David Gamage, Emmanuel Saez, Darien Shanske · Policy report on California bill AB2088 (2019-20), 10 August 2020; SSRN 3782747

The California Extreme Wealth Tax: Revenue, Economic, and Constitutional Analysis

18 problems found: 7 moderate · 11 minor

What the paper claims

AB2088's annual 0.4% tax on worldwide net worth above $30 million (directly held real estate excluded) would raise about $7.5 billion a year (2019 dollars, after a 20% avoidance/evasion allowance). Revenue relative to the size of the economy would be the same in the short and long run.
About 30,000 California families (the top 0.15%) would be liable.
The table gives revenue and liable families under alternative exemption thresholds. In Panel A the threshold is the same for all filers; in Panel B it is halved for non-joint filers.
Economic responses would be limited. California's rising share of US millionaires and billionaires after the 2012 surtax shows 'we are still far from a tipping point'. The 10-year trailing liability makes the tax 'much harder to avoid than the income tax'. The newcomer phase-in, entrepreneur deferral and anti-evasion rules limit harm. Taxing the rich in the recession is better than cutting spending, since 'states cannot borrow through public debt'.
The tax 'was carefully designed to stay clear of constitutional issues'. It taxes residents' worldwide wealth, complies with Prop 13 by excluding real estate and crediting property tax, and credits other jurisdictions' wealth taxes. The California constitution caps the rate on net worth at 0.4%.
Context. California has 12% of the US population but 17% of US millionaires and 25% of billionaires (15.5% and 21% in 2011). California billionaire wealth was $924bn on 29 July 2020, against $706bn in March 2019 and about $300bn in 2011. The CA top-1% income share is above the US share. Billionaires' effective tax rate is 23% against 28% overall. Illustrations: Zuckerberg would owe $340m; fortunes fall about 4% after 10 years and 8% after 20.

Every problem found

1moderateconfirmed · newly identified here · internal inconsistency

The authors say they knocked 20% off their estimate to allow for tax dodging. In their table, though, that cut is not applied to California's 99 billionaires, who hold more than a third of the taxable wealth. Correcting this trims the headline only a little, to about $6.9 billion. It does make the figures for a billionaires-only version of the tax roughly a fifth to a quarter too high.

Technical detail and evidence

Why it matters, in detail. The haircut is applied as reported wealth = 0.8 x true wealth, before the exemption, and only to SCF records. Applying it uniformly removes 0.2 x Forbes wealth (about $138bn of base) at every threshold up to $1.68bn. Panel A revenue then moves from $9.1/7.5/6.2/4.9/3.7/2.4bn to $8.5/6.9/5.6/4.3/3.2/1.8bn (-7% to -24%). Read instead as a liability share (revenue = 0.8 x gross mechanical revenue: $11.8/9.7/7.6/5.9/4.3/2.45bn), the figures are $9.5/7.8/6.1/4.7/3.5/2.0bn. On that reading the headline is 3% conservative but the $1bn row is 22% too high. Either way, the group with the largest mobility and planning margins is assigned zero avoidance.

Evidence. Report p.2 and p.4 notes. Forbes 400 (2019) API: 99 California members, $690.9bn, matching Forbes (G. Tognini, 'Nearly A Quarter Of America's Richest People Live In California', 4 Oct 2019). Adjudicator scripts: forbes_checks.py; best_fit_detail.py (with Forbes x0.8, the best-fitting SCF scales leave the $1bn base at or below $463bn); bound_1bn_row.py (with a haircut Forbes component, even 2x SCF wealth reaches only $563-583bn while putting 318-394 taxpayers in that row); final_reproduction.py, which writes reproduction_table.csv (24-cell RMS log error 0.6%, max 1.2%). The authors' own convention elsewhere is to haircut billionaires: the Warren letter (18 Jan 2019, point 4) puts Forbes wealth 'for tax purposes' at 85% of the Forbes estimate; wealth.do (Sept 2019) applies (1-e) to all records including the Forbes 400; the 2019 OPA workbook applies a 'Uniform discount of 15%'. Their March 2021 AB310 billionaire bracket ($704bn base, 172 taxpayers, from $960bn excluding Musk, 'aged to end of 2022') implies aging of about 1.13 with a haircut but about 0.90 (a decline) without one, which suggests the haircut was applied in 2021.

