Emmanuel Saez, Gabriel Zucman · Unpublished response note, June 25, 2019

Response to Summers and Sarin, 'A wealth tax presents a revenue estimation puzzle'

12 problems found: 1 major · 6 moderate · 5 minor

What the paper claims

Senator Warren's wealth tax (2% on net worth above $50m plus a 1% surtax above $1bn) would raise $212bn in 2019 and $2.75tn over 2019-2028, paid by about 75,000 families (less than 0.1% of households).
Absent any avoidance the tax would raise $250bn in 2019; this number 'is broadly accepted as being in the right ballpark' and 'not contested by Summers and Sarin'; the dispute is only about avoidance.
Based on 'the four published academic studies of wealth taxes abroad', the rich would shelter 15% of their wealth; the estimate is 'not a best case scenario but is itself a middle-ground', and with Warren's enforcement design avoidance could be below 15%.
Summers and Sarin's $25bn 'amount[s] to assuming that the rich would shelter 90% of their assets' (80% even if the rich were only half as rich), which 'is not a serious revenue estimation'; S&S 'claim that our revenue estimates are exaggerated by a factor of 8'.
S&S's estate-tax benchmark ($10bn collected from estates above $50m at a 40% rate, 1-in-50 mortality, hence $25bn) has three flaws: (i) the effective estate-tax rate is far below 40% (S&S's own $75bn variant); (ii) estate data understate top wealth by a factor of two because of avoidance (top 0.1% below 10% vs 20% elsewhere; Forbes decedents report half of Forbes wealth; 'about 50% of estate taxes owed are avoided'); (iii) estate-tax enforcement collapsed (65% of the 29,000 largest returns audited in 1975 vs 8.6% of 34,000 returns in 2018; revenue relative to household wealth five times higher in the mid-1970s, 'the bulk' due to enforcement), so it is no guide to a well-enforced wealth tax.
Avoidance would be limited because about 80% of top-0.1% wealth is in assets with a clear market price (only about 20% in private businesses); the Forbes 400 alone would pay $82bn (more than 3x S&S's total) and the 15 richest $28bn (more than S&S's total); for the 12 of those 15 whose fortunes are in listed giants 'avoiding the wealth tax is impossible'.

Every problem found

1majorconfirmed · previously known · speculative assumption

The $212 billion figure depends on a guess that the very rich would hide or shelter only 15% of their wealth. That guess came from averaging four foreign studies in a way that leaned toward small responses; for example, it used the small Danish estimate for the moderately rich rather than the much larger one the same study (co-written by Zucman) found for the very rich. Reasonable alternative assumptions give about a quarter less money in the first year, so the estimate is better described as optimistic than as middle-of-the-road.

Technical detail and evidence

Why it matters, in detail. With fixed thresholds and a Pareto tail (a~1.4), revenue scales roughly with (1-h)^a, so the haircut h is first-order: SZ's homogeneous 15% net-worth haircut already lowers revenue 20.5% on their data. Replacing the Danish bunching semi-elasticity (~0.5) by the Danish ceiling-DD estimate (~20.8, PWBM conversion) inside SZ's own four-study average, or using PWBM's avoidance-only -13 scaled to 2%/3% marginal rates, lowers 2019 revenue from $203.5bn to ~$152bn (-25%; -28% vs $212bn); applying SZ's own estimate of estate-tax avoidance (~33%) gives a similar ~-22% on the 2% component. Dropping the Swiss outlier raises revenue ~21%, so the plausible range is ~$123-246bn, with the evidence most relevant to the super-wealthy (DiD, estate data) below $212bn. With a five-year phase-in of the response the ten-year total is ~$2.1tn rather than $2.75tn.

