← All papers
Emmanuel Saez, Gabriel Zucman · Oxford Open Economics 3(Supplement 1): i1130-i1132 (Dimensions of Inequality: The IFS Deaton Review)
Top Incomes and Tax Policy
9 problems found: 3 moderate · 6 minor
What the paper claims
- Modern tax systems do not satisfactorily tax the wealthiest: corporate income taxes have shrunk under tax competition, the realization-based income tax can be avoided while gains stay unrealized and profits are retained, and estate taxes are easy to avoid.
- A modern progressive net wealth tax with a high exemption and a comprehensive base differs from past European wealth taxes and is more enforceable today: third-party information can pre-populate returns (whereas 'European wealth taxes were based on self-reported asset values'), tax havens now report automatically, elasticities can be reduced by design, and private businesses can be valued with listed-firm multiples or paid for in kind.
- An annual 0.2% tax on the market value of listed corporations would raise substantial revenue and would be progressive.
- A one-off tax on billionaires' unrealized capital gains (deemed realized on a set date, taxed under the individual income tax 'just like regular capital gains', payable over 10 years) would not distort behaviour if credibly one-off, would be hard to avoid, and would raise about $1 trillion in the US.
- Given the size and concentration of wealth, the revenue potential of these policies is significant.
Every problem found
1moderateconfirmed · newly identified here · internal inconsistency
The '$1 trillion' assumes the gains would be taxed at about 40%. Yet the paper says they would be taxed like ordinary capital gains, whose top US rate is about 24%. At that rate the same calculation gives roughly half a trillion dollars. That is still a lot of money, but the headline depends on a rate rise that was proposed and never passed, and the paper does not say so.
Technical detail and evidence
Why it matters, in detail. The $1T is 0.40 x $2.5T gross. The source gets $1.04T net as 0.434 x $2.705T minus a $0.13T offset for lost future realizations. The text's design is deemed realization taxed as regular capital gains. The current-law top federal rate is 23.8% (20% + 3.8% NIIT), which gives $0.60-0.64T gross and $0.45-0.50T net under the authors' own method, about -50% against the headline. The 40% presupposes ordinary-rate taxation of high-income gains. That was a live proposal in 2021-23 but was never enacted. Narrower current-rate designs score in the same range (PWBM: $507bn over 10 years for the Wyden plan). The bias runs one way (upward), and the order of magnitude ('hundreds of billions') survives.
Evidence. Paper p. i1131 (both quoted sentences). Saez & Zucman (2021b) 'How to Get $1 Trillion from 1000 Billionaires', p. 3 ('This is presumed to be 39.6% if the Biden campaign plans are enacted plus the existing 3.8% ... for a total rate of 43.4%. If the top tax rate on capital gains does not change, the tax rate would be the existing 20%+3.8%'). US Treasury, General Explanations of the Administration's FY2023 and FY2024 Revenue Proposals ('Tax capital income for high-income earners at ordinary rates', income above $1 million; 43.4% including NIIT). Roll Call (27 Oct 2021): Sen. Sinema 'wouldn't support dialing up ... the capital gains rate'; the Wyden Billionaires Income Tax therefore used current-law rates of up to 23.8%. PWBM (8 Nov 2021) scored that plan at $507bn over FY2022-31. Adjudicator re-scoring (scripts/oneoff_check.py): the SZ base of $2.705T gives $1.174T gross / $1.043T net at 43.4% and $0.644T gross / $0.501T net at 23.8%. The paper's $2.5T at 23.8% gives $0.595T gross / $0.452T net. 40% on $2.5T is $1.000T gross / $0.861T net of the source's lost-realization offset.
Why this severity. The $1T is arithmetically right for the stated 40% rate. But under the design as described and under current law, the figure roughly halves (about $0.5T net). That is an important undisclosed caveat on the headline number. The qualitative claim of substantial revenue still survives.
The authors' response. The source note states the conditionality explicitly (SZ 2021b, p. 3). It adds that 'a one-time tax on 50% of the gains would raise about $.5 trillion' (p. 5). No response specific to this paper was located.
