Jasper Boll, Emmanuel Saez, Gabriel Zucman · NBER WP 35218 (May 2026, revised August 2026); forthcoming in Tax Policy and the Economy, vol. 41 (NBER/University of Chicago Press, 2027)

California Billionaires: Wealth, Taxes, and Wealth Tax Revenue Estimates

16 problems found: 4 moderate · 12 minor

What the paper claims

California billionaires' wealth is very large and has grown very fast in recent years.
Over the long run, the real wealth of California's 'billionaire class' (top 0.001% of families) grew vastly faster than average family income.
California billionaires pay little California income tax relative to their wealth and wealth gains.
The top 4 pay even less relative to their company wealth, and all their taxes are low relative to economic income.
Counting all taxes at all levels of government, billionaires pay little relative to wealth and wealth gains.
The one-off 5% California billionaire tax (payable over 5 years) could raise about $100B, with comparatively minor income-tax losses.
A permanent annual wealth tax on California billionaires could raise substantial revenue even after mobility-induced income-tax losses.

Every problem found

1moderateconfirmed · previously known · methodology

The paper argues Forbes misses hundreds of smaller billionaires, which would push the tax take up to $115-128 billion. That estimate comes from a curve-fitting exercise that gives wildly different answers depending on the year and on whether the very richest people are included. A reconstruction from state tax data finds far fewer missing billionaires, adding about $5 billion rather than $24 billion.

Technical detail and evidence

Why it matters, in detail. Replacing +28.2% missing wealth with Jaros & Rauh's +6.1% (same 20% evasion) moves scenario 2 from $127.9B to about $109B and scenario 4 from $114.8B (or $112.8B after F10) to about $94B; the missing-wealth offset to departures falls from about 2 times (in wealth) to about 0.4 times. The method projects counts from a mean-excess b dominated by the extreme top, a mismatch; count-based estimators roughly halve the gap, and the 2022 placebo shows the method can imply implausibly large undercounts.

Evidence. BSZ pp.22-23/PDF 24-25; App. Fig. A4 (PDF 53); Table 5 rows 2 and 4 (PDF 39). Workbook Pareto-missing (D23 = 4.479, H5 = 617.1, I23 = 0.2819) and data_rtb_20260921_2023 (31 Dec 2022 and 21 Sep 2026 snapshots). scripts/pareto_check.py. Jaros & Rauh (2026) App. F, pp.46-48.

Why this severity. Affects only the upside scenarios (rows 2 and 4) and the paper's suggestion that the Forbes-based benchmark is conservative; the $104B benchmark itself is unaffected.

The authors' response. The paper notes that the FTB has no direct way to identify these billionaires, applies a higher 20% evasion rate to them and keeps them out of the benchmark (pp.22-23).

2moderateconfirmed · newly identified here · methodology

The '25 times richer since 1982' figure starts in a year when stock prices were unusually low and California's top fortunes were measured from a very small group; starting a year later gives about 15 times. Either way, top fortunes grew far faster than ordinary incomes, but the exact multiple is less solid than the headline suggests, and earlier versions of the paper gave 30 and 38.

Technical detail and evidence

Why it matters, in detail. Multiples to July 2026: x24.6 (1982), x15.4 (1983), x15.2 (1984), x17.1 (1985), x14.6 (1986); the observed fixed-size top-0.0002% group is x36.9 from 1982 but x23.4 from 1983. The growth gap over AGI per family falls from 6.1 to about 5.0 pp/yr. The time-varying Pareto b lowers rather than raises the multiple (x36.9 for the observed group), so the extrapolation does not create the trend; the problem is that the start is an outlier low and the end (1 July 2026) a market high.

Evidence. Workbook longrunseries columns AP, AQ, AT, BU, BV (scripts/longrun.py); NBER rev0 (May) and rev1 (July) abstracts and Table 1B; BSZ pp.5-7/PDF 7-9, fn 9, Table 1B (PDF 35).

Why this severity. The qualitative contrast (billionaire-class wealth grew an order of magnitude faster than average income, which grew x1.6-2.0) holds for every base year, but the headline multiple falls by about a third (x25 to x15-17) and the annual growth rate from 7.6% to 6.6-6.9% with any other early-1980s base.

The authors' response. None specific. The paper notes stock-market peaks and says 'the underlying trend is clearly upward' (p.7/PDF 9); 1982 is the first Forbes 400 year.

