Emmanuel Saez · Journal of Economic Perspectives 38(2): 227-244

Gabriel Zucman: Winner of the 2023 Clark Medal

16 problems found: 5 moderate · 11 minor

What the paper claims

About 8% of global household financial wealth, about 10% of world GDP, is held in tax havens [3]; offshore wealth/GDP ranges from a few percent in Scandinavia to ~15% in Continental Europe and 'as much as 60%' in Russia, Gulf countries and some Latin American countries [8].
Hidden wealth is extremely concentrated (Scandinavian leaks and amnesties) and it is 'plausible' to apply this distribution to other countries; adding offshore wealth raises top-0.01% wealth shares (2000-09 averages) [8], [9].
Evasion is much higher at the top than random audits show: the top 0.01% of Scandinavians evade ~25% of taxes (audits: <5%) [9]; in the US, unreported income rises from 7% of true income (bottom 50%) to >20% (top 1%) [10], so true income is more concentrated than reported income.
US multinationals book more than half of their foreign profits in tax havens ('by the 2010s' per [11]; 'now' per Figure 3), but only ~20% of foreign tangible capital and ~10% of foreign wages are there.
Globally 36% of multinationals' foreign profits end up in tax havens; US multinationals shift almost twice as much as others; haven affiliates are an order of magnitude more profitable than local firms [12].
The US top-1% wealth share rose from ~25% around 1980 to almost 40%; the Fed Distributional Financial Accounts ('with small adjustments') show the same, and Smith, Zidar & Zwick (2023) find 'remarkably similar results' [14], [16].
The Piketty-Saez top-1% share of reported (fiscal) income more than doubled since the 1970s and hit a record in 2021; reported income is about two-thirds of national income, its share having fallen from 70% to ~60% [16].
In distributional national accounts the top-1% pre-tax national income share doubled from ~10% to ~20% since the 1970s; the effect of missing income is modest, and offsetting the rise would require 'extreme, and hence highly unrealistic, equalization of unreported income' [19], [20].
Growth was evenly shared 1946-80; since 1980 the bottom 50% saw hardly any real pre-tax gains and barely rising disposable cash income, while the top 1% saw 'astonishing' growth; COVID-era transfers were temporary [16], [21].
Zucman's work shaped policy: automatic information exchange (FATCA 2010; CRS, >100 countries), the 15% global minimum tax (>130 countries, Oct 2021), and US wealth-tax proposals whose budget effects he helped estimate [24].

Every problem found

1moderateconfirmed · previously known · version instability

The essay quotes a 2021 draft saying the richest 1% fail to report over 20% of their income, against 7% for the bottom half. Before the essay was published the same authors revised this to roughly 16% against 10%. The rich still under-report more, but the gap is about half as large as the essay says.

Technical detail and evidence

Why it matters, in detail. The 2021 benchmark applied DCE multipliers return by return, which re-ranks and concentrates undetected under-reporting at the top; the Dec-2023 MA-DCE allocates it in proportion to exam-corrected income by income type. Bottom-half rate ~7% -> ~10-10.5%; top-1% rate 20-25% -> ~16-18%; top-1% fiscal share correction +1.5 pp -> +1.1 pp. The sophisticated-evasion component (~6 pp at the top) is unchanged and the top 1% still earn about one-third of unreported income, so the direction holds; the essay's 'dramatically shifts the picture' fits the 2021 numbers better than the revised ones.

Evidence. w28542 (Mar 2021): abstract; introduction ('from about 7% ... to 21% in the top 1%'); Sec. 6 ('hovers between 20 and 25% in most of the top 1%'); top-1% share 20.3% -> 21.8%. w28542 (cover: 'March 2021, Revised December 2023'): abstract (top-1% unreported income +50%, top-1% share +~1 pp); introduction ('rises from about 10% in the bottom 90% ... to 16% in the top 1%. Out of this 16%, 6 percentage points correspond to sophisticated evasion'; 20.3% -> 21.4%); Sec. 6.2. Online appendix Table A8 (2008-13 averages, recomputed by the adjudicator): 10.5% of true income unreported for P0-90, 18.0% for the top 1%. Files: /mnt/zdata/downloads/glrrz-versions/w28542_rev0_Mar2021_nber.txt, w28542_rev_Nov2023_nber.txt; glrrz-audit-primary/GLRRZ_Nov2023_Appendix.pdf.

