where the money comes from

We modeled the effects of English.Villa tax“by analyzing land registry data on every property trans­action since 1995. This allows us to estimate how much each postcode and constituency will pay. It is an approx­imate estimate with a lower limit — see details of the method­ology below. Because property taxes are devolved in nature, there is currently no property tax in Wales and Scotland — although I tend to expect this to be the case soon.”

This inter­active map shows the results of our model and highlights each zip code that contains “mansion tax” properties. It also shows English and Welsh data on current council tax rates, median house prices and changes in house prices since 1995 (showing how outdated council tax is):

Here you can view the map in full screen mode. It’s important to emphasize that only zip codes are shown here — the markers on the map are in the center of the zip code and do not represent individual properties.

This graph shows our estimate of villa tax revenue for each constituency. It will come as no surprise that most of the properties affected are in London (you can hover over individual constituencies to see full details).

In my opinion the tax is good policy. As with all taxes, there will be ineffi­ciencies and injus­tices, but the basic concept is right: to end the anomaly that someone in a £10 million house pays the same council tax as someone in a £1 million house — and just twice the council tax as someone in a £400,000 house. I wrote more about my views here.

This graphic shows how the “manor house tax” affects council tax something more progressive:

Methodology and limitations

Our method­ology was simple: use the Paid database of the land registry office to find every real estate trans­action since 1995 that looks like residential real estate and increase prices the change in average house prices in each constituency (while also displayed Council tax data).

The code for the analysis and web app is available on our GitHub.

Techni­cally, it would be easy for our analysis and map to show the assessed value and property tax for individual properties, but we were unhappy with the privacy impli­ca­tions (although there are many property price websites where you can see “price paid” data for specific properties). Therefore, we limit the map to zip codes (which has the side effect of making the app load and respond much faster).

Our very simple approach has obvious limita­tions:

• The open land register data does not contain title numbers or other identi­fiers for properties. Therefore, we need to “deduplicate” repeated trans­ac­tions in the same property so that we only count the most recent ones. This is error prone and we err on the side of conser­vatism — so we will be missing some features.
• The open land register data also does not differ­en­tiate between residential and commercial properties. We use the “Property Type” field, which indicates whether a property is a detached property, a semi-detached house, a terraced apartment, an apartment/maisonette or “Other”. In principle, “Other” should be commercial property and the other types should be residential property – but there will be numerous cases where this is not the case. For example, a farmhouse sold together with the farm may be classified as “detached” and recorded in our data as if the full price was for the house, although realis­ti­cally this is not the case.
• Inflation is higher in some areas of a constituency than in others
• We cannot take into account improve­ments etc. to properties, conver­sions (e.g. dividing a property into apart­ments) and other changes after a sale.
• Our approach completely ignores properties that have not been sold since 1995.
• In some cases (especially high-value properties) the price is hidden or too low.
• Another problem is portfolio trans­ac­tions – for example, where several low-value properties are purchased for a high price of more than £2 million, but no separate price is regis­tered for each property. In this case, the land registry office sees every Properties listed for sale worth more than £2 million and we end up with several misiden­ti­fi­ca­tions of “villas”. We are trying to address this issue by detecting when multiple purchases are made in the same zip code, on the same day, and at the same price. However, this does not cover all such cases (e.g. when a portfolio trans­action spans multiple zip codes). There is no easy way to fix this.

Taken together, our approach likely results in a lower estimate of the actual static revenue from the tax. Total estimated sales are £510 million.

That’s a lot less than that OBR’s static revenue estimate is £600m – This is because they have used more sophis­ti­cated approaches, for example more detailed correc­tions for house price inflation, better detection of home ownership and the inclusion of properties that are not included in the trans­action data. The OBR then adjusts the static estimate to reflect behav­ioral effects (clustering below thresholds) and losses on other taxes — bringing their total estimated revenue to £400 million.

So our numbers and our map are missing properties and under­valuing the properties, and that adds up to an error of about 20%. We make no attempt to adjust for behav­ioral effects. Therefore, none of the figures we present will be individ­ually accurate, but the overall picture should be an accurate reflection of the constituencies and postcodes from which the ‘mansion tax’ revenue will come.

Contains HM Land Register data © Crown Copyright and Database Law 2025. This data is licensed under the Open Government License v3.0.