3moderatelikely · previously known · speculative assumption

The $100 billion figure is 5% of what Forbes says California's billionaires own, minus a 10% allowance for avoidance and departures. But a few people own a large share of that wealth: Larry Page alone holds about 13%, and the paper concedes he may have left the state in time. If Page and Sergey Brin both escaped the tax, the yield would be about $77 billion, so the $100 billion is a reasonable central case rather than a firm number.

Technical detail and evidence

Why it matters, in detail. Revenue is linear in the base: R = 0.05(1-h)W with W = $2,307B, so each percentage point of haircut costs $1.15B. Page alone is 12.8% of W, 2.6 times the 5% mobility allowance, so the benchmark implicitly treats Page and Brin (24.6% of W together) as taxable residents. Scores: Page (+Thiel, Hankey, Kalanick) out $88.8B (-14%); Page and Brin out $76.6B (-26%); haircuts of 20-30% give $81-92B; private-wealth valuation shortfalls of 20-30% cost only $3-7B because 60-65% of the base is listed stock valued at market. Combined with the weak upside from 'missing' billionaires (F3), a balanced central range is roughly $70-110B before any litigation discount, versus 'about $100B'.

Evidence. BSZ p.21/PDF 23 (10% assumption from Galle et al. 2026a); p.22/PDF 24 (half mobility); pp.23-24/PDF 25-26 (leavers; Page 'likely the only one among the top 4 who could possibly successfully argue he left'); Table 5 (PDF 39). Workbook Tab5!B20-B29 (Page 294.6, Brin 271.7, Zuckerberg 210.3, Huang 170.9; total 2,307). Tracker file rtb_ca_movers_out.csv: Page to Florida and Brin to Nevada dated 2026-10-11 (still taxable under the paper's 1 Jan rule; tracker 24 Oct: $2,322B, $104.5B). Jaros & Rauh (2026, NBER w35899) Sec. 4.3, Table 7. Balkir, Saez, Yagan & Zucman (2025, NBER w34170), 'Effective estate tax rates' paragraph. Initiative 25-0024A1 Sec. 50303(c)(3)(E) (book value + 7.5 x profits), (c)(10) (funding-round floor), Sec. 50304 (ODA). Adjudicator recomputation (scripts/c6_scoring.py): Page+Thiel+Hankey+Kalanick out $88.8B; also Brin out $76.6B; uniform haircut 20%/30% gives $92.3B/$80.7B; 20%/30% valuation shortfall on the 40% non-public wealth with 5% mobility gives $100.8B/$96.4B.

Why this severity. The score reproduces exactly and is explicitly conditional, and the paper shows an $89-128B range. Plausible alternatives (Page alone, or Page and Brin, ruled non-resident; modest valuation shortfalls) lower the headline by about 10-26%, reaching but not clearly exceeding the 'major' threshold, while the qualitative conclusion (tens of billions, large relative to current taxes) survives.

The authors' response. The paper presents the aggressive-leaver scenario ($89B), argues that Brin and Zuckerberg remain California residents because they keep working at Alphabet and Meta in Silicon Valley, cites practitioners who doubt successful moves (Manes 2026), and stresses the 1 Jan 2026 residency date and the comprehensive base (pp.19, 23-24). A reply to Rauh et al. (Galle et al. 2026c) exists but was not accessed.

4moderatelikely · previously known · speculative assumption

The paper's claim that a yearly wealth tax would lose little revenue to billionaires moving away depends on how mobile billionaires are, and it uses an estimate drawn from European millionaires. Evidence on US billionaires suggests they may move more readily, in which case a 1% annual tax might keep about half of its paper revenue rather than 88%. A modest annual tax would still probably raise money, but less than the paper implies.

Technical detail and evidence

Why it matters, in detail. In the paper's model R(tau) = (t0+tau)Z0 exp(-e tau) - t0 Z0 with t0 = 0.2%, the share of mechanical revenue kept at tau = 1% is 0.89 (e=10), 0.78 (20.2), 0.72 (26.4), 0.57 (45), 0.43 (64), 0.24 (100), 0.07 (150); the revenue-maximizing add-on 1/e - t0 is 9.8%, 4.8%, 3.6%, 2.0%, 1.4%, 0.8%, 0.5%; maximal additional revenue on a $2,000B base is $71B, $34B, $25B, $14B, $9B, $5B, $3B a year. With d=15 erosion the 1% kept share is 0.73 at e=10 versus 0.46 at e=45. A small tax loses revenue only if e exceeds 500; Lucerne's total semi-elasticity (187) and Poterba-converted Moretti-Wilson values (141-326) are within a factor of 1.5-3.5 of that threshold, so 'no doubt' is too strong, although those figures include non-mobility responses or rest on strong conversion assumptions (the highest would imply that California's existing 0.2% income tax already shrinks its billionaire base by 25-50%).

