← All papers
Brian Galle, David Gamage, Emmanuel Saez, Darien Shanske · Expert report (SSRN 5996554); versions Dec 2025, Feb 2026, Apr 2026, updated July 20, 2026
Expert Report on Proposition 40, the California 2026 Billionaire Tax: Revenue, Economic, and Constitutional Analysis
17 problems found: 1 major · 6 moderate · 10 minor
What the paper claims
- Proposition 40 would raise about $100 billion over 2027-2031 (about $20 billion a year) from the ~250 wealthiest Californians; the benchmark is described as conservative.
- The tax base is large and growing fast: California has 12% of the US population but 25% of US billionaire wealth, and billionaire wealth now about equals total California AGI.
- Billionaires are under-taxed relative to their income and wealth, and their taxes relative to wealth have fallen sharply since the Reagan era.
- Mobility and income-tax losses are negligible: residence is fixed at January 1, 2026; pre-2026 departures are 'very unlikely'; the super-wealthy 'do not generally move because of taxes'; lost income tax would be 'very small' relative to $100bn ('a point noted by the LAO') and income-tax revenue is more likely to rise.
- Avoidance and valuation risk is small, justifying a 10% allowance, and the Forbes base is accurate or conservative.
- The tax will not harm businesses or start-up finance, and its valuation rules are simple, Swiss-tested and do not tax founders' voting control.
- The tax is constitutional: it is not impermissibly retroactive (cf. Prop 30, 2012), states may tax residents' worldwide wealth, it is consistent with Prop 13, and the constitutional amendment lifts the 0.4% cap on intangibles.
Every problem found
1majorlikely · previously known · speculative assumption
Two people, Google's founders, hold about a quarter of all the wealth the tax would hit. Both bought homes and moved business entities out of California in the weeks before the January 1, 2026 cutoff. One of the report's own authors wrote in a companion paper that Page might be able to show he had already left. If they don't owe the tax, $100 billion becomes about $78 billion, but the report dismisses the risk instead of showing it.
Technical detail and evidence
Why it matters, in detail. The score is linear in the taxable base, and the base is extremely concentrated: the top 4 hold 41.1%, Page 12.8% and Brin 11.8%. January-1 residency is a facts-and-circumstances test (R&TC ss.17014, 17015.5) that will be litigated taxpayer by taxpayer. Removing Page alone costs $13.3bn; removing both costs $25.5bn ($103.8bn -> $78.3bn). Both have since asserted non-California residency, so the same base would also escape if a court reformed the January-1 date (Act s.50311(b); see F4). The report gives an unweighted point estimate and omits BSZ's -14.5% row while citing its +23% row.
Evidence. Report p.5, p.4 fn 7; April 25, 2026 version fn 1. BSZ workbook Tab5 scenario-3 worksheet (July-1 values: Page 294.6, Brin 271.7, Thiel 27.3, Hankey 8.15, Kalanick 3.56; F8 = $88.80bn). BSZ July 2026 pp.21-22 (Page 'reportedly spends most of his time in Florida with little physical presence in California since late December 2025'; Thiel, Hankey and Kalanick 'may have moved before January 1'). Pre-report press: The Real Deal, Jan 7, 2026 (Page bought two Miami estates for $173m in Dec 2025/Jan 2026; three Page LLCs incorporated in Florida in December; NYT reported Page 'was looking to leave California'); Fox Business, Jan 12, 2026, citing NYT (45+ Page-linked California LLCs moved out or went inactive in December; a Brin entity terminated or relocated 15 California LLCs, several to Nevada). Post-report: Forbes, Oct 16, 2026 (Brin registered to vote in Nevada on Dec 10, 2025 and bought a $42m Nevada home in December; spokespeople for both now assert non-California residency); authors' tracker movers-out file (Forbes relabelled Page to Florida and Brin to Nevada on 2026-10-11). Adjudicator computation (scripts/reproduce_core.py), July-1 base: Page out -> $90.6bn (-12.8%); Page, Thiel, Hankey, Kalanick out -> $88.8bn (-14.5%); Page and Brin out -> $78.3bn (-24.5%); all five out -> $76.6bn (-26.2%).
