Annette Alstadsæter, Niels Johannesen, Gabriel Zucman · Journal of Public Economics 162: 89-100

Who Owns the Wealth in Tax Havens? Macro Evidence and Implications for Global Inequality

18 problems found: 3 major · 3 moderate · 12 minor

What the paper claims

Global household offshore financial wealth is about 10% of world GDP ($5.6 trillion in the mid-2007 benchmark), with no particular trend over 2001-2015. Switzerland held ~40% of it in 2001, 45-50% in 2006-07 and ~30% recently, while Asian centres (Hong Kong x6 in 2007-15) rose.
Offshore wealth varies greatly across countries: a few percent of GDP in Scandinavia, about 15% in Continental Europe, and about 60% in Gulf countries and some Latin American economies. The text adds Russia ('as much as 60%', p. 90/95; 'more than 50%', p. 100).
Because offshore wealth is very concentrated, accounting for it raises the top 0.01% wealth share substantially in Europe. Scandinavia rises from about 4% to 5%; 30-40% of the wealth of the top 0.01% is held abroad in the UK, Spain and France; France goes from below to above Scandinavia; the UK becomes comparable; the US effect is muted.
Russia: the vast majority (around 60%) of the wealth at the top is held offshore.
Secondary (introduction and conclusion, not in the abstract): accounting for offshore wealth 'markedly' increases the rise in top wealth shares seen in tax data. France's top 0.01% share is now significantly higher than in the early 1950s; European and Scandinavian concentration has returned to its 1950s level (U-shape); the US top 0.01% share is as high as in early-20th-century Europe.

Every problem found

1majorconfirmed · previously known · speculative assumption

The paper assumes that in every country the richest 0.01% of households own about half of all hidden offshore wealth, because that is what Scandinavian data showed. It says it checked that other assumptions make little difference, but those checks are not in its published appendix or data. With a plausible lower figure, the extra inequality it reports for France, Britain and Spain shrinks by half or more, and France no longer overtakes Scandinavia.

Technical detail and evidence

Why it matters, in detail. The workbook formula incl = s/(1+h) + h*q, with h = offshore/(onshore+offshore) of about 2-4%, makes the increment roughly h*(q - s); with s of about 3-4%, it is driven almost entirely by q. With stocks fixed, the 2000-09 increments for France/UK/Spain fall from +1.57/+1.68/+1.02 pp to about +0.70/+0.77/+0.47 pp at q = 0.25, and to about +0.1 pp at q = 0.07. France falls below Scandinavia for q < 0.374, and the France 2014 vs 1950-54 ordering reverses for q < 0.30. External evidence on q ranges from 7% (Netherlands) to about 50% (Scandinavia, Colombia). The 'second-order' robustness claim is unsupported.

Evidence. Adjudicator recomputation (scripts/repro_top_shares.py; work/repro_top_shares.log), 2000-09 averages, all other inputs held fixed. Published q = 0.516: France 5.24, Scandinavia 4.78, UK 4.41, Spain 3.48. q = 0.40: France 4.86, Scandinavia 4.60. q = 0.25 (half): France 4.37 vs Scandinavia 4.38, so the reversal disappears; the offshore part of top-0.01% wealth falls to FR 16%, UK 22%, ES 16% (vs 30/38/29%). q = 0.07 (the Dutch amnesty value in Leenders et al. 2023): France 3.78, UK 2.88, Spain 2.56, i.e. increments of about 0.1 pp. With Scandinavia held at its observed q, France drops below Scandinavia for q < 0.374. France's 2014 share falls below its 1950-54 average for q < 0.30 (auditor B: below 1950 alone for q < 0.44). Leenders, Lejour, Rabate & van 't Riet (2023, JPubE 217:104785, abstract and Section 4) report on Dutch amnesty wealth: the top 0.01% own 7%, P90-P99.9 own 67%, and the top-0.1% correction is only 8.7% -> 9.3%. They also report about 50% for Colombia (Londono-Velez & Avila-Mahecha 2021) and for Scandinavia. Spain 2000-09: Martinez-Toledano's series gives +0.28 pp vs AJZ's +1.02 pp (3.7x).

