Thomas Wright, Gabriel Zucman · NBER Working Paper 24983, September 2018

The Exorbitant Tax Privilege

16 problems found: 1 major · 4 moderate · 11 minor

What the paper claims

In the oil sector, taxes paid to oil-producing states have been 'contained', allowing US firms high after-tax returns; foreign taxes fell abruptly after the first Gulf War (a possible return on US military protection).
In sectors other than oil, the effective foreign tax rate of US multinationals has fallen by (almost) half since the late 1990s.
Almost half of this decline owes to the rise of profit shifting to tax havens.
The low foreign taxes paid by US multinationals can explain half of the US cross-border (direct-investment) return differential.
In 2015 about half of non-oil foreign profits are booked in havens at a ~7% effective rate and half in non-havens at ~27%; the haven share rose from ~20% (early 1990s) to ~50% because reported profitability, not factor use, boomed.
After the 2017 Tax Cuts and Jobs Act the total (foreign + US) tax rate on non-oil foreign profits can be computed for the first time; it fell from ~35% in the early 1990s to ~20%.
US multinationals use tax havens more than other OECD multinationals and account for about half of global shifted profits.

Every problem found

1majorconfirmed · newly identified here · replication failure

The paper says almost half of the fall in the foreign taxes paid by US multinationals came from moving more profits into tax havens. Redoing the calculation the paper describes, with the authors' own data, gives roughly one-sixth to one-quarter. Lower tax rates in ordinary countries did most of the work. The paper's numbers appear to measure how much profit was sitting in havens in 2015, not how much more was moved there after 2000.

Technical detail and evidence

Why it matters, in detail. With ETR = s*tH + (1-s)*tN, the 2000 -> 2015 values are: s 30.1 -> 45.4%, tH 13.2 -> 6.9%, tN 40.5 -> 26.7%, ETR -14.5 pts. Holding haven absolute PTR per wage (per unit of net PP&E) at 2000 levels adds 3.8 (3.2) ETR pts with re-booking in non-havens and 2.6 (2.1) without. Holding relative haven/non-haven profitability fixed adds 1.9 (0.6). A 4-factor Shapley split (haven factor share, relative profitability, tH, tN) assigns 2.2 (0.7) pts to rising relative profitability, 1.4 (2.9) to relocation, 8.5 to tN and 2.4 to tH. The paper's '30 pts / 6 pts' equal the 2015 level of excess haven profit share: s - f_K = 30.7 pts x 20.2-pt gap = 6.2 (Aug-2018 data). That level was already 20.0 pts / 5.45 pts in 2000, so it measures the level of shifting, not its rise. 72% of the decline (10.5 pts) occurred in 2000-08, with s nearly flat. The implied contribution of rising shifting is 0.6-5.1 pts (central about 2-3 pts, 15-20%), against 6 (about 40%) claimed. 'Almost half' requires counting relocation and falling haven ETRs as shifting.

Evidence. Paper: Abstract; Sec. 1 pp. 1 and 3; Sec. 4.1 pp. 9-10. Workbooks: an adjudicator text search of every sheet in both vintages found no decomposition. The only related cells are the hard-coded Data!AS39 = 0.36 (2001) and Data!AS54 = 0.28 (2016). Series used: - Assets(MOFA) col. AR (haven non-oil PTR), AX (non-oil PTR), GG-GH (haven non-oil taxes), GO (non-oil taxes); - T.A11b/T.A12b col. B (net PP&E) and C (compensation); main Data cols AJ-AL. Values, live (Sept-2018) / archive (Aug-2018): - non-oil ETR 32.24% -> 17.75% / 17.56%; - haven profit share 30.1% -> 45.4% / 46.5%; haven PP&E share 10.2% -> 16.1% / 15.8%; - haven ETR 13.2% -> 6.9% / 6.7%; non-haven ETR 40.5% -> 26.7% / 27.0%. Midpoint shift-share (live): non-haven rates -8.5, haven rates -2.4, haven share -3.6 pts. Sub-periods: - 2000-08: -10.5 pts (non-haven rates -9.0, share -0.8, haven rates -0.7); - 2008-15: -4.0 pts (share -2.1, haven rates -1.6, non-haven rates -0.3). BEA's published Swiss 2000 PTR (-3,739; 2000r Table III.G1) raises the literal estimate to 4.3-5.1 pts. Scripts: final/scripts/extract_series.py, decompose.py, haven_rate_decomp.py. Outputs: final/out/decompose_live_2000_2015.txt, decompose_sensitivity.txt, decompose_archive_baseyears.txt, decompose_archive_beaswiss.txt, decompose_subperiods.txt, haven_rate_decomp.txt.

