by the concepts that underlie the measure of personal income in the NIPA and the measure of market wealth contained in the U.S. Financial Accounts. One question I have about these choices is whether the treatment of taxes and transfers in the preferred measures makes good sense, given the likely reasons for data users to be interested in them. A further issue is the adoption of an annual observation window for measuring the concentration of income among those at the top of the income distribution.

The treatment of taxes and transfers. As with the estimates based on tax data and the SCF estimates, the preferred estimates of the top 1 percent and top 0.1 percent shares of income proposed in the paper are constructed on a pretax basis. The tax data-based income concept excludes transfer income altogether, and the SCF income concept adds cash transfers but not in-kind transfers. The more encompassing preferred income concept, which corresponds more closely to the NIPA personal income concept, incorporates both cash and in-kind transfers. A key point here is that the preferred income estimates are pretax but posttransfer, a conceptual formulation that from the perspective of thinking about the distribution of income seems to me to be rather betwixt and between. Measures of the concentration of market income convey information about the distribution of the returns to market activity. Measures of the concentration of income based on a post-tax and posttransfer income concept, which would map to the NIPA disposable personal income concept, convey information about the ultimate distribution of control over resources after societally determined redistributions have been made. It is less clear to me, however, how to think about the hybrid pretax and posttransfer concept that is adopted for the calculation of the paper's preferred income concentration measures.

With respect to measuring the concentration of wealth, the tax data-based estimates, the SCF estimates, and the preferred estimates all rest on a market wealth concept that generally corresponds to the concept used for the measurement of wealth in the U.S. Financial Accounts. Though sensible in isolation, making this choice for the preferred wealth measure introduces an inconsistency between the preferred income concept, which includes the value of both cash and in-kind transfers rather than being purely market based, and the preferred wealth concept, which includes only market wealth. As already noted, I am not entirely comfortable with the pretax and posttransfer concept underlying the preferred measures of income concentration. That said, I also am not entirely comfortable with the income concept and the wealth concept being defined on different bases. On this point, I would note that, with respect to thinking about household wellbeing, the expected present values of Social Security and Medicare benefits are the largest "assets" that many households possess (Steuerle and Quakenbush 2015). In a very real sense, ignoring the anticipated value of these transfers provides a misleading picture of the resources available to lower- and middle-income households. Defined more comparably to the preferred measures of income concentration, the estimated concentration of wealth in the hands of the top 1 percent or top 0.1 percent of households would look less extreme.

An annual frame of reference. A second potentially important issue about the preferred income estimates that I would like to flag is the adoption of an annual frame of reference for the calculation of measures of income concentration. Although this is entirely standard in the literature, I have reservations about whether such a measure in fact tells us what we really want to know. When a layperson thinks about the concentration of income, my guess would be that he or she has in mind something more akin to a measure of the concentration of income averaged over some period of years.

As noted in the paper, there is considerable idiosyncratic year-to-year volatility in incomes, especially at the very top of the income distribution. This is illustrated in my figure 1, constructed using data from the paper's figure 6, kindly supplied by the authors. The numbers plotted in the figure come from unpublished tabulations of tax data prepared by the Statistics of Income Division at the Internal Revenue Service. The figure shows, for the sample of tax-filing units with an adjusted gross income of more than $500,000 in 2011, the share of units experiencing various percentage decreases or percentage increases in adjusted gross income between 2011 and 2012. It is clear that the incomes of such households can change dramatically from one year to the next. More than half of those with annual incomes in excess of $500,000 in 2011 experienced a drop or an increase in income of more than 25 percent the following year.

[FIGURE 1 OMITTED]

Because the peaks and valleys will tend to average out over time, household income averaged over several years will tend to be less concentrated than household income in any single year. This same phenomenon can be seen clearly in related research on the inequality of earnings that has compared estimated inequality based on annual earnings with earnings averaged over several years (Kopczuk, Saez, and Song 2010). Further, changes in the volatility of income over time could impart a trend to measured annual income concentration relative to the concentration of income averaged over several years, though it is an empirical question as to whether this has been important in practice. In any case, it would be good to know whether and to what extent conclusions with respect to recent trends in the inequality of income are robust to the use of multiyear rather than single-year income data.

LOOKING BEYOND THE VERY TOP My final comment about the paper pertains to its exclusive focus on the shares of income and wealth among those at the very top of the distributions--the shares of income and wealth controlled by the top 1 percent or top 0.1 percent of households. It is undoubtedly of interest to know whether and by how much the shares of income and wealth controlled by these groups have changed. That said, I also would very much like to know at whose expense these gains are coming. My feeling about gains at the very top will be quite different to the extent that they are coming at the expense of households in the 90th through 98th percentiles rather than at the expense of households in the bottom 20 percent or bottom 40 percent of the income and wealth distributions.

Producing estimates of changes in shares for groups further down in the distribution is of course easier said than done, especially in the case of estimates based on tax data, given that there are a significant number of households that do not file tax returns and for which income would need to be estimated in some other way in order to calculate all the relevant income shares. Nonetheless, this seems to me to be a worthy objective for future research by this team of authors.

REFERENCES FOR THE ABRAHAM COMMENT

Groves, Robert M., Floyd J. Fowler Jr., Mick P. Couper, James M. Lepkowski, Eleanor Singer, and Roger Tourangeau. 2009. Survey Methodology, 2nd Edition. New York: Wiley.

Kopczuk, Wojciech, Emmanuel Saez, and Jae Song. 2010. "Earnings Inequality and Mobility in the United States: Evidence from Social Security Data since 1937." Quarterly Journal of Economics 125, no. 1: 91-128.

Saez, Emmanuel, and Gabriel Zucman. 2016. "Wealth Inequality in the United States since 1913: Evidence from Capitalized Income Tax Data." Quarterly Journal of Economics 131, no. 2: 519-78.

Steuerle, C. Eugene, and Caleb Quakenbush. 2015. "Social Security and Medicare Lifetime Benefits and Taxes, 2015 Update." Research Report. Washington: Urban Institute.

