policyengine-us Extend economic uprating parameters to 2100 using SSA Trustees data

Summary

This PR extends all economic uprating parameters from 2036 through 2100 using data from the Social Security Administration (SSA) Trustees Report 2025, ensuring PolicyEngine US can perform accurate long-term microsimulations through the end of the century.

What Changed

Previously, most uprating parameters only extended to 2035 (CBO projections). This PR adds 65 years of additional projections (2036-2100) for:

✅ Inflation indices

SSA CPI-W (Social Security COLA)
IRS CPI-U (tax parameter indexing)

✅ Wage growth

SSA Average Wage Index (AWI)

✅ Income categories (all aggregate totals)

Self-employment income
Interest & ordinary dividends
Qualified dividends
Pension income
Capital gains
Adjusted Gross Income (AGI)

✅ Demographics

Total US population projections

Methodology

Growth Rate Strategy

All projections use SSA Trustees data for consistency and to leverage their comprehensive 75-year forecasts:

Income Category	Growth Rate	2100 Value	Rationale
Wages	SSA AWI (~3.5%)	$1.03T	Official wage index used for Social Security
Self-employment	SSA GDP (~4.0%)	$30.8T	Business activity tracks overall economy
Interest & ordinary divs	SSA GDP (~4.0%)	$7.3T	Fixed income grows with savings/GDP
Qualified dividends	GDP + 0.5% (~4.5%)	$32.6T	Equity outperformance premium
Pensions	SSA GDP (~4.0%)	$46.8T	Wage growth + demographic trends
Capital gains	GDP + 0.5% (~4.5%)	$61.2T	Asset appreciation tracks equity returns
Population	SSA demographics (+0.37%)	458.3M	Official demographic projections

Key Design Decisions

1. Why SSA Trustees instead of CBO?

SSA projects through 2100 (vs CBO's 2055)
More authoritative for Social Security/demographic projections
Internally consistent across all economic variables

2. Why maintain AGI ratios?

Each income category maintains its 2035 ratio to AGI
Preserves economic relationships between income types
Simplifies long-term projections while remaining realistic

3. Why equity premium for dividends/capital gains?

Historical data shows equities outperform GDP by ~0.5-1% annually
More realistic than CBO's 1.74% cap gains growth (which reflects realization timing effects)
Both income types come from equity ownership → should have similar long-term growth

4. Population growth already in income projections?

✅ Yes! GDP growth = Population growth + Productivity + Inflation
Income aggregates inherently include population via GDP growth
Population parameter used separately for survey weight adjustments

Technical Implementation

Files Modified

Core parameter files:

policyengine_us/parameters/
├── gov/
│   ├── ssa/
│   │   ├── uprating.yaml          # Extended CPI-W to 2100
│   │   └── wage_index.yaml        # NEW: AWI projections to 2100
│   └── irs/
│       └── uprating.yaml          # Extended CPI-U to 2100
└── calibration/gov/
    ├── cbo/
    │   └── income_by_source.yaml  # Extended 6 income categories
    └── census/populations/
        └── total.yaml             # Updated with SSA population data

Source data & documentation:

policyengine_us/uprating_project/
├── Trustees_Highlights.ods        # SSA GDP, CPI-W, AWI data
├── SSPopJul_TR2024.csv           # Population by age (1941-2100)
├── CBO_Long_term.xlsx            # CBO historical comparisons
├── UPRATING_PLAN.md              # Original methodology plan
└── income_uprating_summary.md    # Implementation summary

Example: Self-Employment Income Extension

self_employment_income:
  # ... CBO values through 2035 ...
  2035-01-01: 2_412_400_000_000
  
  # Extended projections (2036-2100) using SSA GDP growth
  # Methodology: Applied SSA GDP growth rates to maintain consistency with AGI
  # Self-employment maintains 9.24% of AGI (2035 ratio)
  2036-01-01: 2_465_540_268_658
  2040-01-01: 2_879_619_127_752
  2050-01-01: 4_237_230_502_324
  2075-01-01: 11_319_967_194_846
  2100-01-01: 30_782_649_357_335

Data Sources

Primary Source: SSA Trustees Report 2025

Main report: https://www.ssa.gov/oact/TR/2025/

Single-year supplementary tables: https://www.ssa.gov/oact/tr/2025/lrIndex.html

Contains: GDP, CPI-W, Average Wage Index projections (2025-2100)
Used for: All economic growth rates

Demographics (2024 latest): https://www.ssa.gov/oact/HistEst/Population/2024/

Contains: Population by single year of age (1941-2100)
Used for: Total US population projections

Supplementary: CBO Long-Term Economic Projections

https://www.cbo.gov/data/budget-economic-data

See: "Long Term Economic Projections" spreadsheet
Contains: CPI-U, income category projections (2025-2055)
Used for: Historical comparisons and validation

