Gabriel Zucman · Economics for Inclusive Prosperity (EfIP) Research Brief, September 2018

Taxing Multinational Corporations in the 21st Century

16 problems found: 1 major · 3 moderate · 12 minor

What the paper claims

Globally, 40% of multinational profits (profits made outside the parent's country) are shifted to tax havens each year, about $600 billion (relayed from Torslov, Wier & Zucman 2018).
The United States loses about 15% of its corporate income tax revenue because of profit shifting (relayed from TWZ 2018).
US multinationals book a very large share of their foreign profits in a handful of low-tax havens, where reported profits are huge relative to wages and taxed at very low rates (author's computations from BEA 2016 data).
Profit shifting reduces global corporate tax payments (close to $5 avoided in high-tax countries per $1 collected by havens) and redistributes income to multinationals' shareholders; US multinationals use havens particularly extensively (55% vs 40% globally).
The high profits booked in havens are mostly explained by shifting of paper profits, not by relocation of tangible capital ('machines don't move massively to low-tax places; paper profits do').
Apportioning firms' global consolidated profits by where they make their sales would end profit shifting and tax competition for real activity; it can be adopted unilaterally and is 'tried and tested' in US states.
Under the reform the corporate tax would fall, as in a closed economy, mostly on capital, so raising the rate to 35%-50% would raise the progressivity of the US tax system.
Background on tax competition: the global average statutory corporate tax rate fell by more than half between 1985 and 2018, and effective rates fell almost in line with statutory rates.

Every problem found

1majorlikely · previously known · overclaiming

The brief's best-known number, that 40% of multinationals' foreign profits are moved to tax havens, rounds up an estimate its source puts at 36%. About half of that estimate comes from the authors' own top-ups to tax havens' incomplete statistics. Other reasonable methods give figures nearer a quarter to a third; some company-level studies find even less, and one finds more. Profit shifting is clearly very large, but the 40% is probably too high and much less certain than the brief suggests.

Technical detail and evidence

Why it matters, in detail. S = sum over havens of (Pi_f - pi_l x W_f); with pi_l and wages fixed, dS/dU = 1 for the $306bn of unrecorded haven profit U added from partner-reported DI income, CDIS-stock returns and a pro-rata plug. So S at face value is $310bn (18.2% of foreign profits). TWZ's own +45pp (Paul-Isaka) benchmark gives 29.7%. CbC regression counterfactuals give 13-22% (different sample and coverage); the Garcia-Bernardo-Jansky log model gives more (>$850bn, 2017). The brief reports 40% against the source's 36.2% (+10.5%), and it is plausibly 10-60% above a defensible 25-36% range. Because TWZ allocate a pooled S to source countries, any error passes proportionally into the US loss (F2).

Evidence. TWZ2018.xlsx Table1: C9/C8 = 616.462/1,703.229 = 36.19%; TWZ (NBER w24701, rev. Aug. 2018) p.12 'more than $600 billion, i.e., 36%'; abstract 'close to 40%' (the June-2018 text had '$600 billion, i.e., close to 40%'). TWZ2018AppendixTables.xlsx Table A.6!D91 = 306.391 (Ireland 48.8, Luxembourg 31.8, Netherlands 35.3, Belgium 3.6, non-OECD havens 187.0). Table A.7 col. 15 = C - O x TableA4!D, where C = TableA6!E includes the additions (OECD havens: E = F+G+H, H = additions; non-OECD: E = B-K, B includes the additions, K = 0.48 x local wages). Removing the additions gives $310.1bn = 18.2% of foreign profits (22.2% if they are also removed from the denominator) [scripts/twz2018_check.py]. TWZ REStud 2023 Table 2 (TWZ2022.xlsx): +45pp benchmark $506.2bn (29.7%); 'removing corrections to haven data' $507.2bn; bounds $369.0bn (21.7%) to $739.9bn (43.4%). The project's adjudication of TWZ (2023) finds that the published face-value column removes only about $109bn of the $306bn (a unit error) and puts defensible alternatives at roughly $425-550bn (25-32%); this adjudication confirms the one-for-one pass-through in the 2018 workbook but did not re-derive that range. External evidence: Fuest, Greil, Hugger & Neumeier (CESifo WP 9757, 2022, pp.25-26; JEEA 23(5) 2025, doi:10.1093/jeea/jvaf007): 13% of foreign profits (regression counterfactual) or 22% (payroll counterfactual) vs TWZ's 36%; Dharmapala (AEA P&P 109, 2019, doi:10.1257/pandp.20191043): micro elasticities imply less than 20%; Garcia-Bernardo & Jansky (World Development 177, 2024, doi:10.1016/j.worlddev.2023.106527): more than $850bn shifted in 2017. Blouin & Robinson (JPubE 252, 2025, doi:10.1016/j.jpubeco.2025.105512; Blouin, NeSt conference slides 2025, slides 25-26): the $306bn rests on bilateral direct-investment comparisons that assume common reporting conventions, which do not hold (the US uses the equity method).