Why this severity. The measured decline itself is unaffected, and rising haven profitability did contribute. But under the calculation the paper describes, it explains roughly a sixth to a quarter of the decline, not 'almost half'. The headline attribution is therefore overstated by about 1.5-3 times on the authors' own data and rests on unreported choices. The qualitative message that havens contributed survives, so the issue is not critical.

2moderateconfirmed · previously known · overclaiming

The paper says low foreign taxes explain half of why US investments abroad earn more than foreign investments in the US. The arithmetic checks out. But only about one-eighth of the gap comes from US firms using tax havens to cut foreign taxes. The rest of the 'half' rests on an assumption about how much oil-producing countries could have taken, and on low US taxes rather than foreign ones. Since 1990, when the gap was largest, the share explained is about a third.

Technical detail and evidence

Why it matters, in detail. The 4.52-pt differential decomposes into: oil normalization -0.82 (18%), harmonization at the non-haven rate -0.59 (13%), and the US federal + state rate on non-oil profits -0.83 (18%), for 49.8% in total. The third term changes the yield concept (post-foreign-tax -> post-all-tax) on the asset side only. The first is a full-rent-appropriation scenario, not a tax-rate change. Sub-period shares: 75% (1966-89, gap 3.3) vs 36.5% (1990-2016, gap 5.6). The profit-shifting term is a stable 13% of the gap in both. Partial equilibrium with fixed capital; Auditor B's specification range is 41-52%.

Evidence. Paper: Abstract; p. 12 (oil 0.8; shifting 0.6; US rate 'further reduces ... by 0.6'); p. 13 ('half of the U.S. DI yield differential of 4.5 points'). Aug-2018 appendix workbook (inside WrightZucman2018.zip): - ' T.A8'!E64 = -0.00824, J64 = -0.00594, M64 = -0.00831; - ' T.A9'!K62 = 0.04519, K63 = 0.03298 (1966-89), K64 = 0.05605 (1990-2016). Adjudicator recomputation: final/scripts/returns_check.py -> out/returns_check.txt. Auditor B's variants: codex/additional_results.json. E. McGrattan, discussion slides (July 2018), 'Take-Aways': 'Need evidence of capital adjustment costs/frictions => Puzzle isn't resolved'.

Why this severity. The paper's own accounting counterfactual reproduces (about half), so the claim stands as arithmetic for a broadly defined 'tax privilege'. But the abstract's attribution to low foreign taxes is overstated: foreign profit shifting accounts for 13% (31% with the oil rent scenario). The share is about a third in the period when the differential is largest.

The authors' response. The NBER version states that the counterfactuals hold US firms' investment decisions fixed (p. 12) and describes all three components in Sec. 5; no response on the attribution point located.

3moderateconfirmed · newly identified here · overclaiming

The paper links the sharp fall in taxes paid by US oil companies abroad to the 1991 Gulf War, and suggests it may have been a reward for US military protection of oil states. The authors' own figures show the sudden fall happened outside OPEC, mostly before the war, while OPEC rates barely moved that year. About three-quarters of the long-run fall happened outside OPEC, so the Gulf-War story can explain only a small part of it.

Technical detail and evidence

Why it matters, in detail. OPEC/non-OPEC shift-share of the pooled oil ETR change (66.7% in 1982-89 -> 45.5% in 1991-2015): OPEC-rate -4.1 pts, non-OPEC-rate -16.1, composition -1.1. The 1989-90 break (60.3 -> 45.3) is entirely non-OPEC (54.1 -> 28.0), with OPEC up (69.5 -> 70.3) in a fiscal year that largely predates the war. A 'return on military protection' to OPEC/Gulf states can explain at most about 4 of the ~21-pt decline. The level claims (68.9% vs 44.6% annual means) are unaffected.

