Extrapolate post-2022 weights using a CBO returns (not a CBO population) projection

Makefile

@@ -16,7 +16,7 @@ clean:
 	rm -f tmd/storage/output/preimpute_tmd.csv.gz
 	rm -f tmd/storage/output/make_data_*.log
 
-# Each of the five build stages below tees its stdout+stderr to a
+# Each of the six build stages below tees its stdout+stderr to a
 # per-stage log file in tmd/storage/output/.  On-screen output during
 # `make data` is unchanged; the log files are a byproduct so that any
 # warnings emitted during the run (pandas, numpy, taxcalc, scipy,
@@ -31,20 +31,26 @@ clean:
 # appear silent on the terminal until the buffer flushes.  The `-u`
 # flag preserves the pre-tee behavior of seeing progress in real time.
 
-tmd/storage/output/tmd.csv.gz:
-	python -u tmd/create_taxcalc_input_variables.py 2>&1 \
-	    | tee tmd/storage/output/make_data_tmd.log
+# The two growth-factors files are built first because subsequent
+# stages use growfactors.csv as the Policy-parameter indexing file and
+# tmd_growfactors.csv as the Records-variable extrapolation file.
 
-tmd/storage/output/tmd_weights.csv.gz:
-	python -u tmd/create_taxcalc_sampling_weights.py 2>&1 \
-	    | tee tmd/storage/output/make_data_weights.log
-
-tmd/storage/output/tmd_growfactors.csv:
+tmd/storage/output/growfactors.csv:
 	python -u tmd/create_raw_growth_factors.py 2>&1 \
 	    | tee tmd/storage/output/make_data_raw_growfactors.log
+
+tmd/storage/output/tmd_growfactors.csv: tmd/storage/output/growfactors.csv
 	python -u tmd/create_taxcalc_growth_factors.py 2>&1 \
 	    | tee tmd/storage/output/make_data_tmd_growfactors.log
 
+tmd/storage/output/tmd.csv.gz: tmd/storage/output/tmd_growfactors.csv
+	python -u tmd/create_taxcalc_input_variables.py 2>&1 \
+	    | tee tmd/storage/output/make_data_tmd.log
+
+tmd/storage/output/tmd_weights.csv.gz: tmd/storage/output/tmd.csv.gz
+	python -u tmd/create_taxcalc_sampling_weights.py 2>&1 \
+	    | tee tmd/storage/output/make_data_weights.log
+
 tmd/storage/output/cached_files:
 	python -u tmd/create_taxcalc_cached_files.py 2>&1 \
 	    | tee tmd/storage/output/make_data_cached.log
@@ -54,9 +60,10 @@ tmd/storage/output/preimpute_tmd.csv.gz:
 	    | tee tmd/storage/output/make_data_preimpute.log
 
 .PHONY=tmd_files
-tmd_files: tmd/storage/output/tmd.csv.gz \
-  tmd/storage/output/tmd_weights.csv.gz \
+tmd_files: tmd/storage/output/growfactors.csv \
   tmd/storage/output/tmd_growfactors.csv \
+  tmd/storage/output/tmd.csv.gz \
+  tmd/storage/output/tmd_weights.csv.gz \
   tmd/storage/output/cached_files \
   tmd/storage/output/preimpute_tmd.csv.gz
 

README.md

@@ -8,32 +8,16 @@ For Tax-Calculator results generated when using these TMD input files,
 see [this
 folder](https://github.com/PSLmodels/Tax-Calculator/tree/master/taxcalc/cli/input_data_tests).
 