Why this severity. The headline moves only modestly. Under the authors' own uniform wealth-haircut convention it falls 8% ($7.5bn to $6.9bn). If '20%' is read as 20% of gross liability, it barely moves (+3%, $7.8bn). The threshold table, however, is materially distorted: the $200m and $1bn rows are 7-15% and 18-24% too high under either reading. A billionaire-only design therefore looks more lucrative than the stated assumption supports.

2moderateconfirmed · previously known · documentation gap

The authors did not publish how they built their numbers, which can be rebuilt only by working out several hidden choices. Reasonable alternative choices would put the number of affected families anywhere from about 23,000 to 36,000, and would move revenue by several hundred million dollars either way. The report gives single precise numbers with no margin of error.

Technical detail and evidence

Why it matters, in detail. Reconstructed specification: SCF 2016 with the bill's real-estate exclusion; weights wgt x 1.009^3; networth x (101 - 2.955)/89.24; a 20% pre-exemption haircut on SCF records; 17.5% CA weight; Forbes at face value. This gives, at $30m (Panel A), a base of $1,866bn, 30,142 families and $7.47bn. With the authors' Feb-2019 $94tn aggregate it gives $6.90bn and 25,253 families; with the 2019 SCF vintage, $6.86bn and 23,087. Sampling SE is about +/-$0.50bn and +/-3,500 families. The counts imply irregular Pareto slopes (alpha about 2.2 between $30m and $50m, about 1.0 between $50m and $100m), reflecting lumpy SCF weights.

Evidence. Adjudicator scripts: share_scale_grid.py (searching CA shares of 15-22.5% and aggregates of $88-112tn, the best fit lands at the stated 17.5% and $101tn, RMS error 0.6%) and fine_scale.py. Authors' 2019 materials: rawdata_wealthtax.do and documents_materials_wealthtax-simulation.xlsx ('wealth inflated uniformly to match 2019 aggregate total of $94Tr'). SaezZucman2020JEPData.xlsx, sheet DataF1, 2019 total wealth denominator $93,688bn. Sensitivities: at $94tn the same method gives $6.9bn and 25,253 families (reconstruct_table.py). The actual SCF 2019, released Sept 2020, gives $6.86bn and 23,087 families with Forbes at face value, or $6.30bn with a uniform haircut (scf2019_check.py). A family bootstrap puts the SE of the SCF part at 10.7% (about $0.50bn of revenue) and of the family count at 11.7%, with a 90% range of 24,400-35,900 families (base_composition.py). Claude's independent bootstrap: 11.3% (about +/-$0.47bn).

Why this severity. Once the hidden choices are found, the computation is internally consistent and reproducible, so the headline is not shown to be wrong. But equally defensible inputs move the family count by 15-25% and revenue by about 8-10%, before any behavioural uncertainty. The report's precision (30,443 families; $7.5bn) is not warranted.

3moderateconfirmed · previously known · overclaiming

The report says the tax was built to avoid constitutional problems, but it never discusses the part most likely to be challenged in court: taxing people for up to ten years after they move away. If a court struck that part down, the rest of the tax would survive, but the main protection against wealthy people leaving would be gone.

Technical detail and evidence

Why it matters, in detail. Without an enforceable trailing liability, moving out removes the whole 10-year liability; with it, 45% remains (90%+80%+...+10% = 4.5 of 10 years), so moving avoids 55%. The legal risk therefore bears directly on the extensive-margin elasticity at the top (F4) and on the implicit 0% avoidance allowance for billionaires (F1).

