Emmanuel Saez · In O. Blanchard and D. Rodrik (eds.), Combating Inequality: Rethinking Government's Role, Cambridge MA: MIT Press, 2021, pp. 153-160

Should We Tax Wealth?

10 problems found: 1 major · 2 moderate · 7 minor

What the paper claims

US wealth concentration is high and has risen sharply since about 1980. Figure 16.1 shows four estimates of the top 0.1% share (SCF+Forbes 400; capitalization; revised capitalization; estate multiplier 'adjusted for accurate mortality differentials by wealth'). The share of the 'top 400 richest Americans' rose from 0.9% in 1982 to 3.3% in 2018. Wealth is more uncertain than income because there is no US wealth tax.
A well-enforced wealth tax has 'great revenue potential'. If the top 0.1% owns 20% of household wealth, which equals 100% of national income, then taxing 1% of their wealth raises 1% of national income. The potential base above $50 million (about the top 0.05% of families) was about $10 trillion in 2019. Note 2: the SCF (combined with the Forbes 400) and capitalized tax data 'generate approximately the same scoring ($9.4 and $10.9 trillion, respectively)'.
Saez and Zucman (2019a) 'shows that a low evasion rate of around 15% is possible with a well-designed and enforced wealth tax that learns from and hence avoids the mistakes of the European wealth tax experiences'. In an advanced economy 'the government can reliably obtain or create asset valuations'.
The US tax system (all taxes, all levels of government, relative to income 'consistent with national income') is mildly progressive below the top 0.01% but regressive at the very top. The top 400 richest Americans paid 'only 23%' in 2018, 'lower than for any other income group'. Two reasons are given: realization-based income taxation (Buffett: $65bn of wealth in 2015, about $3bn of economic income at 5%, $12m of reported income) and the 2018 corporate-tax cut.
Adding the initial Warren wealth tax (2% above $50m, 3% above $1bn; 'we assume an evasion rate of 15%') raises the top 0.01%'s rate 'by 14%' and doubles the top 400's rate from 23% to 46%. This 'would restore tax progressivity at the top to levels last observed in 1980'. Warren's 6% billionaire rate gives 68%. The Sanders tax gives 75%, 'equaling or even surpassing' mid-twentieth-century top effective rates.
Wealth concentration and oligarchy feed each other. Billionaires spend heavily on politics (Trump $66m in 2016; Bloomberg $900m and Steyer $270m in 2020; the Kochs and Adelson). Early-20th-century progressive income, estate and corporate taxation 'led to a large and sustained reduction in income and wealth concentration that reversed after tax progressivity went away'. A wealth tax is 'a potentially more powerful tool' than income, estate or corporate taxes to deconcentrate wealth.
Figure 16.3: had the Warren or Sanders tax been in place since 1982, the top 400's 2018 wealth share would have been 2.0% or 1.3% instead of 3.3%. That is still above 1982, and even the Sanders tax ('5%-6% of billionaires' wealth annually on average') 'is not even quite sufficient to beat back the surge'. Decabillionaires would not stay so for as long, and founders would lose control sooner. Concerns about the capital stock and entrepreneurship can be offset by how revenue is used and by earlier interventions.

Every problem found

1majorconfirmed · previously known · overclaiming

The chapter's most quoted claim, that in 2018 the 400 richest Americans paid a lower tax rate than any other group, depends on a gap of half a percentage point. That gap disappears or reverses with the authors' own updated figures, released before the chapter came out, or with the standard way of counting tax credits for low-income workers. What does hold is that the very richest pay noticeably less than the merely rich just below them.