Evidence. Note p.2 para 1, p.6 para 3. Letter (18 Jan 2019) point 2, fn.2, point 4. SZ, 'How would a progressive wealth tax work?' (5 Feb 2019), sec. on avoidance ('Four of these countries have been studied'). Jakobsen, Jakobsen, Kleven & Zucman, NBER WP 24371 (Mar 2018), pp.2-3 (ceiling DD ~30% after 8 years; mechanical effect ~one quarter) and p.13 (bunching caveat). SZ BPEA Fall 2019 pp.471-472. PWBM (12 Dec 2019) Table A1: Seim -0.17, JJKZ bunching -0.22, Londono-Velez & Avila -2.0, Zoutman -11.6, JJKZ DiD -20.8, Duran-Cabre et al. -29.1, Brulhart et al. -40.5; weighted avoidance-only -13. Kopczuk, BPEA comment (2019) pp.1, 3-4 ('a severe understatement'; 26% of base between $50m and $100m). Adjudicator re-scoring of SZ's Sept-2019 simulator microdata (scripts/rescore_simulator.py): uniform 15% $203.5bn; SZ semi-elasticity 8 scaled to 2%/3% marginal rates $190.7bn (-6%); PWBM -13 at 2%/3% (26%/39%) $152.6bn (-25%); JJKZ DiD substituted for JJKZ bunching in SZ's own average $152.0bn (-25%); Advani-Tarrant 7-17 $123-199bn; Swiss study dropped $246-248bn (+21%). SZ's own estate-implied avoidance (~33%, BPEA p.462; base reduction 37.8% above $50m, Table 2) implies a 2% component of ~$140bn vs $181-187bn. Ten-year (scripts/tenyear_and_gifting.py): PWBM-style five-year phase-in $2.10tn vs $2.62tn for uniform 15% on the same data (-20%) and vs the letter's $2.75tn (-24%); 24.9% of the base lies between $50m and $100m.

Why this severity. The headline magnitude rests on one unvalidated parameter; alternatives within the range used by independent scorers and by the authors' own co-authored Danish evidence lower 2019 revenue by about a quarter (to ~$150-165bn), meeting the rubric's 'major' threshold. The qualitative conclusion (revenue far above S&S's $25bn) survives across the whole defensible range, so it is not critical.

The authors' response. Letter fn.2 and BPEA (2019): the Swiss estimate is an outlier, weakly identified and driven by the absence of third-party reporting and by cantonal mobility; avoidance is 'a policy choice' that the Warren design (no exemptions, third-party reporting, audits, exit tax) would keep low. BPEA (pp.471-472) concedes bunching captures responses 'but perhaps not its full magnitude' and that the Danish DiD estimates are 'substantially larger'. No published SZ rejoinder to PWBM located.

2moderateconfirmed · previously known · misrepresentation of source

The note says the rich escape about half of the estate tax, citing two studies. But one of those studies (by the same authors) said the gap was largely due to the rich living longer than assumed and declined to blame avoidance, and the IRS researchers said much of the gap reflects legal differences and generous Forbes estimates. The authors' own later work put estate-tax avoidance at about one-third, not one-half. The note's broader point, that estate-tax data understate how rich the rich are, still stands.

Technical detail and evidence

Why it matters, in detail. Under a Pareto tail (a~1.5) a 1.9x mortality-rate error inflates estate-based top shares by 1.9^(2/3)~1.53 (SZ BPEA), taking the top-0.1% share from ~10% to ~15%; the residual relative to the ~20-22% capitalized share implies ~33% (31.6-37.8%) estate-tax avoidance/evasion, about two-thirds of the note's 50%. For S&S's calculation the deduction (x3) and mortality (x1.9) corrections scale $25bn to ~$140bn for the 2% component, still ~25% below SZ's $187bn, consistent with ~33% estate-type avoidance rather than 50%. If part of RJN's 0.50 ratio reflects Forbes generosity, debt or family attribution (as RJN and Kopczuk argue), the note's Forbes-based figures and the surtax base are upper bounds.

Evidence. Note pp.4-5, fn.3-4. SZ (2016) QJE 131(2) sec. VII.B, pp.570-572 ('We leave to future research the difficult task of identifying what part of this phenomenon owes to rising mortality differential by lifetime income, to rising estate tax avoidance, and to changes in how the approach of death affects real wealth', p.572). Raub, Johnson & Newcomb (2010) NTA Proceedings pp.130-134: mean ratio 0.50 for 181 decedents listed at death; married 0.46 to 0.53 with jointly held assets; post-2004 subgroup 0.47 to 0.54 with discounts and 0.58 with discounts and spousal wealth; p.134 conclusion. SZ BPEA Fall 2019 pp.456-457 (mortality factor 1.9; estate-based top-0.1% share ~10% to ~15.3%; 'close about half of the gap'), p.462 ('approximately 33 percent'), Table 2 (31.6-37.8%), pp.462-463 (S&S reconciled by deductions x3 and mortality x1.9 = 5.7).

