Thomas Tørsløv, Ludvig Wier, Gabriel Zucman · Review of Economic Studies 90(3): 1499-1534

The Missing Profits of Nations

19 problems found: 3 major · 7 moderate · 9 minor

What the paper claims

36% of multinational (foreign-affiliate) profits were shifted to tax havens in 2015, with a lower bound of 22% and an upper bound of 43%.
Affiliates of foreign multinationals are an order of magnitude more profitable (profits-to-wage) than local firms in low-tax countries.
US multinationals shift twice as much profit as other multinationals relative to the size of their foreign earnings; about half of shifted profits accrue to US parents.
Reallocating shifted profits to source countries would raise domestic profits by about 20% in high-tax EU countries, 10% in the US and 5% in developing countries; EU countries are the prime losers.
Corrected for profit shifting, the corporate capital share has risen in high-tax European countries too, not only in North America.
Two fully independent methodologies and data sources (excess profitability; excess high-risk service and interest receipts) give the same magnitudes, globally and haven by haven.
Most profits shifted out of the EU go first to EU havens, but about half ultimately end up in non-EU havens.

Every problem found

1majorconfirmed · previously known · arithmetic or transcription

The paper checks what happens if one trusts tax havens' own (incomplete) statistics instead of the authors' corrections, and reports that the estimate would fall by only 18%. Because of a units mistake in the spreadsheet, the correct fall is about half, and the paper's stated minimum of 22% becomes roughly 10-12%. The main 36% figure is unchanged, but it leans on the authors' data corrections far more than readers are told.

Technical detail and evidence

Why it matters, in detail. About half of S = $616bn consists of 'unrecorded profits' U = $306bn: partner-reported DI income replacing haven data, stock-based returns for the BVI and Curacao, and a pro-rata plug closing the global DI-income gap. Shifted profits move one-for-one with U when wages and pi_l are held fixed. The face-value scenario is therefore $310bn, and the combined lower bound (benchmark +45pp, sigma = 1.3, no corrections) is $169-209bn: 10-12% of foreign profits (12-15% on a denominator net of U), not $369bn or 22%. Face value is an extreme scenario: Bermuda's national accounts set offshore operating surplus to zero, and Cayman excludes its offshore sector. The corrected figure is thus a bound, not a plausible estimate. But it is the bound the paper reports and propagates into the Fig. 8 band.

Evidence. TWZ2022ReplicationGuideTables.xlsx (Zenodo 6790852; identical values to the author-site copy): TableE4!E24 '=TableA7!C24*(TableA7!M24-TableA7!I24-TableA7!O24)/TableA7!M24'; TableA7!I24 '=TableA6!I24/TableA2!$C24'; TableA7!M24 '=C24/TableA4!D24'; Table A.6 col. 3 haven total $306.39bn. Adjudicator recomputation (final/scripts/face_value.py): the published formula removes $109.3bn and reproduces $507.16bn; removing U in full with wages and pi_l fixed gives $310.07bn (-49.7%; 18.2% of $1,703bn, 22.2% of $1,397bn). By haven: Ireland 106.3 -> 57.6 (published 94.8), Luxembourg 46.8 -> 15.0, Netherlands 57.4 -> 22.0, Singapore 70.4 -> 14.9, Bermuda 24.0 -> 0.3, Caribbean 96.7 -> 31.1. Lower bound (TWZ2022.xlsx Table2 col. J = E + I + G - 2C by haven group) with the corrected col. 7: $169.5bn (10.0%) without floors and $208.7bn (12.3%) with group floors at zero, vs the published $369.0bn (21.7%). Non-OECD foreign wages are 40% of national-accounts compensation or derived from DI-based operating profits (Table A.4 col. 3), not from the additions, so $310bn is the exact face-value counterpart within TWZ's framework. Both auditors obtained the same figures independently.

Why this severity. The 36% point estimate is unchanged, but the paper's quantified robustness to its most contested assumption is wrong by a factor of about 2.8 (-18% vs -50%), and the advertised 22% lower bound roughly halves. The headline relies on data reconciliation much more than the paper discloses.

The authors' response. No response to the formula error located. The paper argues the haven data 'must be corrected' because some are clearly incomplete (Cayman excludes its offshore sector) and because, at face value, global inward DI income would fall short of outward income (pp. 1520-1521).

2majorconfirmed · newly identified here · internal inconsistency

The paper claims that, once profit shifting is accounted for, the share of output going to corporate profits rose in Germany, France and Italy. That result depends on a correction that assumes about three times as much shifted profit as the paper itself estimates for those countries. Using the paper's own numbers, the rise is about 1 to 1.5 points rather than 3 to 4, and Germany's share actually falls.