Evidence. Authors' series: Assets(MOFA) col. AU/AV (oil and OPEC PTR) and GK/GL (oil and OPEC taxes). Adjudicator: final/scripts/oil_check.py -> out/oil_check_live.txt (matches Auditor A's claude/out/oil_rates.txt). Paper: Abstract ('Foreign taxes fell abruptly after the first Gulf War'); p. 2 ('return on military protection'); p. 7 (OPEC 70% -> 43%); Fig. 2A label 'First gulf war'.

Why this severity. The descriptive averages (about 70% -> 45%) are correct. But the abstract's dating of the abrupt fall to the Gulf War, and the military-protection interpretation offered in the introduction, can account for at most about a fifth of the long-run decline. The paper does caveat that the evidence is not causal.

The authors' response. Sec. 3.1 says 'the available evidence does not prove that the first Gulf War had a causal impact', lists alternatives (mid-1980s incentives, tax competition, new producers, profit shifting) and notes the early non-OPEC (UK) fall; no later response located.

4moderateconfirmed · previously known · methodology

From 1994 the paper switches to a definition of oil-company profits that no longer subtracts the cost of using up oil reserves. Profits therefore look bigger, and tax rates lower, after 1994: by about 2-5 points in normal years. This makes the fall in oil tax rates look about an eighth bigger than it is, and the paper does not mention it.

Technical detail and evidence

Why it matters, in detail. Splicing NI-based (net of depletion) profits before 1994 with PTR (gross of depletion) after biases post-1994 oil ETRs down by 2-5 pts in typical years, and by far more in low-price years (2015 O&G 47.7% vs 70.4%). Re-basing raises the 1991-2015 mean from 44.6% to about 47.6% (Auditor A's approximation), cutting the measured fall from 24.4 to about 21 pts. The bias also inflates post-1994 oil after-tax returns used in Fig. 1B and in the Sec. 5 oil normalization (likely a small effect on the 0.82-pt average). The non-oil ETR moves by less than 0.3 pt.

Evidence. Formulas: Assets(MOFA) rows 27-38 (formula-based) vs rows 39+ (hard-coded PTR). BEA 2000r Tab3E2/Tab3G2: final/out/bea2000_check.txt. BEA 2015r II.D2 (NI 1,498; taxes 10,854; equity income 5,223; capital gains -8,278), II.F2 (PTR 22,745; CCA 27,947), II.J2 (depreciation and depletion 35,284), all for oil and gas extraction. Auditor A: claude/out/oil_rates.txt and bea_panel_compare.txt. Blouin & Robinson (2020), Sec. 4.2, quote BEA: PTR is net income 'before income taxes or depletion charges'.

Why this severity. The break overstates the oil-tax decline by about an eighth (about 3 of about 24 pts) and is not disclosed. The qualitative fall survives.

5moderateconfirmed · newly identified here · overclaiming

The paper says the 2017 US tax law lets it measure, for the first time, the full tax US companies pay on foreign profits: about 20%. In fact it adds only the one-off 2017 repatriation tax to foreign taxes. It leaves out the ordinary US taxes companies paid when they brought profits home before 2018. The 'total' is therefore incomplete. The missing amount has not been measured, but it would have to be large to overturn the finding that the overall rate fell substantially.

Technical detail and evidence

Why it matters, in detail. Total-rate proxy = foreign ETR + 10% x BoP haven retained earnings/DI equity income x (1-ETR), giving 21.8% in 2015 (20.8% with fn. 7's 7.6%; 21.2% with the AH variant). Omitted, biasing the total down: pre-2018 residual US tax on repatriations and Subpart F (net of credits), and transition tax on non-haven E&P. Ambiguous direction: treating flows of immediate-counterpart haven retained earnings as the transition-tax base. The size of the omitted ordinary US tax was not estimated by either auditor or by the adjudicator. The '17% + 3-4%' decomposition should be read as foreign tax plus transition tax only.

