Subtract the expected Part D premium from medicare_cost

[MaxGhenis]

Draft: waiting on the microsim and independent review. Stacked on #10107. That PR sets calibration.gov.hhs.medicare.per_capita_cost to CMS NHE Medicare spending per enrollee, which covers Parts A through D, gross of the premiums that finance it. medicare_cost, the net value of Medicare (a program statistic in policyengine.py's US_PROGRAMS), subtracted only the Part A and gross Part B premiums. So for every Medicare enrollee it overstated Medicare's net value by the Part D premiums that pay for part of that spending.

Change

medicare_cost also subtracts an expected Part D premium:

part_d.enrollment_share × (part_d.premiums_per_enrollee + income_adjusted_part_d_premium_surcharge)

No variable records Part D enrollment, so the premium is weighted by the Part D enrollment share. Both series are new, under calibration.gov.hhs.medicare.part_d, and come from the 2026 Medicare Trustees Report (cites are file pages):

Series	Definition	Years
premiums_per_enrollee	(Table III.D3 premium income − Table V.E4 aggregate Part D IRMAA) ÷ Table V.B3 Part D enrollment, annual $	2015–2035; 2036–2100 grow at the 2034–2035 rate (3.3%)
enrollment_share	Table V.B3 Part D ÷ total ("HI and/or SMI coverage") enrollment	2015–2035 annual, then every fifth year to 2100

Year	Premiums per Part D enrollee	Share	Expected premium per Medicare enrollee, before IRMAA
2015	$277.60	0.7517	$208.67
2024	$307.70	0.8117	$249.76
2025	$220.25	0.8191	$180.41
2026	$289.79	0.8195	$237.48
2030	$637.06	0.8382	$533.98
2035	$722.05	0.8485	$612.66

2025 dips because of the voluntary Part D premium demonstration (page 109). 2030 jumps with the Inflation Reduction Act's reset of the base beneficiary premium percentage (page 152).

medicare_gross_cost is unchanged.

Why this premium, and not the base beneficiary premium

The brief suggested base beneficiary premium × enrollment share + IRMAA. The Trustees Report shows that would overstate what enrollees pay:

• "The average paid premium has always been lower than the base beneficiary premium." In 2025 the average paid premium was $23.95 a month (pages 216–217), against a $36.78 base beneficiary premium.
• Low-income subsidy enrollees "pay substantially reduced premium and cost-sharing amounts" (page 219). There were 14.2 million of them among 56.8 million Part D enrollees in 2025 (Table IV.B7, page 153). Medicare's low-income subsidy payments are Part D account expenditures, not enrollee premiums (page 114).
• So base premium × Part D enrollment comes to $25.0 billion in 2025, against $12.5 billion of premiums net of IRMAA ($14.9 billion including IRMAA, Table III.D1, page 109).

What the series measures, from the sources:

1. NHE spending is gross of Part D premiums. NHE Medicare spending covers Parts A–D and the plans' non-medical costs (NHEA methodology, pages 15 and 23). The SMI trust fund "is financed through general revenues, premiums (Part B, Part D, and Medicare premium buy-in programs by Medicaid), state phase-down payments, and interest income" (pages 29–30).
2. Premium income counts every channel. Table III.D3, footnote 1: "Premiums include both amounts withheld from Social Security benefits or other Federal payments, direct payments to the Federal government, and those paid directly to Part D plans." Premiums paid directly to plans finance part of Part D benefits and are included in both income and expenditures (page 110).
3. IRMAA is inside premium income, so it is taken out and charged per person instead. Part D IRMAA "shall be paid through withholding from benefit payments" (42 U.S.C. 1395w-113(c)(4)(A)) and "is deposited into the Part D account" (footnote 73, page 151). Table V.E4 gives its aggregate. medicare_cost subtracts the person's own income_adjusted_part_d_premium_surcharge, times the share, as it already subtracts the person's own Part B IRMAA.
4. Timing. Table III.D3 is on a cash basis. Footnote 6 (page 117) credits one January's withheld premiums on December 31 of the prior year: $0.2 billion in 2015, $0.1 billion in 2020 and an estimated $0.2 billion in 2026. Those amounts are moved to 2016, 2021 and 2027.
5. Identity. share × premiums_per_enrollee = (premium income − IRMAA) ÷ total enrollment. So the uniform part reproduces the Trustees' aggregate per Medicare enrollee. 2025: 0.8191 × $220.25 = $180.41, and $180.41 + $2.4B ÷ 69.289M = $215.04 = $14.9B ÷ 69.289M.
6. Extension. The premium grows at its own 2034–2035 rate (3.3%), not at per_capita_cost's 5.4%. The Trustees project Part D cost per beneficiary growing 3.1% in 2035 against 5.4% for all of Medicare (Table V.D1, page 210), and Part D near 0.76% of GDP through 2045 (Table III.D6).

Alternatives rejected:

• One uniform premium that includes IRMAA. It drops the person-level IRMAA that medicare_cost already applies for Part B.
• Full IRMAA without the share. It assumes every high-income enrollee is in Part D.
• Modeling Part D or LIS enrollment. That needs a new program model and is out of scope. LIS enrollees pay little or no premium, so the uniform average understates their Medicare value and overstates everyone else's slightly.

medicare_gross_cost stays gross

medicare_gross_cost feeds cbo_household_social_insurance_benefits and is a gross concept by name and docstring, so it stays unchanged. CBO's standard household-income measure, however, counts Medicare "net of premiums paid by enrollees", for Parts A, B and D. "Gross Medicare spending" appears only in CBO's expanded sensitivity measure (CBO, January 2026). Aligning pe-us's default CBO household income with that is a separate change, proposed as its own follow-up session.