Technical detail and evidence

Why it matters, in detail. Adding S dollars to GP gross profits and value added requires rho = S/Pi_GP. TWZ use rho = tau*S/T, which equals S/Pi only if Pi_GP = T/tau, the statutory-rate-implied tax base. German and Italian national-accounts corporate sectors include partnerships, and effective rates are below statutory, so T/tau is a fraction of Pi_GP. With rho consistent with Table 3 (Germany 0.076, France 0.099, Italy 0.083 on official GVA), the corrected 1985-2015 change is +1.2pp. It is +1.5pp on a 75% base and +1.0-1.3pp if interest is also excluded. That compares with +2.98pp published and 'about 4' in the introduction. By country: Germany -2pp, France +3.9pp, Italy +1.6pp.

Evidence. RG sheet GP2020data: L275 = 0.28, L527 = 0.21, L711 = 0.19; corrected share O = K(1+L)/(1+K*L); L back-cast by X = haven share of global profits. RG Table A.2 corporate gross value added: Germany 2,073.0, France 1,308.5, Italy 872.8. TWZ2022.xlsx Table3: Germany 54.9, France 32.1, Italy 22.7. Adjudicator (final/scripts/capital_share.py; the stored series is reproduced exactly): - published method: +2.98pp, 2015 level adjustment 4.46pp; - rho consistent with Table 3: +1.16pp on the official GVA base to +1.54pp on a 75% base, with Germany at -2.05 to -1.69pp; - consistency with the published rho would require the GP base to be 27% (Germany), 47% (France) and 44% (Italy) of official corporate GVA; - also excluding interest: +0.97 to +1.30pp (Auditor B: interest exclusion alone gives +2.54pp); - the consistent 2015 level adjustment is 1.5-2.3pp, in line with Table 4's 1.8-2.1pp for these countries.

Why this severity. The corrected European trend, the paper's macroeconomic headline, is overstated about 2-3 times. On magnitudes consistent with the paper's own shifted-profit estimates, the average rise is about 1-1.5pp and Germany's corrected share falls. The claim that Europe's capital share rose substantially, approaching the US pattern, is therefore not supported, although a small average rise survives.

The authors' response. The exercise is described as 'merely illustrative' (p. 1529). Sec. 6.1.1 motivates tax-based ratios by the inclusion of non-corporate income in German and Italian national-accounts profits.

3majorlikely · newly identified here · internal inconsistency

The paper says US companies shift about twice as much profit as other multinationals, relative to their foreign earnings. But the amount it attributes to US firms is more than US government survey data say US firms' affiliates in tax havens booked in the first place. With that limit imposed, US firms shift perhaps 1.2 to 1.7 times as much as others: still more, but not twice.

Technical detail and evidence

Why it matters, in detail. Allocating a flow (shifted profits) by a stock key (ultimate-owner FDI positions, swollen for US multinationals by pre-2018 deferred earnings) attributes about 49.5% of S to US parents. The BEA ceiling implies US shifting of at most $225-259bn, i.e. at most 40-46% of US foreign profits. Others' share is 27% if the excess is an over-count of the total, or 31-34% if it is reallocated, so the intensity ratio is about 1.2-1.7 rather than 2.0. 'About half of shifted profits accrue to US parents' falls to about 36-42%.

Evidence. RG sheet FractionShifted: D5 = 562.8, E5 = 305.41, L46 = 267.76, M46 = 305.41, note J14. RG Table C4 col. 12 (N45 = TableA7!P91 x TableC3!C43/TableC3!C91, a stock-based key). BEA revised 2015 Table II.F.1. Adjudicator bound (final/scripts/us_share_check.py): - excess of BEA profit-type return over TWZ's own local-profitability benchmark: $187.3bn; - plus Puerto Rico ($37.7bn): $225bn; - with $10bn for havens hidden in BEA aggregates and +10% for minority-owned affiliates: at most $259bn; - so the US share is 40-46% of foreign profits, and the ratio to others is 1.2-1.5 if the global total is unchanged or 1.5-1.7 if the excess is simply removed (Auditor A: 1.1-1.4).

Why this severity. Once US-attributed shifting is constrained by BEA's own count of profits booked by US affiliates in havens, the abstract's 'twice' factor falls to roughly 1.2-1.7, a 15-40% reduction. The qualitative ranking (US multinationals shift more) survives.

The authors' response. None located. The workbook note attributes the 'small ~10% gap' between the GBJZ US haven-profit series and TWZ to small havens excluded from GBJZ (FractionShifted J14).

4moderateconfirmed · newly identified here · overclaiming

The authors say a second, completely separate method gives the same answer, which is meant to reassure readers. In fact about a third of that second estimate was built from the first estimate, or from goods-export data the method was supposed to exclude. Where the second method really is independent, it agrees only in total, not country by country.

Technical detail and evidence

Why it matters, in detail. Agreement between the methods is presented as validation of the level of S and of its distribution across havens. With $160.5bn allocated pro rata to the baseline and Puerto Rico based on merchandise exports, the comparison is informative only for the EU havens, Switzerland, Singapore and Hong Kong. There the totals are $441bn vs $404bn, and haven-level ratios range from 0.35 (Malta) to 2.6 (Belgium). Inferences drawn from the gap between the methods are therefore partly circular: the 5% attributed to haven-headquartered firms (Sec. 5.2.3) and the Replication Guide's conclusion that goods mispricing is small.