COMMENT BY

WOJCIECH KOPCZUK Jesse Bricker, Alice Henriques, Jacob Krimmel, and John Sabelhaus have produced very careful estimates of the magnitude and trends (from 1989 to 2013) in top wealth and income shares in the United States, relying on data from the Survey of Consumer Finances (SCF). This is of course not a new question, and the existing estimates of the top 1 percent share and the like have been highly influential, both in the economic literature and in broader public discussions. This paper adds to the existing evidence by providing high-quality estimates and by reconciling discrepancies between different methods. The authors' key contribution is their estimation of the top wealth shares, a topic on which there has been recent controversy.

Before delving into the details of the paper, it is useful to comment on the broader question of why one might be interested in wealth inequality, and in the top shares in particular. The paper's opening paragraph signals one reason: There is much popular interest in this topic. I take as given that we may be interested in inequality--but why in wealth? Wealth is a much more complicated outcome than income. Income itself does not measure the inequality of well-being or opportunities, and it comingles them with decisions about skill acquisition, occupational choice, hours of work, effort, saving, and portfolio choice. Focusing on wealth shares has the same problems, and adds some. It is inherently linked to the life-cycle dynamics of wealth accumulation--it is an outcome of the income, transfer, spending, and investment decisions that individuals make up to a particular point in time when they happen to be observed. In the natural economic approach, wealth reflects potential consumption (including that done in the form of transfers to others). Correspondingly, it is related to lifetime resources--and it does have advantages over permanent income, in that it responds to intergenerational transfers. However, if this is the objective of analyzing wealth inequality, then one should make clear how lifetime resources and wealth are related, and should at least account for age distribution--having the same amount of wealth means something very different at age 20 than at 65. Alternatively, one may be interested in the distribution of wealth itself rather than in learning from it about the distribution of consumption opportunities. The economic rationale for separating wealth from its consumption value is more speculative, but one can certainly consider the notions of political and personal power, and of control or status, that are tied to it. Arguably, the higher one goes in the distribution, the more important these issues become, providing some cover for focusing on the top wealth shares as they are, without a more carefully specified conceptual framework. This is not a complaint about this paper--measurement is important--but just a discussant's reminder that there is a considerable distance between what we can measure and the interpretation of what wealth inequality represents.

This paper provides estimates for both income and wealth, but its findings about wealth stand out as its key contribution. This is because estimates of the top wealth shares are much less settled than those of the top income shares, and there is substantial controversy about how they have evolved in recent years. (1) The paper provides estimates using the SCF, and it offers evidence that enables us to understand the sources of the differences between these estimates and the most prominent recent alternative: the capitalization approach offered by Emmanuel Saez and Gabriel Zucman (2016). The share of wealth of the top 0.1 percent, as estimated in the paper by Bricker and his colleagues, grew between 1989 to 2013 by about 4 percentage points--from a bit under 11 percent to close to 15 percent of aggregate wealth. In contrast, the estimated share of wealth of the same group, as analyzed by Saez and Zucman (2016), doubled from the similar level in 1989 to more than 20 percent in 2013. Both methods show that wealth concentration has increased, but the difference in trends is massive. And the temporal dynamics are also different; using the SCF approach, the top 0.1 percent share fluctuated somewhat but did not change much between 1995 and 2010, so the increase over the whole period is accounted for by changes between 1989 and 1995 and since 2010. In contrast, the capitalization approach shows relentless growth, with just a short break in about 2000.

With such a large difference in results, one would expect there to be a smoking gun as evidence for what is going on--and there is one here: Looking at the composition of assets of the top groups, the bulk of the discrepancy is due to the amount of fixed-income assets that both approaches yield. Both Saez and Zucman (2016) and I (Kopczuk 2015) have noted this discrepancy before, and this paper makes it clear that this is the mechanical source of the differences. How does it come about?

First, let us start with a potential problem on the SCF side. There is a discrepancy between capital income in the data from the Internal Revenue Service on which Saez and Zucman (2016) rely and what is observed in the SCF data. This could potentially mean that the SCF is not accurately capturing the very top of the distribution. This is certainly true in a narrow and obvious sense; the SCF explicitly excludes those individuals on the Forbes 400 list (to preserve confidentiality), but this particular issue is explicitly dealt with in wealth estimates by adding the estimated wealth of this group to the top shares. (2) There is an extensive and very informative discussion in the paper about the approach to and quality of sampling in the SCF that compares presurvey income tax information for respondents and non-respondents. This discussion indicates that the role of the sampling bias is limited, though it cannot prove it plays no role. In principle, it is still possible that even though respondents and nonrespondents are similar in prior years, their income trajectories could potentially diverge in the survey year (and perhaps be related to the reason for the difference in response behavior). This discussion is also limited to sampling for the 2013 SCF, leaving open the possibility of changes in the quality of the SCF's coverage. However, if by 2013 the survey is of a high quality, then the improved coverage of the top shares should strengthen rather than weaken the observed trend.

The paper's authors also note that the overall level of income of the top groups is consistent between the SCF data and the Internal Revenue Service data on which Saez and Zucman (2016) rely, and only its composition between capital income and other sources (primarily wages) differs. They speculate that the explanation may have to do with varying notions in the tax and survey data of what constitutes labor versus capital income, especially for business owners. I am quite sympathetic to this argument--as any public finance economist working on capital taxation knows, the line between labor and capital is inherently imprecise, and it is certainly possible that tax accounting differs from the common-language way of separating labor from capital. I also find persuasive the argument that the close match of the overall income concentration measures suggests differences in the classification of income rather than bias. Still, at the end of the day, there is a difference in capital income observed in the two sources, and this is clearly an important future research area for improving our understanding of the SCF's concepts and quality of sampling. Also, perhaps more can be done with the existing data to further explain which components of capital income are a problem and how these discrepancies evolved over time.