Why this severity. The conclusion the note draws (estate data understate top wealth, so S&S's base is too low) survives because the mortality bias undermines S&S's 1-in-50 multiplier just as avoidance would; what fails is the stated mechanism and the 50% figure, which weakens a secondary argument rather than the core rebuttal.

The authors' response. Implicitly revised in SZ BPEA (Fall 2019, pp.456-463), which attributes about half of the gap to mortality and puts estate-tax avoidance/evasion at ~33%.

3moderateconfirmed · newly identified here · selective presentation

The note says audits of big estates collapsed from 65% to under 9%. But it compares the biggest estates in 1974 with all estates in 2017, most of which were small. Comparing like with like, audits of large estates fell from about 65% to about 19-31%, a real drop but a much smaller one. And the note gives no evidence that weaker enforcement, rather than law changes such as letting spouses inherit tax-free, explains most of the fall in estate-tax revenue.

Technical detail and evidence

Why it matters, in detail. Audit coverage fell 22.7% to 8.6% for all returns (x2.6) and, for large estates, 64.8% to 18.7% (>= $5m, x3.5) or 31.0% (>= $10m, x2.1); the 1974 >= $300k class is about $1.5m in 2017 dollars, i.e. poorer in real terms than the 2017 >= $5m class. The note's pairing implies x7.5. Revenue relative to household wealth fell ~4x (5x only with FY1977 or relative to top-0.1% wealth). With spousal and charitable deductions now removing ~57% of gross estates >= $50m, a large part of the long-run decline reflects legal design (deferral through the unlimited marital deduction, exemptions, rates), which S&S already partly correct for in their $75bn and which cannot also be counted as weaker enforcement.

Evidence. IRS, Annual Report of the Commissioner 1975, Table 2, p.89 (scan): gross estates >= $300,000, 18,834 of 29,069 examined (64.8%); all estates 48,948 of 215,537 (22.7%). IRS Data Book FY2018 Table 9a: all estate returns 2,898 of 33,690 (8.6%); under $5m 566 of 21,223 (2.7%); $5-10m 12.6%; $10m or more 1,283 of 4,143 (31.0%); $5m or more 2,332 of 12,467 (18.7%). OMB Historical Table 2.5 and Fed Z.1 B.101 (scripts/estate_revenue_wealth.py): estate and gift receipts relative to prior-year household net worth, FY1974-76 average vs FY2017 3.8x, vs FY2018 4.2x; FY1977 vs FY2018 5.1x. SZ BPEA p.457 (0.7% vs 0.13% of top-0.1% wealth, 5.4x; attribution sentence). SOI 2017, gross estates >= $50m: spousal bequests $25.8bn (37.6% of gross estate), charitable $13.0bn (18.9%).

Why this severity. The qualitative point (enforcement intensity fell and the estate tax is a weak guide to a better-enforced annual tax) survives, but the evidence offered for what the note calls its 'fundamental' objection is overstated by a factor of roughly 2-3.5 and the causal claim is unsupported; this weakens a supporting argument, not the headline.

The authors' response. SZ BPEA (Fall 2019, p.457) repeats the 65%/8.6% comparison but attributes the 1976-2017 difference 'primarily [to] fewer deductions (especially no marital deduction), higher rates, and better enforcement'.

4moderateconfirmed · previously known · misrepresentation of source

The note answers the critics' lowest number ($25 billion) as if it were their forecast, even though the critics had said it was probably too low and had offered $75 billion as an optimistic alternative. It also says the critics assumed the rich cannot be taxed, which they did not say. The note's central point, that the critics' method implies an implausible amount of avoidance, still holds, but the gap between the two sides was smaller than the note suggests.

Technical detail and evidence

Why it matters, in detail. Relative to SZ's no-avoidance 2%-tax revenue ($228bn on their data), $25bn implies an 89% revenue loss and $75bn a 67% loss (54% net-worth haircut), both far above SZ's ~20% revenue loss; the 15%-vs-90% contrast thus compares SZ's assumption with the critics' explicitly labelled lower bound. 'Refuse to engage' and 'premise that the rich cannot be taxed' are contradicted by the op-ed text, which anticipated the design-purity reply.

