Gabriel Zucman · Annual Review of Economics 11: 109-138

Global Wealth Inequality

14 problems found: 1 major · 5 moderate · 8 minor

What the paper claims

US: surveys (SCF) and capitalized income tax data both show a top-1% wealth share of about 40% in 2016 versus 25-30% in the 1980s; the 2016 levels are 'virtually identical' and the rises since 1989 similar.
US very top: top-0.1% share back near 1920s levels; Forbes top-0.00025% share x4 since the early 1980s, in line with the tripling of the capitalized top 0.01% and with a 'strikingly similar' year-to-year evolution; capitalization bias 'not very large'.
Global proxy (China + Europe + US): the top-1% wealth share rose while the bottom-75% share hovered around 10%; the top 10% owns >70%, the bottom 50% <2%, the middle 40% <30%.
Global wealth growth incidence 1987-2017: fast growth at the bottom, squeezed middle, fastest growth at the top.
Country trends: top-1% shares almost doubled in China and Russia; more moderate rises in France and the UK.
Offshore wealth is about 10% of world GDP and highly concentrated; accounting for it raises top wealth shares.
Traditional survey and tax data likely underestimate the level and rise of wealth concentration; in Norway the offshore correction 'erases almost half' of the decline in the top-0.1% share since the 1930s; Norway's effect is likely a lower bound for most countries.

Every problem found

1majorconfirmed · newly identified here · methodology

The paper's 'world' wealth figures treat hundreds of millions of people in Eastern Europe and the former Soviet Union as if they were as rich as people in Britain, France and Spain, and convert currencies in a way the paper never mentions. With reasonable alternatives, the rise in the richest 1%'s share shrinks from about 5 points to between about 1 and 4 points, and the poorer three-quarters' share rises instead of staying flat. The richest 1%'s share still goes up in every version.

Technical detail and evidence

Why it matters, in detail. The pooled distribution is a population-weighted mixture in which ~250m Eastern European/Soviet adults carry UK/FR/ES distributions at Western levels, inflating European wealth by roughly 60-85% and the global upper-middle. Correcting composition at MER (Western Europe only, or Eastern Europe at 7-30% of the Western per-adult level) lowers the 1980-2016 rise from +4.8 to +1.8-3.6 pp and raises both endpoints by 3-6 pp; PPP conversion (the same team's later choice for whole-distribution analysis) lowers it to +2.8 pp and makes the bottom-75% share rise from 9.6% to 14.3%; combined fixes give about +1 pp. Sign robust, magnitude and bottom-75% dynamics not.

Evidence. Code lines 20-38, 61-65; WIR 2018 population inputs (QE = QX + QM; QM includes SU until 1990; SU 1980 = 175.8m adults); adjudicator re-implementation (scripts/global_merge.py) reproduces published top-1% shares within 0.09 pp and bottom-75% shares within 0.01 pp in every year 1980-2016, incl. the 1991 break, and the 2016 top 0.1%/0.01%/middle 40% (15.5/7.8/28.8%). Sensitivity (scripts/outputs/global_scenarios.txt), top 1% 1980->2016 and bottom 75%: as coded 27.9->32.7 (+4.8), 8.7->10.9; consistent QE +4.4; Western Europe only 32.2->35.8 (+3.6), 3.9->9.4; Western + Eastern at 30%/20%/7->13% of Western level +3.2/+2.9/+1.8 (34.4->36.2; bottom 75% 3.7->9.6 for the last); PPP 27.4->30.2 (+2.8), 9.6->14.3; PPP + Western only 31.5->32.3 (+0.9), 5.1->13.3; PPP + Eastern at 30% +1.0. Donor weights matter little (equal +5.2, population-proportional +5.4, no UK rescaling +4.8). Growth-incidence shape survives all alternatives (outputs/gic.txt). Blanchet (2017, WID TN 2017/7) discloses MER and 60/20/20 weights; WIR 2018 data note says 'a world represented by China, the EU and the US'; WIR 2022 ch. 4 uses PPP for global wealth 'because we focus both on the bottom and top'. AR pp.109, 111, 124, 126-128.

Why this severity. Under defensible alternatives the published +4.7-point rise of the global top-1% share shrinks by about 25-80%, the 1980 level is understated by 4-6 points, and the 'bottom 75% hovered around 10%' description fails (it rises 4-8 points whenever Europe's composition or the currency conversion is changed); the rise itself survives in every scenario, so the core finding is materially weakened but not overturned.

