Gabriel Zucman · Paris: Seuil (La République des idées), 2013 (2nd augmented ed. 2017); English ed. University of Chicago Press, 2015 (trans. Teresa Lavender Fagan, foreword by Thomas Piketty)

La richesse cachée des nations: enquête sur les paradis fiscaux / The Hidden Wealth of Nations: The Scourge of Tax Havens

13 problems found: 2 major · 7 moderate · 4 minor

What the paper claims

About 8% of households' (net) financial wealth is held in tax havens, about one-third of it in Switzerland.
Offshore wealth kept growing despite the post-2009 crackdown: 'tax evasion is doing just fine. There has, in fact, never been as much wealth in tax havens as today.'
About 80% of offshore wealth is undeclared, costing governments about $190bn a year in personal income, inheritance and wealth taxes.
Offshore wealth and the resulting losses differ sharply by owner region.
US multinationals book a majority of their foreign profits in tax havens, pay almost no tax on them, and profit shifting cuts their tax bill substantially.
Policy: a world financial registry with automatic information exchange, sanctions proportional to havens' costs, taxation of consolidated corporate profits, and a global tax on capital.
A century of offshore finance: the share of Europeans' financial wealth held offshore rose over the twentieth century.

Every problem found

1majorconfirmed · previously known · speculative assumption

The book's $190 billion yearly tax loss assumes hidden money earns 8% a year and that all of that gain is taxed every year at full rates, and it uses the highest inheritance-tax rates as if nobody qualified for exemptions. The author himself used 6%, 7% and 6.5% in other editions. With more realistic assumptions the loss is still large, but more like $110-170 billion than 'at least $200 billion'.

Technical detail and evidence

Why it matters, in detail. dLoss/dr is about $15.9bn per percentage point. Taxing the full nominal total return annually at personal dividend rates is an upper-bound concept for the income-tax base: realization, deferral, step-up at death, preferential gains rates and fees all lower the taxable yield, and 2013-14 deposit and bond yields were near zero. Top marginal estate rates applied without spousal exemptions (and at 30-40% in regions where many countries levy little or no inheritance tax) bias the $51.8bn inheritance term upward. A defensible range is roughly $110-170bn against $189bn; the 'at least $200bn' framing is not supported.

Evidence. Zucman2015TablesFigures.xlsx, sheet Data-Fig4_Tab1: O3 = 0.08, P3 = 0.03, H = 0.2 for all regions, K19 = 0.3, K20 = K21 = 0.4, note A29(iii) ('Nominal rate of return on offshore assets changed to 8%'); M25 = 189.39 = O25 127.09 + P25 51.82 + Q25 10.48. EN Online Appendix (Sept. 2015), 'Mutual fund returns': 'well in line with the 6% return average nominal return on offshore accounts I retain'. FR 2013 Données-T3!N3 = 0.06; JEP DataTab1!N3 = 0.07 (note A29); FR 2017 DataTab4!O3 = 0.065. Author's slides (Sept. 2015): 'costing at least $200bn'. Adjudicator re-implementation (scripts/revenue_model.py, out_revenue_model.txt) reproduces $189.39bn; r = 6%: $157.6bn (-17%); 5%: $141.7bn (-25%); 4%: $125.8bn (-34%); inheritance at half of top statutory rates: $163.5bn (-14%); 10% inheritance rates for Asia/Latin America/Africa: $174.4bn (-8%); r = 5% with half inheritance rates: $115.8bn (-39%). Pellegrini, Sanelli & Tosti (2016), Banca d'Italia QEF 367, pp.26-28: with instrument yields of 0.5-3.3% capital-income tax evasion is $20-42bn a year (at most $49bn in 2013) against the book's $127bn, a gap 'almost exclusively due to the use of quite different rates of return'; their yields exclude taxable realized gains and are a lower bound. Auditor B also reports, from the chapter text (not verified by the adjudicator), that the book justifies a 5% real return including capital gains by citing Piketty & Zucman (2014), whose return measure is income-based.

Why this severity. The income-tax term is two-thirds of the headline and is linear in an assumption the author himself varied between 6% and 8%; defensible returns (4-6%) and effective rather than top statutory inheritance rates cut the total by roughly 15-40%, so $190bn is near the top of a plausible range rather than a minimum. Large losses (tens of billions to over $100bn a year) survive.