The alternative explanation for the discrepancy has to do with how capitalization estimates are constructed. In Saez and Zucman's (2016) capitalization approach, observed capital income must be multiplied by a capitalization factor in order to arrive at the underlying level of wealth. Thus, if unobserved asset worth A generates observed return rA, one needs to multiply rA by the capitalization factor, 1/r, to arrive at the original stock. If realized r were known, this would be an uncontroversial--and trivial--procedure. However, r varies over time, it varies on average across asset classes, and it varies across individual portfolios within an asset class. In a nutshell, Saez and Zucman's (2016) approach is to use aggregate information about flows and stocks by asset classes to construct average capitalization factors, while assuming that they do not vary across income distributions and providing a battery of approaches and outside data to test sensitivity. This procedure still allows for differences in rates of return across income groups, because their portfolio compositions might differ, but this can only be due to differences in portfolio composition across very broad asset classes, which include fixed income, equities, business assets, and housing--categories that match the limited level of detail observed in data on income tax returns.

For all this approach's reliance on microeconomic data, the capitalization factor for a particular asset class is a single number for a particular year, which is constructed on the basis of aggregate data. Any bias in this factor skews the estimated value of the whole asset class. Any bias in its trend generates a trend in the estimated value of the underlying asset. In an environment with a low rate of return, a seemingly small bias in the estimated rate of return has large consequences. The capitalization factor for taxable interest income used by Saez and Zucman (2016) for 2009 is 96.6, which corresponds to the estimated rate of return of 1.04 percent for the asset class that it reflects in the economy as a whole. Hypothetically, imagine that we are underestimating the true rate of return by 1 percentage point. In this case, the true capitalization factor would be 50 (or, if it were instead an upward bias, it could be 2,500 ...), so the assumed 96.6 capitalization factor would erroneously double the amount of wealth estimated in this particular asset category! In an environment with a higher rate of return, however, the implications of mismeasurement will be more benign. In the 1990s, the capitalization factor for taxable interest income was about 25. In that case, increasing the corresponding rate of return by 1 percentage point, from 0.04 to 0.05, would modify the capitalization factor to 20--still a bias, to be sure, but the value of the assets would be overestimated by 25 percent rather than 100 percent.

Moving beyond hypothetical situations, the paper's figure 9 shows that directly observed rates of return on some fixed-income assets (Treasuries, bonds) are higher than those implied by observed interest income on individual income tax returns, as analyzed by Saez and Zucman (2016), so that relying on them would translate into large differences (by a factor of 2 or more, by the end of the period) in capitalization factors. The paper's figure 10 then shows that reducing the capitalization factor for fixed-income assets brings the estimates of the SCF and Saez and Zucman (2016) much closer to each other, especially in the 2000s, when they track each other fairly closely.

The paper's authors suggest that the overestimation of the capitalization factor is the reason for the discrepancy in fixed-income estimates that constitutes the bulk of the difference. I have also suggested so in the past (Kopczuk 2015), and thus--not surprisingly--I concur. The key series for me are those capitalization factors that rely on the linked estate and (pre-death) income tax data; this approach constructs the rate of return that is specific to a high-net-worth population and, in particular, it reflects a wealthy-specific portfolio composition within asset classes. One can still worry about the quality of information for the estate tax versus the income tax, the timing of when income and wealth are observed, and the representativeness of those who died for the whole wealthy group. However, the fact that it moves closely in sync with the Treasury rate and its growing discrepancy with the series assumed by Saez and Zucman (2016) over the 2000s strongly suggest the existence of a trending bias in their capitalization factor. If the capitalization factor based on the estate-income rate of return was the approach used in the baseline figures of Saez and Zucman (2016) (rather than that reported in their figure B27b, on page 79 of their 385-page-long online appendix (3)), we would be left with an understanding of the remaining discrepancies in figure 10 rather than of the major differences in trends shown in figure 1.

Having said this, the remaining and interesting question is why the interest income observed on income tax returns would imply too low a rate of return. Let us assume that there are no problems with measuring the underlying aggregate stock of fixed-income assets. There are two main possibilities. One is that some interest income is not reported or that some fixed-income assets generate no interest income (my checking account!). The other possibility (which is closely linked) is that fixed-income assets are still a broad category that, in particular, includes checking accounts, savings accounts, certificates of deposit, and bonds. In practice, these different types of investments correspond to different rates of return, but Saez and Zucman's (2016) capitalization factor is based on the average rate of return for the whole class. The much lower implied capitalization factor, which is based on an income-estate link that is not far from the Treasury rate, suggests that the portfolios of the wealthy are tilted toward higher-yield assets (for example, bonds) relative to the general public's low-interest deposits. This would always result in bias; but in a world where the general public earns 3 percent and the top of the distribution earns 5 percent, this bias is much smaller than in a world of 0 percent versus 2 percent earnings. If, for simplicity, each group had half the aggregate assets, we would be back to my original example, with average rates of 4 percent and 1 percent and a 1 percentage point difference between the average rate of return and the one that should be used for the wealthy population.

I am not aware of any outside evidence (other than the capitalization method) that would indicate that between 2000 and 2012, the top 0.1 percent did indeed rebalance their portfolios to increase their holdings of fixed-income assets from 21 to 43 percent of their net worth, as implied by the approach taken by Saez and Zucman (2016, table B5b). This finding is driven by declining fixed income, multiplied by strongly increasing capitalization factors. Given the issues with constructing the capitalization factors, I find the evidence in this paper that indicates no such rebalancing in the SCF much more plausible.

In conclusion, this very valuable paper provides timely and careful estimates of the top wealth shares and makes a persuasive argument for the source of the discrepancy between these results and those of Saez and Zucman (2016). This is not a mortal blow to the capitalization method; nor is it intended to be one. The two methods are certainly complementary, and one way of describing the discrepancy's source is that it is due to a particular implementation of the capitalization method rather than the method itself. Adjusting capitalization factors to match the portfolios of the rich is certainly a feasible task. However, the paper does highlight how the capitalization approach is very sensitive to hard-to-estimate capitalization parameters and how the assumption of the constant rate of return across income groups for broad asset classes is potentially problematic. This approach is also heavily based on tax reporting, with all its associated conceptual problems. Hence, I view it as a complement to approaches that are based on observing wealth directly (such as surveys, the administrative data on wealth available in some countries, and estate tax data) rather than the preferred alternative. In the United States, the SCF remains the prime source of information for understanding wealth distribution.