Evidence. Main workbook formulas: Data!AA53 = Z53*0.1; Z53 = MAX(X53*(1-U53),0); AB53 = U53+AA53 = 0.2179; AH53 = (U53*AE53+AG53)/AE53 = 0.2120; AG1 = 0.076. Chart5 plots AB (per Auditor B). App. Table A.19: M6 = 3.5, M7 = 3.81, M12 = 7.63%. Paper: Sec. 1 p. 3; Sec. 2.1 pp. 4-5; Sec. 3.2 p. 9 and fn. 7. Auditor B: IRS 2014 CFC Table 2 shows $96.1bn of dividends to US parents, $81.2bn of Subpart F income and $306.4bn of previously taxed E&P (not convertible mechanically into tax).

Why this severity. This affects a secondary result. The omitted US taxes bias the 'total' rate down and the 10% rate biases it up by about 1 pt in 2015. The net size is not quantified, but it is unlikely to overturn the large fall from the early-1990s level. The claim of a comprehensive 'total' rate is nonetheless not met.

6minorconfirmed · newly identified here · arithmetic or transcription

One number in the text (0.6) does not match the authors' spreadsheet (0.83). The correct number makes the paper's 'half' add up better, so this is a typo-level problem.

Technical detail and evidence

Why it matters, in detail. Text 0.6 vs workbook 0.83 pts (1966-2016); the 0.6 matches the 1990-2016 average (0.56). Total explained 2.25/4.52 = 49.8%.

Evidence. WP p. 12. Aug-2018 workbook: ' T.A8'!M64 = -0.008307 (1966-2016) and M66 = -0.005644 (1990-2016). Recomputed: final/out/returns_check.txt.

Why this severity. A transcription slip. Correcting it makes the 'half' arithmetic more, not less, consistent.

7minorconfirmed · newly identified here · data error

The authors changed one 2000 figure for Switzerland from negative to positive, believing the official number was a mistake. The official tables are internally consistent. The change does not affect the headline tax rates, and it slightly reduces, rather than inflates, the role the paper gives to tax havens.

Technical detail and evidence

Why it matters, in detail. With BEA's published value, the 2000 haven share is 24.4% (not 30.1%) and the 2000-15 rise in the haven share is 21 pts (not 15). The non-haven-rate term falls from 8.5 to 6.8-7.0 pts. The shifting terms rise: literal counterfactual 4.3-5.1 pts, Shapley relative-profitability term 1.9-3.3 pts. So the decomposition shares in F1 move by about 1.5-2 pts depending on this single cell.

Evidence. BEA USDIA 2000 revised Tab3G1 and Tab3E1 (final/out/bea2000_check.txt); Assets(MOFA)!AG45, AG46 and column note AG4; final/scripts/decompose.py --swiss=asis|bea|consistent (out/decompose_live_2000_2015.txt).

Why this severity. The headline 32% -> 17.7% is unaffected, and the edit works against the authors' own shifting attribution. It is still an unjustified, asymmetric change to the decomposition's base year.

The authors' response. Workbook note only: '2000 and 2001 values are negative, likely to be a mistake in the tabulations; I fix here' (Assets(MOFA)!AG4).

8minorconfirmed · previously known · speculative assumption

The paper calls havens' huge profits 'shifted' because havens employ few workers and little machinery. Measured against the money and patents held there instead, havens are no more profitable than other countries. That is partly because that money itself comes from shifted profits, so this caveat matters for interpretation more than for the numbers.

Technical detail and evidence

Why it matters, in detail. Excess profitability is measured against wages and net PP&E, which omit intangible and financial capital. On owner's equity, haven and non-haven pre-tax returns are similar (2015: 8.9% vs 8.2%). If part of the haven IP or financial capital is a genuine factor, 'shifted' profits are overstated. Because haven equity is accumulated out of shifted earnings, the direction of the bias on the trend is unclear.

Evidence. App. workbook T.A5d col. I; T.A16 cols E-F and notes P1-P7; T.A11b/T.A12b. final/out/decompose_live_2000_2015.txt (relative profitability R). Corpus: offshore-havens-10 (Dharmapala 2019, AEA P&P 109:488-492) and offshore-havens-14 (Hodge 2018, Tax Foundation).

Why this severity. This is an interpretive caveat. It does not change any measured tax rate, and the benchmark the authors chose is defensible; it bears on how much of the haven profit share should be called 'shifted'.

The authors' response. The paper counts the booking of intangibles in low-tax affiliates as one of the two shifting channels (p. 10); Torslov, Wier & Zucman (REStud 2023) defend wage/tangible benchmarks against these critiques.