Why this severity. The abstract's claim that offshore wealth raises top 0.01% shares 'substantially in Europe', and the 30-40% figure, scale almost one-for-one with a parameter that has not been tested for France, the UK or Spain. Halving it halves the European corrections and removes the France-over-Scandinavia reversal. The direction of the effect survives; its size does not.

The authors' response. AJZ (2019 AER, p. 2075) 'certainly do not claim' that the Scandinavian estimates hold 'as a universal law', but argue that nothing unique to Scandinavia explains evasion at the top. AJZ (2018, p. 90) concede that 'one cannot exclude that offshore wealth is more or less concentrated in certain countries'. No direct reply to Leenders et al. was located. The authors' Global Tax Evasion Report 2024 (ch. 1, p. 24) lists Leenders et al. among studies documenting 'similarly steep gradients', without discussing its 7% top-0.01% share.

2majorconfirmed · newly identified here · overclaiming

The claim that most of the Russian super-rich's wealth is hidden abroad is not measured. It comes from subtracting a very large estimate of Russian offshore money, which the paper itself says also covers property and the wealth of people who have left Russia, from a separate estimate of how rich the top is. With the paper's own narrower estimate, roughly a fifth to a third of top wealth would be offshore. The calculation even yields an impossible negative figure for 2009.

Technical detail and evidence

Why it matters, in detail. Excl. = (incl*(W+O) - q*O)/W, with incl taken from NPZ (Forbes plus generalized Pareto) and O = cumulated NEO ($609bn mid-2007, 3.54x the BIS-based $171.8bn, including real assets and emigrants' wealth). The offshore fraction q*O/(incl*(W+O)) is therefore set by q and O, not observed. It is 60% on 2000-09 means and 57-102% annually, with a negative onshore share in 2009. On the BIS scale it is about 19-34%.

Evidence. Adjudicator (scripts/repro_country_global.py, scripts/repro_top_shares.py; work/*.log). Published offshore part of top-0.01% wealth: 60% (ratio of 2000-09 means) and 68% (mean of annual ratios); 2009 excl. share -0.22% (T.A8!G110; Online Appendix Table A8). Scaling the NEO path to the BIS-based level (x171.8/609.0) gives 19% if NPZ's inclusive top-0.01% wealth is held fixed, and 34% if the backed-out onshore share is held fixed (2009 excluded). Auditor A: 21-42%; auditor B: 17.8% for 2007. The authors' 2025 BIS-based update raises Russia's 2007 stock only to about $175-256bn (FGMZ 2025 updated series, per auditor A). NPZ 2018 pp. 205-206 and 221.

Why this severity. The abstract's 'vast majority' comes from an undisclosed modelling step with an inconsistent stock. On the paper's own stock the offshore part of top wealth is about 20-35%, not a majority, so C4 as stated is not supported. Large Russian offshore holdings remain plausible, and the other core findings do not depend on this one, which is why this is major rather than critical.

The authors' response. None located. The authors' 2025 update (Faye, Godar, Moura & Zucman 2025, pp. 11-12) restates the AJZ result that 'the majority of wealth of the richest Russian residents was held outside of Russia already between 2000 and 2009' to justify an upward Russia adjustment.

3majorconfirmed · newly identified here · data error

Venezuela is the paper's only Latin American example of offshore wealth near 60% of GDP. That figure depends on an exception not mentioned in the paper, which counts a large pool of money the authors' own notes describe as Venezuelan government funds. Under the paper's stated rule Venezuela falls to about 25% of GDP, and no Latin American country comes close to 60%.

Technical detail and evidence

Why it matters, in detail. Using the stated 2003-04 average for Venezuela lowers its renormalized Swiss key from 5.03% to 1.63% (k' = k*r/(1 - k + k*r), r = 0.313). Its stock falls from $147.7bn to $57.1bn (64.1% -> 24.8% of GDP) and other Swiss keys rise by 3.6%. The 2006 value was kept knowingly, although the target concept is household wealth, so the Latin American upper tail of C2 is an artefact of non-household money.