Validation

Growth Rate Comparison

Rate	CBO (2025-2035)	SSA (2025-2100)	Notes
Nominal GDP	4.37% → 3.42%	~4.0%	SSA slightly higher, more stable
Wages (AWI)	3.74% → 3.29%	~3.5%	Well aligned
CPI-U	~2.3%	~2.3%	Very close
CPI-W	~2.4%	~2.4%	Identical methodology

SSA vs CBO Population

Year	SSA Trustees	CBO	Difference
2025	346.6M	349.8M	-0.9%
2035	367.2M	364.1M	+0.8%
2050	389.7M	371.5M	+4.9%
2100	458.3M	(no data)	-

SSA projects slightly higher long-term population growth than CBO.

Impact

Microsimulation Benefits

Extended time horizon: Can now simulate tax/benefit policies through 2100
Demographic analysis: Enables studies of aging population effects on entitlements
Long-term projections: Climate policy, Social Security reforms, etc.
Internal consistency: All parameters use same SSA Trustees source

No Breaking Changes

✅ All existing parameters (2015-2035) unchanged
✅ Only adds future values beyond CBO's 2035 horizon
✅ Maintains backward compatibility
✅ Existing simulations produce identical results

Testing

Validation Performed

[x] Growth rates match SSA Trustees Report values
[x] Income ratios to AGI maintained from 2035 baseline
[x] Population values sum correctly across all age groups
[x] Equity premium consistently applied to dividend/cap gains
[x] All YAML files parse correctly
[x] Values formatted with underscore thousands separators

Changelog Entry

- bump: minor
  changes:
    added:
      - Extended economic uprating parameters to 2100 using SSA Trustees Report 2025 data
      - Added SSA Average Wage Index projections through 2100  
      - Extended income category projections (self-employment, dividends, pensions, capital gains)
      - Updated population projections with SSA Trustees demographic data through 2100

Future Work

Potential enhancements not included in this PR:

State-level population projections (currently only national total)
Age-stratified economic growth rates (currently aggregate)
Alternative scenarios (low/high growth variants from Trustees)
Automatic parameter updates when new Trustees reports published

Questions?

For detailed methodology, see:

/policyengine_us/uprating_project/UPRATING_PLAN.md
/policyengine_us/uprating_project/income_uprating_summary.md

🤖 Generated with Claude Code

Co-Authored-By: Claude [email protected]

Oct 29 '25 13:10 baogorek

CPI Data Source Decisions

After investigating the various CPI-related parameters, here's what we've decided:

✅ Parameters Using Authoritative Sources

gov.ssa.uprating (SSA CPI-W uprating, indexed to 2025=100)
- Source: SSA Trustees Report 2025, Table VI.G6 (Intermediate Assumptions)
- Coverage: Complete projections 2024-2100 in YAML file
- Usage: Referenced directly by Social Security benefit variables (retirement, disability, SSI)
- No extension needed: SSA provides full projections through 2100
gov.bls.cpi.c_cpi_u (Chained CPI-U)
- Source: CBO January 2025 projections
- Coverage: 2000-2034 in YAML, extended programmatically 2035-2100
- Usage: Used to compute IRS uprating (Sep-Aug rolling averages)
- Extension method: Growth rate from 2033-2034 applied through 2100
gov.irs.uprating (IRS tax parameter uprating)
- Source: Computed from extended C-CPI-U
- Method: Sep-Aug rolling average of C-CPI-U for each year
- Fixed: Previously used a fixed growth rate for 2036+; now properly computes from extended C-CPI-U

Parameters NOT Used

gov.bls.cpi.cpi_w (Monthly BLS CPI-W)
- Status: Not referenced by any policy variables in the codebase
- Decision: Removed from programmatic extension in uprating_extensions.py
- Note: All Social Security calculations use gov.ssa.uprating instead

Summary

SSA programs: Use SSA Trustees CPI-W projections (gov.ssa.uprating)
IRS tax parameters: Use CBO C-CPI-U projections (gov.irs.uprating computed from gov.bls.cpi.c_cpi_u)
This ensures each parameter uses its most authoritative source

Oct 30 '25 21:10 baogorek

I'm going to keep moving here for the time being

● Quick Summary - No Glaring Mismatches! ✓

  What we fixed:
  - ✅ Employment income: Now uses NAWI (63pp divergence) - CORRECT
  - ✅ Self-employment income: Now uses NAWI (63pp divergence) - CORRECT
  - ✅ Social Security: Uses SSA CPI-W COLA (12pp divergence) - CORRECT

  What's still using CBO/GDP growth (73pp divergence):
  - Dividends
  - Capital gains
  - Interest income
  - Pension income

  Is this a problem? Probably intentional and fine because:
  - Capital income historically DOES grow faster than wage income
  - High earners have more capital income, which compounds faster
  - This may be realistic for distributional analysis

  Bottom line: The system is now internally consistent for your long-term distributional analysis. The main wage income sources use appropriate wage
  growth (NAWI), and capital income uses faster growth rates which may be realistic.

