Ending Corporate Tax Avoidance and Tax Competition: A Plan to Collect the Tax Deficit of Multinationals

Why it matters, in detail. The remedial tax's apportionment key (US-destination share of global sales, with a throw-out of sales to low-tax states) requires destination-based sales that CbCR does not contain; residence-based revenue can be zero for a haven seller with 100% US customers. Using exchanged CbCRs as an assessment base would also fall outside the MCAA's 'appropriate use' terms, so implementation needs new domestic reporting, as Pillar Two's GloBE Information Return did. No effect on Tables 1-2.

Evidence. CSZ p.11, p.12 and fn 20-21, p.13. Form 8975 Schedule A Part I instructions: revenues of the constituent entities resident in the jurisdiction (Auditor B; adjudicator checked). OECD Multilateral Competent Authority Agreement on the Exchange of CbC Reports, s.5(2): information 'will be used for assessing high-level transfer pricing, base erosion and profit shifting related risks, and, where appropriate, for economic and statistical analysis ... transfer pricing adjustments will not be based on the CbC Report'; US Model CbC Competent Authority Arrangement (12 Sept 2017), s.5(2), similar (texts retrieved by the adjudicator). Laffitte & Toubal (2022), AEJ: Economic Policy 14(4):371-396, on the absence of destination sales in CbCR.

Why this severity. Undermines the stated feasibility argument for the plan's third pillar (in the abstract), but not the possibility of such a tax with new reporting rules; the revenue estimates do not depend on it.

The authors' response. fn 21 (p.12): 'There are important legal and administrative issues that may take time to iron out, similar to those generated by sales-based formulary apportionment. If rules that would disregard sales to low-tax states prove infeasible, Avi-Yonah and Clausing (2019) have suggested a possible look-through rule'; conclusion (p.13): such measures 'will only be approximate'.

Why it matters, in detail. Incremental revenue should be sum_j max(0, m*P_j - F_j - US_j), where US_j is current-law US residual tax on P_j (Subpart F, branch income); CSZ effectively set US_j = 0 apart from the GILTI deduction. IRS CFC proxies put the overlap at ~$9-12bn/yr in 2017 terms (two-thirds of which enters the US figure), i.e. -10% to -21% of the ten-year net, depending on how CFC tax E&P by country of incorporation maps onto CbC accounting profit by tax residence; branch overlap is unquantified.

Evidence. CSZ p.7 ('we assume this tax replaces the present global minimum (GILTI) tax, so we subtract ... GILTI revenues') and A.1 pp.16-17. Form 8975 instructions (June 2017; Rev. Dec 2020), line 3 and PE example: home-country income tax on PE income is reported on the home jurisdiction's Schedule A (adjudicator re-download). 26 USC 951(a)(1)(A) and 951A(c)(2)(A)(i)(II) (Auditor B). IRS SOI CFC Table 2, TY2016, col.17: Subpart F income $95.3bn for all CFCs, about $70bn in the havens of Tables 1-2 (Netherlands 21.6, Luxembourg 12.3, Bermuda 11.2, Ireland 10.3, Switzerland 5.9, Cayman 3.9, Singapore 2.6, Hong Kong 2.1). Adjudicator (scripts/dividend_and_subpartf.py): overlap = sum of max(0, 21% - ETR) x min(Subpart F, CbC profit) = $9.2bn/yr (15% of Table 1's top-up) and $12.3bn/yr (17% of Table 2's); ten-year net $432bn -> $347bn and $537bn -> $422bn; with half the overlap $390bn and $479bn. Auditor A obtained $347bn and $430bn.

Why this severity. A clear omission from the incremental-revenue baseline, plausibly worth 10-20% of the ten-year net: below the rubric's ~25% threshold for 'major', and the conclusion that the tax raises substantial revenue is unaffected.

