Fall back to the earliest input year's MAGI for Medicare IRMAA, and add a same-year income switch

changelog.d/irmaa-lookback-fallback.added.md

@@ -0,0 +1 @@
+Added gov.simulation.medicare_irmaa_uses_current_year_income, which sets Medicare Part B and Part D IRMAA from the benefit year's own MAGI (medicare_irmaa_magi, medicare_irmaa_tax_year_magi) for marginal tax rate and behavioral response analysis.

changelog.d/irmaa-lookback-fallback.fixed.md

@@ -0,0 +1 @@
+Medicare IRMAA no longer drops to zero when the year two years before the benefit year precedes every input (single-year datasets in their first two years, and households entered for one year): the earliest input year's MAGI stands in, deflated to that year by per-capita CBO AGI. For a household entered for one year this imputes the lookback income from that year's income, so its IRMAA now moves with that income.

policyengine_us/parameters/gov/simulation/medicare_irmaa_uses_current_year_income.yaml

@@ -0,0 +1,18 @@
+description: PolicyEngine sets Medicare income-related monthly adjustment amounts from the benefit year's own modified adjusted gross income, rather than the income from two years earlier that the law uses, if this is true.
+# The law sets IRMAA on income from two years before the benefit year, so a
+# change in this year's earnings meets the surcharge two years later. Marginal
+# tax rate and behavioral response analyses that treat a lasting earnings
+# change as meeting the surcharge in the same year turn this on. Even when it
+# is off, a simulation whose earliest input year is the benefit year (a
+# household entered for one year) imputes the income two years prior from the
+# benefit year's income, deflated, so its charges already move with that
+# income; see medicare_irmaa_magi_two_years_prior.
+values:
+  0000-01-01: false
+metadata:
+  unit: bool
+  period: year
+  label: Set Medicare IRMAA from current-year income
+  reference:
+    - title: 42 U.S.C. 1395r(i)(4)(B)
+      href: https://www.law.cornell.edu/uscode/text/42/1395r#i_4_B

policyengine_us/tests/policy/baseline/gov/hhs/medicare/eligibility/medicare_irmaa_magi.yaml

@@ -0,0 +1,76 @@
+# The IRMAA brackets apply to MAGI from two years before the benefit year
+# unless gov.simulation.medicare_irmaa_uses_current_year_income is true, when
+# they apply to the benefit year's own MAGI.
+
+- name: By default the 2025 charges use 2023 income.
+  period: 2025
+  input:
+    age: 66
+    employment_income:
+      2023: 50_000
+      2025: 300_000
+  output:
+    medicare_irmaa_magi: 50_000
+    # Below 106,001: the $185 standard premium only.
+    gross_medicare_part_b_premium: 2_220
+    income_adjusted_part_d_premium_surcharge: 0
+
+- name: With the switch on the 2025 charges use 2025 income.
+  period: 2025
+  input:
+    gov.simulation.medicare_irmaa_uses_current_year_income: true
+    age: 66
+    employment_income:
+      2023: 50_000
+      2025: 300_000
+  output:
+    medicare_irmaa_magi: 300_000
+    # 2025 single, 200,001 up to 500,000: 12 x (185 + 406.90) = 7,102.80.
+    gross_medicare_part_b_premium: 7_102.8
+    # Part D: 12 x 78.60.
+    income_adjusted_part_d_premium_surcharge: 943.2
+
+- name: With the switch on a provided lag MAGI does not set the charges.
+  period: 2025
+  input:
+    gov.simulation.medicare_irmaa_uses_current_year_income: true
+    people:
+      person:
+        age: 66
+        employment_income: 50_000
+    tax_units:
+      tax_unit:
+        members: [person]
+        medicare_irmaa_magi_two_years_prior: 300_000
+    households:
+      household:
+        members: [person]
+  output:
+    medicare_irmaa_magi: 50_000
+    gross_medicare_part_b_premium: 2_220
+
+- name: A joint return's MAGI applies to both spouses with the switch on.
+  period: 2026
+  input:
+    gov.simulation.medicare_irmaa_uses_current_year_income: true
+    people:
+      head:
+        age: 70
+        employment_income: 150_000
+      spouse:
+        age: 68
+        employment_income: 150_000
+    tax_units:
+      tax_unit:
+        members: [head, spouse]
+        filing_status: JOINT
+    marital_units:
+      marital_unit:
+        members: [head, spouse]
+    households:
+      household:
+        members: [head, spouse]
+  output:
+    medicare_irmaa_magi: 300_000
+    # 2026 joint, 274,001 up to 342,001: $202.90 + $202.90 a month each.
+    gross_medicare_part_b_premium: [4_869.6, 4_869.6]

policyengine_us/tests/policy/baseline/gov/hhs/medicare/eligibility/medicare_irmaa_magi_two_years_prior.yaml