The authors' response. No direct reply to Pellegrini et al. located. The 2017 edition lowered r to 6.5% and raised the declared share to 25% ($170bn = EUR 155bn; DataTab4). JEP 2014 (p.142) states that 'the main uncertainty relates to the fraction of those funds that evade taxes'.

2majorlikely · previously known · methodology

The statistics the book uses to show where US companies' foreign profits sit count some profits twice and credit the tax havens with profits earned and taxed in other countries, because that is where holding companies are registered. Corrected figures put the havens' share nearer a third than a half, their tax rate nearer 10% than 3%, and the yearly cost nearer $80 billion than $130 billion. Profit shifting is still big, but the book's figures overstate it.

Technical detail and evidence

Why it matters, in detail. DII by first-tier country conflates where holding companies book income with where it is earned; with equity income at 75% of haven affiliates' net income, both the haven share and the 3% tax ratio are distorted. Replacing DII with adjusted pre-tax income cuts the haven share to ~35% and the cost to ~$80bn; a 25% rather than 35% counterfactual gives ~$92bn (both changes ~$58bn). The equity-income subtraction may remove some genuine foreign-to-foreign shifting (Clausing 2020), so a defensible range is roughly $60-120bn. The effective-rate attribution is an accounting residual, not a causal decomposition.

Evidence. Zucman2015TablesFigures.xlsx DataFig8-9 row 89: L89 = 657, CQ89 = 0.5583, CR89 = 0.1741, AG89 = 474.3; AY74 = 0.293 (1998), AY89 = 0.199 (2013). BEA USDIA preliminary 2011, table II.D1 (bea2011p-income.xls), haven group (Ireland, Luxembourg, Netherlands, Switzerland, Other Western Hemisphere, Singapore): net income $611.9bn, of which equity income $456.7bn (74.6%), foreign income taxes $18.4bn; tax/net income = 3.00%, tax/(net income - equity income + tax) = 10.59%; all countries 14.8% vs 28.0% (scripts/corporate_checks.py; first established by Auditor B). Blouin & Robinson (2020), 'Double Counting Accounting', SSRN 3491451, sec. 4.1 and 5.2.1: adjusted pre-tax income puts the 2013 haven share at ~35% and the cost at ~$80bn ($650bn x 35% x 35%). Alternatives on the book's base: 35% share: $80.5bn; 25% counterfactual rate: $91.7bn; both: $57.5bn. JEP 2014 p.129: balance-of-payments data 'do not reveal the real source of profits, but mainly the location of the holding companies'; pp.131-133 for the residual attribution.

Why this severity. A published, data-based correction lowers the $130bn by about 40% (to ~$80bn) and the haven share from 55% to ~35%; the 3% rate is an artefact of mixing accounting bases. Large shifting to havens survives (later non-double-counted estimates by the author's team still find it), so the qualitative claim stands.

The authors' response. Tørsløv, Wier & Zucman (REStud 2023, sec. 4.2.1) accept that BEA net-income tables double count but say their later estimates use profit-type-return and FATS data free of the problem; the book's 55%/$130bn were not revised. Clausing (2020) argues the Blouin-Robinson correction strips out genuine foreign-to-foreign shifting. The author's 2015 WSJ reply clarifies that the $130bn is a loss to all governments, not only the US Treasury (book p.105).

3moderateconfirmed · newly identified here · circularity

The book's world total of $7.6 trillion was obtained by assuming that the 8% share measured for 2008 still held, not by adding up the new data, so the reported growth of 'about 25%' is just the growth of the estimated world wealth figure. The author's own updated data give a similar total and show that hidden wealth did grow after 2008, so the main message stands, but the growth numbers quoted are not measurements.

Technical detail and evidence

Why it matters, in detail. With the share fixed at 8%, every published total, regional amount and growth rate scales with an assumption-laden denominator (world GDP x 0.85 x a 4.4 wealth-income ratio x 0.30 net financial share / 0.94). Bottom-up totals from the author's own gap bracket the 2013 level at $6.8-8.2tn (7.1-8.6%), about +/-10%; gap growth is +31-45% from 2008-09 but +20% from the 2007 peak, so any 'over 25%' claim is base-year sensitive and in any case was not computed from the evidence. Gap vintages differ by up to ~10%.