Evidence. BSZ pp.26-27/PDF 28-29 and fn 35; workbook sheet Fig9 (R = t*Z0*exp(-e(t-t0)), t0=0.2%, d=15). Brulhart et al. (2022, AEJ:EP 14(4)) abstract and pp.130-133 (33.7pp response to a 0.18pp differential; 8.1pp from net movers; 'less than 30 percent ... attributable to mobility'). Agrawal, Foremny & Martinez-Toledano (CESifo WP 11048, 2024; AEJ:Applied 2025) pp.2-3 and the Swiss comparison paragraph (elasticity 7.96-10.85; 43 x 24% = 10.3, with 45 as the Lucerne alternative). Moretti & Wilson (NBER w26387) abstract, p.2 (8.25% effective rate), pp.20-21 (Poterba conversion; sensitivity rising with age). Jaros & Rauh (2026) Sec. 4.5 (10.32 baseline; 20.16-26.40 for large reforms). Jakobsen, Kleven, Kolsrud, Landais & Munoz (NBER w32153) abstract. scripts/c7_sensitivity.py reproduces the paper's 88.6% at e=10.

Why this severity. The qualitative claim (a modest annual tax raises net revenue) survives for any e up to about 100 at a 1% rate, i.e. across nearly all the evidence. But the quantitative claims are fragile: within the range used in the literature and by the critics (10-45) the share of mechanical revenue kept by a 1% tax falls from 89% to 57% and the revenue-maximizing add-on rate from 9.8% to 2.0%; under converted billionaire estate-tax evidence (64-104) the kept share is 24-43%.

The authors' response. The paper notes that the Spanish and Swiss estimates concern millionaires, that Spain polices residency weakly while California enforces it strictly, and that Switzerland is small, making mobility easier; it says e=10 is also the calibration of Rauh et al. (2026), and the workbook lets readers change e and d (pp.26-28; Fig. 9 note).

5minorconfirmed · previously known · overclaiming

The paper says billionaires who left to dodge a one-time tax would not stay away for good, but people who have already paid the cost of moving often do not move back. This matters less than it sounds: the billionaires most likely to leave pay very little California income tax, so even if they never returned the lost income tax would be a small fraction of the wealth tax collected.

Technical detail and evidence

Why it matters, in detail. Under the paper's own income-tax figures, a perpetual loss discounted at 3% is worth $5.1B in the benchmark ($0.153B/yr) and $17.2B in the aggressive scenario ($0.515B/yr), against $104B and $89B of revenue; on a contemporaneous base the benchmark loss is $0.24B/yr ($7.9B PV). Persistence therefore matters only if departures are much larger and leavers pay much more CA income tax per dollar of wealth than the SEC evidence shows for Page, Brin and Zuckerberg (0.38% fiscal income/wealth in 2025).

Evidence. BSZ pp.19-20/PDF 21-22, fn 29 (PDF 22), p.25/PDF 27; workbook Tab5!H6 = -0.05*Tab2!C14 and F18/B18; Auditor B fixed-cost example (T=4, K=2, V=1) and PV table; scripts/c6_scoring.py. PV at 3%: benchmark $0.153B/yr gives $5.1B in perpetuity; aggressive scenario $0.515B/yr gives $17.2B.

Why this severity. The theoretical claim is overstated, but because the paper's SEC evidence shows that the most likely leavers pay very little California income tax, even permanent losses are small relative to the one-off revenue ($5-17B in present value against $89-104B), so 'comparatively minor impacts on income tax revenue' survives.

The authors' response. Table 5 'does not take a stand on how long the annual income tax losses will last' (p.25); fn 29 argues that permanence requires expected future wealth taxes, which Section IV.B analyses.

6minorconfirmed · previously known · methodology

When the paper compares the four richest Californians' tax rate with the national average, it measures their income a little more generously than everyone else's. Fixing this nudges their rate up by about a point, but they still pay a much lower share of their income in tax than the average American.