Evidence. Note p.1 para 1; p.2 paras 1-3; p.4 para 3 (acknowledges $75bn); pp.5-6. Summers & Sarin, Washington Post, 4 Apr 2019 (paragraphs beginning 'We recognize', 'In response to this', 'We would be surprised'). Summers & Sarin, 'Be very skeptical about how much revenue Elizabeth Warren's wealth tax could generate', Washington Post, 28 Jun 2019 (syndicated 2 Jul 2019). scripts/rescore_simulator.py: S&S's $75bn implies a 67% loss of SZ's no-avoidance 2%-tax revenue (a 54% net-worth haircut), not 90%.

Why this severity. The arithmetic about the $25bn is right, but the note presents the critics' self-described lower bound as their estimate and omits their caveat (an important undisclosed caveat), overstating the distance between the positions (S&S implied SZ were too high by a factor of ~2.5 to 8, not 8). The rebuttal still holds against $75bn, so the core claim is not overturned.

5moderateconfirmed · newly identified here · selective presentation

The note's striking claim that just 15 people would pay more than the critics' total for all 75,000 families mixes two different taxes: the critics' number was for the basic 2% tax, while the note's number adds the extra 1% charged on billionaires and assumes no avoidance at all. Compared fairly, the 15 richest would pay about three-quarters of the critics' total and the 400 richest about twice it: still a strong point, but less dramatic than stated.

Technical detail and evidence

Why it matters, in detail. Like-for-like (2% rate, no surtax): Forbes 400 $57.4bn (2.3x S&S) without avoidance or $48.8bn (2.0x) with SZ's own 15% haircut; top 15 $18.8bn (0.75x) or $16.0bn (0.64x). Adding the surtax to S&S's frame would scale their figure by ~212/187 to ~$28bn, so the top-15 'more than' claim fails on any consistent basis.

Evidence. S&S (4 Apr 2019): 'a 2 percent wealth tax would raise a total of $25 billion. That's around one-eighth of the Saez and Zucman estimate'; '$75 billion, or approximately 40 percent' (75/187 = 40%). Forbes 400 2018 (API extract, 400 entries, $2,891.7bn) and SZ simulator Forbes records (scripts/forbes_checks.py, rescore_simulator.py): Forbes 400 2%+1% with no avoidance $82.35bn; 2% only $57.4bn; 2% only with 15% haircut $48.8bn; 2%+1% with haircut $69.3bn. Top 15 ($942.5bn): $28.11bn; 2% only $18.8bn; with haircut $16.0bn; 2%+1% with haircut $23.9bn. SZ BPEA p.462 later restates the point like-for-like (a $2.9tn Forbes base is 'more than twice' S&S's $1.25tn base).

Why this severity. One of the note's two headline illustrations is false on a consistent basis and the other is overstated by about a third; the point they illustrate (S&S's total is implausibly low given how much wealth sits at the very top) survives, so only a secondary result is affected.

The authors' response. SZ BPEA (Fall 2019, p.462) restates the comparison as a tax base ('more than twice' S&S's), which is like-for-like; the June note was not corrected.

6moderatelikely · previously known · overclaiming

The note says four-fifths of the richest families' wealth has an obvious market price, so it would be hard to under-value. That figure comes from the authors' own estimation method; the Federal Reserve's wealth survey and the estate-tax returns of very large estates both suggest that two-fifths to a half is in private businesses with no market price. Even owners of listed shares can shrink their taxable wealth in other ways, such as giving shares to children or to their own foundations.

Technical detail and evidence

Why it matters, in detail. The hard-to-value share is ~20% in SZ's capitalized series but ~40-50% in the SCF and estate data. The SCF 'business' concept (self-reported; may include holding companies with marketable assets) and estate valuations (discounted; decedents older than the living rich) are not identical concepts, so the true share is uncertain, but the 80% is source-dependent rather than a fact. Roughly doubling the base exposed to appraisal and discount games strengthens the case for a larger avoidance parameter (F1); independently, ~25% of the base lies between $50m and $100m and is exposed to gifting whatever the valuation regime.