Evidence. Zucman2015TablesFigures.xlsx Data-Fig3!H3 '=0.85*80000', H4 '=H3*0.3*4.4/(1-0.06)', G5 = 0.08; Tab1!B13 '=ROUND(0.08*J13,-2)'; Data-Fig4_Tab1!D25 '=7600-B25'; FR 2013 Données-T2!F5 '=ROUND(F4*G5,-2)' with G5 = 0.08 and H4 = 95,489; JEP 2014 Table1!B13 same formula. QJE 2013 Table III: $5,878bn, household financial wealth $73,625bn (Credit Suisse 2010). Author's securities gap (Zucman2015MissingWealth.xlsx, Table A3, col. E): 2007 $5,170bn, 2008 $4,266bn, 2009 $4,720bn, 2013 $6,185bn (series maximum); growth 2008-13 +45.0%, 2009-13 +31.1%, 2007-13 +19.6%. Bottom-up 2013 totals: gap + $1.5tn deposits (the figure Faye, Godar, Moura & Zucman 2025, p.8, attribute to the book) = $7,685bn (8.05%); gap/0.75 (QJE rule) = $8,247bn (8.6%); gap x 1.10 (Swiss deposit ratio) = $6,804bn (7.1%). The gap for 2011 was revised from $5,442bn (2013 workbook) to $4,911bn (2015 workbook), -9.8% (scripts/headline_checks.py).

Why this severity. The level itself is supported (the updated gap plus the book's deposit figure gives $7.69tn), but the published growth figures and regional amounts are not measurements; since the measured gap grew at least as fast from end-2008, the qualitative claim survives and the problem is an undisclosed construction rather than a wrong conclusion.

The authors' response. The 2017 preface (p.7) says offshore holdings have grown 'au même rythme environ que le patrimoine financier mondial', i.e. a constant share; Faye, Godar, Moura & Zucman (2025) find the share 'fluctuated around 7%' of household financial wealth (about 10% of world GDP) over 2001-2023, which supports the level and the constant-share approximation. No explicit response on the construction.

4moderateconfirmed · newly identified here · speculative assumption

The claim that 80% of hidden wealth escapes tax comes from Swiss data on European savers, adjusted with an assumed interest rate, and is then applied to every country and every tax haven. The English appendix also quotes the Swiss figure inaccurately. Under other reasonable assumptions the untaxed share could be anywhere from about 55% to 90%, which moves the tax-loss estimate by up to a quarter either way.

Technical detail and evidence

Why it matters, in detail. Revenue is proportional to (1 - h). The only direct evidence is Swiss AFC data on directive-covered interest of EU residents; converting them into a wealth-based compliance rate needs an assumed yield and portfolio mix, and the result is then extrapolated to all havens and owners. h is 10% (FR 2013/JEP, taxing regions), 20% (EN), 25% (FR 2017); 2013 data are consistent with h of roughly 15-45%. The bias can go either way, but the documentation misstates the statistic and the uncertainty is not shown.

Evidence. Zucman2015Appendix, 'Fraction of offshore wealth which is not declared on tax returns'; Zucman2013LivreAnnexe pp.14-15 ('about a third ... while 2/3 is not'; overall ~20% with 10% compliance for shell accounts); Zucman2014JEPAppendix (Dec. 2014), 'The fraction of offshore money that evades taxes'; JohannesenZucman2014ExcelData.xlsx Data-Fig7: declared/(declared + taxed) = 16.0% (2008), 26.4% (2011), unweighted 2011 average 21.7%. FR 2013 Données-T3!H5 = 0.1, H18 = 1, H24 = 0.808; JEP DataTab1!H5 = 0.1, H18 = 1; EN Data-Fig4_Tab1 H = 0.2 everywhere; FR 2017 0.25. revenue_model.py: declared 30%: $165.7bn (-12.5%); 40%: $142.0bn (-25%); 10%: $213.1bn (+12.5%).

Why this severity. The compliance share scales the entire revenue estimate and its evidential base is narrow and partly misdescribed, but pre-2013 evidence (US Senate findings of 85-95% undeclared accounts before 2008; modest voluntary disclosures) supports high non-compliance, and plausible values move the total by roughly 15-25% in either direction.