REFERENCES FOR THE KOPCZUK COMMENT

Kopczuk, Wojciech. 2013. "Taxation of Intergenerational Transfers and Wealth." In Handbook of Public Economics, Volume 5, edited by Alan J. Auerbach, Raj Chetty, Martin Feldstein, and Emmanuel Saez. Amsterdam: North-Holland.

--. 2015. "What Do We Know about the Evolution of Top Wealth Shares in the United States?" Journal of Economic Perspectives 29, no. 1: 47-66.

Raub, Brian, Barry Johnson, and Joseph Newcomb. 2010. "A Comparison of Wealth Estimates for America's Wealthiest Decedents Using Tax Data and Data from the Forbes 400." In Proceedings of the 103rd Annual Conference on Taxation. Washington: National Tax Association.

Saez, Emmanuel, and Gabriel Zucman. 2016. "Wealth Inequality in the United States since 1913: Evidence from Capitalized Income Tax Data." Quarterly Journal of Economics 131, no. 2: 519-78.

(1.) For an extensive discussion, see Kopczuk (2015).

(2.) Note, however, that this approach takes at face value the estimates of net worth reported in Forbes publications. There are reasons to be skeptical about precision here; these estimates sometimes mix the wealth of a whole family with an individual's wealth, and they may miss some components of net worth, in particular debt. Raub, Johnson, and Newcomb (2010) compared the Forbes estimates with estate tax reports for individuals who died while on the list and found that reported estates are only about 50 percent of the Forbes numbers. Though some of this may reflect tax avoidance, the magnitudes are substantially larger than existing evidence of the extent of tax avoidance (Kopczuk 2013), suggesting that Forbes is likely to somewhat overestimate the net worth of these individuals. Hence, if anything, I suspect that the approach taken by the Forbes list leads to upward bias in estimated top shares.

(3.) The online appendix is found at http://eml.berkeley.edu/~saez/SaezZucman2016QJE Appendix.pdf.

GENERAL DISCUSSION Moderator James Stock began by posing two simple questions intended to help frame the discussion to come. First: What, from an economist's perspective, do we mean by income, and is that something that is available for current consumption? And second: What do we mean by wealth? Noneconomists tend to conflate the two terms, so he thought that a really clear statement about the definitions of the ultimate economic objects was a good place to start.

Justin Wolfers suggested that when the general public asks about what the level of wealth and inequality is, economists generally have two choices: The first is simply to measure it, and the second is to patiently explain to the general public that wealth is a less useful concept than casual intuitions would suggest. He suggested that the next time someone asks him about the level of wealth inequality, he may just refuse to answer the question. He added that if economists' role in this debate is to educate the public, what is important is actually understanding the concept that the public is after, which he argued is not actually the economic concept of wealth.

Martin Feldstein made three comments following his observation that although income distribution research is a very interesting subject, one real obstacle is that the interesting parts of the income distribution--the very top and the very bottom--are where the data are most uncertain. He first noted that one thing that was not included in the authors' tabulations was Social Security wealth, despite the fact that for most people, it is the thing that they count on for retirement income for themselves and for their survivors. According to Feldstein, the Social Security trustees estimate that current Social Security wealth is about $59 trillion. Household net worth, by comparison, is about $80 trillion. Additionally, Medicare and Medicaid wealth is estimated to be roughly $50 trillion. Between Social Security wealth, Medicare wealth, and Medicaid wealth, the total far exceeds official household net worth. He wondered if the authors had an explanation for these seemingly large omissions.

Second, Feldstein was interested in wealth's relation to political influence and power, and what that might mean for high-income people. He suspected that, in reality, one probably does not gamer a lot of political influence and power from being in the 99th, or even the 99.9th, percentile of the wealth distribution, where annual incomes are only about $500,000 and $1.6 million, respectively. Last, Feldstein commented on measuring incomes over time. Data from Thomas Piketty show that a big shift in inequality at the top income percentiles started to happen in the 1980s, and Feldstein believed that a lot of that was driven by tax changes. For example, the top marginal tax rate on investment income declined from 70 percent in 1980 to 50 percent, and eventually to 28 percent, meaning the net-of-tax share rose from 30 percent to 72 percent. Not surprisingly, people chose to recognize more income on their tax returns, and that is where one begins to see all the data for the very top percentiles. Even more important, he argued, were the major changes in tax rules brought about by the Tax Reform Act of 1986. The reforms induced individuals who had a separate small corporation in their own name to shift that income from the corporation into their regular income tax returns, which led to a big increase in reported personal incomes. He therefore concluded that it could be misleading or inappropriate to compare post-1986 personal incomes with pre-1986 data.

Matthew Shapiro protested against obsessing over the shares of wealth. More important, he argued, was what levels go with the values of these shares, and how they have evolved over time. It is very different if wealth has doubled and all the wealth has gone to some group versus wealth being stagnant and one group getting more. The reality is that the situation is probably something in between. Likewise, Shapiro added that one should probably also care about the composition of the changes, which he believed was buried in the authors' numbers. He thought it would be nice if the paper included some of the level information and some ways one might think about its shifts. While the shares are interesting, so are the levels and the sources of the changes in levels.

Scott Winship believed that the paper lent support to the usefulness of the Congressional Budget Office's income concentration data. He wondered if the authors had any plans to look into measuring income from gains on an accrual basis rather than when they are realized, which Jeff Larrimore, Richard Burkhauser, and others have shown does make a big difference. Realized gains are lumpy, and if one could produce estimates that distribute those gains in some sense over the period in which assets are held, that could make a noticeable difference in the estimates, especially for the levels.

