Annette Alstadsæter, Sarah Godar, Panayiotis Nicolaides, Gabriel Zucman · EU Tax Observatory report, October 2023

Global Tax Evasion Report 2024

24 problems found: 3 major · 7 moderate · 14 minor

What the paper claims

Thanks to the automatic exchange of bank information, offshore tax evasion declined by a factor of about three in less than 10 years: households still hold the equivalent of about 10% of world GDP in offshore financial wealth, but only about 25% of it now evades taxation.
Profit shifting to tax havens remains large, with no discernible effect of BEPS or the US Tax Cuts and Jobs Act so far.
The 15% global minimum tax has been 'dramatically weakened' since the 2021 political agreement by a growing list of loopholes.
Global billionaires have personal effective tax rates equivalent to 0-0.5% of their wealth (US about 0.5%, France about 0%), and counting all taxes their rates are 'significantly lower than those of all other groups'.
A global minimum tax on billionaires equal to 2% of their wealth would raise 'close to $250 billion' a year from fewer than 3,000 individuals.
Secondary: new forms of tax competition are emerging: preferential personal tax regimes in Europe and a green-energy subsidies race.

Every problem found

1majorconfirmed · newly identified here · version instability

The report says profit shifting costs governments nearly 10% of the world's corporate tax revenue and has not declined. For 2021-22 these numbers were largely assumed rather than measured, and the same team's later, more complete estimates put the cost at about 7% and show a modest fall after 2018. The central fact, that roughly $1 trillion of profits is booked in tax havens each year, still stands.

Technical detail and evidence

Why it matters, in detail. In the workbook, 63% of 2022 shifted profits (non-US multinationals) is fixed at the 2020 rate by assumption, and the 2022 revenue denominator is an assumed figure 21% below the later-measured $3,488bn; the projection therefore produced 'constancy' and a high loss share mechanically. In the later vintage the 2022 loss share is 6.9% (-2.7pp, -28%); the 2016-20 losses also fall by 0.8-1.6pp because the revenue series was revised up by about 7-15%; the shifted share is 31% rather than 35%; and US multinationals' share falls from 51% (2018) to 39% (2022) instead of staying near 46-50%. The dollar amount shifted is little changed (-4%).

Evidence. Workbook 2-profit-shifting.xlsx 'Data Fig 2.7-2.8' rows 45-52 (B51 =96510*0.84, D51 =0.2, D52 =0.19, E51 =I51*C51, I51 =I50, AO51 =AO50, BA51:BA52 typed values) and 'Tab 2.2' rows 10-24 (I24 = J24 = H24 = 0.3093); report pp. 8-9, 36, 39 (fn 56), 46-50, Tables 2.1-2.2, Figure 2.8 note; GWZ (2026) abstract, s.4, Table 1 and Table A1 (re-extracted by the adjudicator). The report's own 2022 dollar loss ($265.6bn) divided by GWZ's 2022 revenue gives 7.6%.

Why this severity. The $1tn level and the 'large and persistent' message survive (GWZ: $0.96-1.04tn; 31-34% of foreign profits), but the headline revenue-loss share is about 30% lower in the authors' own later series (6.9% vs 9.6% in 2022; about 8% vs 9.3% on average in 2016-20), and 'stagnation' becomes a modest decline. That is a change of a quarter or more in a headline level.

The authors' response. The report labels 2021-22 'preliminary and subject to revision' (Figure 2.8 note) and says policies may have prevented even higher shifting (p. 9). GWZ (2026) is the authors' own update; it still describes shifting in 2023 as 'at a historically high level', two points below 2015.

2majorconfirmed · previously known · misrepresentation of source

The report says the global minimum tax on companies was weakened 'by a factor of two' after the 2021 deal by a growing list of loopholes. But the largest of these, an exemption tied to companies' real activity, was already written into the October 2021 agreement, and the authors had written about it at the time. Measured against what was actually agreed, the report's own model shows a smaller loss: about 40% rather than half.

Technical detail and evidence

Why it matters, in detail. In datafig3, a 15% tax without carve-out yields 9.47% of 2023 corporate tax revenue, the October 2021 design (15% with year-1 carve-outs; $220bn, matching the OECD's January 2023 central estimate, which also includes the carve-out) 7.73%, and the final scenario 4.81%. Relative to the 2021 design the post-agreement reduction is 1.61x (-38%), not 1.97x (-49%); with the US-step correction (F4) it is 1.31x (-24%). The '3x' versus a 20% tax is arithmetically 3.47x but compares the outcome with a rate that was never agreed.