Evidence. Report p.3. Bill s.50310(b)(3)-(4), with severability in s.50304 and s.50310(d). NTU Foundation, https://www.ntu.org/foundation/detail/california-wealth-and-exit-tax-would-be-an-unconstitutional-disaster ; SCOCAblog, https://scocablog.com/exit-taxes-in-california-not-so-fast/ ; North Carolina Dept. of Revenue v. Kimberley Rice Kaestner 1992 Family Trust (U.S. Supreme Court, 21 June 2019). Galle, Gamage, Saez & Shanske (23 Mar 2021), Parts II-III. Cal. Const. art. XIII s.2 (Wayback capture of leginfo).

Why this severity. The omission concerns the feature that underpins the anti-avoidance argument. If that provision were struck down, severability (s.50304, s.50310(d)) would preserve the tax on residents, so the static base is not at stake, but the revenue protection against leavers would be. The legal merits are unresolved, so the impact cannot be quantified.

The authors' response. No AB2088-specific response was found. The 2021 successor proposal shortened the trailing period to four years. A 2026 note by Galle, Gamage & Shanske defends a different design, based on residence on a fixed date, against a right-to-travel critique.

4moderatelikely · newly identified here · speculative assumption

The report says the tax would bring in the same share of the economy every year, starting right away. But the bill lets some owners of private companies delay payment and phases newcomers in, and the tax itself slowly shrinks the fortunes it taxes. Revenue in the first years would probably be noticeably below $7.5 billion, and it would rise and fall with the stock market.

Technical detail and evidence

Why it matters, in detail. A static accrual score is not a cash forecast. The upper bound on deferral-eligible tax is about $2bn of $7.5bn. The shortfall from residence apportionment is unmeasured (for example, 20% of liability at an average fraction of 0.5 would cost 10%). Mechanical erosion is about -4% per decade against the counterfactual. Swings of +/-30% in billionaire wealth imply revenue volatility of about +/-$1bn.

Evidence. Report p.1, p.2 and p.4 notes. Bill s.50307(c)(1)-(2) and s.50310(b)(1)-(3) (Wayback capture, 29 Aug 2020). base_composition.py: business equity is 59% of liable families' non-real-estate net worth, and 43.9% of the SCF-part base belongs to families with business at 80% or more. Codex's illustrative cash scenarios: 10-25% deferral gives $5.6-6.7bn of initial cash. Boll-Saez-Zucman (2026) California billionaire series, end-year wealth: $821bn (2019), $967bn (2020), $1,212bn (2021), $843bn (2022).

Why this severity. This is mostly about timing and the steady-state level. Deferred tax is collected later, with growth, and trailing liabilities eventually offset the newcomer phase-in, so the order of magnitude survives. First-year cash, which the report ties to the fiscal crisis, could plausibly be 10-25% below the static figure.

5moderatelikely · previously known · overclaiming

The report argues the rich will not leave, because California's share of millionaires kept rising after a 2012 tax increase. That trend says little about how people react to a new tax, and a study of the same 2012 increase found real responses. A few billionaires hold a big slice of the taxable wealth, so even a handful of departures could cut revenue noticeably. Two of the richest Californians did move away within months of the report.

Technical detail and evidence

Why it matters, in detail. For a billionaire whose realized income is 1-2% of wealth, a 0.4% wealth tax is equivalent to a 20-40 percentage-point income-tax surcharge, far more than Prop 30's 1-3 points. Moving cuts the 10-year liability by 55% (the sum of k/10) and ends it after that, provided the trailing tax is enforceable (F5). With 13% of the base in four families, revenue risk depends on the extensive margin at the very top, not on average millionaire-migration elasticities.

Evidence. Report pp.2-3. Rauh & Shyu, May 2020 revision, abstract (/mnt/zdata/downloads/gss2020-ab2088/rauh-shyu-may2020.txt). Varner, Young & Prohofsky, Table 4.2. forbes_checks.py: the top 4 contribute $243bn (13.0%) and the top 10 $338bn (18.0%) of $1,874bn. rtb_check.py: on 29 Jul 2020 there were 174 CA-resident billionaires with $958.2bn; Musk ($71.0bn) and Ellison ($72.4bn) held 15%. Departures: the authors' March 2021 report, fn.4 ('Elon Musk who is now listed as a Texas resident'); Boll-Saez-Zucman (2026), on Ellison: 'he left for Hawaii in late 2020'.