Why this severity. A single unvalidated legal assumption covers about a quarter of the base. Between the authors' own 'aggressive' scenario (-14.5%) and the case in which both Google founders prevail (-24.5%), the headline falls to $78-89bn at July-1 prices. 'Tens of billions' survives, so major rather than critical.
The authors' response. Response to Rauh et al. (Mar 17, 2026): buying real estate or announcing a move does not sever California residence, which requires 'uprooting most of the individual's personal and business ties'; 'there is no real evidence that any significant number' of Forbes-listed billionaires will avoid the tax; Rauh et al. present 'only the worst case'. BSZ (July 2026, p.21) argue Brin and Zuckerberg remain residents because they keep working in Silicon Valley. These are reasonable legal arguments, but they support presenting the departures as a quantified risk, not omitting it.
2moderateconfirmed · newly identified here · misrepresentation of source
The report says California's independent budget office agreed that lost income taxes would be tiny next to $100 billion. The office said no such thing. It estimated only 'tens of billions' from the tax and warned that the income-tax losses could continue year after year while the new money would be temporary.
Technical detail and evidence
Why it matters, in detail. The LAO's quantities are 'tens of billions' of one-time revenue against an ongoing loss of $0.1bn to more than $1bn a year. Whether that is 'very small' depends on duration and discounting (F8), which the LAO explicitly declined to judge. The attribution adds no evidence and misstates the independent benchmark.
Evidence. Report p.9 and fn 12 (same sentence in the Dec 31, 2025 version, pp.6-7); LAO letter to the Attorney General, Dec 11, 2025, p.2, 'Fiscal Effects' and 'Summary of Major Fiscal Effects' (oag.ca.gov fiscal-impact-estimate-report(25-0024A1).pdf, SHA-1 0db246a5...); LAO Proposition 40 analysis for the Nov 3, 2026 ballot.
Why this severity. No number changes, but the report borrows the authority of the official nonpartisan estimate for a conclusion, and an implied $100bn, that the LAO did not reach. The LAO's framing (ongoing losses against temporary gains) points the other way.
The authors' response. Response to Rauh (Mar 17, 2026), concluding points: 'Our prior analysis as well as nonpartisan analysis by California's Legislative Analyst's Office each conclude those policies would on net raise tens of billions'. That paraphrase is closer to the LAO's text than the report's.
3moderateconfirmed · previously known · selective presentation
The report says very rich people rarely move to avoid taxes. But the best study of American billionaires, which the authors' own companion paper relies on, found they move quite a lot in response to state estate taxes, and several California billionaires did announce moves in 2026.
Technical detail and evidence
Why it matters, in detail. Moretti-Wilson's estimates imply large stock responses among the ultra-wealthy to anticipated taxes. That bears on (a) whether pre-cutoff moves occurred and (b) persistent post-tax departures driven by expected future wealth taxes, which BSZ acknowledge. Studies of millionaires under income taxes (Young et al. 2016; Money Moves 2025) find modest migration, so the literature brackets the question rather than settling it the report's way.
Evidence. Report pp.5, 8; fn 15 (citing Galle-Gamage-Shanske, Money Moves, 113 Cal. L. Rev. 635 (2025)); Moretti & Wilson, NBER WP 26387 (2019, rev. 2020), abstract, published AEJ: Economic Policy (2023); BSZ July 2026 pp.16-17; Bruelhart, Gruber, Krapf & Schmidheiny (2022) AEJ: Economic Policy 14(4). Later facts: Forbes relabelled eight January-1 California billionaires to other states during 2026 (tracker movers-out file).
Why this severity. It does not change the one-time revenue if the January-1 date holds. It does weaken the evidence for the claims that departures and future income-tax losses are negligible, and for the dismissal of pre-2026 moves (F1).
The authors' response. BSZ July pp.16-18: the one-time tax fixes residence before announcement effects can operate, a permanent response to a one-time tax 'does not make economic sense', and Moretti-Wilson find smaller responses among younger billionaires.
4moderatelikely · previously known · selective presentation
The report says its estimate is probably too low because Forbes misses some billionaires. But its count of missing billionaires is far higher than other sources suggest. It also ignores the authors' own finding that the values billionaires' estates report to the tax authorities are well under half what Forbes says they were worth. So the claim that the estimate is cautious is not shown; it could just as well be too high.
