Emmanuel Saez, Gabriel Zucman · Letter/note, September 22, 2019

Scoring of Senator Sanders' Wealth Tax Proposal

13 problems found: 2 major · 5 moderate · 6 minor

What the paper claims

The Sanders wealth tax would raise about $4.35 trillion over the ten-year budget window 2019-2028, about 1.6% of GDP per year.
About 180,000 households (about the top 0.1%) would be liable and the tax would raise $335bn in 2019, assuming each family hides 16% of its wealth.
The Sanders tax raises $135bn (68%) more than the Warren tax in 2019 ($335bn vs $200bn); halving singles' brackets adds $11bn (8%), the 1% bracket $19bn (14%), graduated rates above $250m $106bn (78%).
The tax would 'fully eliminate' / 'entirely close' the gap between wealth growth for billionaires and for the middle class (average family).
The tax would cut the wealth of the typical billionaire in half after 15 years relative to no wealth tax.
Had the tax been in place since 1982, Forbes 400 wealth would be 40% of its 2018 level, the Forbes 400 share about 1.3% instead of 3.3%, and the top 15 fortunes $196bn instead of $943bn (Warren: 2.0%, $434bn).

Every problem found

1majorconfirmed · newly identified here · overclaiming

The note says the tax would completely close the gap between how fast billionaires' fortunes grow and how fast ordinary families' wealth grows. The authors' own simulation of the tax - the one drawn in the note's figure - shows the richest 400 still pulling ahead, only more slowly: about a third of the gap remains, and more if some evasion is allowed. The tax would shrink the gap a lot, but not eliminate it.

Technical detail and evidence

Why it matters, in detail. Group-level excess growth of the population-normalised Forbes 400 over average wealth falls from 3.53 to 1.14 pts/yr in the zero-evasion Sanders counterfactual (67.5% of the log excess removed); scaling the effective rate by the note's 16% hiding (x0.82) leaves about 1.6 pts/yr. A static comparison of a ~6% average rate with a 4.0-point gap is valid for a fixed set of incumbent fortunes, not for a top group with entry: with finite long-run exposure e_inf the taxed/untaxed ratio converges to (1-tau)^e_inf, so the long-run growth gap is unchanged and only its level shifts. Measured against middle-40% wealth (2.24%/yr vs 2.88% mean) the residual gap is roughly 1.7-2.2 pts/yr.

Evidence. SZ2019BPEAfinal.xlsx, sheet DataFig6 (source of the note's figure; row 2: 'Assumes zero evasion rate'): Forbes 400 share 1982 -> 2018, actual 0.934% -> 3.261% (x3.49; +3.53 pts/yr relative to average wealth); with the Sanders tax 0.890% -> 1.337% (x1.50; +1.14 pts/yr), i.e. 68% of the excess growth removed, not 100% (scripts/02_fig6_checks.py). Sub-periods: 1990-2018 actual +3.07 vs Sanders +0.97 pts/yr; 2002-2018 +2.34 vs +0.98. The note itself says the share 'would only be slightly higher than it was in the early 1980s' (p.2): 1.34% vs 0.93% in 1982 is +44%. With the note's own 16% hiding the simulated 2018 share is roughly 1.57% (residual about +1.6 pts/yr; scripts/12_counterfactual_evasion.py, indicative). Middle class: in the authors' Triumph of Injustice appendix (TA1, equal-split adults), real wealth of the middle 40% grew 2.24%/yr vs 2.88% for the mean, 1982-2018 (scripts/14_middle_class.py).

Why this severity. C4 is one of the two claims in the note's opening sentence. The authors' own simulation shows the gap cut by about two-thirds under perfect enforcement, and by roughly half to three-fifths under the note's own 16% evasion assumption, so the claimed effect is overstated by a third or more; the direction (a large reduction) survives.

The authors' response. None located. The note itself concedes the counterfactual share 'would only be slightly higher than it was in the early 1980s' (p.2) and 'still slightly higher than in early 1980s' (p.6).

2majorlikely · previously known · speculative assumption

The note assumes every rich family hides the same 16% of its wealth whether it pays a 1% or an 8% tax. But the authors arrived at 16% by assuming that hiding grows with the tax rate, and they worked it out for a 2% tax. If hiding grows with the rate, as their own method implies, the richest families - who would pay most of the tax - would hide much more, and first-year revenue would be roughly one-fifth to almost one-third lower.

