← All papers
Mona Baraké, Paul-Emmanuel Chouc, Theresa Neef, Gabriel Zucman · Intertax 50(10): 689-710
Revenue Effects of the Global Minimum Tax Under Pillar Two
14 problems found: 1 major · 5 moderate · 8 minor
What the paper claims
- A 15% minimum tax collected by headquarters countries (income inclusion rule, polar case) would raise about EUR 179bn a year globally without carve-outs, for 83 parent jurisdictions (2017 CbCR data complemented by TWZ, uprated to 2021 EUR).
- EU Member States could collect EUR 67bn a year from a 15% top-up tax without carve-outs (about 19% of EU corporate tax revenue).
- Substance-based carve-outs reduce revenues by about 14-22% over the full sample (EUR 139bn in year 1, 154.5bn in the long run); the year-1 figure is 'broadly in accordance' with the OECD's USD 150bn.
- Under the agreement (long-run carve-outs of 5% of payroll and tangible assets) the EU can expect EUR 55bn a year, about 16% of corporate tax revenue and 4% of health spending; 20-40% of EU revenue comes from undertaxed domestic profits.
- The distribution of revenue depends on who collects: under the IIR the US (EUR 54.4bn), Ireland (11.7), Germany (10.1), the UK (5.9) and Luxembourg (5.1) gain most; under host-country collection (QDMTT) low-tax hubs gain (Netherlands 14.1, Luxembourg 12.5, Cayman 11.4, Bermuda 8.1) and the US falls to EUR 3.4bn; G7 collect 89.5bn vs 17.4bn.
- Extending the minimum tax to large purely domestic EU groups (ORBIS, 170 groups) adds only about EUR 35m a year.
- (Online Appendix 2, illustrative) Behavioural responses relocate EUR 108-255bn of profits and lower IIR revenue (no carve-outs) to EUR 150.5-167.2bn while raising ordinary CIT in higher-tax countries (US up to EUR 15.3bn).
Every problem found
1majorlikely · previously known · methodology
About two-fifths of the EU's EUR 55 billion comes from supposedly low-taxed profits that big European companies record in their home countries. Those profits look low-taxed mainly because they include dividends passed between companies of the same group, and similar income the new rules ignore. Correcting for this alone brings the EU figure down to about EUR 42 billion or less.
Technical detail and evidence
Why it matters, in detail. Domestic cells' PBT includes non-GloBE income (exempt intra-group dividends, equity-method and disposal gains, owner-taxed flow-through income), so ETR_pp is biased down and the base up, and (15% - ETR) x (PBT - SBIE) is overstated. Direction: upward. Magnitude: EU long-run -24.6% (55.2 -> 41.6bn) under the authors' 0.6028 factor, and -38.5% (-> 33.9bn) if no domestic top-up were due. Germany -50% (10.1 -> 5.0bn); world -8.8% to -13.7%. The downward within-cell heterogeneity bias offers no demonstrated offset for domestic cells, and official estimates for DEU, NLD, AUT, DNK and BEL lie below the paper's domestic components alone.
Evidence. (1) Reproduced (scripts/run_sensitivities.py): with the authors' correction, EU 55.184 -> 41.582bn (-24.6%), EU domestic 21.24 -> 7.64bn, Germany 10.13 -> 5.05bn, world 154.53 -> 140.93bn (-8.8%). (2) National-base consistency test, 2017 (scripts/national_base_check.py; OECD Revenue Statistics code 1200 divided by the statutory rate). German MNEs' CbCR domestic profit (USD 205.1bn, tax paid 23.4bn, ETR 11.4%) equals 82% of Germany's implied corporate tax base, while their tax equals 31% of German corporate tax revenue. That implies a 112% ETR for the rest of the corporate sector. The same test gives Austria 52%, Denmark 53% and Sweden (unadjusted) 256%. Luxembourg MNEs' domestic profit exceeds the whole implied national base (110%). (3) Huber & Maiterth (2019, arqus DP 246, abstract) find German corporations' tax burden near the statutory ~30% once partnerships and the 95% dividend exemption (s. 8b KStG) are accounted for. The authors cite this paper (OA B.4, p. 15) but keep Germany's EUR 5.1bn domestic top-up. (4) Later official ex-ante estimates using tax-return data are smaller than the paper's domestic component alone. The German Minimum Tax Act bill gives +EUR 0.2bn full-year direct effect for the entire minimum tax (Bundestag Drucksache 20/8668, pp. 2, 88-90), against the paper's domestic 5.1bn. The CPB-certified Dutch estimate is EUR 0.47bn structural, including behavioural effects, with the domestic top-up estimated from 2017-2020 returns (CPB 2023, Tabel 2.1), against the paper's 2.0bn. A Tax Foundation compilation (not verified here) gives Austria EUR 0.1bn, Denmark DKK 2-3bn and Belgium EUR 0.33bn, against the paper's domestic components of 1.8, 1.0 and 0.9bn. (5) Later CbCR vintages are mixed: German domestic paid ETR 11.9% (2018), 13.2% (2019), 12.8% (2020), 16.0% (2021), 13.9% (2022), 22.4% (2023). Austrian (5.9-10.5%) and Luxembourg (0.7-2.5%) domestic ETRs remain implausibly low for 25% statutory rates, consistent with persistent exempt holding income. The current OECD release was used.