Why this severity. The quoted gradient (about 14 pp, a three-fold rise) is roughly halved in the authors' own pre-publication revision (about 6-8 pp, ~1.6-1.8-fold) and the effect on the top-1% share shrinks by about a quarter; but the conclusion the essay draws from it - true income is more concentrated than reported income because audits miss sophisticated evasion - survives.

The authors' response. The GLRRZ revision itself (Dec 2023) adopts the lower figures and argues that, although the magnitude is uncertain, a decrease in the top share is 'difficult to rationalize'. Reck, Risch & Zucman (Oct 2021) replied to Auten & Splinter. No response by the essay's author located.

2moderateconfirmed · previously known · misrepresentation of source

The chart of US companies' profits in tax havens credits the wrong source, and its last point (57%) is a projection drawn as if it were data. Figures released later, including by Zucman himself, show the share stayed around 50% rather than climbing. The chart's main lesson - profits moved to tax havens far more than jobs or factories - still holds.

Technical detail and evidence

Why it matters, in detail. Non-oil MOFA profit-type return in havens: 45.4% (2015), 49.5% (2016) in [13]; with Puerto Rico (2016 CbC: $38.9bn profits, $16.4bn tangible assets) 53.8% in 2016 (observed) - the 1966-2015 uplift is a constant-ratio back-cast. The 2017 point (57.2%) is a geometric extrapolation (+3.4 pp); the observed 2017 PTR share fell to 45.8% excl. PR (50.1% with the 2016 PR ratio). Capital (17-19%) and wage (10%) shares are essentially unaffected (see F7).

Evidence. Essay p. 236 and Fig. 3 (p. 237). SZ2019MainFigures.xlsx DataF4b: B1 'Excluding Puerto Rico: BEA data from Wright-Zucman (2018)'; N1 'Including Puerto Rico ... extrapolating backwards'; P5:P54 '=P$55' (1.1875); U56 '=U55*((U55/U49)^(1/6))' = 0.5720; V56 = 0.1927; L56 '=L55'. WrightZucman2018.xlsx Data AL53/AL54 = 45.4%/49.5% (2015/2016); Wright & Zucman (2018) p. 10 ('almost half of their pre-tax profits are booked there'). TOI Fig. 4.2 note ('almost 60% today'). GBJZ replication data (gbjz-tcja/repo/data/combined_dataset.tsv, BEA_return, their haven list): 2016 49.0%, 2017 45.8%, 2018 47.8%, 2019 45.7%, 2020 50.8%, 2021 45.7%; GBJZ text: 'in 2017 ... 46% of the profit-type return of majority-owned affiliates was in tax havens'. Scripts: scripts/haven_profit_share.py, scripts/gbjz_haven_share.py.

Why this severity. The figure's message - profits moved to havens far more than capital or workers - is unaffected, but its end-point overstates the haven profit share by about 7 pp relative to later observed data, the wrong attribution stops readers checking it, and the text's 'more than half ... now' depends on the undisclosed Puerto Rico construction; 'about half' is what the authors' own later data show.

The authors' response. None located. The TOI note refers readers to taxjusticenow.org. Puerto Rico is plausibly a low-tax location for US firms and is absent from BEA outward data, so some adjustment is defensible.

3moderateconfirmed · previously known · selective presentation

The essay says it is reasonable to assume hidden wealth everywhere is as concentrated among the ultra-rich as in Scandinavia, citing one supporting study. A large Dutch study, which thanks Saez and Zucman for comments, found hidden wealth there was much less concentrated. If other countries look more like the Netherlands, the essay's estimates of how much hidden wealth adds to the richest Britons' share would be far too big.

Technical detail and evidence

Why it matters, in detail. The corrected top share is s/(1+h) + h*q, where q is the top group's share of hidden wealth; with h ~3.4% of UK household wealth offshore the increment is nearly linear in q, so q = 0.07 instead of 0.516 cuts the UK increment from +1.68 pp to +0.15 pp. Leenders et al. note the Dutch super-rich may evade through trusts and foundations or emigrate, so 7% is not a validated replacement either; the point is that the transfer is contested and its sensitivity is large.