The authors' response. None specific. The AR calls the estimates 'tentative' and 'imperfect, provisional, and subject to revision' (pp.124, 126); Blanchet (2017) discloses MER and the 60/20/20 donor weights and asserts the weighting has 'a limited impact on the overall trend'; WIR 2022 later reported global wealth at PPP for whole-distribution analysis.

2moderateconfirmed · previously known · methodology

The tax-based US estimate assumes that rich and ordinary families earn the same interest rate on their savings. When researchers - including the author in later updates - let the rich earn more, the estimated rise in the top 1%'s share since 1989 is about a quarter to a third smaller, and up to half smaller for the top 0.1%. The headline that the top 1% owns about 40% still holds, and survey data confirm a large rise.

Technical detail and evidence

Why it matters, in detail. Equal yields make the fixed-income capitalization factor (~1/r) explode after 2008; allowing top yields above the average lowers the capitalized 1989-2016 top-1% rise from +11.3 pp (tax units) to +7.7-8.5 pp in the authors' own 2020/2022 vintages and to +6.6 pp (equal-split) in SZZ; the top-0.1% rise falls by 22-46%. The 2016 tax-unit level (38.1-38.9%) and the SCF trend are unaffected, so the 'both sources show a large rise' conclusion holds while precise magnitudes and very-top dynamics do not.

Evidence. Top 1%, 1989->2016: AR vintage tax units 27.56->38.89 (+11.3; DataF2); authors' Sept-2020 update 30.30->38.79 (+8.5; SZ 2020 JEP data workbook); PSZ Feb-2022 update 30.35->38.10 (+7.75; Table E10). Equal-split adults: AR 26.57->36.90 (+10.3) vs PSZ 2022 28.59->35.52 (+6.9). SZZ (2023) supplemental TotalWealthShare.xlsx: baseline 27.11->33.70 (+6.59) vs equal returns 28.74->38.91 (+10.17), -35%; top 0.1% +4.63 vs +8.60 (-46%). BHH (2018, FEDS 2018-024, abstract): 2011 top 1% 40.5% -> 33.9%. SZ (2016, pp.549-550): estate interest premium lowers the 2012 top 0.1% from 22.0% to 18.7%. SCF microdata (independent of capitalization): +8.6 pp for households, +10.5 pp for tax-unit-sized groups (scripts/outputs/scf_units.txt).

Why this severity. The core US level (~40% in 2016 vs 25-30% in the 1980s) survives in the authors' later vintages and the SCF independently shows a ~9-point rise, so C1 stands; but the capitalized rise is uncertain by about a quarter to a third, and the very-top (top 0.1%) rise by up to a half.

The authors' response. Saez & Zucman (BPEA 2019; JEP 2020) adopted higher interest rates for the rich in revised series; Saez & Zucman (2022, NBER WP 30396) argue SZZ underestimate billionaire equity (x2.1), top interest-bearing assets (x1.6) and business wealth and that, once corrected, the estimates 'are reconciled'; SZZ dispute these counter-corrections.

3moderateconfirmed · previously known · misrepresentation of source

To defend its tax-based numbers, the paper cites studies as saying things they do not say - for example, that the richest earn lower interest rates, or that the two US data sources agree within statistical error - and leaves out evidence, including the author's own earlier finding, that the richest earn higher interest rates. Read correctly, that evidence suggests the tax-based method overstated the rise at the very top in the 2000s.

Technical detail and evidence

Why it matters, in detail. Correctly read, the evidence cited points the other way for fixed income: estate data show a 1.6x top interest premium after 2008 (top 0.1% 22.0 -> 18.7% in 2012), SCF top-1% fixed-income yields are several times the average, and the only confidence-interval reconciliation of the top 0.1% requires re-capitalizing top interest (top 0.1% ~19% instead of ~22%). The window chosen to rebut Kopczuk (2010-16) avoids the 2001-10 period in which the capitalized series rose 6.6 pp vs 1.9 pp in the SCF. The defence of equal-returns capitalization at the very top is therefore much weaker than presented.