Why this severity. The domestic component is 38.5% of the EU long-run headline, and the authors' own correction removes 24.6% of the EU total. The national-base test and official estimates indicate this is not an over-correction for the largest contributors. The EU level (C2/C4) is therefore plausibly overstated by a quarter or more. The global total moves 9-14% and the qualitative conclusions survive.
The authors' response. Acknowledged in the paper. A rule-of-thumb correction (OA B.4, Table B.4.2) is presented as a 'lower bound' because some countries may already exclude dividends (fn 20). Low ETRs are argued to be plausible for Ireland, Luxembourg and Poland (OA B.4, pp. 14-15). No later reply located.
2moderateconfirmed · newly identified here · methodology
For Poland and a few smaller countries, the model applies the minimum tax to the profits of every locally owned company, small firms included. The tax only covers groups with more than EUR 750 million of turnover. On Poland's own data for large groups, its 'almost EUR 4 billion' (EUR 2.7 billion after exemptions) shrinks to a few hundred million.
Technical detail and evidence
Why it matters, in detail. Population mismatch: the GloBE applies to groups with at least EUR 750m of turnover, whereas the TWZ series covers the whole domestically owned corporate sector, including SMEs, and uses a national-accounts tax/profit ratio as the ETR. Upward bias of about EUR 3.26bn (5.9% of the EU long-run total; 2.1% of the world total). Poland's base is 7.5 times the domestic profit of its in-scope CbCR groups.
Evidence. TWZ2020AppendixTables.xlsx, sheet TableA6 ('Corrected corporate profits (2015)'): K35 = B35 - E35 = USD 68.40bn; L35 = TableA3!H35/TableA6!B35 = 10.004%. codes/2b_twz_localfirms.do imports columns A, K and L. data/TWZ/2017/twz_domestic.xlsx gives POL 74.491bn at 0.100041 (2015 x 1.089). Poland's own CbCR all-sub-groups domestic row for 2017 shows profit USD 9.95bn, tax paid 1.03bn and tax accrued 1.65bn (2016-17 paid ETR 12.1%; 2017 accrued 16.6%). Latvia's paid ETR is 17.5%. Removing all TWZ local-firm top-ups gives EU 55.18 -> 51.93bn (-5.9%) and Poland 2.70 -> 0.05bn. Using Poland's own CbCR row gives a domestic top-up of about EUR 0.2bn instead of 2.65bn (scripts/run_sensitivities.py, scripts/run_combined.py).
Why this severity. Changes Poland's country estimate by about 90% and the EU total by about 5-6%; the global order of magnitude is unaffected.
The authors' response. Generic caveat on p. 694 (the fourth upward bias). OA B.4 (p. 14) cites low backward-looking Polish ETRs (10.3%) as corroboration.
3moderateconfirmed · previously known · methodology
Where two data sources disagree, the model always picks the one that yields more revenue. France's EUR 3.6 billion comes mostly from this choice; France's own detailed company data give about EUR 1.2 billion.
Technical detail and evidence
Why it matters, in detail. A one-sided maximum over two noisy, conceptually different measures is upward-biased by construction. The two measures are booked low-taxed CbCR profit at observed ETRs, and TWZ shifted profit allocated by ultimate owner at an assumed 10% ETR. Total effect: +EUR 3.56bn on the world total, of which +2.41bn for France.