Evidence. SNB, Banks in Switzerland 2003/2004/2006, Table 38: Venezuela fiduciary liabilities of CHF 1,964m (2003), 2,064m (2004) and 6,154m (2006). At bank-office level in USD this is $2,385m (2003-04 average) vs $7,614m (2006), a factor of 3.19. Adjudicator recomputation (scripts/repro_country_global.py) on the stated rule: Venezuela's Swiss key falls from 5.03% to 1.63%; its stock from $147.7bn to $57.1bn; its share of GDP from 64.1% to 24.8%. Every other country's Swiss key rises by 3.6%. The highest Latin American values become Argentina 37.7% and Paraguay 37.4% (all Latin American rows of Table A3 checked). The Gulf is unaffected (Saudi Arabia 57.9%, UAE 75.3%). Both auditors obtained the same numbers independently.

Why this severity. Venezuela is the only Latin American country near 60% of GDP, so the abstract's '60% in Gulf countries and some Latin American economies' loses its only Latin American instance. The regional maximum falls to about 38%, a change of more than a quarter of the stated level. The global total and the Scandinavia / Continental Europe / Gulf contrasts are untouched.

The authors' response. None in print. The choice is kept in the 2025 update code. Part of the 2005-07 rise could be private capital flight or public funds appropriated by individuals, but that is not shown.

4moderateconfirmed · previously known · speculative assumption

To decide which country owns the money in each tax haven, the paper uses several rules of thumb. For example, it assumes that money held through shell companies belongs to countries in the same proportions as money held directly. Other reasonable rules, including one the authors later used themselves, change some countries' figures by a factor of two or three. The broad pattern survives, but individual country numbers are less certain than presented.

Technical detail and evidence

Why it matters, in detail. Country stock = $2,667bn x Swiss key + $2,957bn x non-Swiss key. The keys depend on pro-rata reallocation of the ~50% of Swiss fiduciary deposits booked to havens, on ad hoc household shares and shell haircuts, and on a hard-coded Cyprus. A sector-based change to Swiss custody (55% household) shifts about $950bn from Swiss to non-Swiss havens and cuts the Gulf by about 11 pp of GDP. An FDI-based shell reallocation changes several country estimates by 2-3x.

Evidence. Auditor A's comparison with the authors' 2025 updated AJZ-method data (AJZ2017DataUpdated.xlsx, FDI-based reallocation of haven-booked Swiss deposits) gives these 2007 changes: Saudi Arabia 60.6 -> 31.5%, Greece 42.0 -> 12.7%, Italy 14.4 -> 5.6%, Spain 13.5 -> 7.1%, Norway 4.8 -> 7.9%, Denmark 4.4 -> 7.0%. FGMZ (2025, pp. 10-11, 15-16) present this variant as an alternative that shows 'the sensitivity of individual country results'. A 55% household share of Swiss custody securities (Schmitz 2021, Table 1), holding the global total fixed: Gulf 54.1 -> 43.1%, Venezuela 64.1 -> 45.3%, Argentina 36.5 -> 26.0%. Scandinavia (4.0 -> 4.1%) and Continental Europe (14.3 -> 13.4%) barely move (auditor A; adjudicator checked the Gulf figure by hand). Cyprus: 24.3/692.0 of European-haven deposits (totbyhaven sheet) times $1,504bn gives $52.8bn. Supporting the baseline: the HSBC leak, where shell accounts can be linked to owners, correlates 0.84 with the Swiss allocation (137 non-haven countries; adjudicator recomputation). FGMZ judge their FDI variant not preferred, since it gives unrealistic results for Czechia and Hungary.

Why this severity. Individual country levels and rankings, including the Gulf and Southern Europe, move substantially under defensible alternatives. But the broad pattern (low Scandinavia, about 15% Continental Europe, high Gulf) survives the best-supported alternatives, and the HSBC check partly validates the main key.

The authors' response. FGMZ (2025) move to centre-specific household shares and add an FDI-corrected variant. They acknowledge that 'our estimates suffer from uncertainty about who ultimately owns Swiss fiduciary deposits held through shell companies in other financial centers' (p. 16).