Evidence. Essay pp. 234-235. Leenders et al. (2023) abstract; introduction pp. 1-2; Sec. 4 p. 7 ('the top 0.1% increases from 8.7% to 9.3%, less than half the adjustment for Norway and below a fifth of that in Colombia'); acknowledgments. AJZ2017bData.xlsx T.A9b col I (0.5156, every year). Adjudicator sensitivity (scripts/ajz2018_corrections.py): replacing 51.6% by 7% for the UK, holding aggregates and ranks fixed, gives a 2000-09 top-0.01% share of 2.88% instead of 4.41% (increment +0.15 pp vs +1.68 pp); the US increment falls from +0.71 pp to about zero.

Why this severity. It concerns secondary results (the country corrections), which could shrink by an order of magnitude under a Dutch-like distribution, and the essay presents the transfer as supported while omitting the one directly contrary peer-reviewed test known to its author; the Scandinavian finding itself and the Colombian corroboration stand, and amnesty samples may miss the most sophisticated evaders, so the true bias is unknown.

The authors' response. AJZ (2019, p. 2075) state they 'certainly do not claim' the Scandinavian estimates hold everywhere and argue top-end evasion may be even higher elsewhere. No reply to Leenders et al. located.

4moderateconfirmed · previously known · selective presentation

The essay says the Federal Reserve's own wealth data confirm that the richest 1% of Americans hold almost 40% of all wealth, after 'small adjustments'. Those adjustments add about 6 percentage points, and on its own terms the Fed put the figure at about 30% when the essay was written. The Fed data do confirm a big rise since 1989; the 'almost 40%' level depends on the authors' definitions.

Technical detail and evidence

Why it matters, in detail. Level, 2018: Fed concept/households 30.6% (Jan-2024 vintage) vs 38.1% plotted; harmonized with [16]'s rules ~36% (approximate). The harmonization choices (durables are not SNA assets; unfunded DB pensions are claims on governments; tax units via a Pareto factor) are documented in [16] and defensible, but they raise the level by ~18% and add ~1 pp to the 1989-2018 rise. For comparison SZZ (2023) give 33.7% for 2016 in equal-split units vs 36.3% for SZ in the same unit.

Evidence. Essay pp. 237-238 and Fig. 4. SaezZucman2020JEPData.xlsx DataF1-F2(Wealth): O89 = 23.49%, O118 = 32.39% (raw DFA); AW89 = 25.89%, AW118 = 35.40% (excl. durables and unfunded DB); R89 = 28.10%, R118 = 38.14% (adjusted); I80 = 24.92%, I118 = 37.91% (SZ, tax units). Saez & Zucman (2020 JEP) fn. 3. Jan-2024 DFA: /mnt/zdata/downloads/saez2024-adjudicator/dfa_20240121.zip (sha256 ef29b89b...e13e); scripts/dfa_top1_vintages.py, scripts/dfa_excl_durables.py (durables exclusion 31.5%, x1.08 = 34.0% in 2018Q3; adding the 2020-vintage unfunded-DB effect of +2.2 pp gives ~36.4%). Batty et al. (2019), FEDS 2019-017, Sec. 4.1.

Why this severity. The Fed data do confirm a sharp rise (+7.7 pp from 1989 to 2018Q3 in the vintage current at writing), so the trend survives; but the level the essay presents as jointly confirmed rests on ~6 pp of author adjustments described as 'small' and on a superseded vintage, while the Fed's own published estimate was about 30%.

The authors' response. Saez & Zucman (2020 JEP, pp. 8-9 and fn. 3): durables are not assets in the SNA, unfunded pensions are promises not backed by wealth, tax units are the relevant unit, and with harmonized definitions both sources give 38% in 2018.

5moderatelikely · previously known · overclaiming

The essay says the share of national income going to the top 1% of Americans doubled since the 1970s. That is the authors' own estimate. Another published estimate, which the essay does not mention but which Saez had argued against only months earlier, finds a rise about half as large. Both agree inequality rose; the essay presents a disputed size as settled.

