Add November 2025 Autumn Budget income source tax rate increases

[MaxGhenis]

Summary

Implements the income source-specific tax rate increases from the November 2025 Autumn Budget, as detailed in the OBR Economic and Fiscal Outlook November 2025, paragraph 3.33.

Dividends (from April 2026)

• Basic rate: 8.75% → 10.75% (+2pp)
• Higher rate: 33.75% → 35.75% (+2pp)
• Additional rate: unchanged at 39.35% (per OBR, only basic and higher rates mentioned)

Savings (from April 2027)

• Basic rate: 20% → 22% (+2pp)
• Higher rate: 40% → 42% (+2pp)
• Additional rate: 45% → 47% (+2pp)

Property (from April 2027)

• Basic rate: 20% → 22% (+2pp)
• Higher rate: 40% → 42% (+2pp)
• Additional rate: 45% → 47% (+2pp)

Changes

• Updated dividends.yaml with new rates from April 2026
• Created savings.yaml with separate savings income tax rates
• Created property.yaml with separate property income tax rates
• Updated savings_income_tax.py to use the new savings-specific rates
• Created property_income_tax.py variable to calculate property income tax separately
• Added comprehensive tests for all rate changes (pre/post reform)

Test plan

• All 14 new tests pass for income source tax rate changes
• All 598 policy tests pass
• All 26 pytest tests pass
• Code formatted with black

Closes #1383, #1384, #1385

🤖 Generated with Claude Code

[MaxGhenis]

Implementation Assumptions

The following assumptions were made based on the OBR Economic and Fiscal Outlook November 2025, which may need revision once formal legislation is enacted:

1. Dividend additional rate unchanged: The OBR only mentions increases to basic and higher dividend rates (to 10.75% and 35.75%). The additional rate is assumed to remain at 39.35%. If legislation increases this rate too, we'll need to update.

2. Simple rate structure: The implementation assumes the new rates simply replace the existing rates at each band, with no additional complexity (e.g., new thresholds, tapers, or allowance changes).

3. Property income stacking order: Property income tax is calculated assuming property income is stacked on top of other earned income (employment, pension, self-employment) when determining which band it falls into. This matches current UK tax treatment but may need adjustment if the legislation specifies a different ordering.

4. No interaction with Scottish rates: The property income tax implementation currently doesn't have special handling for Scottish taxpayers. If Scotland adopts different property/savings rates, this will need to be updated.

5. Effective dates: We used April 6 (start of tax year) as the effective date. The OBR references "April 2026" and "April 2027" without specifying exact dates.

6. No behavioral adjustments: The OBR notes behavioral effects (e.g., forestalling for dividends, shift to ISAs for savings), but these are not modeled in the parameter/variable changes - they're OBR's revenue estimates rather than policy mechanics.

These assumptions follow the most natural interpretation of the OBR description but should be validated against the actual Finance Act when published.

[MaxGhenis]

Revenue Estimates vs OBR Projections

Using PolicyEngine's microsimulation with the enhanced FRS dataset:

Policy	PolicyEngine (2029)	OBR Estimate	Ratio
Dividends (+2pp basic/higher)	£3.8bn	£1.2bn	3.2x
Savings (+2pp all rates)	£0.02bn	£0.5bn	0.04x
Property (+2pp all rates)	£0.8bn	£0.5bn	1.6x
Combined	£4.6bn	£2.1bn	2.2x

Analysis of Discrepancies

Dividends (3.2x higher than OBR)

• Taxable dividend base in PolicyEngine: £184bn
• OBR behavioral adjustment: -£0.3bn forestalling in 2026-27, plus ongoing behavioral responses
• Possible data issue: FRS may overestimate dividend income at higher income levels
• Note: We correctly only increased basic/higher rates (not additional) per OBR description

Savings (25x lower than OBR)

• Taxed savings base in PolicyEngine: only £1.1bn
• Most savings income shielded by: ISAs (tax-free), Personal Savings Allowance (£1k/£500/£0), Savings Starter Rate
• FRS likely under-captures savings interest income
• OBR has access to HMRC administrative data with better coverage

Property (1.6x higher than OBR)

• Taxable property base: £39bn
• OBR includes behavioral effects (higher taxes → lower rents → lower property values)
• Reasonably close given uncertainties

Implications

The parameter values are correct per the OBR description. The revenue estimate discrepancies are due to:

1. FRS data limitations (especially for savings/dividend income)
2. OBR behavioral adjustments not modeled here
3. Different population/income projections

These issues are separate from the policy implementation itself.

[MaxGhenis]

Revenue Estimate Analysis

I compared PolicyEngine's revenue estimates against OBR projections for these three policies:

Methodology

Used microsimulation comparing baseline vs reform scenarios for 2026/27 to 2029/30.

Results Summary

Income Source	PolicyEngine Revenue	OBR Revenue	Ratio
Dividends	£4.7bn (2026/27)	£2.0bn (2026/27)	2.4x
Savings	£0.18bn (2027/28)	£0.9bn (2027/28)	0.2x
Property	£0.79bn (2027/28)	£0.6bn (2027/28)	1.3x

Investigation: Calibration Target Discrepancy

The dividend income discrepancy traces to the incomes_projection.csv calibration targets in policyengine-uk-data:

Income Type	SPI Raw (2021)	Projection (2024)	Ratio
Dividends	£71.6bn	£185.6bn	2.6x
Savings	£2.5bn	£6.4bn	2.6x
Property	£25.8bn	£66.5bn	2.6x

The raw SPI data shows £63bn weighted dividend income, matching the £71.6bn target reasonably well. However, when create_income_projections() reweights and simulates forward, all income types are inflated ~2.6x.

Impact on Revenue Estimates

The calibration targets used in loss.py for the Enhanced FRS are significantly higher than raw SPI data, causing:

• Dividend revenue estimates to be ~2.4x higher than OBR
• This appears to be a data issue in policyengine-uk-data, not a policy implementation issue

Recommendation

The policy implementation in this PR is correct. The revenue estimate discrepancy should be addressed separately in policyengine-uk-data by investigating the create_income_projections() reweighting process.

🤖 Generated with Claude Code