5moderateconfirmed · newly identified here · internal inconsistency

To extend its estimates back to the 1950s, the paper says it assumed offshore wealth grew like the wealth held in Swiss banks, but the spreadsheet actually uses a faster-growing series covering all tax havens. Under the method the paper describes, offshore wealth adds little to the rise in inequality since the 1980s, and France's top share today is about the same as in the early 1950s rather than 'significantly higher'.

Technical detail and evidence

Why it matters, in detail. Pre-2007 country shares h(t) = h(2007) x Index(t)/Index(2007). With the all-havens index, h rises about 55% from the 1980s to the 2000s, against about 7% with the Swiss index the text describes. Since the increment is about h*(q - s), the amplification of the 1980s-2000s rise depends almost entirely on this undocumented choice: for France +0.52 pp under the all-havens index vs +0.07 to +0.15 pp under the Swiss index.

Evidence. Adjudicator (scripts/france_index_variant.py; work/france_index_variant.log), same 2007 benchmark and q, Swiss index (M) instead of N. France's 1980s -> 2000s rise including offshore is +1.88 to +1.96 pp vs +1.81 pp excluding: an amplification of 4-8% instead of 29% under N. UK: +1.18 to +1.24 pp vs +1.08 pp, i.e. 9-15% instead of 51%. France 1950-54: 4.13-4.17% vs 2014: 4.64%; 1950 alone is about 4.60%. Under N the figures are 3.96%, 4.64% and 4.40%. Claude's and Codex's logs agree on the Fig. 10 composition and the caption error.

Why this severity. These dynamic claims are highlighted in the introduction and conclusion but are not in the abstract. Under the method the paper describes, the 'marked' amplification of the rise largely disappears and the France-1950s comparison becomes a near tie. The U-shape and the US-Europe gap survive.

The authors' response. The paper notes a 'sizable margin of error' for pre-2000 estimates (p. 98) and argues the broad patterns are robust. The all-havens index is arguably the more appropriate concept, but its pre-2001 path is assumed.

6moderateplausible · previously known · speculative assumption

The global total comes from a gap in international statistics: more investments are recorded as owed than as owned. The paper treats all of that gap as belonging to households. Some of it may belong to companies or be statistical noise, which would make the total smaller, perhaps by up to a quarter. Other independent estimates reach a similar size, so the broad order of magnitude is probably right.

Technical detail and evidence

Why it matters, in detail. About 80% of the 2007 stock is the residual liabilities-minus-assets gap. If 15-33% of it is reserves, NFC/institutional custody or FDI misclassification, the total falls about 12-27% (9.8% -> roughly 7-8.5% of world GDP), and every country stock and top-share increment falls proportionally. Independent mirror-statistics estimates and the authors' 2025 sector-adjusted update bound the downside.

Evidence. Schmitz (2021) pp. 3-4 and Table 1 (read). Auditor B's illustration: removing 15% or one third of the 2013 securities component cuts the total by 12-27% ($7.69tn -> $6.76tn / $5.62tn). Country stocks and top-share increments scale with the total. Pointing the other way: independent mirror-statistics estimates ($6.1-7.1tn for 2013, Pellegrini et al. 2016; $7.8tn for 2016, Vellutini et al. 2019). The authors' 2025 update, which corrects for reserves and Cayman-listed firms, finds $6.5tn at end-2007 vs $6.3tn here. Non-financial offshore assets are excluded entirely.

Why this severity. In an adverse case this could lower the '10% of world GDP' headline by up to about a quarter, with proportional effects downstream. Independent estimates corroborate the order of magnitude, and omitted real assets bias the other way, so the headline likely stands as an order of magnitude.

The authors' response. No direct reply located. FGMZ (2025) adjust for reserve assets and Cayman-listed companies (they add a share of these firms' liabilities, not remove it) but still attribute the residual gap to households.

7minorconfirmed · newly identified here · documentation gap

A few points in the global series were set by hand or copied from a consulting firm's growth rates rather than estimated, and this is not fully disclosed. The changes are small (about 4% for 2007) and do not alter the 'about 10% of world GDP' conclusion.