Janice Eberly was struck by the observation that so much of the divergence in measurement was driven by the fixed-income sector. She argued that although discussant Wojciech Kopczuk did a nice job of pointing out in his presentation how sensitive those valuations are to low levels of interest rates, it did not absolve anyone from trying to figure out what is going on in the U.S. Financial Accounts. The measure that comes out of the Financial Accounts might be implausible, in that it is out of line with other measures, but it is still important to know why it is implausible. Saying that the results are sensitive indicates that one should be worried about the effect of small errors, but it does not absolve anyone from still trying to get the best point estimate. Eberly argued that simply ignoring the Financial Accounts might not be the right way to go, and wondered if the authors or others in the room knew why the Financial Accounts were not only off but also increasingly off, and how this is related to the measurement of fixed-income assets. She added that the U.S. Treasury Department has put a lot of effort into getting better disclosure and better information on international accounts through the Foreign Account Tax Compliance Act, and wondered if those efforts might be driving some of the divergence.

Alan Blinder wondered if there was more that could be said about the underreporting of closely held business assets, which he believed to be quite substantial, and whether it had increased or decreased. William Brainard wondered how much the likelihood that some individuals in the very-high-income distributions move their money to places that minimize their estate tax obligation biases the authors' estimates. Another speaker noted that a lot of the wealth held by the very wealthy usually takes the form of things that are objectively very hard to value; they are not things that have a daily market value, and one can only know the value once they are sold. She wondered if it was important to acknowledge that imprecision, and how much of an issue it might be for the estimates.

In response to Brainard's question about estate tax avoidance, Kopczuk noted that it is definitely a big issue, and that it may show up in various estimates. He cited a paper published in 1977 about the very issue as evidence that estate tax avoidance is not a new concern. (1) In fact, the incentives were stronger in the past than they are now because the tax rates were higher. He found it implausible that there has been a huge trend in tax avoidance that would be severely skewing the data. He agreed with Feldstein that the tax reform of the 1980s had a substantial effect on income observable on personal tax returns, skewing measurement of the distribution of income and, possibly, of wealth. He cited evidence from Norway showing that such reporting effects are quantitatively important.

Discussant Katharine Abraham agreed with Eberly that it is certainly worth looking more carefully at the Financial Accounts to understand what is going on with them, though she warned that the problems may not lie there. Rather, there may be problems related to the reporting of interest income. She explained that in an environment of very low interest rates, there are a lot of people who have significant assets with very low interest earnings, so low that they are not reporting them on their tax returns.

She pointed to some interesting calculations done by the authors that suggest that that could account for a big chunk of the growth in the observed concentration of fixed-asset wealth at the top of the distribution because the total unreported interest incomes of many people at the bottom of the distribution who hold small accounts may add up to a substantial amount.

In response to Stock's opening comment about the concepts and what exactly the present paper is trying to measure, John Sabelhaus noted that measuring wealth is essentially measuring potential consumption. Measuring income, on the other hand, has more to do with behavior, and asks the question: How much of that potential consumption are people actually consuming? He agreed with Wolfers that it is important to communicate exactly what economists are measuring when they talk about wealth, and that the concept of wealth may not actually be what people think it is.

On the questions of measurement and capitalization factors, Sabelhaus noted that the authors had done a lot of digging into what could be going wrong with the measurement of not-for-profit holdings in the Financial Accounts, and the measurement of taxable interest income on individual tax returns, much of which he noted was slipping over into tax-preferred retirement accounts. He also added that low interest rates exacerbate the measurement problems.

On the question of the underreporting of small businesses' assets, Sabelhaus stated that they are really tricky to value, but added that the authors did not see that as a particular problem, and that he believed they did a good job on that front.

(1.) George Cooper, "A Voluntary Tax? New Perspectives on Sophisticated Estate Tax Avoidance," Columbia Law Review 77, no. 2 (1977): 161-247 (reprinted in 1979 by the Brookings Institution Press).

JESSE BRICKER

Federal Reserve Board

ALICE HENRIQUES

Federal Reserve Board

JACOB KRIMMEL

University of Pennsylvania (1)

JOHN SABELHAUS

Federal Reserve Board

(1.) This paper was written while Jacob Krimmel was a research assistant at the Federal Reserve Board.

(2.) Notable exceptions include, for the top income shares, Congressional Budget Office (2014); Burkhauser, Larrimore. and Simon (2012); Burkhauser and others (2012); and Smeeding and Thompson (2011). For the top wealth shares, notable exceptions include Kopczuk (2015b).

(3.) Bricker and others (2014) describe the results from the latest SCF, conducted in 2013. A slow rise in the top wealth shares is also consistent with estimates derived from administrative estate tax data (Kopczuk and Saez 2004).

(4.) Piketty and Saez regularly update the tables and statistics from their 2003 paper. The most recent version, updated to 2014, is available at http://eml.berkeley.edu/~saez/ TabFig2014prel.xls. We refer to these updated data throughout this paper.

(5.) SCF income values are for the year preceding the survey.

(6.) These issues are not unique to the United States. See, for example, Atkinson, Piketty, and Saez (2011), who provide a multinational and longer-run view of rising income inequality.

(7.) The top share estimates from Piketty and Saez (2003) and Saez and Zucman (2016) are regularly updated and published in the World Wealth and Income Database, which is maintained by Facundo Alvaredo and Tony Atkinson, along with Thomas Piketty, Emmanuel Saez, and Gabriel Zucman. This database is accessible at www.wid.world.

(8.) The Financial Accounts of the United States (Statistical Release Z.1) are available from the Federal Reserve Board (http://www.federalreserve.gov/releases/z1).

(9.) Greenwood (1983), among others, provided the foundational work for the capitalization approach. Capitalization is used in conjunction with other approaches in the SCF sampling procedure. See the online appendix to this paper and Kennickell and Woodburn (1999) for more details. The online appendixes for this and all other papers in this volume may be found at the Brookings Papers web page, www.brookings.edu/bpea, under "Past Editions."

(10.) The SCF, administrative income tax, and our preferred measures of wealth and income can be biased by mismeasurement. The mismeasurement in the SCF can come from a respondent misreporting wealth or income components, and the income tax data can suffer from mismeasurement by tax avoidance and evasion. For this to matter in the analysis of top share trends, however, mismeasurement must have changed in a nonrandom way over our time series.