Evidence. Note p.3 and fn.1; pp.2-3 (top 15). scripts/scf_top_portfolio.py: SCF 2016 top 0.1% (125,615 households): privately held business 50.4% of assets (active 42.4%, non-active 8.0%); with the 2016 Forbes 400 ($2.4tn) added at zero private business 42.5%, at 20% 45.6%. scripts/soi_estate_50m.py (SOI, Estate Tax Returns Filed in 2017, Table 1, gross estate >= $50m, 423 returns, $68.66bn): closely held stock 17.3%, real-estate partnerships 6.2%, other LPs 5.3%, other noncorporate business 5.3%, farm 1.4%, PE/hedge funds 3.5% (39.0%) vs publicly traded stock 22.3%. scripts/fn1_check.py: PSZ 2019 vintage, 2016 top 0.1%: fixed income 7.24 of 18.38 points (39%); PSZ Sept-2020 vintage rule gives 24.2%. SEC Ex. 99.1 (13 Nov 2005): Georgia-Pacific 'to become a privately held, wholly owned subsidiary of Koch Industries'. SZ Feb-2019 note sec.3.6 itself anticipates accelerated charitable giving.

Why this severity. This is the note's main argument that valuation avoidance would be small, i.e. support for the 15% assumption; other credible sources roughly double the hard-to-value share, but concepts differ and the revenue consequence is not quantified, so it weakens a supporting argument rather than overturning a headline number.

The authors' response. SZ (Feb 2019 note; BPEA 2019): private businesses can be valued by formula or third parties, financial institutions could be required to report valuations, and the Warren design allows no discounts; foundations controlled by the wealthy should be taxable.

7moderatelikely · previously known · overclaiming

The note says its estimate of what the tax would raise with no avoidance ($250 billion) is widely accepted. In fact, how much wealth the richest families hold is itself debated: one of the two data sources the authors averaged gives notably higher numbers than the other, and later work by other economists suggested the taxable base could be substantially smaller. The figure is reasonable, but less settled than the note implies.

Technical detail and evidence

Why it matters, in detail. On the letter's own inputs an SCF-only base gives a 2% component of ~$164bn and a DINA-only base ~$210bn (average $187bn), so source choice alone moves revenue by about +/-12%; SZ's non-estate BPEA bases span $9.4-10.9tn. Heterogeneous-returns estimates (SZZ 2019-2023) cut the 2016 top-0.1% share from 18.6% (PSZ) to 15.7% and in their 2019-20 drafts implied ~40% lower revenue; SZ dispute those yields and later report the series 'reconciled'. 'Broadly accepted' hides this two-sided but material uncertainty.

Evidence. Note pp.1-2. Letter point 3 (SCF $8.2tn and DINA $10.5tn, 'fairly close', averaged). Adjudicator reproductions: SCF base $7.94-8.21tn (scripts/scf_base_reproduction.py); DINA base $9.91tn, 78,153 filers (scripts/dina_base_reproduction.py). SZ BPEA Fall 2019 Table 2: base above $50m $10.9tn (SZ 2016 capitalized), $9.4tn (revised capitalized), $9.7tn (SCF + Forbes). PWBM (12 Dec 2019) sec. D: DINA top-0.1% share 'the highest'. SZ, 'Comments on Smith, Zidar and Zwick (2020)', 14 May 2020, p.1: a 2%/3% tax would have raised $202bn in 2016 on SCF + Forbes but 'only $117 billion' on SZZ's estimates.

Why this severity. Base uncertainty passes one-for-one (or more, with fixed thresholds) into revenue, and the note presents a contested input as consensus; but the uncertainty demonstrable in June 2019 was moderate (the two averaged sources differ by ~25%) and the larger later downward revisions are themselves disputed, so this is a robustness caveat rather than a demonstrated error.

The authors' response. SZ (May 2020 comments on SZZ; NBER w27921 2020; w30396 2022): SZZ's top yields are too high and dividend-based capitalization misses billionaire equity; after corrections the series are reconciled. The letter calls the SCF and DINA bases 'fairly close'.

8minorconfirmed · newly identified here · arithmetic or transcription

The note mixes two different percentages (the share of wealth hidden and the share of tax revenue lost) and gets the 'half as rich' calculation wrong (about 70%, not 80%). Its basic point, that the critics' $25 billion implies an enormous amount of hiding, still holds.

Technical detail and evidence

Why it matters, in detail. The only substantive implication is in the half-as-rich robustness check: halving the base would by itself reconcile SZ's 15% assumption with S&S's 'maximally optimistic' $75bn, which the note's framing obscures; for the $25bn the implied avoidance stays very large.