Technical detail and evidence

Why it matters, in detail. The 2006 +$400bn adds $200bn (3.6%) to the two-year mid-2007 benchmark. The 2014-15 points are BCG-scaled, so the post-2013 trend and its agreement with BCG hold by construction. Fig. 1's conclusion rests on 2001-13, which matches Zucman (2015) except for 2006.

Evidence. Adjudicator: the mid-2007 benchmark is $5,623.7bn (9.79%) with the +400 and $5,423.7bn (9.44%, -3.6%) without. T.A1 formulas read directly. Both auditors report the same.

Why this severity. This changes the 2007 level by 3.6% and makes 2 of 15 years non-independent. The 'about 10% of world GDP, no trend' conclusion is unaffected.

The authors' response. The Table A1 note discloses that the 2006 point was corrected and why, but not by how much.

8minorconfirmed · newly identified here · internal inconsistency

The text says Russia's offshore wealth is up to 60% of its economy, but the paper's own chart shows about 47% under the broadest measure and 13% under the paper's main method.

Technical detail and evidence

Why it matters, in detail. The text's 50-60% for Russia exceeds both of the paper's own mid-2007 estimates (NEO 46.9%; BIS 13.2%). Only the end-2007 NEO value (54.5%) exceeds 50%, and it is 3.5x the paper's own method.

Evidence. DataF4!E37 = 608.99/1299.71 = 46.9%; Table A3 row 133 = 13.2%; T.A9!T107/GDP = 54.5%.

Why this severity. A textual overstatement for one country. The figure itself is shown correctly.

9minorconfirmed · newly identified here · documentation gap

The paper quietly counts money held by residents of Monaco as French. There are arguments for doing so (France's official balance-of-payments statistics include Monaco), and it changes France's figures by only about 10%, so no conclusion changes.

Technical detail and evidence

Why it matters, in detail. Monaco is about 19.5% of the merged French fiduciary amount. Reallocating it like other havens lowers France's offshore wealth by 9.7% and its top-0.01% correction by about 0.15 pp. The issue is disclosure, and a mismatch between a BoP-territory numerator and a tax-data top share.

Evidence. Auditors A and B, consistent with each other: treating Monaco as a haven lowers France's stock from $409.3bn to $369.7bn (15.4% -> 13.9% of GDP) and France's 2000-09 top 0.01% share incl. offshore from 5.24 to 5.09. SNB 2003/2004 Table 38 values checked by the adjudicator.

Why this severity. This changes France's stock by under 10% and no qualitative claim: France stays about 14-15% of GDP and above Scandinavia. The merge is defensible under balance-of-payments residence.

10minorconfirmed · newly identified here · documentation gap

The paper says anyone can reproduce its figures from public data, but about half of its country allocation uses bank data that are not public (the authors offer them on request). The calculations that can be checked do check out.

Technical detail and evidence

Why it matters, in detail. The non-Swiss allocation ($2,957bn, 52.6% of the stock) uses non-public BIS bilateral files and the unpublished '1N' aggregate, so its keys cannot be rebuilt from public data. Released keys reproduce Table A3 to rounding.

Evidence. AJZ2017bPrograms readme.txt; build_offshore07.do lines 11-12 and 55-65. Both auditors and the adjudicator reproduced all downstream tables from the released keys but could not rebuild the non-Swiss keys.

Why this severity. This limits independent verification of about half of the allocation (the non-Swiss share of 52.6%) but is disclosed in the readme, and everything downstream checks out.

The authors' response. The readme discloses the restriction and offers the files on request.

11minorconfirmed · newly identified here · selective presentation

The paper checks its country estimates against the Panama Papers leak for a group of high-tax countries, but the group it uses is not the one it describes. Depending on how the rule is applied, the match is weak or strong, so this check is less informative than presented.

Technical detail and evidence

Why it matters, in detail. The Panama-owners/GDP vs offshore/GDP correlation ranges from 0.10 to 0.87 depending on which year's CIT rates define the sample, against 0.67 for the hand-coded list. The nominee filter in the code is not the one described. The validation therefore offers limited independent support for the country allocation.

