How house price indices work, and why repeat-sales is the good one

Every house price statistic you see is built one of three ways, and the differences matter enough to reverse conclusions.

Median sale price

Take everything that sold in a period and report the middle price. Simple, timely, and widely quoted.

Its flaw is severe: it measures what sold, not what things are worth. If a quiet quarter happens to include several large houses, the median rises even if every individual property is worth exactly what it was. During the 2020 to 2021 period, a genuine shift in what people were buying, larger homes further out, pushed medians up by more than underlying values rose.

Useful for "what does a typical transaction look like right now". Useless for "what happened to my house".

Hedonic regression

Model price as a function of attributes: floor area, bedrooms, age, location, garden, parking. Fit it on observed sales, then use the coefficients to price any property, including ones that did not sell.

This is the machinery behind most commercial automated valuations. Done well, with enough data, it is powerful. It has two weaknesses. It needs rich attribute data on every property, which is expensive and often wrong in public records. And it can only value what it can measure, so it systematically misses the things that do not appear in a database: the view, the noise, the neighbour, the light.

Repeat-sales

Only look at properties that have sold more than once, and measure the change in price between their own sales. Discard everything that sold only once.

This throws away most of the data, which sounds wasteful, and buys something valuable in exchange: it holds the house constant. When the same property sells twice, the difference is market movement rather than a difference in what was sold. That is the whole idea, and it is why the FHFA index in the United States and the Land Registry index in the United Kingdom are both built this way.

A repeat-sales index answers "what happened to the value of a given house" rather than "what did the average transaction cost".

Why that matters for you specifically

If you know what you paid and when, a repeat-sales index for your area is not an approximation of a better method. It is the method. The index was constructed to measure exactly the quantity you want: how much the value of one house in that area has changed over a period. Applying it to your own purchase price is the intended use.

This is why a valuation tool that asks what you paid can be more accurate than one that does not, even with far less data. It is using a better estimator, not a bigger one.

Where repeat-sales still fails

Three limitations, and they are real.

• It describes an area, not your house. If you extended the property or let it deteriorate, you have diverged from the local average and no index will notice.
• Renovation contaminates it. A house that sells, gets gutted, and sells again shows a price rise that is partly market and partly the new kitchen. Index builders filter aggressively for this, but they cannot catch everything.
• Thin markets are noisy. An index for a small area with few repeat sales moves around for statistical reasons rather than economic ones. Prefer the metro or district level over something smaller.

What to take from this

When you read a house price figure, check which construction produced it. A median telling you prices rose 8% and a repeat-sales index telling you they rose 3% are not contradicting each other; they are answering different questions. For your own property over your own holding period, the repeat-sales number is the one you want.