(11.) This is described in the online appendix.

(12.) Most of the discussion here is focused on concepts in FA table B.101, though the reconciliation between SCF and FA aggregates also involves details on pensions from subtables, such as table L.117. For details on the SCF and FA reconciliation, see the online appendix, Henriques and Hsu (2014), and Dettling and others (2015).

(13.) The SCF collects information on the value of such charitable trusts and foundations, and wealth held in these entities. Including these assets along with SCF household wealth would have only marginal effects on our top share estimates presented later. In the 2010 SCF, for example, the wealth share held by the top 1 percent would increase from 34.5 to 34.7 percent. Further, these assets only constitute about 9 percent of the total assets held by nonprofits (authors' calculations; McKeever 2015).

(14.) Devlin-Foltz, Henriques, and Sabelhaus (2016) estimate the present value of Social Security benefits for the cohort of near-retirees in 2013, for whom future taxes are inconsequential, and show that inequality in total retirement claims is effectively eliminated when Social Security is included. Specifically, the ratio of average total retirement claims (individual retirement accounts, defined-contribution accounts, and the present value of defined-benefit pensions and Social Security) to average income is roughly constant across most lifetime income groups, and lowest at the very top of the distribution.

(15.) "Family" is defined here as the economic core of a household and all people at that address whose finances are intertwined with that person.

(16.) Net worth is generally calculated as households' total assets (financial and nonfinanal) minus their total liabilities (debts to other sectors). However, because households effectively "own" the other private sectors (such as corporations) through ownership of equities and debt, household sector net worth effectively represents all private net worth claims.

(17.) Fagereng and others (2016) test this assumption and reject it. Families at the upper tail of the wealth distribution have much higher rates of return than other families. Tabulations from the SCF are consistent with this finding as well.

(18.) The algorithm for distributing SCF DB pension wealth is described in the online appendix and in greater detail by Devlin-Foltz, Henriques, and Sabelhaus (2016).

(19.) Most of the discussion here is focused on broad income concepts in NIPA table 2.1, though a comprehensive reconciliation with the micro data also involves details from other parts of the NIPA, such as tables 1.12, 3.12, 7.9, 7.10, 7.11, and 7.20. For a detailed reconciliation of NIPA and SCF incomes, see Dettling and others (2015).

(20.) One aspect of income concentration we do not (and cannot) address in this paper is the conceptual issue of what frequency should be used to measure top shares. Wealth is generally more straightforward, because concentration is measured at a point in time, though we will see frequency also plays a role there in terms of what can and cannot be measured. One can argue that income concentration should be measured at lower frequencies, in order to sort out transitory income effects, and also to address some of the conceptual issues we raise, such as measuring retirement income when the claim is established versus when the income is actually received. The decision here to focus on annual measures is largely driven by what data are available over long periods.

(21.) The evolving differences in the concept of income in administrative versus survey data are also emphasized by Burkhauser, Larrimore, and Simon (2012); and by Armour. Burkhauser, and Larrimore (2014).

(22.) Statistics on tax units here and later in the paper are from Emmanuel Saez's website, in the regularly updated file http://eml.berkeley.edu/~saez/TabFig2014preI.xls. The actual unit of observation in the SCF is the "primary economic unit," which is somewhere between the census "family" and "household" concepts. See the appendix to Bricker and others (2014) for a precise definition. The number of families in the SCF is benchmarked to that found in the Current Population Survey. The number of tax units includes an estimate of nonfilers.

(23.) See the online appendix for a detailed discussion of the SCF sampling strategy. See Sabelhaus and others (2015) for direct estimates of the relationship between income and unit nonresponse. O'Muircheartaigh, Eckman, and Weiss (2002) provide a comprehensive description of the National Opinion Research Center's national area probability sample.

(24.) The sampling frame technically excludes other "public" figures as well, but assuming that those families have observational equivalents who are not public figures, there is no bias in the estimated wealth distribution.

(25.) They estimate that nonfliers have 20 percent of the average income of filers, where income is defined using the same taxable income concepts of the filers.

(26.) Sabelhaus and others (2015) show this is the case for the Consumer Expenditure Survey and Current Population Survey (CPS). Burkhauser and others (2012) show that at least some of the divergence between the CPS and administrative incomes is also due to top-coding of very high incomes in the CPS. Attanasio, Hurst, and Pistaferri (2015) use household budget data to study inequality; and in addition to the nonresponse issues, they find that reporting problems further confound consumption-based inequality estimates.

(27.) The online appendix has extensive details about the SCF sampling process. At the time the list sample was drawn, the most recent complete administrative data were those from two years before the survey year. The sample includes individual and sole proprietorship tax filings from the Internal Revenue Service's administrative tax data. These data are made available by the Statistics of Income Division in its annual publication no. 1304, available at https://www.irs.gov/uac/SOI-Tax-Stats-Individual-Income-Tax-Returns-Publication- 1304-(Complete-Report).

(28.) See, for example, the discussion by Kennickell and Woodburn (1999).

(29.) One would perhaps like to compare respondent and nonrespondent incomes in the survey year itself, or to compare respondent-reported and administrative incomes for the survey year, but any such comparison would involve an implicit audit and thus violate the explicit agreement the SCF has with respondents to not audit their data.

(30.) Capital income here includes taxable and nontaxable interest, dividends, Schedule C and Schedule E business income, Schedule F farm income, and capital gains.

(31.) Results across income concepts, strata, and for earlier years are available upon request.

(32.) In 2013, the differences for the second-highest stratum were significant at the 5 percent level. Again, results for other years, income measures, and stratum are available upon request.

(33.) See, for example, Debacker and others (2013); Guvenen, Kaplan, and Song (2014); and Parker and Vissing-Jorgenson (2010).

(34.) We are grateful to the Internal Revenue Service's Statistics of Income Division for the unpublished growth rate distributions and threshold comparisons described here.

(35.) Almost 19 percent of SCF families in the top two sampling strata had not yet filed their taxes as of the interview date but planned to do so; only 4 percent of all other SCF families had not yet filed taxes. Many high-wealth families file their taxes late in the year, after getting an extension.