Evidence. Note p.2 paras 1-2. scripts/rescore_simulator.py (SZ Sept-2019 microdata): no avoidance $255.9bn (2% only $228.0bn); 15% haircut = -20.5% revenue; $25bn = 89.0% revenue loss / 78.3% haircut; $75bn = 67.1% / 54.3%; half as rich: 2% only $85.7bn; $25bn = 70.8% / 56.6%; $75bn = 12.5% / 8.7%. Pareto illustration (a = 1.4): 0.85^1.4 = 0.797. SZ BPEA p.462 later gives 'around 85 percent'.

Why this severity. With correct arithmetic the contrast survives (about 20% vs 89% of revenue, or 15% vs 78% of wealth; ~71% rather than 80% in the half-as-rich case); conclusions about the $25bn are unaffected.

The authors' response. SZ BPEA (Fall 2019, p.462) restates the implied avoidance as 'around 85 percent'.

9minorconfirmed · newly identified here · arithmetic or transcription

A technical slip in estimating how much wealth billionaires just below the Forbes 400 list hold means the extra 1% billionaire tax was overstated by about $3 billion a year, which is small compared with the total.

Technical detail and evidence

Why it matters, in detail. Correct B(1) = N(>1)/(a-1) ~ $2.20tn vs the letter's ~$2.55tn; surtax $22.0bn vs $25bn; with the $187bn 2% component the total is ~$209bn. Together with a DINA base that reproduces ~6% below the letter's, SZ's own later data give $203.5bn (-4% vs $212bn).

Evidence. Letter point 4 and fn.3. scripts/forbes_checks.py: a = 1.412 (6.12/1.785) or 1.419 (6.1/1.8); Forbes excess above $1bn $2,040-2,048bn; full Pareto excess $2,200-2,201bn; non-Forbes increment $153-160bn; surtax $22.0bn. scripts/rescore_simulator.py: surtax $22.03bn at the 15% haircut; total $203.5bn.

Why this severity. Lowers 2019 revenue by about $3bn (1.4%) and the ten-year total by about $39bn; negligible next to the avoidance uncertainty.

10minorconfirmed · newly identified here · misrepresentation of source

The letter said its growth assumption matched official budget projections; it did not, because official projections were slower. Because of how the tax thresholds work this may not change the ten-year total much, but the stated justification was wrong.

Technical detail and evidence

Why it matters, in detail. Multiplier 13 vs 11.9 (revenue growing with nominal GDP) or ~12.9 (wealth growing with GDP, thresholds fixed in nominal terms, Pareto a~1.4); SZ's own 12.0 implies $2.54tn. The letter's further claim that top wealth will keep outgrowing average wealth despite a 2-3% annual tax rests on 1980-2016 history and assumes no behavioural accumulation response (see F1).

Evidence. Letter point 6 and fn.4. CBO, 51135-2019-01-economicprojections.xlsx, sheet '2. Calendar Year', row 7: GDP $21,478bn (2019) to $30,144bn (2028), annual growth 3.67-4.75%; sum of GDP 2019-2028 / GDP 2019 = 11.91. SZ BPEA conference draft, Table 2 note ('multiply by 12.0 to get the standard 10-year projection'). scripts/multiplier_fixed_threshold.py: on SZ's data a 3.84% wealth growth with fixed nominal thresholds gives a 12.87x multiplier (14.46x at 5.5%).

Why this severity. A misattributed basis for a disclosed assumption; depending on whether thresholds are indexed, the ten-year total is either ~8% too high (if revenue grows with GDP) or roughly unaffected.

The authors' response. SZ's BPEA conference draft (Sept 2019) uses a 12.0 ten-year multiplier; no explicit correction of the letter located.

11minorconfirmed · newly identified here · external inconsistency

The note says European wealth taxes typically started at around $1 million; most started far lower. This does not help the authors' case (if anything their point is stronger), but it is inaccurate.

Technical detail and evidence

Why it matters, in detail. European thresholds were mostly two orders of magnitude below $50m; the larger true contrast also means the populations behind the European elasticities in F1 were far less wealthy than Warren's base, which weakens rather than strengthens their transferability.