Technical detail and evidence

Why it matters, in detail. Counting 50% of the top 4's $11.8B realized gains lowers their economic income from $142.9B to $137.0B and raises their total tax rate from 19.5% to 20.3%; together with F7 (about +0.8pp) the rate is about 21%, still about 10pp below the 31.4% macro average. The effect of alternative transfer and payroll-tax conventions (Splinter 2025) on this specific comparison was not re-estimated.

Evidence. BSZ Table 4 (PDF 38); Fig. 7 text (p.15/PDF 17); p.9/PDF 11; workbook data_dina!G formulas and Note 3; Auditor A repro_tables.py: 27.8/(142.9 - 0.5 x 11.8) = 20.3%.

Why this severity. A consistent capital-gains treatment raises the top-4 rate by under 1 point; the ranking against the economy-wide average is unchanged.

The authors' response. The paper follows the national-accounts conventions of Saez & Zucman (2023) and Balkir et al. (2025).

7minorconfirmed · newly identified here · methodology

The paper assumes companies pay property tax only on land and buildings, but California also taxes business equipment such as servers. Counting equipment would raise the top four's estimated tax rate by at most about one percentage point, which does not change the conclusion.

Technical detail and evidence

Why it matters, in detail. Applying 1% to all net PP&E, pro-rated by the ownership shares implied by w_pi over company pre-tax income, gives $1.93B of property tax for the top 4 over 2019-25 versus $0.78B in the workbook (+$1.15B), raising total taxes/economic income from 19.45% to 20.26% (Meta alone +$97M in 2024 and +$159M in 2025). This is an upper bound (equipment abroad, data-centre abatements); the gross-versus-net slip partly offsets it. No effect on C3 or C6.

Evidence. BSZ App. C.4(g) (PDF 63); BOE Personal Property FAQ ('Business personal property is all property owned or leased by a business except real property... may be subject to property taxes'); Meta 2025 Form 10-K Note 6 (land $3,687M/$2,561M, buildings $55,568M/$47,076M, net PP&E $176,400M/$121,346M for 2025/2024); workbook data_sec_top4 w_tax_ppent; XBRL net PP&E for Alphabet, Meta and Nvidia; scripts/property_tax_check.py.

Why this severity. Quantitatively small (under 1 point on the top-4 total tax rate) and unrelated to the California income-tax and revenue findings.

8minorconfirmed · newly identified here · methodology

Part of the reported jump in billionaire wealth comes from new people joining the billionaire list, not just existing fortunes growing, and the '322%' for the top four is for people picked because they are the richest today. Measured more neutrally, growth is still very large.

Technical detail and evidence

Why it matters, in detail. Incumbent growth (160%) is about 30pp below aggregate growth (192%) over Dec 2022-Sep 2026; the ex-ante top 4 grew +248% vs +322% for the ex-post top 4 (Huang's wealth rose 11.8-fold). Both remain far above the 144% aggregate, so 'more pronounced at the top' holds.

Evidence. Workbook data_rtb_20260921_2023 and data_sec_all; scripts/entrants.py; BSZ Table 1A (PDF 35), Fig. 1, abstract.

Why this severity. A framing point: the aggregate is the relevant tax base and both corrected growth rates remain far above income growth, so no conclusion changes.

The authors' response. The paper defines the top 4 as 'the top 4 richest today' and notes that new billionaires enter the list (p.3, p.21).

9minorconfirmed · newly identified here · arithmetic or transcription

A spreadsheet formula mixes two slightly different measures of total state income tax when projecting 2024-25. Newly released 2024 tax data show the projected number was nonetheless right, so the 'about 0.2% of wealth' finding is unaffected.

Technical detail and evidence

Why it matters, in detail. Using row 19 consistently would raise 2024/2025 from $3.045B/$4.345B to $3.223B/$4.600B (2023-25 mean $3.28B to $3.43B; tax/wealth 0.206% to 0.214%). However, the authors' Method I applied to the FTB TY2024 data published in September 2026 gives $3.051B, so the 2024 projection was in fact accurate; the inconsistency has no demonstrated effect on the 0.2% finding.

Evidence. Workbook billionairesCAinctax rows 15, 18, 19, 50, 51; BSZ App. B (PDF 55-56); scripts/method1_ty2024.py (FTB PIT Annual Report 2025, Table B-4A, TY2024: $10M+ bracket 6,207 returns, CA AGI $190.38B, taxable income $174.49B, tax $16.92B).

