uk-energy-reforms

PolicyEngine UK reforms and analysis for household energy bill support. Each reform is installed into policyengine-uk at runtime as parameters and variables. Reforms are switched and tuned with ordinary parameter-change dicts, so the same code serves household calculations, microsimulation runs and, later, a dashboard.

Reforms

Targeted energy discount (reforms/targeted_energy_discount)

The Resolution Foundation's proposal in Billing me softly (Brewer, Clegg and Marshall, 10 August 2026). A household in Great Britain is eligible if:

• it receives a passporting benefit (the Warm Home Discount list: Universal Credit, Pension Credit, Housing Benefit, income-related ESA, income-based JSA, Income Support); or
• the highest taxable income of any member is below the threshold. Taxable income is total_income, so pensions count as well as earnings.

Support follows a schedule by the household's assessed income. Passported households are placed at the lower of their own income and passport.assessed_income.

The parameters live under gov.contrib.targeted_energy_discount (see parameters.yaml):

• in_effect switches the scheme on. It is off by default, which leaves the system numerically identical to the policyengine-uk baseline.
• amount is a three-bracket schedule by assessed income, in GBP per household.
• bill_share.in_effect with bill_share.rate switches to a bill-share discount: a percentage off the household's annual gas and electricity spend. RF proposes a cut in unit prices (pence per kWh); the microdata hold annual spend, not kWh, so this also discounts standing charges and gives low-consumption households relatively more than a per-kWh cut would. The rates default to zero; each run's receipt records the rates used.
• passport.in_effect, passport.benefits and passport.assessed_income control passporting.
• income_test.in_effect and income_test.sources control the income test. income_test.sources is the list of person-level income variables; use [employment_income, self_employment_income] for an earnings test. income_test.household_equivalised switches to RF's comparator: total household income divided by the modified-OECD BHC equivalisation factor.
• gb_only leaves out Northern Ireland, whose bills sit outside the Ofgem cap.
• count_in_hbai_income makes the discount count as HBAI income, as DWP does for the Warm Home Discount. Set it to false to treat the discount as a price cut instead, as policyengine-uk does for the 2022 Energy Price Guarantee.

The discount always adds to household_benefits, and so to household_net_income, and to gov_spending.

Presets (presets.py) use RF's published averages for a £2bn scheme:

• rf_flat: £175 per household below £24,000, and for every passported household.
• rf_tiered: £220 below £18,000, £85 from £18,000 to £24,000. Passported households get £220.
• rf_tiered_own_income: as rf_tiered, but passported households are tiered on their own income, with £85 as the floor. RF does not say which tier passported households get.
• rf_household_income: RF's comparator, £175 where equivalised household income is below £30,000, or passported.
• passport_only: passporting alone at £175, a Warm-Home-Discount-style comparator.

calibrate.py turns any fixed-amount run into a bill-share schedule. Brackets paying the same amount share one rate, so the average payment matches across those brackets combined. It also rescales any schedule to a budget.

Usage

uv sync --extra dev
export HF_TOKEN=...   # read access to the pinned private datasets

Household calculation:

from policyengine_uk import Simulation
from uk_energy_reforms.reforms.targeted_energy_discount import preset, scenario

sim = Simulation(situation=..., scenario=scenario(preset("rf_tiered")))
sim.calculate("targeted_energy_discount", 2026)

Microsimulation with baseline and reform side by side:

from uk_energy_reforms.simulate import run

# r.frame: one row per household
r = run("microcosm_national", 2026, preset("rf_tiered"))

Full receipt (JSON plus markdown with breakdowns by region, household type and decile, poverty impacts, winners and losers, inequality, coverage of households in poverty or with high energy costs, eligibility across five income measures, counts around the income cut-offs and the baseline the scheme acts on):

uv run uk-energy-reforms run --datasets microcosm_national efrs_1573 --year 2026 \
  --presets rf_flat rf_tiered --bill-share --budget 2e9 --out analyses/<name>/results

Our estimates beside the figures RF publishes (2024-25 like-for-like, and 2026-27), and the pre-computed data files for the dashboard (added in a follow-up PR): the results for every results-<year> folder in the analysis, and the household-calculator cases the dashboard's JavaScript calculator is tested against:

uv run uk-energy-reforms rf-compare --out analyses/<name>
uv run uk-energy-reforms export-dashboard --analysis analyses/<name> \
  --out dashboard/public/data/targeted_energy_discount_results.json \
  --calculator-out dashboard/public/data/calculator.json

--take-up sets take-up among households eligible only through the income test (they must self-declare). Passported households are enrolled automatically. Draws are seeded and are set as an input, because policyengine-core forbids randomness inside formulas.

Datasets

datasets.py pins each dataset by Hugging Face revision and sha256:

• microcosm_national: the Microcosm UK 2024-25 national release published on 4 October 2026 (microcosm-uk-2024-25-national, cut microcosm-uk-2024-25-national-20261002T230158Z-5c6b3f68). It replaced the staged national attempt uk-frs-calibration-attempt-20260923T134002Z-c1be1c9f that the first results used.
• efrs_1573: Enhanced FRS 2024-25 from policyengine-uk-data 1.57.3.

Use Microcosm for anything below national totals. eFRS weights are highly concentrated (Kish effective sample size about 1,100).

policyengine-uk does not uprate electricity_consumption or gas_consumption between years. policyengine-uk#1860 added CPI uprating; policyengine-uk#1868 removed it again, because policyengine-uk-data prices the Enhanced FRS at Ofgem April–June 2026 unit rates. Spend therefore stays at each dataset's stored price level in every year: April–June 2026 for the Enhanced FRS and 2024-25 for Microcosm. Ofgem's typical-use cap averaged within 2% of its April–June 2026 level over 2024-25, so the two datasets sit at about the same price level. Both sit below this winter's cap (October–December 2026 is 17–18% above April–June 2026), so bill levels, bill-share payments and energy-burden shares are understated in the scheme years on both. Fixed amounts and eligibility do not depend on spend.

Notes

• Situations cannot name reform-only input variables such as claims_targeted_energy_discount, because policyengine-uk rebuilds a baseline from the same situation. Set them with sim.set_input(...) instead.
• Neither dataset records prepayment meters, off-grid heating fuel, EPC bands or the HBAI heating-affordability item. Hardship is proxied by poverty, the four lowest income deciles and energy spend above 10% of net income.

Layout

src/uk_energy_reforms/
  reforms/targeted_energy_discount/   parameters.yaml, variables.py, reform.py, presets.py
  datasets.py        pinned microdata
  simulate.py        baseline + reform household frames
  calibrate.py       bill shares and budget scaling
  household_types.py RF-aligned household types
  analysis.py        impacts, winners and losers, inequality, coverage, cliffs,
                     breakdowns, baseline, eligibility across income measures
  report.py, cli.py  results receipt
  rf_comparison.py   RF's published figures beside ours
  sources.py         official statistics and other organisations' figures
                     (external_sources.json) shown beside the baseline
  calculator.py      household cases the dashboard calculator is tested against
  dashboard_data.py  compact results for the dashboard
analyses/            one folder per analysis (results and write-ups)
tests/               household-level reform tests and helper unit tests