Technical detail and evidence

Why it matters, in detail. Margin to the next-lowest group: -0.48 pp (2019 vintage). Like-for-like later vintages (income-ranked top 400, pre-tax income plus pure-KG denominator, full adult population): - top 400 minus bottom half: -1.9 pp (2019), +1.8 (September 2020), +3.7 (February 2022), +4.7 (2026); - top 400 minus all adults: -4.8, -1.1, +0.6, +0.6. Credits netted: four bottom deciles 2.0-4.9 pp below the top 400. Payroll add-back: P10-P50 0.2-1.0 pp below. The top 400 remain below the top 0.1% in every case (by 4.4-8.3 pp across vintages). Holding the Warren increment (+22.8 pp) fixed, a 27-29% base gives about 50-52%, so the 'doubling' survives as a near-doubling. The income-ranked group's income is about 10.7% of its wealth, against about 5.5% for the wealth-ranked Forbes 400 (BSYZ 2025). For the actual richest, the wealth-tax increment would be larger.

Evidence. Chapter p. 155 ('only 23%, lower than for any other income group (Saez and Zucman 2019b, chapter 7)'); Fig. 16.2 caption, p. 156. BPEA DataFig5 column C (C3:C17): P30-40 23.52%, top 400 23.04% (scripts/tax_rates_check.py; output/tax_rates_check.txt). Authors' 'taxrates' sheets, 2018 values (scripts/vintages_top400.py; output/vintages_top400.txt): | Vintage | Top 400 | All adults | Bottom 50% | Top 0.1% | |---|---|---|---|---| | TOI book file PSZ2019Distributions.xlsx | 23.04 | 27.82 | 24.97 | 31.35 | | September 2020 (PSZ2020AppendixTablesII(Distrib).xlsx) | 27.23 | 28.34 | 25.40 | 33.77 | | February 2022 | 29.31 | 28.74 | 25.65 | 33.72 | | SaezZucman2026Data.xlsx | 29.44 | 28.84 | 24.71 | 34.34 | The PSZ update log (September 2020) says: 'Distributional series for 2017, 2018 and 2019 are provisional and subject to revision.' Authors' simulator taxsimulator2.xlsx (eml.berkeley.edu/~saez; readme 'revised on 1/20/2020'), sheet calculator, col. AY 'refundable tax credits (EITC+refundable CTC)/income': -7.13, -6.14, -4.50, -2.47, -0.71 pp for P0-10 to P40-50, about $91.6bn in total; net rates 18.49, 18.11, 19.99, 21.05, 23.49% (scripts/sim_credits.py). Splinter (2020, NTJ appendix, Fig. B7; 2010 data): the bottom quintile falls by at least 17 points. TOI 2018 DINA micro-file usdina2018.dta: ssuicontrib is 6.5-7.5% of peinc for deciles 1-9. Adding it back lowers the TOI rates by 1.5-1.7 pp, e.g. P30-40 22.00%, P10-20 22.63% (scripts/dina2018_payroll.py). TOI online appendix pp. 6-8 (45% cap; income taxes scaled proportionally) and pp. 9-10 (pure gains = 40% of 2018 realized taxable gains). BPEA Table 3: 50.8%, 50.4%, 49.9% and 40.9% (the last for the Forbes 400). The flip occurs at r = 0.4734. TOI full2018 top-400 row: tax $105.2m, pre-tax income $416.0m, denominator $456.6m -> 23.04% vs 25.29%. Balkir, Saez, Yagan & Zucman (2025), NBER w34170: abstract, pp. 1-2 and fn. 10.

Why this severity. The ranking is the chapter's sharpest claim about the current system. It is reversed by the authors' own revision of the same series, which appeared before publication, and separately by the agency-standard treatment of credits and by a consistent payroll denominator. In the later vintages the level itself is 4-6 points (18-28%) higher. The qualitative finding that the very top pays less than those just below survives, so this is major rather than critical.