Technical detail and evidence

Why it matters, in detail. A rate-invariant hiding share h = 0.16 is equivalent to a semi-elasticity e(tau) = 0.16/tau (16 at tau = 1%, 2 at tau = 8%), whereas the calibration source assumes a constant e = 8. With constant e = 8 at each family's marginal rate (q = exp(-e*tau), bunching at kinks) 2019 revenue is $269.8bn (-19%); with the linear 1 - e*tau form implied by the authors' '2% x 8' arithmetic, $238.5bn (-29%). Losses concentrate in the $10bn+ bracket, where the exp form hides 47% of wealth. Combined with mechanical decumulation the 2019-28 total is $3.07T (-29%), and $3.24T / $3.42T on 2020-29 / FY2021-30 windows, within about 2% of the Tax Foundation ($3.26T) and PWBM ($3.35T) conventional scores (scripts/13_windows.py). The authors' best defence - hiding is an enforcement parameter, listed equity cannot be hidden, billionaires face 100% audits - could justify a flat or even declining share at the top, but it is not argued in the note and conflicts with how the 16% was derived; legal margins (gifts below thresholds, foundations, valuation of private stakes, expatriation) plausibly scale with the rate.

Evidence. Adjudicator re-implementation (scripts/04_avoidance.py; results/04_avoidance.log): each family chooses its reported fraction q to minimise T(qW)/W + c(q), with c calibrated so the interior optimum is q = exp(-8*tau) (exponential) or q = 1 - 8*tau (linear) at its marginal rate tau, bunching at thresholds allowed. 2019 revenue $269.8bn (exp, -19.3%) or $238.5bn (linear, -28.7%) vs $334.45bn, identical to both auditors. At $20bn of married wealth q = 0.53 (exp) / 0.44 (linear). Revenue by marginal bracket: 5%-8% brackets $177.6bn (53.1%), 8% bracket $83.3bn (24.9%). Average-rate (rather than marginal-rate) variants give -2% to -16% depending on whether the rate is measured on true or reported wealth. External: PWBM Warren brief (12 Dec 2019, technical appendix D): 'as Max Ghenis notes, the Warren estimate appears to apply the elasticity to a rate of 2 percent for all taxpayers', whereas PWBM computes 'household-specific avoidance responses as a function on the specific marginal wealth tax rate'. Tax Foundation Fiscal Fact 691 (Li & Smith, Jan 2020, p.12) applies the same -8 rate by rate ('about 38 percent' hidden at a 6% rate) and scores the Sanders plan at $3.261T for 2020-29 (Table 4, p.10).

Why this severity. Applying the authors' own elasticity rate by rate, as independent scorers did, lowers the first-year score (C2) and hence the decade total (C1) by 19%-29%, and by about 29% when combined with base erosion (F3). A core number thus rests on an unvalidated and internally inconsistent assumption whose change within a defensible range substantially weakens it; not critical because revenue stays in the trillions.

The authors' response. No response specific to rate scaling located. The authors' general position (BPEA 2019 sec. I.C and III; 'Response to Summers and Sarin', 25 June 2019, pp.2-4) is that avoidance is a policy choice set by enforcement, that 80% of top-0.1% wealth has market prices, and that for the 12 of the 15 richest who hold listed stock 'avoiding the wealth tax is impossible'.

3moderateconfirmed · previously known · speculative assumption

The ten-year figure simply multiplies the first year's revenue by 13, as if the tax would not shrink the fortunes it taxes. The note itself says it would shrink them - by about half for billionaires within 15 years - so revenue from the very richest would fall over time; on standard assumptions that trims the ten-year total by roughly a tenth to a fifth. In fairness, billionaire wealth actually grew far faster after 2019 than the note assumed, which would have offset much of this.