Evidence. OECD/G20 Statement on a Two-Pillar Solution (8 Oct 2021), 'Carve-outs' (p. 4); BCNZ (2022) p. 696 and fn 17; report pp. 6, 9, 37, 51-52, 54, 74, fn 70, fn 107; Bunn (Tax Foundation blog, 15 Nov 2023). Workbook 0-summary datafig3: 15% without carve-out 9.47% of 2023 corporate tax revenue, with year-1 carve-out 7.73% ($220bn), final 4.81%.

Why this severity. The carve-out step (-1.74pp) is 37% of the claimed 4.66pp weakening. Measured against the October 2021 design, the report's own model gives a reduction factor of 1.6 (-38%) rather than 2 (-49%); combined with F4 it is about 1.3 (-24%). The direction survives (post-2021 provisions do cut revenue), but the size of the 'weakening since 2021' is overstated by well over a quarter.

The authors' response. None located. The report's normative position is that the carve-out is a loophole whatever its date, and its arithmetic relative to a loophole-free 15% tax is correct.

3majorlikely · previously known · speculative assumption

The 'three-fold fall' in offshore tax evasion comes from measurements made in Denmark, which has an unusually effective tax administration, applied to the whole world. The calculation leaves out money held in the United States and by residents of countries that do not swap bank information, and assumes that everyone whose account is reported pays the tax due. Evasion clearly fell a great deal, but the true fall could be closer to two-fold than three-fold.

Technical detail and evidence

Why it matters, in detail. s(2022) = u*nc_u + (1-u)*nc_r = 0.30*0.9 + 0 = 27% (3.2% of world GDP); the decline factor 0.80/s is 2.96 against the workbook's 2013-15 value (2.65-3.07 in % of GDP depending on the base year). Treating US-located wealth (10%) and non-CRS residents (2.4-6.2%) as unreported with 50-90% non-compliance raises s to 30-37% (factor 2.2-2.7); the Danish lower-bound 16% evasion on CRS-reported income alone gives 38% (2.1); weaker reporting and enforcement for the ~30% owned by low- and middle-income residents gives 30-41% (1.95-2.6). Conversely, the later 17% value-weighted non-reporting share alone gives 15% (5.3), and 28.6% (2.8) combined with the 16% evasion rate. The headline is a point inside a ~2-5x band whose omitted channels mostly lean toward the lower half; untaxed offshore wealth in 2022 is plausibly 3.2-4.6% of world GDP rather than a precise 3.2%.

Evidence. Report pp. 27-29 and fn 27; Figure 1.4; workbook 1-offshore-wealth.xlsx dataFig1.4 G23:G25 (=70%+0.1*0.3), H23:H25, F23:F25; dataFig1.3 C23 (LMIC share 29.8%); Online Appendix Ch.1 s.2 (non-participating list). Boas, Johannesen, Kreiner, Larsen & Zucman (NBER w32714, rev. Oct 2024) s.6.3 ('evasion rate of 16% ... a lower bound'), s.7.1 (extrapolation caveat), s.7.2 ('17% of the foreign wealth ... lower than the 30% when counting individuals'; low match rates for non-OECD non-havens), s.8 (64-73% of the gap closed); FGMZ (2025) s.3.2 (US share 2% to 10%); Bachas et al. (2026) abstract (Colombia: CRS deterrence plus enforcement; FATCA gives 'limited visibility into U.S.-held assets'). Adjudicator recomputation (scripts/reproduce_core.py): coverage-gap variants give 30-38% untaxed (decline factor 2.1-2.7 against the workbook's 80% for 2013-15); weaker LMIC compliance gives 30-41% (factor 1.95-2.6); 16% evasion on reported wealth alone gives 38% (2.1); the report's own high-end gives 37% (2.2) and low-end 15% (5.3).

Why this severity. The direction of C1 is supported by well-identified country evidence (the CRS closed 64-73% of Denmark's offshore tax gap; Norway and Colombia show deterrence and compliance effects), so the finding survives qualitatively. Its global magnitude, however, rests on transplanting one high-capacity country's parameters, and within defensible alternatives the decline factor falls to about 2-2.7, more than a quarter below 'about three'; one later revision (17% value-weighted non-reporting) pushes the other way.

The authors' response. The report calls its estimates 'tentative and provisory', labels the 30% 'provisional and subject to revision' (fn 27) and shows 15%/37% scenarios. The authors' later Danish paper (Boas et al. 2024/2026) finds that the CRS closed about 70% (64% under actual enforcement) of Denmark's offshore tax gap, consistent with a factor of about three for Denmark; no global validation has been published and no direct reply to TJN (2024) was located.