-The **current TMD version is 2.1.4**, which was released on July 13,
-2026, and is the same as TMD version 2.1.3 except that a new version
-of `scikit-learn` (1.9.0), which fixed a bug in its weighted imputation
-algorithm, is used (see #534 for details).  Tax-Calculator version 6.8.1
-or higher is required to generate the TMD files.
-
-When using version 2.1.4 to generate **sub-national weights**, there
-will be (presumably small) differences from the sub-national weights
-generated using earlier versions.  Note that the 118th Congressional
-district weight fingerprints have not yet been updated.
-
-The earlier TMD 2.0.0 version introduced the following significant
-improvements:
-- generate national, state, and Congressional district, input files
-for **2022**:
-[#470](https://github.com/PSLmodels/tax-microdata-benchmarking/pull/470)
-[#471](https://github.com/PSLmodels/tax-microdata-benchmarking/pull/471)
-[#472](https://github.com/PSLmodels/tax-microdata-benchmarking/pull/472)
-[#473](https://github.com/PSLmodels/tax-microdata-benchmarking/pull/473)
-[#474](https://github.com/PSLmodels/tax-microdata-benchmarking/pull/474)
-- improve the selection of CPS tax units to represent nonfilers:
-[#438](https://github.com/PSLmodels/tax-microdata-benchmarking/pull/438)
-- vastly improve the reweighting algorithm:
-[#416](https://github.com/PSLmodels/tax-microdata-benchmarking/pull/416)
-- impute values for three variables used in new OBBBA deductions:
-[#397](https://github.com/PSLmodels/tax-microdata-benchmarking/pull/397)
+The **current TMD version is 2.2.0**, which was released on September
+9, 2026, and is the same as TMD version 2.1.4 except that post-2022
+weights are grown using a CBO returns projection (rather than using a
+CBO population projection as in prior versions).  See PR #546 for details.
+
+When using version 2.2.0 to generate national or sub-national weights,
+the 2022 weights (and hence, the fingerprints) are the same as for
+version 2.1.4.  However, all the post-2022 weights are larger.  Note
+that the 118th Congressional district weight fingerprints have not
+been updated.
 
 ## Usage instructions
 

tests/conftest.py

@@ -7,7 +7,7 @@
 import numpy as np
 import pandas as pd
 import taxcalc
-from tmd.storage import STORAGE_FOLDER
+from tmd.storage import TMD_VARIABLES_PATH
 from tmd.imputation_assumptions import TAXYEAR
 
 # convert all numpy floating-point execeptions into errors
@@ -50,24 +50,4 @@ def tests_folder():
 
 @pytest.fixture(scope="session")
 def tmd_variables():
-    return pd.read_csv(STORAGE_FOLDER / "output" / "tmd.csv.gz")
-
-
-@pytest.fixture(scope="session")
-def tmd_variables_path():
-    return STORAGE_FOLDER / "output" / "tmd.csv.gz"
-
-
-@pytest.fixture(scope="session")
-def tmd_weights_path():
-    return STORAGE_FOLDER / "output" / "tmd_weights.csv.gz"
-
-
-@pytest.fixture(scope="session")
-def tmd_growfactors_path():
-    return STORAGE_FOLDER / "output" / "tmd_growfactors.csv"
-
-
-@pytest.fixture(scope="session")
-def policy_growfactors_path():
-    return STORAGE_FOLDER / "output" / "growfactors.csv"
+    return pd.read_csv(TMD_VARIABLES_PATH)

tests/expect_tmd_revenues

@@ -3,18 +3,18 @@ TMD vs CBO aggregate individual income tax revenue
 
 YEAR     ACTUAL     EXPECT  PCTDIFF
 2022     2149.4     2149.6    -0.01
-2023     2218.1     2171.9    +2.13
-2024     2407.9     2408.4    -0.02
-2025     2491.9     2493.7    -0.07
-2026     2637.4     2665.0    -1.04
-2027     2697.5     2776.8    -2.85
-2028     2757.4     2860.0    -3.59
-2029     2898.3     3039.1    -4.63
-2030     3049.1     3196.7    -4.62
-2031     3171.2     3310.2    -4.20
-2032     3302.1     3452.3    -4.35
-2033     3438.6     3599.4    -4.47
-2034     3581.8     3760.4    -4.75
-2035     3730.6     3929.3    -5.06
+2023     2241.8     2171.9    +3.22
+2024     2441.3     2408.4    +1.36
+2025     2520.3     2493.7    +1.07
+2026     2685.1     2665.0    +0.75
+2027     2760.4     2776.8    -0.59
+2028     2835.8     2860.0    -0.85
+2029     2991.6     3039.1    -1.56
+2030     3158.0     3196.7    -1.21
+2031     3302.8     3310.2    -0.22
+2032     3451.1     3452.3    -0.03
+2033     3604.9     3599.4    +0.15
+2034     3765.1     3760.4    +0.12
+2035     3932.5     3929.3    +0.08
 