Technical detail and evidence

Why it matters, in detail. With R_t = sum_i T(W_it; z_t) and W_i,t+1 = (W_it - s*T_it)(1+g), the authors' R_t = R_0 (1.055)^t is exact only if the taxed group's after-tax wealth grows like average wealth. With distribution-neutral pre-tax growth and s = 1, 2028 revenue is only 14% above 2019 instead of 62% and the 2019-28 total is $3.56T (-18%); s = 0.75 or CPI indexation gives $3.73T (-14%), fixed nominal thresholds $3.96T (-9%, bracket creep offsets erosion). Erosion is concentrated in billionaires (Forbes 400 revenue -42% by 2028). If top wealth keeps its historical excess growth, the total is $4.41T; realised billionaire wealth growth of ~15%/yr over 2019-2025 suggests the distribution-neutral baseline would have understated the very top, though this is pre-tax and cannot validate a counterfactual.

Evidence. scripts/05_decade.py on the authors' file (per-family wealth +4.5%/yr, families +1%/yr, tax paid out of wealth): thresholds indexed to average wealth (the authors' implicit convention) $3.56T (-18%); only 75% of the tax financed from saving $3.73T (-14%); CPI-indexed thresholds $3.73T (-14%); fixed nominal thresholds (the plan text is silent on indexation) $3.96T (-9%); CBO-consistent 3.84% growth $3.30T (-24%). Forbes 400 revenue in 2028 is 42% below the no-erosion path. Under the authors' 'top grows faster' view (top +2.8 pts/yr, Forbes 400 +4.0) the total is $4.41T (+1%). Both auditors obtained $3.55T for the indexed full-erosion case. Ex post, the authors' own note of 2 March 2026 (on a different Sanders proposal; cached copy) reports US billionaire wealth rising from $3,528bn (2019) to $8,189bn (end-2025), about 15% a year nominal, far above the 5.5% assumed.

Why this severity. Under the distribution-neutral baseline used by conventional scorers the decade total falls 9%-18% (more with CBO growth); under the authors' alternative that top wealth keeps outgrowing average wealth - which realised top-wealth growth after 2019 supports - it is roughly unchanged. A noticeable, assumption-dependent effect on C1, not a transformative one.

The authors' response. Warren letter (18 Jan 2019, p.3 step 6): 'This uniform growth assumption is conservative as the wealth of the rich has grown substantially faster than average in recent decades ... Therefore, even with the wealth tax, it is most likely that top wealth would continue to grow at least as fast as the average.' Not repeated in the Sanders note.

4moderateconfirmed · newly identified here · internal inconsistency

The note's dramatic 'what if the tax had existed since 1982' numbers assume nobody evades, while its revenue estimate assumes everyone hides 16%. With the same 16% hiding, the fall in billionaire fortunes would be noticeably smaller - the top 15 might hold around $250 billion rather than $196 billion - though still very large.

Technical detail and evidence

Why it matters, in detail. In the exposure model the 2018 taxed/untaxed ratio is (1-tau)^e_T = 0.41 at zero evasion (DataFig6). Hiding 16% lowers the effective rate on true wealth by about 18% (static liability ratio 0.818), so ln(ratio) scales to about 0.82 of its value: ratio about 0.48, share about 1.57%, average about $3.5bn, top-15 ratio 0.21 -> 0.27; residual excess growth about 1.6 rather than 1.1 pts/yr. A 15-year simulation of the Forbes 2019 list gives median ratios of 0.536 (h = 0) vs 0.602 (h = 0.16) and aggregate 0.434 vs 0.504. These are approximations; the authors' panel simulation cannot be rerun.

Evidence. Static 2018 Forbes 400 liability $180.0bn at full reporting vs $147.2bn with 16% hidden (ratio 0.82). Scaling the zero-evasion log effect by that ratio (authors' exposure model; indicative, scripts/12): Forbes 400 counterfactual about 48% of actual instead of 41% (about $3.5bn average, 1.57% share instead of 1.34%), top 15 about $254bn instead of $196bn. Typical billionaire (Forbes 2019 list, 607 US citizens; scripts/06): median 15-year wealth ratio 0.54 at zero evasion vs 0.60 with 16% hidden (0.50 vs 0.57 with fixed thresholds and 6.6% growth); aggregate 0.43 vs 0.50. Billionaire rate 5.95% of reported wealth at zero evasion vs 4.86% of true wealth with 16% hidden (scripts/03).

Why this severity. Applying the note's own evasion assumption shrinks the reported deconcentration effects by roughly a sixth to a quarter (for example top-15 wealth about $254bn rather than $196bn); the qualitative conclusion of large deconcentration is unaffected.