The authors' response. The authors' responses: - Refundable credits are transfers under national-accounts conventions (Zucman, as reported by Gleckman 2019). - The half-minimum-wage restriction avoids uninformative rates (TOI appendix p. 5). - The 45% ratio is a triangulation that 'could be refined' (TOI appendix p. 8). - Pure gains are added because taxes on realized gains are in the numerator (TOI appendix pp. 8-10). The September 2020 update flags 2017-2019 as provisional. BSYZ (2025, p. 3) describe their 23.8% as giving 'similar bottom-line results' to the 2019 estimate. No acknowledgment of the higher 2018 top-400 rates in the 2020-2026 releases was found.

2moderatelikely · previously known · selective presentation

The chapter's figures for how much wealth the very richest hold, and so how much a wealth tax could raise, come from the high end of the available estimates. Other careful estimates, including one reported in the authors' own paper, put the taxable wealth up to about a third lower. That would shrink the revenue but not make it small.

Technical detail and evidence

Why it matters, in detail. Revenue per 1% levy = share x (household wealth / NI): 20% x 5.03 = 1.0% of NI. With the authors' revised 17.8% it is 0.90%; with SZZ's 15.7% it is 0.79%. Base above $50m, 2019 (families): $10.9tn (capitalization), $9.7tn (SCF+Forbes), $9.4tn (revised), $6.8tn (estates, after the authors' own mortality correction). Scaling the authors' series by SZZ's 2016 base ratio (0.70) gives $6.6-7.6tn. The SCF+Forbes leg depends on Forbes valuations. Estate returns of Forbes decedents report about 30-50% of the Forbes figure, which the authors read as estate-tax avoidance and critics as Forbes error. The trend survives in every series, but the 'adjusted' estate series supplies little independent support: its raw rise is +1.2 pp, against +8.3 pp after adjustment.

Evidence. Chapter p. 154 (Fig. 16.1 caption; 'If the top 0.1% owns 20%...'; '$10 trillion') and p. 158 n. 2. Note 2 was recovered by the adjudicator from the print-edition Google Books volume snOdEAAAQBAJ (final/chapter_p158_recovered.md). BPEA Table 2, SZ2019BPEAfinal.xlsx sheet Table2: H8 = 10.885 (capitalization), H13 = 9.401 (revised), H18 = 9.655 (SCF+Forbes), H23 = 6.767 (estates). The BPEA text says the estate base is 'about a third lower'. DataFig2, 2016 (top 0.1% share): - B = 19.61% (capitalization); - D = 19.35% (SCF+Forbes); - E = 17.83% (revised); - M = 16.37% (interest correction); - N = 13.88% (AAA/SZZ-type). The adjusted estate series (J) is 16.10% in 2012. Decomposition of the J rise, 1981-2012 (+8.33 pp): raw +1.20, smoothing +1.34, mortality +5.34 (64%), unit conversion +0.45. Mortality factor H goes from 1.0 (1981) to 1.9 (2012) (scripts/fig161_check.py; output/fig161_check.txt). BPEA pp. 455-457 describe this assumption and its caveat. Smith, Zidar & Zwick (2023), QJE 138(1), Table I, p. 551: top 0.1% 15.7% (baseline) vs 18.4% (Revised SZ) vs 20.4% (equal returns), 2016. Local Pareto interpolation gives a base above $50m of $4.43tn vs $6.33tn (-30%) (scripts/szz_base.py). Fed DFA (2026Q2 vintage): top 0.1% of households 12.5% (2016Q4) and 12.4% (2019Q4); the unit and wealth concept differ. Summers (2021), ch. 15 of the same volume, p. 142: SZZ's work 'suggests that the Saez and Zucman estimates of wealth share of the top 0.1% should be revised downward by 25%'. Raub-Johnson-Newcomb (2010) and Balkir et al. (2025): estate values of Forbes decedents are about 30-50% of Forbes estimates (us-wealth-taxes-11).

Why this severity. Revenue potential scales with the top share, and more than proportionally against a fixed $50m threshold. Credible alternatives lower the illustrative 1%-of-NI yield to about 0.8-0.9% and the base by up to about a third (6-35%). A large base remains under every estimate and the rise in concentration is common to all series, so the qualitative claims survive. This is a noticeable, not a transformative, change.