The authors' response. JEP 2014 (p.142) acknowledges that 'the main uncertainty relates to the fraction of those funds that evade taxes' and cites US Senate evidence of 85-95% undeclared UBS/Credit Suisse accounts before 2008; the 2017 edition raises the declared share to 25%; the Zucman-led Global Tax Evasion Report 2024 later estimates a much lower untaxed share after automatic exchange of information.

5moderateconfirmed · newly identified here · selective presentation

The book says money in Swiss banks grew 18% after world leaders vowed to end bank secrecy in 2009. That is true in dollars, but mostly because stock markets recovered and the dollar fell against the Swiss franc; measured in francs the total shrank, and it stayed well below its 2008 peak. So the Swiss numbers do not show that evasion was 'doing just fine', even if hidden wealth worldwide did not fall.

Technical detail and evidence

Why it matters, in detail. Nominal dollar growth from a near-trough base mixes valuation and exchange-rate effects with net flows. A currency-neutral benchmark from the same custody statistics (residents' foreign portfolios) shows non-resident holdings underperforming by about 15 points in 2009-15, i.e. net outflows (relocation to other havens per Johannesen-Zucman 2014, or repatriation/declaration). Evidence for C2 therefore rests on the global gap, which rose 31-45% from 2008-09 in nominal dollars but was roughly flat as a share of household financial wealth.

Evidence. scripts/swiss_series.py on Zucman2014JEPdata.xlsx (TableS.1, SNB_FX): April 2008 $3,100.6bn; April 2009 $2,102.6bn / EUR 1,593.4bn; June 2014 $2,462.0bn / EUR 1,811.0bn; CHF per USD 1.1482 to 0.896; total in CHF 2,414bn to 2,206bn (-8.6%); fiduciary deposits $435bn to $191bn. scripts/snb_custody_chf.py on SNB cube bawebewa: non-resident foreign-issuer custody CHF 1,595bn (Apr 2009) to 1,867bn (Apr 2015): raw +17.1%, +10.0% after the author's 6% break adjustment; residents CHF 828bn to 1,060bn (+28.0%; residents' December-2012 reporting jump only +2.2%). FR 2017 workbook DataTab1 note A32: April 2017 Swiss total $2,252bn; FR 2017 introduction p.12: EUR 2,100bn, +25%. Auditor A's primary-SNB reconstruction: +17.8% to April 2015, +11.0% to March 2015 (USD).

Why this severity. The published numbers are arithmetically right in the book's dollar convention, but the inference that evasion is 'doing just fine' is not supported by Swiss holdings that were below their pre-crisis peak and shrinking in francs and relative to a valuation benchmark; the global nominal-record claim survives on the gap series.

The authors' response. JEP 2014 (p.141) says post-2008 growth 'reflects both valuation effects ... and also net new inflows'; Johannesen & Zucman (2014) document relocation to non-cooperative havens; Faye et al. (2025) later report that Switzerland's share of global offshore wealth declined significantly after 2006.

6moderateconfirmed · previously known · speculative assumption

How much hidden wealth belongs to Europeans, Africans or Russians is not measured directly; it is divided up using 2004 Swiss bank-deposit shares plus the author's judgment calls. Those judgment calls changed between editions and moved Africa's estimated offshore share from 30% to 44%. The regional numbers are rough guides, not measurements.

Technical detail and evidence

Why it matters, in detail. The gap identifies a world total, not a geographic ownership matrix. Regional amounts are the world total times assumed keys, so regional shares move by up to half with key changes that are mostly judgmental (Russia excepted); regional revenue losses inherit the keys, and small onshore denominators (Russia $0.2tn, Gulf $0.6tn, Africa $1.2tn) amplify them.

Evidence. Zucman2015TablesFigures.xlsx DataFig2 C4:C23 and notes A32-A38; Data-Fig4_Tab1 E5 = 0.25, E19 = 0.2, G22 hard-coded, note A29(i). Zucman2017LivreTableauxGraphiques.xlsx DataTab1 C4 '=56%-10%', C18 '=10%+3.5%', C20 '=6.5%+3.5%', C23 '=3%+3%', notes A33-A39; Tab4: Russia 600 (54.1%), Africa 900 (43.7%), Latin America 1,000 (26.5%). Auditor B: with the world and Swiss totals fixed, a 15% or 35% instead of 25% European share of non-Swiss havens gives 8.9% or 12.8% instead of 10.9% of European financial wealth.

