Skip to article frontmatterSkip to article content

Methodology

PolicyEngine US Version

This analysis was conducted using PolicyEngine US version 1.398.0.

The simulation was last run on 2025-09-21 using Python 3.13.0.

Analysis period: 2026-2035

Dataset: 56,839 households and 146,133 individuals

Dataset and Microsimulation Methodology

This analysis uses the PolicyEngine US microsimulation model, which relies on the Enhanced CPS (2024) dataset.

Microdata Construction

The Enhanced CPS is constructed by PolicyEngine to address known limitations in the raw Current Population Survey (CPS), such as underreporting of income and lack of tax-specific variables. The construction process involves:

  1. Base Data: The 2024 Current Population Survey (CPS) Annual Social and Economic Supplement (ASEC).

  2. Tax Unit Construction: PolicyEngine groups CPS households into tax units (filing units) to accurately model the tax system.

  3. Imputation: Missing or underreported variables are imputed using machine learning techniques (quantile regression forests) trained on administrative data. This includes:

    • Imputation of capital gains and other capital income from the IRS Public Use File (PUF).

    • Imputation of itemized deductions and other tax-specific fields.

  4. Reweighting: The dataset is reweighted to match over (TODO: number from docs) administrative targets from IRS and SSA data, ensuring that aggregate estimates for income, taxes, and benefits align with official benchmarks.

Long-Term Projections: Economic Uprating and Demographic Calibration

Projecting tax revenue over a 75-year horizon requires simultaneously modeling two distinct but interrelated dynamics: how the economy evolves (wage growth, inflation, tax parameters) and how the population structure changes (aging, longevity, household composition). Traditional approaches typically sacrifice either economic sophistication or demographic realism. Our methodology preserves both through a two-stage process.

Two-Stage Projection Methodology

For each projection year (2025-2100), the model executes two sequential stages:

Stage 1: Economic Uprating PolicyEngine’s microsimulation engine projects each household’s economic circumstances forward using official macroeconomic assumptions. This stage adjusts all monetary values (wages, benefits, tax parameters, etc.) to reflect the target year’s economic conditions while preserving the distributional characteristics of the base microdata.

Stage 2: Demographic Calibration Household weights are adjusted to match Social Security Administration demographic and fiscal projections. This reweighting ensures the synthetic population reflects the target year’s age distribution and aggregate fiscal totals while maintaining consistency across all calculated variables through household-level operations.

Key Innovation: By performing all tax calculations at the household level before aggregation, this approach avoids person-to-household mapping inconsistencies that can arise in traditional methods where different variables are calculated at different units of analysis.

Economic Uprating Details

The model uses intermediate assumptions from the 2025 Social Security Trustees Report for key macroeconomic variables:

Different income categories are uprated using category-specific growth rates:

Tax parameters (brackets, standard deductions, credits) are uprated according to their statutory indexing rules (typically CPI-U for federal income tax provisions).

Demographic Calibration Method

After economic uprating, household weights are recalibrated using Generalized Regression (GREG) calibration:

Calibration Constraints

The GREG method can enforce up to three types of constraints simultaneously:

  1. Age Distribution

    • 86 categories: ages 0-84 individually, 85+ aggregated

    • Source: SSA Single Year Age demographic projections (2024 publication, latest available)

  2. Social Security Benefits

    • Total OASDI (Old-Age, Survivors, and Disability Insurance) benefit payments in nominal dollars

    • Ensures aggregate Social Security income matches SSA fiscal projections

    • Source: SSA Trustees Report 2025

  3. Taxable Payroll

    • Total earnings subject to Social Security taxation, properly accounting for the annual wage base cap

    • Calculated as: taxable_earnings_for_social_security + social_security_taxable_self_employment_income

    • Source: SSA Trustees Report 2025

    How the wage base cap is enforced:

    The two-variable calculation prevents double-counting and correctly handles the cap at the individual level:

    • taxable_earnings_for_social_security = min(wage_base_cap, total_W2_wages)

      • Caps each person’s W-2 wages at the annual wage base ($168,600 in 2024)

      • Sums across all employers for that person

    • social_security_taxable_self_employment_income = min(SE_income, wage_base_cap - taxable_W2_earnings)

      • Self-employment income can only fill remaining room under the cap

      • Explicitly subtracts W-2 taxable amount from the cap before applying to SE income

    Examples:

    Person with two 100kW−2jobs+100k W-2 jobs + 20k SE income:

    • W-2 taxable = min(168.6k,168.6k, 200k) = $168.6k

    • SE taxable = min(20k,20k, 168.6k - 168.6k)=168.6k) = 0

    • Total taxable payroll = $168.6k (cap enforced, no excess)

    Person with 150kW−2+150k W-2 + 30k SE income:

    • W-2 taxable = min(168.6k,168.6k, 150k) = $150k

    • SE taxable = min(30k,30k, 168.6k - 150k)=150k) = 18.6k

    • Total taxable payroll = $168.6k (SE income fills remaining cap room)

    This ensures the national aggregate correctly reflects total taxable earnings with no double-counting of wages above the cap.