The authors' response. The chapter notes 'more uncertainly [sic] about the distribution of wealth and its trend than for income' (p. 154). The authors argue: - Estate-based estimates are biased down by estate-tax avoidance (BPEA 2019; response to Summers & Sarin, June 2019). - The AAA yields used by SZZ overstate top interest rates (BPEA fn. 21). - Financial Accounts private-business values are too low (SZ 2020, NBER w27921). SZ (2022, NBER w30396) say the remaining gap is concentrated in the top 0.01% and 'reconcile' the series after correcting billionaire equity. SZZ dispute this.

3moderatelikely · previously known · speculative assumption

How much a wealth tax raises depends on how much wealth the rich manage to hide or shelter. The chapter assumes only 15% and says research shows this is achievable. In fact it is an informed guess from a few foreign studies, and the same 15% is used even for much higher tax rates, where hiding pays off more. With plausible higher avoidance, the revenue and the increase in billionaires' tax rates are noticeably smaller, though still large.

Technical detail and evidence

Why it matters, in detail. Taxable base = (1-h) x true wealth, with the threshold fixed. The authors' calibration h = 2% x mean semi-elasticity (8) = 16% is linear in the tax rate, but it is applied as a constant. At marginal rates of 3%, 6% and 8%, the same rule gives h = 24%, 48% and 64%. On the authors' own TD1 inputs: - Warren top-400 rate: 43.5-43.9% (vs 45.9%); - Warren 6%: 49.9-51.4% (vs 66.9% reproduced, 68% printed); - Sanders: 45-57% (vs 74.8%). Flat alternatives grounded in the literature (h = 23-33%) lower 2019 Warren revenue from $203bn to $146-177bn. The authors' 50% weak-enforcement case gives $96bn. None of these estimates is validated for a US tax. Critics' alternatives are judgment-based too, and part of the Swiss response (mobility, house prices) is less relevant to a federal tax.

Evidence. Chapter pp. 154-155 ('shows that a low evasion rate of around 15% is possible'; 'we assume an evasion rate of 15%'). BPEA DataFig5 headers D2/F2: '(& 15% avoidance)' for Warren and Sanders. Saez-Zucman letter to Sen. Warren (18 Jan 2019), p. 2 and fn. 2: '2%*(.5+.5+2.5+28.5)/4=16%'; '...to be on the conservative side... 15%... This is a large response in light of existing estimates'. SZ Sanders note (Sept 2019): 'an evasion tax rate of 16%' with rates up to 8%. Brülhart, Gruber, Krapf & Schmidheiny (2022), AEJ:EP 14(4), abstract: '...raises reported taxable wealth by at least 43 percent after 6 years'. Of this, 24% is mobility and 21% house prices. PWBM (12 Dec 2019), Table A1 and text: -13 vs the Warren estimate's -8; the Warren estimate 'appears to apply the elasticity to a rate of 2 percent for all taxpayers' (citing Ghenis). Kopczuk (2019, BPEA comment): 'optimistic empirical assumptions that do not highlight uncertainty, which is likely to run mostly in one direction'. BPEA Table 2 rows 28-30. Summers (2021, ch. 15, n. 5, p. 149): the four studies give 'between 0 and 34%'. Adjudicator re-implementation of wealth.do on SZ's micro-file (scripts/revenue_sensitivity.py). 2019 Warren revenue: - $203.5bn at 15%; - $177.2bn at 23% (Brülhart 43 in the authors' average), $167.6bn at 26% (PWBM -13 x 2%), $145.8bn at 33%, $96.1bn at 50%; i.e. -13%, -18%, -28%, -53%. With rate-scaled avoidance (8 x marginal rate), revenue changes by -6% (Warren), -27% (6% variant) and -34% (Sanders). Top-400 rates under the same alternatives (scripts/tax_rates_check.py): - Warren: 43.5-43.9% (rate-scaled), 41.0% (flat 33%); - 6% variant: 49.9-51.4%, against the chapter's 68%; - Sanders: 45-57% (rate-scaled), 63.8% (flat 33%), against 75%.