2moderateconfirmed · newly identified here · external inconsistency
The paper says Europe's past wealth taxes relied on people declaring their own wealth, while today governments could use information from banks. But Denmark and Sweden already received bank information decades ago, as the authors' own research shows, and wealthy Danes still cut their taxable wealth substantially over time. What is genuinely new is information-sharing with tax havens. So the claim that a modern wealth tax would be much easier to enforce is weaker than presented, though not refuted.
Technical detail and evidence
Why it matters, in detail. The administrability argument implies that low historical compliance reflected self-assessment. In Denmark, with third-party reporting of deposits, listed securities, mortgages and real estate, JJKZ estimate an effect of the 1989 reform on the taxable wealth of the very wealthy of 31% after 8 years, with implied long-run elasticities of 0.77-1.15 with respect to the net-of-tax return. Those responses include real saving responses and evasion through the self-reported classes, which remain hard to report today. Other evidence (Brülhart et al. 2022; Advani & Tarrant 2021 for the UK, 7-17% base erosion per 1% tax) shows design-dependent but non-trivial elasticities. Automatic exchange with tax havens (FATCA/CRS) is the genuinely new element. The paper gives no revenue figure, so the effect is on the strength of the feasibility claim, not on a number.
Evidence. Paper p. i1130 ('Today's enforcement context is also markedly different. European wealth taxes were based on self-reported asset values.'). Jakobsen, Jakobsen, Kleven & Zucman (2020) QJE 135(1): pp. 332-333 ('Most assets were third-party reported under the Danish wealth tax-thus limiting the scope for evasion'; implied long-run elasticity of taxable wealth with respect to the after-tax return 0.77 for the moderately wealthy and 1.15 for the very wealthy), p. 336 (list of third-party-reported and self-reported items), pp. 341-342. SZ letter to Sen. Warren (18 Jan 2019), fn 2 ('Sweden and Denmark in two countries with systematic third party reporting of wealth'; 'Switzerland has no systematic third party reporting').
Why this severity. One of the paper's three stated reasons why a modern wealth tax would be more enforceable is factually overstated, against the authors' own research. The other arguments (high threshold, comprehensive base, automatic cross-border reporting) are untouched, and no number changes.
The authors' response. None located for this wording. In the 2019 Warren letter (fn 2) the authors themselves contrast Scandinavian third-party reporting with its absence in Switzerland, to explain the larger Swiss responses.
3moderatelikely · previously known · speculative assumption
The estimate assumes billionaires would report their wealth at exactly the values Forbes publishes, with no undervaluation or avoidance. Yet the same authors assume 15% would be hidden under an annual wealth tax, and argue that billionaires already undervalue their estates. Because only the gain is taxed, undervaluing wealth by 10% cuts the taxable amount by about 16%. The real yield would probably be lower than stated, though nobody can say exactly by how much.
Technical detail and evidence
Why it matters, in detail. Gains equal reported value minus a fixed basis (~$1.56T on a $4.26T base), so a valuation shortfall h cuts taxable gains by h x 4.26/2.705, about 1.6h. h = 0.10 gives -16% and h = 0.15 gives -24%. The authors' own Warren assumption (85% of Forbes) and their estate-tax evidence point to h > 0. Private shares make up 64% of gains above $100m. Forbes also attributes family and trust holdings to individuals (corpus us-wealth-taxes-11). There are offsets. The retroactive valuation date rules out post-announcement restructuring. Listed stakes, which dominate the largest fortunes, cannot be undervalued. And the 50% 2019 gains-share assumption looks conservative against SCF 2019 (59-61% of wealth above $100m), worth ~+10% on the base. Net direction: downward, magnitude uncertain.