Evidence. TWZ2022.xlsx Table2!D9 '=(D16-D4-D5-D6-D7-D8-D11-D12-D13-D14)*C9/(C16-C14-C13-C12-C11-C8-C7-C6-C5-C4)'; D10 is analogous and D15 is the residual. Together 90.68 + 22.45 + 47.41 = $160.5bn (24.8% of $646.3bn); each is 0.9375 x the baseline. Puerto Rico col. 2 = $44.5bn, built from TWZRawDataC sheet 'Puerto Rico' (Planning Board Table 23: FY2015 recorded merchandise exports $69,391m, of which pharmaceuticals $48,764m) x 73% (RG Sec. C.1.1). Havens with own data (Belgium, Ireland, Luxembourg, Malta, Netherlands, Switzerland, Singapore, Hong Kong): $441bn vs $404bn baseline. By haven: Belgium 34 vs 13, Luxembourg 88 vs 47, Malta 4 vs 12, Hong Kong 28 vs 39, Netherlands 74 vs 57, Ireland 89 vs 106. Swiss, Singapore and Hong Kong exports are scaled up 25% (RG Sec. C.1.1).

Why this severity. The baseline is unaffected, but the main corroboration advertised for it is weaker than claimed. About a third of the alternative ($205bn) is mechanically tied to the baseline or to goods data, and the genuinely independent part agrees in aggregate, not haven by haven.

The authors' response. The imputation is disclosed in the Replication Guide (Sec. C.1.1) but not in the paper. No response located.

5moderateconfirmed · newly identified here · internal inconsistency

The summary says profits would rise about 20% in high-tax EU countries and 10% in the US if shifting stopped. By the paper's own profit figures the rises are about 12% and 8%; the 20% is actually a tax-revenue number, and the US 10% matches neither measure. Europe still loses more than the US, but by less than the summary suggests.

Technical detail and evidence

Why it matters, in detail. On the profit measure the high-tax-EU/US ratio is 1.6 (12.1/7.5); on the tax measure it is 1.5 (20.6/14.1). The abstract's 20/10 implies 2.0. The tax measure is defensible for Germany and Italy, whose national-accounts profits include non-corporate income, but it is not a profit increase and is not applied to the US.

Evidence. TWZ2022.xlsx Table3: H47 = 0.1208, I47 = 0.2058, H46 = 0.1115, I46 = 0.1818, H34 = 0.0755, I34 = 0.1408, H49 = 0.0396, I49 = 0.0537. WIL WP 2020/12 abstract. REStud abstract p. 1499; Sec. 1 p. 1500; Sec. 6.1.1 p. 1523.

Why this severity. An abstract-level quantitative claim overstates the EU's profit loss (about 20% vs 12% on the stated measure) and the EU-US contrast (2.0x vs about 1.5x on either consistent measure). The ranking EU > US > developing holds.

The authors' response. The body explains why tax-based ratios better capture losses where national-accounts profits include non-corporate income (Sec. 6.1.1-6.1.2). The abstract nonetheless labels the figures as profit increases.

6moderateconfirmed · newly identified here · documentation gap

For the smaller tax havens the authors do not actually observe how profitable local firms are. They assume a common figure (48%) and then show it in a chart as if it were data. Because the estimate subtracts this assumed 'normal' profit, reasonable alternative assumptions would change the global total by up to about a tenth, mostly downward.

Technical detail and evidence

Why it matters, in detail. For non-OECD havens S_h = B_h - pi_l*W_h with pi_l imposed, so dS/dpi_l = -$315bn per unit. A change of +/-0.12 in pi_l therefore moves S by about $37-44bn in the opposite direction. Local-sector profitability in developing economies in TWZ's own data (aggregate 0.69) exceeds the OECD-weighted 0.48, so if anything the imposed benchmark biases S upward.

Evidence. RG Table A.7: O53 '=(E9+E45)/(TableA4!G9+TableA4!G45)' = 0.480, and O54:O89 = O$53. TWZ2022.xlsx DataF4 D2, D6, D7 = 0.48. RG Table A.4 B97 = 0.4. Developing countries' local pi_l in the same table ranges from 0.42 (Brazil) to 1.24 (India), with an aggregate of 0.69. Adjudicator sensitivity (non-OECD compensation $314.7bn): pi_l 0.41 -> +$22bn (+3.6%); 0.60 -> -$37bn (-6.1%); 0.68 (Ireland's observed local level) -> -$62bn (-10.1%).

Why this severity. The assumption is undisclosed in the paper, and only the favourable direction of sensitivity is reported. Plausible alternative benchmarks move the global total by -6% to -10%, and the paper's validation of the benchmark is circular for these havens.

The authors' response. Disclosed in RG Sec. A.3.3. The paper reports that pi_l = 41% would raise the estimate by 3% (Sec. 5.2.3).