Evidence. Saez & Zucman (BPEA 2019, p.465 and fn 20) describe Bourne et al. fig. 4 as reported capital income = 3% of 2007 wealth for estates above $100m; IRS manuscript (estatestaxbourne.pdf) fig. 4 = capital gains share of capital income. SZ (2016, pp.566-567): 'After correcting the SCF, there remains a residual gap' (adjusted top 0.1% 17.2% in 2013). BHKS (2016): top 0.1% 'all but the most recent periods are reconciled' (p.295); within the confidence interval only with 10-year-Treasury capitalization, 'just under 19 percent' (p.298); fn 48: SCF 2013 fixed-income return ~1% overall vs almost 6% for the top 1%. SZ (2016, pp.549-550): premium 1.6x; 2012 top 0.1% 22.0% -> 18.7%. DataF2: 2001-10 capitalized +6.6 pp vs SCF +1.9 pp; 2010-16 -0.3 vs +4.3. BHH (FEDS 2018-024, March 2018): 40.5% -> 33.9% (2011); absent from AR references. Fagereng et al. (2016): capitalization 'can lead to misleading conclusions about the level and the dynamics' (Norwegian top-5% share overstated 5.4 pp on average; top-1%/0.1% bias of ambiguous sign).

Why this severity. These points leave the US 2016 level unchanged, but they remove the cited support for dismissing the fixed-income critique and for the claimed reconciliation at the very top, and every one leans in the paper's favour.

The authors' response. None to these specific points; SZ (2016) had disclosed the estate premium, and SZ (BPEA 2019) later described Bourne et al. as showing low reported capital income relative to wealth and adopted higher interest rates for the rich.

4moderateconfirmed · previously known · methodology

The paper says the survey and the tax records give 'virtually identical' figures, but it compares a group of 1.26 million households with a group of 1.7 million tax-filing units. Compared on an equal footing, the survey shows the top group owning more - about 43% - so the conclusion that wealth is highly concentrated is not hurt, but the claimed precise agreement is overstated.

Technical detail and evidence

Why it matters, in detail. Ranking the same number of units in the SCF raises its top-1% share by ~12% (38.6 -> 43.3% in 2016; 30.0 -> 32.9% in 1989), 4-5 pp above the capitalized estimate; the AR's level agreement arises from comparing a smaller SCF group on SCF totals with a larger tax-unit group on Financial Accounts totals. The trend agreement (+10.5 vs +11.3 pp) and the high level are unaffected.

Evidence. Adjudicator SCF microdata (rscfp2016, SCFP1989; 5 implicates): households top 1% 38.55% (2016), 29.97% (1989); top 1.70m/1.17m households (= 1% of tax units, from the author's CompF3 col. G) 43.33% / 32.85%; rise +10.48 vs capitalized +11.33 (scripts/outputs/scf_units.txt; Auditor A identical). AR p.121: $17.6m x 1.70m = $29.9tn vs $26.8m x 1.26m = $33.8tn (+$2.4tn Forbes). CompF2 cols. [2]-[3] blank. SZ (2020 JEP, pp.8-9): DFA household top 1% x1.08 -> tax units; 38% in both series in 2018 on Financial Accounts concepts.

Why this severity. The paper's headline demonstration that surveys and tax data agree is not like-for-like, an important undisclosed caveat; but correcting it raises rather than lowers the survey figure, and a later concept-and-unit-harmonized comparison by the authors finds agreement, so the ~40% conclusion is not weakened.

The authors' response. Saez & Zucman (2016, Table C4b) performed these adjustments; Saez & Zucman (2020, JEP) applied a household-to-tax-unit factor of 1.08 when comparing with the Fed DFA and found agreement at 38% in 2018.

5moderateconfirmed · previously known · overclaiming

The paper says the Forbes rich list and the tax-based estimate move together 'strikingly' from year to year; in the paper's own data they move in opposite directions about half the time. They agree only over the whole 35-year period, which supports a big long-run rise but not its timing.

Technical detail and evidence

Why it matters, in detail. With r = 0.11 on annual log changes and opposite movements in 2000-06 and 2010-16, Forbes validates only the 35-year multiple (x3.95 vs x3.42); it cannot validate the timing or the fixed-income-driven 2000-2012 rise of the capitalized top 0.01% (x1.59 vs Forbes x1.04).