When using all three constraints, GREG calibration achieves less than 0.1% error on each target, ensuring the microsimulation replicates official demographic and fiscal aggregates with high precision.

Data Sources for Long-Term Projections

Demographic and fiscal targets come from official SSA publications:

Validation of Long-Term Projections

Each projected dataset is validated against three benchmarks to ensure calibration accuracy:

  1. Population Demographics: Age-specific population counts match SSA projections (exact match by construction for GREG)

  2. Social Security Benefits: Aggregate benefits match SSA fiscal projections within 0.1% (when using GREG with Social Security constraint)

  3. Taxable Payroll: Aggregate taxable earnings match SSA economic projections within 0.1% (when using GREG with payroll constraint)

Validation Example: 2027 (Near-term)

from policyengine_us import Microsimulation
import numpy as np

# Load calibrated 2027 dataset
sim = Microsimulation(dataset="hf://policyengine/test/2027.h5")

# 1. Validate Social Security benefits
ss_estimate_b = sim.calculate("social_security").sum() / 1e9
ss_trustees_b = 1_800  # SSA Trustees Report 2025, Table VI.G10
assert round(ss_estimate_b) == ss_trustees_b
print(f"✓ Social Security 2027: ${ss_estimate_b:.1f}B (target: ${ss_trustees_b}B)")

# 2. Validate taxable payroll
taxable_estimate_b = (
    sim.calculate("taxable_earnings_for_social_security").sum() / 1e9 +
    sim.calculate("social_security_taxable_self_employment_income").sum() / 1e9
)
ss_trustees_payroll_b = 11_627  # SSA Trustees Report 2025, Table VI.G6
assert round(taxable_estimate_b) == ss_trustees_payroll_b
print(f"✓ Taxable payroll 2027: ${taxable_estimate_b:.1f}B (target: ${ss_trustees_payroll_b}B)")

# 3. Validate population demographics (example: 6-year-olds)
person_weights = sim.calculate("age", map_to="person").weights
person_ages = sim.calculate("age", map_to="person").values
total_age6_est = np.sum((person_ages == 6) * person_weights)
ss_age6_pop = 3_730_632  # SSA Single Year Age Projections 2024
assert round(total_age6_est) == ss_age6_pop
print(f"✓ Age 6 population 2027: {total_age6_est:,.0f} (target: {ss_age6_pop:,})")

Output:

✓ Social Security 2027: $1800.0B (target: $1800B)
✓ Taxable payroll 2027: $11627.0B (target: $11627B)
✓ Age 6 population 2027: 3,730,632 (target: 3,730,632)

Validation Example: 2100 (Long-term)

# Load calibrated 2100 dataset
sim = Microsimulation(dataset="hf://policyengine/test/2100.h5")

# 1. Validate Social Security benefits
ss_estimate_b = sim.calculate("social_security").sum() / 1e9
ss_trustees_b = 34_432  # SSA Trustees Report 2025, Table VI.G10 (nominal dollars)
assert np.allclose(ss_estimate_b, ss_trustees_b, rtol=0.0001)
print(f"✓ Social Security 2100: ${ss_estimate_b:.1f}B (target: ${ss_trustees_b}B)")

# 2. Validate taxable payroll
taxable_estimate_b = (
    sim.calculate("taxable_earnings_for_social_security").sum() / 1e9 +
    sim.calculate("social_security_taxable_self_employment_income").sum() / 1e9
)
ss_trustees_payroll_b = 187_614  # SSA Trustees Report 2025, Table VI.G6
assert round(taxable_estimate_b) == ss_trustees_payroll_b
print(f"✓ Taxable payroll 2100: ${taxable_estimate_b:.1f}B (target: ${ss_trustees_payroll_b}B)")

# 3. Validate population demographics (example: 6-year-olds)
person_weights = sim.calculate("age", map_to="person").weights
person_ages = sim.calculate("age", map_to="person").values
total_age6_est = np.sum((person_ages == 6) * person_weights)
ss_age6_pop = 5_162_540  # SSA Single Year Age Projections 2024
assert round(total_age6_est) == ss_age6_pop
print(f"✓ Age 6 population 2100: {total_age6_est:,.0f} (target: {ss_age6_pop:,})")

Output:

✓ Social Security 2100: $34432.0B (target: $34432B)
✓ Taxable payroll 2100: $187614.0B (target: $187614B)
✓ Age 6 population 2100: 5,162,540 (target: 5,162,540)

Data Sources for Validation Targets:

This validation ensures that policy impact estimates are grounded in official demographic and economic projections, providing a realistic foundation for 75-year fiscal analysis.

Limitations and Data Caveats

While the Enhanced CPS improves upon raw survey data through tax record integration and machine learning imputation, estimates remain subject to sampling error and imputation uncertainty inherent in survey-based microsimulation.