Evidence. OECD (2018), The Role and Design of Net Wealth Taxes in the OECD, Table 4.2, p.81 (single taxpayer, EUR): France 1,300,000; Spain 700,000; Iceland 473,248; Finland 250,000; Sweden 166,214; Norway 157,833; Netherlands 90,756; Ireland 88,882; Switzerland (Zurich) 67,550; Germany 61,355; Austria and Luxembourg no threshold.

Why this severity. A factual misstatement in a contextual argument; it cuts against, not for, the authors' point and has no revenue implication.

12minorconfirmed · newly identified here · documentation gap

A few footnotes cite the wrong source year, describe the population slightly differently from the data, or overstate a comparison number. None of this changes the argument.

Technical detail and evidence

Why it matters, in detail. None of these changes the revenue estimate: the fn.1 decomposition matches a later PSZ vintage within 0.2 points, durables are small at the top (SCF top 0.1%: ~1.8% of assets), and the estate-vs-capitalized contrast is real even if the SCF comparator is ~15-17% rather than 20%.

Evidence. scripts/fn1_check.py (PSZ 2019 vintage, 2016 top 0.1%: 18.38 / 9.60 / 6.98 / 1.80 vs note 18.6 / 9.6 / 7.1 / 1.9). Fed Z.1 B.101 (Sept-2021 vintage), 2016Q4: consumer durables $5.16tn of net worth $94.74tn (5.4%). SZ BPEA data workbook, sheet 'Estates-SCF': 'Top 0.1% SCF (adjusted)' 13.9% (2004), 15.3% (2007), 15.6% (2010), 17.2% (2013); 'Top 0.1% estates' 8-10% in 2004-2012 (12.0% in 2011).

Why this severity. Documentation and labelling slips with no bearing on the conclusions; the fn.2 overstatement even makes the authors' exclusion of these assets look more conservative than it is.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

The note's wealth-to-revenue equivalence means the illustrative $250bn baseline 'becomes $199.125bn rather than $212.5bn' at a 15% wealth haircut; rated moderate (Auditor B, I2). — Partly rejected. The $212bn was computed correctly in the January letter with the haircut applied to net worth before exemptions (as B acknowledges); the $250bn in the note is the linear back-out 212/0.85 and is, if anything, slightly too low (SZ's own data give $255.9bn), which is conservative for the note's argument. The metric conflation is real but does not change any conclusion, so it is retained as minor (F8), not moderate.
fn.1 arithmetic is wrong: (1.9 + 0.3 x 7.1)/18.6 = 21.7%, not 21.5% (Auditor A, I14). — Within rounding of the inputs: the unrounded PSZ 2019-vintage values give 21.2%, so the printed 21.5% is consistent with the source. Not an error.
Because Gates held only 1.34% of Microsoft (~$10bn of $97bn), the '83%' overstates the market-priced share of top-15 wealth (upper bound ~74%) (Auditor A, I15). — The 83% is a classification of people by principal wealth source, as both auditors note, not an asset-level measure, and no evidence was assembled that Gates's non-Microsoft wealth lacks market prices. The Gates holding is confirmed (Microsoft DEF 14A 2018) but does not establish an error; the Koch/Bloomberg part of the claim is retained within F2.
Using 15% rather than the computed 16% is one of the upward-tilting discretionary choices (+1.7%) (Auditor A, I7). — Trivial as a standalone point (about $3bn); mentioned within F1 only as part of the overall direction of discretionary choices.
SZ's own sources span $6.8-10.9tn for the base above $50m, a 38% spread, which shows base uncertainty (Auditor A, I6). — Partly rejected. The $6.8tn is SZ's mortality-corrected estate-based figure whose shortfall SZ attribute to estate-tax avoidance, i.e. the quantity in dispute, not an independent measure of the true base. The relevant spread among non-estate sources is $9.4-10.9tn (BPEA Table 2) or $8.2-10.5tn after the haircut (letter). F3 is retained on those grounds and on the later SZZ estimates.
'Five times more revenue relative to household wealth' is really about four times (Auditor A, I5). — Partly rejected as a standalone error: with the note's vague endpoints ('mid-1970s', 'today') the ratio is 3.8-4.2x for FY1974-76 vs FY2017-18 but 5.1x for FY1977 vs FY2018 and 5.4x relative to top-0.1% wealth (SZ BPEA). It is an imprecision, kept only as context in F5.
With CBO-consistent growth the ten-year total would be $2.52tn instead of $2.75tn (-8%) (Auditor A, I10). — Magnitude not established: with unindexed nominal thresholds the base grows faster than wealth, and on SZ's data CBO-pace (3.84%) wealth growth still yields a 12.9x multiplier. The misattribution to CBO/JCT is upheld (F10, minor); the -8% is only one of two plausible readings.
The 30% unlisted-equity rule understates S-corporations at the top because Z.1 S-corporation equity is ~2.4x the S-corporation wealth in the DINA microfile (Auditor A, I4). — Not independently verified by the adjudicator and not needed for F2, which rests on the SCF and estate-return evidence; set aside.
'The four published academic studies' is a standalone misrepresentation because only one was journal-published by June 2019 (Auditor A, I12). — Facts confirmed (only Seim 2017 was published; JJKZ appeared online in the QJE on 9 Oct 2019; the other two were working papers), but the 'published' wording alone is immaterial since all four were public. The substantive problem, that 'the four' is an incomplete and selective set, is folded into F1.