Technical detail and evidence
Why it matters, in detail. The upside rests on a model-dependent Pareto imputation that two independent benchmarks put about 4-5x too high in revenue terms (+$24bn vs ~+$5bn). The downside channels go unquantified: statutory-valuation shortfalls on unlisted equity (Skar et al. 2025: Norwegian statutory values at a median 61% of sale prices), the formula-rebuttal asymmetry, and family and trust attribution. The net sign of the Forbes-vs-statutory gap is unknown, so 'conservative' is not established. On this dimension the benchmark is best described as roughly central, with error bands of at least +/-10%.
Evidence. Report p.4 fn 7; Dec 2025 version fn 4 (Bloomberg, Oct 17, 2025; top-500 threshold $7.33bn); BSZ July pp.20-21 and Table 5 row 2; Response to Rauh (Mar 17, 2026) p.4 (WSJ 255 vs Forbes 196 in 2024); Jaros & Rauh (2026) App. F, pp.46-48; Balkir-Saez-Yagan-Zucman (2025) NBER WP 34170 pp.2-3 and p.12; Act s.50303(b), (c)(3)(E)-(F), (c)(10); scripts/persons_sep21.py ('& family' entries); scripts/reproduce_core.py.
Why this severity. This is not a demonstrated bias in the $100bn, which rests on Forbes values, but the claim that the estimate errs low is unsupported. Valuation and attribution shortfalls on the private 40% could plausibly cut the base by 5-15%, partly offset by a ~5% missing-billionaire upside.
The authors' response. Report fn 7 and Response to Rauh p.4: Forbes likely misses heirs and new billionaires (WSJ 255 vs 196). BSZ p.21: the FTB can find them with income-tax data and capitalization methods. BSYZ attribute estate gaps to estate-specific undervaluation techniques, which the Act's design (no discounts, funding-round and insurance floors) largely closes.
5moderateplausible · previously known · speculative assumption
The report assumes only 10% of the money will be lost to avoidance and evasion, but gives no evidence for that number and no range. If 20% were lost instead, the total would be about $92 billion.
Technical detail and evidence
Why it matters, in detail. Revenue = 0.05(1-h)W, so each 5pp of allowance costs ~$5.8bn. The one-time design and retroactive residency date close many of the margins behind large recurrent-tax elasticities (Bruelhart et al. 2022; Jakobsen et al. 2020). The margins that remain are valuation of the ~40% private wealth, trust and family attribution, and residency, and the 10% (only 5% net of mobility) is not calibrated to any of them. It is an assumption, not an estimate, and should be shown as a range.
Evidence. Report p.4; Dec 31, 2025 version p.4 ('72%'); BSZ workbook Tab1 column E (2025: 0.652; July 2026: 0.6) and Tab5 note ('half due to mobility'); scripts/persons_sep21.py (Forbes public share 60.0% on 2026-09-21). Adjudicator sensitivity (scripts/reproduce_core.py), July-1 base: 15% -> $98.0bn; 20% -> $92.3bn; 30% -> $80.7bn; Page out plus 15% -> $85.5bn; Page and Brin out plus 15% -> $74.0bn.
Why this severity. Within a defensible 10-20% range the headline moves by up to -11%. That is noticeable but not transformative on its own, though it compounds F1 and F3.
The authors' response. Report p.4; BSZ pp.15-19: a comprehensive worldwide base, trust and gift add-backs, funding-round floors, no valuation discounts and an audit of every taxpayer make avoidance hard; BSZ say the 10% 'captures all dimensions of behavioral responses'. These design features genuinely make a low allowance defensible, but not a validated point value.
6moderateplausible · previously known · speculative assumption
The tax will almost certainly be challenged in court. The law itself says that if judges reject the January 1, 2026 residency date, a later date should be used, which would let billionaires who left during 2026 escape. The report doesn't mention this and presents $100 billion as if there were no court risk.
Technical detail and evidence
Why it matters, in detail. Expected revenue = sum over margins of P(state prevails) x base at risk; the report sets every probability to 1. The date-reformation margin dominates: ~$578bn (Oct-2026 values) of January-1 residents have since left. Jaros-Rauh's flat 50% survival discount is itself arbitrary (borrowed from the ASC 740-10 'more likely than not' threshold), so no point estimate of this risk is defensible. A range, or at least a statement that $100bn assumes full legal survival, is.