9minorconfirmed · newly identified here · internal inconsistency

A few points in two charts were smoothed or taken from a different year without saying so on the chart. Using the actual values would not weaken the paper's message; if anything it would make the pattern look stronger.

Technical detail and evidence

Why it matters, in detail. Undisclosed (Fig. 3B) or weakly disclosed (Fig. 2B) substitutions temper outliers whose small or volatile denominators produce extreme ETRs. Restoring 2015 values (France 72.5%, Brazil 95.8%) strengthens the negative profitability-tax association, and restoring 2001 (37.8%) enlarges the decline from 2001.

Evidence. Main workbook Data!U39 = 0.31236; App. T.A2!L45 '=(L44+L46)/2' and note A63 (both vintages); DataF3bc V14 = U14, V19 = T19, V20 = U20; BEA 2015 preliminary II.D.1/II.F.1: France 3,497/4,826, Brazil 1,382/1,443, Canada 5,636/32,722 (adjudicator check).

Why this severity. Presentation only. The actual values would make the high-tax/low-profitability pattern more extreme and the post-2000 fall larger, so no conclusion depends on the substitutions.

10minorconfirmed · newly identified here · version instability

The spreadsheets on the authors' website contain errors that stop readers recomputing one of the paper's results. An older version in the authors' download bundle reproduces the numbers exactly, so this is a housekeeping problem, not a flaw in the results.

Technical detail and evidence

Why it matters, in detail. The live workbook cannot compute T.A8/T.A9 averages; the archive reproduces 49.8% (49.6% excluding the imputed 2016). The 2015 BEA revision moves the haven share by about 1 pt and the non-oil ETR by 0.19 pt.

Evidence. Website WrightZucman2018Appendix.xlsx: ' T.A8'!E64/J64/M64 = #VALUE!, T.A5b!J59/K59 = #VALUE!, Assets(MOFA)!KD60 = #VALUE!; archive values as in F2/F6; main Data!AL53 and AC53 in both vintages; App. PDF p. 10.

Why this severity. Reproducibility maintenance only: the authors' own archived files reproduce every Sec. 5 number exactly, and the revisions are small.

11minorconfirmed · newly identified here · documentation gap

Some early-year figures are filled in with rough adjustment factors, and a few changes in the underlying data are not described in the paper. Checks suggest these choices barely affect the results, but readers can find them only by digging through the spreadsheets.

Technical detail and evidence

Why it matters, in detail. Interpolations are transparent in the formulas, and the 1966-90 oil mean is robust to them (66.3-71.4% per Auditor A's sensitivity). The undocumented 2009 bank inclusion biases post-2009 ETRs up. These are documentation, not substance, issues.

Evidence. Assets(MOFA) col. GK/GM rows 11-27 (formulas checked by adjudicator); Auditor A's references for BoP scalings (Assets(BoP) AV67, AV71) and the bank coverage change; adjudicator text search of all workbook sheets (scripts/grep_workbook.py) for the decomposition.

Why this severity. Per Auditor A, the oil levels are robust to the scaling factors (1966-90 mean 66-71%) and the coverage change is conservative. The missing decomposition is graded under F1.

12minorconfirmed · previously known · selective presentation

The paper says no other rich country books as large a share of its foreign profits in tax havens as the US. In the authors' own data, Hungary and Belgium do, though they act as havens themselves. The broader point that US companies use havens more than most others still holds.

Technical detail and evidence

Why it matters, in detail. The US is not the maximum in the sample unless Belgium and Hungary are reclassified as havens ad hoc. The aggregation choices move the European benchmark between 21.3% and 23.6%. All shares are on an immediate-counterpart DII basis (conduit bias disclosed), not the PTR basis used elsewhere in the paper.

Evidence. App. workbook 'outw share' K10 = 0.550 (Belgium), K21 = 0.605 (Hungary), K43 = 0.475 (US), K48 = 0.219 (Europe; D46/E46 include Hungary), E26 sums outw!F275:F286 (Japan: LU, NL, SG, CH only); T.B6 notes; WP p. 11; App. pp. 19-20.