Why this severity. The assumption drives every quantitative wealth-tax result. Defensible alternatives at the 2% rate cut Warren revenue by 13-28% and the Warren top-400 rate from 46% to 41-44%: noticeable, but a large effect remains. The 68% and 75% illustrations are more fragile, falling to about 50% and 45-64% under rate-scaled or higher avoidance. They are secondary to the qualitative case, which survives.

The authors' response. The authors argue that avoidance is a policy choice. The Warren design (comprehensive base with no exemptions, third-party reporting, minimum audit rates, exit tax) differs from the European taxes. In their response to Summers & Sarin (June 2019) they say 80% of top-0.1% wealth is in assets with market prices, and: 'Our estimate is not a "best case" scenario but is itself a middle-ground... it is possible that avoidance would be below 15%'. The Warren letter (fn. 2) attributes the large Swiss response to bank secrecy and cantonal mobility. No response to the rate-scaling point (PWBM/Ghenis) was located.

4minorconfirmed · previously known · internal inconsistency

The chart showing how much a wealth tax would have shrunk the biggest fortunes assumes nobody avoids the tax, while the rest of the chapter assumes 15% avoidance. Using the chapter's own 15% makes the effect somewhat smaller, but the picture does not change.

Technical detail and evidence

Why it matters, in detail. The cumulative log effect scales by about (1-h). For h = 0.15: Warren 2018 share 1.96% -> about 2.11%; Sanders 1.34% -> about 1.53% (actual 3.26%; 1982 0.93%). Behavioural erosion of the pre-tax trajectory (Jakobsen et al. 2020) would cut either way: lower pre-tax growth deconcentrates more, while avoidance deconcentrates less in reported terms. The figure is an illustration, not a forecast.

Evidence. BPEA DataFig6, A2: 'Assumes zero evasion rate'. 2018 values: B44 3.261%, C44 1.955%, D44 1.337%; 1982: 0.934%. Chapter p. 157 (description of the calculation) and p. 158 (2.0% and 1.3%; 'still higher in 2018 than ... in 1982'; Sanders 'not even quite sufficient'). Kopczuk (2019, BPEA comment, fn. 3): the simulations consider the 'mechanical' case. With 15% evasion (log-scaling of the cumulative effect, scripts/tax_rates_check.py): about 2.11% (Warren) and 1.53% (Sanders).

Why this severity. At the chapter's own 15% rate the deconcentration effect shrinks by about 12-15%: the 2018 reduction falls from 1.31 to 1.15 points (Warren) and from 1.92 to 1.73 points (Sanders). The chapter's own tempered conclusions hold even more strongly, and the mechanics are disclosed.

The authors' response. The chapter describes the calculation as mechanical ('uses the existing individual wealth trajectories'). No response on the evasion assumption was located.

5minorconfirmed · newly identified here · arithmetic or transcription

The chapter says one group's tax rate would rise 'by 14%' when it means 14 percentage points, a much bigger increase (from about 30% to 44%). It is a wording slip; the underlying numbers are correct.

Technical detail and evidence

Why it matters, in detail. +13.95 pp (relative +46.0%) for P99.99-top400; +16.04 pp for the inclusive top 0.01%. Read literally as a relative change, '14%' understates the effect about threefold.

Evidence. Chapter p. 155: 'would rise by 14 %', identical in the ebook (NBkSEAAAQBAJ) and print-edition (snOdEAAAQBAJ) Google Books indexes. BPEA DataFig5 C16 = 30.36%, D16 = 44.32%. BPEA p. 468 ('would rise by 14 points'). Inclusive top 0.01%, weighted by TOI TD1 incomes: 28.64% -> 44.68% (Auditor B; recomputed).