(36.) The archive of SOI Bulletins is available at https://www.irs.gov/uac/SOI-Tax-StatsSOI-Bulletins. For the most recent "Individual Tax Shares" report, see Dungan (2015). We are grateful to SOI for providing thresholds and counts in the early SCF years not covered in the published time series.

(37.) One subtle point about negative incomes affects the very top end in an important way. A taxpayer experiencing a capital loss may have that loss limited in a given tax year, but, for example, a business loss may be fully deductible against other positive incomes. Thus, if an SCF respondent accurately reports a loss, but misreports the type of loss, he could be misclassified based on "total" income. The analysis here is based on the SCF "total income" measure, which is, at the end of the day, the respondent's best estimate as to what he actually received during the year.

(38.) The wage share of income of the top 1 percent of SCF families was 47 percent in the 2001 SCF and was 49 percent in 2013 (authors' calculations). In the tax data, comparable wage share of families reporting more than $200,000 in AGI (roughly comparable to the top 1 percent) was 45 percent and decreased to 44 percent (SOI table 1.4; see note 27).

(39.) We also show in the online appendix that the income tax data may be missing some forms of capital income for lower-income families in recent years, which would lend an upward bias to capital income concentration estimates in the income tax data in figures III and X of Saez and Zucman (2016). Further, the shares reported in the final year of these figures are undoubtedly biased up because 2012 was a year when many wealthy families chose to realize capital income (Wolfers 2015).

(40.) "Bulletin" wealth derives its name from the fact that this is the consistent series published in the Federal Reserve Bulletin after each triennial survey. For the most recent survey, see Bricker and others (2014). Our estimate of Forbes 400 wealth is found by summing up the wealth of the families from the list, which was $2,021 trillion in 2013, or about 3 percent of total household wealth. We add this total to the total wealth in the SCF to create a new estimate of total U.S. family wealth. To compute a new top 1 percent estimate, we remove from the SCF top 1 percent those families that represent the 400 lowest-wealth families (weighted) and add the total Forbes 400 wealth, then divide by the new estimate of total U.S. family wealth (which includes Forbes 400 wealth). Alternatively, we can estimate the top shares after including the Forbes 400 families by using inferences from a Pareto distribution (Vermeulen 2014). The answers are qualitatively similar, though we prefer to use the data rather than make the inherent assumptions necessary for the Pareto distribution.

(41.) There are a few minor differences between the preferred measure and FA household sector net worth, described in the online appendix, and introduced to make the estimates more consistent with Saez and Zucman (2016). Primarily, we start with SCF Bulletin net worth, subtract vehicles, miscellaneous financial and nonfinancial assets, cash value of whole life insurance, and miscellaneous debt.

(42.) The slower growth of top shares in the SCF is also consistent with patterns in the top shares derived from estate tax data, as in Kopczuk and Saez (2004). Saez and Zucman (2016) include updates of the estate tax estimates, but these estimates are very sensitive to assumptions about mortality differentials for decedents affected by the estate tax.

(43.) See Dettling and others (2015) for a comparison of aggregate SCF and FA balance sheets for the 1989-2013 period. Also, Brown and others (2013) show that SCF debt by category generally tracks Equifax aggregates very well, though some categories such as credit cards are difficult to compare because of point-in-time versus revolving balance accounting for debt outstanding.

(44.) The differences in SCF and FA housing stock valuations are driven by the very different methodological approaches. In the aggregate FA data, the housing stock is valued using a perpetual inventory that involves new investment, depreciation, and a national house prices index. In the SCF, house values are owner-reported. Henriques and Hsu (2014) discuss how house values in the SCF compare favorably with other micro-based estimates, such as the American Housing Survey, and Henriques (2013) provides evidence that SCF respondents' house valuations generally track local area house price indexes quite well. See the online appendix for more details.

(45.) In practice, this constraint is imposed by simply changing the target counts of families in a given fractile to match the estimated number of tax units in a given fractile, which is the same as saying that every household at the top is also a tax unit. As noted earlier in the paper, there were about 30 percent more tax units than families in 2013, so one can think of the constrained "top 1 percent" as really representing the top 1.3 percent of families. The online appendix has details about the distributions of tax units versus families.

(46.) The online appendix and SCF website have details about how to use replicate weights and bootstrapping for generating confidence intervals consistent with the dual-frame sample design.

(47.) For reference, the gross capitalization model used in the SCF sampling exercise (see the online appendix) uses the Moody's Aaa rate to capitalize SOI interest income. It is also worth noting that the bond series in the B. 101 table of the FA has been subject to downward revision as new source data have become available.

(48.) The rate of return on these sorts of assets does appear to vary across the wealth distribution in the SCF. In the 2013 SCF, the average rate of return on fixed-income assets (found by the ratio of SCF interest income to SCF fixed-income assets) across all households is about 1 percent, but the average rate of return for the top 1 percent of families is almost 6 percent. Fagereng and others (2016) also show that families at the upper tail of the wealth distribution have much higher rates of return than other families.

(49.) Some of these issues may impart serious bias to the capitalization factors. The online appendix describes these issues in more detail, and some back-of-the-envelope calculations suggest that substantial biases in capitalization factors are likely introduced by these inconsistencies between micro income and macro balance sheet estimates.

(50.) The Forbes 400 is based on estimated wealth holdings, and Forbes makes no attempt to produce estimates of the incomes those families earn during the year. We estimate their incomes using information on income and wealth for the top 0.1 percent of families in the SCF sample, for which we know both income and wealth. For those top families, we compute the median ratio of income to wealth, and then we apply that ratio to the estimated Forbes 400 wealth. Although the Forbes 400 account for about 3 percent of total wealth, our approach suggests they account for less than 1 percent of income.

(51.) The distribution of the in-kind transfers is, as with our wealth imputations, driven by the available data in the SCF. Employer-provided health care benefits are distributed across families based on their reported employer-sponsored health care coverage, Medicare is distributed equally for eligible families, and the means-tested transfers are all distributed to the bottom 99 percent by income.