The authors' response. BPEA 2019 p.498 notes that the model can be extended to evasion using its sec. I.C framework; no Sanders-specific response located.

5moderatelikely · previously known · selective presentation

The estimate relies partly on a wealth dataset that shows more wealth at the very top than other sources - including a revised version the same authors published the same month. Using those alternatives would lower the revenue by roughly a tenth; the note does not show this uncertainty.

Technical detail and evidence

Why it matters, in detail. The DINA micro-file's tail is fatter than the Saez-Zucman (2016) series benchmarked in BPEA: base above $50m $12.68T vs $10.9T (capitalized) and $9.4T (revised for heterogeneous fixed-income yields and pass-through valuation). Compressing the DINA half's wealth above $50m to those bases lowers 2019 revenue by 6.8% and 12.1%; SCF+Forbes alone gives -9.3%. The 936 non-Forbes families between $1bn and $2.1bn pay $35.5bn (10.6%); a Pareto alpha of 1.4 from the Forbes 400 implies about 730 in that band, so only part of the excess over Forbes's 217 is implausible. DFA- and SZZ-based tails imply larger cuts but rely on contested concepts (households, durables, DB pensions, Financial Accounts private-business values; heterogeneous yields).

Evidence. Base above $50m (true wealth, 2019): DINA+Forbes in the file $12.68T vs BPEA Table 2 (published p.461) $10.9T (capitalized), $9.4T (revised), $9.7T (SCF+Forbes), $6.8T (estates) (scripts/08-10). Public PSZ files: usdina2016 aged to $94T gives $12.07T (top 0.1% 21.4%), usdina2019 $12.76T (22.2%) (scripts/09). Revenue sensitivities on the file: SCF+Forbes only $303.5bn (-9.3%); DINA half compressed to the BPEA capitalized base -6.8%, to the revised base -12.1% (crude proportional compression above $50m); DINA+Forbes only +9.3%. The file contains 1,336 billionaire families (936 between $1bn and $2.1bn) vs 607 US citizens on Forbes's 2019 list and 705 in Wealth-X 2019; reweighting that band to Forbes's count lowers revenue 3.4%-8.2% (scripts/03), but a Pareto extrapolation from the Forbes 400 (alpha about 1.4, the Warren-letter method) implies about 1,130 billionaires, so rich lists likely undercount that band. Fed DFA (2026 vintage): top 0.1% of households $12.8T in 2019:Q2 vs $17.7T for the file's top 128.6k units; Auditor A's DFA-calibrated tail gives -32% (contested concept).

Why this severity. Using the authors' own concurrent bases or the SCF alone lowers C2 (and C1) by about 7%-12%; larger cuts depend on contested alternatives (DFA, Smith-Zidar-Zwick). A noticeable robustness gap and an undisclosed choice rather than a demonstrated error.

The authors' response. BPEA 2019 p.462: 'The tax bases are quite close across the first three sources' ($10.9T, $9.4T, $9.7T above $50m). The authors contest the DFA and Smith-Zidar-Zwick estimates elsewhere; the SCF half is benchmarked down to DINA's lower aggregate, and PWBM notes that DINA's lower total and mid-year valuation partly offset its higher concentration.

6moderateplausible · previously known · speculative assumption

How much of their wealth the rich would hide is the biggest unknown in any wealth-tax estimate. The authors chose a figure from the low-to-middle part of the research range and showed no alternatives; several careful studies find much larger responses, which would cut revenue further - though studies from countries with strong reporting find almost none.

Technical detail and evidence

Why it matters, in detail. Revenue is steeply decreasing in the semi-elasticity: holding the rate-by-rate exponential form, moving e from 8 to 13 lowers 2019 revenue from $269.8bn to $208.4bn (-23% relative to F1's central case; -38% vs the published $334.5bn). The authors' e = 8 is itself driven by the Swiss estimate (without it the average is 1.2), so they can argue the haircut is conservative relative to the Scandinavian evidence; PWBM's -13 also averages weak-enforcement settings. No US implementation evidence exists.