Evidence. Report pp.5, 11-13; Act ss.50306(a)-(b), 50308(a),(n),(o), 50311(b) (initiative 25-0024 A1, pp.21-27); LAO letter Dec 11, 2025 ('very hard to predict'); LAO Prop 30 analysis (2012: higher rates 'starting in the 2012 tax year'); Forbes RTB snapshot 2026-10-25 (scripts/forbes_us_share.py: eight relabelled January-1 residents, $578bn); BSZ July p.21 (Zuckerberg bought Florida real estate).
Why this severity. An undisclosed caveat, not a demonstrated error. Revenue scorers rarely discount for litigation, but an adverse ruling on the residency date alone would now remove roughly a quarter or more of the base.
The authors' response. Response to Rauh (Mar 17, 2026) fn 2: invalidating the date is 'an unlikely scenario, as we have explained in depth elsewhere'; the Galle-Gamage-Shanske legal memo (Jan 9, 2026) relies on United States v. Carlton (1994), the 1913 income tax and Prop 30.
7moderateplausible · newly identified here · documentation gap
The report says the money will arrive at about $20 billion a year for five years. But the law lets billionaires whose wealth is mostly in private companies put off much of their payment until they cash out or die, possibly long after 2031, while others may pay everything at once. The yearly figure is a rough average, not a forecast.
Technical detail and evidence
Why it matters, in detail. Cash in 2027-2031 = lump-sum payments + installment principal + 7.5% charges - ODA-deferred principal. ODA tax (5% of future distributions or terminal value) is roughly PV-neutral at market discount rates. The window's cash flow could therefore be front-loaded (lump sums if borrowing costs are below 7.5%), higher in nominal terms (+15% with universal installments) or lower (ODA). None of this is modelled. With 10-20% ODA use, principal inside the window would be $83-93bn.
Evidence. Report pp.1, 3, 5; Act s.50301(c) (p.11 of 32) and s.50304(a)-(l) (pp.18-21); scripts/persons_sep21.py (126 eligible; $35.5bn deferrable upper bound, reproducing Auditor A's R12); scripts/reproduce_core.py (all-installment charges $6.23, 4.67, 3.11, 1.56bn = $15.6bn, matching Auditor B).
Why this severity. The present-value total need not be lower, but an unknown and possibly material share (up to ~29%, an upper bound) could arrive after 2031 or depend on future valuations. That matters for a tax justified by an immediate funding gap, and the headline cash profile is unsupported.
8minorconfirmed · newly identified here · overclaiming
The report points to a boom in venture funding for California start-ups as proof the tax scares no one away. But the boom came from a couple of giant AI companies and would likely have happened anyway, so it can't show whether the tax discourages other start-ups.
Technical detail and evidence
Why it matters, in detail. A causal 'no effect' claim needs a counterfactual (e.g., a comparison with other AI hubs, or deal timing around the valuation date). An aggregate share driven by AI mega-rounds has essentially no power against a marginal deterrent effect on ordinary start-ups; absence of evidence is presented as evidence of absence.
Evidence. Report p.6; BSZ July pp.18-19 and Figure 8; BSZ workbook data_venturemonitor_quarterly ($336.8bn of $412.6bn = 81.62%; 2,285 of 7,491 deals = 30.5%, per Auditor B).
Why this severity. A secondary claim. The data rule out a funding collapse but not a marginal effect, and the revenue estimate is untouched.
9minorconfirmed · newly identified here · misrepresentation of source
The report says its formula for valuing private companies is one Switzerland has used for decades. Switzerland actually uses a different formula that gives a company's book value much less weight, and it allows discounts for minority stakes that this measure bans.
Technical detail and evidence
Why it matters, in detail. With equal earnings coefficients, Prop 40 = Swiss + (2/3)S, so the default overvalues asset-heavy, low-ROE firms by up to ~2x. Because taxpayers can rebut overvaluation by appraisal, the formula will bind mostly when it undervalues, which feeds the asymmetry in F3.