Evidence. Paper p. i1131 ('It would be hard to avoid ... daily estimates compiled by Forbes and Bloomberg'). SZ (2021b): p. 3 fn 3 (estate-tax under-reporting; Raub, Johnson & Newcomb 2010) and fn 5 (cap at rate x (wealth - $1bn)); p. 4 (billionaires 'report the value ... and basis' on a new form; the score applies no haircut). SZ letter to Sen. Warren (18 Jan 2019), pp. 2-3: 15% avoidance/evasion; billionaire base = '85% of the Forbes estimate'. Saez-Yagan-Zucman (2021) Table 1, reproduced in Zidar (2022, IFS): above $100m, 54% of wealth is unrealized gains; of those gains, 26% are in public stock, 64% in private business shares, 10% in real estate. Treasury FY2023 Green Book: its own minimum tax on unrealized gains values non-tradable assets at 'the greater of the original or adjusted cost basis, the last valuation event ... or other methods approved by the Secretary', not by annual market valuation. Adjudicator scenarios (scripts/oneoff_check.py), with basis fixed at $1.555T: each 10% shortfall in reported wealth cuts gains by ~16%. A 15% shortfall gives -24% ($0.77T net at 43.4%, $0.69T at 40%, $0.35T at 23.8%). The notch cap costs -4.6% on the 2021 Forbes list.
Why this severity. The zero-shortfall assumption biases the one-off estimate upward by an uncertain amount; each 10% valuation shortfall removes ~16% of the base. Combined with F1, the plausible figure falls to ~$0.35-0.5T. But a realistic shortfall for a retroactive, fully audited levy on fewer than 1,000 returns has not been estimated, so this is not graded major.
The authors' response. Saez & Zucman read the Forbes-to-estate gap as estate-tax avoidance under collapsed enforcement rather than Forbes error (response to Summers & Sarin, 2019, as summarised in corpus entry us-wealth-taxes-17). SZ (2021b, p. 4) argue that the IRS can audit each of fewer than 1,000 returns and that valuation rules can be strengthened. They add that Forbes/Bloomberg lists and business balance sheets give the government 'an enormous leg-up'.
4minorconfirmed · newly identified here · data error
The paper puts the combined economic output of the G20 at $100 trillion. For the year the figures refer to, it was about $76 trillion. The dollar revenue estimate (about $180 billion a year) is roughly right. The slip makes the tax look smaller relative to the economy than it really is, so it does not help the authors' case.
Technical detail and evidence
Why it matters, in detail. Holding $180bn fixed, revenue is 0.236% of 2019 G20 GDP, not 0.18%. With reconstructed 2019 market capitalization it is $157-168bn (0.21-0.22% of GDP); with 2020 values, $188bn (0.25%). The capitalization-to-GDP ratio is 103-110% (2019) or ~126% (2020), not 90%. The error biases the revenue/GDP figure downward (conservative).
Evidence. Paper p. i1131. Saez & Zucman (2022), Economic Policy 37(110), summary page and pp. 217, 223; April 2021 draft p. 9. World Bank WDI NY.GDP.MKTP.CD (vintage 8 Oct 2026; adjudicator script scripts/g20_check.py): G20 union $76.25T (2019), $74.57T (2020), $85.27T (2021), $88.21T (2022), $92.62T (2023), $96.33T (2024); world $88.80T (2019). Market capitalization from WFE-based reconstructions by Auditor A ($78.6-81.3T in 2019; $93.8T in 2020) and Auditor B ($79.1T in 2019; $84.0T including Hong Kong).
Why this severity. The stated GDP and ratios are wrong by ~30%. But the dollar revenue ($180bn) is within ~10% of independent reconstructions, and the error understates revenue relative to GDP, so it does not prop up the conclusion. Codex graded this moderate; the adjudicator grades it minor because it does not undermine any core claim.
5minorconfirmed · newly identified here · methodology
The revenue figures are simple multiplications that assume nothing changes once the tax exists. The authors' own fuller paper says share prices would fall by about 5%, lowering revenue to match, and that taxing only listed companies would encourage firms to stay private. The short paper leaves both out. Some of the money would also come from shares owned by governments. Together these trim the estimate by perhaps a tenth or more, which does not change the conclusion.
Technical detail and evidence
Why it matters, in detail. Capitalization alone gives $171bn instead of $180bn. Codex's perpetuity derivation gives $164-176bn for discount-minus-growth of 2-8%. Excluding tax paid by public-sector owners (~14% globally, concentrated in China) brings revenue to roughly $147-155bn net of intra-government transfers. Delisting or staying private (if the private leg is not enacted) and possible corporate-income-tax deductibility of cash payments would lower it further, unquantified. The direction is downward, by well under a quarter.