Why it matters, in detail. The 2/3 factor is an ad hoc behavioural haircut, not an estimate. With static haven rates it understates US collection (100% under unilateral adoption); with the haven rate response the paper predicts, which materialised under Pillar Two, the residual US top-up on haven profit taxed locally at 15% is 6/21 (about 29%) of the static amount, so for affected years US revenue from haven rows would fall by roughly two-thirds relative to the paper's path. The net bias is therefore time-dependent and unidentified; revenue scales one-for-one with the share.

Evidence. CSZ p.5 (fn 11), p.7 (fn 15), p.8, A.1 p.16. Adjudicator sensitivity (scripts/repro_tables.py), matching both auditors: ten-year net $873.7bn (100% US share), $536.8bn (2/3), $368.4bn (1/2), $199.9bn (1/3), $115.7bn (1/4). Johannesen (2022), 'The global minimum tax', Journal of Public Economics 212:104709 (open access): a minimum tax raises equilibrium haven rates 'up to the global minimum', 'which transfers resources from non-haven firms to haven governments'; it cites CSZ's 'around $50 billion per year' as an early estimate of the mechanical channel. Post-publication: under Pillar Two, low-tax jurisdictions adopted domestic minimum top-up taxes; Bermuda's Corporate Income Tax Act 2023 imposes 15% on Bermuda entities of groups with revenue of at least EUR 750m for fiscal years beginning on or after 1 January 2025. At a 15% local rate, a 21% US top-up on zero-taxed profit falls by about 70%.

Why this severity. The largest single swing factor in the ten-year figure, but it is disclosed, it is a haircut relative to static scoring, and independent scorers (TPC $442bn, AEI $457bn, Treasury $533.5bn) reach similar magnitudes: the point estimate is uncertain while the qualitative revenue claim survives.

The authors' response. fn 11 (p.5): 'while the minimum tax per se would raise less revenue, the part of that revenue that was due to tax-motivated profit shifting would show up in the regular US corporate tax base'; p.8: the numbers 'should be considered the total effect of both minimum tax revenue and the buttressing of the regular corporate tax base'.

Why it matters, in detail. The hybrid multiplies full-sample (net) profit by positive-profit ETRs; if positive-sample profit includes exempt intra-group dividends or profit sheltered by loss carryforwards, ETR_pos is biased down and the top-up up. CFC-data proxies imply that 35-70% of the hybrid's $11.1bn/yr increment over Table 1 is artefact, lowering the ten-year net from $537bn to $463-501bn; the $4.5bn/yr charged in the UK, Canada, Australia and Japan is most likely spurious. Combined with the Subpart F overlap (F1), the proxies give $369bn (full) to $448bn (half).

Evidence. Form 8975 instructions (June 2017 and Rev. July 2019), line 2: 'Enter the aggregate profit or loss before income tax'; Rev. Dec 2020 adds that it 'does not include payments received from other constituent entities in your group that are treated as dividends' (adjudicator re-download from irs.gov). OECD (2020) Tax Challenges Arising from Digitalisation - Economic Impact Assessment, ch.3 paras 25-26 (text checked). Adjudicator (scripts/repro_tables.py): hybrid step +$11.1bn/yr (+$104bn over ten years); UK full-sample tax $9.34bn on $18.09bn profit (51.6%) while positive-sample profit is $81.67bn at 10.5%, giving a $1.89bn top-up; Canada $1.63bn; Australia $0.79bn; Japan $0.23bn. IRS SOI CFC Table 2 TY2016: dividends from related CFCs / current E&P of CFCs with positive E&P: Netherlands 52.9%, Luxembourg 49.9%, UK 36.9%, Switzerland 30.8%, Canada 20.0%. Purging positive-profit ETRs at these shares (capped at the full-sample ETR) lowers Table 2 from $72.0bn to $64.1bn/yr and the ten-year net from $537bn to $463bn ($501bn with half the shares): Auditor A's figures, reproduced exactly (scripts/dividend_and_subpartf.py). IRS CbC 2018: seven-haven profit jumped from $289bn to $430bn in the record repatriation year.

Why this severity. Lowers the paper's preferred estimate by roughly 7-14% of the ten-year net (bounded by the $104bn hybrid step); the full-sample Table 1 estimate ($432bn) is largely unaffected and the qualitative revenue conclusion survives.