Evidence. DataF3 (Fig. 3 data), 1982-2016: annual log-change correlation r = 0.11; same sign in 17 of 34 years; 3-year overlapping changes r = 0.54; 2000-06 Forbes x0.73 vs top 0.01% x1.11; 2010-16 x1.17 vs x0.97; 2000-12 x1.04 vs x1.59 (x1.22 in the authors' 2022 vintage, per Auditor A); 1982-2016 x3.95 vs x3.42 (scripts/outputs/forbes_vs_top001.txt).

Why this severity. The long-run agreement holds, so the headline is not affected; but the stated year-to-year agreement is false and Forbes gives no support to the 2000-2012 surge that is most in dispute.

6moderatelikely · previously known · speculative assumption

The paper assumes that money hidden in tax havens is owned mostly by the very richest everywhere, as in Scandinavia, and therefore calls the Norwegian effect a likely minimum for most countries. A later Dutch study found hidden money there was spread among the merely well-off, so the effect could be smaller in some countries and larger in others.

Technical detail and evidence

Why it matters, in detail. The corrected share is (s + q*h)/(1 + h); the 'lower bound' inference scales by larger aggregate offshore stocks h while holding the Scandinavian q fixed. With Dutch q (~7% for the top 0.01%) the top-0.1% correction is +0.6 pp vs +1.4 pp in Norway, while Colombia shows +3.2 pp, so the cross-country correction can be smaller or larger than Norway's; 'lower bound for most countries' is unproven.

Evidence. AJZ (2018) T.A9b: top 1%/0.1%/0.01% allocation 93.6/76.9/51.6%. Leenders, Lejour, Rabate & van 't Riet (2023, JPubE 217:104785, Sec. 4.2): Dutch top 0.01% own 7% of amnesty wealth (P90-99.9: 67%); top-0.1% share 8.7% -> 9.3% vs Norway 8.4% -> 9.8%; Colombia 15.9% -> 19.1%. Schmitz et al. (2021) on non-household holders of the asset-liability gap (offshore-havens-03). AR pp.131, 134.

Why this severity. Affects a secondary, explicitly hedged claim and the country corrections in Fig. 8; the point that recorded top shares omit offshore wealth stands.

The authors' response. The transfer assumption is disclosed in the AR (p.131) and presented by Alstadsaeter, Johannesen & Zucman (2018) as an approximation; Colombian evidence (Londono-Velez & Avila-Mahecha) points to larger corrections.

7minorconfirmed · newly identified here · arithmetic or transcription

The paper says counting hidden wealth wipes out 'almost half' of Norway's long-run fall in top wealth; its own numbers show about one third. The broader point - that the fall looks smaller once hidden wealth is counted - still holds.

Technical detail and evidence

Why it matters, in detail. From Fig. 9's own data the correction removes 33.5% of the 1930-2010 fall in Norway's top-0.1% share (12.03 -> 8.40 uncorrected vs 12.25 -> 9.84 corrected) and 25% from 1940; the 2000-09 average correction is 8.35 -> 9.62, not 8.4 -> 9.8.

Evidence. DataF9: top 0.1% excl./incl. hidden wealth 1930 12.03/12.25, 1940 13.22/13.44, 2000 8.35/9.62, 2010 8.40/9.84 -> 33.5% of the 1930-2010 decline erased (25.3% from 1940); top 0.01% 2010 3.846% -> 4.849% (+26%). DataF8 (2000-09 averages) Norway top 0.1% 8.35/9.62.

Why this severity. Overstates by about half an explicitly illustrative exercise with a stated 'sizable margin of error'; the qualitative inference (a less steep long-run decline once hidden wealth is counted) survives, and the claim is not in the abstract or conclusion.

8minorconfirmed · newly identified here · arithmetic or transcription

A few numbers in the global section are slightly off - for example, average income grew about 1.4% a year, not 1.9% - because they were copied from an earlier report. They do not change the paper's message.

Technical detail and evidence

Why it matters, in detail. Average income per adult in China+Europe+US grew 1.39%/yr, not 1.9%, so wealth (2.8%/yr) outpaced income by 1.4 pp rather than 0.9 pp; 2016 top-10%/bottom-50% shares are 68.9%/2.3%; the 7-8% Forbes figure mixes a world-level 1987 base with regional end points (effect unknown without the Forbes microfile).