Evidence. Adjudicator recomputation (scripts/panama_fig7.py; work/panama_fig7.log): authors' sample r = 0.667 (n = 16); stated rule with 2007 OECD rates r = 0.097 (n = 19); with 2016 rates r = 0.866 (n = 15); authors' sample without Greece r = 0.559. These match auditor A.

Why this severity. This only affects a supporting validation exercise. The correlation is fragile in both directions (strong under 2016 rates) rather than systematically flattering.

The authors' response. Footnote 15 argues that Eastern European shell owners are mostly nominees, which supports leaving those countries out.

12minorconfirmed · newly identified here · internal inconsistency

The authors' own spreadsheet says some offshore money (5-20%) was declared to tax authorities, but the calculation treats all of it as hidden. The effect is small for the 2000s and somewhat larger for recent years.

Technical detail and evidence

Why it matters, in detail. Setting B = 1 instead of 0.95/0.90 in the 2000s raises increments by 5-10%. For 2013-15 the authors' own 80% unseen would cut the increments by a fifth, reinforcing F6 on the France 'today vs 1950s' comparison.

Evidence. Workbook formula scan (no reference to T.A1 column G). Auditor A: using the declared shares lowers France's 2000-09 incl. share from 5.24 to 5.14 and its 2010-14 average from 4.78 to 4.51. Adjudicator: 90% unseen gives France 5.08, UK 4.25, Spain 3.38.

Why this severity. The effect on the 2000s results is small (about 0.1-0.15 pp for France); it is larger for the 2010s endpoints of the long-run comparisons.

The authors' response. Footnote 19 flags possibly higher declaration since 2013.

13minorconfirmed · newly identified here · arithmetic or transcription

There are a few small calculation and labelling slips in the inequality tables and charts. None of them changes any conclusion.

Technical detail and evidence

Why it matters, in detail. The denominator slip overstates shares by s*h^2/(1+h) (at most 0.011 pp). The stacked-bar labelling slightly understates the literal offshore fractions, which if anything strengthens the 30-40% statement within the paper's own assumptions.

Evidence. Adjudicator: the exact formula gives France 2000-09 5.236 vs 5.240. Literal offshore fractions of top-0.01% wealth: FR 33%, ES 31%, UK 41%, RU 66% vs 30/29/38/60% implied by the figure. Auditor B: maximum formula error +0.011 pp (US 2014); Finland 1967 6.32% vs 6.62%. T.A10 vs T.A7 Finnish cells checked.

Why this severity. Each effect is at most a few hundredths of a point on the main series. The Finnish cells and the caption do not enter the main comparisons.

14minorconfirmed · newly identified here · documentation gap

The UK figure for the very top rests on only four years of data with no stated source, and the Norwegian numbers differ slightly from the authors' companion paper. These are small documentation problems.

Technical detail and evidence

Why it matters, in detail. The UK excl. average (2.73%) is driven by four estate-based observations, including a low 2000 value; without it the average is 3.03%. Norway's increment is 1.27 pp vs 1.4 pp in the companion paper.

Evidence. Sheet 'source-top-shares' row 16; T.A8 K101, K102, K104, K110; DataF7 K12/K14; AJZ 2019 p. 2094 (all checked).

Why this severity. Documentation and consistency problems. Dropping the 2000 UK outlier changes the UK offshore fraction from 38% to about 35%.

15minorlikely · newly identified here · overclaiming

Two statements go further than the evidence: that top income shares are understated by the same proportion as top wealth shares, and that the method cannot overstate inequality because the global total is cautious. Neither is demonstrated, though neither is central.

Technical detail and evidence

Why it matters, in detail. A conservative global total does not make country-level corrections conservative, because the allocation is zero-sum and several country choices push upward. The income-share extrapolation needs evidence on returns to hidden wealth that the paper does not provide.

Evidence. Footnote 17 (p. 98); p. 97; see F1-F4 and F12.

Why this severity. These are interpretive statements, not inputs to the results.

16minorplausible · newly identified here · methodology

Part of the offshore money is bank deposits, and some countries already count those in their official statistics. Adding all of it again may double count a little, at most about a fifth of the effect.

