How to Get $1 Trillion from 1000 Billionaires: Tax their Gains Now

Why it matters, in detail. Taxable gains = assessed value minus basis; with the note's implied basis of ~$1.56T held fixed, each 1% of Forbes value lost to discounts, attribution (non-grantor trusts, family members) or evasion lowers gains by ~1.6% and the levy by ~$18bn. Using the authors' 2019 rule (reported wealth = 85% of Forbes), gains fall from $2.71T to $2.07T and net revenue from $1.04T to ~$0.77T. Applying their 2021 15% rate to gains gives ~$0.87T. The retroactive April-1-2021 valuation date and fewer than 1,000 auditable returns justify a smaller haircut than for an annual wealth tax, but not zero. About 58% of top-400 wealth (more below the top 400) has no market price, and estate values of Forbes decedents average half of Forbes.

Evidence. Score: note pp.4-5; fn 3 (p.3): 'Billionaires manage to largely under-report their estate through tax avoidance'. Authors' own practice: Saez & Zucman letter to Sen. Warren, Jan 18, 2019, p.2 item 4: 'their reported wealth for tax purposes would be 85% of the Forbes estimate due to avoidance and evasion'; letter of Feb 24, 2021, p.2 ('reduce their tax liability by 15% through a combination of tax evasion and tax avoidance') and p.4 table note ('a 15% tax avoidance/evasion rate which is realistic given the robust proposed enforcement'); Saez, Yagan & Zucman, 'Capital Gains Withholding' (Jan 2021 draft cited by the note), p.16: 'we assume that the tax evasion rate would be 15% at the federal level' for a tax on the same unrealized gains. Hard-to-value share: Balkir, Saez, Yagan & Zucman (2025, NBER w34170, Sec. 3.1, p.9): 42% of top-400 wealth in listed C-corporations, 58% in private businesses (and listed partnerships), 74% for the next 300; SYZ (2021) Table 1: for households above $50m/$100m only 25-26% of gains are in publicly traded stock. (In their June 2019 reply to Summers & Sarin, p.3, the authors had put the private-business share of top-0.1% assets at 'only about 20%'.) Estate analogue: Raub, Johnson & Newcomb (2010, NTA Proceedings, pp.130-134): estate values average 0.50 of Forbes; discounts on 64.4% of post-2004 returns (restoring them raises the ratio 0.47 -> 0.54; with spousal wealth 0.58); Forbes credits trust assets to income beneficiaries; gaps are largest for portfolios of hard-to-value assets. Official scorers of a related design (Wyden's 2021 Billionaires Income Tax: JCT $557bn, PWBM $507bn) taxed only tradable assets up front. Adjudicator computations (scripts/adjudicator_sensitivity.py): 15% haircut on gains -> net $0.868T (-17%); reported wealth 85% of Forbes with basis unchanged -> $0.767T (-27%); 15% / 30% valuation shortfall on a 58% private share -> $0.883T / $0.722T (-15% / -31%).

Why this severity. The zero-shortfall assumption is unvalidated, is the most favourable possible choice, and departs without explanation from the authors' own treatment of the same population (85% of Forbes in 2019; 15% avoidance/evasion called 'realistic' in Feb 2021). Replacing it with the authors' own numbers removes a sixth to a quarter of the headline ($0.77-0.87T), and plausible valuation shortfalls on the private share give $0.72-0.94T. The qualitative claim survives, but $1T does not survive as a central estimate.

The authors' response. No response specific to this note was located. The note argues that the April 1, 2021 valuation date, fewer than 1,000 auditable returns, Forbes/Bloomberg lists, business balance sheets and NY-bill-style valuation rules would prevent gaming (pp.3-5), and concedes the experiment may 'not work as well as planned' (p.5). In their June 2019 reply to Summers & Sarin the authors argued that private businesses can be valued and that the estate-tax gap reflects weak enforcement. Saez & Zucman (2024, Oxford Open Economics) add that tax authorities can apply listed-firm multiples and accept payment in kind.

Why it matters, in detail. Revenue is nearly linear in the rate. At 23.8% the gross is $0.64T and the offset $0.14T (no realization response), leaving ~$0.50T. At a literal 2021 ordinary rate (40.8%) it is ~$0.97T. The authors' later restatement (SZ 2024) quotes '40%'. Neither Biden's 39.6% capital-gains rate nor the levy was enacted, so the policy-relevant yield under prevailing law was ~$0.4-0.5T.