The authors' response. p.18: 'the definition of profit may be unclear, and companies are free to supply data as they see fit. Still, since the data are known to be used for transfer pricing risk assessment, it is unlikely that companies will have an incentive to overstate their income, especially in tax havens. It is also reassuring that foreign totals are similar to those reported from other sources that are known to exclude double-counting.' p.16: the positive-profit sample is 'more conceptually appropriate'.

Why it matters, in detail. Consistent GDP-weighted combined rates give about 47.5% (world, 83 countries) or about 49% (OECD) in 1985 and 25.8-26.4% in 2019, a fall of about 22-23 pp (-46%) rather than 26 pp (-53%). The 23% value matches unweighted means (OECD 23.9%; world 22.6%; the OECD's published 23.2% for 2020) or GDP-weighted central-government rates (OECD-38 22.1%). No effect on the revenue analysis.

Evidence. Adjudicator (scripts/global_cit.py, scripts/oecd_cit_1985_2019.py): Tax Foundation worldwide dataset, GDP-weighted combined rates: 2019 25.81% (181 jurisdictions), OECD members 26.37%; 2018 26.01% / 26.40%; 1985 46.96% (82 countries, Japan missing), 47.3-47.6% with Japan imputed at 50-53%, OECD 48.6% (48.8-49.2% with Japan). OECD Table II.1 (38 members), 2019: combined GDP-weighted 26.38%, combined unweighted 23.95%, central-government GDP-weighted 22.12%. World unweighted 2019: 22.63% (224 jurisdictions). OECD Corporate Tax Statistics 2nd ed. (2020): unweighted OECD average 23.2% in 2020. Auditor A: OECD CTS combined x World Bank GDP 26.3% (2019); central-government 23.5%. Auditor B: 26.30% world, 26.53% OECD (Tax Foundation 2019, Table 4). TWZ (NBER w24701, p.1; REStud 2023, p.1499) and Saez & Zucman (2019, ch.4): '49% to 24%' for 1985-2018.

Why this severity. The qualitative claim that the global rate roughly halved survives (about 49% -> 26%, -46%), but the abstract's opening statistic overstates the fall by about 3-4 points (26 vs about 22-23 points, i.e. 13-17%), and the stated source and weighting cannot produce the 23% endpoint.

Why it matters, in detail. DII attributes lower-tier income to first-tier haven holding companies (Blouin-Robinson), so Figure 1's ~60% level overstates where profits are booked relative to an equity-income-adjusted measure (~40%). The 2011-2016 rise contradicts 'steady since 2011', while the post-2016 plateau does support 'no decline' after BEPS and the TCJA.

Evidence. Figure 1 (p.3, rendered and inspected by the adjudicator). BEA AMNE 2017 preliminary Table II.D1 (adjudicator): equity income / net income = Bermuda 101.7/90.4 (113%), Luxembourg 116.8/121.4 (96%), Netherlands 205.6/259.2 (79%), UK Caribbean islands 55.1/75.1 (73%). BEA DII seven-haven shares: 50.1% (2011), 65.5% (2016), 60.5% (2019) (Auditor A; Auditor B 50.1% in 2011 and 62.4/60.3/62.4% in 2017-19, current vintage). Table A2 (p.20): haven share 53% (pre-tax DII) vs 40% (Blouin-Robinson adjusted).

Why this severity. Affects the level of an illustrative series and the wording of a footnote; the narrow claim that the 2019 haven share equals that of recent years, and the CbC-based revenue estimates, are unaffected.

The authors' response. A.3 (p.19): the DII series 'is free of any double-counting concerns' and 'there is no reason to suspect that the tax rates of reporting countries will be systematically lower'; the adjusted series 'may exclude foreign-to-foreign profit shifting and consequently allocate too little income to tax havens' (e.g. hybrid dividends).

Why it matters, in detail. Excluding Puerto Rico lowers the CbC haven shares by 1.2-2.8 pp, narrowing the gap with the BR-adjusted BEA share (40%) that the paper highlights; Puerto Rico is properly foreign for the US tax computation.