Why this severity. The 40% is 4 points above the source's own 36.2%, and that estimate's magnitude depends heavily on contestable reconciliation and benchmark choices: taking haven statistics at face value would give 18-22%, and defensible alternatives give roughly 25-36%, so the headline plausibly overstates the share by a quarter or more. Large-scale shifting (hundreds of billions of dollars a year) survives every variant, so the qualitative finding stands.

The authors' response. TWZ (REStud 2023, Secs. 4.2.3, 5.2) argue that haven statistics must be corrected (e.g. the Cayman Islands exclude their offshore sector; global DI income paid must equal income received), that haven excess profitability survives capital, R&D and industry controls, that an independent excess-transactions method gives $646bn, and that Orbis micro data miss most haven profits. They report a +45pp benchmark scenario ($506bn) and bounds of 22-43%. No reply to Blouin & Robinson (2025) or to the CbC studies located; no response specific to the brief located.

2moderateconfirmed · previously known · selective presentation

The '15%' rounds up an estimate of 14% for 2015, before the 2017 US tax reform. It assumes every dollar of shifted profit would otherwise have been taxed at the full headline rate of 40%; at the rates companies actually paid on average it would be about half as large. And because it is a fixed slice of the global estimate, any overstatement there carries over.

Technical detail and evidence

Why it matters, in detail. L_US/T_US = tau x s_US x S / T_US, with tau = 0.40, s_US = 0.2304, S = $616.5bn and T_US = $405bn. Using tau_eff = T/Pi = 0.214 gives 7.5%. With S of $425-550bn the figure is 9.7-12.5% at statutory rates, or 5.2-6.7% at effective rates. Before 2018, deferral made part of tau x S_US a deferred claim, partly collected through the 2017 transition tax (15.5%/8%), so the present-value loss is lower. The ownership-based allocation (30.2%) answers a different, residence-based question and does not bound the source-based figure.

Evidence. TWZ2018AppendixTables.xlsx: Table C4!C45 = 1000 x TableA7!P91 x (US share of excess payments) = 142,045 ($m); Table C4d row 45: revenue 404.992, rate 0.40, loss 56.818 -> 14.03% (TWZ Aug-2018 Table 2: 14%). At 39%: 13.7%. At the 21.4% US average effective rate (Table A3!J44; Table C4c 'share of tax base' 0.0752): 7.5%. TWZ Appendix C.3.1: with effective rates 'the tax loss falls to 6 percent' (globally). Scaling with S [scripts/twz2018_check.py]: +45pp benchmark 11.5%; S = $425-550bn 9.7-12.5%; haven data at face value 7.1%. TWZ REStud 2023 p.1524: CFC rules mean a lost dollar of profit need not cost tau cents. Wright & Zucman (2018, fn.7): present-value US rate on repatriated profits 7.6%. Forstater (CGD Note, Nov. 2018, pp.6-7): the ~$180-200bn global loss 'might be considered a high-end estimate, with recoverable revenues as some fraction'.

Why this severity. Statutory-rate valuation is the standard marginal convention (Devereux & Maffini), and TWZ's US shifted amount is broadly consistent with an independent BEA-based estimate (Guvenen et al., as cited by TWZ), so 14% is a defensible upper-middle figure rather than an error. But 'about 15%' sits at the top of a plausible range of roughly 7-14% once the global-total uncertainty (F1), effective-rate valuation and deferral are allowed for.