@@ -1,23 +1,36 @@
-# IRMAA uses MAGI from two years before the benefit year. PolicyEngine models
-# 2015 onward, so for 2015 and 2016 benefit years the income year (2013, 2014)
-# is outside the model: only income provided for that year counts, and the
-# MAGI is otherwise zero.
+# IRMAA uses MAGI from two years before the benefit year. When the simulation
+# holds no input for that year or any earlier one, the MAGI of the earliest
+# year with inputs (no earlier than 2015, the first modeled year) stands in,
+# deflated to the year two years prior by
+# calibration.gov.cbo.income_by_source.adjusted_gross_income_per_capita (CBO
+# AGI divided by population). That index starts in 2015 and is held at its
+# 2015 value before then:
+#   2015: 9,033,840,000,000 / 326,078,000 = 27,704.54
+#   2016: 9,033,840,000,000 / 328,964,000 = 27,461.49
+#   2022: 14,821,800,000,000 / 339,654,000 = 43,637.94
+#   2023: 15,347,400,000,000 / 342,835,000 = 44,766.14
+#   2024: 16,685,900,000,000 / 346,588,000 = 48,143.33
+# For 2015 and 2016 benefit years the income year (2013, 2014) precedes the
+# model: income provided for it counts, and only that.
 
-- name: Case 1, 2015 benefit year reads no 2013 income.
+- name: Case 1, 2015 benefit year with only 2015 inputs uses 2015 income.
   period: 2015
   input:
     age: 66
     employment_income: 500_000
   output:
-    medicare_irmaa_magi_two_years_prior: 0
+    # 500,000 x 27,704.54 (2013, held at 2015) / 27,704.54 (2015).
+    medicare_irmaa_magi_two_years_prior: 500_000
 
-- name: Case 2, 2016 benefit year reads no 2014 income.
+- name: Case 2, 2016 benefit year with only 2016 inputs uses 2016 income.
   period: 2016
+  absolute_error_margin: 0.01
   input:
     age: 66
     employment_income: 500_000
   output:
-    medicare_irmaa_magi_two_years_prior: 0
+    # 500,000 x 27,704.54 (2014, held at 2015) / 27,461.49 (2016).
+    medicare_irmaa_magi_two_years_prior: 504_425.34
 
 - name: Case 3, 2015 benefit year counts 2013 income that is provided.
   period: 2015
@@ -52,28 +65,153 @@
   output:
     medicare_irmaa_magi_two_years_prior: 155_000
 
-- name: Case 6, 2015 Part B premium is the base premium with no IRMAA.
+- name: Case 6, 2015 Part B premium has no IRMAA for 2015 income below the first threshold.
   period: 2015
   input:
     age: 66
-    employment_income: 500_000
+    employment_income: 100_000
     is_medicare_eligible: true
     # 2015 standard Part B premium of $104.90 per month, supplied directly so
     # the case tests only the IRMAA amount.
     base_part_b_premium: 1_258.8
   output:
+    medicare_irmaa_magi_two_years_prior: 100_000
     gross_medicare_part_b_premium: 1_258.8
 
-- name: Case 7, 2016 Part D IRMAA is zero for an enrolled high earner.
+- name: Case 7, 2016 Part D IRMAA is zero for an enrolled earner below the first threshold.
   period: 2016
+  absolute_error_margin: 0.01
   input:
     age: 66
-    employment_income: 500_000
+    employment_income: 100_000
     is_medicare_eligible: true
     medicare_enrolled: true
   output:
+    # 100,000 x 27,704.54 / 27,461.49, under the first threshold of 103,001.
+    medicare_irmaa_magi_two_years_prior: 100_885.07
     income_adjusted_part_d_premium_surcharge: 0
 