Technical detail and evidence

Why it matters, in detail. Top-1% pre-tax share, 1979-2019: PSZ +8.2 pp vs AS +4.4 pp (-47%); after tax: PSZ +6.6 pp vs AS +1.4 pp. The gap reflects (i) fiscal-income corrections (AS: about a third of the Piketty-Saez fiscal rise occurred around TRA86), (ii) individual vs equal-split units, (iii) audit-based vs proportional allocation of under-reported business income (~2 pp of the 6 pp 2014 gap), (iv) retirement income (~1 pp; PSZ corrected a rollover error), (v) other items. Gale, Sabelhaus & Thorpe (2023) judge that 'almost all the choices AS have made tend to push top income share estimates down' but that magnitude and timing remain open. On the bottom 50%, AS find real pre-tax growth of more than one-third since 1979 vs +14.5% (1980-2019) in the Feb-2022 PSZ series.

Evidence. Essay pp. 239-240. AutenSplinter-IncomeIneq-2024.xlsx sheet T-A1: C24 = 9.407 (1979), C64 = 13.786 (2019) (AS pre-tax); K24 = 10.854, K64 = 19.078 (PSZ, update of 14 March 2023). PSZ (2019) WID WP 2019/01 Sec. II 'Comparison with Auten and Splinter (2018)' ('one needs extreme and hence unrealistic assumptions'). Piketty, Saez & Zucman, 'Comment on Auten and Splinter (2023)', 13 Dec 2023. Auten & Splinter, reply to PSZ (March 2025). Gale, Sabelhaus & Thorpe (2023), per corpus us-income-21. Identity check (scripts/offset_identity.py): on Piketty-Saez fiscal income, a full offset would need a negative top-1% share (-5.5%) of untaxed income in 2019, so the narrow claim holds, while AS's 2019 level corresponds to ~3% vs ~18% for PSZ.

Why this severity. The direction (a large rise) is not in dispute and the essay's narrow claim that a full offset would need implausible assumptions is arithmetically correct; but the magnitude it presents as settled is about twice the leading published alternative and readers are given no hint of that range. If Auten & Splinter were right the 1979-2019 rise would be ~45% smaller, which would be a major change; because the debate is unresolved, the demonstrated impact is an uncertainty band, not a correction.

The authors' response. PSZ (2019 [20]; Dec-2023 comment) and Saez & Zucman (2020) argue AS's results imply implausibly low concentration of untaxed income and rest on methodological mistakes; Auten & Splinter (JPE 2024; March-2025 reply) maintain their allocations and argue PSZ are outliers relative to BEA, CBO and Census-based series. Unresolved.

6minorconfirmed · previously known · internal inconsistency

The essay repeats an older figure that more than half of US firms' foreign profits were in tax havens by the early 2010s. That figure partly counts where holding companies sit rather than where profits are booked. Zucman's own better measure, shown in the essay's next chart, puts it at about 35-40% then, reaching about half only around 2016.

Technical detail and evidence

Why it matters, in detail. Direct-investment income assigns lower-tier affiliates' income to the (often haven) holding company; profit-type return by affiliate location avoids this. 2013: 55.8% (DII) vs 37.8% (PTR; 41.9% with PR). The PTR share was 45.7-50.8% (excl. PR) in 2017-2021.

Evidence. Zucman (2014 JEP) pp. 128-129; Zucman2014JEPdata.xlsx DataFig1-5 col CQ: 51.7% (2010), 55.8% (2013); WrightZucman2018.xlsx Data col AL: 33.4%, 37.8%, 45.4% (2015), 49.5% (2016); SZ2019MainFigures.xlsx DataF4b col U (with PR): 37.4%, 41.9%, 49.7%, 53.8%; GBJZ (IMF Econ Rev) Sec. 4 ('direct investment income series can be seen as an upper bound'; ~55% DII vs ~45% PTR in 2015-2020); Blouin & Robinson (2020) ~35% for 2013. scripts/haven_profit_share.py.

Why this severity. The sentence faithfully reports [11], the 'enormous rise' holds in every series, and the essay's own Figure 3 shows the better series; but the early-2010s level is overstated by 14-18 pp relative to the authors' preferred measure.

