Fix birth_year calculation for two-child limit projections

[MaxGhenis]

Summary

Fixes the bug where two-child limit budgetary impact was incorrectly constant across projection years.

Problem

The birth_year variable was being loaded from the dataset as a static value, preventing it from being recalculated based on the projection year. This caused:

• Birth years to remain constant (e.g., 2006-2023 range in both 2026 and 2029)
• Number of children exempt from two-child limit to stay constant
• Cost of repealing the limit to be incorrectly flat across years

With a consistent age distribution in the microdata:

• 2026: 10-year-olds should represent children born in 2016 → exempt (< 2017)
• 2029: 10-year-olds should represent children born in 2019 → NOT exempt (≥ 2017)

Solution

Skip loading birth_year from the dataset in simulation.py, allowing the Variable formula (period.year - age) to calculate it fresh for each projection year.

# Skip birth_year - it should be calculated from age and period
# to properly reflect the projection year in multi-year datasets
if variable == "birth_year":
    continue

Results

Population Demographics

Before fix:

• 2026: 14,870 children born <2017 (exempt), birth years 2006-2023
• 2029: 14,870 children born <2017 (exempt), birth years 2006-2023
• Same cost in both years ❌

After fix:

• 2026: 10,638 children born <2017 (exempt), birth years 2009-2026
• 2029: 6,504 children born <2017 (exempt), birth years 2012-2029
• 4,134 more children subject to limit in 2029 ✓

Budgetary Impact (Repeal Cost)

With the fix applied:

• 2026: £3.28bn total (£0.00bn CTC, £3.28bn UC)
• 2029: £3.76bn total (£0.00bn CTC, £3.76bn UC)
• Change: +£0.47bn (+14.4%)

The cost now correctly increases over time as more children born after April 2017 enter the population ✓

Dependencies

• Companion PR in policyengine-uk-data: Computing marginal tax rates #198 (removes birth_year from dataset generation)

Test Plan

• Verified birth years now update correctly by projection year
• Confirmed number of affected children increases over time
• Ran microsimulation showing 14.4% cost increase from 2026 to 2029
• Changelog updated

🤖 Generated with Claude Code

[MaxGhenis]

Closing as unnecessary. Removing birth_year from the dataset (PolicyEngine/policyengine-uk-data#198) is sufficient - the Variable formula will automatically calculate it from age and period when it's not present in the input data.