7moderatelikely · previously known · internal inconsistency

The authors add about $119 billion of profits to Ireland, Luxembourg, the Netherlands and Belgium, because US statistics record more investment income coming from those countries than they report paying out. But the paper's own comparison of company statistics shows almost no shortfall, and the US government's survey of US companies' foreign affiliates puts their profits there $70-80 billion below what the authors' corrected figures imply. Fixing this would trim the global estimate by roughly a tenth to a sixth.

Technical detail and evidence

Why it matters, in detail. The upgrade adds the US-reported excess of DI income over haven-reported DI payments to FATS-based foreign profits. DI income is recorded against the immediate counterpart, with US equity-method reinvested earnings. For holding-company affiliates it therefore carries lower-tier earnings that host-country FATS already count where earned: the Blouin-Robinson mechanism. The paper's own Table A.8 shows a net GOS gap of about $0. The gross shortfall of $12.5bn is 5-10 times smaller than the upgrade, even granting TWZ's R&D-depreciation explanation for Ireland. TWZ's implied US-owned profits exceed BEA MOFA profit-type return by $76-83bn. RG fn 20 argues that a smaller EU upgrade would only move profits to non-EU havens through the global DI balance. That defence is limited for two reasons: BEA caps total US-owned haven profits (F6), and the un-netted Luxembourg/Belgium receipt upgrades ($65.5bn) are what enlarge the residual gap the plug fills. Net effect on S: -$35bn (netting rule only) to -$100bn.

Evidence. RG Tables A.6 (cols. 3, 8-10), A.8, A.10, B.10, B.11; BEA revised 2015 Tables II.F.1 and II.D.1. FATS gross operating surplus of US-controlled affiliates, haven-reported vs US-reported ($bn; Table A.8 row 'USA'): Belgium 7.82 vs 7.86; Ireland 88.95 vs 76.35 (Ireland reports more); Luxembourg 2.41 vs 5.63; Netherlands 17.01 vs 26.23. Gross shortfall $12.5bn; net about $0. Adjudicator (final/scripts/eu_haven_us_check.py): TWZ-implied US-owned profits = US share of haven FATS GOS x (FATS operating profits + SPE net) + US upgrade = Belgium 12.4, Ireland 107.4, Luxembourg 32.8, Netherlands 44.3, total $196.9bn. BEA majority-owned-affiliate profit-type return is $120.9bn (TWZ's preliminary vintage, Table A.10) or $113.7bn (revised). BEA income from equity investments: Netherlands $172.3bn, Luxembourg $96.2bn, Ireland $67.2bn, vs profit-type return of $23.3bn, $9.6bn and $73.6bn. Receipt upgrades (Table B.10 col. 14): Luxembourg 48.8, Belgium 16.7; not netted in Table A.6. Auditor A's scenario (BEA profit-type return for US-owned affiliates, haven FATS for the rest) gives -$93bn (TWZ vintage) to -$100bn (revised). Applying the non-OECD netting rule alone removes Luxembourg's and Belgium's additions (-$35.3bn).

Why this severity. Resolving the inconsistency toward the paper's own FATS/BEA benchmarks lowers the global estimate by about $64-100bn (10-16%). That is noticeable but not transformative for C1 on its own. It changes the haven ranking (Ireland about 106 -> 71, Luxembourg 47 -> 15-17, Netherlands 57 -> 24-34) and is the main mechanism behind the overstated US share (F6). Combined with F3, the reduction in C1 reaches roughly 30%.

The authors' response. RG Sec. A.3.3 (pp. 18-21) argues that US data are reliable (they match non-haven partners, OA Fig. G), that FATS show similar discrepancies, and that Ireland's FATS excess reflects R&D capitalisation (RG Sec. A.4.1). It adds that a smaller EU upgrade would only change the geography of shifting (fn 20). Paper Sec. 4.2.3 acknowledges that US statisticians may misassign profits to Ireland (CSO 2016). No reply to Blouin & Robinson (2025) located.

8moderatelikely · previously known · speculative assumption

For smaller havens the authors fill gaps in the statistics with estimates, including a top-up that makes world investment-income statistics balance and partner countries' figures added on top of the havens' own accounts. In Singapore, Hong Kong and Malta, which have complete national accounts, these top-ups add about $92 billion; for Singapore they imply more shifted profit than all the corporate profit its statistics record. Removing such top-ups would lower the global estimate by roughly 15%.

Technical detail and evidence

Why it matters, in detail. Within TWZ's framework, S_h = B_h - 0.48*W_h for all 36 non-OECD havens. Every dollar added to corrected corporate profits B_h is therefore a dollar of shifting. The plug assumes that the residual global DI-income discrepancy, after max-of-mirror corrections to both credits and debits, is unrecorded haven profit. It allocates that residual by a key that rewards havens with large reported DI flows (Hong Kong). Partner-DI and stock-based additions for holding-company jurisdictions (Singapore, Hong Kong, BVI) can include lower-tier earnings of affiliates in countries whose DI payments are not observed (e.g. China). That is equity income, which TWZ's own profit definition excludes (Sec. 4.2.1). For Singapore the addition is inconsistent with its national accounts.