The authors' response. Zucman (2014, p. 129) caveated the balance-of-payments data; Torslov, Wier & Zucman (2023, Sec. 4.2.1) accept that BEA net income double counts but say profit-type return is unaffected; Clausing (2020) argues PTR-type measures can understate foreign-to-foreign shifting.

7minorconfirmed · newly identified here · arithmetic or transcription

A small spreadsheet slip makes the share of US companies' factories and equipment located in tax havens look about 0.3 percentage points bigger than it should. It changes nothing in the argument.

Technical detail and evidence

Why it matters, in detail. The overstatement is S*R/(D*(D+R)) ~0.30 pp in 2016 (18.63% vs 18.33%) and proportionally similar in back-cast years; the profits-versus-capital contrast is unaffected.

Evidence. SZ2019MainFigures.xlsx DataF4b: V55 '=S55/D55', S55 '=R55+H55', Q55 '=N55+(B55-C55)'; Puerto Rico values match IRS 2016 country-by-country Table 1A row 61 (checked by codex). scripts/haven_profit_share.py (last column).

Why this severity. A 0.3 pp slip in a ~19% share; 'about 20 percent' remains a fair rounding.

8minorconfirmed · previously known · selective presentation

The essay presents '36% of multinational profits go to tax havens' as a discovery. It is the authors' central estimate: their own range runs from about 22% to 43%, and some other researchers find 13-22%. Profit shifting is large either way, but the exact share is uncertain.

Technical detail and evidence

Why it matters, in detail. The 36% depends on the counterfactual (haven foreign affiliates' profit-to-wage ratio set to that of local firms) and on raising havens' inward data to partners' outward figures; TWZ's own bounds span roughly 22-43%, and CbC-based estimates with other counterfactuals are 13-22% (large MNEs, 2016-17).

Evidence. TWZ (REStud 2023) abstract, Table 1 ($616.5bn / $1,703.2bn), Table 2 cols 1, 8, 9 (616 / 369 / 740), text 'and 43% in our upper bound scenario'; Fuest et al. (2022, CESifo WP 9757) pp. 25-26 (EUR 133bn of EUR 1,048bn = 13%; EUR 229bn = 22%).

Why this severity. The preferred estimate is reported accurately and even the lower bounds imply large shifting; the omission affects how certain the number appears, not the conclusion.

The authors' response. TWZ (2023) report the sensitivity table and argue that firm-level data miss most haven profits.

9minorconfirmed · newly identified here · data error

The spreadsheet behind the country estimates has small flaws, including one year in which Russia's richest 0.01% would own a negative share of domestic wealth. Fixing them does not change the essay's conclusions.

Technical detail and evidence

Why it matters, in detail. The Russian residual (incl. share x (W+O) - q*O)/W turns negative when NPZ's top-0.01% share falls to 7.5% in 2009, showing the key can assign the Russian top more offshore wealth than they own in total. The UK/US formula overstates by s*h^2/(1+h) ~0.003 pp.

Evidence. AJZ2017bData.xlsx T.A8!G110 '=(T.A11!G110*(T.A6!G110+T.A9!T109)-T.A9b!I109*T.A9!T109)/T.A6!G110' = -0.0022 (W = $4,013bn, O = $703bn, incl. share 7.5%, q = 0.5156); T.A9!K107 '=T.A3!$K$93/(T.A3!$K$93+AVERAGE(T.A6!K107,T.A6!K108))'; T.A11!K101 '=T.A8!K101/(1+T.A9!K100*T.A9b!$B100)+T.A9!K100*T.A9b!$B100*T.A9b!$I100'. scripts/ajz2018_corrections.py; work/ajz2018_corrections.out.

Why this severity. None of the three changes a rounded figure in the essay: Russia still 'more than doubles' and the UK correction moves by 0.003 pp; they mainly show how fragile the transfer assumption is (F3).

10minorconfirmed · newly identified here · arithmetic or transcription

The essay says hidden wealth reaches as much as 60% of GDP in Russia and several Latin American countries, but its own chart shows Russia at 13% (or 47% on an alternative method) and only Venezuela above 50% in Latin America. The general point that some countries hold far more hidden wealth than others is right.

Technical detail and evidence

Why it matters, in detail. Russia 13.2% of GDP under the BIS-deposit allocation, 46.9% under net errors and omissions; one Latin American country (Venezuela) above 50%.

