Publication Details
Publisher: "GLOBAL RESEARCH NETWORK" LLC (USA)
Issue: Vol 9, No 11 (2026)
ISSN: 2576-5973

Abstract

This study tests how accurately open British transaction data can value homes in two thin, lower priced housing markets, and whether machine learning improves on standard methods without bias against cheaper homes. Using 51,899 standard residential sales in Newcastle upon Tyne and Sunderland from HM Land Registry Price Paid Data for 2018 to 2024, it compares a comparable sales baseline, a fixed effects hedonic regression, a random forest, gradient boosting and an ensemble. Every feature is built only from earlier sales, and the models are tested on all 6,857 sales of 2024 with a quarterly rolling origin. The ensemble achieved a median absolute percentage error of 12.6 per cent, against 20.6 per cent for the hedonic regression, and valued 41.4 per cent of homes within 10 per cent of the price. Accuracy rested mainly on the price history of the postcode unit, the street and the dwelling’s own previous sale, and held up where local comparables were scarce. Grouped by sale price, all models overvalued the cheapest homes, but grouped by predicted value the ensemble’s ratios stayed between 0.98 and 1.01. Bank Rate added no accuracy. Without floor area or condition data, the models suit screening rather than lending decisions on their own.

Keywords
automated valuation models machine learning hedonic pricing vertical equity thin housing markets HM Land Registry Newcastle upon Tyne Sunderland ValuInsight AI