4moderateconfirmed · newly identified here · internal inconsistency

One step in the report's chart subtracts revenue for 'exempting US companies' that its own starting calculation had never counted, so the same thing is taken off twice. Fixing this raises the final estimate from 4.8% to about 6% of world corporate tax revenue, so the minimum tax was weakened less than the chart shows.

Technical detail and evidence

Why it matters, in detail. Consistent treatment gives a final 5.91% of corporate tax revenue ($168bn) within the IIR family (US step = 0), or 6.33% ($180bn) within the full-domestic-coverage family (246 -> 200 -> credits), versus 4.81% ($137bn), i.e. +23-32%. The 'factor of 2' becomes 1.60 (IIR family) or 1.80 (full family, against its own 11.4% no-carve-out benchmark); relative to the 2021 design it becomes 1.31-1.61.

Evidence. Workbook 0-summary datafig3 C21, D24, D28, D29, D6 (=D5*D29/D28) and its cell note; Online Appendix pp. 6-7; Figure 3 note (p. 10). EU Tax Observatory tax-deficit-simulator saved outputs HQscenario_firstYearCO.xlsx and partialCoop_EUandOthers_{no,20}StatRateCond.xlsx (column sums re-computed in scripts/pillar2_scenarios.py and reproduce_core.py).

Why this severity. Final revenue is understated by 1.1-1.5pp of corporate tax revenue (23-32%), which moves the 'factor of 2' to about 1.6-1.8 against a loophole-free 15% tax. The qualitative point survives, because in a full-implementation world the US safe harbour does remove domestic UTPR revenue.

The authors' response. The Online Appendix (p. 7) calls the treatment of domestic deficits 'simplified and subject to revision'; no reply located.

5moderateconfirmed · newly identified here · speculative assumption

Part of the claimed weakening comes from a rule that lets some company tax credits count as income rather than as tax cuts. The rule is real, but the report simply assumes it is worth the same as lowering the minimum rate from 15% to 13%, without data to back that size. If the effect is smaller, the weakening is smaller too.

Technical detail and evidence

Why it matters, in detail. Without the haircut the final is 6.29% (with it 4.81%); with half the effect about 5.5% (other steps as in the report). On a consistent basis (F3-F4) the post-2021 reduction factor is 1.31 with the assumed credit effect and about 1.1-1.2 with half of it.

Evidence. Online Appendix p. 6; report pp. 53-54 and Figure 3; workbook datafig3 D25 (=143), D7 (=D6*D25/D24), cell note I7; saved simulator output HQscenario_firstYearCO.xlsx sheet '13%' (142.2bn).

Why this severity. The channel is real (GloBE treats qualified refundable and marketable transferable credits as income rather than as tax reductions), but its size is assumed; it moves the final figure by up to a third and, with F3-F4, determines whether the post-2021 weakening is ~1.3x or ~1.1x. It is disclosed as provisional.

The authors' response. The Online Appendix flags the assumption as 'provisional and subject to revision'; no later calibration located.

6moderateconfirmed · newly identified here · internal inconsistency

The summary says a 2% minimum tax on billionaires would raise 'close to $250 billion' a year, but the report's own table, printed on the next page, says $214 billion. The larger figure is roughly what the tax would raise before subtracting the taxes billionaires already pay.

Technical detail and evidence

Why it matters, in detail. 2% x $12,916bn = $258.3bn gross, minus $44.3bn of current personal taxes (0.5% of North American and 0.25% of other billionaire wealth) = $214.0bn net. The headline exceeds the net figure by $36bn (16.8%).

Evidence. Report pp. 7, 12-13, 73 (Table 5.1), 76-77 (Table 5.3); workbooks 0-summary Table2 F12 = 213.975 and 5-policies '5.3 - wealthtax' G12 = 258.32, H12 = 213.975; recomputed in scripts/reproduce_core.py.

Why this severity. The headline overstates the report's own static estimate by 17%; on its own this is below the 'quarter' threshold for major, but it compounds F8.

The authors' response. The later G20 blueprint (Zucman 2024) gives $200-250bn using 2024 Forbes wealth ($14.2tn) and a 0-20% avoidance range; $250bn thus becomes attainable only with 2024 wealth and no avoidance.

7moderateconfirmed · previously known · overclaiming

The finding that billionaires pay personal taxes worth only 0-0.5% of their wealth comes from two countries, the US and France, and later official tax data have confirmed it for both. But the report presents it as true of billionaires worldwide, shows an estimated (not measured) Dutch figure as if it were data, and overstates how far US billionaires' total tax rate falls below that of every other income group.