Evidence. scripts/04_avoidance.py: semi-elasticity 13 at marginal rates gives $208.4bn (exp, -37.7%) or $156.1bn (linear, -53.3%) in 2019; a flat hiding share of 26% (13 x 2%) gives about $284bn (-15%; uniform 25% = $283.7bn). Elasticities as tabulated in PWBM's Warren brief (Table A1) and corpus entries optimal-tax-behavior-17/-19/-20; Kopczuk, BPEA Fall 2019 comment, p.516 ('of limited quantitative usefulness ... a severe understatement').

Why this severity. Plausible higher elasticities would lower C1-C2 by a further 15%-25% beyond F1, but the evidence is two-sided and context-specific (third-party-reported Scandinavian settings show near-zero responses), so a bias is plausible but its size is not demonstrated.

The authors' response. Warren letter fn.2: the Swiss estimate 'is extrapolated from very small variations ... not as compellingly identified', and Switzerland lacks third-party reporting. 'Response to Summers and Sarin' (25 June 2019): their benchmark implies 80%-90% avoidance, and with third-party reporting, no exemptions, high audit rates and an exit tax 'it is possible that avoidance would be below 15%'. BPEA 2019: enforcement is a policy choice.

7moderateplausible · previously known · speculative assumption

Almost half the revenue comes from the 400 richest Americans, valued with Forbes magazine's estimates. When these people die, their estates typically report about half what Forbes estimated, partly because of tax planning and perhaps partly because Forbes overestimates. If Forbes overstates what could actually be taxed, revenue would be lower; for fortunes held in publicly traded shares the overstatement is probably small.

Technical detail and evidence

Why it matters, in detail. Because most Forbes 400 wealth sits in the 7%-8% brackets, each 10% overstatement of Forbes values removes about 5% of total revenue. The estate/Forbes gap is not identified: the authors read it as estate-tax avoidance under weak enforcement (which a well-enforced annual tax would curb), critics as Forbes overstatement (which would persist). The corpus reports a later administrative match (Balkir, Saez, Yagan & Zucman 2025) with estate values around 30% of Forbes estimates (not re-verified here).

Evidence. Raub, Johnson & Newcomb (2010), 'A Comparison of Wealth Estimates for America's Wealthiest Decedents Using Tax Data and Data from the Forbes 400', NTA Proceedings 103: 128-135: 'In aggregate, the average Forbes Ratio was 0.50'; valuation discounts 'can be as large as 50 percent'. Revenue with Forbes values x0.85 / x0.7 / x0.5: $309.2bn (-7.6%) / $284.4bn (-15.0%) / $252.4bn (-24.5%) (scripts/08). Forbes 400 share of revenue 44.0%.

Why this severity. A robustness concern of limited demonstrated impact: a 15%-30% overstatement of taxable value by Forbes would lower C2 by 8%-15%, but most top-15 wealth is listed equity with observable prices and the 16% haircut already absorbs part of any shortfall.

The authors' response. 'Response to Summers and Sarin' (25 June 2019, pp.3-5): the Raub et al. gap shows estate-tax dodging ('about 50% of estate taxes owed are avoided'); 12 of the top 15, with 83% of that group's wealth, hold listed stock with 'a well-defined market value'.

8minorconfirmed · newly identified here · documentation gap

The authors posted data and code that exactly reproduce the first-year figures, which is commendable. But the ten-year total and the 'since 1982' simulations rely on steps and data that are not posted or explained in the note, so outsiders cannot fully check them.

Technical detail and evidence

Why it matters, in detail. Reproducibility tiers: (1) the 2019 score is fully reproducible (exact match from wealth.dta + wealth.do; SCF half rebuilt from the Fed extract; DINA half matches public usdina2019 totals); (2) the decade total is reproducible only by importing the Warren-letter rule, with growth, indexation and erosion unstated in the note; (3) the historical simulations reconcile with SZ2019BPEAfinal.xlsx (DataFig6; datatable4 linked-family columns) but cannot be re-run, so their zero-evasion and indexation choices cannot be varied.

Evidence. Fresh download of https://eml.berkeley.edu/~saez/wealthtaxcalc.zip on 25 Oct 2026 (md5 e2139493952e9afb90f4ad6e20da2ce3, Last-Modified 22 Sep 2019; identical to all cached copies). SZ2019BPEAfinal.xlsx, DataFig6 rows 1-3. BITSS-OPA/opa-wealthtax rawdata/wealthtax.do: married collapsed at line 63 but dropped by the group collapse at line 76; SCF part keeps only weight, data and networth (line 51); usdina2019 used (line 61). http://wealthtaxsimulator.org/simulator_app/ returned HTTP 502 on 25 Oct 2026.

