Story perspectives

Researchers Trim 19k Outliers, Yield 12M England Repeat-Sale Pairs

6/22/2026

Summary

Researchers Trim Outliers

Researchers removed 19,438 extreme outliers from England's price-paid property dataset, leaving 27,758,154 transactions.
They set price thresholds at £3,718,654.30 and £5,557.30 using a log transformation and four-standard-deviation limits.
They excluded rows outside those thresholds, kept postcodes as of January 2023, and dropped entries missing postcode, PAON, SAON, street name, sales of £100, or classified as “Other.”
They created unique identifiers by concatenating address fields and produced 12,009,213 repeat-sale pairings for analysis.

Summary

Analysts Update LADs

Analysts updated each postcode-district’s Local Authority District (LAD) to the 2023 LAD dataset, addressing marked changes from historical English local-government reorganisations.
Postcode-district geometry was sourced from the Ordnance Survey Code-Point with Polygons dataset, the most detailed GB spatial layer.
Address coordinates came from AddressBase, which provides National Grid references for every delivery point.
Local Authority District boundaries were drawn from the Office for National Statistics.

Summary

Researchers Track Floods

The repeat-sales hedonic model gives each English postcode a 0-to-1 Flood Zone 3 score to isolate flood effects on house prices.
NFHU flags sale pairs that bracket a flood in the same local-authority district, using a flood record of 78,025 polygons covering 2,723 km² to form treatment and control groups.
After removing pre-1995 events and duplicates, the age-period-cohort framework and Kaplan-Meier analysis measure how flood exposure alters home-ownership length.

1 of 3

Researchers Trim Outliers

Researchers removed 19,438 extreme outliers from England's price-paid property dataset, leaving 27,758,154 transactions.
They set price thresholds at £3,718,654.30 and £5,557.30 using a log transformation and four-standard-deviation limits.
They excluded rows outside those thresholds, kept postcodes as of January 2023, and dropped entries missing postcode, PAON, SAON, street name, sales of £100, or classified as “Other.”
They created unique identifiers by concatenating address fields and produced 12,009,213 repeat-sale pairings for analysis.

1 / 3