Why this severity. The US vs typical-OECD contrast (about 47.5% vs about 21-24%) survives; only the 'no other country' wording and the aggregation details are inaccurate.

The authors' response. App. B.2.1 notes that Belgium and Hungary have far larger shares and interprets them as conduits ('we might consider as havens themselves').

13minorconfirmed · newly identified here · misrepresentation of source

The paper credits a Danish central-bank study with showing that half of tax-haven profits belong to US firms. That study measures investment stocks, not profits. The 'half' actually comes from a related paper by Zucman and coauthors that used those stocks to estimate profits; the number may be reasonable, but its source is described inaccurately.

Technical detail and evidence

Why it matters, in detail. Converting ultimate-owner position shares into income shares assumes haven income is distributed across owners like positions. The 'half' is therefore model-based (TWZ), not an observed income statistic, and the EU share is 30% in the cited source, not 25%.

Evidence. D&E (2017) abstract and Secs. IV-V, Annex II ('FDI positions are broken down by the UIE'); TWZ (2018) NBER w24701 Sec. 4.3 and p. 31; WP p. 11.

Why this severity. A citation and description error on a secondary claim. The figure exists (in TWZ) and rests on a stated allocation assumption.

14minorconfirmed · newly identified here · overclaiming

An appendix chart linking countries' tax rates to their investment returns depends almost entirely on two countries, the US and Ireland. Without them the relationship disappears, so it cannot be read as evidence that taxes cause the pattern.

Technical detail and evidence

Why it matters, in detail. With 26 country clusters and marginal significance driven by two countries, the cross-section cannot identify a causal tax elasticity of profit shifting.

Evidence. Authors' Stata dataset and .do file (WrightZucman2018Programs.zip); final/scripts/appB_regs.py -> out/appB_regs.txt (t(G-1) p-values, Stata small-sample correction); App. B.1.3 pp. 17-18.

Why this severity. This is an appendix result that the main text does not rely on; the US and Ireland are, however, the cases of interest.

15minorconfirmed · newly identified here · arithmetic or transcription

Several descriptive statements in the text don't quite match the authors' own numbers: some ranges are off, a few figures are rounded generously, and one country count is wrong. None of this changes the conclusions.

Technical detail and evidence

Why it matters, in detail. These are descriptive overstatements of profitability ranges and shares that do not enter any computed result. The largest, the claim that main non-haven countries have profit/wage ratios below 35%, understates non-haven profitability and so slightly exaggerates the haven contrast.

Evidence. Main workbook Data cols M, P; DataF3bc col. E; App. T.A1 col. M; T.A2 K68/L68; Data!AS39/AS54; BEA 2015p II.D.1/II.F.1 by country (adjudicator check); WP pp. 1, 6-10; App. pp. 5-6.

Why this severity. Loose wording and rounding; no conclusion depends on these statements.

16minorplausible · newly identified here · methodology

Oil-producing countries collect money from oil companies in several ways besides income tax, such as royalties and production-sharing deals. The paper counts only income tax, so its 'tax rate' does not fully measure how oil profits are split. That matters for the paper's story about oil states, not for its figures on what US oil firms kept after tax.

Technical detail and evidence

Why it matters, in detail. Government take is broader than income taxes: royalties and production-sharing arrangements are excluded, and downstream affiliates dilute the aggregate. Adding royalties raises the take by about 4-5 pts in the 2009 and 2014 checks. Shifts in the legal form of take can move the measured income-tax rate without any change in rent shares; the effect on the 70 -> 45 trend has not been quantified.

Evidence. Auditor A's BEA panel (claude/out/bea_panel_compare.txt; BEA II.J2 royalties), not re-verified by the adjudicator; WP p. 7 ('The taxes levied by oil States determine how natural resource rents are split').