Evidence. Reproduced (scripts/pair_checks.py, run_sensitivities.py). France's long-run haven top-up is 1.07bn from CbCR against 3.48bn from TWZ. TWZ Table C4 allocates USD 47.5bn of profits 'shifted to Belgium' to French owners, whereas French CbCR shows USD 7.0bn in Belgium at a 21.7% ETR, which carries no top-up. French CbCR haven profits total USD 32.0bn against TWZ's USD 77.8bn. Without the rule: France 3.56 -> 1.15bn (-68%), Canada 7.62 -> 6.82bn (-10.5%), EU 55.18 -> 52.77bn (-4.4%), world 154.53 -> 150.97bn (-2.3%). Canada's CbCR has no individual haven partner except the Netherlands. Its havens sit in 'Other Europe' (USD 25.0bn at 3.9%) and 'Other Americas' (USD 25.1bn at 5.9%), which are already topped up as non-havens. Adding TWZ's Luxembourg, Switzerland and 'Rest' allocation on top therefore likely double counts. The strict CbCR share of the global total is 87.5%, not about 91%. France's own ex-ante estimate (EUR 1.5bn a year; Tax Foundation compilation) is closer to the CbCR-only figure.
Why this severity. Changes France's estimate by about two-thirds and Canada's by about a tenth; EU -4.4% and world -2.3%.
The authors' response. The June 2021 report justified the rule by incomplete 2016 CbCR reporting. It is not discussed in the article.
4moderatelikely · newly identified here · methodology
The US already has its own minimum tax on foreign profits, in force since 2018. The paper counts the money that tax already raises as new revenue, so the US gain is overstated by roughly EUR 8-12 billion a year.
Technical detail and evidence
Why it matters, in detail. Baseline mismatch: a pre-GILTI 2016-17 tax baseline is set against 2021 law. GILTI taxes are covered taxes that raise US MNEs' jurisdictional ETRs. One-for-one netting is approximate because GILTI involves global blending, the QBAI exclusion and expense allocation. Offsetting factors push the US figure up: within-cell heterogeneity (aggregation) bias and excluded stateless income.
Evidence. JCT ex-ante GILTI revenue is USD 9.5bn for FY2021 (OECD 2020 EIA, Table 3.14, p. 115). CSZ (2021) Table 1 shows 'less expected GILTI revenue -137'. Netting at the paper's 2021 exchange rate: US 54.38 -> 46.35bn (-15%) with JCT, or -> 42.8bn (-21%) with the CSZ 10-year average. World 154.5 -> 146.5bn (-5.2%) or 142.9bn (-7.5%) (scripts/run_combined.py). OECD (2020) notes that actual GILTI revenue may exceed the ex-ante estimate.
Why this severity. Reduces the largest country figure by 15-21% and the global total by 5-7.5%; the qualitative distributional conclusions are unchanged.
The authors' response. The paper discusses how GILTI interacts with QDMTTs (s. 6.4, pp. 704-705) but does not net GILTI in the IIR scenario.
5moderatelikely · overclaiming
The figures rest on the first, admittedly imperfect years of company data, projected forward with economic growth, and are presented as single numbers with no margin of error. Using a neighbouring year moves the totals by a quarter. The authors say their errors roughly cancel out, but for the EU almost all the known errors push the same way, upwards.
Technical detail and evidence
Why it matters, in detail. Model uncertainty is large relative to the reported precision: vintage choice changes the world total by -25% to +22% and the EU total by -35% to +12%. For the EU, the quantified upward biases sum to about -30% against no demonstrated offset. A range of roughly EUR 35-45bn for the EU long-run figure, and EUR 125-170bn for the world, would better reflect the evidence than the EUR 55bn and 154.5bn point estimates.