Evidence. WIR 2018 C4.1.xlsx T4.1.1: C11 = 0.01387 (C+E+US income), D11 = 0.01344 (world income), D10 = 0.01920 (world wealth); C5:C6 = 0.0782/0.0700 (C+E+US Forbes), D5:D6 = 0.0639/0.0532 (world Forbes). WIR 2018 Part IV text repeats '1.9% for average income'. Adjudicator replication (scripts/outputs/global_levels.txt): 2016 top 10% 68.9%, bottom 50% 2.3%, middle 40% 28.8% (WIR published 28.83%). R code lines 162-170 ('Add year 1987 from C21C': 1.6e9 and 2.7e9).

Why this severity. Small slips copied from the World Inequality Report; none changes a conclusion (the income-growth slip, if anything, understates the gap between wealth and income growth).

9minorconfirmed · newly identified here · internal inconsistency

The introduction says both data sources show a rise of more than 10 percentage points since the early 1990s; the survey actually shows about 9. It is still a large rise.

Technical detail and evidence

Why it matters, in detail. SCF top-1% share rises +8.7 pp (1992-2016; +9.6 with Forbes) and +9.1 pp (1989-2016); the 'more than 10 points in both' statement holds only for the capitalized series (+10.3/+11.3).

Evidence. DataF2/CompF2: SCF 1992 30.10% -> 2016 38.79% (+8.69); with Forbes 31.17% -> 40.76% (+9.59); 1989 29.70%/30.83%; capitalized tax units 28.62% (1992) / 27.56% (1989) -> 38.89%.

Why this severity. Internal inconsistency on an intermediate statistic; the body gives the right figure and the conclusion of a large rise holds.

10minorconfirmed · newly identified here · arithmetic or transcription

The paper quotes the Federal Reserve's figure as 38.8% when the Fed's report says 38.6%, and adds the Forbes 400 in a way it does not explain. Both differences are tiny.

Technical detail and evidence

Why it matters, in detail. The cited Bulletin value is 38.6% (0.2 pp below the AR's 38.8%); the Forbes-adjusted series uses [s(W_FA - F) + F]/W_FA, 0.34 pp above simple addition to the SCF aggregate.

Evidence. Federal Reserve Bulletin Sept 2017, Box 3: 'climbed from 36.3 percent in 2013 to 38.6 percent in 2016'; public SCF 2016 microdata 38.55%. CompF2!E36 = (0.38791 x (74,350 - 2,397.4) + 2,397.4)/74,350 = 40.76%; adding Forbes to the SCF total (CompF2!M39 = $87,192bn) gives 40.43%.

Why this severity. Differences of 0.2-0.3 points that leave the ~40% level unchanged.

11minorconfirmed · newly identified here · arithmetic or transcription

One chart labels the increase in the top group's share as 'offshore wealth', which is not quite the same thing; done properly, the hidden part is slightly bigger than shown.

Technical detail and evidence

Why it matters, in detail. The true offshore component of the corrected share is T - s*W/(W+H), not T - s; the plotted offshore fractions are understated by roughly s*z/(1+z) relative to the total, i.e. by about 1.5-2 pp of the top group's wealth.

Evidence. DataF8 rows 7-9 (Total = Excl + Offshore; offshore fraction row 23: UK 38.0%, France 30.0%, Spain 29.4%, US 9.1%); AJZ2017bData.xlsx T.A9 row 107 (x = H/(H + recorded wealth)) and T.A11 formulas. Auditor B's consistent decomposition: UK 40.1%, France 32.2%, Spain 30.8%, US 10.5%; total-share effect of the formula slip negligible (UK 4.413% vs 4.410%).

Why this severity. A labelling/decomposition issue whose correction slightly strengthens the paper's point; the '30-40%' statement remains approximately right.

12minorconfirmed · newly identified here · data error

The computer code behind the global figures mishandles Spain in two small ways. Fixing them changes the reported growth rates by about a tenth of a percentage point and the shares hardly at all.

Technical detail and evidence

Why it matters, in detail. Excluding Spain (20% of 'Europe') from the wealth total but not the population overstates top-group real growth by ~0.11 pp/yr (3.53 vs 3.42% for the top 1%) and average-wealth growth by 0.04 pp/yr; the Spain unit mismatch has a negligible effect on shares in our replication (Auditor A estimated 0.2-0.3 pp).