Why this severity. This does not overturn the revenue magnitude: the rate is low, and the trailing liability lowers the gain from moving. But the claim that responses will be small is asserted exactly where the base is most concentrated, and where the table assumes zero avoidance (F1). Departures by a handful of billionaires could remove about a tenth of the base.

The authors' response. The authors' March 2021 successor report cites Young, Varner, Lurie & Prisinzano (2016, ASR) as evidence that millionaire migration is small (p.2 n.7). It keeps the 'far from a tipping point' inference and shortens the trailing period to four years.

6moderateplausible · previously known · external inconsistency

The estimate takes magazine estimates of billionaires' fortunes at face value. When rich people die, the wealth reported on their estate-tax returns is typically only about half of what Forbes said. The bill tells officials to value wealth the way the estate tax does, so taxable amounts could be lower than assumed. The authors argue that this gap mostly reflects weak estate-tax enforcement, which their design would fix.

Technical detail and evidence

Why it matters, in detail. The multiplier comparison implies a 2-5x gap in levels. Several factors shrink it: spouses splitting wealth below the filing threshold, trusts and gifts that sit outside the estate, and higher multipliers for the long-lived rich. Even so, the administrative base could be materially below SCF/Forbes measures, especially wealth in completed-gift trusts, which estate-tax rules exclude.

Evidence. Raub, Johnson & Newcomb (2010), NTA Proceedings: 'In aggregate, the average Forbes Ratio was 0.50' (/mnt/zdata/downloads/raub-johnson-newcomb2010/raub2010.txt). IRS SOI estate tax by state, filing year 2019: California 1,282 returns, 18.7% of $159.7bn of gross estates (rerun of Claude's soi_estate_ca_share.py). Bill s.50305(b)(1) and s.50307(b). Claude's estate_benchmark.py.

Why this severity. If assessed values fell short of Forbes and SCF values by more than the 20% allowance, revenue would fall in proportion; a 30% shortfall on the Forbes component alone would cost about $0.8bn (-11%). But what the estate-tax gap means is disputed: it may reflect avoidance under a weakly enforced estate tax, or overstatement by Forbes. This is a plausible, unquantified downside risk.

The authors' response. Saez & Zucman (June 2019, 'Response to Summers and Sarin') argue that estate-tax values are about half of Forbes values because of estate-tax avoidance and collapsed enforcement (65% of the largest estate returns audited in 1975 against 8.6% in 2018). They argue that a comprehensive, no-discount, well-enforced wealth tax would not inherit that erosion.

7moderateplausible · newly identified here · methodology

The bill lets wealthy owners subtract the property taxes their companies already pay on buildings and land. The report mentions this, but the revenue estimate ignores it. That probably lowers true revenue by somewhere between a few percent and about a seventh, though public data cannot pin it down.

Technical detail and evidence

Why it matters, in detail. The credit is about 0.4% x the pro-rata value of indirectly held real estate that pays property tax of at least 0.4% of value. Business equity is about 59% of the SCF-part base. If 25-50% of it is real estate and 50-80% of that carries at least 0.4% property tax, the credit is 7-24% of SCF-part revenue (about $0.35-1.1bn), plus a small amount on public equity. Prop-13 assessment lags push toward the lower end.

Evidence. Bill s.50308(b)-(c); report p.3. base_composition.py: business equity is 59.1% of liable families' non-real-estate net worth, public equity 23.8%. Claude's z1_business_realestate.py (Z.1 Sept-2021 vintage, tables B.103/B.104).

Why this severity. The effect runs one way (it lowers revenue) and is plausibly worth about 3-15% of revenue ($0.3-1.2bn). Public data cannot measure the real-estate share of top business wealth or take-up of the credit, so the impact is not demonstrated.