Behavioral Modeling Limitations

The current analysis incorporates labor supply elasticities but does not model several additional behavioral responses that could affect long-term fiscal estimates:

1. Social Security Claiming Age Optimization

The model treats retirement claiming decisions as static, without optimizing claiming age (62-70) in response to tax policy changes. In reality, individuals facing higher marginal tax rates on Social Security benefits may strategically delay claiming to take advantage of actuarial adjustments and potentially lower future tax rates. This optimization behavior could amplify or dampen the revenue effects of Social Security taxation reforms.

2. Employer Compensation Structure Responses

The analysis treats employer-provided compensation components (wages, health insurance, retirement contributions) as exogenous inputs. It does not model how employers might restructure total compensation packages in response to differential tax treatment. For example, policies that increase the tax advantage of certain benefits could lead firms to shift compensation from wages to tax-preferred benefits, affecting both revenue estimates and distributional outcomes.

3. Generational Differences in Labor Supply Elasticities

While the model incorporates labor supply responses using elasticity estimates, it applies uniform elasticities across time periods and age groups (with a blanket doubling for workers 65+). This approach faces two limitations:

Age variation within current population: Empirical evidence suggests labor supply elasticities vary substantially across the lifecycle, with older workers and those near retirement potentially showing different behavioral responses than younger workers.

Cohort variation across the 75-year horizon: More critically, the model assumes that labor supply elasticities estimated from today’s older workers will apply to future cohorts over the entire 2025-2100 projection period. In reality, workers reaching age 65 in 2050 or 2080 will have fundamentally different financial characteristics than today’s 65-year-olds:

These generational shifts affect labor supply decisions: workers with lower pension wealth and home equity may exhibit different work-retirement tradeoffs than current retirees, potentially making them more (or less) sensitive to tax policy changes. The current model’s elasticity parameters are calibrated to contemporary populations and held constant across the projection horizon, which may not accurately capture behavioral responses of future cohorts.

4. Cohort-Specific Income Composition Changes

The model projects aggregate income by source through 2100 using time-varying targets based on CBO projections (2025-2035) and SSA Trustees Report extensions (2036-2100). These projections account for economy-wide growth in employment income, pensions, Social Security benefits, capital gains, and other sources. However, the model assumes the age profile of income sources remains relatively constant over time.

The critical limitation is that future cohorts reaching retirement age will have fundamentally different income compositions than today’s retirees:

What the model currently assumes:

What cohort-specific modeling would capture:

This shift reflects structural changes already underway:

For Social Security taxation reforms, this matters significantly: policies primarily affecting benefit recipients will have increasing budgetary and distributional impacts as Social Security becomes a larger share of retirement income for future cohorts. The current model’s constant age-income profile may underestimate the long-term fiscal and distributional effects of reforms targeting Social Security benefits or other retirement income sources.

The infrastructure for time-varying projections exists in PolicyEngine’s calibration framework, but the granularity for cohort-specific structural changes to income composition is currently absent.

These limitations are inherent to the microsimulation approach used here, which focuses on mechanical tax calculations with standard labor supply responses rather than full dynamic behavioral modeling. However, the PolicyEngine architecture provides pathways for addressing several of these limitations:

Potential Enhancements

Cohort-Specific Elasticity Parameters

The existing labor supply elasticity framework (substitution and income elasticity by earnings decile) could be extended to vary by birth cohort. PolicyEngine already tracks birth year and has time-varying parameter infrastructure through 2100. The elasticity parameters could be stratified:

Birth-Cohort Income Composition Targets

PolicyEngine already has Rhode Island’s Social Security taxability model that varies by birth year, demonstrating the technical feasibility of cohort-specific parameters. The income-by-source calibration targets could be extended:

Claiming Age Optimization Module

The current binary retirement indicator (is_retired.py) could be enhanced to:

Employer Behavior Linkages

Currently employer contributions are exogenous inputs. A “total compensation” framework could:

The infrastructure for time-varying, cohort-specific projections largely exists in PolicyEngine’s parameter system and microsimulation architecture. Implementing these enhancements would require expanding calibration targets and adding birth-cohort dimensions to existing behavioral parameters rather than fundamental architectural changes.

Implementation

Reform Specifications

Scoring Methodology

This analysis presents two sets of conventional scoring estimates:

Without Labor Supply Responses: Holds taxpayer behavior constant, isolating the direct mechanical effect of policy changes.

With Labor Supply Responses: Incorporates labor supply elasticities based on CBO estimates, with elasticities doubled for workers aged 65 and older based on meta-analysis findings. This captures behavioral responses to changes in effective tax rates but does not include broader macroeconomic feedback effects.

For most Social Security taxation reforms, estimates with and without labor supply responses are very similar (typically within 5%), with the main exceptions being the Roth-style swap options (Options 5 and 6) which show larger differences due to the behavioral effects of taxing employer payroll contributions.

Projection Period

All estimates cover a 75-year projection window from 2026 through 2100, providing both near-term (10-year) and long-term fiscal impact assessments.

Social Security Taxation Reform: Policy Analysis and Impact Assessment
Policy Options
Social Security Taxation Reform: Policy Analysis and Impact Assessment
Household Impacts