Evidence. Reproduced sensitivities: 2016 data -25% world, -35% EU (OA B.5.1); 2017-only ETRs +6.3% world, +11.0% EU; accrued taxes -0.5% world, +2.1% EU; leaving the Belgian/Singapore anomalies in place +18% world, +29% EU. The authors' 2023 update with 2018 data gives USD 201.9bn world (about EUR 189bn in 2021 terms, +22%) and EU USD 65.9bn. The identified biases for the EU mostly point upward: dividends -24.6%, TWZ local firms -5.9%, max rule -4.4%, and minority interests of 8-9% of after-tax profits (OA D). Combining dividends, no max rule and Poland's/Latvia's own CbCR rows gives EU 37.3bn (-32%) and world 135.5bn (-12%), or 127.5bn (-17.5%) after GILTI netting (scripts/run_combined.py). The downward bias from within-cell heterogeneity is real. Fuest, Hugger & Neumeier (2022) find that 47.7% of low-taxed foreign profit sits in country pairs whose average ETR is at least 15%, and Auditor A's OECD Table IV test gives +20-36%, mostly for US parents. This makes the global total defensible, but no comparable offset has been shown for EU domestic cells.
Why this severity. Does not by itself change the numbers. However, the paper presents point estimates with a claim of no systematic bias while the documented biases for the EU headline mostly point upward, by roughly 25-35% combined.
The authors' response. The paper stresses that the estimates are first-round and static (pp. 689-690, 705) and provides robustness checks in Online Appendix 1, section B.
6moderatelikely · newly identified here · external inconsistency
When the German and Dutch governments later estimated what the minimum tax would actually raise for them, their numbers were many times smaller than this paper's; figures compiled for Austria, Belgium and Denmark point the same way. Part of the gap reflects different scenarios and years, but not all of it. The paper's figures for individual countries should be treated with caution.
Technical detail and evidence
Why it matters, in detail. In actual implementation, each country's QDMTT collects the top-up on its own low-taxed domestic and host profit, which is the component directly comparable to the paper's domestic and QDMTT figures. For DEU, NLD, AUT, BEL and DNK, official totals are 3-30 times below the paper's domestic or host components alone. Possible explanations: contaminated CbCR cells (F1; dividends received by intermediate holdings in hubs); structural change after 2017; GloBE adjustments (deferred taxes, losses, exclusions). The foreign IIR component for Germany is corroborated by micro data.
Evidence. Germany: the paper gives IIR EUR 10.1bn (domestic 5.1 + foreign 5.0) and QDMTT 5.5bn. External estimates: the government bill gives +EUR 0.2bn full-year direct effect (Bundestag Drucksache 20/8668); an ifo study for the BMF gives 1.5-1.7bn (as reported by tagesschau.de, 31.07.2023, not read in the original); Heidecke & Liebe (2023, ITPJ, abstract) give 1.3-1.8bn. Fuest, Hugger & Neumeier (2022, ifo Schnelldienst 75(4): 41-49), using CbCR micro data, find 5.1-6.7bn for the static IIR on foreign profits only, close to the paper's 5.0-6.0bn foreign component, and about 2bn if low-tax countries levy the top-up themselves. Netherlands: paper IIR 2.0bn and QDMTT 14.1bn, against a CPB-certified EUR 0.47bn structural. A Tax Foundation compilation (Bunn & Perez Weigel 2023/2024; underlying documents not verified) gives the following, against the paper's IIR/QDMTT figures. Austria: 2.3/2.6 vs EUR 0.1bn. Belgium: 3.4/3.3 vs EUR 0.33bn. Denmark: 1.6/1.0 vs DKK 2-3bn. France: 3.6 (1.15 without the max rule) vs EUR 1.5bn. UK: 5.9/7.3 vs GBP 2.3bn. Czech Republic: 0.1 vs CZK 4-6bn (here the paper is lower). Globally, OECD (2024, WP 68) gives USD 155-192bn including about one-third from reduced profit shifting, the same order as the paper's total (output/external_estimates.csv).
Why this severity. Bears on the country-level results (C5) and corroborates F1 rather than adding a further EU correction. The comparisons are not like-for-like (later years, actual rather than polar implementation, safe harbours, behavioural effects), so they show that the country magnitudes are fragile rather than measuring a precise overstatement.
7minorconfirmed · newly identified here · internal inconsistency
The paper's description of which foreign countries generate the EU's extra revenue uses the wrong figures. The totals are unaffected.
Technical detail and evidence
Why it matters, in detail. The named partner contributions are overstated by factors of about 3-15. The dominant source, a single aggregate cell for Irish parents, is not mentioned.