Evidence. RG Tables A.2 (rows 72, 77, 83), A.4 (H83 = -16.78), A.6 col. 3 (Singapore 55.53, Hong Kong 21.29, Malta 11.31, Cyprus 3.76) and formulas (D77 includes TableB10!E77), A.7, and B.10 (E72 = F72 = 137,000; G72 = 22,138; M72 = 20,280; M220 = 45,354; L64 = 31,376; L66 = 12,940; E77 = 10,100; F83 = 0; G83 = 48,949; $m). Adjudicator scenarios (final/scripts/nonoecd_scenarios.py, reproducing col. 3 exactly): - dropping the stock imputations and the plug gives -$87bn (Auditor A: -$91bn); - no additions in the four havens with SNA-compliant national accounts (Singapore, Hong Kong, Malta, Cyprus) gives -$92bn; - both together give -$147bn ($469bn, 27.6%). The authors' stated reconciliation target is global DI-income balance. Closing only the raw IMF global DI gap ($157-187bn; RG Tables B.9-B.10) would need about $165-196bn of additions instead of $306bn, i.e. about $475-506bn shifted.

Why this severity. Each alternative lowers the global estimate by about 14-15%. Some imputation is unavoidable for havens whose statistics exclude offshore activity, so the true overstatement is uncertain. Together with F2, plausible alternatives give roughly $425-550bn (25-32% of foreign profits) instead of 36%.

The authors' response. RG Sec. A.3.3: non-OECD havens publish no or very incomplete DI statistics, and the guiding principle is that global DI income paid must equal DI income received. The Cayman Islands exclude their offshore sector (paper fn 24). RG fn 17 notes Malta's balance-of-payments/national-accounts inconsistency.

9moderatelikely · previously known · methodology

Deciding which countries 'lose' the shifted profits requires a rule, and the paper follows where royalty and interest payments come from, which points to Europe. A study that instead follows where firms' workers and capital are located assigns much more of the Irish, Dutch and Luxembourg profits to the US. So Europe's position as the biggest loser depends on that choice.

Technical detail and evidence

Why it matters, in detail. The source split is a key-dependent apportionment of a fixed total, not an observed quantity. On the three EU havens alone, Guvenen et al.'s attribution to the US exceeds TWZ's by about $72bn, a magnitude comparable to a third of TWZ's EU loss of $216bn. Because the implementation scales every source's payments by the global total (F17), any over- or under-statement of S (F1-F3) passes proportionally into every country's loss.

Evidence. RG Sec. C.1.1-C.2; RG Table C4 cols. 1-11. Guvenen et al., NBER w23324 rev. Feb 2022, fn 24. Laffitte & Toubal, AEJ: Policy 14(4) (2022). Coffey (2020, 2021; corpus offshore-havens-16). Paper p. 1522: 'no unambiguously correct way'.

Why this severity. The EU-as-prime-loser magnitude (1.5% of GDP vs 0.8% for the US) depends on a payments-based key that the authors acknowledge has no unambiguously correct alternative. A production-based key shifts losses toward the US. The global total is unaffected.

The authors' response. The paper acknowledges there is 'no unambiguously correct way' to allocate the profits and presents parent-country allocations as complementary (pp. 1522, 1525). Fn 35 links destination-based allocation to Pillar One; fn 16 notes that production location is often indeterminate.

10moderateplausible · previously known · external inconsistency

How much counts as 'shifted' depends on what one assumes multinationals would earn in havens without tax motives. Some studies using confidential company tax reports find a much smaller share (13-22%); another finds even more shifting than this paper. The paper's 36% is plausible, but sits toward the upper end of a wide range.

Technical detail and evidence

Why it matters, in detail. S depends on a counterfactual in which haven affiliates earn local firms' profit/wage ratio. Within the same CbC dataset, Fuest et al.'s estimate moves from 13% to 22% when the counterfactual switches from regression-based to payroll-based. TWZ's own +45pp Paul-Isaka bound lowers S by 18% (22% if the joint effect is transformed correctly; F13). The comparison literature differs in coverage, year, ownership population and denominator (foreign vs worldwide profits), so no single alternative supersedes 36%. But the evidence spans roughly 13% to over 40% of foreign profits.

Evidence. Fuest, Greil, Hugger & Neumeier, CESifo WP 9757 (2022), pp. 25-26 (published JEEA 2025, doi:10.1093/jeea/jvaf007). Fuest, Hugger & Neumeier, JEBO 194 (2022). Dharmapala, AEA P&P 109 (2019). Garcia-Bernardo & Jansky, World Development 177 (2024), 106527 (author version: $817-880bn across linear, quadratic, logarithmic and misalignment models). Guvenen, Mataloni, Rassier & Ruhl, NBER w23324 rev. Feb 2022 (AER 2022), fn 21. TWZ Table 2 col. 3: +45pp benchmark -> $506bn (-18%). Auditor A: +100pp -> about $409bn.

