Add LA-level household land value targets and calibrate on them

[vahid-ahmadi]

Summary

• Add 360 LA-level ons/household_land_value/{code} calibration targets
• Generalise mhclg_regional_land.py methodology to local-authority granularity
• Wire the new targets into the LA calibration loss matrix so the LA reweighter trains on them
• Ship canonical input CSV, target source module, loss-matrix wiring, and 27 unit tests

Closes #370.

What this PR does

Extends the regional methodology to LA level

Each LA's share of national household land value is proportional to its total property wealth (households × avg_house_price), scaled so the LA totals match the ONS national household-land series. Exactly the same formula as mhclg_regional_land.py::_compute_regional_shares, one geography deeper.

share_la  = (households_la × avg_house_price_la) / Σ_LAs
target_la = share_la × HOUSEHOLD_LAND_VALUES[year]

Wires the targets into LA calibration

datasets/local_areas/local_authorities/loss.py now adds an ons/household_land_value column to the LA target matrix:

• matrix entry: per-household household_land_value from policyengine-uk.
• y entry: 360-vector of per-LA targets at the calibration year, taken from la_land._compute_la_targets and reordered to match local_authorities_2021.csv so the country mask and target indices agree at every position.

The year is selected from time_period; if it is outside HOUSEHOLD_LAND_VALUES (defined for 2021–2026) the latest known year is used as a fallback.

Files

New

• policyengine_uk_data/storage/la_land_values.csv — 360 rows: code, name, households, avg_house_price.
  • households from the existing local_authority_weights.h5 (sum of each LA's 2025 weight row) — keeps household-count semantics aligned with the rest of the LA calibration.
  • avg_house_price from HM Land Registry UK HPI (Dec 2025). Primary match on ONS code, name-based fallback for LAs with re-allocated codes (e.g. Sheffield E08000019 → E08000039 in HPI), NI country-level HPI fallback for missing NI LGD months, national-avg fallback for Isles of Scilly.
• policyengine_uk_data/targets/sources/la_land.py — _compute_la_shares(), _compute_la_targets(), get_targets() returning 360 Target objects with geographic_level=LOCAL_AUTHORITY.
• policyengine_uk_data/tests/test_la_land_value_targets.py — 18 unit tests on the targets themselves.
• policyengine_uk_data/tests/test_la_loss_land_value.py — 9 unit tests on the loss-matrix wiring.
• changelog.d/370.md.

Modified

• policyengine_uk_data/datasets/local_areas/local_authorities/loss.py — adds the ons/household_land_value column (matrix + y) so the LA reweighter trains on the new targets.
• policyengine_uk_data/tests/test_regional_land_value_targets.py — test_target_registry_includes_regional now filters by GeographicLevel.REGION (the regional and LA targets share the ons/household_land_value/ name prefix, so filtering by prefix alone now pulls both).

Tests

Targets — test_la_land_value_targets.py (18 tests)

CSV data quality

• Row count matches local_authorities_2021.csv (360)
• Columns match schema exactly
• No missing values; covers E/W/S/NI
• House prices in [£50k, £2m], households positive
• Households bounded [500, 500_000] (regression test for the Isles of Scilly fallback leak)

Share / target aggregation

• Shares sum to 1; all positive
• LA targets sum to the national ONS household-land series within 1e-6 for every year
• Kensington and Chelsea avg household land > 3× Blackpool
• London LAs dominate the top quintile of LAs by avg household land
• London total land dwarfs North East total by >3×

Registry integration

• get_targets() returns exactly 360
• Target names follow ons/household_land_value/{code}; geo_code == code
• All carry GeographicLevel.LOCAL_AUTHORITY
• Every target has values for every year in HOUSEHOLD_LAND_VALUES
• get_all_targets(year=2024, geographic_level=LOCAL_AUTHORITY) returns 360 LA land targets

Loss-matrix wiring — test_la_loss_land_value.py (9 tests)

Light layer (no Microsimulation)

• Every LA code in local_authorities_2021.csv has an LA land target.
• Reindexing by la_codes order yields a finite, all-positive 360-vector.
• LA target sum equals HOUSEHOLD_LAND_VALUES[year] for 2024 / 2025 / 2026 within 1e-6.

Full-build layer (gated on the enhanced FRS fixture)

• ons/household_land_value column present in both matrix and y.
• y vector length 360.
• y sum equals ONS national household-land for the calibration year (within 1e-6).
• y ordering matches la_codes (np.testing.assert_array_equal).
• All y entries positive.
• Matrix column equals sim.calculate("household_land_value").values.

