Comparison of the two analyses of California Proposition 40: Overview · Open Output (here) · Open Analysis: Report and Slides · Open Materials Also in this project: BSZ analysis overview · JR analysis overview

Why do the two analyses of Proposition 40 disagree?

The BSZ analysis (Boll, Saez and Zucman, August 2026) finds that the one-time billionaire tax raises money for California, net of the income tax it changes. The JR analysis (Jaros and Rauh, September 2026, a working paper of the National Bureau of Economic Research, NBER) finds that it loses money, because it counts the income tax lost to billionaires who leave for ever. This page puts both on one definition, the net fiscal effect. Each slider moves one disputed assumption from one analysis's value towards the other's. The bars show what one estimating function returns for each panel.

BSZ main estimate1 $106.8B Net total over 5 years: the wealth tax plus the income tax it changes, in today's dollars
JR main estimate1 −$38.9B Net present value: the wealth tax minus the income tax lost for ever, in today's dollars, averaged over the authors' model runs

Today's dollars (present value): what money received or lost in future years is worth now.

What separates the two numbers

Start from the BSZ estimate and take away what each disagreement costs: each term is that input's effect on the net fiscal effect when switched from the BSZ value to the JR value, averaged over every order the inputs could be switched in (its Shapley value). The terms add up exactly. The JR estimate here is the exact mean of the authors' Monte Carlo, −$39.0B; their seeded run prints −$38.9B. The report gives every step.

Two panels, one estimating function

Each panel holds the assumptions the two analyses disagree on, with the JR values taken from its latest version (September 2026). The BSZ panel starts every slider at the BSZ value, the JR panel at the JR value. A slider has five positions: the two analyses' values and three steps between them. The BSZ value is printed at the left end and the JR value at the right end. Each box opens with its formula; hover over a dial's symbol, or tap it, for what it is and why the two analyses differ. Three inputs that barely move the estimate (non-US-citizen residents, the real-estate deduction, extra income tax from asset sales) share one switch.

The bars show what the common estimating function returns for each panel. The left sliders move only the left bar and the right sliders only the right bar. Move every slider in one panel to its far end and the two bars meet. The report states the function line by line; your browser runs a copy of it, which a test checks against the pipeline.

BSZ: $106.8B

The BSZ analysis's main estimate1 (Boll, Saez and Zucman, August 2026)

With these sliders $106.8B

  • Wealth tax collected = 5% × (W − D) × (1 − a) × q

    The wealth on the list, less what has already left, less avoidance, times the chance the tax survives in court. moreThe four dials multiply each other, so the effect of one depends on where the others stand. W and D are valued on the same Forbes list. Hover over a dial for what each analysis assumes.

    • W, forbes source. Which Forbes list the base comes from, and so the wealth the tax applies to before departures and behaviour. Forbes 2026: the BSZ analysis values California's billionaires on 1 July 2026 (250 people, $2,157 billion without the non-US citizens). Forbes 2025: the JR analysis takes the 2025 list (212 people, with Ellison, Houston and Snyder removed and Sacks added) and grows it 7% to the 31 December 2026 valuation date ($2,027 billion). Two positions: one list or the other. Kind: data.
      Forbes 2026Forbes 2026Forbes 2025
    • D, departed wealth. Billionaires counted as already gone before the 1 January 2026 residency date, whose wealth the tax cannot reach. Three stops. None: the BSZ analysis's main estimate taxes every resident on that date, since an announced move is not a legal change of residence. BSZ four: Page, Thiel, Hankey and Kalanick, the BSZ analysis's own aggressive scenario (Table 5 row 3), $334B on the BSZ list. JR seven: those four plus Brin, Spielberg and Sacks, the JR analysis's confirmed departures, $614B on the BSZ list and $578B on the JR list. Each stop is valued on the list chosen by the Forbes source dial. Kind: guesswork resting on data.
      NoneNoneJR seven
    • a, no avoidance. The share of the tax lost to avoidance and evasion other than leaving. The BSZ analysis allows 10%. The JR analysis allows none. Kind: guesswork.
      10.0%10.0%0.0%
    • q, litigation. The chance that the Act survives a constitutional challenge, applied to the revenue. The BSZ analysis takes the Act as written and collected in full. The JR analysis weights the revenue by 50%, the accounting threshold for a tax that is more likely than not to stand; its simulation draws the revenue anywhere between zero and the ceiling, which has the same average. Kind: guesswork.
      100%100%50%

  • Income tax lost = f × C × A(H)

    The share of billionaires' income tax that leaves, times that income tax per year, counted over H years. moreA(H) is what $1 a year for H years is worth today, at the 1.5% to 4.5% discount rates both analyses use: about 5 for 5 years, about 37 for ever. The three dials multiply each other. In the BSZ analysis f also follows D in the box above. Hover over a dial for what each analysis assumes.