The authors' response. TWZ Appendix C.3.1 justifies statutory rates as the marginal rate at which shifted profits would have been taxed (Devereux & Maffini 2007). REStud 2023 fn.37 reports that forward-looking effective marginal rates (Spengel et al.) give similar results for EU countries, and p.1524 acknowledges that CFC rules mean lost profits need not translate one-for-one into lost revenue. No response specific to the brief located.

3moderateconfirmed · newly identified here · overclaiming

If one country adopts the plan, companies can no longer escape that country's tax by moving profits or factories elsewhere; that part holds. But they can still save tax by moving profits or production between other countries that keep today's rules. So the plan on its own does not end profit shifting or tax competition worldwide, as the brief's summary implies.

Technical detail and evidence

Why it matters, in detail. T_A = t_A s_A P is invariant to an amount D moved between B and C, but T_world = t_A s_A P + t_B P_B + t_C P_C changes by (t_C - t_B) x D. The system is location-neutral only if every country uses a destination-sales-only factor on a common consolidated base. With heterogeneous formulas (payroll or property factors) or residual source taxation, production and booking incentives persist among those countries. The 'annihilates tax competition' step in the incidence argument holds only for the adopter's own tax.

Evidence. Brief abstract (p.1), p.2, pp.4-5. Codex scripts/reproduce.py (theory_checks): with P = $10bn, s_A = 20% and t_A = 35%, moving $1bn from a 30% to a 5% source country leaves A's $0.7bn unchanged but cuts total tax by $0.25bn. GewStG s.29(1) no.1 (payroll key); Canada Income Tax Regulations s.402(3)(a) (1/2 gross revenue + 1/2 salaries and wages); FTA state apportionment table for tax year 2018 (6 states use equally weighted three-factor formulas; Massachusetts sales or double-weighted sales). Brief p.4 itself: firms may 'start moving more tangible assets to low-tax places if policy-makers reduce profit shifting opportunities but tax rates remain different across countries'.

Why this severity. The operative claim is that a country can make its own base largely insensitive to where profits are booked and production located, and so tax multinationals at high rates without coordination. That claim is valid; the counterexample defeats only the categorical, system-wide wording of the abstract and pp.4-5. It is moderate because these are headline policy statements and the qualifications they need (universal adoption, compatible formulas, credit design) are not disclosed.

The authors' response. Not directly addressed. The brief notes (p.4) that firms may move more tangible assets if shifting is curbed while rates differ, and that international cooperation is 'always preferable'. Zucman (2014, p.134) noted that formulas including capital or employment leave incentives to move real activity. No response to these specific objections located.

4moderateplausible · previously known · speculative assumption

The plan works only if tax authorities can tell where a company's customers really are. Research shows US companies already record many sales in tax havens even though the goods are sold elsewhere, and sales can be routed through middlemen; the brief simply asserts that such tricks are 'easy to detect'. The author's own earlier work also warned that a sales-only rule can give taxing rights to the wrong countries, a caveat the brief leaves out.

Technical detail and evidence

Why it matters, in detail. Under destination apportionment the base becomes s_A x P, so avoidance migrates from transfer prices to s_A: routing sales through unrelated distributors or intermediaries outside A, or misreporting destination. Countering this requires look-through and market-based sourcing rules. L&T estimate booked-sales shifting by US MNEs of $66-85bn in 2013 under current rules, showing that firms already exploit sales location when it matters for tax. Moving taxing rights to market countries (Zucman 2014) and non-uniform formulas create double taxation or non-taxation and treaty conflicts. None of this is addressed, so 'end to profit shifting' is an upper-bound characterisation.

Evidence. Brief pp.4-5. Laffitte & Toubal, AEJ: Economic Policy 14(4):371-396 (2022), doi:10.1257/pol.20200203. Abstract; author manuscript pp.24-25 on third-party distributors and the administrative cost of look-through rules, citing Fleming, Peroni & Shay (2014). They 'do not argue that sales-based policies should be discarded'. Zucman (2014) JEP 28(4) pp.134-135. Clausing (2016) NTJ 69(2):353-385, doi:10.17310/ntj.2016.2.04: state economic activity is not very sensitive to state corporate tax choices (supportive). Gordon & Wilson (1986) Econometrica 54(6):1357-1373, doi:10.2307/1914303: formula apportionment turns the corporate tax into implicit taxes on the apportionment factors.