Why this severity. A units slip in the text. The figure and the source numbers are right, so no conclusion changes.

6minorconfirmed · newly identified here · internal inconsistency

A small factual slip about Warren Buffett's 2015 fortune ($65bn rather than about $62bn). It does not change the point that his taxable income was tiny compared with what his wealth earned.

Technical detail and evidence

Why it matters, in detail. Implied economic income at 5%: $3.10bn instead of $3.25bn. The ratio of reported to economic income stays below 0.4%.

Evidence. Chapter p. 155. Forbes 2016 Forbes 400 press release (Wayback copy in cache). BPEA fn. 36: '$12 million in 2015 when his wealth was $62 billion'. TOI (2019) ch. 7: 'According to Forbes, Buffett owned $65.3 billion in wealth in 2015'.

Why this severity. This is a likely year slip that does not affect the illustration: 5% of $62bn is $3.1bn, still 'around $3 billion' against $12m of reported income.

7minorconfirmed · newly identified here · arithmetic or transcription

A footnote attaches one of the paper's estimates to the wrong data source. The numbers are close, so nothing important changes.

Technical detail and evidence

Why it matters, in detail. SCF+Forbes base above $50m = $9.655tn, not $9.4tn; the revised capitalization series is $9.401tn. The range of the three non-estate series is $9.4-10.9tn either way.

Evidence. Chapter p. 158, note 2. This text was recovered by the adjudicator from the print-edition Google Books index (final/chapter_p158_recovered.md); both auditors lacked p. 158. SZ2019BPEAfinal.xlsx Table2: H13 = 9.401 ('Revised capitalized incomes'), H18 = 9.655 ('SCF+Forbes 400'), H8 = 10.885. BPEA text: '$10.9 trillion ... $9.4 trillion in the revised capitalized income series, and $9.7 trillion in the SCF'. These values are identical in the September 2019, October 2019 and final versions.

Why this severity. A $0.3tn mislabelling with no effect on the 'about $10 trillion' conclusion.

8minorconfirmed · newly identified here · other

A one-word typo; it does not affect anything the chapter argues.

Technical detail and evidence

Why it matters, in detail. None.

Evidence. Chapter p. 154, identical in both Google Books volumes (the query 'uncertainly' returns only p. 154).

Why this severity. A trivial typo with no effect on content.

9minorlikely · newly identified here · documentation gap

The 1982 starting point is for the richest roughly 220 families, scaled to the smaller population then, not the actual 400 richest. It makes the rise look slightly bigger than it was for the literal top 400; the conclusion that their share more than tripled is unaffected.

Technical detail and evidence

Why it matters, in detail. The 1982 group is about 220 families with $83bn (0.93%), against the 400 families with $92bn (1.03%) on the same denominator. The 2018 endpoint (3.26%) is unaffected.

Evidence. BPEA p. 500. DataFig6 B8 = 0.934%, i.e. about $83bn on the authors' 1982 household wealth of $8.90tn (DataFig1: 3.14 x $2,834bn). Kaplan & Rauh (2013, AEA conference paper, p. 6): the Forbes 400 held $92bn in 1982. BSYZ (2025, fn. 1): the same fixed fractile is 221 tax units in 1982. Chapter pp. 154 and 157-158; p. 153 not recovered.

Why this severity. This is a labelling issue only. The fixed-fractile comparison is defensible, arguably preferable, and the rise is 3.2-fold for the literal Forbes 400 against 3.5-fold as shown.

The authors' response. Disclosed in BPEA 2019 (p. 500) but not, as far as recovered, in the chapter.

10minorplausible · previously known · overclaiming

The chapter states that progressive taxes caused the fall in inequality in the mid-twentieth century, and that a wealth tax would not hurt investment or entrepreneurship. Both are reasonable arguments, but the chapter asserts them rather than demonstrating them.

