Repository navigation
Social Security, pensions and self-employment grow per record at aggregate rates #9666
Description
Problem
For Microsimulation, the dataset is extended to later years by multiplying each record's dollar inputs by the growth of a national total (extend_single_year_dataset in policyengine_us/data/economic_assumptions.py). For three income sources, that total moves for reasons that should not reach each record's amount. Replacing those three rates removes most of the senior miss and about a third of the child miss in our 2025 SPM poverty forecast.
Growth from 2024 to 2025 in policyengine-us 2.2.1 (the version policyengine==6.0.0 pins), measured on the default population, against the CPS ASEC 2025 and 2026 public-use files (cross-sections, calendar 2024 and 2025):
| Input | Per-record growth | Index | CPS ASEC |
|---|---|---|---|
social_security_retirement, _disability, _survivors, _dependents |
+8.31% | calibration.gov.irs.soi.social_security → calibration.gov.cbo.social_security |
+4.04% per recipient 66 and over at the 2024 age mix (+3.45% all ages; unadjusted mean +3.9%, median +4.0%); COLA +2.5% |
| taxable and tax-exempt pensions; IRA, 401(k), 403(b), SEP and Keogh distributions | +19.47% | calibration.gov.irs.soi.taxable_pension_income / tax_exempt_pension_income → CBO taxable pension income |
pensions and annuities per recipient 65 and over: median +3.7%, mean +6.6%; retirement distributions: mean +4.7%, median unchanged |
self_employment_income_before_lsr, sstb_self_employment_income_before_lsr |
−5.76% | calibration.gov.irs.soi.self_employment_income → CBO net business income |
SEMP_VAL per recipient with positive income: median +8.7%, mean +13.1% |
employment_income_before_lsr |
+4.94% for everyone | calibration.gov.irs.soi.employment_income → CBO employment income |
+4.1% to +5.3% in every wage decile; by decile of SPM resources over threshold, no pattern distinguishable from sampling error (bottom decile +1.8 points over all earners, bootstrap SE 2.1) |
Weights grow with total population (0.93%). Recipient counts in the survey are not a usable comparison because the ASEC's population controls changed between the two files (the weighted 65+ population rises 6%).
#9526 took the population part out of these series. By its rule, main should now grow Social Security about 7.3%, pensions about 18.4% and self-employment about −6.6% per record in 2025 (our arithmetic from the per-capita rule, not model runs), so the self-employment problem is slightly worse on main.
Effect on the 2025 SPM forecast
The same 2025 run with only these inputs regrown, percentage-point change from 2024:
| 2025 run | All people | Under 18 | 65 and over |
|---|---|---|---|
| Census | +0.07 | −0.06 | +0.24 |
| As released (registered forecast) | +0.22 | +0.86 | −0.52 |
Social Security at gov.ssa.uprating (COLA), pensions at CPI-U |
+0.47 | +0.98 | +0.21 |
| Self-employment at the wage index | +0.01 | +0.47 | −0.59 |
| Wages by the survey's growth per decile of SPM resources over threshold (schedule within sampling error) | +0.09 | +0.71 | −0.59 |
| Wages by the survey's growth per wage decile | +0.24 | +0.85 | −0.52 |
| Social Security at the COLA, pensions at CPI-U annual average, self-employment at the wage index | +0.25 | +0.57 | +0.15 |
| Same, Social Security at the survey's 4.04% | +0.17 | +0.53 | −0.02 |
Among people whose 2024 SPM resources were 75–150% of their threshold, Social Security at 8.3% per record contributes 5.7 of the 8.4% mean resource growth for people 65 and over. None of the three Census changes is statistically significant at 90%, so these runs identify which assumptions matter, not their right values. Code, results and survey comparisons: analysis/prediction-grade-2025/uprating in PolicyEngine/spm-threshold-paper#10.
Questions and options
- Social Security series. The parameter file labels 2024 as SSA actual and 2025 as a CBO February 2026 projection. If both are CBO fiscal-year outlays, 2025 also includes the Social Security Fairness Act's retroactive payments, which should not raise every record's ongoing benefit. Worth confirming before choosing a fix.
- Option 1: Social Security per record at the COLA, with growth in beneficiaries coming from weights fitted to projected beneficiary counts and benefit totals. COLA-only growth leaves out new retirees' higher benefits and the Fairness Act increases unless the weights carry them; without reweighting it also lowers projected Social Security totals and the income tax on benefits below CBO, which changes budget estimates.
- Option 2: self-employment on a per-person earnings index (wages, or an SOI Schedule C series) rather than CBO's aggregate net business income.
- Option 3: review the pension series. If CBO's 19.5% reflects more recipients or one-time distributions, apply it through weights or a recipient split, not per record.
- Option 4 (lower priority): wage growth by family income. The CPS can't distinguish a family-income gradient in 2025 wage growth from sampling error, but the run above shows one would move child poverty noticeably. Worth revisiting with a source that is not selected on the outcome; growth for the same workers over time also includes mean reversion, which would overstate growth at the bottom if applied to a fixed population.
These choices change budget and distributional results as well as poverty, so they need a decision before a PR.
axiom: n/a: microsimulation uprating, not a policy rule.
Activity