Evidence. Reproduced (scripts/pair_checks.py). All parents with no carve-outs: AUS 2.68, GBR 7.72, USA 4.58, OTE 5.47, OAM 4.58. EU parents with long-run carve-outs: AUS 0.85, GBR 1.37, USA 1.25, OTE 0.49, OAM 0.31. The largest single EU non-haven item is the Irish parents' unallocated foreign total (USD 90.8bn at a 4.2% ETR, EUR 9.2bn).
Why this severity. A descriptive error about where EU revenue comes from; no total changes.
8minorconfirmed · newly identified here · internal inconsistency
Several small numbers in the text, appendix tables and references do not match each other or are mislabelled. None changes the main results.
Technical detail and evidence
Why it matters, in detail. No effect on the reproduced main tables. Item 9 overstates the correspondence with the OECD benchmark: the USD 150bn includes reduced profit shifting, whereas the paper's figure is static.
Evidence. Article pp. 690, 696, 701-702, 705, 707-708; OA1 pp. 2-5, 17-18; OECD (2020) EIA Executive Summary and Table 3.16 (p. 118); Note No. 2 (Oct 2021) p. 2; reproduction outputs (output/table1_adj.csv, sensitivities_adj.csv).
Why this severity. Editing and version-synchronisation slips; apart from the introduction's 165 upper bound (+7%), each is below EUR 1bn or a matter of labelling.
9minorconfirmed · newly identified here · version instability
A widely quoted earlier version of this study put the EU's gain at EUR 72-83 billion. The journal article lowers it to EUR 55-67 billion, mainly by correcting a few suspicious data points, without explaining the difference to readers.
Technical detail and evidence
Why it matters, in detail. Between versions, the EU long-run figure fell by 23% and the world total by 24% for the same scenario; individual countries moved by up to 97%. The 2023 update with 2018 data then raised the world total by about 22% in 2021 terms.
Evidence. Note No. 2 (repository copy, Note-2-November-2021.pdf) Table 3 pp. 11-12 and summary p. 2; article Table 1. Adjudicator's model without the anomaly replacements: world 182.5bn, EU 71.1bn. Authors' 2023 update (repository results, 2018 data): world USD 201.9bn, EU USD 65.9bn, US USD 79.1bn.
Why this severity. The article is the corrected version and discloses how the anomalies are treated (fn 7; OA B.5). The problem lies in reconciliation and communication. It also shows that the headline depends on a handful of cells, but the published figures are not shown to be in error.
10minorconfirmed · previously known · methodology
The exemption for real business activity is computed a little too generously, which slightly understates the revenue. The authors say so themselves.
Technical detail and evidence
Why it matters, in detail. The carve-out effect in C3 is overstated by about a fifth in relative terms (22.3% -> 17.7% in year 1; 13.7% -> 10.8% in the long run); revenue levels are understated by 3-5%.
Evidence. Reproduced: excluding inventories (assets x 0.76) raises long-run revenue to world 159.73bn (+3.4%) and EU 57.64bn (+4.5%). OA B.6.1 reports year-1 figures of 147.3 and 51.3. The full-sample carve-out effect becomes 10.8-17.7% instead of 13.7-22.3%.
Why this severity. Disclosed, quantified by the authors, and conservative for revenue; it narrows the abstract's carve-out range by 3-5 points.
The authors' response. Acknowledged: 'the results displayed below would overestimate the effect of the substance-based income exclusion and (to a lesser extent) understate revenue gains' (p. 693).
11minorconfirmed · newly identified here · data error
A labelling mix-up in the input data makes Nicaragua appear to gain almost EUR 1 billion in one appendix scenario; the gain belongs to Mexico. The main results are not affected.
Technical detail and evidence
Why it matters, in detail. The data error misallocates about EUR 0.9bn of behavioural CIT gains from Mexico to Nicaragua; the aggregate ordinary-CIT increment changes by only about EUR 0.07bn.
Evidence. IRS SOI 17it01bcbc.xlsx rows for Jamaica and Mexico match the archived MEX and NIC rows. In the current OECD Data Explorer (Table I), USA-MEX in the positive-profit panel shows USD 17.73bn and 1,112,216 employees. The adjudicator's cubic re-computation for US parents (scripts/behavioral_nic_check.py) gives 'Nicaragua' EUR 904.6m under the archived labels, which moves to Mexico once the labels are corrected (Auditor B: Nicaragua 906.7m -> 2.0m, Mexico 3,758m -> 4,758m).