Why this severity. A genuine formula inconsistency, but the 2024 figure it produced is confirmed by later data and the 2025 figure is a projection either way.

10minorconfirmed · newly identified here · arithmetic or transcription

One alternative scenario forgets a small reduction applied elsewhere, overstating it by about $2 billion ($115 billion should be $113 billion). The main estimate is unaffected.

Technical detail and evidence

Why it matters, in detail. Correct value F8 + (F7 - F6) = $112.8B versus $114.8B printed (-$2.0B, -1.7%).

Evidence. Workbook Tab5!F7, F9; BSZ p.25/PDF 27 and Table 5 (PDF 39); scripts/c6_scoring.py.

Why this severity. A $2.0B slip in a secondary scenario; the benchmark is unaffected.

11minorconfirmed · newly identified here · data error

For 2024 the paper credits Page and Brin with a full year of Alphabet dividends although Alphabet started paying only in mid-2024. That slightly overstates their taxes, which if anything makes the paper's 'low taxes' point more conservative.

Technical detail and evidence

Why it matters, in detail. Overstatement about $0.30B (+33% of 2024 Page/Brin dividends; 7.5% of the top 4's 2019-25 dividend total of $4.0B), CA tax about +$35-40M (about 2% of the top 4's 2019-25 CA tax); 2025 ($0.84 vs $0.83) is negligible. The 3.7% payout ratio becomes about 3.5%.

Evidence. BSZ App. C.3(g) (PDF 61-62); workbook data_sec_top4 (dividend 2024); Alphabet XBRL CommonStockDividendsPerShareDeclared: CY2024 0.60, CY2025 0.83; scripts/top4_table.py.

Why this severity. Overstates the top 4's taxes slightly, i.e. in the direction that weakens rather than strengthens the paper's thesis.

12minorconfirmed · newly identified here · misrepresentation of source

The paper says billionaires had only a few weeks to leave before the tax's cut-off date, but the proposal and its 1 January 2026 residency date were public from late October 2025, giving about ten weeks. That makes successful departures somewhat more plausible than the paper suggests.

Technical detail and evidence

Why it matters, in detail. The window between public filing (22 Oct 2025) and the 1 Jan 2026 snapshot was about 10 weeks rather than about 5, which makes a pre-snapshot relocation somewhat more feasible than the paper's narrative implies; the score itself is unaffected.

Evidence. BSZ p.19/PDF 21; original initiative 25-0024 cover letter (stamped Oct 22 2025) and definitions section ('as assessed on January 1, 2026'); amended initiative 25-0024A1 cover letter dated November 24, 2025, stamped Nov 26 2025; Jaros & Rauh (2026) 'over two months'.

Why this severity. Does not change any computation; it modestly weakens the qualitative support for the 5% mobility allowance discussed in F1.

13minorconfirmed · newly identified here · documentation gap

The authors posted their yearly results but not the full filing-by-filing data behind their estimates for the four richest Californians, so others cannot rebuild them from scratch. The pieces we could check against public securities filings matched exactly.

Technical detail and evidence

Why it matters, in detail. Lot-level basis choices and gift classification drive realized gains and deductions (removing all modeled charitable deductions would raise top-4 2019-25 CA tax from $1.88B to $2.45B, an extreme bound per Auditor B), but the spot-checked inputs match exactly, so no bias in C4 is indicated.

Evidence. BSZ cover note, App. C.1-C.4 (PDF 56-63); workbook data_sec_top4, data_sec_all, data_sec_codebook, data_sec_propublica. Adjudicator spot checks: EDGAR shows no Page Form 4s in 2018-2020 or 2023-2026 and only conversion and gift codes in Brin's 2019, 2020 and 2023 filings; Huang's EDGAR open-market sales (6,000,000 shares, $713.0M in 2024; 6,000,000, $1,048.6M in 2025) equal the workbook; Auditor A's XBRL pre-tax income check implies plausible ownership shares.

Why this severity. A reproducibility gap without demonstrated error: every independently checkable input matched public filings.

The authors' response. The workbook posts person-year data and formulas; Appendix C.3(i) discloses the ProPublica comparison and suggests ProPublica may have missed a high-income year for Zuckerberg.

14minorconfirmed · newly identified here · arithmetic or transcription

There are a number of small typos, rounding slips and stale numbers, such as a table average that leaves out one year. None changes any conclusion.