Evidence. Report p.7; SSK Kreisschreiben 28 (2022 version) RZ 32, 34-35, 61-62; SSK/CSI 2025 commentary (capitalization-rate table); Act s.50303(b), (c)(3)(E); scripts/swiss_formula.py: at K = 8.75-10%, Prop 40 exceeds the Swiss value by 1.24-1.39x (ROE 30%), 1.60-1.75x (ROE 10%) and 1.93-2.06x (ROE 5%).
Why this severity. No effect on the Forbes-based score, and Switzerland does show that formula valuation is administrable. But the specific validation claimed is wrong, and the divergence means appraisal disputes will be common for asset-heavy firms.
10minorconfirmed · previously known · internal inconsistency
Critics said founders with extra voting rights might be taxed as if they owned far more of their companies than they do. The report calls that a bad-faith misreading, but the authors' own first version admitted the rule could apply in some cases, and the law leaves the outcome to the tax agency.
Technical detail and evidence
Why it matters, in detail. Classifying Class B shares as 'publicly traded assets' depends on FTB discretion under s.50308(j). If they are not so classified, the presumption applies and rebuttal carries a clear-and-convincing burden. The authors' interpretation will likely prevail by regulation, but the categorical 'does not apply' and 'can always' wording understates both steps.
Evidence. Report pp.5 (fn 9), 7-8; Dec 31, 2025 version fn 5 (p.4); Feb 16, 2026 version p.7; authors' controlling-shares memo (Feb 5, 2026); Act s.50303(c)(1), (3)(C), (3)(F) and s.50308(j); Meta 2025 Form 10-K Item 5 (Class B not listed; per Auditor B).
Why this severity. The scoring values founder shares at market, and the authors' reading (Feb 5 memo; FTB regulatory power) is plausible. The defect is overstated legal certainty and an unacknowledged shift from the hedged December text; there is no revenue effect.
The authors' response. Controlling-shares memo (Feb 5, 2026): s.50303(c)(3) expressly excludes 'assets and entities governed by paragraph (1)', the s.50308(j) catch-all covers convertible super-voting shares, scoring and Forbes value such shares at market, and the FTB can confirm this by regulation.
11minorconfirmed · newly identified here · arithmetic or transcription
A handful of small factual mistakes and out-of-date details, such as Anthropic's valuation being misstated by $20 billion. None changes the report's conclusions.
Technical detail and evidence
Why it matters, in detail. Transcription and staleness errors; the largest is a $20bn (2.1%) overstatement of one firm's valuation, with no propagation into the $2,307bn base.
Evidence. Report pp.2-4, 6, 8-9; Anthropic Series H announcement, May 28, 2026 ('$65 billion in Series H funding ... valuing the company at $965 billion post-money'); BSZ July p.18; BSYZ p.11; tracker CSV; scripts/persons_sep21.py; Act s.50303(c)(7); Dec 2025 version pp.3, 6.
Why this severity. Each is small and none changes a conclusion.
12minorconfirmed · internal inconsistency
Two examples are oversimplified. Selling 5% of your shares won't quite pay a 5% tax once you owe tax on the sale, and the Google and Meta founders do pay some state income tax on dividends even without selling. The broader points still hold.
Technical detail and evidence
Why it matters, in detail. Sale gross-up q = 0.05/[1 - t(1-b)]. Dividend income on founder stakes (yields ~0.3-0.4%) makes California income tax small but not zero. Neither enters the revenue score.
Evidence. Report pp.3-4, 9; BSZ workbook Tab3 and data_sec_top4 (columns sale, dividend, ca_income_tax, 2024-2025 rows for larry-page, sergey-brin, mark-zuckerberg); Auditor B's sale-financing identity q[1 - t(1-b)] = 0.05; Meta 2025 Form 10-K (dividends since 2024, per Auditor B).
Why this severity. Both are simplifications in examples; the underlying points (dilution is small; most billionaire income is unrealized) survive.
13minorconfirmed · newly identified here · internal inconsistency
A footnote calls the richest 45 Californians 'the billionaire class' although there are about 250 billionaires, and uses a broad economic average as 'family income'. Using the broader group shows wealth growing 25-fold rather than 37-fold, which is still enormous.