Evidence. Abstract ('39.6%+3.8% under Biden's plan'); p.3 rate paragraph; p.5 ('maximalist version'). Same method at 23.8% (no realization response because the rate is unchanged): gross 0.238 x $2.705T = $0.644T; offset 0.238 x $50bn x 12 = $0.143T; net $0.501T (-52%); with the authors' own 15% haircut (F1) ~$0.40T (scripts/core_reproduction.py, adjudicator_sensitivity.py). Auditor B obtained the same $499.8bn.

Why this severity. The condition is disclosed in the abstract and on p.3, so the 43.4% estimate is not misstated. But the title and headline present $1T unconditionally, and under the alternative the note itself names (the law that actually prevailed) the yield is halved. This is a material presentation gap, not an error in the estimate.

The authors' response. The note labels the rate 'under Biden's plan', states the 23.8% fallback (p.3) and calls its estimate 'the maximalist version' (p.5). Saez & Zucman (2024, Oxford Open Economics 3(S1), p. i1131) restate the result as $1 trillion at 'a rate of 40%'.

Why it matters, in detail. The correctly specified liability is sum_i 0.434 x min(G_i, W_i - $1bn), not 0.434 x sum_i G_i. The loss provision is a put-like refund min(loss_i, G_i) x 0.434 on individually volatile fortunes: under a broad 'refund on decline' reading its expected cost is of the same order as the $130bn offset; under a narrow realized-carryover reading, and with timing at the end of the window, it is much smaller. Net of the +$112bn interest, plausible combined effects lie between roughly -$105bn and +$60bn under a narrow-to-moderate reading of the loss provision (more negative under a broad 'refund on any decline' reading).

Evidence. Cap: with a uniform 63.4% gain share on the ATF/IPS list it binds for 286 of 657 taxpayers and costs $43.5bn (3.7% of gross); uniform shares of 40-75% give $11-79bn, and a wealth-graded share (50% below $2bn, 65% above) gives $30bn (scripts/core_reproduction.py; both auditors obtained the same $43.5bn). Loss adjustment: Auditor A's ex-post illustration (refund on any Forbes wealth decline, matched lists, scaled to $1.17T) gives $113-203bn by 2026 and $209-263bn at the 2023 trough, reproduced here. These are upper bounds, because Forbes declines include giving and the provision covers only realized-loss carryovers at the end of the period; the note does not specify the provision tightly enough to score it. The two largest fortunes (Bezos and Musk, $340bn) carry roughly an eighth of gross liability. 2021 credit ~$9bn. Scaling the offset from the top 400 (90% of wealth) to all billionaires raises it by $14.5bn. Interest (equal principal, 1.74%): +$112bn nominal.

Why this severity. The omissions point in both directions. The cap ($20-45bn under plausible distributions), the 2021 credit and full coverage lower revenue; interest raises it by about 11%. So the central estimate moves little. But the unpriced asymmetric loss refund adds material downside uncertainty that the note does not acknowledge.

Why it matters, in detail. Lump-sum status requires that the levy is unanticipated (true for an April-1-2021 base announced on April 14) and that future levies are not expected (undercut by the note's own framing as a pilot for a recurrent or broadened tax). If repetition is expected, the levy acts as an expected tax on future accumulation and entrepreneurial returns. There is no effect on the fixed-base receipts, and the magnitude of real effects is not identified.

Evidence. Note abstract and p.5 ('Economic effects'; 'How does this one-time tax fit with more structural reform?'). Kopczuk (2019), comment on Saez & Zucman, BPEA Fall 2019, pp.514-515: 'a onetime, swift, and unexpected tax on principal is ... non-distortionary ... though "unexpected" and "swift" are difficult to execute and "onetime" is hard to commit to'. Scheuer & Slemrod (2021), JEP 35(1), NBER WP 28150 Sec. 6.2 (p.19): the appeal 'critically hinges on the ability of policymakers to implement such on short notice and on their commitment not to make such taxes permanent or to reintroduce them'.

Why this severity. This qualifies a headline claim of the abstract (C5), not the revenue figure. The claim survives as a conditional benchmark, but the note's own text weakens the condition, and the authors had been warned of exactly this point on their 2019 BPEA paper. The size of any behavioural effect is unidentified.

The authors' response. Saez & Zucman (2024, Oxford Open Economics 3(S1), p. i1131) add the missing condition: the argument holds 'at least if the government can credibly commit that this is a one-off tax'.