Why this severity. The 36% is a broad-concept estimate at the upper-middle of the credible range. Micro CbC studies with regression counterfactuals find roughly a third to two-thirds as much, while another CbC study finds more, so the external evidence brackets TWZ rather than refuting it.

The authors' response. Sec. 2.1: Orbis misses 83% of multinationals' profits (Fig. 1). Sec. 3.3: haven excess profitability survives capital, R&D and industry controls. Sec. 5.2.1: the +45pp benchmark bound. No direct reply to the CbC studies located.

11minorconfirmed · newly identified here · overclaiming

The finding that much of the profit routed through EU havens ends up in places like Bermuda is partly built by adjusting the flow data so that the totals match the authors' other estimates. It is a reasonable reconstruction, but not something the data show directly.

Technical detail and evidence

Why it matters, in detail. Haven-to-haven payment data are observed (e.g. Luxembourg receives $88bn and pays out about half), but the ultimate shares rest on a reallocation matrix after forced balancing equal to 13% of S. Uncertainty in the haven-level estimates (F2, F3) therefore passes into the routing shares.

Evidence. RG Table C4x rows 42-58 (B46 'Total negative' = 81,245; $m); RG Sec. C.3.2.

Why this severity. This is a supporting result, and the reconciliation is disclosed in the Replication Guide. The 77%/46% split is partly a construct of forcing flows to match the excess-profit estimates, not a pure observation.

The authors' response. RG Sec. C.3.2 describes the adjustments.

12minorconfirmed · previously known · selective presentation

The paper converts shifted profits into lost tax using official headline tax rates. Using the rates companies actually pay, which the authors computed but did not report in the paper, the loss would be about 6% of corporate tax revenue instead of about 9-10%.

Technical detail and evidence

Why it matters, in detail. Revenue loss = tau*S. Using effective rather than statutory tau lowers the global loss from about 9-10% to about 6% of corporate tax revenue. Netting CFC taxation and withholding taxes, which the paper acknowledges but does not quantify, would lower it further.

Evidence. RG Sec. C.3.1; TWZ2022.xlsx Table3 I44 = 0.0913; paper Sec. 6.1.2 and fns 37-38 (p. 1524).

Why this severity. Statutory rates are a defensible benchmark for marginal shifted profits, and the profit-based ratios are shown in Table 3. The omission concerns the revenue framing, not the shifting estimate.

The authors' response. Fn 37: Spengel et al. effective marginal rates give similar results for the EU. Fn 38: CFC rules and transfer prices of intangibles matter for revenue.

13minorconfirmed · newly identified here · arithmetic or transcription

Several of the paper's 'what-if' calculations contain small formula slips. Fixing them changes those alternative estimates by a few percent in different directions and does not change the paper's conclusions.

Technical detail and evidence

Why it matters, in detail. The benchmark scenario understates the paper's own calibrated profitability premium, giving a 22% rather than 18% reduction. The linear CES mapping exaggerates both capital-intensity scenarios: exact values are +4%/-3% vs the published +8%/-5%. The '1+' omissions cut the sigma = 0.7 column, and hence the upper bound, by about $9bn.

Evidence. Adjudicator (final/scripts/benchmark_check.py, ces_check.py): - +56.8pp benchmark gives $481.4bn vs $506.2bn published; - copy-error repair alone gives sigma = 0.7 at $672.9bn and sigma = 1.3 at $586.0bn (published $663.5bn and $585.9bn); - exact CES with copy fixes gives $641.9bn and $597.0bn. Auditor B's figures ($481.41bn; $641.34bn/$597.33bn) are confirmed to within $1bn.

Why this severity. Each slip moves a robustness column by at most about 4% of the baseline, in offsetting directions. Conclusions are unchanged.

14minorconfirmed · newly identified here · arithmetic or transcription

One supporting table in the authors' published database has two rows built from the wrong input. It does not affect the paper's headline numbers.

Technical detail and evidence

Why it matters, in detail. The two rows fail flow conservation (shortfalls of $39.3bn and $58.8bn) and some cells are negative. Users of the database's ultimate-destination series for developing countries and the rest of the world would be misled.

Evidence. TableC7!Q46 '=1000*TableA7!$P$91*TableC3!E44/TableC3!$C$91' (row 91 analogous) vs SUMPRODUCT formulas in other rows. Developing-country ultimate total: $63.3bn vs source flow $102.6bn. Rest of World: $19.9bn vs $78.7bn.

Why this severity. This is an error in a supporting table of the released database. The global total, the source totals and the EU routing shares cited in the paper are computed elsewhere and reproduce.

15minorconfirmed · newly identified here · documentation gap

One of the paper's checks left out two control variables it says it included. Putting them back reduces the measured gap a little, but havens still look vastly more profitable.

Technical detail and evidence

Why it matters, in detail. With the described controls the 2014 haven profit/wage premium is 179pp rather than 211pp; haven profitability is still several times the non-haven level (about 46%). Only seven haven clusters exist (p. 1510).