Technical detail and evidence

Why it matters, in detail. The causal attribution of the mid-century compression to tax progressivity is not identified. The magnitude of the compression is about one-third smaller in Geloso et al. (2022), and the post-1980 decline in top rates is about half as large under CBO-style incidence. The efficiency claims rely on a citation that does not test incentive effects of a wealth tax.

Evidence. Chapter p. 157 (history; 'more powerful tool') and p. 158 ('Other Economic Effects'; notes 3-4 cite Acemoglu & Robinson 2012, Piketty 2020, Mayer 2017, and Page, Seawright & Lacombe 2018). Geloso, Magness, Moore & Schlosser (2022, EJ): the PS overstate the levelling by about a third (us-income-07). Splinter (2020, NTJ): federal progressivity rose after 1979 under CBO conventions (us-wealth-taxes-16). The size of the decline in top rates depends on the shareholder-only incidence choice (us-wealth-taxes-13). Kopczuk (2019) and Jakobsen et al. (2020) on real responses.

Why this severity. These are supporting arguments in a policy essay. They do not enter any of the chapter's numbers, so the quantitative findings are unaffected.

The authors' response. The chapter cites Saez and Zucman (2019b, ch. 7), Acemoglu & Robinson (2012), Piketty (2020) and Bell et al. (2019). SZ (2023/2026, 'Distributional Tax Analysis ...') defend the shareholder-incidence convention.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

Excluding 'pure capital gains' from the denominator raises the top-400 rate to 25.8%, so the top 400 is no longer the lowest group (Auditor A, I3). — Partly rejected; retained only as a disclosure point inside F3. - The authors' own top-400 row (TOI full2018: tax $105.2m, pre-tax income $416.0m) gives 25.29%, not 25.8%. The auditor's 0.40 x realized-gains shortcut overstates pure gains (10.7% vs 8.9% of the denominator). - Dropping pure gains while keeping the taxes paid on them is the PSZ-2018 convention that the authors reject because it makes top rates spike in high-realization years. - The internally consistent variant also removes those taxes; Balkir et al. (2025, fn. 10) report it leaves rates 'virtually unchanged'. It gives 22.4-23.3%, still below P30-40 (scripts/tax_rates_check.py). This convention alone does not overturn the ranking, although other conventions do (F3).
The chapter misdescribes the Fig. 16.2 denominator as 'consistent with national income in national accounts' when it includes pure capital gains. — Rejected as immaterial. The denominator is national income plus pure gains: 2.05% of NI in 2018 (TOI Table A3), concentrated at the top. 'A comprehensive basis consistent with national income' is a fair summary, and the treatment is documented in the TOI appendix. BSYZ (2025) describe the same concept as 'slightly broader than basic-price national income'. The resulting rate difference is reported in F3.
The 45% fiscal-to-economic income cap for the top 400 is a favourable choice because three of the authors' four triangulation sources give 50-51%. — Partly rejected. The three sources at 50-51% concern broader groups (estates above $100m, SCF top 0.001%/0.01%). The only top-400-specific source gives 41% (BPEA Table 3). The authors' later administrative study finds fiscal income of about 40% of economic income for the top 400 (BSYZ 2025), which would lower, not raise, the rate. The 47.3% flip threshold is kept in F3 only as a sensitivity.
Figure 16.1's 'estate multiplier (adjusted)' series is 'built to reconcile' with the capitalization estimates. — Partly rejected. The decomposition is correct (64% of the 1981-2012 rise comes from the mortality path) and the series is not independent corroboration (kept in F1). But the correction rests on evidence (Chetty et al. 2016) and is disclosed with caveats in the BPEA (pp. 455-457), so intent to reconcile is not shown.
Figure 16.3 is presented as illustrating 'the power of the wealth tax' without qualification (the overclaiming element of Auditor A's I5). — Rejected. Page 158, unavailable to both auditors and recovered here, qualifies the result: the top-400 share 'would still be higher in 2018 than it was in 1982', and even the Sanders tax 'is not even quite sufficient to beat back the surge in billionaires' wealth'. The zero-evasion inconsistency is kept as F4.
The '14%' wording is only 'likely' because a Google Books OCR/indexing substitution cannot be excluded. — Rejected. The identical wording ('would rise by 14 %') appears in the separately indexed print-edition volume (snOdEAAAQBAJ) as well as the ebook. The slip is confirmed (F5).
No concentration-level issue should be retained: Figure 16.1 shows alternative estimators and p. 154 acknowledges uncertainty. — Overturned. The recovered note 2 (p. 158) cites only the two concordant base estimates. It omits the estate-based $6.8tn that the same BPEA table and text report as 'about a third lower'. The chapter's recovered references cite none of the lower estimates or critiques (SZZ, Kopczuk, Summers-Sarin). Retained as F1 (moderate).
On tax and income concepts, no chapter error can be asserted without full notes and a matched alternative calculation. — Overturned. Matched alternatives exist in the authors' own files: the September 2020 'taxrates' release, the January 2020 simulator's refundable-credit column, and the 2018 DINA micro-file's contribution variable. Each reverses the 'lower than any other group' ranking (F3).