-sum over 2022-2035 of squared PCTDIFF: iitax = 174.05
+sum over 2022-2035 of squared PCTDIFF: iitax = 19.00

tests/expected_tax_exp_2022_data

@@ -1,9 +1,9 @@
-YR,KIND,EST= 2023 paytax 1431.0
-YR,KIND,EST= 2023 iitax 2332.3
-YR,KIND,EST= 2023 ctc 129.6
-YR,KIND,EST= 2023 eitc 66.9
-YR,KIND,EST= 2023 social_security_partial_taxability 47.3
-YR,KIND,EST= 2023 niit -47.4
-YR,KIND,EST= 2023 cgqd_tax_preference 179.6
-YR,KIND,EST= 2023 qbid 55.0
-YR,KIND,EST= 2023 salt 23.8
+YR,KIND,EST= 2023 paytax 1446.3
+YR,KIND,EST= 2023 iitax 2357.2
+YR,KIND,EST= 2023 ctc 130.9
+YR,KIND,EST= 2023 eitc 67.7
+YR,KIND,EST= 2023 social_security_partial_taxability 47.8
+YR,KIND,EST= 2023 niit -47.8
+YR,KIND,EST= 2023 cgqd_tax_preference 181.6
+YR,KIND,EST= 2023 qbid 55.6
+YR,KIND,EST= 2023 salt 24.0

tests/test_imputed_variables.py

@@ -7,17 +7,18 @@
 import numpy as np
 import pytest
 import taxcalc
+from tmd.storage import (
+    POLICY_GROWFACTORS_PATH,
+    TMD_GROWFACTORS_PATH,
+    TMD_VARIABLES_PATH,
+    TMD_WEIGHTS_PATH,
+)
 from tmd.imputation_assumptions import TAXYEAR, SOI_IITAX_SPEC
 from tests.conftest import create_tmd_records
 
 
 @pytest.mark.obbba_deduction
-def test_obbba_deduction_tax_benefits(
-    tmd_variables_path,
-    tmd_weights_path,
-    tmd_growfactors_path,
-    policy_growfactors_path,
-):
+def test_obbba_deduction_tax_benefits():
     """
     Estimate percent of tax-units affected by each new tax deduction and
     average size of the 2026 dollar effect among those affected.
@@ -56,7 +57,7 @@ def actual_results(rdf, bdf):
             # https://taxpolicycenter.org/taxvox/
             #         budget-laws-tax-cuts-overtime-and-
             #         tips-are-popular-few-will-benefit
-            "exp_totben_2022": 24.42,
+            "exp_totben_2022": 24.86,
             "exp_affpct_2022": 8.78,
             "exp_affben_2022": 1423,
         },
@@ -73,7 +74,7 @@ def actual_results(rdf, bdf):
             # https://taxpolicycenter.org/taxvox/
             #         budget-laws-tax-cuts-overtime-and-
             #         tips-are-popular-few-will-benefit
-            "exp_totben_2022": 7.19,
+            "exp_totben_2022": 7.32,
             "exp_affpct_2022": 2.64,
             "exp_affben_2022": 1397,
         },
@@ -88,7 +89,7 @@ def actual_results(rdf, bdf):
             # because the Tax Policy Center did not provide any
             # statistics for this new deduction.  However, see the
             # following reform analysis for FOUR reforms.
-            "exp_totben_2022": 8.14,
+            "exp_totben_2022": 8.28,
             "exp_affpct_2022": 12.84,
             "exp_affben_2022": 324,
         },
@@ -108,7 +109,7 @@ def actual_results(rdf, bdf):
             # https://taxpolicycenter.org/model-estimates/T25-0257
             # Note that the $1081 TPC estimate is derived by dividing
             # the all-unit average of $320 by the 0.296 affpct.
-            "exp_totben_2022": 60.07,
+            "exp_totben_2022": 61.15,
             "exp_affpct_2022": 28.97,
             "exp_affben_2022": 1061,
         },
@@ -119,13 +120,13 @@ def actual_results(rdf, bdf):
     ]
     # create Tax-Calculator Records object
     recs = create_tmd_records(
-        data_path=tmd_variables_path,
-        weights_path=tmd_weights_path,
-        growfactors_path=tmd_growfactors_path,
+        data_path=TMD_VARIABLES_PATH,
+        weights_path=TMD_WEIGHTS_PATH,
+        growfactors_path=TMD_GROWFACTORS_PATH,
     )
     # create baseline_sim Calculator object for simyear and get its output
     policy_gf = taxcalc.GrowFactors(
-        growfactors_filename=str(policy_growfactors_path)
+        growfactors_filename=str(POLICY_GROWFACTORS_PATH)
     )
     pol = taxcalc.Policy(gfactor=policy_gf)
     pol.implement_reform(SOI_IITAX_SPEC)

tests/test_revenue_levels_cbo.py

@@ -18,14 +18,19 @@
 import pytest
 import taxcalc
 
+from tmd.storage import (
+    POLICY_GROWFACTORS_PATH,
+    TMD_GROWFACTORS_PATH,
+    TMD_WEIGHTS_PATH,
+)
 from tmd.imputation_assumptions import TAXYEAR, SOI_IITAX_SPEC
 