Evidence. wealth-evolution-cn-eu-us.R lines 53-56, 61-65, 82-88, 187-193, 252-261. Adjudicator emulation (scripts/outputs/growth_rates.txt): archived-code logic gives 3.529/4.496/5.687/2.805%/yr (published T4.1.1 3.528/4.495/5.684/2.805) vs consistent 3.422/4.388/5.578/2.769 for top 1%/0.1%/0.01%/average. Units: US 2016 average 247,764 implies ~$76tn of household wealth only if in euro-PPP (x1.31); WID xlceup(US) = 1.31; Spain-unit fix changes global top-1% shares by <=0.1 pp (scenario B).

Why this severity. Code inconsistencies with effects of about 0.1 point per year on growth rates and under 0.1 point on shares.

13minorconfirmed · newly identified here · documentation gap

The 'today' global numbers reuse older country data (2012-2015) and an older version of the US figures, and the paper describes as an update what is a copy of the 2018 report. This makes the latest years less informative but changes the headline only slightly.

Technical detail and evidence

Why it matters, in detail. Post-2012/2015 movements of the pooled shares reflect only between-country changes in average wealth, not within-country distributional change; the US input is ~1 pp above the AR's own US series, worth ~0.3-0.6 pp on the 2016 global top-1% share.

Evidence. gpinterized.dta: top-1% shares CN 15.78-15.79% (1980-95) and 29.63% (2015-17); US 38.26% (2014-17); FR 23.38% (2014-17); GB 19.88% (2012-17). DataF2 col. F (US equal-split) 2014 37.24%, 2016 36.90%. import-spain.R loops at 1.03. AR p.124. Auditor A: substituting the AR's or the PSZ 2022 US series lowers the 2016 global top 1% from 32.4% to 32.1% / 31.8% (not re-run by the adjudicator).

Why this severity. A documentation gap about the vintage of 'today'; the effect on the headline level is a few tenths of a point.

The authors' response. The WIR 2018 technical note (Blanchet 2017) states that recent-year wealth is extrapolated assuming constant wealth/income ratios.

14minorconfirmed · newly identified here · documentation gap

A few references and labels are wrong or slightly overstated, such as citing the wrong paper for Russia's figures. They do not affect the findings.

Technical detail and evidence

Why it matters, in detail. Mis-citation of the Russia source, a rounding slip (38.8 vs 38.9%), a dangling reference label, a typo, and a mild overstatement of Kuhn et al.'s comparison; none affects any estimate.

Evidence. AR pp.117 (fn 9), 121, 125, 132 and reference list pp.134-138; NBER WP 25462 (Sec. 4.1) cites 'Piketty, Novokmet and Zucman, 2018'; DataFA1 2016 col. H 0.38893; Kuhn, Schularick & Steins (2017/2018 Minneapolis Fed IWP 9) Sec. 3.2.

Why this severity. Editorial slips with no effect on conclusions.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