Technical detail and evidence

Why it matters, in detail. Personal-tax-to-wealth ratios are observed for about 30% of billionaires (US and France); the 'global' 0-0.5% assigns them to the rest. IRS-based evidence confirms the US level (0.4-0.6% of wealth; total rate 24% vs a 30% population average). The 'all other groups' comparison holds clearly for France (26% vs 46-54%) but in the US is within 0.5-3pp of the bottom deciles on the SZ (2019) curve; under conventions that count refundable credits as negative taxes, bottom-quintile rates fall by at least 17pp (Splinter 2020, in his 2010 comparison), below the billionaire rate.

Evidence. Report pp. 6, 11-12, 66-70; workbook 0-summary Table2 J4:J6 notes; 4-progressivity datafig4.1 (US P30-40 0.2352, P10-20 0.2425, billionaires 0.2304; NL billionaires 0.18); CPB Discussion Paper 436 ('only 21% for the top 0.01%'); Forbes 2021/2023 lists (US 724/734, France 42/43, Netherlands 12 of 2,754/2,640); Balkir, Saez, Yagan & Zucman (2025), NBER w34170, abstract and s.3.3 (top-400 individual income taxes 0.6% of wealth, top 100 0.4%; total 24% vs 30% for the population); Splinter (2020), NTJ 73(4) (corpus us-wealth-taxes-12/-16).

Why this severity. The magnitudes for the two observed countries are corroborated by later administrative evidence, so C4's core numbers survive for the US and France. The issue is the extension to 'global billionaires' and to 'all other groups', which affects how far the claim reaches rather than the numbers themselves.

The authors' response. The report notes that the evidence 'remains at this stage very limited globally' (p. 67). Balkir, Saez, Yagan & Zucman (2025) later confirm the US magnitudes with IRS data, and Bach et al. (2025) the French ones.

8moderateconfirmed · previously known · misrepresentation of source

The chart says it counts all taxes for everyone, but the French billionaire figure leaves out VAT and some business and property taxes that are counted for other French groups. French billionaires still pay much less than other high earners, but the gap shown is somewhat larger than a like-for-like comparison gives.

Technical detail and evidence

Why it matters, in detail. Like-for-like, within IPP Note 92, the rate falls from 46% (top 0.1%) to 26% (top 0.0002%); the figure instead juxtaposes 50% from the DINA profile with 26% from the narrower direct-tax concept. Allocating production taxes (and VAT on billionaires' small consumption) to the billionaire point would raise it by an unknown, probably single-digit, number of points, narrowing but not closing the gap.

Evidence. Report pp. 67-68 and Figure 4.1 note; workbook 4-progressivity datafig4.1 C19 (0.5026), C21 (0.26); Bach, Bozio, Guillouzouic & Malgouyres (June 2023), Note IPP 92, pp. 3-4; Bach et al. (2025) (45% at the top-0.1% threshold falling to 25% for billionaires, same direct-tax scope).

Why this severity. The collapse of French billionaires' tax rates is real within the IPP's own definition (46% for the top 0.1% to 26%), so the qualitative result stands. But the plotted drop of about 24 points mixes concepts, an undisclosed caveat in a headline figure with a modest, unquantified effect.

9moderatelikely · previously known · selective presentation

The report says new company-by-company tax data confirm that about 35% of multinationals' foreign profits are shifted to tax havens. Some studies using those data agree, but a major study using confidential reports for more than 3,600 firms finds 13-22%, and the report does not mention it. Profit shifting is real, but its exact size is disputed.

Technical detail and evidence

Why it matters, in detail. The C2 level applies TWZ's local-firm profit-to-wage counterfactual to FATS and balance-of-payments data. Regression or payroll counterfactuals on country-by-country micro-data give 13-22% of foreign profits; Garcia-Bernardo & Janský (2024) obtain more than $850bn for 2017, close to TWZ. Revenue-loss estimates differ less (Fuest et al. extrapolate about EUR 162bn, ~12% of large multinationals' taxes, versus TWZ's ~10% of global revenue).

Evidence. Report pp. 39, 43 and fn 62-63; Online Appendix Ch. 2 s.1; GWZ (2026) Table A1 notes (US foreign profits = BEA direct-investment equity income); corpus entries offshore-havens-08, -09, -10, -11, -12 (Fuest et al. WP pp. 25-26: 13% and 22%); Blouin & Robinson (May 2020 WP) p. 16.

Why this severity. This bears on the level in C2: the share of foreign profits shifted is method-dependent by up to a factor of about 2-3. It bears less on the revenue at stake, which converges more across methods. The dispute is unresolved, and the report's omission makes its 35% look more corroborated than it is.