Evidence. Paper p. i1131. Saez & Zucman (2022), Economic Policy: p. 219 (§3.1, 0.2/4 = 5%), p. 220 (§3.3, private businesses), p. 222 (§4.1, 'reduce capitalization (and thus revenues) by 5%'), p. 218 (payment 'in cash or in-kind'). De La Cruz, Medina & Tang (2019), OECD 'Owners of the World's Listed Companies': 'Today 14% of global stock market capitalisation is held by the public sector'; China accounts for 57% of public-sector holdings.
Why this severity. The omitted adjustments cut net revenue by roughly 5-20%, below the quarter threshold. The authors disclose them in the longer source paper.
The authors' response. The source paper itself labels the figures mechanical and proposes the 5% adjustment and a private-company leg (Economic Policy 2022, §§3.1, 3.3, 4.1).
6minorconfirmed · newly identified here · documentation gap
The published version says billionaires owned $4.26 trillion 'on April 1 of this year'. The figure is actually from 2021, not the publication year, and two references are missing or garbled. These are editing slips. They change no conclusion, because billionaire wealth has not fallen since 2021.
Technical detail and evidence
Why it matters, in detail. No number changes between the 2022 and 2024 versions. Only the date label became ambiguous. Later Forbes annual totals ($4.49T in 2023, $5.70T in 2024) show the 2021 base is not stale in the direction that would overstate revenue, although the gains share at later dates was not re-estimated.
Evidence. Paper pp. i1130-i1132; IFS Deaton Review version (April 2022), pp. 3-4; SZ (2021b), p. 2. Forbes annual lists (adjudicator script scripts/oneoff_check.py): US-citizen billionaires $4.398T (724 people) in 2021, $4.489T (734) in 2023, $5.701T (813) in 2024.
Why this severity. Editorial slips with no effect on any number or conclusion. The billionaire base has not shrunk since 2021.
7minorlikely · previously known · internal inconsistency
The paper says a one-time tax on past gains would not discourage work or investment, provided people believe it will never be repeated. But the authors' own proposal calls it a trial run for a possible permanent tax, which makes that belief harder to sustain. This weakens the 'no harm to incentives' argument somewhat; the revenue figures are unaffected.
Technical detail and evidence
Why it matters, in detail. A capital levy is lump-sum only if it is unanticipated and not expected to recur. The source explicitly frames the levy as an experiment toward recurrent mark-to-market taxation, which lowers the credibility of the one-off commitment. If recurrence is anticipated, future accumulation faces an expected accrual tax with standard capital-tax distortions, offset in part by reduced lock-in. The effect is qualitative.
Evidence. Paper p. i1131 ('For economists ... credibly commit that this is a one-off tax'). SZ (2021b), p. 5 ('How does this one-time tax fit with more structural reform?'). Kopczuk (2019), comment on Saez & Zucman, BPEA Fall 2019, manuscript p. 3: a one-time unexpected capital tax is non-distortionary, but 'unexpected' and 'swift' are difficult to execute and 'one-time' is hard to commit to.
Why this severity. This qualifies a secondary efficiency claim that the 2024 text already hedges. It does not affect the revenue estimates.
The authors' response. The 2024 version adds the credible-commitment qualifier (p. i1131); no further response located.
8minorplausible · newly identified here · documentation gap
The paper says the UK stock market is worth about 125% of national income but gives no source. Official UK statistics put UK companies' listed shares at roughly 90-105% of national income in recent years. On that basis the UK revenue would be a little smaller, about 0.2% rather than 0.25% of national income. The difference is small.
Technical detail and evidence
Why it matters, in detail. On the ONS measure of UK-incorporated listed equity, the 0.2% tax raises 0.18-0.21% of GDP (2014-2022), so the stated 0.25% is 17-38% too high. Against the WDI/WFE 2021 ratio (118.9%) the overstatement is only ~5%. Direction: slight overstatement of UK revenue.
