Technical Summary

1. Model Overview

The Gadsden Valuations AVM produces point estimates of residential property market value for properties in England and Wales, consistent with the RICS definition of Market Value (IVS 104).

For each property, the model produces:

• A point estimate of market value (£) with upper and lower bounds
• A confidence tier (1–3, or declined) with Forecast Standard Deviation mapped to the EAA common 0–7 scale
• Feature contribution breakdown showing which inputs influenced the valuation
• Comparable transactions supporting the estimate
• Downloadable PDF report with full audit trail

1.1 Intended Use Cases

The AVM is designed as a specialist tool under IVS 105.20, providing model outputs for review by a RICS Registered Valuer who applies professional judgement before issuing a Red Book compliant valuation. It is not intended as a standalone replacement for physical valuation where required by lender policy or regulation.

• Portfolio revaluation for Basel 3.1 compliance (expected January 2027)
• Remortgage AVM-assisted valuations where physical inspection is not proportionate
• Independent cross-check against physical valuations or competing AVM outputs
• Desktop review support for Registered Valuers (VPS 1, restricted service)
• Market monitoring for the 10% revaluation threshold

1.2 Coverage

England and Wales. Four standard residential property types: detached, semi-detached, terraced, and flats/maisonettes. The model does not cover Scotland, Northern Ireland, commercial property, new-build prior to first sale, or non-standard constructions.

---

2. Methodology

The model is a gradient-boosted decision tree ensemble (LightGBM) trained on 8.4 million arm’s-length residential transactions from HM Land Registry Price Paid, enriched with data from multiple government open data sources.

2.1 Feature Categories

The model uses 47 features across six categories. The specific feature list and importance rankings are documented in GV-AVM-DOC-001 (available under NDA).

2.2 Validation Approach

All accuracy metrics use strict walk-forward testing: the model is trained on transactions up to a cutoff date and tested exclusively on future transactions it has never seen. The current test set comprises 295,026 properties completing after July 2025. This prevents data leakage and ensures published figures reflect real-world predictive performance.

2.3 Benchmark Methodology

All accuracy metrics are measured against Land Registry completion prices — the actual amount paid in arm’s-length transactions. This is a materially harder test than benchmarking against surveyor opinion (which is itself an estimate with a ±5–10% margin against eventual sale prices).

The industry leader, Hometrack, reports ~80% PE10 benchmarked against surveyor valuations. A like-for-like comparison against actual sale prices would produce lower figures. Published PE10 figures from providers using the surveyor benchmark are not directly comparable with our figures.

---

3. Accuracy Metrics

3.1 By Property Type

3.2 By Price Band

---

4. Confidence Model

Each valuation receives a confidence tier based on data availability and comparability factors. This satisfies the IVS 105 requirement for accuracy measures and the EAA ESSVM requirement for a per-property confidence score translatable to Forecast Standard Deviation.

Confidence can be upgraded by the client. The most impactful upgrade is provision of an EPC (£60–120), which supplies floor area, construction age, and energy characteristics — typically sufficient to move a property from Tier 2 to Tier 1. This creates a practical path to maximise high-confidence coverage without sacrificing accuracy.

---

5. Known Limitations

The model values properties based on recorded data. It cannot account for:

• •
  Internal condition — renovation status, kitchen/bathroom quality, structural issues. This is the primary limitation of any AVM and the principal reason IVS 105 requires professional judgement (PS 1.3).
• •
  Bespoke improvements — extensions or conversions not yet reflected in EPC or planning data
• •
  Legal encumbrances — restrictive covenants, rights of way, leasehold complications beyond Land Registry data
• •
  Micro-location — specific view, aspect, noise, or neighbourhood factors a local agent would consider
• •
  Market lag — Land Registry data lags completion by 2–4 months; ONS HPI temporal adjustment partially compensates

The confidence model flags properties where unobserved factors are more likely to affect accuracy.

---

6. Data Sources

All data sourced under Open Government Licence v3. No proprietary datasets.

---

7. Regulatory Alignment

---

8. Further Information

Full model documentation (GV-AVM-DOC-001) is available under NDA to organisations conducting formal vendor due diligence. This includes the complete feature list with importance rankings, model hyperparameters, SHAP analysis, bias assessment, and governance procedures.

A sample valuation report is available at gadsdenvaluations.com/sample-report.

Contact: [email protected] • gadsdenvaluations.com

This document is for informational purposes. AVM outputs require review by a RICS Registered Valuer before forming part of an IVS-compliant written valuation.