Technical detail and evidence

Why it matters, in detail. Largest effects: the Table 2 top-4 average (0.08% vs 0.072%), the US long-run growth rate (6.2% vs 6.0%/yr) and the 2023-25 CA tax ($3.2B vs $3.28B). All other items are labels, citations or approximations.

Evidence. scripts/show.py dumps of Tab1, Tab2, TabA1, top4taxes, billionairesCAinctax; Auditor A repro_tables.py; Auditor B reproduce.py; BSZ pages as listed (PDF = printed + 2).

Why this severity. Each slip is a rounding, labelling or range error of at most a few tenths of a percentage point; none changes a conclusion.

15minorlikely · newly identified here · misrepresentation of source

The paper suggests the state's official analysts were neutral on how long lost income-tax revenue would last; in fact they called the loss 'likely ongoing'. Their revenue guess ('tens of billions') is also lower than the paper's.

Technical detail and evidence

Why it matters, in detail. The LAO characterises losses as ongoing and revenue as 'tens of billions', below the paper's $104B; citing it as agnostic on duration overstates its support for treating losses as temporary. No number in the paper changes.

Evidence. BSZ p.25/PDF 27; LAO/DOF fiscal impact letter on 25-0024A1, 11 Dec 2025, 'Fiscal Effects' and 'Summary of Major Fiscal Effects'.

Why this severity. Presentational: the LAO gives no duration in years, but its framing (ongoing losses, 'tens of billions' of revenue) is less favourable than the paper conveys.

16minorlikely · newly identified here · overclaiming

The paper points to record venture investment in California as evidence that the wealth-tax proposal did not deter investors. The surge comes from a handful of giant AI deals, the number of deals was flat, and there is no way to know what would have happened without the proposal.

Technical detail and evidence

Why it matters, in detail. The data show that no collapse occurred, not that the initiative had no effect; the flat count share (32.2% in 2025, 30.5% in 2026H1) is uninformative either way.

Evidence. BSZ pp.20-21/PDF 22-23; Fig. 8 (PDF 47); workbook data_venturemonitor_annual columns H (count share) and I (value share).

Why this severity. A secondary descriptive claim; no core number depends on it.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

The paper's e of about 10 misrepresents Brulhart et al. (2022) by combining the cross-canton 43% with a 24% mobility share from a different sample (issue I1, categorised as misrepresentation of source). — Rejected as a misrepresentation: Brulhart et al.'s own abstract states that '24 percent of the effect arises from taxpayer mobility' after reporting the 43% effect, and Agrawal et al. (CESifo WP 11048) and the critics Jaros & Rauh (2026) make the identical 43 x 24% = 10.3 calculation. The legitimate point, that applying the share to the Lucerne response gives 45, is retained as a sensitivity in F2.
Agrawal et al.'s finding that wealth-tax mobility generates income-tax losses six times larger than wealth-tax losses contradicts the paper (part of issue I1). — The six-to-one ratio reflects Spanish wealth taxpayers' income tax being six times their wealth tax. The paper's model already includes the income-tax loss term; it is small for California billionaires because their CA income tax is only about 0.2% of wealth, so the Spanish ratio does not transfer.
The statement that e=500 is 'at least one order of magnitude larger than any semi-elasticity estimated in empirical work' is unsupported, and Moretti-Wilson implies e of about 137-326 (part of issue I1). — Partially rejected. For directly estimated mobility-only semi-elasticities (at most 45) the statement holds. The 137-326 range comes from a Poterba-style conversion that would imply California's existing 0.2% income tax already shrinks its billionaire base by 25-50%, hard to square with California holding about a quarter of US billionaire wealth; the critics themselves use 10-26 and Scandinavian evidence gives about 2. F2 keeps the softened point (Lucerne's total 187 and converted estate-tax evidence are within a factor of a few of 500).
Defensible alternatives put the one-off revenue at $67-96B (issue I4). — Partially rejected. The $67B case removes Zuckerberg, who by all accounts, including Jaros & Rauh's, was a resident on 1 Jan 2026 and remains taxable under the Act as written; a 50% valuation shortfall on all private wealth is an upper bound given funding-round floors and listed-stock valuation. The adjudicated range is about $70-110B (F1).
Litigation risk to the retroactive residency date and revenue timing under the Optional Deferral Account are flaws of the score (part of issue I4). — Not flaws of a conventional score, which assumes enactment as written (the LAO does the same); Jaros & Rauh's 50% litigation weight is an accounting convention (ASC 740), not an estimate. Recorded as caveats in F1.
The paper asserts without support a 'consensus' that the initiative is written to minimise avoidance (part of issue I6). — Immaterial: the initiative's anti-avoidance provisions (no discounts below pro rata, funding-round floors, inclusion of trusts and post-October-2025 transfers) are in the text, and critics' main complaints (e.g. control-share valuation) concern over- rather than under-taxation. The LAO mischaracterisation is retained (F13).
The main text ('about 10%') and the appendix ('about 13% higher') give inconsistent figures for how much Method I overshoots the Balkir-based benchmark (part of issue I10). — Not a real inconsistency: the same comparison is 13% when Method I is expressed relative to the alternative ($1.92B vs $1.70B) and 11.7% when the alternative is expressed relative to Method I; 'about 10%' is loose rounding.
Burden comparisons omit the statute's 7.5% deferral charge, so the installment burden is 5.75% rather than 5% (issue I8). — The tax is 5% of net worth; the 7.5% charge is the price of an optional deferral (interest-like), so for a taxpayer who elects it the present-value burden is at most 5% at their own discount rate. Omitting it neither overstates revenue (the charge would add revenue) nor understates the economic burden.
The Table 2 top-4 0.08% and the text's 0.07% are different statistics and should not be treated as contradictory. — Tab2!J14 = AVERAGE(J7:J12) omits 2025; the seven-year average of the annual ratios is 0.072%, consistent with the text's 0.07%. The printed 0.08% is a formula range error (F16).
Alternative start years for the long-run series 'are not evidence that the 1982 calculation is wrong or cherry-picked', and the May/July/August headline changes are disclosed, not a problem. — Partially rejected. 1982 is not cherry-picked (it is the first Forbes 400 year), but it is an outlier low for California (+70% in one year vs +25% US) and the abstract's x25 falls to x15-17 with any 1983-86 base; the headline also moved from x30 to x38 to x25. Retained as moderate F4.