I10 (moderate): US trend conclusions are specific to marketable wealth; including Social Security wealth largely offsets the 1989-2019 rise (Catherine, Miller & Sarin 2025). — Not an error in the work: the AR explicitly defines marketable wealth and argues for excluding Social Security and unfunded pensions (pp.112-113, fn 4); CMS estimate a different object whose results depend heavily on discount-rate assumptions. Recorded as a scope caveat in the limitations. Auditor B concurs.
I13(b): the introduction's 7-8% Forbes growth contradicts the body's 5.3%/6.4%. — Rejected as a numerical error: the intro refers to the China+Europe+US Forbes series (7.0%/7.8%, plotted in Fig. 7) and the body explicitly switches to the world ranking (5.3%/6.4%); retained only as a presentational note in F8(c), as Auditor B argued.
The pooled China+Europe+US 28%->33% result is robust: equal and single-country European donor stress tests preserve the increase, and global coverage concerns are disclosed. — Robustness not upheld: donor weights matter little (+5.2/+5.4 pp), but Auditor B did not test the Europe population composition or the currency conversion, which change the rise by 25-80% and are not disclosed in the AR (F1). The reproduction itself is confirmed.
I1: understated capitalization uncertainty is a major issue for the US findings. — Severity not upheld (issue retained as moderate, F2): the SCF independently shows the US rise (+8.6 pp) and the authors' later vintages keep the 2016 tax-unit level at 38.1-38.8%; only the size of the capitalized rise (-25% to -35%) and the very-top dynamics are materially affected.
I8: the Norway 'almost half' overstatement is a moderate issue. — Severity not upheld (retained as minor, F7): illustrative exercise with an explicitly stated large margin of error, absent from the abstract and conclusion; the qualitative inference survives with one third.
I8 sub-claim: only 14% of the decline is erased if AJZ's 1910s top-0.1% value (18.0%) is used, and the AR's '12-14%' early-century description omits it. — Not used in grading: the AR's stated base is 'since the 1930s', and the adjudicator did not verify the 1910s value in the AJZ appendix.
I2: frozen input distributions, the older US vintage and the 'update' wording constitute a moderate issue. — Severity not upheld (retained as minor, F13): documentation points worth at most a few tenths of a point on the 2016 level; the captions credit WIR 2018.
I16: the Spain unit inconsistency raises the global top-1% share by about 0.2-0.3 pp. — Magnitude revised: in the adjudicator's replication (which matches the published series within 0.09 pp) the Spain unit fix changes the top-1% share by <=0.1 pp; the issue is retained as minor (F12b).
I4(d): the AR misrepresents the Scandinavian validation studies (Fagereng et al. 2016; Lundberg & Waldenstrom 2018). — Only partly upheld (weakest sub-point of F3): Fagereng et al. find sizable level biases, but their top-1%/0.1% bias is of ambiguous sign and Lundberg & Waldenstrom find correlated trends, so 'not very large' is a contestable characterization rather than a clear misstatement.

Audit notes

What reproduced. Reproduced: (1) the global China+Europe+US series from the archived WIR 2018 inputs with independent code (published top-1% shares within 0.09 pp and bottom-75% shares within 0.01 pp in every year 1980-2016, incl. the 1991 break; 2016 top 0.1%/0.01%/middle 40% exact to 0.1 pp); (2) the WIR T4.1.1 growth rates by emulating the code's aggregation (within 0.003 pp), and the consistent rates without the Spain omission; (3) SCF top shares from public microdata (2016 38.55% vs Bulletin 38.6%; 1989 29.97%; +8.6 pp; tax-unit-sized groups 43.3%/32.9%, +10.5 pp); (4) all workbook-based US, country and offshore numbers checked (DataF1-F9, CompF2-F3, DataFA1); (5) the authors' later vintages (Sept-2020, Feb-2022) and SZZ's alternatives from their published workbooks. Not reproducible here: the capitalized US series itself (confidential IRS microdata), the Forbes global microfile behind the regional Forbes growth rates, and gpinter's exact interpolation (approximated). Statements that do not reproduce from the paper's own data or cited sources: '1.9%' income growth (1.39%), 'more than 10 points' for the SCF (8.7-9.9), Norway 'almost half' (~one third), 'top 10% >70%/bottom 50% <2%' for 2016 (68.9%/2.3%), 'official' SCF 38.8% (Bulletin 38.6%), and Forbes 'year-to-year strikingly similar' (r = 0.11).

Limits of this audit. Confidential IRS, HSBC and tax-amnesty microdata were not available, so the capitalized US series and the offshore allocations are taken as published; the WIR Forbes microfile, the AR supplemental appendix and the published NTJ text of Bourne et al. were not accessed (the latter's content was established from Saez & Zucman's BPEA 2019 description and the IRS manuscript). The global replication approximates gpinter (power-curve within-bracket quantiles) and proxies WID run-time population and exchange-rate series with the WIR 2018 inputs; the exact year-by-year match validates both, but the scenarios assigning Eastern Europe relative wealth levels are assumptions rather than measured distributions (using Russia's own, more concentrated distribution would add a few tenths of a point to recent top shares). Later vintages and critiques (SZ 2020/2022, SZZ 2023, Leenders et al. 2023, Catherine-Miller-Sarin 2025) post-date the AR and are used to gauge sensitivity, not as proof of error at publication; SZ dispute SZZ's corrections. The AR's estate-tax revenue ratios (0.20% -> 0.03-0.04% of household wealth) were not verified with state death taxes (Auditor A's federal-only figure is 0.135% for 1970-74). Social Security wealth is treated as a scope caveat: under a total-resources concept the US trend would be much smaller (CMS 2025).