Evidence. Do-file read as text. Adjudicator re-estimation (final/scripts/rd_regression.py, weighted within-estimator with year x industry effects): 2014 haven coefficient 2.108 (N = 2,556) as published; 1.788 (N = 2,500) with GDP and population retained. This matches Auditor B.

Why this severity. The haven premium falls 15% but remains very large, so the validation evidence for C2 stands.

16minorconfirmed · newly identified here · documentation gap

The earliest ten years of the European data in the key chart were filled in by the authors from a related series, but the chart says the data come from another study. Other reasonable fill-in methods change the result only slightly.

Technical detail and evidence

Why it matters, in detail. The 1985 base of the 1985-2015 comparison is a TWZ construction. Alternative extensions span +2.7 to +3.5pp under TWZ's correction. The splice is second-order next to F5.

Evidence. GP2020_main_series.xlsx (author download): Data col. F blank 1985-1994 for DE, FR and IT. All 63 values for 1995-2015 match. Adjudicator (final/scripts/splice_check.py): corrected 1985-2015 change +2.98pp (TWZ splice), +3.46pp (additive business-share extension), +2.69pp (multiplicative unadjusted corporate share); raw change -0.05, +0.41 and -0.33pp respectively. This matches Auditor B.

Why this severity. The splice is reasonable, and alternatives move the corrected trend by about +/-0.5pp. Its nature is undisclosed and it determines the 1985 starting level.

17minorconfirmed · newly identified here · documentation gap

The paper describes one way of assigning lost profits to countries but its spreadsheets use a slightly different one. The method used is reasonable, but it makes Europe's losses about 14% larger than the described method would.

Technical detail and evidence

Why it matters, in detail. Pooling makes source losses depend only on the global total and the payment key, not on haven-level estimates. It implicitly treats payments to conduit havens as reaching final havens with the global average yield. The described haven-specific method would cut EU losses by $31bn and raise those of the US, developing countries and the rest of the world.

Evidence. TableC4!C20 '=1000*TableA7!$P$91*(TableC2!D18+TableC2!O18)/(TableC2!$D$91+TableC2!$O$91)'. Adjudicator (final/scripts/allocation_check.py), haven-by-haven allocation as described vs TWZ: - non-haven EU $185.5bn vs $216.2bn (-14%); - high-tax EU $157.8bn vs $185.6bn (-15%); - US $146.9bn vs $142.6bn (+3%); - main developing $111.9bn vs $102.6bn (+9%); - EU loss as % of GDP 1.28% vs 1.49%.

Why this severity. Pooling is defensible, because conduit havens pass profits on to other havens, and the EU remains the largest loser relative to GDP under either method. The discrepancy between description and implementation changes the EU loss by about 14%.

18minorconfirmed · newly identified here · internal inconsistency

There are a number of small mislabelled rows, misstatements and outdated data points. None of them changes the paper's conclusions.

Technical detail and evidence

Why it matters, in detail. Labelling and transcription problems in tables, text and references. The BEA-vintage point slightly weakens the case for classifying Belgium (2% of S) and the Netherlands (9%) as havens on effective-tax-rate grounds. Item (j) shows that pi_f > pi_l is not unique to havens.

Evidence. TWZ2022.xlsx Table3 row 44 (C44 = C52, which includes C51) and Table4; RG Tables C4c, A.7, A.10, A.11, F1b; RG figures DataFN1; BEA revised 2015 Tables II.D.1 and II.F.1. Adjudicator computation for (j).

Why this severity. Individually small; none changes a conclusion. Item (j) would raise rather than lower the estimate.

19minorconfirmed · newly identified here · version instability

Before publication the authors described their 36% estimate as 'close to 40%'. The published article uses the accurate 36%.

Technical detail and evidence

Why it matters, in detail. Rounding 36.2% up to 'close to 40%' in widely cited working papers overstated the headline by about 4 points. The journal version corrects the framing.

Evidence. NBER WP 24701 (June 2018, rev. August 2018) abstract and text ('close to 40% of multinational profits ($616 billion out of $1.7 trillion)'); WIL WP 2020/12 abstract; REStud p. 1499.

Why this severity. This is a presentation issue in pre-publication versions; the estimate itself did not change across versions.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