The authors' response. TWZ (2023, Table 2) report robustness bounds ($506bn, $507bn, $369bn for 2015); Garcia-Bernardo & Janský (2024) find comparable magnitudes with country-by-country data. No direct reply to Fuest et al. or Blouin & Robinson located.

10moderatelikely · previously known · overclaiming

The revenue figure assumes billionaires would not avoid the tax at all, and it uses 2021 fortunes, which were higher than in 2023, so it is not 'conservative' as the report says. Using the avoidance assumptions the authors use in other work, the tax would raise roughly $150-200 billion a year: still a large sum from fewer than 3,000 people.

Technical detail and evidence

Why it matters, in detail. With April-2023 Forbes wealth the static figure is $202bn. Applying WIR 2022's benchmark haircuts (base x 0.9 x 0.85) gives $153bn (current taxes unscaled) to $164bn (scaled); a 20% revenue loss gives $171bn, 15% about $182bn. A plausible October-2023 range is about $150-205bn. Further downside: Forbes-to-tax-base valuation gaps (estate-tax values of Forbes decedents are reported at about half the Forbes estimate). Small upside: heterogeneity in current taxes, since the tax is a floor.

Evidence. Report p. 77 and Table 2/5.3 notes; Forbes 'World's Billionaires' facts pages 2021 and 2023 (Wayback copies); WIR 2022 Full Report ch. 7 ('benchmark estimates ... take into account tax evasion and depreciation (respectively 10% and 15%)'); Zucman (2024) G20 blueprint s.2.3 ($242bn and $193bn). Recomputation in scripts/reproduce_core.py: 2023 wealth $202bn; WIR haircuts $153-164bn; 20% loss $171bn; 2024 wealth $235bn.

Why this severity. The authors' own conventions lower the $214bn by about 5-28% (to roughly $150-205bn). The claim that a 2% tax would raise large sums from a few thousand people survives; combined with F7, the '$250bn' headline is about 20-60% above such estimates.

The authors' response. Report p. 78 argues that mobility is 'moot' under a global agreement and that private firms can be valued with listed-peer multiples. The 2024 G20 blueprint later adds a 0-20% avoidance range.

11minorconfirmed · newly identified here · documentation gap

The year-by-year path of how much offshore wealth was hidden before the reforms is typed into the spreadsheet without explanation. That makes the timeline hard to check, though the assumed pre-reform levels match the older evidence.

Technical detail and evidence

Why it matters, in detail. The decline factor is 2.65 (from 2013), 2.91 (2014) or 3.07 (2015) in % of GDP, 2.96 as a share of offshore wealth and 2.0-2.3 in current dollars.

Evidence. Workbook 1-offshore-wealth.xlsx dataFig1.4 G4:G22 (typed constants); Figure 1.4 note ('See text for the construction'); scripts/reproduce_core.py.

Why this severity. The assumed pre-reform levels are consistent with the cited 85-95% pre-2010 evidence (the 80% for 2013-15 is, if anything, conservative for the decline), so the omission affects transparency and timing, not the headline.

12minorconfirmed · newly identified here · version instability

The report gives two different sizes for offshore wealth today (10% and 12% of world output), and a later update by the same team puts 2022 at about 10%. This is a presentation problem; it does not weaken the finding that evasion fell.

Technical detail and evidence

Why it matters, in detail. With the FGMZ (2025) level, 2022 untaxed wealth is 2.8% rather than 3.2% of GDP; 2015-21 levels are also 6-13% lower, so the ratio is broadly preserved.

Evidence. Report pp. 7, 21, Figures 1 and 1.4; workbooks 0-summary datafig1 F25 vs 1-offshore-wealth dataFig1.4 J25; Online Appendix p. 1; FGMZ (2025) Figure 1 (digitised).

Why this severity. Using the later series consistently leaves the decline factor at about 3.1-3.3, so the revision does not weaken C1; it is a presentation and vintage issue.

The authors' response. The 2022 projection method is disclosed in the Online Appendix; FGMZ (2025) is the authors' own update.

13minorconfirmed · newly identified here · misrepresentation of source

One sentence says a quarter of all offshore financial wealth may have moved into property; the study it relies on found that a quarter of the money that left tax havens did, about 1.5-3% of the total. The report states it correctly later, and the slip does not affect its numbers.

Technical detail and evidence

Why it matters, in detail. 25% of the stock vs about 1.5% (global haven stock) or 2.6-3.0% (early adopters' stock); the real-estate leakage it implies is 8-17 times too large.