Why this severity. Affects secondary regional results (shares for Africa, Russia, the Gulf and Latin America and their revenue losses), whose apparent precision is not supported; the world total is unaffected.

The authors' response. Adjustments are disclosed in workbook notes; 2004 is chosen as the last year before the EU Savings Directive triggered the shift to shell companies; the 2017 Russia figure draws on Novokmet, Piketty & Zucman (2017).

7moderateconfirmed · newly identified here · methodology

The book's century-long chart of Europeans' wealth in 'all tax havens' copies the Swiss figures for the early decades and draws a straight line for 1980-2010. It looks like a measured history but is mostly an assumption; the Swiss part of the chart is based on real archives.

Technical detail and evidence

Why it matters, in detail. Of 12 plotted all-haven points, 7 copy the Swiss series and 4 are linear interpolation, so the slope is (L2013 - L1970)/43 by construction. Setting non-Swiss havens to zero before 1980 mechanically steepens the measured rise from 4.5% to 9.9%; the 2017 endpoint rule shifts the 2000 value by 1.9 points (+24%) with no new data.

Evidence. Zucman2015TablesFigures.xlsx Data-Fig1: L3:L9 '=K', L10:L13 '=L9+(L$14-L$9)/5', L14 '=Tab1!C5', F7:F10 '=F$14', F11 = 0.44, I3:I14 constants; Zucman2017LivreTableauxGraphiques.xlsx Data-Fig1 L10:L12 '=L9+(L$13-L$9)/4', L13 '=K13*2', note A17 (verified by adjudicator; first reported by both auditors).

Why this severity. A secondary historical result: only the endpoints are estimated, and the 1970 all-haven level assumes no non-Swiss havens, so the timing and part of the size of the rise are imposed; the archival Swiss series is unaffected.

The authors' response. The FR 2017 workbook note A17 discloses the 2010s proxy (twice the Swiss share); no other response located.

8moderateconfirmed · previously known · speculative assumption

The book argues that a 30% tariff on Swiss goods by France, Germany and Italy would cost Switzerland more than it earns from hiding money. That holds only if Swiss exporters, rather than European buyers, absorb most of the tariff; if buyers pay part of it, a much higher tariff would be needed. The same table's proposal for Hong Kong fails its own test.

Technical detail and evidence

Why it matters, in detail. The cost of a tariff to the exporting country is the exporter-borne share of the price wedge plus lost producer surplus on reduced volumes, not gross tariff revenue. With high pass-through to importers (as found for the 2018 US tariffs, though transferability to Swiss specialty exports is uncertain) the break-even tariff could be 30-70%; quantity effects could raise Swiss losses, so the sign of the error is ambiguous but sufficiency is not shown.

Evidence. Zucman2013LivreTableauxGraphiques.xlsx 'Tableau 4': D8 '=0.8*C8*0.01/I8', L8 '=J8*F8*K8', row 9 (Hong Kong), rows 10 and 12 without cost cells; scripts/tariff_checks.py. EN Introduction p.5. Sternberg (2015), Wall Street Journal review, p.2 (importer incidence). Auditor B's calculation from the English chapter text (EUR 60bn of imports, 60/1.3 = 46.15).

Why this severity. A secondary policy calibration: the claim that 30% would be enough rests on an unestimated, favourable incidence assumption; it does not affect the measurement findings.

The authors' response. The author's published reply to the WSJ review (20 Oct. 2015) does not address tariff incidence; according to Auditor B the chapter suggests enlarging the coalition if needed and notes that a credible threat may suffice.

9moderatelikely · previously known · speculative assumption

The method treats every unexplained gap in world investment statistics, and every foreign account in Swiss banks, as belonging to wealthy households. Swiss data show most of those accounts are registered to companies and institutions; some are fronts for individuals, but some are genuine funds and banks. The Swiss figure may therefore be too high, possibly by a lot, though this matters much less for the world total.

Technical detail and evidence

Why it matters, in detail. Genuine institutions that report their holdings do not create a gap, so the bias in the world total is limited to unreported non-household holdings (up to a third of the 2016 gap on Schmitz's crude French-proportion extrapolation; much less after the adjustments in Faye et al. 2025): an overstatement of 10-25% is possible but not demonstrated. For Switzerland, attributing institutional custody wholesale to households drives the level; shells, trusts and nominees classified as institutional (NOGA 642) cut the other way, and the true household share is unidentified.