    • f, share lost. How much of billionaires' yearly California income tax the state loses when some of them leave. Two rules, so two positions. What leavers paid (the BSZ analysis): the state loses the income tax the people who leave were paying, from their SEC filings and a wealth-based rate for the rest, plus its 5% of movers. With nobody gone that is $0.15B a year; at the D dial's middle stop, $0.36B a year more. 45% of all (the JR analysis): the state loses 45% of every billionaire's income tax, the middle of its 30% to 60% range, whoever leaves. Kind: data (BSZ) against guesswork (JR).
      what leavers paidwhat leavers paid45% of all
    • C, income tax level. Billionaires' yearly California income tax, the amount at risk each year. The BSZ analysis: about $3.1 billion from tax data, since top wealth holders earn about half of what top earners do. The JR analysis: $4.55 billion, the midpoint of a $3.3 billion to $5.8 billion range from two model bounds. Kind: research (BSZ) against guesswork (JR).
      $3.06B/yr$3.06B/yr$4.55B/yr
    • H, horizon. How many years the income tax loss is counted. The BSZ analysis: 5 years, about the payment period, with responses temporary. The JR analysis: for ever, discounted at 1.5% to 4.5%. The slider's five settings are 5, 10, 20 and 50 years, and a perpetuity. Kind: scenario.
      5 years5 yearsperpetuity
  • Three inputs that each move the estimate by less than $7 billion when moved alone: the non-US-citizen residents ($150 billion of wealth; the BSZ analysis taxes them, the JR analysis leaves its 28 international billionaires out), the real-estate deduction (none in the BSZ analysis; the JR analysis nets out directly held homes, about 0.5% of the base) and the extra income tax from assets sold to pay the tax (the BSZ analysis counts it, the JR analysis does not). Off keeps this panel's own values; on takes the other analysis's values for all three.
    BSZ valuesBSZ valuesJR values
  • Show the three inputs one by one
    • The non-US-citizen residents: $150 billion of wealth that the BSZ analysis taxes and the JR analysis leaves out with its 28 international billionaires. Kind: data.
      $150B$150B$0B
    • The real-estate deduction: none in the BSZ analysis; the JR analysis nets out directly held homes, about 0.5% of the base. Kind: data.
      0.0%0.0%0.5%
    • Extra income tax from assets sold to pay the tax: the BSZ analysis assumes a third of the tax is paid by selling assets, whose gains are taxed; the JR analysis does not count this. Kind: guesswork.
      1/31/30%

Starts at the BSZ analysis's net total, $106.8B.

The BSZ panel's estimate is $106.8B and the JR panel's is −$39.0B.
Net fiscal effect, $ billions. Left bar: the BSZ panel. Right bar: the JR panel.

JR: −$38.9B

The JR analysis's main estimate1 (Jaros and Rauh, NBER, September 2026)

With these sliders −$39.0B

  • Wealth tax collected = 5% × (W − D) × (1 − a) × q

    The wealth on the list, less what has already left, less avoidance, times the chance the tax survives in court. moreThe four dials multiply each other, so the effect of one depends on where the others stand. W and D are valued on the same Forbes list. Hover over a dial for what each analysis assumes.

    • W, forbes source. Which Forbes list the base comes from, and so the wealth the tax applies to before departures and behaviour. Forbes 2026: the BSZ analysis values California's billionaires on 1 July 2026 (250 people, $2,157 billion without the non-US citizens). Forbes 2025: the JR analysis takes the 2025 list (212 people, with Ellison, Houston and Snyder removed and Sacks added) and grows it 7% to the 31 December 2026 valuation date ($2,027 billion). Two positions: one list or the other. Kind: data.
      Forbes 2026Forbes 2025Forbes 2025
    • D, departed wealth. Billionaires counted as already gone before the 1 January 2026 residency date, whose wealth the tax cannot reach. Three stops. None: the BSZ analysis's main estimate taxes every resident on that date, since an announced move is not a legal change of residence. BSZ four: Page, Thiel, Hankey and Kalanick, the BSZ analysis's own aggressive scenario (Table 5 row 3), $334B on the BSZ list. JR seven: those four plus Brin, Spielberg and Sacks, the JR analysis's confirmed departures, $614B on the BSZ list and $578B on the JR list. Each stop is valued on the list chosen by the Forbes source dial. Kind: guesswork resting on data.
      NoneJR sevenJR seven
    • a, no avoidance. The share of the tax lost to avoidance and evasion other than leaving. The BSZ analysis allows 10%. The JR analysis allows none. Kind: guesswork.
      10.0%0.0%0.0%
    • q, litigation. The chance that the Act survives a constitutional challenge, applied to the revenue. The BSZ analysis takes the Act as written and collected in full. The JR analysis weights the revenue by 50%, the accounting threshold for a tax that is more likely than not to stand; its simulation draws the revenue anywhere between zero and the ceiling, which has the same average. Kind: guesswork.
      100%50%50%

  • Income tax lost = f × C × A(H)

    The share of billionaires' income tax that leaves, times that income tax per year, counted over H years. moreA(H) is what $1 a year for H years is worth today, at the 1.5% to 4.5% discount rates both analyses use: about 5 for 5 years, about 37 for ever. The three dials multiply each other. In the BSZ analysis f also follows D in the box above. Hover over a dial for what each analysis assumes.