Why this severity. Limits independent verification of the decade total and the deconcentration results but does not by itself change any number; the one-year score is unusually transparent and the long-run outputs reconcile with the authors' workbook.

9minorconfirmed · newly identified here · internal inconsistency

The ten-year figure assumes the taxed wealth grows 5.5% a year - faster than the economy was officially projected to grow - and rounds the result up slightly. The authors' own research paper from the same month used a smaller multiplier that would give about $4.0 trillion. These are small effects.

Technical detail and evidence

Why it matters, in detail. 13 vs [(1.055)^10-1]/0.055 = 12.875 adds $42bn (+1%). With 5.5% revenue growth and 3.84% GDP growth, revenue/GDP rises from 1.56% (2019) to about 1.79% (2028), averaging 1.70%. BPEA's 12.0 factor corresponds to about 3.9% growth. The choice of the 2019-28 window does not inflate the total: on later windows the authors' method gives more ($4.84T for 2021-30).

Evidence. scripts/01_reproduce.py: 13 x $334.45bn = $4.348T; 12.875x = $4.306T; 12.0x = $4.013T. CBO 51135-2019-01-economicprojections.xlsx, sheet '2. Calendar Year', row 8: GDP $21,478.1bn (2019) to $30,144.1bn (2028), sum $255.7T. Warren letter p.3 step 6 and fn.4. BPEA conference draft (Sept 2019), Table 2 note.

Why this severity. Shifts C1 by 1%-8%; household wealth has historically grown faster than GDP and PWBM itself assumes 6%, so the growth rate is defensible even if its attribution to CBO is not.

The authors' response. Warren letter p.3 step 6: nominal taxable wealth 'would grow at the same pace as the economy, at 5.5% per year as in standard projections of the Congressional Budget Office or the Joint Committee on Taxation'.

10minorconfirmed · newly identified here · arithmetic or transcription

A small bookkeeping slip: the extra revenue from giving single people lower thresholds is about $9 billion a year, not $11 billion. It does not change the total.

Technical detail and evidence

Why it matters, in detail. Same-file decomposition: S(halved singles) - S(married thresholds) = 334.450 - 325.708 = $8.742bn (6.5% of $134.313bn); (Warren + 1% bracket) - Warren = $18.880bn (14.1%); S(married thresholds) - (Warren + 1% bracket) = $106.691bn (79.4%). The stated $11bn = 335 - 324 includes a dataset difference of about $1.4bn.

Evidence. scripts/01_reproduce.py: married thresholds for all $325.708bn; Sanders $334.450bn; increment $8.742bn; 1% bracket $18.880bn; upper-bracket progressivity $106.691bn; Warren $200.137bn. Both auditors obtained $8.74bn.

Why this severity. A $2.3bn misattribution within the Sanders-Warren decomposition; the totals and the conclusion that higher-bracket progressivity drives most of the difference are unaffected.

11minorconfirmed · newly identified here · arithmetic or transcription

A few numbers in the text do not quite match each other or the authors' related papers (for example 3.5% versus 3.3%). These are small slips that do not affect the conclusions.

Technical detail and evidence

Why it matters, in detail. (a) DataFig6!B44 = 3.261%. (b) Share ratio 3.49 implies 3.53%/yr relative growth, vs (1.066/1.026)^36 = 3.96 or 3.90%/yr; the Warren letter's 4.5 points refers to 1980-2016 (7% vs 2.5%). (c) 5.954% wealth-weighted, all coded married, on Forbes's 2019 list; 'around 6%' holds.

Evidence. DataFig6 B44 = 0.03261; share ratio 1982-2018 of 3.49 vs 3.96 implied by 6.6%/2.6% (scripts/02_fig6_checks.py); BPEA 2019 published p.503 and conference draft; Warren letter p.3 step 6; scripts/03_billionaires.py. The 6.6% itself reproduces (6.59%-6.67%) from Gomez's independent 1982 Forbes panel with BPEA's population normalisation (top 250-265 in 1982; scripts/07_growth_gap.py).

Why this severity. Typos and rounding or definition differences; none changes a conclusion.