Evidence. Zucman2015Appendix, 'Official Swiss National Bank data' (sector breakdowns 'deeply misleading'/'almost meaningless'; 'I treat Middle-East central banks / SWFs as private wealth'). SNB cube bawebedomsecwja (non-resident holders, all currencies, foreign issuers): private customers 63.7% (1999), 48.3% (2005), 29.7% (2013), 23.8% (2017); institutional investors 26.4%, 42.0%, 64.4%, 72.1% (scripts/snb_sectors.py; reproduces Schmitz and Auditor A). Schmitz (2021), ISI World Statistics Congress IPS214, Table 1 (2014): AJZ $2.34tn; reported household holdings (SNB/BIS) $0.72tn; 55% household share $1.27-1.29tn. Auditor A: 50% / 25% household shares of commercial and institutional custody lower the May-2013 Swiss total by 29% / 44%. Faye, Godar, Moura & Zucman (2025) adjust for unreported reserves and sovereign assets, Cayman-listed firms and Irish fund holdings, still assign the residual to households, and find about 10% of world GDP.

Why this severity. The Swiss component could be overstated by up to ~45% under a published alternative, which affects the 'one-third in Switzerland' split and the regional allocation, but the world total rests on the global gap, where the main known non-household components were later adjusted without changing its order of magnitude.

The authors' response. The appendix argues that commercial/institutional accounts are mostly wealth-holding companies, trusts, nominees and sham corporations; the post-2005 shift coincides with the EU Savings Directive, which applied only to individuals. Faye et al. (2025) use centre-specific household shares for deposits but still attribute the residual gap to households; no direct reply to Schmitz was located.

10minorconfirmed · newly identified here · internal inconsistency

Some of the regional percentages are computed inconsistently; for Europe the slip makes the share of hidden wealth look slightly smaller (10% rather than 11%). These are bookkeeping errors that do not change the book's conclusions.

Technical detail and evidence

Why it matters, in detail. Europe's offshore share is biased down by ~1.1 points by the double count; regional denominators overshoot the world denominator by ~4%; measured against gross household financial assets rather than net financial wealth the world share would be closer to 6%. All are definitional or bookkeeping effects of second order.

Evidence. Zucman2015TablesFigures.xlsx Tab1!J5 = 23,980 ('Net onshore financial wealth'); Data-Fig3!H9 = H10 + Data-Fig1!P14 = 2,600 + 21,379.5; Tab1!C5 = 0.0987 (corrected 0.1094); Data-Fig3!G10 = 0.110; sum Tab1!J5:J12 = 94,456, less 2,600, plus 7,600 = 99,456 (scripts/headline_checks.py). Same J5 in JEP 2014 Table1. University of Chicago Press book page (blurb).

Why this severity. Shifts shares by about one point at most and in both directions (Europe's share is understated); no effect on the world total or the revenue arithmetic.

11minorconfirmed · newly identified here · arithmetic or transcription

There are several small copying and labelling mistakes in the spreadsheets and appendix, such as a table in euros that adds up to the dollar total. They do not change the book's findings but show the figures were not always carefully updated between editions.

Technical detail and evidence

Why it matters, in detail. Item (c) changes the meaning of the '80% undeclared' stock (about a tenth is untaxable Gulf wealth) without affecting revenue, since Gulf tax rates are zero; the others are labelling, vintage or comparison slips.

Evidence. Zucman2017LivreTableauxGraphiques.xlsx Tab1(FR) G47 '=250+600+500+200+750' vs C47 = 2,100; Zucman2015TablesFigures.xlsx Fig2 J14 '=600/1800*2300'; Data-Fig4_Tab1!H18 = 0.2 vs Données-T3!H18 = 1; Fig4!J13, J26; Zucman2015Appendix, Luxembourg section; WDI 2012 GNI/GDP (scripts/lux_rank.py: Puerto Rico 0.670, Cayman 0.677, Equatorial Guinea 0.686, Liechtenstein 0.698, Côte d'Ivoire 0.714, Luxembourg 0.768).

Why this severity. Presentation and documentation slips in auxiliary material; none changes a core number.

12minorconfirmed · newly identified here · arithmetic or transcription

A small arithmetic slip in one 1945 figure, too small to matter.