Why this severity. The claim that the reform 'would put an end to profit shifting' rests on an unvalidated assumption: that destination is cheap to identify and hard to manipulate. Documented sales-booking behaviour shows the channel exists, but its size under a properly destination-based factor is unknown, and US state experience is partly reassuring. This is a noticeable but unquantified qualification of the central policy claim.

The authors' response. Zucman (2014, pp.134-135) presented sales-only apportionment as 'a more radical proposal', with the caveats quoted above. The brief itself anticipates manipulation and invokes anti-abuse rules. No direct reply to Laffitte & Toubal located.

5minorconfirmed · newly identified here · internal inconsistency

The brief compares two different things: profits estimated to have been artificially moved (worldwide) and all profits recorded in havens (US firms). Compared properly, US companies still use havens more than others, so the conclusion survives; only the numbers are not like for like.

Technical detail and evidence

Why it matters, in detail. The booked share B/Pi_f and the shifted share S/Pi_f differ by (B - S)/Pi_f = 6.6pp globally ($112.6bn). Comparing global S/Pi_f (36%, rounded to 40%) with US B/Pi_f (about 52-55%) does not overturn the ranking. A like-for-like shifted comparison also preserves it: TWZ give 54% vs 27%, which the project's TWZ adjudication puts at about 1.2-1.7x on a BEA-consistent basis.

Evidence. TWZ2018AppendixTables.xlsx Table A.7: haven foreign profits (col. 2) $729.07bn vs shifted profits (col. 15) $616.46bn [scripts/twz2018_check.py; Codex reproduce.py gives identical values]. Auditor A: TWZ Table 2 foreign-firm profits in the 11 listed havens $668.8bn (39.3%). Brief pp.3-4.

Why this severity. Mixing concepts muddles the comparison, but on consistent definitions US multinationals still book a larger share of profits in havens (52-55% vs 39-43%). The qualitative conclusion holds.

6minorconfirmed · previously known · selective presentation

The '$5 lost for every $1 havens collect' figure assumes shifted profits would otherwise have been taxed at countries' full headline rates. At the rates companies actually pay on average, it is closer to $3 for every $1. That is still a large net loss for high-tax countries.

Technical detail and evidence

Why it matters, in detail. Ratio = sum_j(tau_stat,j x S_j) / sum_h(tau_eff,h x S_h). Valuing source-country losses at average effective rates (TWZ's alternative) cuts the numerator by about 35%, giving about 3.0-3.3. To first order the ratio does not depend on the level of S, since both terms scale with it, so F1 does not affect it.

Evidence. TWZ2018AppendixTables.xlsx Table C4d!E93 = 182.85. TWZ2018.xlsx 'Data F8b' rows 10 and 13 are both 'Netherlands' (F16 = SUM(F3:F14) = 41.73; 35.72 without the duplicate); 'Data F8c' total 33.26. Ratios: 4.38, 5.12, 5.50. At 6% of non-haven revenue ($1,986.6bn) the loss is about $119bn, giving about 3.3:1 [scripts/twz2018_check.py]. TWZ Aug-2018 p.27; TWZ Appendix C.3.1.

Why this severity. The ratio illustrates the redistribution argument. It is 'close to 5' under the stated convention and still well above 1 under the alternative, so the conclusion that shifting reduces global tax payments survives.

The authors' response. TWZ Appendix C.3.1: statutory rates approximate the marginal rate on shifted profits (Devereux & Maffini 2007).

7minorconfirmed · documentation gap

The brief's 'about 55%' cannot be traced to a stated list of havens; its own five havens plus Puerto Rico give about 52%. The difference is small and does not change the picture.

Technical detail and evidence

Why it matters, in detail. Share = (sum_h PTR_h + PR)/(sum PTR + PR), with PR of about $40-43bn from TWZ's national-accounts estimate for 2015, mixed with BEA 2016 MOFA profit-type return. Reaching 55% needs Belgium (TWZ: 'borderline', effective rate 21.8%) and Hong Kong (Wright & Zucman: 'borderline').