Results of running the new tests plus adjacent suites (regional land, land targets, target DB, target registry, release manifest): 67 passed, no regressions.

Sanity check — top 10 LAs by avg household land value (2024)

LA	Avg household land	Avg house price (UK HPI)
Kensington and Chelsea	£622k	£1,178k
Westminster	£465k	£880k
Camden	£414k	£784k
Richmond upon Thames	£410k	£777k
Elmbridge	£392k	£743k
City of London	£391k	£740k
Hammersmith and Fulham	£377k	£714k
Islington	£369k	£700k
Wandsworth	£364k	£689k
Haringey	£331k	£627k

Bottom 10 are all post-industrial / deprived areas (Inverclyde, East Ayrshire, West Dunbartonshire, Hull, Burnley, Hartlepool, Aberdeen, North Ayrshire, Hyndburn, Blackpool — all at £60–72k).

Sources

• HM Land Registry UK House Price Index — Dec 2025
• ONS National Balance Sheet 2025

Related

• Regional property wealth calibration: WAS-imputed main_residence_value doesn't match UK HPI by region #314 — regional intensity ratios (✅ merged)
• Add regional household land value calibration targets #315 — regional household land value targets (the pattern this PR generalises)

[vahid-ahmadi]

Note for whoever picks up #357: this PR mirrors mhclg_regional_land.py's property-wealth-share methodology at LA level, so it inherits the same flat-land-share assumption. When the regional shares are updated to multiply by property_wealth_intensity[region], the LA CSV here (storage/la_land_values.csv) needs regenerating in the same PR — same multiplier, with each LA using its region's intensity — so the LA and regional targets stay numerically consistent with each other and with the region-aware formula in policyengine-uk.

[MaxGhenis]

Blocker: data bug in la_land_values.csv — Isles of Scilly (E06000053) has 2,492,115 households. The real figure is ~1,000 (pop ~2,000, 1,115 households per ONS mid-2023 estimate).

Impact: IoS alone absorbs 8.6 % of the national household share (2.49M / 29.0M), which the methodology then multiplies by the national household-land value — £65 bn of UK household land 'lives' in Scilly under this target, depressing every other LA's share by that amount. London LAs take the biggest hit because their share-of-average-price is highest.

Quick verification:

$ awk -F',' 'NR>1 {sum+=$3} END {print sum/1e6}' la_land_values.csv
31.5   # with IoS bug
$ awk -F',' 'NR>1 && $2 != "Isles of Scilly" {sum+=$3} END {print sum/1e6}' la_land_values.csv
29.0   # without — matches ONS ~29.4M

Looks like a UK-HPI 'national-total-as-fallback' path leaked into one LA row. Likely two lines to fix:

1. Correct la_land_values.csv row to E06000053,Isles of Scilly,1115,308582 (or whatever the canonical source gives).
2. Add a test that bounds households per LA, e.g. assert (df.households.between(500, 500_000)).all(). None of the current 18 tests catch a 1000x outlier.

Happy to approve once that's in. The methodology itself is sound — mirrors mhclg_regional_land.py::_compute_regional_shares correctly and the target shape aligns with the regional targets.

[vahid-ahmadi]

@MaxGhenis thanks — fixed in 3ed729c.

Data fix

• Patched E06000053,Isles of Scilly from households=2,492,115 to 1,115 (ONS mid-2023 estimate). avg_house_price=308,582 kept as-is.
• Post-fix UK household total across the CSV drops to 29.73M, in line with the ~29.4M ONS figure you quoted (was 31.5M pre-fix, matching your awk).

Quantified impact of the fix

• IoS share of national property-wealth share: 7.85% → 0.0038%.
• K&C 2024 target: £42.6bn → £46.2bn (+8.5%). Every non-IoS LA gets the same ~8.5% uplift; London LAs were the most suppressed.

Tests added

• test_households_within_plausible_range — bounds every LA to [500, 500_000] per your suggestion. A future fallback leak of this class fails immediately.
• test_isles_of_scilly_households_are_thousands_not_millions — explicit [500, 5,000] bound on E06000053 so the specific row that leaked has a named regression test.

Full suite: 20/20 pass locally via uv run pytest policyengine_uk_data/tests/test_la_land_value_targets.py.

Generation-path note: the 2,492,115 figure matches the South West regional household total, so the fallback that fired during CSV generation was a regional sum, not "national-avg" as the PR body suggested. I'll correct the PR description; worth flagging for whoever regenerates the CSV next.