Audit notes

What reproduced. Reproduced: Table 1A/1B (144%, 322%, x24.6-24.7); the Table 2 series and ratios (with two AVERAGE range errors); Method I CA income tax for 2018-2023 from FTB/IRS inputs (Auditor A; 2023 re-run by the adjudicator) and, out of sample, 2024 on FTB TY2024 data ($3.051B vs $3.045B projected); Tables 3-4 sums; Table 5 rows 1-3 exactly (row 4 overstated by $2.0B); the Figure 9 model; the Pareto-missing extrapolation (617, 368, $615B); and SEC spot checks (Page/Brin no-sale years, Huang's 2024-25 sales, Alphabet dividends, Meta property-tax base). Not reproducible from public materials: the 1982-2018 Forbes 400 California inputs (Stata output not posted), the Compustat-based inputs and transaction-level SEC pipeline end to end, PitchBook data, and the 2025 income-tax figure (a projection; no FTB data yet). Robustness analyses run: base-year sensitivity of x25; Hill, rank-size and 2022 placebo checks of the missing-billionaire extrapolation; e-sensitivity of Figure 9; alternative leaver, haircut and valuation scenarios; entrant decomposition; property-tax convention.

Limits of this audit. Galle et al. (2026a-c), Walczak (2026) and the SSRN version of Rauh et al. (2026) could not be accessed (SSRN blocked for both auditors), so the 78.4% and 2.25% attributions are unverified. The 1982-2018 Forbes 400 California inputs and Pareto coefficients come from unposted Stata output, so the cause of the 1982-83 jump could not be traced. Compustat and PitchBook are proprietary; the SEC pipeline was spot-checked (Page, Brin, Huang; XBRL aggregates), not rebuilt. The workbook is a post-August vintage (it contains a 21 Sep 2026 Forbes snapshot and its 'last update' cell is =TODAY()). The legal residency of Page, Brin and others on 1 Jan 2026, litigation outcomes and the December 2026 valuation cannot be determined here, and Forbes residence labels are not legal determinations. Conversions of estate-tax evidence into annual wealth-tax semi-elasticities are approximations reported as ranges. Jaros & Rauh (Oct 2026) postdates the audited version and was used as a comparison, not as a standard the authors could have met. Moretti & Wilson were read in the NBER working-paper version.