    • f, share lost. How much of billionaires' yearly California income tax the state loses when some of them leave. Two rules, so two positions. What leavers paid (the BSZ analysis): the state loses the income tax the people who leave were paying, from their SEC filings and a wealth-based rate for the rest, plus its 5% of movers. With nobody gone that is $0.15B a year; at the D dial's middle stop, $0.36B a year more. 45% of all (the JR analysis): the state loses 45% of every billionaire's income tax, the middle of its 30% to 60% range, whoever leaves. Kind: data (BSZ) against guesswork (JR).
      what leavers paid45% of all45% of all
    • C, income tax level. Billionaires' yearly California income tax, the amount at risk each year. The BSZ analysis: about $3.1 billion from tax data, since top wealth holders earn about half of what top earners do. The JR analysis: $4.55 billion, the midpoint of a $3.3 billion to $5.8 billion range from two model bounds. Kind: research (BSZ) against guesswork (JR).
      $3.06B/yr$4.55B/yr$4.55B/yr
    • H, horizon. How many years the income tax loss is counted. The BSZ analysis: 5 years, about the payment period, with responses temporary. The JR analysis: for ever, discounted at 1.5% to 4.5%. The slider's five settings are 5, 10, 20 and 50 years, and a perpetuity. Kind: scenario.
      5 yearsperpetuityperpetuity
  • Three inputs that each move the estimate by less than $7 billion when moved alone: the non-US-citizen residents ($150 billion of wealth; the BSZ analysis taxes them, the JR analysis leaves its 28 international billionaires out), the real-estate deduction (none in the BSZ analysis; the JR analysis nets out directly held homes, about 0.5% of the base) and the extra income tax from assets sold to pay the tax (the BSZ analysis counts it, the JR analysis does not). Off keeps this panel's own values; on takes the other analysis's values for all three.
    BSZ valuesJR valuesJR values
  • Show the three inputs one by one
    • The non-US-citizen residents: $150 billion of wealth that the BSZ analysis taxes and the JR analysis leaves out with its 28 international billionaires. Kind: data.
      $150B$0B$0B
    • The real-estate deduction: none in the BSZ analysis; the JR analysis nets out directly held homes, about 0.5% of the base. Kind: data.
      0.0%0.5%0.5%
    • Extra income tax from assets sold to pay the tax: the BSZ analysis assumes a third of the tax is paid by selling assets, whose gains are taxed; the JR analysis does not count this. Kind: guesswork.
      1/30%0%

Starts at −$38.98B, the exact average of the authors' Monte Carlo simulation (a model run many times with its uncertain inputs drawn at random). The authors' own run of it gives −$38.9B, the main estimate in the header; the two differ by sampling error.

Notes on the two estimates and the conventions

1 Each analysis's own main estimate, defined its own way. The BSZ analysis's is a net total over 5 years: the wealth tax (Table 5 row 1), plus extra income tax from assets sold to pay it, minus the income tax lost to billionaires who leave, for 5 years at 3%. BSZ's own headline is the wealth tax alone; the net total is this OPA's choice, explained on the BSZ analysis overview. The JR analysis's is a perpetuity net present value, averaged over the authors' Monte Carlo draws (Jaros and Rauh, NBER chapter c15504, Tax Policy and the Economy vol. 41, September 2026).

The net fiscal effect puts both on one definition, and the panels start at its two ends. The BSZ end is $106.8B and the JR end −$39.0B, a gap of $145.8B. The BSZ end discounts the same 5 years over the JR analysis's 1.5% to 4.5% rate range instead of at 3%. The JR end is the exact average of the Monte Carlo whose own run gives −$38.9B.

Conventions. All amounts are nominal billions of dollars. The wealth tax is counted at face value when assessed and undiscounted, as both analyses do (the BSZ analysis values the base on 1 July 2026; the JR analysis grows its 2025 Forbes base by 7% to 31 December 2026). Present values are at the assessment date. The 1.5% to 4.5% range is what the JR code draws; the code labels it r − g and the text r (the JR analysis's mismatch 3).

Notes and data

Where the numbers come from. Every number on this page is computed by comparison/R/ from the two analyses' export contracts (bsz-analysis/export/r/, rjkdc-analysis/export/r/: the files each pipeline writes for others to read) and the transcriptions in comparison/data/. The estimating function is score_common() in comparison/R/model.R, with a Python twin (R and Python agree to 1e-9 on every exported number; see comparison/tests). The browser runs model.js, a copy of it, and a test runs that copy with node over 600 slider mixes against the R values, to 1e-9. The report derives each number step by step and shows the bridge from the BSZ end to the JR end; the materials page lists the code, tests and files.

Data. ends.csv (the two ends, at each horizon setting), slider_sample.csv (300 slider mixes per panel with their values, the sample the tests run model.js on), inputs.csv (every disputed input and its sources), DISPUTES.md (the crosswalk of the disagreements, one row each, with what each analysis says and where), bridge.csv (the bridge, step by step), comparison-data.js (what this page reads). A test fails if comparison-data.js or this page differs from the generator's output.

Kinds. Each slider's explanation ends with the kind of its two values: data (settled by observable facts), research (a published estimate), guesswork (asserted without data or a study behind it) and scenario (a choice of which world is estimated).