(52.) The working paper version of this paper (Bricker and others 2015) has more details on the variability of top incomes, particularly with respect to the capital income shares. Saez and Zucman (2016) emphasize that the failure of the SCF to capture top capital incomes is indicative that the survey is missing the top wealth holders, but we show there that most of the capital income at the top is captured as well after doing the same reconciliation exercise we do here for total incomes, and the remaining modest differences are likely associated with some of the reporting issues discussed in section I of this paper.

(53.) For a somewhat contrary position on the economic stability effects, see Bordo and Meissner (2012).

(54.) The view that markets underlie rising inequality is well described by Kaplan and Rauh (2010, 2013). See also Jones (2015) for a discussion of how competition among innovators affects the top shares.

(55.) The Social Security actuaries estimate that the present value of future Social Security benefits for current workers is currently about $58 trillion, which is nearly the size of conventionally measured household sector net worth. Social Security wealth is also rising faster than other forms of wealth. Devlin-Foltz, Henriques, and Sabelhaus (2016) show how the distribution of Social Security wealth for near-retirees interacts with other forms of retirement wealth. Not surprisingly, given the progressive nature and cap on earnings in the benefit formula, Social Security wealth is disproportionately important for the bottom half of the wealth distribution.

Table 1. Measuring Household Wealth in the Survey of Consumer
Finances and Capitalized Administrative Tax Data

                     Survey of Consumer
Concept                   Finances             Administrative tax data

Owner-occupied    Direct report on value of   Allocate FA housing total
housing           primary residence           by capitalizing property
                                              tax paid on Form 1040
                                              (among itemizers)

+ Businesses      Direct report on value of   Allocate FA total by
                  businesses                  capitalizing business
                                              income on Form 1040

+ Nonretirement   Direct report on value      Allocate FA total by
financial         of checking accounts,       capitalizing interest,
                  savings accounts,           nontaxable interest,
                  certificates of deposit,    dividend income on Form
                  mutual funds, directly      1040
                  held stocks, annuities,
                  trusts, managed accounts

- Mortgage        Direct report on value of   Allocate FA outstanding
liabilities       mortgage balances           mortgages by capitalizing
                                              mortgage interest
                                              deduction reported on
                                              Form 1040

- Other           Direct report on value of   Unallocated
liabilities       lines of credit, car
                  loans, education debt,
                  credit cards, other
                  consumer debt

+ Defined-        Direct report on value of   Allocate FA pension total
contribution      individual retirement       using wages and pension
(DC) retirement   accounts, DC pensions       payments (defined-benefit
                  on current and past jobs    [DB] and DC are not
                                              separated)

= Marketable      SCF Bulletin concept        Allocate FA pension total
net worth + DB    Allocate FA DB total        using wages and pension
retirement        using wages and direct      payments (DB and DC are
                  report on plan              not separated)
                  participation and
                  benefits

= Private net     Preferred estimate
worth +                                       Saez and Zucman (2016)
Unallocated                                   estimate
liabilities

Table 2. Income Concepts and Data Sources

                 Survey of                             National Income
                 Consumer           Administrative     and Product
Concept          Finances           tax data           Accounts

Wages and        Concepts           Concepts           Concepts
salaries,        generally          generally          generally
business         consistent with    consistent with    consistent with
income,          income tax-based   income tax-based   income tax-
interest and     reporting          reporting          based reporting
dividends paid
directly to                                            Adjusts for
incomes                                                underreporting
                                                       of proprietors'
                                                       income, various
                                                       rental and other
                                                       capital income
                                                       imputations

+ Realized       Concepts           Concepts           Capital gains
capital gains    consistent with    consistent with    not included in
                 income tax-based   income tax-based   NIPA PI
                 reporting          reporting
                                                       Adjusts timing
+ Retirement     Excludes           Excludes           to match micro
income cash      employer           employer           data concepts
flow timing      contributions to   contributions to
adjustment       and earnings on    and earnings on    Effectively
                 pension balances   pension balances   subtracts part
                 and Social         and Social         of "net saving"
                 Security           Security           in retirement
                                                       plans from NIPA
                 Includes           Includes taxable   PI
                 withdrawals and    withdrawals and
                 payments from      payments from
                 retirement plans   retirement plans

= Market                            Piketty and Saez
income                              (2003) concept

+ Government     Social Security    No information     Includes all
cash transfers   collected          on nontaxable      government cash
                 separately in      cash transfers     transfers
                 work and
                 pensions module
                 and as a
                 component of
                 total in income
                 module

                 Supplemental
                 Security Income,
                 Temporary
                 Assistance for
                 Needy Families,
                 and other cash
                 transfers
                 collected in
                 income module
                 (known to be
                 somewhat under-
                 reported, as in
                 other surveys)

= Total cash     SCF Bulletin
income           concept

+ In-kind        No direct          No direct          Includes all
transfers        information on     information on     employer-and
and benefits     employer-or        employer-or        government-
                 government-        government-        provided health
                 provided health    provided health    care, and other
                 care, or other     care, or other     government in-
                 in-kind benefits   in-kind benefits   kind benefits

                 Distribute
                 between top
                 shares using
                 proportionality

= Total cash     Preferred                             PI less
and in-kind      measure                               imputations and
income                                                 partially
                                                       adjusted for
                                                       retirement
                                                       income timing

Table 3. Population Coverage and the Unit of Analysis across Income
and Wealth Data Sets

                 Survey of
                 Consumer         Administrative
                 Finances            tax data

Unit of          Families            Tax units
  analysis
Coverage     Entire non-         Tax-filing
               institutional       population only
               population
             Corrects for low    Supplement
               participation       with
               at high end         information on
               using list          non-filers from
               sample              other data
             Excludes Forbes       sources
               400

                National
               Income and
                 Product          Financial
                Accounts          Accounts

Unit of         Aggregate         Aggregate
  analysis
Coverage     Households        Households
               and nonprofit     and nonprofit
               institutions      institutions
                               Possible to
                                 separate out
                                 nonprofit
                                 holdings of
                                 real estate