Technical detail and evidence
Why it matters, in detail. Growth of a repeated top fractile (not a fixed cohort) depends on the fractile chosen: 8.5%/yr for the top .0002% vs 7.6%/yr for the top .001%. A GDP- or AGI-per-family comparator is a macro or fiscal average, not household income (cf. CORPUS us-income-03/-04).
Evidence. Report p.2 fn 2; BSZ July Table 1 Panel B and longrunseries (Auditor B: $0.974bn -> $35.97bn, x36.93, 8.55%/yr; GDP per tax unit x1.88); current BSZ workbook Tab1 Panel B (top .001%: x24.7, 7.56%/yr; AGI per family x1.97).
Why this severity. The qualitative point (top wealth has grown far faster than average income) holds under any definition; only the labels and the multiple (37 vs 25) change.
14minorconfirmed · newly identified here · version instability
Early versions of the report already said $100 billion, but they counted a billionaire who had left California years earlier and missed two dozen others. The number stayed the same only because stock prices rose.
Technical detail and evidence
Why it matters, in detail. At constant January-1-2026 prices the corrected base scores $92.5bn; the +$11bn to July is market movement, which the report says can move the score by +/-$10bn within months ($87-107bn range).
Evidence. Dec 31, 2025 version p.3; Feb 16, 2026 version p.3; April 25, 2026 version p.2 and fn 1; July version fn 1; Response to Rauh p.4 (Ellison $245bn; 24 non-citizens, $110bn); tracker 2026-01-01 row; scripts/reproduce_core.py.
Why this severity. The corrections were disclosed (fn 1) and improve accuracy, and scoring at current prices is standard. The point is that the headline's stability is coincidental, not evidence of robustness.
The authors' response. Fn 1 of the April and July versions discloses both corrections.
15minorlikely · previously known · methodology
The report compares a one-time pot of money with tax losses that could recur every year, without saying for how many years. On the authors' own numbers the losses would still be much smaller than the gain, but not 'very small' if the departures are permanent.
Technical detail and evidence
Why it matters, in detail. PV(loss) = L/(r-g) if permanent and ~5L if temporary. With L ~ $0.3-0.5bn (BSZ method), net revenue stays strongly positive. The NPV turns negative only with Jaros-Rauh's larger income-tax base ($3.3-5.8bn/yr, 30-60% lost permanently) and a low discount rate, and BSZ's SEC evidence on the top three, checked by Auditor B, makes that base less credible.
Evidence. Report p.9; BSZ workbook Tab5 (H6 = -0.153, H8 = -0.515; F18/B18 = 0.2008% of wealth; F20, F25), Tab2-Tab3; scripts/reproduce_core.py. Auditor B's horizon table ($0.5bn/yr at 3%: $2.29bn over 5 years, $7.44bn over 20, $16.7bn perpetual) reproduces.
Why this severity. Even if losses are permanent, on the authors' SEC-based income estimates their PV is ~10-25% of the one-time revenue. 'Very small' overstates it, but the net-positive conclusion survives.
The authors' response. Response to Rauh and BSZ pp.18, 22-23: under rational expectations, responses to a truly one-time tax should fade; Rauh et al. set permanent losses against one-time revenue at a 1.5% discount rate, magnifying the loss 67-fold. That critique of Rauh et al. is valid, but persistence vs fading is an assumption either way.
16minorlikely · newly identified here · misrepresentation of source
The report says federal cuts will leave California's health budget $19 billion short every year. The source it cites estimates about $15 billion a year on average, with most of the cuts arriving after 2030.
Technical detail and evidence
Why it matters, in detail. KFF's 10-year midpoint for California is $15.0bn/yr. Scaling by the national path gives ~$7.6bn in 2027 rising to ~$21bn in 2031 (~$14.4bn/yr over 2027-2031). The report's $19bn matches KFF's high-end 10-year average (or a late-years average), not its central estimate.
Evidence. Report p.1 (hyperlink to kff.org, 'Allocating CBO's Estimates of Federal Medicaid Spending Reductions Across the States: Enacted Reconciliation Package', published 2025-07-23, modified 2026-06-15). Embedded Datawrapper data in the cached page: California 149,824,000,000 (low 112,368,000,000; high 187,280,000,000; 17% of baseline). Datawrapper chart 36eMx earlier versions: California 164,316,000,000. Chart v0mna: national annual path. Copies in /mnt/zdata/downloads/ggss2026-prop40-adjudicator/.