8minorconfirmed · newly identified here · misrepresentation of source

The report says its 20% allowance for evasion is a bit higher than the IRS estimate for income and corporate taxes. The IRS figure for those taxes is actually about 21%. The slip weakens the justification but does not change the numbers.

Technical detail and evidence

Why it matters, in detail. The all-taxes rate is presented as the income-plus-corporate rate. The tax gap is in any case a compliance rate on income taxes, not a behavioural response to a wealth tax. The implemented pre-exemption haircut already removes about 32% of SCF-part liability.

Evidence. IRS Pub. 1415 (Sept 2019), Table 3 and s.3.4.8 (/mnt/zdata/downloads/irs-taxgap-pub1415/p1415--2019.txt). Report p.2. The sentence is repeated in the March 2021 report.

Why this severity. This affects the justification of the assumption, not the computation. At a 0.4% rate, elasticity-based estimates mostly imply responses below 20%; billionaires are the exception (F1).

9minorconfirmed · newly identified here · internal inconsistency

The 2011 number is measured differently from the later ones. Measured the same way, California's share of millionaires rose even more than the report says.

Technical detail and evidence

Why it matters, in detail. Count share: 14.83% (2011) against 16.78% and 16.72% (2016, 2017). AGI share: 15.46% (2011) against 18.4% and 17.4%.

Evidence. IRS SOI Historic Table 2: TY2011 (11in54cm.xlsx); TY2016-17 (16in54cmcsv.csv, 17in54cmcsv.csv). Claude's soi_ca_millionaires_2011.py, rerun by the adjudicator; Codex's primary_checks.py gives the same 14.83%. Repeated in the March 2021 report.

Why this severity. Measured consistently, the rise is larger (14.8% to 16.7%), so the descriptive point is strengthened, not weakened.

10minorconfirmed · newly identified here · arithmetic or transcription

A footnote gives the wrong US population, but the percentage the report actually uses is correct.

Technical detail and evidence

Why it matters, in detail. The US population is overstated by 3.56m (1.1%). It affects no estimate.

Evidence. Census nst-est2019-alldata.csv, POPESTIMATE2019: US 328,239,523; CA 39,512,223. The 2011 share is 12.081%. The error is repeated in the March 2021 report.

Why this severity. A typo-level slip; the share actually used is correct.

11minorconfirmed · newly identified here · documentation gap

The report's assumption that 17.5% of America's rich live in California is reasonable. But the numbers it cites in support say 16.7-16.8%; other tax data, which it does not cite, support 17.5% or more.

Technical detail and evidence

Why it matters, in detail. The SCF part of revenue ($4.7bn) scales linearly with the share, so +/-0.75 points moves revenue by about -/+$0.2bn. The Forbes part is observed directly.

Evidence. Report p.2 and fn.1. share_scale_grid.py: the best-fit share is 17.5%. Claude's ftb_soi_5m_share.py and soi_estate_ca_share.py, both rerun by the adjudicator (estate-tax return shares 20.4%, 20.1%, 20.0%, 20.8%).

Why this severity. Using 16.75% would lower revenue by about $0.2bn (-2.7%). Shares measured at higher wealth or income levels point to 17.5% or more, so the assumption is reasonable.

12minorconfirmed · previously known · internal inconsistency

The report says nothing is exempt, but homes and other directly owned real estate are. Wealthy people could move money into such property to cut their bill, and the report does not account for this.

Technical detail and evidence

Why it matters, in detail. Moving financial wealth into directly held real estate, or keeping it there, removes it from a base taxed at 0.4% a year. With Prop-13 assessment lags, the effective property tax on such holdings can be below the wealth-tax rate.

Evidence. Report p.2 and fn.6. Bill s.50306 and s.50309. Saez-Zucman Warren letter (18 Jan 2019), point 8 (/mnt/zdata/downloads/sz2019-warren-claude/sz-warren-2019.txt, lines 143-144).