Evidence. BEA 2016 preliminary Table II.F.1 [scripts/bea2016_check.py]: five havens 47.9%; plus Puerto Rico at $40bn 52.3% (at $42.9bn 52.6%); plus Luxembourg 52.4%; plus Luxembourg and Hong Kong 53.5%; plus Luxembourg, Hong Kong and Belgium 55.1%. TWZ2018AppendixTables.xlsx Table A.7!C89 = 42.913 (Puerto Rico). Wright & Zucman (2018), NBER w24983, p.6.

Why this severity. The share lies between 52% and 55% depending on the haven list, so 'about 55%' is at the top of a reproducible range. No conclusion changes.

8minorconfirmed · previously known · methodology

When companies place patents and brands in tax-haven subsidiaries, the brief counts the resulting profits as 'paper profits'. It says so openly, and the choice fits its argument. But some economists would count part of these profits as genuine, which would make 'shifting' look somewhat smaller.

Technical detail and evidence

Why it matters, in detail. The decomposition attributes to real effects only differences in tangible capital intensity, so IP returns fall into the residual. Whether IP relocation (e.g. via cost-sharing) counts as avoidance is definitional: Dharmapala's broad vs narrow concepts. It does not affect measured haven profits or the tangible-capital comparison, but it raises the macro share relative to micro concepts.

Evidence. TWZ Aug-2018 pp.13-14 ('We treat movements of intangible as profit shifting'; 'movements of intangibles have no particular reason to affect workers' productivity'); intangibles are not in TWZ's capital measure (TWZ Aug-2018 Sec. 4.2, per Auditor A). Dharmapala (2019), AEA P&P 109:488-492. Brief pp.2 and 4.

Why this severity. The definition is stated in the brief itself and is coherent for its distributional argument. The descriptive fact that havens hold little tangible capital relative to the profits booked there is robust. The caveat matters mainly when 'shifting' is read as revenue recoverable under current rules.

The authors' response. TWZ (2018) justify treating intangibles as shifting by their distributional equivalence to transfer mispricing; TWZ (REStud 2023, Sec. 3.3) show that haven excess profitability persists after controls for capital, R&D and industry.

9minorconfirmed · newly identified here · overclaiming

The brief says very low tax rates of about 5% bring small havens the most revenue. The data cited do not show this: havens with rates of 12-14% collect about as much as those near 5%.

Technical detail and evidence

Why it matters, in detail. R(t) = t x Q(t) peaks where the elasticity of Q with respect to t equals -1. A cross-section of nine heterogeneous countries (differing in size and sector mix) cannot trace Q(t), and the correlation in TWZ's own data is weak and statistically insignificant.

Evidence. TWZ2018.xlsx 'Data F8c' (cols. B, I) and 'DataF8a' (col. D) [scripts/haven_rate_revenue.py]; TWZ Aug-2018 p.26. Codex's identification example: Q(t) = 100 x exp[-b(t - 0.05)] with b = 10, 20 or 40 gives the same observed base and revenue at 5% but optima of 10%, 5% and 2.5%.

Why this severity. An aside explaining havens' behaviour; it affects no headline number.

10minorconfirmed · newly identified here · internal inconsistency

The text says havens tax US companies at 5-10%, but the Netherlands figure in the brief's own chart is almost 12%. Tax rates in havens are still very low; the stated range is just slightly off.

Technical detail and evidence

Why it matters, in detail. The upper end of the range is understated by 1.75pp (2.3pp on revised data). The denominator nets out loss-making affiliates, which inflates non-haven rates; haven rates are less affected because haven affiliates are highly profitable.

Evidence. BEA 2016 preliminary Table II.D.1 col. 8 / Table II.F.1 col. 3: Netherlands 5,088/43,310 = 11.75%; France 3,328/6,072; Brazil 1,063/1,720; Germany 4,058/9,295 [scripts/bea2016_check.py]. Revised 2016 data: Netherlands 12.3%.

Why this severity. A presentation slip: haven rates are low (5.7-11.7%) whatever the exact range, and the plotted points are correct.

11minorconfirmed · newly identified here · version instability

The striking comparison, that Ireland alone books more US profits than five big economies combined, held by a hair in the data available in 2018 and no longer holds in the revised data. Ireland's profits are still about as large as those five countries' combined.

Technical detail and evidence

Why it matters, in detail. The margin moves from +0.2% (preliminary) to -1.3% (revised). With a margin this small, any change in comparators or data vintage reverses the inequality.