Evidence. Report pp. 20 and 33; Bomare & Le Guern Herry (2022), abstract and pp. 35-36 (Table 8 and back-of-the-envelope calculation).

Why this severity. An introductory sentence contradicted by the chapter's own correct statement; it does not enter any estimate and it cuts against the report's headline.

14minorconfirmed · previously known · misrepresentation of source

The report cites a Dutch study as showing that hidden offshore wealth belongs mostly to the very richest, but that study found it was spread more widely among the merely well-off.

Technical detail and evidence

Why it matters, in detail. Relevant to distributional uses of offshore wealth (top-share corrections), not to the aggregate untaxed share.

Evidence. Report p. 25 and fn 17; Leenders, Lejour, Rabaté & van 't Riet (2023), JPubE 217:104785, abstract.

Why this severity. Secondary claim about who owns offshore wealth; no effect on the C1 numbers.

15minorconfirmed · newly identified here · misrepresentation of source

The report describes counts of Dubai properties as counts of people, and an overall comparison as a person-by-person check. The study's main conclusion, that most Norwegian-owned Dubai property was not declared, still holds.

Technical detail and evidence

Why it matters, in detail. Counts are of properties, not taxpayers, and the comparison is aggregate; the source's bound (at least 70% unreported) is close to the report's 'three-quarters'.

Evidence. Report pp. 35, 87; Alstadsæter et al. (2022), EU Tax Observatory WP 1, pp. 17-18 and Figure 10.

Why this severity. A secondary illustration; the qualitative conclusion (most Norwegian-owned Dubai property is undeclared) is supported by the source.

16minorconfirmed · newly identified here · misrepresentation of source

The report says the US system (FATCA) does not require banks to report account balances. Foreign banks reporting to the US do; it is the information the US sends back to other countries that leaves balances out. The report's underlying point, that the US shares less than other countries, is right.

Technical detail and evidence

Why it matters, in detail. The asymmetry that matters for C1 is US reciprocity (no balances, no non-US-source income), which supports treating US-located wealth as weakly covered.

Evidence. Report p. 86; IRS Form 8966 (2022) Part IV; US Treasury Reciprocal Model 1A IGA (6 June 2014), Art. 2(2)(a)-(b).

Why this severity. An inaccurate institutional statement whose intended point is right; no effect on any estimate (it is relevant context for the US-located wealth gap in F1).

17minorconfirmed · newly identified here · arithmetic or transcription

The total amount of wealth reported under the bank-data exchange is quoted three different ways, and a claimed 'coincidence' between two numbers depends on mixing euros and dollars. This does not affect the report's estimates.

Technical detail and evidence

Why it matters, in detail. Currency and year slips; the 30% non-reporting parameter comes from transfer matching, not from this comparison.

Evidence. Report pp. 19 (fn 2), 27, 88; OECD Secretary-General Tax Report to G20 Leaders (Sept 2023) p. 11.

Why this severity. The report itself cautions against reading the coincidence as full coverage and does not use it quantitatively.

18minorconfirmed · newly identified here · internal inconsistency

A description of the European bank data contradicts itself about how many countries the banks come from. It does not affect the main results.

Technical detail and evidence

Why it matters, in detail. Sample-description error; no demonstrated numerical consequence.

Evidence. Online Appendix p. 4.

Why this severity. Affects only the description behind Figure 2.5 (European banks), not the global profit-shifting estimates.

19minorconfirmed · newly identified here · documentation gap

The table of what individual countries could raise on their own at a 25% rate was produced by scaling up the 15% figures rather than by running the model, and the report does not say so.

Technical detail and evidence

Why it matters, in detail. Mixing parent-country and market-country bases makes the 25% unilateral figures approximations of unknown error.

Evidence. Workbook 5-policies.xlsx '5.4 - sales' D5:D11 (=C5*G5/F5 etc.); report Table 5.4 (p. 84).

Why this severity. Secondary, unilateral-policy table; the accuracy of the approximation is unknown.

20minorconfirmed · newly identified here · arithmetic or transcription

A worked example writes billions instead of millions, one cross-reference points to the wrong table, and 'a factor of three' is really about three and a half. These are typos.

Technical detail and evidence

Why it matters, in detail. Unit and cross-reference errors; the factor-3 rounding understates the arithmetic ratio.

Evidence. Report pp. 6, 10, 82, 84; workbook datafig3.

Why this severity. Typographical; none affects a table.

21minorconfirmed · newly identified here · internal inconsistency

Two small numbers do not match: France's billionaire tax rate is given as 0.1% in one place and 0% in another, and the three countries studied hold about 30% of the world's billionaires, not 'almost 35%'. Neither changes the conclusions.