Why it matters, in detail. Incremental revenue = levy minus the baseline tax on pre-2021 gains collected in the window through sales and deaths. With death realization in the baseline, forgone death-tax revenue is P(death within 10 years) of ~0.17-0.25 x (1 - deferral) x (1 - avoidance) x 0.434 x $2.7T, about $40-120bn. Raising the revenue-maximizing rate from 30% to 35-50% (e from 2.33 to 1.0-1.86) adds $20-63bn to the offset. The note's fn-3 claim that estate planning defeats taxation at death would shrink the first term, but the same planning erodes the levy's own base (F1).

Evidence. Note pp.3-5 and fn 8. Saez, Yagan & Zucman (2021) fn 3: if death is a realization event, 'the revenue maximizing tax rate increases somewhat'. Kamin (2015) fn 29: the revenue-maximizing rate of a mixed system 'should be above that of the pure realization system'; Kamin fn 39 flags the estate-tax interaction. Wealth-weighted probability that a 2021 US billionaire dies within 10 years: 16.7-24.9% (HMD 2019 US mortality x elite factor 0.4-0.7; Forbes 2021 ages and genders; scripts/death_baseline.py, reproducing Auditor A). Forgone death-tax revenue on pre-2021 gains: ~$40-120bn after 40-60% spousal/charitable deferral and 30-50% avoidance. A revenue-maximizing rate of 35-50% instead of 30% raises the offset by $20-63bn (scripts/adjudicator_sensitivity.py). Auditor A's elasticities of 0.5-1.0 (+$63-94bn) are the upper end given the authors' own 'somewhat'.

Why this severity. Under the coherent Biden-plan baseline the note invokes, the offset is too small by roughly $60-180bn (6-17% of the headline; central ~10%). That is a noticeable but not transformative reduction, and it depends on a baseline the note invokes but never specifies.

Why it matters, in detail. G2021 = W2021 - 0.5 x W2019 equates the change in wealth with the change in gains for a changing population. Each $100bn of extra basis cuts the levy by $43bn. The authors' Jan-2021 convention (50% of current wealth) gives G = $2.13T and net $0.79T. The SCF tail (which excludes the Forbes 400) ranges from 52-66% (2019) to 53-56% (2022). Founder concentration at the very top argues for >60%. A 55-70% range brackets the note's 63.5%.

Evidence. Note p.4. SYZ (2021, Jan draft) p.16 and fn 21 (Forbes billionaires as of Aug 13, 2020, $3,700B, 'assuming that half their wealth is unrealized capital gains'; $1,850B) and Table 1 (50% for >$50m, 54% for >$100m). Adjudicator SCF re-run (scripts/scf_gain_shares.py; 5 implicates): 2019: 42.7% top 1%, 52.1% top 0.1%, 65.6% top 0.01%, 83.5% top 0.001%, 59.3% NW >= $100m, 75.4% NW >= $500m; 2022: 43.3%, 53.1%, 55.7%, 51.1%, 55.7%, 52.7% (15-90 records per implicate at the top). Forbes 2021 annual list: self-made share of US wealth 72.9%. Sensitivity: 2021 gain share 50% / 55% / 60% / 70% -> net $0.79T / $0.89T / $0.98T / $1.16T (scripts/adjudicator_sensitivity.py). Auditor B: exact-name matching of the 2019 and 2021 annual lists gives 561 continuing names, 163 entrants and 46 exits.

Why this severity. Plausible 2021 gain shares of 55-70% move net revenue by -15% to +12%. The note's figure is plausible (the bottom-up check supports it) but is unvalidated, is presented as a point estimate without sensitivity, and the sign of its error is unknown.

The authors' response. The note calls the 50% starting share conservative and the post-2019 growth 'overwhelmingly new gains' (p.4). SYZ (2021, fn 21) argue that the Forbes 400's gain share is 'substantially higher' than 50%. Saez & Zucman (2024) restate the stock as 'more than $2.5 trillion'.

Why it matters, in detail. The estate-to-Forbes ratio (~0.5, ~0.58 after discount and spousal adjustments) mixes evasion and avoidance with Forbes attribution and valuation error. If the gap is mainly attribution or overvaluation, the Forbes-based levy base is overstated. If it is mainly avoidance through discounts and trusts, the same channels apply to a one-time valuation. Either reading argues for a haircut in F1.

Evidence. Note fn 3 (p.3); IRS SOI Table 1, estate returns filed in 2019 (471 returns of $50m or more, $70.44bn; note [1]: mostly 2018 deaths). Raub et al. (2010), pp.130-134: mean ratio 0.50; married decedents 0.46 -> 0.53 with jointly held assets; post-2004 discounts 0.47 -> 0.54, and 0.58 with spousal wealth; values 'in closest agreement when valuation issues were relatively objective'. Arithmetic: 14.9 estates, $22.3bn, 0.72% of $3.11T (the note says 'less than .7%'). Auditor A's mortality check: filings capture 31-53% of expected billionaire decedent wealth, so a large gap exists but its causes are mixed.