Why this severity. Concerns the motivation, not any revenue or economic finding. 'Likely $19bn per year' overstates KFF's central annual figure by ~25% and hides its time profile.
17minorplausible · previously known · methodology
The figures showing billionaires pay less tax than average Americans depend on accounting choices that economists debate, and part of the drop in taxes relative to wealth just reflects soaring stock prices. The basic finding still holds up.
Technical detail and evidence
Why it matters, in detail. Tax/wealth = (tax/income) x (income/wealth), so rising valuation multiples lower it even at constant tax rates. Shareholder incidence of the corporate tax raises 1980s rates relative to later ones (us-wealth-taxes-13).
Evidence. Report p.3, fn 4-6; BSYZ (2025) pp.10-11 (Figure 2 discussion); Auditor A scripts/reagan_tax_wealth.py (re-run: 1982-88 means 2.93% income-ranked and 2.81% wealth-ranked; TOI 2010-13 2.31% vs BSYZ 2.7%).
Why this severity. The level comparison (24% vs 30%) is broadly confirmed by administrative data (CORPUS us-wealth-taxes-14). The conventions affect interpretation more than direction, and the splice probably understates rather than overstates the decline.
The authors' response. BSYZ (2025) p.10: the decline in effective rates is robust because 'absolute amounts of taxes paid fell'.
Claims examined and rejected
Problems one auditor raised that did not survive checking.
- No critical or major issue is established; the contested departures belong in a single moderate 'revenue uncertainty' finding (Auditor B, I1). — Partly rejected. The departure risk is graded separately as major (F1): it covers ~25% of the base, the co-authored BSZ paper itself concedes Page's case and scores a -14.5% scenario, and the report omits that row while citing the upside row. Auditor B is right that the $89bn scenario is adverse rather than proven; that is reflected in confidence 'likely', not 'confirmed'.
- Appraisal relief and public-company classification of founder shares are misdescribed, a moderate issue (Auditor B, I4). — Severity rejected (kept as minor, F11). The report's own p.7 text mentions the need to 'demonstrate convincingly', the FTB has classification authority under s.50308(j), and the score values founder shares at market. The defect is overstated certainty with no revenue effect.
- The Swiss-precedent misdescription is a moderate issue (Auditor B, I5). — Severity rejected (minor, F10). The earnings multiplier is close to Swiss practice and the broad feasibility analogy stands; the misdescription does not touch the Forbes-based score.
- The venture-capital causal overclaim is a moderate issue (Auditor B, I3). — Severity rejected (minor, F9). It is a secondary claim, the evidence does rule out a funding collapse, and the revenue estimate is unaffected.
- The authors 'reversed without disclosure' their December 2025 reading of the control-share presumption (Auditor A, I10). — Partly rejected. The December fn 5 was conditional ('If such an interest is valued under the non-public business-entity rules'), pointed to the broad s.50308(j) definition, and expected convertible super-voting shares to be valued at the common-stock value. The July text hardens a hedged position rather than reversing it; the inconsistency is kept as minor (F11).
- Removing Ellison on 'recent reporting' while treating equally public reporting on Page as noise shows an inconsistent rule (Auditor A, I1). — Rejected. Ellison's reported change of domicile dates to March 2023, years before the initiative. Distinguishing a long-completed move from late-December 2025 steps is legally coherent; the defect in F1 is the missing sensitivity, not the different treatment.
- Treaty and apportionment challenges by the 24 non-citizen residents (~$150bn, ~$7bn of revenue) are a material legal risk to revenue (Auditor A, I2). — Rejected as a material risk. US tax treaties generally cover federal income taxes (non-discrimination articles apart), and the Act does not distinguish by nationality. Alternative apportionment requires clear and convincing proof that the wealth was neither substantially accumulated in California nor substantially sustained there for 365 days in the 48 months before the valuation date (s.50306(b)(3)), which long-term residents cannot meet. The broader legal-risk point is retained (F4).
