Impute W-2 Wage Share variable for Qualified Business Income Deduction

taxdata/puf/finalprep.py

@@ -2,6 +2,7 @@
 import pandas as pd
 import taxcalc as tc
 from .impute_pencon import impute_pension_contributions
+from .impute_qbid_w2 import impute_PT_binc_w2_wages
 from .constants import UNUSED_READ_VARS
 from pathlib import Path
 
@@ -57,6 +58,9 @@ def finalprep(data):
     # - Replace e20500 with g20500:
     data = replace_20500(data)
 
+    # - Impute PT_binc_w2_wages:
+    data = impute_PT_binc_w2_wages(data)
+
     data["s006"] = data["matched_weight"] * 100
 
     # - Remove variables not expected by Tax-Calculator:
@@ -323,3 +327,4 @@ def replace_20500(data):
     )
     data["g20500"] = np.int_(gross.round())
     return data
+

taxdata/puf/impute_qbid_w2.py

@@ -0,0 +1,87 @@
+"""
+Impute W-2 wages share in Qualified Business Income in PUF.
+
+The objective of this imputation is to improve the accuracy of Qualified
+Business Income Deduction (QBID) revenue estimates produced by Tax-Calculator.
+In the absence of a valid W-2 wage share variable (e.g., when it is set to zero),
+Tax-Calculatorgenerates QBID values that diverge significantly from benchmark
+estimates published by agencies such as the Congressional Budget Office (CBO).
+
+Since no microdata source contains reliable information on the W-2 wage share
+associated with pass-through income, a statistical imputation is required.
+We calibrate the imputation to match CBO's published QBID revenue estimates.
+To be noticed, there is an observed inconsistency on the CBO published QBID
+estimates between the years before 2020 and the years after 2021. So only the
+QBID value from the years 2021 ~ 2025 are selected targets, which would better
+reflect the policy environments of the recent years. Targets will be updated as
+new CBO QBID projections become available.
+
+The imputation strategy is to minimize the distance between Tax-Calculator's
+simulated QBID revenue and CBO's published targets through an optimization
+process. Specifically, we define a loss function to measure the year-by-year
+squared deviations in aggregate QBID amounts and solve for the W-2 wage share
+parameter which minimizes this loss.
+
+The imputation will add the PT_w2_binc_wages into puf.csv file.
+"""
+
+import pandas as pd
+import taxcalc as tc
+import numpy as np
+from scipy.optimize import minimize_scalar
+from functools import partial
+
+
+def impute_PT_binc_w2_wages(data):
+    """
+    Main function in impute_qbid_w2.py file.
+    Argument: puf.csv DataFrame just before imputation is done.
+    Returns: puf.csv DataFrame with imputed W-2 wage share variable.
+    """
+    # add the qualified business income variable, to be noticed the self employment 
+    # part is not included, to simplyfy calculation.
+    data["qbinc"] = data["e00900"] + data["e26270"] + data["e02100"] + data["e27200"]
+    # solve for W-2 wage share in QBI
+    w2_ratio = opt_ratio(data)
+    data["PT_binc_w2_wages"] = w2_ratio * data["qbinc"]
+    del data["qbinc"]
+    return data
+
+
+def qbid_value(data, ratio):
+    """
+    Function that calculate the QBID aggregates through years.
+    """
+    qbided = []
+    df = data.copy()
+    df["PT_binc_w2_wages"] = ratio * df["qbinc"]
+    pol = tc.Policy()
+    recs = tc.Records(data=df)
+    calc0 = tc.Calculator(policy=pol, records=recs)
+    for year in range(2021, 2026):
+        calc0.advance_to_year(year)
+        calc0.calc_all()
+        qbided.append(calc0.weighted_total("qbided"))
+    return np.array(qbided)
+
+
+def loss_function(data, ratio):
+    """
+    Function that estimates loss between between Tax-Calculator's
+    simulated QBID revenue and CBO's published targets on QBID aggregates.
+    """
+    # CBO QBID aggregates for the year 2021 ~ 2025
+    target = [205.1e9, 215.7e9, 229.5e9, 247.1e9, 258.3e9]
+    return np.sum((qbid_value(data, ratio) - target) ** 2)
+
+
+def opt_ratio(data):
+    """
+    Function that solve for the W-2 wage share from the loss function
+    Argument: puf.csv before imputation.
+    Returns: percentage of W-2 wage income in the Qualified Business Income.
+    """
+    loss = partial(loss_function, data)
+    res = minimize_scalar(loss, bounds=(0, 1), method="bounded")
+    # print(f"Optimal ratio: {res.x:.4f}, target: {res.fun:.4f}")
+    return res.x