12minorconfirmed · newly identified here · data error

There are a few small technical slips in how the data file was assembled; their effects on revenue are tiny and partly cancel out.

Technical detail and evidence

Why it matters, in detail. Top-5 SCF records: exact half-weights 1.34-1.36 vs file weight 2; aggregate +$1.60bn (+0.48%) of 2019 revenue. Grouping in fives lowers revenue by about 3% on the DINA+Forbes subfile because the convex schedule is applied to group means. Unit mixing affects population shares, not revenue.

Evidence. scripts/11_scf_rounding.py rebuilds the SCF half exactly from the Fed's 2016 summary extract (31,240 records, weight sum 64,714,672, revenue $78.161bn) and shows the double rounding adds $1.6bn (+0.5%). Auditor A: grouping lowers revenue (DINA+Forbes $365bn grouped vs $377bn ungrouped), i.e. it is conservative. scripts/01_reproduce.py: liable share of file population 0.1166%.

Why this severity. Effects are small (+0.5%) and partly offset by conservative choices (grouping, Forbes records coded married, SCF benchmarked down to DINA's aggregate).

13minorplausible · previously known · methodology

By shrinking the fortunes of the very rich, the tax would also reduce other taxes they pay, and enforcing it would cost money. The note does not subtract these; they would probably reduce the net gain by a few percent.

Technical detail and evidence

Why it matters, in detail. A cumulative mechanical wealth reduction of about $4T by 2028, at 3%-4% reported yields and 20%-30% average income-tax rates, implies $114-190bn lower income-tax receipts over 2019-28; estate-tax and administrative offsets would add to this. Not re-estimated by the adjudicator.

Evidence. Auditor A's rough calculation (its scripts/16_tax_interactions.py; not re-run here): income-tax offset $114-190bn over 2019-28 (2.6%-4.4% of $4.35T) assuming 3%-4% reported yields and 20%-30% average rates. BPEA 2019 p.498 fn.85. Tax Policy Center (19 Mar 2020) scores the wealth tax at $2.3T for 2021-30 assuming 'The budget of the IRS would not increase above the baseline level'.

Why this severity. Likely a few percent of C1; campaign scores commonly report gross receipts.

The authors' response. BPEA 2019 p.498 fn.85: 'Lower top wealth generates a negative fiscal externality in the public economics jargon'.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

'68% more' is an error because 334.4/200.1 = 67.1%. — With the note's own rounded figures ($335bn vs $200bn) the difference is 135/200 = 67.5%, which rounds to 68%; a rounding convention, not an error.
Footnote 3's '6.1% ... 608 American billionaires' is discrepant with 607 billionaires and 5.95% on Forbes's 2019 list. — The authors' snapshot date and selection are unknown; 5.95% supports the text's 'around 6%'. Auditor B also declined to call this an error. Retained only as a minor note in F11.
The 16% here vs '15%' in the Warren letter, and the Warren rescoring ($200bn vs $210-212bn), are inconsistencies. — The 15%/16% mismatch is internal to the January 2019 Warren letter, not this note; the rescoring of the Warren plan is disclosed in the note's fn.4.
BPEA says fewer than 0.1% of families would be liable for the Warren or Sanders tax, but 182,399 is 0.104% of BPEA's 175m tax units. — A statement in a different work, and 0.104% is 'about 0.1%'; no bearing on this note's findings.
The scoring file's 1,336 billionaire families (about twice Forbes or Wealth-X) are a data error. — Not a processing error: the count reflects the DINA tail, and a Pareto extrapolation from the Forbes 400 with alpha about 1.4 (the authors' Warren-letter method) implies about 1,130 billionaires, so Forbes's 607 likely undercounts the $1-2bn band. Retained only as evidence within F5, with 3%-8% of revenue at stake.
Applying the authors' elasticity to each family's average tax rate cuts revenue by 10%-16% (the lower end of the -10% to -29% range). — Definition-dependent: measuring the average rate on true rather than reported wealth gives only -2% to -5% (scripts/04), and average rates are not the margin for hiding decisions. The adjudication uses the marginal-rate range (-19% to -29%), which both auditors and the Tax Foundation's method support.
Scaling the non-Forbes tail to the Fed DFA's top-0.1% wealth cuts revenue by 32%. — Inputs verified ($17.7T in the file's top 128.6k units vs $12.8T in the DFA, 2019:Q2), but the DFA uses a different unit and wealth concept (households; durables, DB pensions, Financial Accounts business values) that the authors dispute; kept only as a contested lower bound in F5, not as a central estimate.
The gap with independent scorers is explained by internal-consistency corrections, not by different data. — Qualified rather than accepted: the arithmetic match is confirmed (corrected totals within -0.6% of the Tax Foundation and +2.2% of PWBM on their windows), but those models use a different base (SCF+Forbes), elasticity (-13 phased in) and growth (6%), so the match partly reflects offsetting differences. It shows the corrections are of the right size, not that they reproduce those models.
Possible arithmetic error in the note's 6.6% Forbes 400 growth rate, since a literal fixed-400 comparison gives 7.43% (examined by Auditor B, who did not assert an error). — Auditor B itself did not infer an error; the adjudicator confirms that with BPEA's population normalisation (top 250-265 in 1982, Gomez panel, authors' price index) the figure reproduces at 6.59%-6.67%.
Possible inconsistency: $4.35T is 1.70% of decade GDP rather than '1.6% of GDP per year' (examined by Auditor B and not treated as an error). — Auditor B rejected this itself and the adjudicator agrees the 1.6% correctly describes 2019; the gap is retained only as a minor symptom of the above-GDP growth assumption (F9).
Possible circularity: the mechanical Forbes simulations might ignore all consumption, giving and inheritance (allegation examined and rejected by Auditor B). — Examined and rejected by Auditor B, and the adjudicator concurs: BPEA pp.495-496 assumes consumption and giving proportional to wealth and links heirs to parents' exposure; a conditional simulation is not circular. The real limitations are recorded in F2 and F6.

