Split MFJ pension/SS 50/50 for both-young couples

CLAUDE.md

@@ -24,6 +24,31 @@ Fragment names can be anything (branch name, issue number, short description). E
 - `changelog.d/fix-age-default.fixed.md`
 - `changelog.d/123.fixed.md`
 
+## Spousal income allocation (MFJ)
+
+TAXSIM gives several income types as a single household column with no
+per-spouse split. For married-filing-jointly records, the emulator allocates
+them between spouses by this rule (evidence-backed against the NBER
+`taxsimtest` binary; see taxsim issues #774, #924, #965, #966):
+
+- **Interest, dividends, capital gains, S-corp income → always split 50/50.**
+- **Pensions and Social Security → age-aware:**
+  - Both spouses on the **same side** of the elderly-eligibility line (both
+    qualify, or both do not) → **split 50/50**.
+  - **Mixed-age** (one qualifies, one does not) → assign the **whole amount to
+    the older spouse**, so age-based state exclusions (CO 55, DE 60, GA 62,
+    MD 65) reach the qualifying filer.
+
+Rationale: 50/50 matches TAXSIM and gives both spouses age-*independent*
+per-person exclusions (e.g. KY, OK), while the mixed-age exception protects
+age-*based* exclusions from being half-wasted on an ineligible younger spouse.
+The full age-aware rule lives in `runners/policyengine_runner.py` (the
+Microsimulation/CLI path), keyed on `_AGE_GATED_SPLIT_AGE` (55, the lowest
+state threshold). Note the single-household path in `core/input_mapper.py`
+(`income_types_to_split`) currently splits these types 50/50 *unconditionally*
+— correct for same-age couples but not yet age-aware for mixed-age pensions/SS;
+align it with the runner if that path is used for mixed-age elderly records.
+
 ## Running tests
 
 ```bash

changelog.d/fix-both-young-pension-split.fixed.md

@@ -0,0 +1 @@
+Fix both-young MFJ pension/Social Security split: the Microsimulation runner now splits these 50/50 when both spouses are on the same side of the elderly-eligibility line (previously it dumped them on the primary filer when both were under the threshold), restoring per-person state exclusions for age-independent states (e.g. KY, OK). Mixed-age allocation to the older spouse is unchanged.

policyengine_taxsim/runners/policyengine_runner.py

@@ -232,21 +232,24 @@ def accessor(row):
     @classmethod
     def _make_age_gated_primary(cls, source_field):
         """Allocate age-gated income (pension, gssi) to the primary filer's
-        share. Both spouses 60+: 50/50. Mixed-age: assign entirely to the
-        older spouse so age-based state exclusions reach the qualifying
-        filer (see taxsim issues #774 for pensions, #924 for gssi). Both
-        under 60: all to primary by default."""
+        share. The income is split 50/50 whenever both spouses fall on the
+        same side of the elderly-eligibility line (both qualify OR both do
+        not); only in mixed-age couples is it assigned entirely to the older
+        spouse, so age-based state exclusions reach the qualifying filer.
+        See taxsim #774 (pensions) and #924 (gssi) for the mixed-age ->
+        older rule, and #965 (KY) / #966 (OK) confirming both-young couples
+        must still split 50/50 (TAXSIM does, and per-person exclusions like
+        KY/OK are age-independent)."""
 
         def accessor(row):
             value = float(row.get(source_field, 0))
             if int(row.get("mstat", 1)) != 2:
                 return value
             page = int(row.get("page", 0))
             sage = int(row.get("sage", 0))
-            both_old = (
-                page >= cls._AGE_GATED_SPLIT_AGE and sage >= cls._AGE_GATED_SPLIT_AGE
-            )
-            if both_old:
+            primary_eligible = page >= cls._AGE_GATED_SPLIT_AGE
+            spouse_eligible = sage >= cls._AGE_GATED_SPLIT_AGE
+            if primary_eligible == spouse_eligible:
                 return value / 2
             primary_is_older_or_equal = page >= sage
             return value if primary_is_older_or_equal else 0.0
@@ -256,19 +259,19 @@ def accessor(row):
     @classmethod
     def _make_age_gated_spouse(cls, source_field):
         """Spouse's share of age-gated income. Mirror of `_make_age_gated_primary`:
-        50/50 if both 60+, full amount to spouse only when spouse is strictly
-        older than primary in mixed-age cases."""
+        50/50 when both spouses are on the same side of the elderly-eligibility
+        line; in mixed-age couples the full amount goes to the spouse only when
+        the spouse is strictly older than the primary."""
 
         def accessor(row):
             value = float(row.get(source_field, 0))
             if int(row.get("mstat", 1)) != 2:
                 return 0.0
             page = int(row.get("page", 0))
             sage = int(row.get("sage", 0))
-            both_old = (
-                page >= cls._AGE_GATED_SPLIT_AGE and sage >= cls._AGE_GATED_SPLIT_AGE
-            )
-            if both_old:
+            primary_eligible = page >= cls._AGE_GATED_SPLIT_AGE
+            spouse_eligible = sage >= cls._AGE_GATED_SPLIT_AGE
+            if primary_eligible == spouse_eligible:
                 return value / 2
             spouse_is_strictly_older = sage > page
             return value if spouse_is_strictly_older else 0.0

tests/test_spouse_income_splitting.py

@@ -139,12 +139,16 @@ def test_pension_goes_to_spouse_when_spouse_is_older():
     np.testing.assert_allclose(values, [0.0, 40000.0])
 
 
-def test_pension_stays_on_primary_when_both_under_60():
-    """Pension: when neither spouse is 60+, state elderly exclusions
-    do not apply. Default to primary."""
+def test_pension_splits_when_both_spouses_under_threshold():
+    """Pension: when both spouses are on the same (younger) side of the
+    eligibility line, split 50/50 — matching TAXSIM (which splits the
+    household pension column) and giving age-independent per-person
+    exclusions (KY, OK) to both spouses. See taxsim #965 (KY) / #966 (OK):
+    both filers age 45, where dumping pension on the primary denied the
+    spouse's exclusion. Verified against the NBER binary."""
     df = pd.DataFrame([_base_mfj_record(page=45, sage=45, pensions=30000)])
     values = _run_allocation(df, "taxable_private_pension_income")
-    np.testing.assert_allclose(values, [30000.0, 0.0])
+    np.testing.assert_allclose(values, [15000.0, 15000.0])
 
 
 def test_gssi_splits_when_both_spouses_are_60_plus():
@@ -172,11 +176,13 @@ def test_gssi_goes_to_spouse_when_spouse_is_older():
     np.testing.assert_allclose(values, [0.0, 40000.0])
 
 
-def test_gssi_stays_on_primary_when_both_under_60():
-    """gssi: when neither spouse is 60+, default to primary."""
+def test_gssi_splits_when_both_spouses_under_threshold():
+    """gssi: when both spouses are on the same (younger) side of the
+    eligibility line, split 50/50 — matching TAXSIM. Mirrors the pension
+    rule (see taxsim #924 for the mixed-age -> older convention)."""
     df = pd.DataFrame([_base_mfj_record(page=45, sage=45, gssi=40000)])
     values = _run_allocation(df, "social_security_retirement")
-    np.testing.assert_allclose(values, [40000.0, 0.0])
+    np.testing.assert_allclose(values, [20000.0, 20000.0])
 
 
 def test_de_elderly_pension_matches_issue_838():