Why this severity. This affects the motivation for preferring a one-time levy over taxation at death, not the score's arithmetic. The gap itself is real. Its main relevance is that it cuts against the note's own reliance on Forbes values (F1).

The authors' response. In their June 25, 2019 reply to Summers & Sarin (p.5) the authors attribute the estate/Forbes gap to 'widespread estate tax dodging' and to a collapse in estate-tax enforcement.

Why it matters, in detail. For a taxpayer who would realize in year n, the levy raises the present value of tax by a fraction 1-(1+r)^-n of the liability at the government's rate r (15.8% for n = 10, r = 1.74%), and by more at a private discount rate. Basis adjustment removes double taxation only in nominal terms.

Evidence. Abstract; p.3 (basis adjustment; 1.74% installment interest). Auditor B's counterexample, verified: $100m of gains, 43.4% in both cases, sale in year 10. The present value of the deferred tax at 1.74% is 43.4/1.0174^10 = $36.52m, against $43.4m for installments with interest, so the levy adds +$6.88m (15.8% of liability).

Why this severity. This is a wording and fairness point about a secondary claim. It has no effect on the revenue estimate, and the note's own framing makes clear that ending deferral is the intent.

Why it matters, in detail. Section 965 deems realized-but-undistributed corporate earnings repatriated, which is income realized at the entity level, and Moore rests on that ground. A deemed realization of individuals' unrealized appreciation raises the Macomber realization question that Moore declined to decide. Litigation risk is therefore a non-trivial, unpriced contingency on the whole $1T.

Evidence. Note pp.5-6. Moore v. United States, 602 U.S. ___ (2024), syllabus (d) ('Nor does this decision attempt to resolve the parties' disagreement over whether realization is a constitutional requirement for an income tax') and opinion fn 2 ('does not address ... (ii) taxes on holdings, wealth, or net worth; or (iii) taxes on appreciation').

Why this severity. This is a feasibility caveat rather than a scoring error, and the decisive case post-dates the note. The precedent is distinguishable, not contrary.

Why it matters, in detail. None of the slips affects the gross-minus-offset arithmetic. The $1.25T figure appears only in the abstract and is irreproducible. The 18% vs 17.1% gap suggests different GDP denominators for the text and the figure. The 2024 restatement keeps the $1T headline with a lower base, a lower rate and no offset.

Evidence. ATF/IPS master table 'Master List' (Forbes real-time): 657 billionaires, $4,263.5bn on Mar 18, 2021; $2,947.5bn on Mar 18, 2020 (scripts/core_reproduction.py). Figure pixel-calibrated (scripts/figure_points.py): 1982 2.21%, 2019 14.11%, 2020 15.64%, April 2021 17.11%. Top 400 of the ATF list / GDP: 17.8% (Q4-2020 SAAR, BEA Mar 25, 2021) or 18.3% (2020 annual). Auditor B (S11) reports 258 units in 1982 in the SZ (2020) JEP data. SZ (2024) Oxford Open Economics 3(S1), p. i1131.

Why this severity. These are typographical and rounding slips that do not enter the $1.04T score, which uses consistent inputs ($4.26T, $1.15T, $2.7T). The abstract's $1.25T is the only number that cannot be reproduced from anything in the note.

Why it matters, in detail. Payments of ~2.75% of wealth a year plus 1.74% interest on the balance are below typical returns, so collateralized borrowing or partial sales are feasible for holders of listed equity. For closely held firms without outside shareholders, or with transfer restrictions, the premise of 'previous rounds of equity finance' does not hold, and forced sales or in-kind payment would be needed.

Evidence. Note abstract and p.4. Balkir et al. (2025) Sec. 3.1: 58% of top-400 wealth in private businesses (74% for the next 300). Penn Wharton Budget Model (Nov 8, 2021), on Wyden's Billionaires Income Tax: borrowing against a family business 'might be impossible, quite costly, or prohibited by contract (e.g., a sporting franchise)'. Annual installment of about 0.434 x 0.635 / 10 = 2.75% of April-2021 wealth, plus interest.

Why this severity. Installments of about 2.75% of wealth a year are manageable for most billionaires, so this qualifies rather than refutes the feasibility claim, and it does not affect the revenue estimate.

The authors' response. Saez & Zucman (2024, Oxford Open Economics) propose allowing payment in kind with private-business shares (in the wealth-tax context).