Evidence. AJZ2017bData.xlsx sheet DataF4 (scripts/ajz2018_fig1.py; world average 9.79%); AJZ (2018) abstract and Sec. 4.3.

Why this severity. A text-figure mismatch in a descriptive sentence; the heterogeneity point stands.

11minorconfirmed · newly identified here · internal inconsistency

The essay says income reported on tax returns is 'about two-thirds' of national income and, a few lines later, 'about 60 percent'. Both can be right depending on whether capital gains are counted, but the essay does not say so. It does not change the argument.

Technical detail and evidence

Why it matters, in detail. With and without capital gains the recent coverage is 66-68% vs ~61%; the 70% -> 60% decline is taken from [16] and is larger than the decline in Piketty-Saez totals over BEA national income (65.7% -> 61.0% excl. KG), possibly because of different totals; not resolved here.

Evidence. Essay pp. 239-240; Saez & Zucman (2020 JEP) p. 4. TabFig2024 Table A0 / FRED A032RC1A027NBEA (scripts/fiscal_vs_ni.py): excl. KG 65.7% (1976-79) -> 61.0% (2015-19); incl. KG 68.3% -> 65.6% (68.1% in 2018-22).

Why this severity. A small internal inconsistency with no effect on any result; the higher figure slightly flatters the 'modest effect' argument.

12minorconfirmed · previously known · misrepresentation of source

The essay says an earlier IRS audit study found that counting hidden income makes incomes slightly more unequal. The study actually found almost no change, and on its main measure slightly less inequality.

Technical detail and evidence

Why it matters, in detail. 2001 NRP: Gini of true AGI 0.5697 vs reported 0.5727; top 1% 17.8% vs 17.8%; top 5% 32.7% vs 32.2%; crossing Lorenz curves. 'Roughly distribution-neutral' is the accurate summary.

Evidence. Johns & Slemrod (2010, NTJ 63(3)) abstract, discussion of Table 5, Table 7; PSZ (2019) WID WP 2019/01 fn. 3.

Why this severity. A background characterization; it slightly overstates how far earlier audit studies already pointed the way the essay argues.

13minorconfirmed · newly identified here · other

Several references point to the wrong item, one co-author is misnamed, and two figure notes misdescribe their sources or definitions. These are housekeeping errors that do not change any result.

Technical detail and evidence

Why it matters, in detail. Items (6) and (7) misdescribe the wealth concept and the provenance of the fiscal-income series; the others are reference-list and labelling slips.

Evidence. Essay pp. 229, 234, 237, 238, 240, 242, 243. AJZ (2019 AER) Fig. 4 panel titles; Saez & Zucman (2020 JEP) Fig. 1-2 note; TabFig2022 Tables A2-A3 vs WID bulk extract (variable sfiinct992, through 2019): scripts/wid_vs_tabfig.py.

Why this severity. Presentation errors; none changes a number, although the Fig. 4 definition and the Fig. 5 source label could mislead a reader trying to check the series.

14minorconfirmed · newly identified here · misrepresentation of source

A few background statements are inaccurate: Pareto used mostly non-Swiss data, the US rise began slightly before Reagan took office, and there were earlier, rougher estimates of hidden offshore wealth before Zucman's. None affects the essay's main claims.

Technical detail and evidence

Why it matters, in detail. Items (1)-(2) are factual slips with no bearing on the series; item (3) is a priority claim that depends on what counts as 'broadly applicable' evidence.

Evidence. Pareto (1896) tables (pareto96.pdf pp. 2-3); TabFig2022 Tables A1/A2 1976-82 (scripts/tabfig_1976_1982.py); AJZ (2019 AER) fn. 28.

Why this severity. Historical and priority remarks outside the core findings.

15minorlikely · previously known · overclaiming

The essay says an independent team using better tax data got 'remarkably similar' results. For the top 1% the results are fairly close, but for the very richest the independent estimates are noticeably lower and rose by about a third less.

Technical detail and evidence

Why it matters, in detail. Top-1% level -2.6 pp (-7%) vs SZ20 in the same unit; top-1% rise -14% vs SZ20 and -34% vs PSZ; top-0.1% level -15%, rise -27% to -42%. Remaining disagreement is concentrated in the top 0.01% (corpus us-wealth-taxes-06).

