California Proposition 40 billionaire tax calculator

Interactive calculator for California's Proposition 40 (the 2026 Billionaire Tax Act, Initiative 25-0024A1) on the November 3, 2026 ballot: a one-time tax of 5% of the entire net worth of California residents (as of January 1, 2026) worth $1 billion or more, valued on December 31, 2026. Between $1.0 billion and $1.1 billion the rate ramps from 0% to 5% (RTC §50301(b) as proposed).

Live app: https://policyengine.org/us/california-wealth-tax

What the model does

One model, scored person by person, in two stages.

• Stage 1: one-time wealth tax. The residency roster is everyone Forbes listed in California on January 1, 2026. Each person is valued at their latest Forbes worth, wherever Forbes lists them now; people Forbes has added to its California list since are added, with their January 1 residency assumed; people Forbes no longer lists at all are carried at the last positive value in the stored daily history. Directly held real property is excluded, then the $1 billion threshold and rate ramp apply to each observed wealth-tax unit. Documented separately listed spouses share one threshold and rate (Jay-Z/Beyoncé and Lynda/Stewart Resnick); other family assets remain unobserved. The model then applies an optional erosion haircut, the legal assumptions, and the payment election (lump sum due in 2027, or five installments with a 7.5% charge on the unpaid balance). The legal assumptions are a probability that the measure survives legal challenge, which scales expected receipts and nothing else, and two switches that take people out of the base without moving them: people who left after January 1, and non-US citizens. Each defaults to the measure as written. With stage 2 off, the headline is nominal receipts.
• Stage 2 (optional): future California income tax. The cohort's total California income tax is an input ($4.3B a year by default; Rauh et al. and Boll, Saez and Zucman overlap on it). It is divided among people either by wealth (Rauh et al.'s f × C) or, by default, with filings-based estimates for the four largest fortunes (data/income_tax_filings.json, from Boll, Saez and Zucman's Table 3) and the rest by wealth. A mover's share is the loss; the stream is attributed, grown, discounted, and netted against receipts, both in present value as of 2026. Income-tax departures remain person-based: a departing spouse can lose California income tax while the unit's wealth-tax liability remains because the other spouse was a resident on January 1. An optional total share of the cohort's income tax lost to departures (Jaros and Rauh's f) counts documented and modeled departures first; any remainder is a proportional share of the people still paying who leave after January 1, owing the wealth tax but no longer California income tax.

The baseline takes no position on contested residency. Eleven documented claims that a roster member was not a resident on January 1, 2026 are individual adjustments, each shown with its evidence and source (lib/residencyAdjustments.js, data/billionaire_metadata.json). Larry Ellison is the one exclusion from every base: Forbes lists him in Florida and both published estimates now exclude him.

The "Berkeley assumptions" and "Hoover assumptions" starting points apply the assumptions of Galle, Gamage, Saez and Shanske and of Jaros and Rauh to this calculator's roster. They are settings inside the same model, not replications of those papers' rosters. Since October 21, 2026 the Hoover set follows Jaros and Rauh's NBER Working Paper 35899 (October 2026), as in their central revenue estimate: their departures and pre-filing exits out of the base, Mark Zuckerberg and non-US citizens left out, 7% a year wealth growth to the valuation date, a 50% chance the measure survives challenge, no unannounced departures in the wealth-tax base, and 45% of a $4.55B cohort income tax lost, divided by wealth. Their NPV section instead centers revenue on half of the $72B ceiling with Zuckerberg and the pre-filing exits in (their formula gives a point NPV of −$32.25B at f = 0.45, C = $4.55B and r − g = 3%) and reports a mean over draws of revenue, f, C and r − g (−$38.9B), so the set's point value is not their published mean. Before, the set followed Rauh et al.'s March 2026 paper; that set stays in lib/presets.js as HOOVER_MARCH_2026_ASSUMPTIONS, and the tests still score it.

Related work

Two other analyses compare versions of the Berkeley and Hoover estimates. The page lists them at the bottom, with summaries kept in lib/relatedWork.js:

• Fernando Hoces de la Guardia, California Proposition 40, Reconciliation of the Two Analyses (Open Policy Analysis report, October 2026; comparison report, code).
• Jeffrey L. Hoopes, Galle v Rauh: Comparing Two Revenue Estimates for California Billionaire Taxation (SSRN Working Paper 6428578, March 2026).

Data

• public/snapshots/<date>.json: daily Forbes California snapshots, indexed in public/snapshots/index.json. 2026-01-01.json is the residency roster.
• data/billionaires_live.{json,csv} and data/billionaires_live_meta.json: the latest snapshot.
• data/roster_valuations.json: current Forbes worth, in any state, for everyone on the residency roster.
• data/billionaire_metadata.json: per-person residency evidence with sources.
• data/citizenship_by_name.json: citizenship by name, for snapshots before 2026-09-17, which carry no citizenship field. The daily fetch only adds names to it, so for everyone already in it a stored snapshot scores the same whatever Forbes lists later. People not in it count as US citizens; a wrong entry is corrected by hand.
• data/wealth_tax_households.json: sourced, separately valued spouse groups.
• data/billionaires_rauh.json: the October 17, 2025 list with Rauh et al.'s directly held real estate values (personal residences; a lower bound).
• data/income_tax_filings.json: filings-based California income tax for the four largest fortunes, 2019-2025, with source.
• data/income_tax_lookup.json: California income tax at billionaire-scale incomes, precomputed from PolicyEngine (used by the uniform-yield method).