Why this severity. The exclusion is disclosed in footnote 6 and has a constitutional motive. Directly held real estate is a small share of top wealth (14% of net worth for families at $30m or more, per Claude), so the unmodelled shifting margin is real but bounded.

13minorconfirmed · newly identified here · overclaiming

The report says wealth is more unequal in California than elsewhere, but shows evidence only for income. Having many billionaires is not the same as wealth being more concentrated.

Technical detail and evidence

Why it matters, in detail. No within-state wealth distribution is cited. A larger share of the nation's billionaires is compatible with either higher or lower concentration within the state.

Evidence. Report p.1 and fn.1-4.

Why this severity. A contextual claim with no effect on the estimate.

14minorconfirmed · newly identified here · other

The report says states cannot borrow, but they can and do, including through a Federal Reserve programme set up in 2020. What they cannot easily do is run long-term deficits.

Technical detail and evidence

Why it matters, in detail. Short-term liquidity borrowing was available. Sustained deficit financing is constrained by state balanced-budget requirements.

Evidence. Federal Reserve press release, 9 Apr 2020 (/mnt/zdata/downloads/ca-ab2088-law/fed-mlf-april2020.html).

Why this severity. It does not affect the revenue estimate; it narrows one strand of the recession argument.

15minorlikely · previously known · documentation gap

The 'top 0.15%' figure compares a count of families with a count of tax returns. Against households, the affected group would be about the top 0.23%. This is a labelling issue, not an error in the estimate.

Technical detail and evidence

Why it matters, in detail. Mixing a family numerator with a tax-unit denominator makes the population share about 40% smaller than it would be against families or households.

Evidence. Table p.4: N/share is about 20.9-21.8m across rows (Claude's table_identities.py; Codex's check_table.py gives about 21.4m). Frank state data reference units: CA tax units in 2019 = 21.38m (per Claude).

Why this severity. Presentation only. The count is unaffected; only the 'top x%' label depends on the unit.

16minorlikely · newly identified here · data error

The billionaire list used covers only US citizens, so a dozen foreign-citizen billionaires living in California were left out. Including them would slightly raise the estimate, not lower it.

Technical detail and evidence

Why it matters, in detail. The omitted base is about $38.9bn - 12 x $0.03bn = $38.5bn; at 0.4% that is about $0.15bn.

Evidence. forbes_checks.py: on the March 2019 list, 159 CA residents with $750.9bn and 147 CA US citizens with $712.0bn. rtb_check.py: Wayback captures of Forbes' real-time list, 29-30 Jul 2020.

Why this severity. Small, and in the conservative direction (about +$0.15bn of revenue, +2%).

17minorplausible · previously known · other

The claim that billionaires pay a lower tax rate than average comes from the authors' own earlier work, which other researchers dispute, and it actually refers to the 400 richest Americans. It is background, not part of the revenue estimate.

Technical detail and evidence

Why it matters, in detail. The comparison is sensitive to incidence and denominator imputations disputed by Splinter (2020), Kopczuk (2019), Gleckman (2019) and Splinter & Auten (2019).

Evidence. Codex's primary_checks.py (SZ2019AppendixTables.xlsx, TC8(2018)!E152 and E155); Claude (book text). Corpus us-wealth-taxes-12 to -15.

Why this severity. Motivational only; it has no bearing on the revenue estimate.

The authors' response. The authors have defended these conventions in published replies (see the corpus entries).

18minorplausible · previously known · methodology

If the tax pushed some wealthy residents to leave, California would also lose the income tax they pay. The state's overall gain would then be smaller than the $7.5 billion headline, and the report does not mention this.

Technical detail and evidence

Why it matters, in detail. Net fiscal gain = wealth-tax receipts - the change in income tax paid by movers and by stayers who change their realizations. With a small extensive margin the second term is small, but each departing top earner removes both bases.