Household-level changes (YAML tests)

Case	Before	After
2025, premium-free Part A, standard Part B	$15,531	$15,350.59
2025, single, $500k MAGI (top Part D IRMAA, $1,029.60 a year)	$10,204.20	$9,180.45
2021, premium-free Part A	$12,511	$12,271.87

New cases cover a 2025 middle IRMAA tier, 2027, the 2030 reset with IRMAA, and the zero floor. Projected-year cases take the Part B premium and Part D IRMAA as inputs, so they hold when #10103's premium projections land.

Microsim impact

Pending: heavy-lock pair (base #10107 head 33a94b4, PR 712abd3), 2026 and 2034, queued 14:57 EDT.

Invariants (tested)

policyengine_us/tests/core/test_medicare_cost_part_d_premium.py:

1. Transcription. Every value equals its formula applied to the Trustees' published cells, rounded as stored: 21 premium values and 34 share values.
2. Conservation. Premiums per Part D enrollee × Part D enrollment + aggregate IRMAA returns Table III.D3's premium income in every year from 2015 to 2035.
3. Bounds (Hypothesis).
   • The share is strictly between 0 and 1, and the premium is positive, in every year from 2015 to 2200.
   • The monthly premium is below the Trustees' average paid premium in 2025 and 2026.
4. Extension (Hypothesis).
   • 2036–2100 equals the closed form 722.05 × (722.05 / 699.07)^(year − 2035), and later years hold the 2100 value.
   • The rate is within 0.5 points of the Trustees' 2035 Part D cost growth.
5. Calendar years (Hypothesis).
   • Every date in a year reads the January value.
   • Between the five-year points, the share holds the earlier one.
6. Consumer (Hypothesis, differential). Over years 2015–2100 and arbitrary Part A, Part B and IRMAA inputs:
   • medicare_cost equals a reference implementation.
   • 0 ≤ medicare_cost ≤ medicare_gross_cost.
   • medicare_gross_cost does not depend on the Part D premium.
   • medicare_cost does not rise with the person's IRMAA.

Mutation checks each failed at least one test:

• dropping the IRMAA term;
• dropping the share;
• dropping the zero floor;
• a one-digit typo in the 2050 share;
• removing the extension.

#10107's consumer invariant (test_medicare_per_capita_cost.py) now includes the Part D term.

Sources

Source	URL	Retrieved	sha256
2026 Medicare Trustees Report	https://www.cms.gov/oact/tr/2026	2026-10-11	ffa56b9137006872300b0346149eae1613d09a172b6ba118aad48e66dfc48fa8
NHEA methodology paper, 2024	https://www.cms.gov/files/document/definitions-sources-and-methods.pdf	2026-10-11	9c5755666847a78103016f04da937326744071a9a6736f4e77e91ce9e7af8629
42 U.S.C. 1395w-113	https://www.law.cornell.edu/uscode/text/42/1395w-113	2026-10-11	page HTML, in evidence folder
CBO, Sensitivity of the Distribution of Household Income to the Treatment of Health Insurance From 1979 to 2021	https://www.cbo.gov/publication/62544	2026-10-11	read in browser

Interactions

• Project Medicare premiums and IRMAA past the last announced year #10103 (draft, stacked on Uprate premium inputs by premium series, not CMS out-of-pocket spending #10085) projects Part D IRMAA amounts past 2026 and adds the base beneficiary premium series. This PR does not use that series. Its new parameters are in a different folder, and its uprating_extensions.py hunk sits next to Use CMS Medicare spending per enrollee for the Medicare per capita cost #10107's.
• No other pe-us variable reads medicare_cost or the new parameters.

Tests run

Locally: test_medicare_cost_part_d_premium.py (85 passed), test_medicare_per_capita_cost.py + test_part_b_msp_offset.py + test_uprating_extensions.py (113 passed), Medicare YAML directory (212 passed). CI runs once the base is main (after #10107 merges).

axiom: n/a: valuation of Medicare benefits from Trustees aggregates, not a statutory provision

🤖 Generated with Claude Code

[MaxGhenis]

Heads-up from #10115, which charges income_adjusted_part_d_premium_surcharge only to Part D enrollees. It adds medicare_part_d_enrolled, defined for medicare_enrolled and driven by a new takes_up_medicare_part_d_if_eligible input that defaults to true.

This PR subtracts part_d.enrollment_share × (premiums_per_enrollee + income_adjusted_part_d_premium_surcharge). That expected-value form is right while the surcharge reaches every Medicare enrollee. Once microdata set Part D take-up, the surcharge is already zero for non-enrollees, and multiplying it by enrollment_share would count non-enrollment twice.

Options, after both merge:

• If medicare_part_d_enrolled is known per person, subtract medicare_part_d_enrolled × premiums_per_enrollee + income_adjusted_part_d_premium_surcharge.
• Otherwise, keep the expected value on the base premium only, and subtract the surcharge as it stands.

With the default input the two PRs give the same numbers today: git merge-tree is clean, and base and #10115 match exactly on the certified file. So this is not blocking. It is a note for whichever PR merges second.