- The statement that only one billionaire is in the $1.0-1.1bn ramp is wrong (six in the Sept 21, 2026 data) (Auditor A, I13). — Partly rejected. The July-1 count is not observable. Six on Sept 21, 2026 shows the sentence is stale (it is copied from the Dec 2025 version), but its substance holds: the ramp changes the Sept 21 liability by $0.23bn. Recorded as a stale detail in F12.
- Income-tax losses compared without discounting are a moderate issue (Auditor A, I7). — Severity rejected (minor, F8). On the authors' SEC-based estimates, even permanent losses from the reported leavers are ~10% of the one-time revenue in PV (~25% at the extreme r-g = 2% with BSZ's aggressive set), so the net-positive conclusion survives; the defect is the 'very small' wording.
- The 61% public share in the July text should be checked against current workbook data (Auditor B, passed checks: later workbook figures differ). — Not an issue in itself. The July Table 1 hard-codes 60% and the report says 61%, so the immaterial gap is folded into F2 (shifting justification) and F12 (attribution to 'Forbes'). Later workbook vintages are not treated as July errors.
Audit notes
What reproduced. Reproduced: the headline ($2,307bn x 0.05 x 0.9 = $103.815bn; tracker 2026-07-01: 250, $2,306.999bn, $103.815bn); the version arithmetic (Dec $98.55bn, stated $99bn; Feb $98.19bn, stated $99bn; Apr $98.01bn); the tracker range ($86.99-106.73bn, Jan 1-Jul 16) and Jan-Jul growth (12.27%); annual growth 2023-25 (BSZ Tab1 and tracker year-ends); AGI ($1,946.17bn -> $2,317.9bn); BSZ Table 5 rows and July-1 person values (Page 294.6, Brin 271.7, Zuckerberg 210.3, Huang 170.9; top 4 = 41.07%); CA billionaires' PIT ($3.06bn/yr) and top-4 PIT ($0.268bn/yr); BSYZ 23.8% vs 30.2%, and 2.7/2.0/1.3% of wealth; California ~25% of US billionaire wealth; the ODA upper bound (126 taxpayers, $35.5bn of $122.7bn) and ramp effect ($0.23bn) on 2026-09-21 data; installment charges ($15.6bn if universal); the Swiss formula comparison; and KFF's California allocation ($149.8bn/10 yrs). Approximately reproduced: the 1981-88 taxes/wealth figure (2.8-2.9% vs 3.1%) and the 2010-13 figure (BSYZ text 2.7% vs report 2.8%). Not reproduced or not verifiable: the Bloomberg +7.5% comparison (no Bloomberg data), July-1 Forbes person-level data beyond BSZ's Tab5 worksheet, BSYZ's confidential microdata, KFF's '1.6 million' coverage-loss figure, and the OBBBA '$1 trillion to the top 1%' claim. New sensitivities (scripts/reproduce_core.py): Page out $90.6bn; BSZ row 3 $88.8bn; Page and Brin out $78.3bn; allowance 15/20/30% gives $98.0/92.3/80.7bn; Page and Brin out with a 15% allowance $74.0bn; PV of permanent income-tax losses (BSZ aggressive set) $12.9-25.8bn.
Limits of this audit. Residency, retroactivity and apportionment are legal questions outside this audit's competence; they are treated as revenue risks, not errors, and no court outcome is predicted. No July-1 Forbes person-level snapshot exists: July-1 values for named individuals come from BSZ's Tab5 worksheet in the October 2026 workbook (consistent with the July BSZ text), and other per-person checks use BSZ's 2026-09-21 cross-section and a 2026-10-25 Forbes snapshot. Post-report facts (Forbes relabelling of Page and Brin on 2026-10-11; Forbes Oct 16, 2026 article) are used only to judge whether the risks the report dismissed were real, not to fault the July text for not knowing them. The ODA estimate is an upper bound, because Forbes' 'private' category includes diversified listed assets. The Bloomberg comparison, the KFF coverage-loss figure, the OBBBA distributional claim and Norwegian/Spanish revenue figures were not verified. The SSRN version was not obtainable (HTTP 403); the four PDF versions on Saez's site and in the Wayback Machine were compared. Jaros & Rauh's March 2026 version (the one the authors answered) was not obtained; their October 2026 NBER version was used. The vote (Nov 3, 2026), the Dec 31, 2026 valuation and any collections lie in the future.