Why this severity. This qualifies the rent-sharing reading of the oil series, not the after-tax returns that drive C4. The level effect is about 4-5 pts in two checked years; there is no evidence on the trend.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

Figure 2B's 2001 substitution is undisclosed, and Appendix Table A.2 (T.A2!L45) and BEA give 37.81% for 2001. — T.A2!L45 itself holds the 2000-2002 average (31.24%, formula =(L44+L46)/2). The Table A.2 note states 'In col. 11, the tax rate for 2001 is the average of 2000 and 2002' in both workbook vintages; 37.81% is the raw ratio F45/G45. Only the Figure 2 notes omit the smoothing, so the residual point is kept as part of minor issue F10.
The appendix's 'Netherlands ... effective rate was 12%' (2015) contradicts the data (16.0%). — The appendix used BEA's preliminary 2015 release, in which the Netherlands is 3,794/32,814 = 11.6% (adjudicator check of BEA 2015p II.D.1/II.F.1). The 16.0% is BEA's later revision, not an error by the authors.
'Oil accounted for more than a third of pre-tax profits on average 1966-2016' is contradicted (32.8%). — In the vintage the paper used (Aug-2018 workbook; data through 2015) the mean annual oil share is 33.4%. The 32.8% arises only after the 2016 observation was added in the September update.
Corpus lead offshore-havens-08 (Blouin & Robinson 2020/2025): BEA double counting and misattribution of affiliate income invalidate WZ's haven shares and tax rates. Both auditors examined and dismissed this; recorded here for completeness. — WZ use BEA profit-type return from 1994 and net income + taxes - equity income - capital gains before, both of which exclude equity income. BR themselves find PTR and their Adj. PTI 'very similar' in aggregate, with haven percentages 'not that different' except 2009 (BR 2020, Sec. 4.2, Fig. 5). BR's Puerto Rico point (fn. 26) concerns Saez & Zucman (2019), not WZ. Their question about pre-1994 PTR (fn. 25) is answered by the workbook formulas, though see F4 for the depletion consequence.
BR's charge (fn. 26) that excluding the oil sector 'masks' the earlier overstatement applies to WZ's core results. — WZ analyse oil separately as a core result (Sec. 3.1, Fig. 2A) and motivate the exclusion explicitly. The charge concerns the reconciliation with Zucman (2014, 2015) figures, not the validity of WZ's own series.

Audit notes

What reproduced. Reproduced exactly: - the non-oil foreign ETR (32.24% in 2000; 17.56% preliminary / 17.75% revised in 2015) from BEA primary tables; - 2015 haven/non-haven shares and ETRs; the oil averages (68.9% / 44.6%); - Ireland's post-1996 jump; the 7.63% transition-tax rate; - the Sec. 5 accounting from the Aug-2018 workbook (4.52-pt differential; 0.82/0.59/0.83 = 49.8%); - the Appendix B regression; the Fig. 4 shares (47.5% / 21.9% / 14.5%) from the authors' OECD extract. Not reproduced: - the Sec. 4.1 '1 + 8 + 6' decomposition and the '30 points / 6 points' counterfactual. The stated counterfactual gives +16-20 pts / +3.2-4.1 ETR pts with re-booking and +10-14 / +2.1-2.8 without; the paper's figures match the 2015 level of excess haven profit (30.7 pts x 20.2 = 6.2); - the text's 0.6-pt US-rate adjustment (the workbook gives 0.83); - the Sec. 5 1966-2016 averages from the website workbook (#VALUE! errors). Not attempted: pre-1982 IRS inputs and 1982-1998 BEA annual tables (internal consistency only), the January-2018 BoP/IIP vintage, and raw OECD/IMF downloads.

Limits of this audit. Not obtained or not re-verified: - the raw-data Dropbox archive (about 1.2 GB) was not downloaded; - pre-1982 IRS inputs and 1982-1998 BEA annual figures were checked only for internal consistency; - BoP/IIP inputs to Sec. 5, the OECD FDI-income records and the IMF BOPS data were taken from the authors' workbook extracts. Inferences: - that the paper's '30 points / 6 points' are the 2015 level effect rests on a close numerical match, not on author confirmation; - the range of profit-shifting contributions (0.6-5.1 pts) covers accounting decompositions, not causal estimates. Not quantified: the size of the ordinary US residual tax omitted from the 'total' rate (F5). Royalty and downstream figures in F9 and some cell references in F12 are Auditor A's and were not re-verified. Other: the published 2025 Blouin-Robinson article could not be read (publisher page blocked), so the 2020 TPC version was used; the McGrattan discussion was read from Auditor A's cached copy. The Gulf-War causal question cannot be tested with these data. No author contact was made and no later version of the paper exists.