Property Tax Lab

AVMs for a Proportional Property Tax

A proportional property tax requires accurate, up-to-date valuations of every liable property. Automated Valuation Models make this feasible. Here we examine how AVMs work in tax systems worldwide, and what it would take to deploy one in England.

What is an AVM?

An AVM estimates property values using historical transactions, property attributes, and location data. In taxation, AVMs operate within mass appraisal systems (IAAO, 2017), applying consistent methods across entire jurisdictions with ongoing statistical validation. For a proportional tax, the model must produce a point estimate of each property's capital value, not just assign it to a band.

The modelling itself is well understood. Hedonic regression and comparable-sales methods remain the backbone because their outputs are transparent and legally defensible; machine-learning methods, gradient boosting especially, cut valuation error but are harder to explain at appeal, so most jurisdictions layer them on an interpretable baseline. Location is the primary driver of value throughout. Data quality matters more than any of this: transaction prices calibrate the model, attribute data (size, type, condition) explains differences between properties, and reliability falls wherever sales are thin; rural areas, unusual property types, and the top of the market.

No jurisdiction relies on full automation. Every mature system is hybrid: the model produces baseline values, and professional valuers handle outliers, unique properties, sales validation, and appeals. Assessment quality is audited through ratio studies (IAAO, 2013), which test uniformity and vertical equity, whether high- and low-value properties are assessed consistently, rather than standard ML metrics.

Valuation frequency is a design choice in its own right. Annual cycles, as in the Netherlands, keep values current but demand strong governance; longer cycles are cheaper to run and produce bill shocks at each reset. England's Council Tax valuations date from 1991; a proportional replacement would need to establish a regular cycle from the outset.

International Case Studies

Jurisdiction Frequency Model approach Governance
US (Cook County) Triennial ML (LightGBM) plus review High transparency, high appeals
US (Maricopa) Annual CAMA mass appraisal Structured appeals
Netherlands Annual Comparable-based plus models Strong central oversight
New Zealand 3-yearly Mass appraisal Audited by central regulator
Sweden 3-6 year cycles Zone-based models Rule-driven system
Estonia Multi-year Land-only mass valuation Centralised

The Netherlands offers perhaps the closest analogue to what a proportional property tax in England would require: annual capital-value assessments of the entire housing stock, centrally audited, with a mature appeals process.

Appeals

A proportional tax would generate significantly more appeals than the current band system, because individual valuations create more grounds for dispute. Appeals serve a dual function: correcting errors and conferring legitimacy. But high volumes carry real administrative cost and create budget uncertainty for local authorities.

International experience shows that systems separating factual corrections (wrong floor area, missing extension) from valuation disputes (disagreement over market value) operate more efficiently (Almy, 2014). Designing the appeals process is as important as designing the model.

Feasibility in England

Transaction volumes

England has good transaction price coverage via HM Land Registry (1.0–1.2 million annual sales in normal conditions), though turnover is thin in rural areas and at the top of the market. This data will bear serious weight: our own residential stock model, PRISM, is built from it, reweighting two years of sales to VOA stock counts to estimate England's homes by current value.

Transaction volumes would support modelling for mainstream properties. But an AVM also needs property attributes; size, type, condition, tenure; linked to each address. This is where England falls short. The feasibility question is less about sale prices and more about whether a reliable attribute dataset can be assembled.

Data availability

Strengths

  • Near-complete transaction price coverage via Land Registry
  • EPC database for floor area and energy attributes
  • Ordnance Survey geospatial data
  • Planning and land-use datasets

Weaknesses

  • Fragmented and inconsistent property attributes
  • Limited data on condition, quality, and internal layout
  • Leasehold complexity, especially in flats
  • No single integrated property database

The Netherlands maintains a unified property register with standardised attributes linked to every address; no equivalent public dataset exists in England. However, the VOA maintains its own property records (dwelling type, area, age, bedrooms, bathrooms, parking, plot size). These are not publicly available but constitute a working attribute dataset, as the Welsh AVM demonstrates.

The VOA's Welsh AVM: a working precedent

The VOA has built and deployed an AVM for the 2028 Council Tax revaluation in Wales, covering 1.5 million properties. The model uses a Gaussian Markov Random Field to estimate continuous spatial variation, and was externally assured by the IAAO as "more than satisfactory".

The approach is explicitly "model assisted": the AVM generates first-pass valuations, then professional valuers review properties near band boundaries and those flagged as unreliable. The VOA estimates this reduces costs by roughly one-third versus purely manual valuation.

This demonstrates the VOA holds workable attribute data and has built the institutional capability for mass appraisal. The question is whether it can scale.

Wales vs England: the scaling challenge

  • Scale: England has roughly 24 million domestic properties; 16 times the Welsh stock. Data quality issues that were manageable at 1.5 million may compound at this scale.
  • Data enhancement: Even for Wales, the VOA undertook a "significant data enhancement exercise" to update property records. The equivalent exercise for England would be substantially larger.
  • Market diversity: England's housing market is far more heterogeneous; from central London flats to rural estates; increasing model complexity and reducing the share of properties the AVM can handle confidently.
  • Bands vs point values: The Welsh AVM assigns properties to Council Tax bands, which are broad value ranges. A proportional tax requires a point estimate of value, demanding higher precision from the model.