.github/workflows/update-forbes.yml refreshes the Forbes data daily. It runs the fetcher's tests first, and scripts/check_snapshot_sanity.py must pass before anything is committed. The fetch retries a Forbes server error (5xx), a rate limit (429), a network or TLS failure, or a truncated or non-JSON body up to three times, waiting 30, 60 and 120 seconds, before the run fails; any other client error, such as a 404, fails at once. Names are joined on a normalized key (aliases, trailing "& family", case), because Forbes changes display names.

The job checks out the branch's tip when it starts, not the commit it was queued for, and rebases its data commit onto the current tip before pushing, re-running the sanity gate on the result. A push made with the workflow's token starts no push run of ci.yml, so after pushing a data commit the job dispatches ci.yml on it. A failed daily run, or a failed CI run on its data commit, opens a "Daily Forbes update failed" issue (scripts/report_daily_update.py). If the issue is still open two days later, the next failure assigns it to a maintainer and mentions them, and mentions them again every week the outage lasts. The next good run closes it: CI passing on its data commit, or no new data while the current data is still fresh. A CI pass leaves open an issue opened after its data commit about something else. If the daily workflow stops running at all (disabled, or GitHub drops its scheduled runs), nothing reports it.

Known gap: the January 1, 2026 list is a backfill of US-citizen Forbes profiles, so non-US citizens Forbes places in California enter the base only through the current list, with their January 1 residency assumed.

Architecture

app/
├── page.js                  # Scenario state, URL sync, data loading, layout
├── components/
│   ├── LiveBridge.js        # The bridge between two assumption sets (hero)
│   ├── WaterfallChart.js    # The bridge's waterfall
│   ├── ResultStrip.js       # Your scenario's number, start-from sets, data, link
│   ├── AssumptionPanels.js  # One panel per assumption group
│   ├── Slider.js            # Slider with sourced quick picks
│   ├── Heatmap.js           # Net present value over two assumptions
│   └── BillionaireTable.js  # Person-level table with valuation source flags
lib/
├── scenario.js              # One scorer: parameters in, results out
├── presets.js               # The statutory, Berkeley and Hoover assumption sets
├── bridge.js                # Assumption groups and the Shapley bridge
├── calculator.js            # Receipt schedule, present values, headline
├── microModel.js            # Roster join, person-level tax and income tax
├── residencyAdjustments.js  # Documented residency claims with sources
├── departureResponse.js     # Additional migration response mappings
├── scenarioUrl.js           # Shareable URL parse/serialize
├── incomeTaxLookup.js       # California income tax interpolation
└── paperNumbers.js          # Every number quoted in the paper
scripts/
├── fetch_forbes.py          # Daily Forbes fetch and roster valuations
├── check_snapshot_sanity.py # Gate between fetching and publishing
├── check_snapshot_freshness.py
├── report_daily_update.py   # Opens, escalates and closes the outage issue
├── paper_numbers.mjs        # Writes paper/paper_numbers.json
└── precompute.py            # Year-keyed income-tax lookup generator

The app is static. PolicyEngine-dependent calculations are precomputed into JSON and loaded client-side.

Present values are expressed in 2026 dollars. Nominal wealth-tax receipts are deflated using the inflation setting before real discounting; the default explicitly assumes constant prices (0% inflation), and 2% is an illustrative quick pick rather than a forecast. Income-tax growth and losses are already real.

To regenerate the income-tax table, run uv run --with "policyengine[us]" python scripts/precompute.py. The generator writes the year-keyed schema consumed by the app. Schema tests use the committed tax amounts without claiming to revalidate them against new law.

Development

bun install
bun run dev
bun run test
uv run --with pytest python -m pytest scripts/test_fetch_forbes.py -q
bun run lint
bun run build

The paper

paper/california-wealth-tax-ssrn-draft.qmd quotes only calculator figures that lib/paperNumbers.js computes from the dated snapshot; lib/paperNumbers.test.js fails if the text and the code disagree.

The paper is one version, dated to its snapshot (PAPER_SNAPSHOT_DATE, now October 23, 2026), and scores the calculator's current assumption sets on it: the Hoover set follows Jaros and Rauh's NBER version. To move the paper to a newer snapshot, copy public/roster_valuations/<date>.json to paper/data/roster_valuations_<date>.json, set PAPER_SNAPSHOT_DATE, rerun the script below, and rewrite the text until the test passes.

bun scripts/paper_numbers.mjs   # regenerate paper/paper_numbers.json
bun run paper:render            # requires Quarto

License

The code is released under the MIT license. The working paper's text (paper/ and the rendered public/papers/) is released under CC BY 4.0.

Data from third-party sources keep their sources' terms:

• Forbes-derived files: public/snapshots/, data/billionaires_live*, the roster valuations (data/roster_valuations.json, public/roster_valuations/ and paper/data/roster_valuations_*.json) and data/citizenship_by_name.json;
• Boll, Saez and Zucman's filings-based estimates, in data/income_tax_filings.json;
• Rauh et al.'s real estate values, in data/billionaires_rauh.json.

scripts/precompute.py writes data/income_tax_lookup.json with the policyengine package, which is AGPL-3.0 and builds on policyengine-core, a fork of OpenFisca-Core. The documented command, uv run --with "policyengine[us]", installs it only for that run. The repository does not include it, and the app reads only the JSON the script writes.