Plan: model benefits-in-kind (BiK) — scope, implementation, and data needs

[vahid-ahmadi]

Tracking issue for adding benefits-in-kind (BiK) modelling to policyengine-uk.

Current state

There is no BiK modelling in the repo today. Searches for benefit_in_kind,
bik, p11d, company_car, etc. return no hits in parameters/,
variables/, or tests/.

The integration hook does exist:

• policyengine_uk/variables/gov/hmrc/income_tax/bases/taxable_employment_income.py
  already adds an employment_benefits term to taxable earnings.
• policyengine_uk/variables/household/benefits/employment_benefits.py
  currently only sums statutory_sick_pay and statutory_maternity_pay —
  i.e. statutory cash payments, not P11D-style benefits-in-kind.

So the slot exists, but nothing fills it.

Scope (priority order)

UK BiK has a long tail. Roughly 80% of the aggregate taxable BiK base is
captured by four categories — proposing we MVP these and defer the rest:

Phase 1 — MVP

1. Company cars (P11D section F) — biggest single category by value
2. Car fuel benefit (F)
3. Private medical and dental insurance (I)
4. Living accommodation (D)
5. Beneficial loans (H) — interest below HMRC official rate

Phase 2 — moderate prevalence

• Vans and van fuel (G)
• Vouchers and credit cards (C)
• Mileage allowances above HMRC approved rates (E)
• Relocation expenses above £8k (J)
• Assets transferred / placed at disposal (A, L)

Phase 3 — long tail / niche

• Services supplied (K), payments on employee's behalf (B), entertainment,
  non-qualifying travel and subsistence, employer-paid income tax,
  educational assistance, etc.

Exempt categories (no charge — out of scope): trivial benefits <£50, one
mobile phone, workplace parking/nurseries, employer pension contributions,
cycle-to-work, eye tests for VDU users, pensions advice up to £500.

Implementation plan

Variables

• New benefit_in_kind variable on Person, summing typed sub-components.
• One sub-variable per BiK category in variables/gov/hmrc/income_tax/bik/
  (e.g. company_car_benefit, car_fuel_benefit, private_medical_benefit,
  living_accommodation_benefit, beneficial_loan_benefit).
• Adds benefit_in_kind to employment_benefits so it flows into
  taxable_employment_income via the existing hook.

Parameters

• New tree under parameters/gov/hmrc/income_tax/bik/:
  • company_car/appropriate_percentage_by_co2.yaml (year-by-year CO2 bands)
  • company_car/fuel_charge_multiplier.yaml
  • company_car/electric_vehicle_rates.yaml
  • car_fuel/multiplier.yaml
  • living_accommodation/cost_threshold.yaml (currently £75k)
  • beneficial_loan/official_rate.yaml
  • beneficial_loan/de_minimis.yaml (currently £10k)
  • vouchers/... (Phase 2)

Knock-on effects (must not be missed)

BiK feeds into adjusted net income, which means it touches:

• Personal allowance taper at £100k
• High Income Child Benefit Charge
• Tapered annual allowance for pensions
• Class 1A NICs (employer side) — not currently modelled in repo;
  filing this here as a flag for separate work.

Tests

• Unit tests per BiK category against worked HMRC examples.
• Integration test asserting BiK flows through to taxable employment income,
  ANI, and HICBC where applicable.

Data needs and FRS coverage

This is the harder half of the project. Summary of where each Phase 1
component sits w.r.t. our current data:

Component	FRS has flag?	FRS has value?	Calibration source
Company car	Yes	No (no list price / CO2 / fuel)	HMRC Benefits in Kind statistics, Company Car Tax statistics
Car fuel	Partially (fuel-provided flag)	No	HMRC BiK statistics
Private medical insurance	Yes (employer-paid flag)	No	HMRC BiK statistics
Living accommodation	Yes (job-tied accommodation flag)	No	HMRC BiK statistics; SPI (restricted)
Beneficial loans	No reliable flag	No	HMRC BiK statistics; SPI (restricted)

Public sources (no special access)

• HMRC "Benefits in Kind" statistics — annual P11D-based release;
  recipient counts and total taxable values by category and income band.
  Primary calibration target.
• HMRC "Company Car Tax" statistics — CO2 and fuel-type distributions.
• HMRC Personal Incomes Statistics (SPI tables) — published totals.
• DVLA / SMMT — new-car registrations for company-car list-price/CO2
  distributions.

Restricted-access (HMRC Datalab)

• SPI microdata — actual P11D values per individual; gold standard
  but requires accreditation and stays in the secure environment.
• RTI / PAYE — payrolled BiK (growing share since 2016 mandatory
  payrolling reforms).

Proposed imputation approach for the FRS-based pipeline

1. Use FRS flags to decide who receives each BiK type.
2. For each type, draw a value from a distribution calibrated so the
   population total and income-band breakdown match HMRC's published BiK
   statistics. (This mirrors the second-stage imputation pattern flagged
   in Align UK model/data pipeline with US patterns: reported-anchoring, second-stage imputation, and downstream bugs #1621.)
3. For company cars in particular: either (a) skip the list-price/CO2
   mechanics and assign a calibrated taxable value directly, or (b) layer
   a simplified CO2-band model over an imputed list-price distribution.
   Recommend (a) for MVP and (b) as a Phase-1.5 refinement.

Effort estimate

• Framework wiring + a flat-value benefit_in_kind variable feeding the
  existing hook: ~1–2 days.
• Phase 1 MVP (five categories, calibrated imputation, knock-on effects
  through ANI/HICBC, tests): ~2 weeks.
• Trustworthy distributional version (CO2 bands modelled, beneficial-loan
  amounts, sensitivity-checked imputation): closer to a month, with most
  time on the data side rather than the rules.
• Class 1A NICs (employer side): separate workstream.

Open questions

• Do we want to model Class 1A NICs in the same effort, or split it?
• Should BiK enter Universal Credit / legacy means tests as earned income?
  (HMRC treats most BiK as taxable but UC treats earned income via RTI;
  payrolled BiK flows through, non-payrolled BiK does not — needs a
  decision.)
• Pursue HMRC Datalab access for SPI/RTI microdata, or stick to
  public-aggregate calibration for now?

Note

Filing this as a planning issue for future discussion with NEF before we
commit to a build order. Happy to split into per-category issues once
the scope and imputation approach are agreed.