Evidence. Paper p. i1131. ONS 'Ownership of UK quoted shares' 2018/2020/2022: £1.97T, £2.17T and £2.42T, against ONS GDP (YBHA) of £2.178T, £2.132T and £2.591T, i.e. 90%, 102% and 93%. Implied 2014 and 2016 values are 107% and 102%. WDI CM.MKT.LCAP.GD.ZS (GBR): 115.7% (2014), 118.9% (2021), 97.3% (2022).
Why this severity. An undocumented input at the high end of the measured range. Corrected UK revenue would be about 0.18-0.24% of GDP rather than 0.25%, a small difference.
9minorplausible · newly identified here · overclaiming
The paper says the tax would mainly hit the rich but backs this only with US figures. In the UK, more than half of the shares of UK listed companies belong to foreign investors, so a UK-only tax would mostly fall on foreigners. How the rest would be shared among UK households is not shown. This is a gap in the evidence, not a sign that the tax would hit the poor.
Technical detail and evidence
Why it matters, in detail. Under owner incidence (capitalization at enactment), more than half of the UK burden is exported. The domestic remainder falls on directly held shares, which are concentrated, and on pension, insurance and fund holdings, which are broader. No UK joint income-equity distribution is shown. The US evidence does hold up: the top 1% by income hold ~30% of equity against 19% of pre-tax income (Fed DFA check: 27-33%; Codex SCF check: 30.9%). Effect: an unevidenced extrapolation, not a demonstrated error.
Evidence. Paper p. i1131. ONS 'Ownership of UK quoted shares': rest of world 55.7% (2018), 56.3% (2020), 57.7% (2022); UK individuals 13.3%, 12.0%, 10.8%. Saez & Zucman (2022) Economic Policy §5.3 (inversion incentive under a unilateral tax).
Why this severity. The UK distributional claim is unevidenced. But nothing suggests the tax would be regressive, and tax exported to foreign owners is a revenue gain from the UK perspective.
The authors' response. The paper recommends G20 coordination, under which cross-border holdings largely net out (Economic Policy 2022, p. 222).
Claims examined and rejected
Problems one auditor raised that did not survive checking.
- I9: The introduction's 'surge in income and wealth concentration at the top' (with 'including the United Kingdom') misrepresents the cited World Inequality Report 2018, which calls UK wealth inequality's rise 'moderate'. — The sentence is a general statement about 'many developed countries', and the cited WIR 2018 does document a sharp rise in top shares (dramatic for the US). Its UK chapter (§4.6, p. 241) reports that the top 0.1% UK wealth share doubled, from 4.5% to 9% (1984-2013), and that the top 1% share rose from 15% to 20-22%. For UK income concentration the 'surge' is uncontroversial. The wording is a framing choice, not a misrepresentation, and no conclusion depends on it.
- Within I1: the ordinary-rate capital-gains increase behind the 40% was 'abandoned in October 2021'. — Overstated. It was dropped from the 2021 reconciliation package, but the Treasury Green Books for FY2023 (March 2022) and FY2024 (March 2023) re-proposed taxing high-income capital gains at ordinary rates (43.4% including NIIT). It has never been enacted. The underlying finding (F1) stands, since the 40% presupposes an unenacted change the paper does not disclose.
- Within I2: the unscored notch cap of SZ (2021b, fn 5) costs about $85bn (~7% of gross revenue). — Overstated. Applying the cap person by person to the actual 2021 Forbes list, with a uniform 63.5% gains share, gives -4.6% (about $55bn at 43.4%); Codex obtained -3.9%. The 7% came from a Pareto approximation. The cap is kept as a small component of F2.
- Within I2: the Raub, Johnson & Newcomb ratio (estate values ~50% of Forbes) implies a ~79% cut in taxable gains. — Not a credible central scenario for this levy. That ratio compares estates at death, after lifetime planning and gifts, with Forbes estimates that include measurement and attribution error. It is not an estimate of under-reporting under a retroactive, fully audited one-off levy. It is retained only as an extreme bound.