Audit notes

What reproduced. Reproduced: (1) the letter's arithmetic ($187bn + $25bn = $212bn; 13 x $212bn = $2,756bn; (1.055^10 - 1)/0.055 = 12.9); (2) the SCF base above $50m ($8.2tn, 72,000 households): $7.94-8.21tn and 72,084-76,850 households from the public SCF 2016 extract, depending on whether the Forbes 400 sits inside the $94tn aggregate; (3) the DINA base ($10.5tn, 78,000 filers): $9.91tn and 78,153 filers from the cached DINA 2016 file (base -6%, vintage probably differs); (4) on SZ's own Sept-2019 simulator microdata: $203.5bn at the 15% haircut (2% part $181.5bn, surtax $22.0bn, 72,078 taxpayers) and $255.9bn with no avoidance, so the headline reproduces to -4% and the $250bn to +2%; (5) Forbes 400 2018 ($2,891.7bn; $82.35bn) and top 15 ($942.5bn; $28.11bn; 12 of 15 = 83.2%); (6) S&S's inputs (SOI 2017: $9.32bn net estate tax on $68.66bn gross estate >= $50m, 13.6%); (7) the audit rates (64.8% and 8.6%); (8) fn.1's composition (PSZ 2019 vintage within 0.2 points); (9) the '90%' as a revenue-loss share (89.0% for the 2% tax; 90.2% against the full Warren no-avoidance revenue). Not reproduced or discrepant: the billionaire surtax ($22.0bn, not $25bn); 'half as rich implies 80%' (~71% revenue loss); 'five times' (3.8-4.2x on FY1974-76 vs FY2017-18 endpoints); '80% market-priced' (SCF ~50% and estate returns 39% private business); 5.5% growth 'as in CBO/JCT' (CBO: 3.84%). Sensitivity (not reproduction): PWBM -13 scaled to marginal rates $152.6bn; Danish DiD in SZ's average $152.0bn; Advani-Tarrant 7-17 $123-199bn; Swiss study dropped $246-248bn; PWBM-style phase-in ten-year ~$2.1tn.

Limits of this audit. SZ's January-2019 computation files are unpublished; reproduction used the public SCF 2016 summary extract (restated in later-year dollars and rescaled to the published $86.9tn), a cached DINA 2016 file from a July-2019 archive (vintage may differ from the one used in January), and SZ's own Sept-2019 simulator microdata. No US wealth tax exists, so the true avoidance rate cannot be adjudicated; sensitivity scenarios use SZ's homogeneous-haircut convention and simple rate scaling, not a behavioural model. The Zoutman, Duran-Cabre et al. and Danish DiD semi-elasticities are taken from PWBM's Table A1, not re-derived; the Danish DiD includes real saving responses as well as avoidance. Summers and Sarin's rejoinder was read in a syndicated copy. The portfolios of Koch Industries, Bloomberg LP and Gates's holding companies were not inventoried. Statutory facts (exemption amounts, the 1976 top rate, the 1981 marital deduction) were not re-verified from statute, and whether Warren's thresholds were to be indexed was not checked. Later literature (PWBM Dec 2019, SZ BPEA Fall 2019, Brulhart et al. 2022, Advani-Tarrant 2021, SZZ 2019-2023) is distinguished from evidence available on 25 June 2019; the Danish DiD results, Zoutman's study, SZ (2016) and Raub et al. (2010) were all available then.