Thirty years of house prices, measured the hard way

Measured from 11,128,734 pairs of sales of the same property, house prices in England & Wales are now 5.2 times their 1995 level, a rise of 419% in cash terms between 1995 and 2026. That is not an average of what changed hands. It is an estimate of what happened to the same homes, sold twice, which is a different and harder question, and it is the number this site publishes when it needs to be sure a price movement is not a change in what was being sold.

What is a repeat-sales index?

A repeat-sales index measures price change using only properties that have sold more than once. The same address, two dates, two prices. Because the property is the same on both occasions, its size, its type, its street and its garden all cancel out, and what is left is the market. Any index built by averaging different homes has to assume the homes were comparable; this one does not have to assume it, because they are the same homes.

The cost is coverage. Only a minority of the register can be paired, and the properties that change hands twice are not a random sample of all housing. What you gain is a series that a shift in the mix of what sells cannot move, which is the single most common way a house price headline goes wrong.

How much have prices actually risen since 1995?

The series runs from 1995 to 2026, with 2015 set to 100. Every point below is an estimate with a 95% interval attached, because it comes from a regression rather than from counting. The intervals are narrow, which is what 11,128,734 observations buys.

Two features are worth naming. The 2008 to 2009 fall is visible and it is smaller than the fall in transaction volumes beside it, which is the usual shape of a housing downturn: the market thins before it cheapens. And the years since 2022 are close to flat in cash terms, which in a period of high inflation means a substantial fall in real terms. This index is nominal throughout; the difference between cash and real terms is its own article.

Which regions have risen most?

This is where a repeat-sales index earns its keep, because the answer depends entirely on when you start counting, and the two answers point in opposite directions.

Since 2015 the strongest region is Wales at 1.61 times its base, and the weakest is London at 1.22. Over the whole span the order is close to reversed: London has risen more than anywhere else, by 6.87 times. London holds both records at once, which is the whole point of publishing both columns.

The reversal is not a quirk of the estimator. It is the well-documented pattern of a market that ran ahead early and then stalled, against markets that started lower and have been catching up. What the index adds is that both halves are measured on the same properties rather than on whatever happened to sell, so neither number can be explained away as a change in the mix.

What can this index not see?

One weakness matters more than all the others, and it is stated on the dataset itself:

RENOVATION. A repeat-sales index cannot tell a rising market from an extended kitchen: both are the same address selling for more. The index therefore carries an UPWARD bias of unknown size, concentrated in the stock most likely to be improved. Price Paid Data has no attributes and the EPC floor area is one current snapshot per address, so we cannot correct it.

In plain terms: if a house sold for £200,000 in 2005 and £400,000 in 2025, this index reads a doubling of the market. It cannot tell that the second sale followed a loft conversion. Every repeat-sales index in the world has this problem, including the best-known ones, and the honest response is to say so rather than to imply a precision the method does not have. The intervals in the table describe sampling uncertainty only. They do not describe this.

There is a second point worth knowing. The renovation bias runs in the same direction as the floor-area problem in this site’s price per square metre series, where an EPC records one measurement per address and a home that was extended after its certificate is measured at its old size. Two independent series with a bias in the same direction agreeing with each other is weaker evidence than it looks.

How is the index built?

The estimator is Bailey, Muth and Nourse (1963): the log price ratio of each pair is regressed on the difference of period dummies, by least squares, with no intercept. It is the simple, defensible baseline. Case and Shiller’s three-stage weighting corrects the heteroskedasticity that builds up over long holding periods and is the better estimator, and it is not applied here, because a weighted estimate is only as good as the variance model behind it and an unweighted one is easier to check.

Pairs are matched on the same normalised address in the same normalised postcode, which is the same address key the floor-area match uses. Of 13,862,635 candidate pairs, these are removed:

• resold too quickly to be a market movement: 348,069 pairs
• a new build selling for the first time: 1,516,573 pairs
• the property type changed between the two sales: 343,367 pairs
• a price move too large to be a market move: 478,610 pairs
• both sales inside one year: 47,282 pairs

A year is published only where at least 30 pairs touch it, and an area only where it has at least 300 pairs and ten publishable years. That is why every region below appears and why some districts do not.

How should you use it?

For a question about the market, use this index: it is the better instrument for whether prices in a region have risen, because the mix cannot move it. For a question about a home, use the price per square metre on the area page instead: it tells you what a square metre costs today, which is what you need in order to judge an asking price. The two are complements, and this site publishes both partly so that each can be checked against the other.

Figures on this page are read live from the published dataset and are current to June 2026. The full series, including the per-district version, is in the same file. Reuse is welcome under the Open Government Licence; the attribution is 2026 and it is stated in full in the sources below.