Evidence. Smith, Zidar & Zwick (2023 QJE) abstract, fn. 1 (pp. 518-519), Table I (p. 550), Sec. VIII; SaezZucman2020JEPData.xlsx DataF1-F2(Wealth) L116 = 36.27% (equal split), I116 = 38.79% (tax units).

Why this severity. The wording overstates agreement mainly for the very top; the direction and the broad top-1% level agree, and SZZ themselves stress agreement on the fundamental story.

The authors' response. Saez & Zucman (2020, NBER w27921; 2022, NBER w30396) argue the remaining gap is concentrated in the top 0.01% and that the series are reconciled after correcting billionaire equity and top interest-bearing assets; SZZ's replies (2021-22) disagree.

16minorplausible · newly identified here · internal inconsistency

The essay uses 'shifted to' and 'booked in' tax havens as if they meant the same thing and quotes about 50% for both. They are different measures, and taken together the two numbers do not quite add up. This does not change the main point that US firms put a lot of profit in tax havens.

Technical detail and evidence

Why it matters, in detail. By construction shifted <= booked. TWZ's US numbers imply foreign profits of ~$566bn vs ~$448bn ($486bn with PR) for BEA non-oil MOFAs, consistent with broader coverage (oil, minority-owned affiliates, DII-based scaling); readers cannot tell that the 'shifted' and 'booked' ~50% figures measure different things.

Evidence. Essay p. 236. TWZ (REStud 2023) Sec. 6: 'if we compare the profits shifted by US multinationals to the profit made by these firms outside of the US, our estimates imply that US multinationals book 54% of their foreign profits in tax havens. The corresponding figure for other multinationals is 27%'. TWZ2022.xlsx DataF6 D4 = 0.4954. SZ2019MainFigures.xlsx DataF4b row 2015: F54-G54 = $203.3bn; O54 = $241.5bn.

Why this severity. A definitional muddle that does not change the qualitative message; the statistics come from different datasets and may be partly reconcilable (TWZ's US figures cover a broader set of affiliates).

Claims examined and rejected

Problems one auditor raised that did not survive checking.

The billionaire minimum income tax was in the FY2023 budget, so 'its 2022 budget' is wrong (I13). — The cited White House release is dated 28 March 2022; calling the budget released that year 'its 2022 budget' is a common calendar-year shorthand, not an error of substance.
'Hardly any income gains in real terms for the bottom 50 percent' misstates the evidence (part of I8). — The cited source [16] (p. 22) reports bottom-50% pre-tax income of $17,500 (1980) and $18,500 (2018), 0.2% a year, and calls it 'quasi-stagnation'; even the Feb-2022 PSZ vintage gives +14.5% over 1980-2019 (0.35% a year) against +69% on average. The paraphrase is faithful to the cited source; the contrast with Auten & Splinter is covered in F4.
The essay's 'about 50 percent vs about 30 percent' misquotes TWZ's 54% vs 27% (part of I10). — Rounding 54% and 27% to 'about 50' and 'about 30', with 'almost twice', is a fair approximation; the substantive shifted-versus-booked problem is kept in F10.
Blouin & Robinson's double-counting critique applies to Figure 3 (considered and rejected by codex). — Agreed with codex: Figure 3's series uses BEA profit-type return, which excludes equity income received (Wright & Zucman 2018 p. 5; TWZ 2023 Sec. 4.2.1). The critique applies to the direct-investment-income statistic in the text (F6).
Offshore-wealth size critiques (TJN: larger; Schmitz et al.: smaller; offshore-havens-01, -03) undermine the 8% (considered by both). — The essay reports [3] accurately; the critiques point in opposite directions and neither provides a demonstrated replacement for the 8% of household financial wealth.
Pellegrini, Sanelli & Tosti's lower revenue-loss estimate (offshore-havens-02) applies (considered by codex). — Not applicable: the essay states no revenue-loss figure.
TJN (2024) disputes the post-CRS fall in offshore evasion (offshore-havens-07), so 'well on its way to being substantially reduced' is unsupported (considered by claude). — The statement matches the evidence then available ([21], the Global Tax Evasion Report 2024); the TJN critique appeared after the essay and is disputed.
Wealth-tax revenue and optimal-tax critiques (us-wealth-taxes-11/-17/-20; optimal-tax-behavior-16 to -20) apply (considered by both). — Not applicable: the essay gives no revenue estimate or optimal-rate calculation, only that Zucman helped estimate budget effects.
Excluding Social Security wealth (us-wealth-taxes-09) invalidates the wealth-concentration trend (considered by codex). — Not counted: the essay's figure note specifies marketable wealth excluding unfunded pensions; including Social Security answers a different question.
A different capital-gains convention would overturn the 24% 'record' in 2021 (considered and rejected by codex). — Agreed: fn. 4 and the Fig. 5 note disclose the convention, and 2021 is a record in all three Piketty-Saez variants (Table A1 20.7-21.2%, A2 24.1%, A3 27.4-27.6%).
The Fig. 4 note's wording omits housing and business assets (part of I6). — Partly rejected: equity in private businesses is a financial asset in the Financial Accounts, so the omission concerns non-financial assets (mainly real estate). The definitional slip itself is kept in F15.

