Add capital income predictors to asset imputation QRF (#546)

* Add capital income predictors to SIPP asset imputation

The QRF model for imputing liquid assets (bank accounts, stocks, bonds)
previously used only employment_income, age, demographics. This adds
interest_income, dividend_income, and rental_income as predictors,
which are strongly correlated with asset holdings and available in
both SIPP (TINC_BANK, TINC_STMF, TINC_BOND, TINC_RENT) and CPS.

Updated in three places to keep them consistent:
- sipp.py (standalone model training)
- cps.py (CPS variable extraction)
- source_impute.py (calibration-time imputation)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Add changelog entry

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Retrigger CI

* Retrigger CI (2)

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: baogorek@gmail.com <baogorek@gmail.com>

Author: 3 people

policyengine_us_data/calibration/source_impute.py

@@ -84,6 +84,9 @@
 
 SIPP_ASSETS_PREDICTORS = [
     "employment_income",
+    "interest_income",
+    "dividend_income",
+    "rental_income",
     "age",
     "is_female",
     "is_married",
@@ -469,6 +472,10 @@ def _impute_sipp(
             "TVAL_BANK",
             "TVAL_STMF",
             "TVAL_BOND",
+            "TINC_BANK",
+            "TINC_STMF",
+            "TINC_BOND",
+            "TINC_RENT",
         ]
         asset_df = pd.read_csv(
             STORAGE_FOLDER / "pu2023.csv",
@@ -484,6 +491,11 @@ def _impute_sipp(
         asset_df["is_female"] = asset_df.ESEX == 2
         asset_df["is_married"] = asset_df.EMS == 1
         asset_df["employment_income"] = asset_df.TPTOTINC * 12
+        asset_df["interest_income"] = (
+            asset_df["TINC_BANK"].fillna(0) + asset_df["TINC_BOND"].fillna(0)
+        ) * 12
+        asset_df["dividend_income"] = asset_df["TINC_STMF"].fillna(0) * 12
+        asset_df["rental_income"] = asset_df["TINC_RENT"].fillna(0) * 12
         asset_df["household_weight"] = asset_df.WPFINWGT
         asset_df["is_under_18"] = asset_df.TAGE < 18
         asset_df["count_under_18"] = (
@@ -495,6 +507,9 @@ def _impute_sipp(
 
         asset_train_cols = [
             "employment_income",
+            "interest_income",
+            "dividend_income",
+            "rental_income",
             "bank_account_assets",
             "stock_assets",
             "bond_assets",
@@ -518,7 +533,14 @@ def _impute_sipp(
             data,
             time_period,
             dataset_path,
-            ["employment_income", "age", "is_male"],
+            [
+                "employment_income",
+                "interest_income",
+                "dividend_income",
+                "rental_income",
+                "age",
+                "is_male",
+            ],
         )
         if "is_male" in cps_asset_df.columns:
             cps_asset_df["is_female"] = (~cps_asset_df["is_male"].astype(bool)).astype(
@@ -537,6 +559,18 @@ def _impute_sipp(
             if "count_under_18" in cps_tip_df.columns
             else 0.0
         )
+        for cap_var in [
+            "interest_income",
+            "dividend_income",
+            "rental_income",
+        ]:
+            if cap_var not in cps_asset_df.columns:
+                if cap_var in data:
+                    cps_asset_df[cap_var] = data[cap_var][time_period].astype(
+                        np.float32
+                    )
+                else:
+                    cps_asset_df[cap_var] = 0.0
 
         asset_vars = [
             "bank_account_assets",

policyengine_us_data/datasets/cps/cps.py

@@ -1750,6 +1750,9 @@ def add_tips(self, cps: h5py.File):
             "person_id",
             "household_id",
             "employment_income",
+            "interest_income",
+            "dividend_income",
+            "rental_income",
             "age",
             "household_weight",
             "is_female",

policyengine_us_data/datasets/sipp/sipp.py

@@ -152,6 +152,11 @@ def get_tip_model() -> QRF:
     "TVAL_BANK",  # Checking, savings, money market
     "TVAL_STMF",  # Stocks and mutual funds
     "TVAL_BOND",  # Bonds and government securities
+    # Income from assets (monthly, person-level)
+    "TINC_BANK",  # Interest from bank accounts
+    "TINC_STMF",  # Dividends from stocks/mutual funds
+    "TINC_BOND",  # Interest from bonds
+    "TINC_RENT",  # Rental income
     # SSI receipt (for validation)
     "RSSI_YRYN",  # Received SSI in at least one month
 ]
@@ -196,6 +201,12 @@ def train_asset_model():
     df["household_weight"] = df.WPFINWGT
     df["household_id"] = df.SSUID
 
+    # Capital income predictors (annualized from monthly SIPP)
+    # Maps to CPS: interest_income, dividend_income, rental_income
+    df["interest_income"] = (df["TINC_BANK"].fillna(0) + df["TINC_BOND"].fillna(0)) * 12
+    df["dividend_income"] = df["TINC_STMF"].fillna(0) * 12
+    df["rental_income"] = df["TINC_RENT"].fillna(0) * 12
+
     # Calculate household-level counts
     df["is_under_18"] = df.TAGE < 18
     df["count_under_18"] = (
@@ -206,6 +217,9 @@ def train_asset_model():
         [
             "household_id",
             "employment_income",
+            "interest_income",
+            "dividend_income",
+            "rental_income",
             "bank_account_assets",
             "stock_assets",
             "bond_assets",
@@ -235,6 +249,9 @@ def train_asset_model():
         X_train=sipp,
         predictors=[
             "employment_income",
+            "interest_income",
+            "dividend_income",
+            "rental_income",
             "age",
             "is_female",
             "is_married",