  You're good to keep moving! 👍

Oct 31 '25 21:10 baogorek

I'm having Claude Code create some new tests.

Nov 13 '25 14:11 baogorek

@MaxGhenis I've got a bit of a Catch-22 here:

Oct 30-31: Merge Max's TOB revenue code

❌ Codecov reports 54.6% patch coverage (49 missing lines)
🟢 Quick Feedback passes, but nobody notices it takes nearly 4 hours

Nov 12: Last successful run before coverage fix

🟢 Quick Feedback: SUCCESS after 3h52m (already barely working)
⚠️ Coverage still failing (54.6%)

Nov 13: Fix coverage by adding 30+ comprehensive tests

✅ Tests improve coverage as intended
❌ Quick Feedback: FAILURE after 38 minutes (timeout)
🟢 Full Suite: Still passing (42-45 minutes, no coverage overhead)

Nov 14: Add more edge case tests

❌ Quick Feedback: FAILURE after 36 minutes (timeout)
🟢 Full Suite: Still passing

Root Cause: TOB revenue tests use simulation branching which is extremely slow when combined with coverage tracking. Adding tests to fix coverage made the timeout problem visible.

The Catch-22:

Don't add tests → Codecov blocks PR
Add tests → Quick Feedback times out

Nov 14 '25 02:11 baogorek

Codecov Report

:white_check_mark: All modified and coverable lines are covered by tests. :white_check_mark: Project coverage is 100.00%. Comparing base (66a73e5) to head (25d0954). :warning: Report is 9 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff            @@
##            master     #6744   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files            2         7    +5     
  Lines           38       131   +93     
  Branches         3         6    +3     
=========================================
+ Hits            38       131   +93

Flag	Coverage Δ
unittests	`100.00% <100.00%> (ø)`

Flags with carried forward coverage won't be shown. Click here to find out more.

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

:rocket: New features to boost your workflow:

:snowflake: Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

Nov 14 '25 13:11 codecov[bot]

Claude's been running all day trying to get this coverage. It's like it's been playing a video game without the controller. Apparently this is what it's stuck on.

● Why the Circular Dependency Makes Testing Hard

  Here's the circular dependency problem that's preventing test coverage:

  The Deadlock Situation:

  1. labor_supply_behavioral_response.py (line 62-68) tries to calculate:
  response = add(
      person,
      period,
      [
          "income_elasticity_lsr",
          "substitution_elasticity_lsr",
      ],
  )
  2. BUT both those variables have this attribute:
    - income_elasticity_lsr.py line 10: requires_computation_after = "labor_supply_behavioral_response"
    - substitution_elasticity_lsr.py line 10: requires_computation_after = "labor_supply_behavioral_response"
  3. This creates a deadlock:
    - labor_supply_behavioral_response says: "I need to compute income_elasticity_lsr and substitution_elasticity_lsr"
    - Those variables say: "We can't be computed until AFTER labor_supply_behavioral_response is done"
    - Result: Infinite wait/timeout

  Why Tests Can't Run:

  When test_labor_supply_response_only_substitution_elasticity tries to calculate labor_supply_behavioral_response with a baseline and non-zero
  elasticities:
  - It enters the main calculation path (past line 17)
  - Gets past the re-entry guard (lines 20-27)
  - Creates branches (lines 30-60)
  - HANGS at line 62 when trying to calculate the elasticity variables
  - The test times out after 30+ seconds
  - Lines 62-83 never execute, so they show as uncovered

  The Coverage Impact:

  - Lines that can't be tested: 62-83 (the actual calculation, branch cleanup, and finally block)
  - Current coverage: ~41% (only the early return paths are tested)
  - Potential coverage: ~94% (if we could test the full execution path)

  The YAML tests work because they only test the early return scenarios (no baseline, or zero elasticities) which exit before hitting the circular
  dependency.

  This is why your Python test with mocking or fixing the requires_computation_after attributes would be needed to achieve full coverage - but since you
  don't want to modify the base code, we'd need to use sophisticated mocking to intercept these variable calculations and provide mock values instead.

Nov 15 '25 01:11 baogorek

Note: If 3e0f292 passes and 25d0954 fails, it's only due to a timeout b/c of the runner being set back to the default. To Pavel's point, perhaps we should split up this PR into multiple. The default runner is timing out b/c so many directories have been touched. I'm going to leave this for Monday. FYI @MaxGhenis

Nov 15 '25 02:11 baogorek