[MaxGhenis]

This adds a California breakdown of the 2026 labor-to-capital shift, a within-state variant of the shift, an accounting that counts health spending and TANF, and a rerun of all of it on the current PolicyEngine bundle: policyengine 6.16.0, policyengine-us 2.2.1 and the 2026-09-15 Microcosm national release. It leaves the site alone.

What it adds

• analysis/california_shift.py runs the sweep's shift (labor_capital_shift._apply_shift) and fiscal accounting (fiscal.revenue_components, fiscal.state_revenue_components) on the installed bundle, with one fresh microsimulation per level. For California it records:
  • the employee SDI contribution (1.3% of wages, no cap), which sits outside the sweep's state measure (household_state_tax_before_refundable_credits holds state income tax and use tax only);
  • the change by within-state market-income group (bottom 50%, 50th to 90th percentile, 90th to 99th, top 1%), with the labor income each group gives up and the capital income it receives;
  • federal net revenue: the all-government change less every state's net change and less the state payroll and local taxes that the all-government total counts and no state's line holds. federal_net_published_definition_change_b keeps the site's definition (all governments less the states), and the drift table below uses it.
• A second measure, the *_with_health_tanf fields, adds what the sweep's state measure leaves out. Household net income excludes health benefits (gov.simulation.include_health_benefits_in_net_income is false), so Medicaid, CHIP, the Medicare Savings Programs, ACA premium tax credits and state premium subsidies sit in no government's total, and tanf sits in household_benefits but in no state's line, so the sweep books it federal. The new fields charge each state:
  • its share of Medicaid, CHIP and Medicare Savings Program costs under federal matching rules (state_benefit_cost);
  • the whole cost of coverage it funds alone for people whose immigration status rules out federal matching, while CHIP's unborn-child coverage keeps its federal match;
  • its own health subsidies, from healthcare_benefit_value's list (for California, the Covered California premium subsidy);
  • the TANF change its line doesn't already hold, because the federal block grant is a fixed amount. The all-government total adds every program in healthcare_benefit_value and counts TANF once.
• Medicaid at the baseline's cost per enrollee. policyengine-us spreads each state's fixed Medicaid spending over the enrollees in each simulation (medicaid_slcsp_state_denominator), so on its own a get_branch branch that adds enrollees cuts every enrollee's cost and keeps the state's total. Each shifted branch here takes the baseline's denominator, so a new enrollee adds cost.
• A within-state variant (_apply_within_state_shift, --within-state-levels, default 10% and 50%). The sweep scales every positive capital-income dollar in the country by one factor, so a state whose share of national capital income exceeds its share of labor income gains market income from other states. The variant gives each state's freed labor income to that state's own capital holders, so each state's market income stays fixed.
• analysis/california_summary.py prints a run against the published sweep: California with and without SDI, both measures, the federal split, groups, drift from the site, and California's gain split into income moving within the state and income moving between states.
• analysis/california_chart.py draws California with and without SDI, against all other states unless --california-only drops that line, on either measure. A run defaults to health and TANF; the sweep holds only the study's measure.
• analysis/outputs/california_shift_pe6.16.0.json, the rerun (state and group aggregates only).

Results (policyengine-us 2.2.1, 2026)

National allocation with health spending and TANF:

Shift	California	California with SDI	All other states	Federal	All governments
10%	+$2.10B	−$0.15B	−$16.90B	−$175.0B	−$194.8B
20%	+$4.62B	+$0.12B	−$26.66B	−$343.7B	−$376.0B
50%	+$15.36B	+$4.10B	−$93.33B	−$863.5B	−$969.9B
100%	+$47.19B	+$24.67B	−$243.19B	−$1,684.4B	−$1,977.7B

California's parts at each level (the net also holds the use tax, whose change stays under $0.02B):

Shift	State income tax	Refundable credits	Medi-Cal	Premium subsidy	CalWORKs	SSP and other benefits	Net
10%	+$5.40B	+$0.24B	+$2.80B	−$0.03B	+$0.16B	+$0.12B	+$2.10B
20%	+$11.73B	+$0.58B	+$5.61B	+$0.04B	+$0.44B	+$0.44B	+$4.62B
50%	+$36.39B	+$2.13B	+$16.34B	+$0.10B	+$1.53B	+$0.92B	+$15.36B
100%	+$101.29B	−$0.13B	+$45.69B	−$0.08B	+$4.94B	+$3.70B	+$47.19B

California comes out ahead at every level. Its top 1% hold 6.3% of the state's labor income and 39.8% of its capital income. At 10%, that group pays $7.92B more state income tax, the 90th to 99th percentile pays $2.30B more, and the 50th to 90th percentile pays $4.40B less. Medi-Cal, CalWORKs and the premium subsidy take $2.94B of the gain at 10%. SDI contributions fall $2.25B at 10%, so counting them turns the gain into −$0.15B.

This run moves $0.07B of federal Medi-Cal share to California at baseline, for people whose status rules out federal matching, and that amount changes by +$0.00B at 10% and +$0.65B at 100%. The line doesn't estimate California's actual state-only Medi-Cal cost. Microcosm treats anyone who reports Medicaid in the CPS as documented, and from 2026 policyengine-us's enrollment freeze keeps undocumented adults without existing coverage out at every income (is_ca_medicaid_immigration_status_eligible), so only the groups the freeze exempts can enter the line as incomes fall.

