Make property_wealth_intensity region-aware (consumes policyengine-uk-data#358)

[vahid-ahmadi]

Summary

Companion issue for policyengine-uk-data#358 and policyengine-uk-data#357.

Today household_land_value uses a single national scalar for the land-to-property ratio:

# policyengine_uk/variables/household/wealth/household_land_value.py
property_wealth_intensity = (
    wealth.land.value.aggregate_household_land_value
    / wealth.property_wealth
)
return property_wealth * property_wealth_intensity + owned_land

That means regional variation in household land value comes only from regional variation in property_wealth, and the formula cannot express that London dwellings are ~80% land value while rural dwellings are ~45%. Regional calibration targets cannot fix that on their own — they can re-weight households, but every household still computes the same land/property ratio, so the distribution is pinned.

What's needed in policyengine-uk

1. Add a region-keyed parameter — e.g. gov.wealth.land.regional_land_to_property_ratio[region] — with one entry per English region (and GB-wide fallbacks for Wales/Scotland as appropriate).

2. Change household_land_value to look up the household's region and use the matching regional ratio rather than the national scalar:

   ratios = parameters(period).gov.wealth.land.regional_land_to_property_ratio
   region = household(\"region\", period)
   property_wealth_intensity = ratios[region]
   return property_wealth * property_wealth_intensity + owned_land

3. Keep the national scalar as the default / fallback for backward compatibility during rollout.

Coordination

• policyengine-uk-data#358 (merged or open) adds the data-side plumbing — _compute_regional_shares now accepts a land_to_property_ratio mapping — but does not ship real ratios yet. The same parameter values should feed both sides once sourced.
• The two changes must land together for the bias fix to take effect end-to-end. Until this issue is done, supplying non-uniform ratios on the data side will make the calibration fight itself, because the formula still forces every household onto the national ratio.
• Real per-region numbers should be sourced (VOA dwelling value minus ONS reconstruction cost, or Savills residential land-value estimates) in a separate PR with modelling-team sign-off, not as part of this one.

Why this matters

Surfaced while modelling a council-tax → flat-rate LVT swap. Without regional land-share variation:

• London is understated as an LVT base.
• South East / East of England / rural areas with high property wealth but lower land share are overstated.
• Regional winners/losers from a council-tax → LVT swap are distorted in both directions.

Out of scope for this issue

• Sourcing the actual regional ratios.
• Behavioural responses to an LVT.
• Wales/Scotland/NI-specific land-value work beyond whatever fallback is appropriate.

🤖 Generated with Claude Code

[vahid-ahmadi]

This is already in, merged as #1537 (commit e2dff2ac, "Replace flat land intensity ratio with region-specific ratios"). The code snippet quoted in the issue body doesn't match the current file — it looks like it was drafted against an older view.

Current state on main

policyengine_uk/variables/household/wealth/household_land_value.py already does a region-keyed lookup:

def formula(household, period, parameters):
    region = household("region", period)
    Region = region.possible_values
    property_wealth = household("property_wealth", period)
    owned_land = household("owned_land", period)

    regional = parameters(
        period
    ).household.wealth.land.intensity.property_wealth_by_region

    property_wealth_intensity = np.select(
        [region == Region.LONDON, ..., region == Region.NORTHERN_IRELAND],
        [regional.LONDON, ..., regional.NORTHERN_IRELAND],
        default=regional.UNKNOWN,
    )
    return property_wealth * property_wealth_intensity + owned_land

And policyengine_uk/parameters/household/wealth/land/intensity.yaml carries real per-region values sourced from MHCLG 2023 land-value estimates + UK HPI Dec 2025:

Region	Ratio
LONDON	0.85
SOUTH_EAST	0.65
EAST_OF_ENGLAND	0.60
SOUTH_WEST	0.58
WEST_MIDLANDS	0.52
EAST_MIDLANDS	0.48
NORTH_WEST	0.47
YORKSHIRE	0.46
WALES	0.47
SCOTLAND	0.44
NORTH_EAST	0.42
NORTHERN_IRELAND	0.673
UNKNOWN	0.673

This matches the "London ~80%, rural ~45%" distribution called out in the issue, and the MHCLG + UK HPI sourcing was done at modelling-team sign-off time.

Naming note

The parameter sits at household.wealth.land.intensity.property_wealth_by_region rather than the gov.wealth.land.regional_land_to_property_ratio path the issue proposed. gov.* is reserved for government policy parameters; land intensity is a market-characteristic calibration, so household.wealth.* is the right namespace.

Coordination

If policyengine-uk-data#358 expects the country-package-side parameter under a different path or with a different shape, the integration point needs updating on the data-side rather than shifting the model-side parameter that already has downstream consumers. Happy to look at the data-side coupling in a separate issue on the data repo if that's still outstanding.

Closing as completed. Please reopen (or file a narrower follow-up) if the behaviour diverges from what policyengine-uk-data#358 needs.