 # Per-year relative tolerance for all three aggregates.
 RELTOL = {
     2022: 0.01,
-    2026: 0.02,
-    2031: 0.05,
-    2036: 0.06,
+    2026: 0.01,
+    2031: 0.01,
+    2036: 0.01,
 }
 
 VARIABLES = ("n_returns_mil", "agi_bil", "iitax_bil")
@@ -37,29 +42,23 @@
     TAXYEAR != 2022,
     reason="expected values are calibrated to TAXYEAR=2022",
 )
-def test_revenue_levels_cbo(
-    tests_folder,
-    tmd_variables,
-    tmd_weights_path,
-    tmd_growfactors_path,
-    policy_growfactors_path,
-):
+def test_revenue_levels_cbo(tests_folder, tmd_variables):
     epath = tests_folder / "expected_cbo_levels_2022_data.yaml"
     with open(epath, "r", encoding="utf-8") as f:
         exp = yaml.safe_load(f)
 
     policy_gf = taxcalc.GrowFactors(
-        growfactors_filename=str(policy_growfactors_path)
+        growfactors_filename=str(POLICY_GROWFACTORS_PATH)
     )
     pol = taxcalc.Policy(gfactor=policy_gf)
     pol.implement_reform(SOI_IITAX_SPEC)
     rec = taxcalc.Records(
         data=tmd_variables,
         start_year=TAXYEAR,
         gfactors=taxcalc.GrowFactors(
-            growfactors_filename=str(tmd_growfactors_path)
+            growfactors_filename=str(TMD_GROWFACTORS_PATH)
         ),
-        weights=str(tmd_weights_path),
+        weights=str(TMD_WEIGHTS_PATH),
         adjust_ratios=None,
         exact_calculations=True,
         weights_scale=1.0,

tests/test_soi_sanity_2022.py

@@ -48,7 +48,12 @@
 import pandas as pd
 import taxcalc
 
-from tmd.storage import STORAGE_FOLDER
+from tmd.storage import (
+    STORAGE_FOLDER,
+    POLICY_GROWFACTORS_PATH,
+    TMD_GROWFACTORS_PATH,
+    TMD_WEIGHTS_PATH,
+)
 from tmd.imputation_assumptions import TAXYEAR, SOI_IITAX_SPEC
 
 SOI_YEAR = 2022
@@ -81,28 +86,23 @@ def _soi_target(soi_df, variable, count):
     return float(rows["Value"].iloc[0])
 
 
-def test_soi_sanity_2022(
-    tmd_variables,
-    tmd_weights_path,
-    tmd_growfactors_path,
-    policy_growfactors_path,
-):
+def test_soi_sanity_2022(tmd_variables):
     """Five weighted 2022 totals from TMD within 1% of SOI targets."""
     # Run TaxCalc at TAXYEAR to obtain c00100 (AGI) and iitax. e00200,
     # e26270, and s006 are available directly on the input frame but
     # using the post-calc_all arrays keeps one consistent source for
     # all five aggregates.
     policy_gf = taxcalc.GrowFactors(
-        growfactors_filename=str(policy_growfactors_path)
+        growfactors_filename=str(POLICY_GROWFACTORS_PATH)
     )
     pol = taxcalc.Policy(gfactor=policy_gf)
     pol.implement_reform(SOI_IITAX_SPEC)
     recs = taxcalc.Records(
         data=tmd_variables,
         start_year=TAXYEAR,
-        weights=str(tmd_weights_path),
+        weights=str(TMD_WEIGHTS_PATH),
         gfactors=taxcalc.GrowFactors(
-            growfactors_filename=str(tmd_growfactors_path)
+            growfactors_filename=str(TMD_GROWFACTORS_PATH)
         ),
         adjust_ratios=None,
         exact_calculations=True,