Technical detail and evidence

Why it matters, in detail. Using 0.1% for France (0.30% for non-North-American regions) gives $209.9bn.

Evidence. Report pp. 67, 70, 76; workbook 0-summary Table2 J4; Forbes 2021 and 2023 lists (cached JSON); scripts/reproduce_core.py.

Why this severity. Revenue changes by about $4bn ($210bn instead of $214bn); the population share is overstated by about 5pp.

22minorconfirmed · newly identified here · internal inconsistency

The chapter on special tax regimes for wealthy newcomers contains small inconsistencies: two countries' numbers are swapped in the text, two different 'average benefit' figures appear, and one currency conversion goes the wrong way. The headline cost of about EUR 7.5 billion a year is unaffected.

Technical detail and evidence

Why it matters, in detail. The EUR 34,300 average has no stated denominator; the aggregate average implied by the totals is EUR 28,392.

Evidence. Report pp. 11 (Table 1), 57-58 (Table 3.1); Online Appendix pp. 9-11; workbooks 3-tax-competition Fig3.1, Table3.1 and 0-summary Table1.

Why this severity. Secondary finding; the EUR 7.5bn total and the 260,000+ beneficiaries are unaffected.

23minorconfirmed · newly identified here · arithmetic or transcription

A spreadsheet adds one too many years of growth to EU corporate tax revenue when estimating the cost of green subsidies, and a date range is misprinted. The corrected figures round to the same percentages.

Technical detail and evidence

Why it matters, in detail. Correcting the exponents raises the EU central ten-year average from 12.99% to 13.38% and the US+EU average from 14.06% to 14.29% (Auditor B's computation; formula verified).

Evidence. Workbook 3-tax-competition.xlsx DataFig3.2 B4, K3, K20, K25; report p. 64.

Why this severity. The corrected shares round to the same 13% (EU) and 14% (US+EU) as the report states.

24minorlikely · newly identified here · data error

The 2015 figure for offshore wealth is an exact copy of 2014's, which makes 2015 look like a higher peak than it probably was. It exaggerates the decline slightly, and only if 2015 is used as the starting year.

Technical detail and evidence

Why it matters, in detail. The duplicate raises 2015 offshore wealth to 12.4% of GDP (about 11.0% on the FGMZ 2015/2014 profile) and untaxed wealth to 9.9% (about 8.8%) of GDP.

Evidence. Workbook dataFig1.4 K17 = K18 = 9299.69175913834; vector extraction of FGZ (2023) and FGMZ (2025) Figure 1 (Auditor A's compare_offshore_versions.py, re-run by the adjudicator).

Why this severity. It changes the decline only when 2015 is used as the base year (about 2.7x instead of 3.1x); from 2013-14 the factor is unaffected.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

(Part of Codex I3) The workbook's tax-credit haircut (23.5%) contradicts the chapter's statement that credits cut revenue 'by about 15%'. — The chapter's percentages are measured against the 15% no-carve-out revenue, not sequentially: carve-out 1.74/9.47 = 18.3% ('nearly 20%'), US step 1.45/9.47 = 15.3% ('about 15%'), credits 1.48/9.47 = 15.6% ('about 15%'). They are consistent with the workbook. The rest of Codex I3 (an uncalibrated assumption) is retained as F5, downgraded from major to moderate because it is disclosed as provisional and its direction is right.
(Part of Claude I19) By the report's own Danish breakdown only ~29% of CRS-reported value is household wealth, so read consistently the '$12 trillion coincidence' implies far lower CRS coverage than 70%. — Not established. In the Danish data about 4-5% of GDP is household, about 5% firms and the rest (about 5-6%) unmatched, so the household share lies between roughly 30% and 70% depending on who owns the unmatched reports; the report's 'at least 30%' discount is a lower bound consistent with that range. The argument is uninformative rather than inconsistent; the currency and year slips are retained in F18.
(Framing of Claude I9) The 2021-22 profit-shifting figures 'partly assume their conclusion'. — Not circularity in the rubric's sense: the non-projected 2015-2020 estimates also show a roughly flat 34-38% share, and the projection method is disclosed. The substantive finding, that the authors' later series revises the revenue loss and the trend, is retained as F2 (major).
(Part of Claude I1) US-located offshore wealth, being outside the CRS, should be treated as 90% non-compliant, giving 34.7-38.1% untaxed. — The 90% variant is not supported: Bachas et al. (2026) find that Colombians self-report more than eleven times the US wealth implied by FATCA, so lack of CRS reporting does not imply evasion. The 50% variant (about 30-31% untaxed, factor about 2.6-2.7) is the more defensible coverage-gap adjustment; F1 reports the full range.
(Part of Claude I15) The US personal-tax rate of 0.5% of wealth is a number 'set by judgment' and presented as an estimate. — Largely rejected as a substantive problem: the workbook note also cross-checks against DINA's 0.5% for the top 400 in 2021, and IRS-based evidence (Balkir, Saez, Yagan & Zucman 2025) finds individual income taxes of 0.4-0.6% of wealth for the top 100-400. It remains a documentation remark within F9.
(Lead examined and rejected by Codex itself) datafig4.1 row 20 (P99.99-99.999: 35.3% vs 30.36% in the SZ 2019 source) is an error in Figure 4.1. — Concur with Codex: Figure 4.1 plots bins only up to P99.9-99.99 and then the billionaire point, so row 20 is not used in any published figure.
(Corpus lead examined and rejected by Codex) offshore-havens-18: the 2% floor wrongly ignores corporate income tax already paid on retained profits attributed to billionaires. — Concur that this is not an error in this report. The proposal is explicitly a floor on personal taxes (Table 5.3 notes), and the $214bn is internally consistent with that design; whether corporate tax should be credited is a normative design choice. Revenue would be far lower if it were credited; final.md notes this sensitivity.
(Severity) Real-estate denominator error (Codex I7) is moderate. — Downgraded to minor (F14): it is one introductory sentence, the chapter states the finding correctly on p. 33, it enters no estimate, and it cuts against the report's own headline.