The Welsh AVM confirms the VOA can build a credible system from existing data. But it does not settle whether the same approach can deliver the accuracy needed for a proportional tax across a property stock 16 times larger and considerably more diverse.

Design Choices for a Proportional Tax AVM

A proportional tax could take several forms, each with different AVM requirements:

1

Residential capital value tax

Closest to a direct Council Tax replacement. Each property valued at its estimated capital value; tax levied as a fixed proportion.

Feasibility
Moderate to high, depending on attribute data quality
Model type
Hybrid (hedonic plus boosting)
Challenge
Political sensitivity of redistribution
2

Annual revaluation system

Values updated every year rather than at long intervals, eliminating the growing inequity of stale valuations.

Feasibility
Moderate to high
Requirement
Institutional overhaul with central oversight
3

Tax on land values only

Simpler data requirements in principle, since only the land need be valued, not the buildings. But separating land from improvements is technically difficult.

The standard approach (e.g. Diewert & Huang, 2025) uses "land residual" models that subtract an estimated structure value from the sale price. These models typically assume structures depreciate steadily with age; a reasonable assumption for newer, standardised housing.

This is a poor fit for England, where period properties (Georgian, Victorian, Edwardian) make up a substantial share of the stock and often trade at a premium precisely because of their age and character. Land/structure decomposition is considerably harder here than in markets where the method was developed.

Feasibility
Low to moderate; land/structure decomposition is poorly suited to England's housing stock
Model type
Land residual models and spatial interpolation

Strategic Trade-offs

Trade-off Implication
Centralised vs local A single national model ensures consistency and enables quality audit; local valuers may better reflect micro-market conditions and carry more political legitimacy
Transparency vs strategic challenge Publishing model methodology builds public trust, but also equips appellants to mount more targeted challenges, increasing appeal volumes and sophistication
Accuracy vs explainability ML models reduce valuation error, but simpler models are far easier to defend at tribunal; the legal system rewards interpretability
Frequency vs stability Annual revaluations keep assessed values current and reduce horizontal inequity; but year-on-year bill changes create political friction and may require transitional relief

Conclusions

1.

Institutional design matters more than model choice. Data quality, governance, and appeals systems are more important than algorithm selection.

2.

Hybrid systems are the proven approach. Full automation is neither necessary nor desirable for a tax that must command public legitimacy.

3.

The VOA's Welsh AVM is a working precedent, not a finished solution. It proves the institutional capability exists, but scaling to England's 24 million properties; with far greater market diversity and the higher precision a proportional tax demands; is a different order of challenge.

4.

The algorithm is not the hard part. Model design is well understood. The binding constraints are data infrastructure; building a unified property dataset with reliable attributes; and institutional reform: a central valuation body, regular revaluation cycles, and a well-designed appeals process.

Practical Steps

  • Build a unified national property dataset combining attributes, transactions, and geospatial data
  • Establish a central valuation authority for standards and audit
  • Begin with residential capital values, not full automation
  • Implement regular ratio studies and public reporting from the outset
  • Design appeals that separate data corrections from valuation disputes
  • Use pilots in data-sparse segments (rural, high-value) to test model adequacy before full rollout

References

  1. IAAO (2017). Standard on Mass Appraisal of Real Property. International Association of Assessing Officers. Defines the framework for mass appraisal systems used in property taxation worldwide.
  2. IAAO (2013). Standard on Ratio Studies. International Association of Assessing Officers. Establishes COD, PRD, and PRB as the standard metrics for evaluating assessment quality.
  3. Almy, R. (2014). Valuation and Assessment of Immovable Property. OECD Working Papers on Fiscal Federalism, No. 19. Comparative survey of property tax valuation and appeals practices across OECD jurisdictions.
  4. Mirrlees, J. et al. (2011). Tax by Design: The Mirrlees Review, Chapter 16. Institute for Fiscal Studies. Makes the economic case for replacing transaction taxes and banded levies with a proportional property value tax.
  5. Cook County Assessor's Office (2024). Residential Automated Valuation Model. Open-source LightGBM model used for mass appraisal in Cook County, Illinois.
  6. VOA (2024). Algorithmic Transparency Record: Automated Valuation Model. GOV.UK. Documents the VOA's AVM for the 2028 Welsh Council Tax revaluation, including model design, IAAO assurance, and the "model assisted" workflow.
  7. Waarderingskamer (2024). The Dutch System of Property Valuation (WOZ). The Netherlands' central oversight body for municipal property valuations; annual cycle with structured objection and appeal processes.
  8. Diewert, W.E. & Huang, N. (2025). Decomposing Residential Resale House Prices into Structure and Land Components. Discussion Paper 24-03 (revised), Vancouver School of Economics, University of British Columbia. Develops a hedonic regression approach to land/structure decomposition using geometric depreciation of structures.