National allocation on the study's measure (no health or TANF):

Shift	California	California with SDI	All other states	Federal	All governments
10%	+$5.04B	+$2.79B	−$2.10B	−$126.2B	−$128.3B
20%	+$10.71B	+$6.21B	+$1.39B	−$243.6B	−$241.7B
50%	+$33.34B	+$22.08B	+$4.33B	−$547.6B	−$538.4B
100%	+$97.73B	+$75.21B	+$86.95B	−$743.0B	−$655.6B

Within-state allocation:

Shift	California	California with SDI	California with health and TANF	All other states	All other states with health and TANF	States losing	States losing with health and TANF
10%	+$5.39B	+$3.14B	+$2.46B	+$2.52B	−$11.67B	18	44
50%	+$35.17B	+$23.91B	+$17.19B	+$27.83B	−$69.47B	15	46

Under the national allocation, California's households give up $179.9B of labor income at 10% and receive $176.9B of capital income, so $3.1B of market income leaves the state. On the study's measure that costs California $0.35B of the $5.39B it collects when each state keeps its own income.

At every level the committed run agrees with itself: California's pipeline figure matches its household frame, the per-state sum matches the national state lines, and the within-state shift holds California's market income fixed.

Drift from the published sweep (policyengine 5.0.1, policyengine-us 1.764.6)

The study's measure, with federal on the site's definition:

Shift	California, published → rerun	All other states, published → rerun	Federal (site's definition), published → rerun
10%	+$5.00B → +$5.04B (+0.8%)	−$2.10B → −$2.10B	−$138.2B → −$131.3B (5.1% smaller)
20%	+$10.61B → +$10.71B (+1.0%)	+$1.40B → +$1.39B	−$266.8B → −$253.9B (4.9% smaller)
50%	+$32.39B → +$33.34B (+2.9%)	+$4.30B → +$4.33B	−$628.8B → −$576.0B (8.4% smaller)
100%	+$96.81B → +$97.73B (+0.9%)	+$86.66B → +$86.95B	−$973.2B → −$840.3B (13.7% smaller)

The state figures barely move, and the federal loss shrinks by 5% to 14%. This PR neither diagnoses the federal drift nor refreshes the site's sweep.

Limits

• The shift sets employment_income and self_employment_income on each branch, so anything that reads their *_before_lsr inputs sees pre-shift earnings. policyengine-us's default TANF earned-income list reads those inputs, as do Massachusetts's, Montana's and South Carolina's lists, so those states' TANF benefits respond to the added capital income and not to the wage cut, and can fall. The TANF, other-states and all-government lines and the study measure's federal lines miss the response to the wage cut. SNAP's gross self-employment income reads those inputs too. California's CalWORKs list reads the shifted variables, so California's line holds the response. The published sweep's shift has the same limit, and the site refresh should decide whether to move it.
• If the state totals that medicaid_cost spreads count federally matched emergency care for people federal law otherwise excludes (42 USC 1396b(v)), moving their whole federal share overstates the state's cost by an amount this accounting can't measure.
• or_healthier_oregon_cost has no enrollment test, and policyengine-us also enrolls Oregon's undocumented residents in Medicaid. This run drops the Healthier Oregon cost for those people, which affects Oregon only.
• policyengine-us 2.2.1 counts Massachusetts's TAFDC in both tanf and household_state_benefits, and leaves five state premium subsidies out of household_health_benefits. This run charges TAFDC once and counts the five subsidies. On main, Count MA TAFDC once in household benefits policyengine-us#9478 and Register the remaining five state premium programs in household_health_benefits policyengine-us#9514 change both lists, and the code handles either version.

Invariants

Property-based (Hypothesis) and example tests in analysis/tests/test_california_shift.py:

• _weighted_rank puts every household strictly inside (0, 1), and a household with a higher value always ranks higher.
• The four groups partition California's households: for every field, the group changes add up to the state total, and the groups' labor and capital shares each add up to one.
• Households outside California never move California's rows, and the change including SDI equals the net change plus the SDI change.
• The within-state shift conserves every state's market income, and a state that frees labor income but holds no capital income makes the shift refuse.
• Differential: with every record in one state, the within-state shift sets the same inputs as labor_capital_shift._apply_shift and as the closed-form national shift.
• Differential: the summary's federal split plus the national state lines equals fiscal.net_fiscal_impact's total; the split leaves no residual, and its parts match policyengine-us's own aggregates.
• All-government change = federal + every state + state payroll taxes + local taxes, on both measures; the state payroll and local taxes stay off the federal line.
• Federal and state health shares add up to the costs they split; a dollar of health spending moved to a state leaves the federal line; state-only coverage never exceeds the federal share it comes from, and the unborn-child coverage keeps its federal match.
• TANF that a state's line already holds gets charged once, and the all-government total counts it once.
• On real policyengine-us, a get_branch branch keeps its state's Medicaid total fixed on its own, and at the baseline's denominator the total rises with enrollment while existing enrollees keep their cost.
• The health classification check refuses any model whose health lists this accounting doesn't cover, with or without the five subsidies listed.
• The chart's title follows the sign of California's change.

192 tests pass and 9 skip in a venv with matplotlib and Hypothesis and without PolicyEngine. With policyengine-us 2.2.1 installed, the nine tests that need it pass: test_california_shift.py runs 38 passed and 4 skipped (those four need matplotlib, which the first venv has), and all 14 tests in test_net_income_exclusions.py pass. Between the two venvs, every test runs.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
ai-inequality	Ready	Preview	Nov 9, 2026 9:25am UTC