Audit notes

What reproduced. Reproduced exactly from the authors' workbooks with independent code (scripts/reproduce_core.py): the C1 scenarios (15/27/37%; 1.8/3.2/4.4% of GDP) and decline factors; the C2 2022 decomposition ($368.8bn US + $627.5bn non-US = $996.4bn, 35.2%) and tax-loss series; the C3 waterfall (16.7 -> 9.5 -> 7.7 -> 6.3 -> 4.8%); the C5 Table 5.3 ($214.0bn) and its WIR 2022 inputs; the C6 totals. Verified against sources: the October 2021 OECD statement (carve-out), the authors' Intertax paper, the Danish CRS study (Oct 2024: 17%, 16%, 64-73%), GWZ (2026) Table 1/A1, FGMZ (2025), IPP Note 92, Balkir et al. (2025), CPB (2022), Forbes 2021/2023 totals, Bomare & Le Guern Herry (2022), Leenders et al. (2023), Alstadsæter et al. (2022, Dubai), IRS Form 8966 and the Model 1A IGA. The simulator's saved scenario outputs (repository vintage, totals within about 1-2% of the report's inputs) confirm the scope inconsistency in the US step. Not reproduced: the global offshore stock from IMF/BIS/TIC inputs, the TWZ country-level profit-shifting and tax-loss construction, the Pillar Two simulator itself (not re-run), and any confidential micro-data (Danish CRS, French and US tax records). Discrepant with later vintages: the 2022 offshore level (+16% vs FGMZ 2025), 2015 offshore wealth (duplicate of 2014), the 2021-22 profit-shifting tax loss (9.6% vs 6.9% in GWZ 2026), and the Danish calibration inputs (30% vs 17%; 0% vs at least 16%).

Limits of this audit. The October 2023 working-paper version of the Danish CRS study cited by the report is not public; the earliest public versions (NBER w32714, July/Oct 2024) report 17% value-weighted non-reporting, so it cannot be ruled out that the 2023 draft said 'up to 30%' as quoted. The Pillar Two simulator was not re-run; the scope finding (F4) relies on the workbook formulas and the repository's saved scenario outputs, whose totals differ from the report's inputs by about 1-2%. The global offshore stock, the TWZ country estimates and the confidential Danish, French, Dutch and US micro-data were not rebuilt. The coverage-gap and developing-country sensitivities in F1 use AJZ 2007 country weights and assumed non-compliance rates; they are illustrative bounds, not estimates. The Blouin & Robinson (2025) journal article could not be downloaded (their 2020 working paper was read); the OECD's January 2024 Pillar Two estimate ($155-192bn, 6.5-8.1% of CIT) was read only through an EY summary and was not used to grade any issue. Several corroborating later studies (GWZ 2026, FGMZ 2025, Boas et al. 2024/2026, Balkir et al. 2025) are co-authored by Zucman; they use new data but are not independent of the report's team. Later evidence is used to test whether claims hold up, not to fault the authors for not knowing it in October 2023.