+- name: A 2025 household with only 2024 inputs has IRMAA set on deflated 2024 income.
+  period: 2025
+  absolute_error_margin: 0.01
+  input:
+    people:
+      person:
+        age:
+          2024: 66
+        employment_income:
+          2024: 200_000
+    tax_units:
+      tax_unit:
+        members: [person]
+    households:
+      household:
+        members: [person]
+        state_code:
+          2024: TX
+  output:
+    # 2024 MAGI of 200,000 x 44,766.14 (2023) / 48,143.33 (2024).
+    medicare_irmaa_magi_two_years_prior: 185_970.28
+    medicare_irmaa_magi: 185_970.28
+    # 2025 single brackets: 167,001 up to 200,001 adds $295.90 a month to the
+    # $185 standard premium: 12 x (185 + 295.90) = 5,770.80.
+    gross_medicare_part_b_premium: 5_770.8
+    # Part D adds $57.00 a month in the same bracket.
+    income_adjusted_part_d_premium_surcharge: 684
+
+- name: The same household in 2026 reads its 2024 income without deflation.
+  period: 2026
+  absolute_error_margin: 0.01
+  input:
+    people:
+      person:
+        age:
+          2024: 66
+        employment_income:
+          2024: 200_000
+    tax_units:
+      tax_unit:
+        members: [person]
+    households:
+      household:
+        members: [person]
+        state_code:
+          2024: TX
+  output:
+    medicare_irmaa_magi_two_years_prior: 200_000
+    # 2026 single brackets: 171,001 up to 205,001 adds $324.60 a month to the
+    # $202.90 standard premium: 12 x (202.90 + 324.60) = 6,330.
+    gross_medicare_part_b_premium: 6_330
+    income_adjusted_part_d_premium_surcharge: 724.8
+
+- name: The data year itself as the benefit year looks back to deflated income too.
+  period: 2024
+  # MAGI is stored as float32, so allow for its rounding.
+  absolute_error_margin: 0.1
+  input:
+    people:
+      person:
+        age:
+          2024: 66
+        employment_income:
+          2024: 200_000
+    tax_units:
+      tax_unit:
+        members: [person]
+    households:
+      household:
+        members: [person]
+        state_code:
+          2024: TX
+  output:
+    # 200,000 x 43,637.94 (2022) / 48,143.33 (2024).
+    medicare_irmaa_magi_two_years_prior: 181_283.42
+    # 2024 single brackets: 161,001 up to 193,001 adds $279.50 a month to the
+    # $174.70 standard premium: 12 x (174.70 + 279.50) = 5,450.40.
+    gross_medicare_part_b_premium: 5_450.4
+
+- name: Income provided for the year two years prior is used as given.
+  period: 2025
+  input:
+    people:
+      person:
+        age:
+          2024: 66
+        employment_income:
+          2024: 200_000
+    tax_units:
+      tax_unit:
+        members: [person]
+        adjusted_gross_income:
+          2023: 50_000
+    households:
+      household:
+        members: [person]
+  output:
+    medicare_irmaa_magi_two_years_prior: 50_000
+    gross_medicare_part_b_premium: 2_220
+
+- name: A provided IRMAA MAGI overrides the fallback.
+  period: 2025
+  input:
+    people:
+      person:
+        age:
+          2024: 66
+        employment_income:
+          2024: 200_000
+    tax_units:
+      tax_unit:
+        members: [person]
+        medicare_irmaa_magi_two_years_prior:
+          2025: 90_000
+    households:
+      household:
+        members: [person]
+  output:
+    medicare_irmaa_magi: 90_000
+    gross_medicare_part_b_premium: 2_220
+
 - name: Amounts excluded under IRC sections 135, 911, 931 and 933 are added back.
   period: 2026
   input:

policyengine_us/tests/policy/baseline/gov/hhs/medicare/senior_household_2015_2016_integration.yaml