Audit notes

What reproduced. The essay has no data or appendix; every number restates a cited source, so reproduction meant tracing each number to the authors' workbooks, recomputing it, and checking key numbers against later or independent data. Reproduced exactly or to rounding: offshore wealth 8% of household financial wealth and Figure 1 (world 9.79%); hidden-wealth concentration (52/77/>90%); UK 2.73 -> 4.41%, US +0.71 pp, Russia 5.02 -> 12.50%; Scandinavian top-0.01% evasion 24.4% vs audits <=4.3%; the GLRRZ gradient (to the 2021 version); [11]'s 55%; TWZ's 36.2% and 54%/27%; SZ top-1% wealth 24.9 -> 37.9% and DFA-adjusted 28.1 -> 38.1%; Piketty-Saez top-1% reported share 8.5% (1970s) -> 24.1% (2021, a record); PSZ/WID pre-tax top 1% ~10-11% -> 19.1% (2019); [16]'s bottom-50% figures. Not reproduced or not supported as presented: the GLRRZ gradient in the version current at publication (~10% -> 16%; top-1% share +1.1 pp, not +1.5 pp); Figure 3's 2017 point (57.2% is a formula extrapolation; later BEA profit-type-return data give 45.8% excluding Puerto Rico, ~50% with the uplift) and its source attribution; [11]'s '>50% by the 2010s' on the profit-type measure (33-38% in 2010-13); the Fed DFA's own top-1% level (30.6% in 2018Q3 in the Jan-2024 vintage vs 38.1% plotted); Russia's 2009 excluding-offshore share (-0.22%). Not reproducible here: confidential IRS (GLRRZ) and Scandinavian leak/amnesty microdata, and TWZ's full country inputs.

Limits of this audit. (1) The essay has no data, so reproduction means tracing to source workbooks; GLRRZ and AJZ microdata are confidential. (2) The January-2024 DFA vintage comes from a Wayback Machine snapshot of the Fed's dfa.zip (21 Jan 2024); the harmonised ~36% figure holds the unfunded-DB adjustment at its 2020 size (+2.2 pp) and is approximate. (3) Post-2017 US haven profit shares come from the processed BEA series in the Garcia-Bernardo-Jansky-Zucman replication package, not rebuilt from raw BEA tables; that series is close to but not identical with Wright-Zucman's non-oil series (49.0% vs 49.5% in 2016). (4) The AS-PSZ, SZZ-SZ and TWZ-CbC debates were not adjudicated on the merits; only the essay's characterisation was judged. (5) The 2025 published Blouin-Robinson and Fuest et al. versions and the December-2025 GLRRZ draft were not read; corpus summaries are flagged as such. (6) No exact 2024 WID snapshot was available. (7) Biographical claims (17 translations; 'most cited solo work') and the OECD statement page were not verified. (8) The essay is a laudatory review by a frequent co-author (Saez co-wrote 9 of the 24 works in Table 1) and is not expected to survey every critique; the issues above concern the accuracy and balance of what it asserts.