Evidence. Report pp.1-2. Rauh & Shyu (May 2020) on erosion of the income-tax base after Prop 30. Corpus optimal-tax-behavior checklist: 'Spillovers onto other tax bases ignored'. NTU Foundation (27 Aug 2020) reports that the top 1% pay 46% of California income tax (not independently verified here).

Why this severity. The direction is clear (net revenue is below gross), but the size depends on the migration response, which is uncertain and possibly small at a 0.4% rate with trailing liability.

Claims examined and rejected

Problems one auditor raised that did not survive checking.

Claude I2: the SCF component excludes only owner-occupied home equity, not all directly held real estate as the bill specifies. The evidence offered was the Panel-A fit (RMS log error 0.020 against 0.064) and the Panel-B increments (+13,838 vs +13,723 printed at $20m; +42 vs +42 at $1bn). — The real-estate treatment is not identified once the SCF aggregate is treated as unknown. The authors' own 2019 code rescales the SCF to a target aggregate, and their August-2020 target is undisclosed. With the bill's full real-estate exclusion and an aggregate of about $101tn, all 24 printed cells reproduce within 1.2% (RMS 0.6%), and the best-fitting CA share lands exactly on the stated 17.5%. The best home-equity-only fit has an RMS error of 3.4-4.2%. Claude's Panel-B evidence was selective: at $30m, $50m and $100m the bill's definition matches the printed increments better (e.g. +3,199 vs +2,917 printed at $30m, against +4,572 for home equity only).
Applied with the report's stated method and the bill's definition, the estimate is about $6.3bn a year and about 25,000 families (-16% and -17%). — This combines the confirmed Forbes-haircut effect (F1, about -8%) with the rejected real-estate inference and the authors' earlier $94tn aggregate. Under the reproduced specification, the uniform-haircut figure is about $6.9bn. The family count barely changes (about 30,100), because every billionaire stays above $30m after a 20% haircut. A figure of about $6.3bn and 23,000 families appears only with the actual 2019 SCF, released after the report. That is a sampling and vintage sensitivity (folded into F2), not the documented method.
Replacing the aged 2016 SCF with the actual 2019 SCF barely changes the base, so the aging method is not a source of bias (listed among Claude's checks passed). — This holds only under the $94tn aggregate Claude used. The printed table implies about $98tn for the SCF (against $96.3tn in the 2019 SCF) and a fatter top tail. With the same treatment, the 2019 SCF gives an SCF component 13% smaller and 23,087 rather than about 30,100 families at $30m (scf2019_check.py).
Codex I1 / reproduction log: the base and taxpayer count 'do not reproduce' from the published information. The closest diagnostic ($7.49bn) used 5.5%-per-year wealth aging. — Partly rejected. The documentation gap stands (F2), but the table is reproducible. The authors' 2019 code rescales SCF wealth to an aggregate rather than compounding 1.055^3. With an inferred aggregate of about $101tn, every base and count in both panels reproduces within 1.2%. Codex's 1.055^3 near-match is a coincidence: it inflates SCF wealth by about 15% while haircutting Forbes, and so misses the $1bn row ($1.84bn vs $2.38bn), as Codex itself reported.
Unresolved discrepancy: Forbes' March-2019 list gives $750.9bn for California, not the printed $706bn. — Resolved. US-citizen California residents on that list held $712.0bn (147 people), within 1% of $706bn. The report's July-2020 figure uses the same citizen basis (F13 notes the citizen/resident distinction).
(Examined and dismissed by Codex; Claude agreed) The 20% avoidance allowance is too small for a state-level wealth tax. — The adjudicator concurs that this is not demonstrated for the SCF component. At a 0.4% rate, 20% implies a semi-elasticity of about 50, and the pre-exemption haircut removes about 32% of SCF-part liability. Most estimates are lower: Brulhart et al. (2022), at least 43, implies about 17%; PWBM's 13 implies about 5%. The real gap is the 0% allowance effectively given to billionaires (F1).
(Examined and dismissed by Codex) All existing residents would start on a 10% residence fraction, cutting first-year revenue by about 90%. — The adjudicator concurs. s.50310(b)(1) counts years of residence before enactment. Only recent arrivals are apportioned (F3).
(Examined and dismissed by Codex) Because revenue = 0.4% x base, the 20% haircut must have been forgotten. — The adjudicator concurs. The printed base is net of the haircut, following the wealth.do convention, for SCF records; the haircut was omitted only for the Forbes component (F1).
(Examined and dismissed by Codex) Rauh-Shyu's 45-61% erosion of income-tax revenue should be applied to AB2088. — The adjudicator concurs. That erosion is mostly intensive-margin retiming of realized income (the extensive margin is 9.5% of the total), which a tax on unrealized wealth is not exposed to in the same way. The extensive-margin risk at the very top remains (F4).
(Examined and dismissed by Codex) The AB310 report's $22.3bn is a silent revision of AB2088's $7.5bn. — The adjudicator concurs. AB310/ACA8 is a different proposal: 1%/1.5% above $50m, starting in 2023 on end-2022 wealth. There is no version instability.
(Examined and dismissed by both auditors) The capitalization-method critiques of Saez-Zucman wealth series (us-wealth-taxes-01 to -08) undermine this estimate. — Not applicable. The estimate uses SCF microdata plus the Forbes 400, not capitalized income. The SCF is also the lower of the two series the authors use elsewhere (DINA's US base above $30m is 26% higher, per Claude).