@@ -1,6 +1,8 @@
 # Households with a member aged 65 or older in 2015 and 2016. Medicare IRMAA
 # looks two years back, to 2013 and 2014, which precede the first modeled year;
-# these calculations raised ParameterNotFoundError before (issue #9613).
+# these calculations raised ParameterNotFoundError before (issue #9613). With
+# inputs only for the benefit year, the IRMAA MAGI is that year's MAGI deflated
+# to two years earlier, far below the first IRMAA threshold here.
 #
 # Single 66-year-old in Texas with $10,000 of wages:
 # - Payroll tax: 7.65% x $10,000 = $765.
@@ -30,7 +32,8 @@
         members: [person]
         state_code: TX
   output:
-    medicare_irmaa_magi_two_years_prior: 0
+    # 10,000 x 27,704.54 (2013, held at 2015) / 27,704.54 (2015).
+    medicare_irmaa_magi_two_years_prior: 10_000
     income_tax: 0
     ssi: 4_306
     snap: 192
@@ -49,7 +52,8 @@
         members: [person]
         state_code: TX
   output:
-    medicare_irmaa_magi_two_years_prior: 0
+    # 10,000 x 27,704.54 (2014, held at 2015) / 27,461.49 (2016).
+    medicare_irmaa_magi_two_years_prior: 10_088.51
     income_tax: 0
     ssi: 4_306
     snap: 192

policyengine_us/tests/policy/baseline/gov/hhs/medicare/test_irmaa_first_modeled_year.py

@@ -2,7 +2,9 @@
 
 Issue #9613: a 2015 or 2016 household with a member aged 65 or older raised
 ParameterNotFoundError because the IRMAA MAGI read 2013 or 2014 income, before
-the first year the parameters are backdated to.
+the first year the parameters are backdated to. With inputs only for the
+benefit year, the IRMAA MAGI is that year's MAGI deflated to two years
+earlier; with inputs for 2013 or 2014, only the provided income counts.
 """
 
 import pytest
@@ -62,7 +64,7 @@ def test_first_modeled_year_is_the_backdating_floor():
 @pytest.mark.parametrize("year", UNMODELED_LAG_YEARS)
 @pytest.mark.parametrize("filing_status", FILING_STATUSES)
 @pytest.mark.parametrize("employment_income", INCOMES)
-def test_magi_is_zero_when_lag_precedes_the_model(
+def test_benefit_year_income_stands_in_when_lag_precedes_the_model(
     year, filing_status, employment_income
 ):
     sim = senior_household(
@@ -74,13 +76,24 @@ def test_magi_is_zero_when_lag_precedes_the_model(
         },
         tax_unit_inputs={"filing_status": {year: filing_status}},
     )
-    assert sim.calculate("medicare_irmaa_magi_two_years_prior", year)[0] == 0
-    # With a zero MAGI, no IRMAA is added to the base Part B premium and no
-    # Part D surcharge applies.
+    index = SYSTEM.parameters.calibration.gov.cbo.income_by_source.adjusted_gross_income_per_capita
+    # The index starts in 2015 and is held at that value for 2013 and 2014.
+    deflator = index(f"{FIRST_MODELED_YEAR}-01-01") / index(f"{year}-01-01")
+    benefit_year_magi = sim.calculate("medicare_irmaa_tax_year_magi", year)[0]
+    magi = sim.calculate("medicare_irmaa_magi_two_years_prior", year)[0]
+    assert magi == pytest.approx(benefit_year_magi * deflator, rel=1e-6)
+    # The brackets apply to that MAGI: the Part B premium adds the IRMAA
+    # amount to the base premium, and Part D charges its surcharge.
+    p = SYSTEM.parameters(f"{year}-01-01").gov.hhs.medicare
+    scale = filing_status.lower()
     gross = sim.calculate("gross_medicare_part_b_premium", year)[0]
     base = sim.calculate("base_part_b_premium", year)[0]
-    assert gross == pytest.approx(base)
-    assert sim.calculate("income_adjusted_part_d_premium_surcharge", year)[0] == 0
+    assert gross == pytest.approx(
+        base + 12 * getattr(p.part_b.irmaa, scale).calc(magi), abs=0.01
+    )
+    assert sim.calculate("income_adjusted_part_d_premium_surcharge", year)[
+        0
+    ] == pytest.approx(12 * getattr(p.part_d.irmaa, scale).calc(magi), abs=0.01)
 
 
 @pytest.mark.parametrize("year", MODELED_LAG_YEARS)