Evidence. BEA usdia2016p and usdia2016r, Table II.F.1 [scripts/bea2016_check.py; both auditors obtained identical figures]. The Germinal 2020 text (Wayback capture of 21 Dec 2025) repeats '76,5 milliards de dollars'.

Why this severity. An illustrative comparison: Ireland's roughly $77bn, about as much as five large economies combined, makes the same point either way.

12minorconfirmed · newly identified here · methodology

The brief says the world's average corporate tax rate fell by more than half after 1985. That is true for rich (OECD) countries, but a consistent worldwide average fell by about 40-45%, which is still very large.

Technical detail and evidence

Why it matters, in detail. Splicing an unweighted OECD-21 average (1985) onto a KPMG world average for about 150 countries (2018) mixes samples. The scaling factor is also inverted (x OECD/world instead of x world/OECD), adding about 1pp. Consistent samples give falls of -40% (unweighted) to -45% (GDP-weighted).

Evidence. TWZ2018AppendixTables.xlsx Table F1b!G43 = 0.4818 (AVERAGEIF > 0); G44 = G43 x (SUM(Y43:AN43)/SUM(TableF1a!C93:R93)) = 0.4867; AN44 = 0.24 [scripts/twz2018_check.py]. Tax Foundation worldwide corporate tax rates panel (github.com/TaxFoundation/worldwide-corporate-tax-rates, final_data/final_data_long.csv) [scripts/world_rate_1985_2018.py]: -40% unweighted, -45% GDP-weighted; balanced panel -40%/-44%; OECD -49%/-46%. Tax Foundation 2018 report: unweighted world average 38.84% (1980) vs 23.03% (2018).

Why this severity. A background claim; the decline is large on any definition (40-50%).

The authors' response. TWZ Appendix Sec. F.1.2 discloses that pre-2003 world averages are imputed from OECD data.

13minorconfirmed · newly identified here · misrepresentation of source

The brief cites an earlier paper to say that the taxes companies actually pay fell worldwide about as much as headline rates. That paper covers only the US, and there the taxes companies actually paid fell much more than the headline rate.

Technical detail and evidence

Why it matters, in detail. Across decades the cited source shows the US effective rate down 59% against a 31% fall in the statutory rate, and in 1998-2013 a 10pp fall against no change. It provides no evidence on other countries.

Evidence. Zucman (2014), JEP 28(4) pp.131-133 and Figure 5. Data appendix workbook, sheet DataFig1-5, 'Effective tax rate paid worldwide': 0.4658 (1950-59) -> 0.1929 (2010-13); 0.2928 (1998) -> 0.1986 (2013). Federal statutory rate: 42% (1950), 50.75% (1951), 52% (1952-59), giving a 1950s average of 50.9%, and 35% in 2010-13.

Why this severity. A background claim. Its direction ('declined a lot') is supported; its scope ('globally') and comparison ('almost in line') are not.

14minorconfirmed · newly identified here · version instability

The French translation published in 2020 twice says 'intangible' capital where the original says 'tangible' (physical) capital, which garbles a key distinction. It also repeats the 2018 figures without updating them.

Technical detail and evidence

Why it matters, in detail. The swap reverses the mechanism (relocation of real capital vs shifting of paper profits) that the brief uses to argue that profit shifting has no wage effects. A French reader gets the opposite statement in two passages.

Evidence. Germinal 2020/1 no.1 pp.222-237 (doi:10.3917/ger.001.0222), full text from a Wayback capture (20251221110859) of shs.cairn.info/revue-germinal-2020-1-page-222?lang=fr&tab=texte-integral. Byline 'Traduit de l'anglais par Cyprien Carraud'; passages corresponding to brief p.2 (two paragraphs) and p.2 (BEA paragraph); note [1].

Why this severity. These are translation errors in a republication; the numbers are unchanged and the English original is correct.

15minorconfirmed · newly identified here · other

There are several typos, a missing reference and a few loose descriptions; none changes what the brief shows.

Technical detail and evidence

Why it matters, in detail. No effect on any estimate; documentation and attribution only.