Why this severity. Affects only the country distribution of an illustrative appendix exercise, not the article's static estimates.
12minorconfirmed · newly identified here · methodology
A shortcut formula in a side exercise slightly overstates how much profit firms would move back; the effect is small.
Technical detail and evidence
Why it matters, in detail. The first-order approximation overstates finite-change unshifting by 4-7% in the linear cases, small relative to elasticity uncertainty.
Evidence. Re-computed (scripts/behavioral_loglinear_check.py): relocated CbCR profits are 108.0 -> 103.6bn (-4.1%) at eta = 0.8 and 192.8 -> 179.1bn (-7.1%) at eta = 1.428; the cubic case gives 254.9bn, matching OA2.
Why this severity. Changes illustrative appendix magnitudes by 4-7%; the static headline is unaffected.
13minorlikely · selective presentation
Claims about how much poorer countries would gain partly reflect money going to tax havens that are counted as 'developing' countries, or to unidentified regions. These breakdowns are less solid than they look.
Technical detail and evidence
Why it matters, in detail. The country-group shares in Table 3 shift by tens of billions depending on how the aggregates and financial centres are classified. The BVI's IIR revenue is an artefact of allocating shifted profits by ultimate owner rather than by UPE jurisdiction.
Evidence. Article Tables 2, 3 and A3 (pp. 698-709); TWZ Table C4 (2017 update); adjudicator's model gives BVI 5.10bn (haven 2.54 plus imputed non-haven 2.56).
Why this severity. Totals are unaffected; the statements about distribution by country group depend on classification choices.
The authors' response. The text notes that low-tax jurisdictions would be among the main QDMTT beneficiaries and that the host-country distribution is not robust to behavioural responses (p. 697).
14minorplausible · newly identified here · speculative assumption
The new rules exempt some kinds of income, such as international shipping, and use accounting rules the model cannot reproduce. Ignoring these probably inflates a few numbers, for example Greece's, by an unknown amount.
Technical detail and evidence
Why it matters, in detail. These exclusions mostly bias top-ups upward: the shipping exclusion potentially removes up to about EUR 1.7bn for Greece; deferred taxes raise GloBE ETRs for loss-using and capital-intensive firms. No quantification is possible with tabulated CbCR.
Evidence. Model Rules (Dec 2021) Arts 3.3, 4.4, 7.1; adjudicator's model, Greek pair cells (no individual-country detail); article fn 11; OA1 B.1.
Why this severity. The mechanisms are real and point mostly towards overstated revenue, but their size is unknown and likely small relative to F1.
The authors' response. Deferred taxes and CFC taxes are acknowledged as not modelled (fn 11; OA B.1).
Claims examined and rejected
Problems one auditor raised that did not survive checking.
- The TWZ haven data used for TWZ-only parents and in the max rule are 'booked haven profit, not mechanically the TWZ estimate of artificially shifted profit' (Assumptions table), so critiques of TWZ shifted-profit estimates do not transfer. — The workbook the authors load (data/TWZ/2017/TWZ.xlsx, sheet TableC4) is titled 'Allocating the profits shifted to tax havens'. Its 'Ultimate ownership' block, the one the code reads, is shifted profit allocated by owner country. Critiques of the TWZ estimates (offshore-havens-09 to -12) therefore apply to the TWZ-based share, about 12.5% of the world total including the max-rule replacements, and to France via F3.
- Sensitivity results: accrued-tax numerator gives EU 53.629 / world 150.089bn; single-year ETRs give EU 55.745 / world 157.499bn. — Not reproduced. The adjudicator's independent model gives EU 56.34 / world 153.76 (accrued) and EU 61.25 / world 164.20 (2017-only ETRs), matching Auditor A's V1 and V6. The 2017-only ETR choice raises the EU figure by 11%, rather than by 1% as Auditor B's figures imply.
- The change between the October 2021 note and the article is not a version-instability finding. — Partly rejected. Concealment is not alleged, but the article does not reconcile a 23-24% change in its headline relative to a cited, widely publicised version; retained as minor issue F9.
