Why the average house price can fall while every house gets dearer

Here is a true statement that sounds impossible: the average price of homes sold in an area can fall by ten per cent in a year in which every single home in that area became more expensive. Nobody has to lie for this to happen, and it happens constantly. It is the single most common way a property headline misleads, and understanding it takes about two minutes.

How can the average fall while every house gets dearer?

An average of sold prices measures two things at once: what homes are worth, and which homes happened to sell. Change the second and the average moves whether or not the first did.

Suppose an area has flats worth £200,000 and houses worth £400,000. Last year twenty houses sold and no flats, so the average was £400,000. This year both rose five per cent, to £210,000 and £420,000, but a new block completed and forty flats sold alongside the twenty houses. The average is now about £280,000. Prices rose five per cent and the average fell thirty per cent. Both facts are correct and they describe the same market.

This is not a contrived example. New-build completions, a change in the mortgage market, a stamp duty threshold, a wave of downsizers, all of them shift what sells without shifting what anything is worth.

What does mix adjustment actually do?

Mix adjustment holds the composition constant. Instead of averaging whatever sold, it computes a figure for each category and then combines those with fixed weights, so a change in how many of each sold cannot move the answer.

Everything depends on two choices: what the categories are, and where the weights come from. Property type, size band, tenure, location and age are all plausible categories, and weights can come from past sales, from the housing stock, or from a survey. Two mix adjustments of the same data with different weights will disagree, and neither is wrong.

How do the three markets handle the mix?

This network handles it three ways, deliberately, and the differences are instructive.

• One market weights property types by their share of past sales. The categories are the ones the register itself records, and the weights come from the market rather than from the stock.
• One weights flat types by the national housing stock. That is available because the housing authority publishes what it built, which is a luxury most countries do not have.
• One does not attempt it at all. France publishes no mix-adjusted rate, because the estimator would need weights that hold at commune grain and that has not been shown. Declining is a defensible answer, and publishing an adjustment whose weights are guesses would be worse than publishing none.

That last point is worth sitting with. A mix adjustment is not automatically an improvement. It is a model, and a model with wrong weights introduces its own error in place of the one it removed.

What does a mix effect look like with numbers on it?

Take a district that sells one hundred homes in a year: sixty flats at £250,000 and forty houses at £500,000. The median is a flat, so the headline is £250,000. Next year every single price rises five per cent, but the mix flips to forty flats and sixty houses.

The true price move is five per cent. The median reports one hundred and ten, because it crossed from one population into the other; the mean reports twenty. Neither number is miscalculated and neither is about prices.

Real markets do not flip that hard in a year, which is why real mix effects are a few percentage points rather than a hundred. But the mechanism is the same one, and a few percentage points is the whole of most years’ reported movement.

How do you avoid being caught by a mix effect?

• Prefer a rate to a price. Price per square metre already removes the largest single component of the mix, which is size. It does not remove type or location.
• Prefer a longer window. Composition wobbles quarter to quarter and settles over a year.
• Ask whether the categories moved. A district where a large development completed is a district whose mix has changed, and the median will say so.
• Use an index for change. Controlling for composition is what an index is for. A repeat-sales index does it without needing weights at all, by using the same property twice.

The general rule: levels from a median, changes from an index. A percentage change computed from two medians is the market plus the mix, and nothing in the number tells you the split. Where a market on this network publishes a mix-adjusted series beside its median, it labels it as one: IndexProp SG and RealScout both do.