Evidence. Brief pp.1-6. BEA 2016 preliminary Table II.F.1 (Bermuda -5,246; Dominican Republic 608; $m). GewStG s.29(1) no.1; Canada Income Tax Regulations s.402(3)(a). Wright & Zucman (2018), NBER w24983, p.5. Avi-Yonah & Clausing (2007), Hamilton Project, doi:10.2139/ssrn.995202. Keen & Konrad (2013), Handbook of Public Economics vol. 5, doi:10.1016/B978-0-444-53759-1.00005-4.

Why this severity. Cosmetic or peripheral; none affects a number or a conclusion.

16minorplausible · newly identified here · speculative assumption

The brief says that under its plan the tax would be paid mainly by shareholders, which is what makes it progressive. That may be right. But because the tax would depend on where sales are made, part of it could show up as higher prices for customers, and the brief gives no evidence either way.

Technical detail and evidence

Why it matters, in detail. For a firm with global profit P, T_A = t_A x (S_A/S) x P. At the margin this taxes sales into A at rate t_A x P/S (the average margin) as well as global profit. How much is passed forward into prices in A depends on market structure. With pure rents the burden falls on owners; once normal returns are in the base, a closed-economy-like burden on capital holds only if the global supply of capital is inelastic.

Evidence. Brief p.5. Gordon & Wilson (1986), Econometrica 54(6):1357-1373, doi:10.2307/1914303. McLure (1981), 'The Elusive Incidence of the Corporate Income Tax: The State Case', Public Finance Quarterly 9(4), doi:10.1177/109114218100900402.

Why this severity. The progressivity point is secondary and explicitly hedged ('an exact quantification of this effect warrants further research'). The tax could plausibly fall largely on owners if the apportioned base is mostly rents, so the claim is uncertain rather than wrong.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

Taking haven data at face value lowers TWZ's estimate by only 17.73% ($507bn, REStud Table 2 col. 7), so the global headline depends only moderately on TWZ's corrections to haven statistics (used to grade I2 'moderate'). — Corrected. In TWZ's own 2018 workbook, the $306.4bn of 'unrecorded profits' added to haven statistics passes one-for-one into shifted profits (Table A.7 col. 15 = foreign profits including the additions minus a fixed local benchmark times foreign wages). Taking haven data at face value therefore gives $310bn (-50%), not $507bn. The published column removes only about $109bn, a unit error identified in the project's TWZ (2023) adjudication. Face value is an extreme bound, but the dependence is about half, not 18%.
Within TWZ's own alternatives the US loss ranges from 7.5% (effective rates) to 30.2% (ultimate-ownership allocation). — The 30.2% assigns to the US all profits shifted by US-owned multinationals, from any country. It answers a residence-based question and is not an alternative estimate of the source-country revenue loss the brief reports. The relevant sensitivity range is about 7-14%.
The US figure is 'rounded up twice' because TWZ hard-code a 40% rate instead of the 39.0% combined rate in their own Table F1b. — Immaterial: 39% gives 13.7% instead of 14.0%, and the 2015 US combined federal-state rate was about 39%, so 40% is a rounding. Only the final 14.0% -> 'about 15%' rounding is kept (F2).
A sales-only formula merely moves the distortion to the location of sales rather than eliminating it (Gordon & Wilson 1986). — Qualified rather than accepted. With a destination (customer-location) sales factor, the distortion of production location does disappear for the adopter's tax, because customers are largely immobile. What remains is an implicit tax on sales into the jurisdiction (an incidence question, F9) and scope for routing sales through intermediaries (F4). Gordon & Wilson analyse three-factor formulas.
The unilateral-scope overstatement (I1) is a major defect of the brief. — Downgraded to moderate (F3). The brief's operative claim is that 'at the country level' the base can be made largely inelastic and that a country can act unilaterally, and the counterexample confirms this. It defeats only the categorical system-wide wording, which needs a disclosed qualification but does not overturn the policy idea.
The booked-vs-shifted comparison (I3) and the revenue-maximizing 5% rate (I4) are moderate issues. — Both confirmed but downgraded to minor (F5, F10). On consistent definitions the US-vs-world ranking holds (52-55% vs 39-43% booked), and the 5% claim is an aside relayed from TWZ that affects no headline number.
'Mostly explained by shifting effects is partly true by definition' is a moderate issue (I4). — Confirmed but downgraded to minor (F8). The brief itself lists 'locating intangibles ... in tax haven subsidiaries' as a shifting channel (p.2), and TWZ state and justify the definition, so the choice is disclosed and consistent with the brief's distributional argument.
Corpus lead offshore-havens-08 (Blouin & Robinson: BEA data double count lower-tier income) applied to the brief's BEA statistics; examined by both auditors. — Does not apply. The brief uses BEA profit-type return (Table II.F.1), which excludes equity income from lower-tier affiliates and capital gains, and endnote 2 discloses this. A Blouin-Robinson-style adjusted pre-tax income gives a five-haven share of 44.4-48.8%, against 47.9%; net income, which double counts, would give 57.6% [scripts/bea2016_check.py].
Corpus lead us-wealth-taxes-13 (the whole corporate tax assigned to shareholders) applied to the brief's progressivity argument. — Does not apply in that form. The brief (p.5) says part of the corporate tax may be shifted to labor and states the elasticity conditions, and it makes no distributional allocation that assumes full shareholder incidence. The narrower incidence point under sales apportionment is kept as F9.
'Canada and Germany use similar apportionment mechanisms' is a factual error. — Partly rejected. Both do use formula apportionment, which is the brief's point; only the implication that they are sales-based is loose. Kept as a wording slip in F16.