Technical detail and evidence

Why it matters, in detail. If a country's financial accounts already contain its residents' offshore deposits, adding them again inflates both W+O and the top group's wealth. At most the deposit share (about 18%) of the offshore increment is affected.

Evidence. Zucman (2013) QJE pp. 1342-1343 and fn. 27; Martinez-Toledano (2017) fn. 23; Table A1 (deposits $1,165bn of $6,335bn at end-2007).

Why this severity. The upper bound is about 18% of each increment (France +1.57 -> about +1.29 pp). Recording is incomplete in most countries, so the true effect is smaller.

17minorplausible · newly identified here · methodology

The method mixes two different ways of deciding who counts as 'the richest', which slightly inflates the corrected inequality figures. The effect is likely small.

Technical detail and evidence

Why it matters, in detail. (reported wealth of the top group ranked by reported wealth + q*O)/(W+O) is at least (reported wealth of the true top group + q*O)/(W+O). The re-ranking bias is upward and second-order.

Evidence. AJZ (2019) fn. 23-24; AJZ (2018) Section 5.1 and the T.A11 formula. AJZ (2019) Fig. E.2 suggests similar gradients under either ranking.

Why this severity. The direction of the bias is clear but its size is probably small.

18minorplausible · previously known · external inconsistency

The paper's comparison of today's American super-rich with Europe a century ago relies on U.S. figures that other researchers think are too high. With lower published alternative estimates, the U.S. today would be well below early-1900s Europe, though still far more unequal than Europe today.

Technical detail and evidence

Why it matters, in detail. Applying the SZZ/PSZ ratio (7.1/9.5) to AJZ's US 2010s excl. share (9.88%) gives about 7.4%, or about 8.5% with the offshore increment, against Europe's 12.5-13.3% in 1910-1929. The 'as high as early-20th-century Europe' claim is therefore not robust to the main published alternative to SZ (2016).

Evidence. Adjudicator: T.A8/T.A11 column L and T.A11b column O; SZZ (2023) p. 518, read from the cached PDF; corpus entries us-wealth-taxes-06 and -02.

Why this severity. This affects one secondary comparison, which is about an imported baseline rather than the paper's offshore contribution. 'US much more concentrated than Europe today' survives (about 8% vs 4.3%).

The authors' response. Saez & Zucman (2020), 'Comments on Smith, Zidar and Zwick', argue that SZZ underestimate interest-bearing and billionaire equity wealth and that, once these are corrected, SZZ's series nearly matches theirs.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

I15: the paper overstates the precision of the US Senate (PSI 2008, 2014) evidence that '90% of the wealth held by U.S. citizens at UBS and Credit Suisse was undeclared'. — PSI (2008, p. 10) reports UBS's internal October 2005 data: about CHF 18.5bn in undeclared vs CHF 2.6bn in declared US accounts, i.e. 88% of assets undeclared. It also reports 19,000 of 20,000 US accounts (95%) undeclared in 2008. PSI (2014) says the 'vast majority' of 22,000 Credit Suisse US customers were undeclared. The paper's '90%' is a fair rounding; only 'in 2008' is loose, and the 100%-unseen assumption does not depend on it. Severity: none.
I6 (part 6): the amplification of the rise in inequality is 'mechanical' and therefore a defect. — Rejected as a standalone defect. Constant concentration combined with a rising offshore index is the paper's stated and disclosed model (p. 98). What matters is which index is used and that it differs from the text; that is covered and quantified in F6.
I1 (sub-claim): 'none of this is disclosed' about the Russian NEO stock. — Partly rejected. The NEO stock's broader scope (real assets, non-residents' wealth) is disclosed on p. 95 and in Fig. 5. What is undisclosed is its use in Section 5, and that the inclusive series is imported and the exclusive series backed out. F2's severity is unaffected.
I6: including Monaco is a 'meaningful boundary discrepancy' with the paper's residence concept, rated moderate. — The inconsistency claim is largely rejected. The Banque de France balance-of-payments methodology (section 2.2) includes Monaco in France's economic territory, consistent with the paper's 'national accounts concepts'. What remains (non-disclosure, and a mismatch with French tax-based top shares) changes France by under 10% and no conclusion, so it is kept as minor (F9).
I4 and I7 rated moderate: BCG-linked endpoints and the public-reproducibility gap. — Both findings are confirmed but downgraded to minor (F7, F10). The 2006 adjustment moves the 2007 benchmark by 3.6%, the BCG extrapolation affects only 2014-15, and the data restriction is disclosed in the readme while all downstream arithmetic reproduces. Neither materially affects a core finding.
I3 rated moderate: the Russian decomposition is a legitimate decomposition of a defensible inclusive estimate. — The finding is upheld but its severity is raised to major (F2). The abstract's quantitative claim depends on a stock 3.5x the paper's own estimate, the decomposition gives a negative 2009 share in the published appendix, and NPZ themselves say the offshore fraction of Forbes wealth is unknown.