Audit notes

What reproduced. Reproduced exactly with an independent Python scorer on a fresh download of the posted package (wealthtaxcalc.zip, unchanged since 22 Sep 2019): the addendum table (all eight Sanders bracket counts, wealth and tax; total 182,399 families and $334.450bn) and the Warren panel (70,187; $200.137bn); the decomposition (1% bracket $18.9bn, upper progressivity $106.7bn) and the corrected singles increment ($8.74bn, not $11bn); $4.35T as 13 x first-year revenue (exact 5.5% factor $4.31T); 1.56% of CBO's 2019 GDP. The SCF half of the file rebuilds exactly from the Fed's 2016 extract; the DINA half matches the public PSZ usdina2019 totals (Auditor A, consistent with the construction code). Footnote 3's 'around 6%' billionaire rate (5.95% on Forbes's 2019 list) and the 6.6% Forbes growth rate (6.59%-6.67% with BPEA's population normalisation) reproduce approximately. The since-1982 counterfactuals (40%, $3.0bn, 1.3%, 2.0%, $196bn, $434bn) reconcile with the authors' BPEA workbook outputs but could not be re-simulated because the program and Forbes panel are not posted. Not reproduced: the exact 2.6% average family wealth growth (and hence the 4.0-point gap) and the authors' 608-billionaire snapshot. Sensitivities on the authors' file: rate-consistent hiding -19% to -29%; erosion -9% to -18% (about +1% under the authors' 'top grows faster' view); both combined -22% to -29%; alternative bases -12% to +9%; Forbes valuations -8% to -25%; elasticity 13 at marginal rates -38%.

Limits of this audit. No US wealth tax was enacted, so the behavioural assumptions (hiding, real responses) cannot be validated; all sensitivity analyses vary one or two assumptions on the authors' file and are not structural forecasts. The since-1982 Forbes simulation could not be re-run (program and panel not posted), so its evasion-adjusted values are exposure-model approximations. The DINA-compression and billionaire-reweighting scenarios are crude. DFA and WID comparisons use 2026 vintages; the CBO comparison uses the January 2019 projections. The PWBM, Tax Foundation and TPC models were not re-run; the close match with their totals partly reflects offsetting differences. Realised top-wealth growth is taken from the authors' March 2026 note (cached copy; the Senate-hosted original could not be re-downloaded), and Forbes's reported $6.6T 2025 Forbes 400 total could not be verified from the primary page. Constitutionality, administrability, compliance costs, general-equilibrium and asset-price effects were not assessed.