Audit notes

What reproduced. Reproduced: - Every checkable number reproduces from the authors' 2019 files: top-400 23.04% -> 45.88% / 74.80%; +13.95 pp for P99.99-top400; Forbes 400 0.934% -> 3.261%; counterfactual 1.955% / 1.337%; Fig. 16.1 series; household wealth 5.03 x NI; base above $50m $9.4-10.9tn (BPEA Table 2) and $11.40tn / $9.07tn before/after 15% evasion on the micro-file. - An independent Python port of wealth.do reproduces Warren revenue of $203.5bn (15%) and Sanders revenue of $334.5bn (16%). - Approximate only: the 68% (66.9% with the authors' method and inputs) and Sanders' 75% (74.1% via Pareto interpolation; the bracket-level micro data for the top 400 are not public). Not reproducible: - Fig. 16.3 cannot be replicated exactly: the 1982-2018 Forbes micro-data and forbes400tax.do are not public. Sensitivities computed here: - Revenue under 23-50% evasion and rate-scaled evasion. - Top-400 rates under alternative evasion, fiscal-income ratio and pure-gains treatments. - The authors' later 'taxrates' vintages (27.23%, 29.31%, 29.44% for 2018). - The authors' simulator with refundable credits netted (bottom four deciles 18.1-21.1%). - The DINA micro-file payroll add-back (P10-50 at 22.0-22.8%). - A Pareto comparison of the SZZ and SZ20 bases above $50m (-30%). - The estate-series decomposition (64% from the mortality path). All scripts are in final/scripts and their outputs in final/output.

Limits of this audit. Text coverage: - The full chapter was not obtained (MIT Press Direct is paywalled and Cloudflare-blocked; Unpaywall/OpenAlex mark it closed). Text comes from Google Books search snippets. - Page 153 (the opening) is not indexed in either volume and was never seen. The book index shows it discusses the policy debate, Warren and Sanders, Forbes and tax evasion, so caveats there cannot be excluded. - Pages 154-158 are reconstructed from overlapping snippets; the order of fragments within a page is mostly but not fully certain. Data and computations: - The later-vintage and simulator figures come from authors' files in the shared cache (gabriel-zucman.eu US DINA archive; eml.berkeley.edu/~saez/taxsimulator2.xlsx); they were checked here, not re-downloaded. - The DINA payroll and credit computations are approximate: decile cuts were rebuilt on the public micro-file rather than the authors' internal files. - The Fig. 16.3 evasion adjustment is a log-scaling approximation. Sensitivities assume proportional scaling of income taxes with fiscal income, and static behaviour. Not re-verified here: - The final FEC campaign totals and the Raub et al. (2010) estate-Forbes ratio (taken from the corpus and BSYZ 2025).