- I1 severity: the G20 GDP denominator error is 'moderate'. — The error is confirmed, but the severity is lowered to minor. The dollar revenue is within ~10% of reconstructions, and the error understates revenue relative to GDP, so it does not undermine any core claim.
- Hypothesis examined by Codex: paying over ten installments makes the $1 trillion overstated in present value. — Dismissed by Codex, and the adjudicator concurs. SZ (2021b, p. 3) specifies an interest surcharge pegged to the 10-year Treasury rate (1.74% at the time) on deferred installments.
- Hypothesis examined by Codex: the 3.4% figure is presented as an all-taxes rate. — Dismissed by Codex, and the adjudicator concurs. The paper says 'paid in federal income taxes ... only 3.4% of their wealth gain', matching ProPublica's $13.6bn / $401bn.
- Hypothesis examined by Codex: the source score omits the offset for lost future realizations, basis adjustment, later-loss relief and the notch protection. — Dismissed by Codex, and the adjudicator concurs. All are present in SZ (2021b, pp. 3-5), and the offset ($0.13T) is netted in the $1.04T. Only the notch cap is uncosted (small; see F2).
- Hypothesis examined by Codex: calling the levy 'a simple way to tax above-normal returns' is wrong because accumulated nominal gains include normal returns and inflation. — Dismissed by Codex as an interpretive boundary, and the adjudicator concurs. The paper does not claim the tax reaches only above-normal returns. For founder fortunes built from near-zero basis, most gains are plausibly above-normal. Not material.
Audit notes
What reproduced. Reproduced by the adjudicator: (1) all of the paper's arithmetic. (2) ProPublica's 3.39%. (3) The SZ (2021b) one-off score of $1.043T net; at the current-law 23.8% rate the same method gives $0.501T net ($0.452T on the paper's $2.5T); valuation-shortfall and notch-cap scenarios are in scripts/oneoff_check.py. (4) G20 GDP of $76.25T in 2019 (and below $100T in every year through 2024) from WDI, versus the paper's $100T (scripts/g20_check.py). (5) UK listed-equity/GDP of 90-107% (ONS, UK-incorporated, 2014-2022) and 116-119% (WDI, 2014/2021), versus 125%. (6) Top-1%-by-income share of equity incl. pensions of 27-33% from the Fed DFA (scripts/dfa_equity_check.py), versus ~30%. (7) Forbes US billionaire totals for 2019-2024. Taken from the auditors after checking their methods, but not rebuilt: WFE-based G20 market capitalization ($79-84T in 2019; ~$94T in 2020; the two auditors agree within ~5%), A's DINA micro-file computation (38-45% at realistic pension-equity shares; files no longer cached) and B's SCF benchmark (30.9%). Not reproducible: the exact $4.26T Forbes real-time figure of 28 March 2021 (the annual list of 5 March 2021 gives $4.40T for 724 people), and the authors' sources for '125% of UK GDP', '$90 trillion' and '$100 trillion' (undocumented in every version). No replication package exists.
Limits of this audit. No replication package exists; the sources of '125% of UK GDP', '$90 trillion' and '$100 trillion' are undocumented, so recomputations use current-vintage WDI (8 Oct 2026), ONS and Fed DFA (2026Q2) data rather than the 2021 vintages the authors used. G20 market capitalization was not rebuilt from exchange-level WFE data; the auditors' two reconstructions were relied on. A's DINA micro-file computation was not re-run (the files are no longer cached). Forbes real-time data for 28 March / 1 April 2021 were unavailable; annual lists were used. Valuation-shortfall effects (F2) are illustrative scenarios, not estimates. No behavioural, general-equilibrium or incidence model was built for any of the three taxes, and the paper gives no wealth-tax revenue figure to test. Legal feasibility (e.g. the US constitutional question of taxing unrealized gains, left open by the Supreme Court in Moore v. United States, 2024) and political feasibility (a G20 agreement) were not assessed. Corroborating evidence for C1 from Leiserson & Yagan (2021) and Balkir, Saez, Yagan & Zucman (2025) is cited from corpus entry us-wealth-taxes-14 and was not opened.