Evidence. Table A2 (p.20). Adjudicator (scripts/repro_tables.py): CbC big-haven shares 50.7% / 47.3% / 55.7% (printed 51/47/56) become 47.9% / 45.1% / 54.5% without Puerto Rico.

Why this severity. Narrows the CbC-vs-BR contrast by about 2-3 points; no effect on the revenue estimates, where including Puerto Rico is correct.

Why it matters, in detail. A $20bn (3%) internal inconsistency in the gross subtotal; the net of GILTI ($536.9bn) is unaffected.

Evidence. Table 2 (p.9). Adjudicator: growth factor 14.0454; $47.975bn x 14.0454 = $673.83bn; net $536.83bn. Both auditors obtained the same.

Why this severity. Typo in an intermediate line; the headline net figure is correct.

Why it matters, in detail. With current vintages the DII-based alternative rises by about 17% and the BR-adjusted Bermuda value turns positive; the CbC-based headline estimates are unaffected.

Evidence. Adjudicator: preliminary 2017 Table II.D1 Bermuda 90,417 + 874 - 101,722 = -10,431; revised 89,058 + 838 - 81,949 = +7,947 ($m); totals 571,007 vs 594,560. BEA release date: 'Activities of U.S. Multinational Enterprises, 2018', 21 Aug 2020 (revised 2017). Auditor A: the DII-based estimate rises from $30.4bn to $35.7bn a year with the revised vintage; the BR estimate is almost unchanged ($23.6bn -> $23.9bn).

Why this severity. Documentation and vintage issue affecting only appendix alternatives, in both directions (raises the DII alternative, removes the negative-Bermuda argument).

Why it matters, in detail. No quantitative effect; the undocumented residual-row rule moves the preferred net by about $2bn (0.4%).

Evidence. CSZ pp.1-2, 7, 12, 15, 18. Crossref doi:10.17310/ntj.2020.4.09 (NTJ 73(4)) and doi:10.5744/ftr.2017.0005 (FTR 21(2)) (adjudicator). OECD (2020) Peer Review of the Automatic Exchange of Financial Account Information: 'nearly 100 jurisdictions exchanged information on 84 million financial accounts' worth about EUR 10 trillion. Adjudicator scripts/repro_tables.py: 13 rows above $1bn (Ireland $1.0008bn); Table 2 variant $534.77bn.

Why this severity. Presentation and referencing slips with no effect on the conclusions.

Why it matters, in detail. Neither the upward biases (F1, F3: jointly -17% to -31% of the ten-year net under IRS CFC proxies; F2: haven responses) nor the downward ones (within-jurisdiction pooling: 2019 sub-group-level static deficit at least 38% above the Table 2 method; stateless exclusion) are quantified in the paper. The net sign is unidentified, so $537bn is a central calibration with wide two-sided uncertainty, not a lower bound.

Evidence. CSZ pp.7-9, 17. Adjudicator (scripts/aggregation_bias.py; OECD CbCR Table IV, US-parented groups by partner jurisdiction and sub-group ETR bin): static sub-group-level 21% deficit at least $121.2bn in 2019 (upper bound $138.2bn), against $112.1bn when positive-profit groups are pooled per jurisdiction (IRS Table 1B gives $112.0bn), $88.0bn under the Table 2 method and $78.7bn under Table 1; 2020: $100.6-116.5bn vs $92.0bn and $72.7bn; 2021: $120.1-140.1bn vs $110.2bn and $86.6bn. Joint dividend and Subpart F proxies lower Table 2's ten-year net to $369bn (full) or $448bn (half) (scripts/dividend_and_subpartf.py). BEA-based alternatives reported by the paper: $30bn/yr (DII) and $24bn/yr (Blouin-Robinson), i.e. ten-year nets of about $290bn and $194bn (Auditor A).

Why this severity. The 'conservative' label is asserted rather than demonstrated, but a large omitted downward bias means it may well be right for the static calculation; it changes how the estimate should be read, not the estimate itself.