Audit notes

What reproduced. Reproduced exactly by the adjudicator from the authors' workbook (values and formulas): - The mid-2007 benchmark of $5,623.7bn = 9.79% of 2007 world GDP, and the Swiss-share path (41.0/49.3/46.0/31.5/25.7%). - The Table A3 country totals from the published keys (they sum to the benchmark). - All 403 non-Russian Table A11 cells (maximum difference 0). - The Fig. 8 values (Scandinavia 4.07->4.78; FR 3.67->5.24; ES 2.46->3.48; UK 2.73->4.41; US 7.14->7.85; RU 5.02->12.50) and the 30-40% offshore fractions. - The Russian back-out identity, including the negative 2009 value printed in Online Appendix Table A8. - The Fig. 7 correlation of 0.667 and the HSBC-Swiss correlation of 0.84. Auditor A rebuilt all 146 Swiss country keys from primary SNB tables; the adjudicator checked the SNB entries for Venezuela, France and Monaco. Not reproducible from public data: - The non-Swiss (BIS) keys, which use a confidential '1N' haven aggregate. - The hidden-wealth distribution (confidential Scandinavian microdata). - The global portfolio gap from primary IIP/CPIS data, which neither auditor nor the adjudicator rebuilt; all relied on the Zucman (2015) file. Counterfactuals computed by the adjudicator, one assumption at a time and not additive: - Venezuela on the stated 2003-04 rule: 64.1% -> 24.8% of GDP. - Russia with the BIS-scale stock: offshore part of top-0.01% wealth 19-34% vs 60-68%. - Concentration parameter q of 0.40 / 0.25 / 0.07: France 4.86 / 4.37 / 3.78 vs Scandinavia 4.60 / 4.38 / 4.12; reversal thresholds q = 0.374 (France vs Scandinavia) and 0.30 (France 2014 vs 1950-54). - The text-described Swiss index: France's 1980s-2000s amplification 4-8% instead of 29%. - Fig. 7 sample rules: r = 0.10 (2007 rates) to 0.87 (2016 rates). - 90% unseen: France 5.08. - Exact denominator: France 5.236.

Limits of this audit. The non-Swiss allocation could not be rebuilt from raw BIS data, because the authors' dataset with the '1N' aggregate is confidential; it was checked only through the released keys. The global portfolio gap was not re-estimated from primary IIP/CPIS data. The national top-share inputs were not verified against their primary sources, except Russia, which the auditors checked against NPZ. A comparison with the October 2026 WID bulk file shows large later-vintage differences: France's 2000-09 top 0.01% share is 4.7% there vs 3.7% in AJZ, and the Nordic levels differ even more. We could not establish whether these later series embed offshore or rich-list corrections, so no finding is drawn from them. Treating Venezuela's 2005-07 Swiss deposits as government money rests on the authors' own code comments and on ICIJ reporting, not on SNB sectoral data. The FDI-variant and Swiss-custody numbers in F4 come from auditor A's scripts; the Gulf figure was checked by hand. All sensitivities vary one assumption at a time and should not be summed. The CEPR DP version was not read; the NBER (2017) and JPubE (2018) versions bracket it. Leenders et al., Schmitz, AJZ (2019), NPZ (2018), FGMZ (2025), GTER (2024) and SZZ (2023) were read from cached copies. The SZZ comparison in F18 mixes tax-unit and household concepts.