Technical detail and evidence

Why it matters, in detail. The 1940s level is understated by CHF 0.22bn (0.98%), negligible next to the stated coverage uncertainty.

Evidence. Zucman2015Appendix, 'Data from the Volcker and Bergier Commissions': 'I assume that the ICEP figure covers 90% of all customers securities deposits in 1945, and retain CHF 21.84bn'. 19.850884/0.9 = 22.057; 19.850884 x 1.1 = 21.836.

Why this severity. A 1% slip in one historical data point, inside a coverage range (82-100%) the appendix itself discusses; no effect on modern figures.

13minorlikely · newly identified here · documentation gap

The book says a different way of taxing multinationals would raise about 20% more corporate tax, but it never shows the calculation. The number can be roughly rebuilt from the book's other figures, so it is plausible, but it is only as good as those figures.

Technical detail and evidence

Why it matters, in detail. The +20% can be reconstructed as h/(1-h) with h = haven share of profits, but it then inherits the DII-based overstatement of h (F2); with a ~35% haven share of foreign profits (h of about 11%) the implied gain would be nearer 12%.

Evidence. Search of all published workbooks and appendices (Auditor A; confirmed for the TablesFigures workbooks by the adjudicator). scripts/corporate_checks.py: 0.174/(1-0.174) = 21.1%; 130/474.3 = 27.4%; 130/(474.3+130) = 21.5%.

Why this severity. A secondary policy figure that is roughly derivable from the book's own numbers; the issue is documentation rather than a demonstrated error.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

The 2017 edition's $8.7tn world total is, like the 2015 one, simply 8% times a constructed denominator. — The 2017 workbook note (Tab4!A24) states that the $8,700bn total is 'taken from AJZ (2017) Appendix I'; the cell formula coincides with 0.08 x 108,252 = 8,660. The constant-share criticism (F3) is upheld for the 2013 French and 2015 English editions, not as a demonstrated construction for 2017.
The English appendix's '20-25%' is the JEP's all-wealth declared share, so calling it an upper bound because shell accounts are excluded double-counts the shell argument. — Not established: 20-25% also matches the unweighted 2011 country average of the directive-covered declared share in the Johannesen-Zucman data (21.7%; 22.8% for treaty countries), in which case the upper-bound reasoning is coherent. The statement is outdated (weighted 26% in 2011, ~48% in 2013) rather than demonstrably double-counted; the substantive concern is retained in F4.
With a counterfactual equal to US firms' worldwide effective rate (19.9% minus 3%), the profit-shifting cost is $62bn, and $39bn combined with the Blouin-Robinson share. — Not adopted as a central alternative: the average effective rate is itself depressed by the shifting being measured, so it is a lower-end counterfactual. F2 uses a 25-35% counterfactual range ($58-128bn).
The 2017 regional changes (e.g. Russia $0.2tn to $0.6tn) come from assumptions, not new measurements. — Partly rejected: the Russian revision uses Novokmet, Piketty & Zucman (2017), and the 2017 notes disclose the changes; the Europe -10 / Asia +3.5 / Africa +3.5 point shifts are judgmental (retained in F7).
Henry (2012) / Tax Justice Network (2014), corpus offshore-havens-01: offshore financial wealth is $21-32tn, so the book's estimate is far too low. — Examined by both auditors and the adjudicator and not accepted as decisive: the book addresses Henry directly (FR 2013 section '7 500 milliards ou 21 billions ?'; FR 2017 '7 900 ou 21 000 milliards ?') and frames its figure as financial wealth only; Henry's figure rests on private-bank data not published in full, and independent mirror-statistics estimates (Pellegrini et al. 2016: $6.1-7.1tn for 2013) and the authors' 2025 reconstruction (~10% of world GDP) lie near the book. The valid kernel (non-financial assets and nominee holdings in reporting countries are missed) is an acknowledged caveat.
Sternberg (2015, Wall Street Journal): the book assumes all offshore wealth is untaxed, puts 8% of total household wealth offshore, and says US firms paid the US Treasury $130bn less. — Misreadings, correctly rebutted in the author's reply of 20 Oct. 2015: the English model treats 20% as declared (Data-Fig4_Tab1 column H), the 8% refers to net financial wealth (Figure 3; appendix), and the $130bn is a loss to all governments (book p.105 as quoted). The review's tariff-incidence point is upheld in F9.
Corpus statement (offshore-havens-02, via Vellutini et al. 2019) that the 2017 French edition assumes a 6% return. — The 2017 workbook uses 6.5% (DataTab4!O3 = 0.065) with 25% declared, giving $170.3bn (EUR 154.8bn); the corpus' second-hand figure is incorrect.
Pellegrini, Sanelli & Tosti's $20-49bn and Blouin & Robinson's $80bn are the correct replacement figures for the book's $127bn income-tax loss and $130bn corporate loss. — Neither auditor adopted them as point replacements, and the adjudicator agrees: PST's yields exclude taxable realized gains and are acknowledged as downward-biased, and the B&R equity-income subtraction may remove genuine foreign-to-foreign shifting (Clausing 2020). They bound the plausible range in F1 and F2 rather than replace the book's figures.