- The paper's static year-1 figure (EUR 139bn, 'about USD 165bn') exceeds the OECD (2024) direct top-up component (about USD 103-128bn). — Not a like-for-like comparison. The paper's figure is in 2017 data uprated to 2021 EUR (about USD 140bn in 2017 USD), whereas the OECD figures are 2017-2020 nominal averages. The OECD direct component is also computed after reduced profit shifting, under partial implementation, for IF members only. The global totals are of the same order of magnitude; no issue is retained.
- The max(CbCR, TWZ) rule is 'undisclosed'. — Partly rejected. The rule is set out in the cited June 2021 predecessor report (pp. 19-20, 37); it is undisclosed only in the article and Online Appendix 1. The issue is retained (F3) with corrected wording.
- Later CbCR vintages show the domestic contamination washing out after the 2020 dividend guidance (EU-domestic top-up 29.6 -> 16.4bn, Germany 13.3 -> 1.7bn by 2021; German domestic ETR 16.0% in 2021). — Qualified rather than accepted as evidence. The current OECD release shows the German domestic ETR back at 13.9% in 2022 (22.4% in 2023), and COVID confounds 2020-21. The time pattern is mixed, so F1 rests mainly on the national-base test, the authors' own correction and the official estimates.
- Corpus critiques offshore-havens-10/-11/-12 (Dharmapala; Fuest et al.: micro-data estimates of shifting far below TWZ) and offshore-havens-16 (Coffey) undermine the headline. — Examined and dismissed as direct critiques by both auditors and the adjudicator. For CbCR parents the paper tops up booked low-taxed profit at observed ETRs, not estimated shifted profit, and Coffey's German-Irish example is not a premise of the simulation. Only the roughly 12.5% TWZ-based share is exposed. Fuest-Hugger-Neumeier's own micro-data IIR estimate for Germany's foreign profits matches the paper's foreign component.
- Blouin & Robinson's double-counting critique (offshore-havens-08/-09) bears materially on the estimates. — Limited bearing, as Auditor A itself concluded. TWZ use profit-type return, not BEA net income (p. 692), and TWZ data drive only about 12.5% of the total. The CbCR dividend-double-counting problem is handled under F1.
Audit notes
What reproduced. Reproduced. The adjudicator re-implemented the model from the paper, OA1 and a text reading of the authors' code (scripts/adj_model.py; no author code executed). It reproduces Table 1 exactly (world 179.094/139.194/154.527bn; EU 67.121/47.445/55.184bn; CbCR parents 162.627/127.849/141.204bn); the OA A.1.1 split (21.244/10.723/23.217bn); OA B.4.2 (EU 41.582bn); OA B.6.1 (159.727/57.645bn); and OA2 relocated profits (108.0/192.8/254.9bn). Both auditors' independent implementations agree; Auditor B also reproduced the US figure from the original IRS tables, and Auditor A reproduced Table 2's host column (max abs. diff. 0.05bn), Table A2 and OA B.5.1. Partial: printed Table A1/A2 totals differ slightly from code and OA (F8); the paper's '83 parents' includes Bonaire with zero revenue. Not reproduced: ORBIS purely-domestic groups (EUR 35m; proprietary data), Table 3 development-group classification, OA2 revenue tables beyond the checks in F12-F13, and Auditor A's OECD Table IV heterogeneity test (not re-run).
Limits of this audit. No access to MNE-level CbCR or GloBE data, so true GloBE domestic ETRs cannot be computed. The magnitude in F1 therefore rests on the authors' own correction, an approximate national-accounts test (CIT revenue / statutory rate; cash timing) and later official estimates. Those estimates are not like-for-like: 2023+ implementation with QDMTTs and safe harbours, later data, and in some cases behavioural effects. Several official figures (Austria, Belgium, Denmark, France, UK, Czech Republic, Canada) come from a Tax Foundation compilation, and the ifo-for-BMF 2023 figure from a press summary; neither was checked against the underlying documents. The Heidecke & Liebe (2023) and Boukal et al. (2026) figures are from abstracts only. Not re-run: Auditor A's OECD Table IV heterogeneity test. Not reproduced: the ORBIS results, Horst & Curatolo (2020) (paywalled) and most of OA2. The HAL deposit of the October 2021 note was not downloaded; the authors' repository copy was used. The national-base test uses the authors' statutory-rate file; Auditor A's used KPMG rates, with the same results.