Audit notes

What reproduced. Reproduced. (1) All 12 identities revenue = 0.4% x base. (2) The whole table: 12 bases and 12 taxpayer counts in Panels A and B, within 1.2% (RMS 0.6%). This used the authors' own nominal SCF 2016 file and aging code (weights x 1.009^3; wealth rescaled to an inferred $101tn 2019 aggregate including Forbes, against $94tn in their 2019 simulator), the bill's exclusion of directly held real estate and its debt, a 17.5% CA share, a 20% wealth haircut before the exemption on SCF records only, and the 99 Forbes-400 Californians at face value. (3) Most descriptive statistics: Census shares; IRS 2016/2017 millionaire shares; Forbes totals for 2011, March 2019 and July 2020 within about 1-2%; Frank top-1% shares; the Saez-Zucman 23%/28%; the arithmetic illustrations. Corrected variants: a uniform haircut gives $6.9bn at $30m and $1.8bn at $1bn; the actual 2019 SCF gives $6.9bn and 23,087 families (Forbes at face value) or $6.3bn (uniform haircut); the authors' 2019 $94tn aggregate gives $6.9bn and 25,253 families. Family-bootstrap SE: 10.7% of the SCF part (about $0.5bn), 11.7% of families. Not reproduced or qualified. The authors' program is unpublished, so the aggregate and the Forbes treatment are inferred from the fit. The 2011 '15.5%' is an AGI-dollar share, not a count share (14.83%). The US population figure (331.8m) is wrong. The IRS '15%' benchmark is misdescribed. No cash-flow or dynamic projection exists to reproduce. The legal conclusions cannot be checked numerically.

Limits of this audit. No code or intermediate files from the authors exist for this estimate. The reproduction infers one parameter (the SCF aggregate, about $101tn including Forbes) and attributes the Forbes treatment from the fit and from arithmetic bounds, not from code. SCF sampling uncertainty comes from a naive family bootstrap, not the official replicate weights. The SSRN version was checked only through metadata (4 pages, dated 10 Aug 2020) and is presumed identical. The bill text is the 13 Aug 2020 amendment (three days after the report), read through a Wayback capture. Constitutional merits were not adjudicated. The magnitudes for the property-tax credit, deferral and the estate-tax benchmark are illustrative. Forbes API values could include later revisions, but the 2019 California subset matches Forbes' contemporaneous article. Later facts (the Musk and Ellison departures; billionaire wealth swings in 2020-2022) are used only to illustrate concentration and volatility risk, not to judge the 2020 forecast. No independent official (FTB or LAO) revenue estimate for AB2088 was found.