Audit notes

What reproduced. Reproduced exactly from primary or author sources with adjudicator scripts: all BEA 2016 statistics in the brief, including Figure 1's 15 points ($434.7bn; $76.5bn; 47.9%; 358% vs 42%); TWZ's 36.19% ($616.46bn / $1,703.23bn) and US 14.03% (0.40 x $142.05bn / $404.99bn); the haven-revenue ratio (5.1:1 after removing TWZ's duplicated Netherlands row, 4.4:1 with it); TWZ's 1985 'world average' (48.18% x 1.0103 = 48.67%); TWZ REStud Table 2 scenarios ($506.2bn, $507.2bn, $369.0bn, $739.9bn). Conditionally reproduced: 'about 55%' (52.3% with the brief's five havens plus $40bn for Puerto Rico; 55.1% only with Luxembourg, Hong Kong and Belgium added). Not reproduced as stated: '40%' (36.2%); 'about 15%' (14.0%); '5%-10%' (Netherlands 11.7%); 'more than half, from 49%' (consistent world averages fall 40-45%; corrected scaling gives 47.7% and a 49.7% fall); 'around 5% is revenue-maximizing' (no reliable relation in TWZ's cross-section: Spearman -0.43, p = 0.24, and -0.02, p = 0.97); the Ireland comparison in BEA's revised data (-$1.0bn). New sensitivities: $306.4bn (49.7%) of TWZ's shifted profits are imputed haven profits, so face value gives $310bn (18-22%); the US loss is 7.5% at average effective rates and scales with the global total (9.7-12.5% if that total is $425-550bn). Not reproduced: TWZ's raw-data pipeline, the CbC micro estimates of comparison studies, and the legal feasibility of unilateral apportionment.

Limits of this audit. Focused adjudication of a six-page policy brief. TWZ's global model was not rebuilt from raw FATS, balance-of-payments and national-accounts data; checks start from the authors' June-2018 and 2022 workbooks. The 'defensible range' of roughly 25-36% for the global share draws on the project's separate adjudication of TWZ (2023), whose reconciliation scenarios were not re-derived here; this adjudication independently confirmed only the one-for-one pass-through of the $306bn imputation and the resulting face-value bound. The full text of Blouin & Robinson (2025, JPubE) could not be read (ScienceDirect HTTP 403); their argument is taken from Blouin's 2025 conference slides and the corpus. Fuest et al. (2022 WP), Laffitte & Toubal (author manuscript), Forstater (2018) and Garcia-Bernardo & Jansky (2023 arXiv version) were read in cached versions; Dharmapala (2019) only via the corpus and auditors. The legal and treaty feasibility of unilateral sales apportionment and the empirical incidence of a sales-apportioned tax were not analysed. The French version was read from a Wayback capture of Cairn's full-text tab (the PDF is blocked). The Tax Foundation panel used is a recent vintage, not the 2018 one. No author replication files exist for the brief.