Guvenen et al. report $156.5bn of profits shifted out of the US by US multinationals, not the $158.3bn TWZ cite (part of I13). — TWZ cite the 2021 working-paper version. The February 2022 revision (published AER 2022) switched from gross to net DI income, 'attenuating' the adjustments (its fn 21), and reports $156.5bn. TWZ's figure is an outdated citation rather than an error, and the difference is immaterial to the comparison with TWZ's $143bn.
The missingprofits.world statement ('close to 40% ... more than $200 billion, or 10%') misstates the published 2015 figures of 36%, $181bn and 9% (part of I14). — The website's figures refer to 2019 ('close to $1 trillion in 2019', from Wier & Zucman 2022), not to the REStud 2015 estimates. Only the rounding of an approximately 36-37% share to 'close to 40%' remains, recorded under F19 for the working papers.
The EU-haven DI upgrade is about ten times the FATS gaps shown in the authors' Table A.8 (part of I2). — Retained in substance (F2) but the ratio is qualified. The upgrade is about 10x the gross shortfall in the Netherlands, Luxembourg and Belgium ($12.5bn). If one accepts TWZ's R&D-depreciation explanation for Ireland's larger FATS figure (RG Sec. A.4.1), a net-of-depreciation Irish gap of about $12-14bn is possible, making the ratio about 5x. The core inconsistency survives either way.
The Replication Guide gives capital-intensity effects of +8%/-4% while the paper gives +7%/-5% (part of I13). — A rounding difference. Table 2 gives +7.6% and -5.0%, so neither text is wrong in substance; not an inconsistency of any consequence.
The $310bn full-removal figure is only a fixed-input diagnostic, because for non-OECD havens some wages and local quantities are imputed from the foreign-profit data, so a complete no-upgrade rerun could differ (I1 limits). — In the workbook, non-OECD foreign wages are either 40% of national-accounts compensation (Table A.4 B97) or derived from DI-based operating-unit profits (Table A.6 col. 5), neither of which depends on the unrecorded-profit additions. With pi_l fixed at 0.48, non-OECD shifted profits equal corrected total profits minus 0.48 x total compensation, independently of the foreign/local split. The $310bn is therefore the exact face-value counterpart within TWZ's framework, not merely a diagnostic.
Severity 'moderate' for the robustness-calculation slips (I5), the routing-table rows (I6) and the pre-1995 splice (I4). — Not rejected as issues (all confirmed and re-run) but downgraded to minor (F13, F14, F16). Each moves a robustness column, supporting table or trend by at most about 4% or 0.5pp, and none affects a stated conclusion.

Audit notes

What reproduced. Reproduced exactly from the authors' Zenodo workbooks (identical to the April 2022 author-site copies for every cell checked): - the baseline $616.46bn, the $1,703.23bn denominator (36.19%) and the 55.0% fall in haven profits; - every Table 2 column ($646.26bn, $506.22bn, $663.48bn, $585.89bn, $692.85bn, $507.16bn, $369.04bn, $739.87bn); - the Table 3 aggregates (EU $216.2bn = 1.49% of GDP; tax losses EU 18.2%, US 14.1%) and the US/other split (54.3% vs 27.3%); - the Fig. 8 series (+2.98pp; 2015 level 4.46pp) and the R&D regression (2014 coefficient 2.108, N = 2,556). Not reproduced as published once errors are corrected or consistency is imposed (adjudicator scripts): - face-value scenario $310.1bn (vs $507.2bn) and lower bound $169-209bn (vs $369bn); - European capital-share rise +1.0-1.5pp (vs +3.0pp); - BEA-consistent US shifting at most $225-259bn (vs $305bn); - R&D coefficient with the described controls 1.788; - exact-CES capital-intensity scenarios $641.9bn/$597.0bn and joint Paul-Isaka benchmark $481.4bn; - haven-by-haven source allocation: EU $185.5bn (vs $216.2bn). Primary-data checks: - BEA revised 2015: US affiliates in havens earn 346% of wages vs 46-49% elsewhere; Ireland 822%; EU-haven US-affiliate profit-type return of $113.7bn vs TWZ-implied $196.9bn; Belgium/Netherlands effective tax rates 26%/16%; - Gutierrez-Piton original series: 1985-1994 corporate shares absent; - Puerto Rico Planning Board exports: $69.4bn of merchandise, $48.8bn pharmaceuticals. Not reproduced: the raw-data pipeline (Eurostat/OECD FATS, IMF and Eurostat BoP, CDIS, haven national accounts), Orbis Fig. 1 (17%), the Compustat haven-headquarter inputs, the Damgaard-Elkjaer ultimate-owner shares, and the confidential CbC estimates of comparison studies.

Limits of this audit. Blouin & Robinson (2025, JPubE 252:105512), the main published critique of TWZ's reconciliation, could not be read (ScienceDirect and SSRN returned HTTP 403). Only Blouin's 2025 NeSt conference slides were read, and they confirm the $306bn criticism and the reporting-conventions argument. The authors' raw-data pipeline (Eurostat/OECD FATS, IMF/Eurostat BoP, CDIS, haven national accounts) was not rebuilt; the checks start from the authors' compiled workbooks plus BEA revised 2015 tables and the Gutierrez-Piton series. BEA-based bounds depend on the coverage of havens suppressed in BEA aggregates and on how minority-owned affiliates and pass-through entities are classified. The Gutierrez-Piton corporate value-added base is not public and was bracketed (100%/75%/60% of official corporate GVA). The alternative reconciliation scenarios (F2, F3) are sensitivities illustrating dependence on contestable conventions, not preferred estimates; TWZ's global-balance argument could restore part of any reduction by reallocation to non-reporting havens. Confidential CbC microdata, Orbis and Compustat inputs, and the Damgaard-Elkjaer ultimate-owner data were not available. Comparison-literature figures refer to the specific versions cited and differ in coverage, year and denominator.