Audit notes

What reproduced. Reproduced by the adjudicator from the author's workbooks and primary data: the world total ($7.6tn = 8% x $95.5tn) and its denominator; the updated securities gap ($6,185bn, 2013) and a bottom-up total of $7.69tn (8.05%) using the book's deposit figure (range $6.8-8.2tn); the $189.39bn revenue model and its sensitivities (r = 6%: $158bn; 4%: $126bn); the Swiss monthly series (USD +17.1%, EUR +13.7%, CHF -8.6% from April 2009 to June 2014; peak April 2008); SNB custody sector shares; the corporate figures (55.8%, $128bn, effective rate 29.3% to 19.9%) and the BEA 2011 haven tax ratio (3.00% on net income, 10.59% excluding equity income); the tariff table (Switzerland passes only at >= 55% exporter incidence; Hong Kong fails); Europe's double-counted share (9.87% vs 10.94%); FR 2013 and FR 2017 revenue totals. Not reproducible or not measured: the global growth figures ('close to 25%', 'over 25%', '+28%') are not computed from measured data; the joint '$2.3tn and +18%' Swiss claim on the April 2009 value in the author's JEP Table S.1 (+9%); FR 2017's '+25%' Swiss growth (+7% USD, +32% EUR on the author's own figures); the '+20% corporate revenue' (no published computation; ~21-27% derivable); the derivation of the $1.5tn deposit component (not in the workbooks). Not attempted by anyone: rebuilding the global gap from raw IMF CPIS/EWN/TIC data, and the 2017 edition's new leaks chapter.

Limits of this audit. (1) The book's chapters were not read by the adjudicator: the English edition is paywalled; the adjudicator used the author-hosted English introduction (pp.1-7), the French 2017 preface and introduction (pp.7-15), tables of contents, the FR 2013 and EN 2015 online appendices, all published workbooks, the author's 2015 slides and WSJ reply (which quotes book pages) and the publisher's page. Auditor B used a complete copy from an unofficial mirror; the adjudicator did not use it, and details resting only on that copy (e.g. the book's justification of the return on pp.49-50) are flagged as unverified; that URL should not be cited in publication. (2) The global asset-liability gap was not rebuilt from raw IMF CPIS/EWN/TIC data; the author's 2013 and 2015 vintages and later reconstructions were compared instead. (3) The 2017 edition's new chapter on the HSBC and Panama leaks (evasion as 'un sport de (très) riches', drawing on Alstadsæter, Johannesen & Zucman) was not available; Dutch evidence (Leenders et al. 2023, offshore-havens-04) suggests that such concentration may not generalize, but this was not checked against the chapter. (4) The global capital-tax and financial-registry proposals are qualitative in the available materials and were not assessed quantitatively. (5) The published 2025 version of Blouin & Robinson was not read (2020 SSRN figures used); the Banque de France evidence is cited through Schmitz (2021). (6) The 'UpdatedZucman2014JEPdata' file behind the 2015/2017 Swiss totals is not public, so the Swiss level/growth pair could not be reconciled exactly. (7) SNB and WDI data have been revised since 2013-15; small differences reflect revisions. (8) Tariff incidence and counterfactual corporate tax rates are judged from general evidence, not Swiss- or firm-specific estimates. (9) Later evidence that automatic exchange of information cut offshore evasion after 2017 (Global Tax Evasion Report 2024; disputed by TJN, offshore-havens-07) concerns the post-publication period and was not scored against the book.