Tradergy
GUIDES

Why house price measures disagree

Three methods are used to measure house prices and each controls for something the others do not. The starting point is stranger than the methods: a median can move a long way in a quarter when not one property in the country has changed value.

Reviewed by the Tradergy editorial team · September 2026
In short

A plain median reflects which homes sold, not what homes are worth. The Reserve Bank put it directly: "even when there are no changes in the value of any houses within a city in a particular period, if the proportion of sales that occur in more expensive suburbs increases, the measured city-wide median price will increase." Stratification controls that mix across chosen strata but not quality change over time; hedonic regression controls both but only as well as its model is specified; repeat-sales compares a property with itself but assumes it was not renovated and discards every home that sold once.

Two house price figures published in the same week can point in different directions without either being wrong. Before comparing any of them it is worth sitting with the problem they are all trying to solve, because it is more awkward than it looks.

Key Takeaways

  • A median reflects which homes sold, so it moves when the mix changes and nothing else does.
  • The Reserve Bank estimated in 2006 that around 60 per cent of quarterly median variation came from mix alone.
  • Only a small fraction of the housing stock transacts in any quarter - around 1½ per cent, on 2006 data - which is why mix moves so much.
  • Stratification fixes composition across strata but not quality change over time.
  • Hedonic models control both, and are only as good as their specification.
  • Repeat-sales compares a home with itself, and assumes it was not renovated in between.
  • The ABS and the main private index draw on the same underlying sales data.

The problem: almost nothing sells

A share index can be computed from the prices of every constituent, continuously. Housing cannot, because housing barely trades. Working on data available in 2006, the Reserve Bank put the share of the Australian housing stock transacted each quarter at "around 1½ per cent", and noted that average turnover is significantly lower in some other countries.

So every house price measure infers a stock from a small, self-selected sample of it. Which homes come to market in a quarter depends on rates, seasons, confidence, tax changes and the age profile of owners. That is the mix, and it moves.

The Reserve Bank stated the consequence as a clean counterfactual: "even when there are no changes in the value of any houses within a city in a particular period, if the proportion of sales that occur in more expensive suburbs increases, the measured city-wide median price will increase."

It also estimated the size of the effect, and the estimate is the reason this is a methodological problem rather than a footnote. Working on data available in 2006, it attributed "around 60 per cent of the quarterly variation in median price changes" to changes in the mix of sales between higher and lower priced suburbs, rather than to price changes reflecting demand and supply. That figure is a finding about a particular period and should not be carried forward as a constant, but the mechanism it describes has not changed.

Method one: the median, and what it is good for

A median needs no data beyond the sale prices themselves, which is its entire advantage: it can be produced quickly, for any geography, with no model to argue about.

The Reserve Bank's research sets the condition under which it is sound. A simple measure of central tendency "will be an accurate guide to pure price changes only when there is little change in the composition of houses sold between periods, and when quality change is limited."

Over a long span in a stable market both conditions roughly hold, and a median is a reasonable description. Over a single quarter in a moving market neither holds, which is when a median is least reliable and most quoted.

Seasonality is a related trap. Medians show a repeating within-year pattern, but it is a pattern in what sells rather than in what things are worth: seasonality in median prices, as the Reserve Bank's researchers put it, "arises because of seasonality in the composition of transactions that occur."

Method two: stratification

Stratification splits the sample into groups by characteristics - price level, location, size, amenities - and tracks each group separately before recombining. Because the comparison happens within groups, a shift of sales between expensive and cheap suburbs no longer moves the result.

Its advantage over simply seasonally adjusting a median is that it also removes compositional effects that are not seasonal. Seasonal adjustment can only remove a pattern that repeats; stratification removes the mix effect whenever it occurs.

Two limits remain. Mix can still shift within a stratum, and the finer the strata the fewer sales each contains, so precision and control trade against each other directly. And, as the Reserve Bank's research notes, unlike hedonic measures "mix-adjusted measures take no account of quality changes in the housing stock over time." A gradual national improvement in the standard of housing is read as price growth.

Method three: hedonic regression

A hedonic model treats a house as a bundle of characteristics and estimates what the market pays for each. With those estimates it can value the difference between two different homes, and so compare periods in which entirely different properties sold. It uses regression techniques, in the Reserve Bank's description, "to control for compositional and quality change", producing direct estimates of pure price change.

The cost is that the answer now depends on a model. The same research is explicit: hedonic models "are only as good as the specifications used to derive them", and omitting relevant variables "can result in biased estimates of pure price changes."

The characteristics that are easy to record - bedrooms, land area, postcode - are not the only ones buyers pay for. Outlook, light, street noise and the quality of a renovation are priced by the market and largely absent from the data, and where an unmeasured characteristic shifts systematically across the stock its effect lands in the price term.

Method four: repeat-sales

Repeat-sales sidesteps comparison between different houses entirely by using only homes that have sold more than once, and measuring the appreciation of each against its own earlier price. It "more accurately controls for the attributes of houses since it is based on observed appreciation rates of the same house."

Its assumption is that "house characteristics (that is, quality) have not changed over time" - and for a country that renovates, that assumption is violated constantly and in one direction. A kitchen replaced between sales appears as price growth.

It is also, in the Reserve Bank's word, inefficient: it discards every property that has sold only once, which is most of them. And the homes that turn over more often are not a random subset of the stock.

The point most comparisons miss

Faced with a divergence between an official figure and a private index, the instinctive reading is that two independent measurements disagree. In Australia that reading is wrong at the source.

The Bureau of Statistics states that all Australian residential property sales data are supplied to it by CoreLogic - now trading as Cotality - combining data from state and territory land titles and Valuers-General offices with agent data. The same underlying transactions therefore feed both the official series and the largest private one.

What differs is method and clock. Cotality's home value index is hedonic, updated daily, and revises its back series for the previous twelve months at the start of each month. The official series waits for the transaction record to fill in and then treats the result as settled; its release timing and revision points are on the house prices page.

Those are two defensible positions on one trade-off: publish immediately and correct as the record arrives, or wait for the record and correct rarely. Neither is more accurate in general. What follows for a reader is narrower and more useful than "the ABS disagrees with CoreLogic": on the same underlying sales, a gap between two published measures at any moment is as likely to be the clock, or the method, as the market.

Where to look next

The current published mean and its history are on the house prices page, and the stamp duty calculator applies each state's rates to a purchase price. The cash rate is the policy rate mortgage pricing responds to, and why the same purchase costs different duty in every state covers the transfer costs that sit on top of a price. For quality adjustment in the consumer price index, which uses the hedonic technique for a different purpose, see what a comparison in real dollars can and cannot tell you.

Sources

Not financial advice, and not a recommendation of any product or institution. Reviewed by the Tradergy editorial team, September 2026.

Common questions

Why do different house price measures disagree?

Because they control for different things. A plain median controls for nothing and moves with the mix of what sold. Stratification controls composition across chosen strata. Hedonic regression controls composition and quality through a model. Repeat-sales compares each property with its own earlier sale.

Can a median rise when no house has changed value?

Yes. The Reserve Bank stated that even with no change in the value of any house, a rise in the proportion of sales in more expensive suburbs will increase the measured median. In 2006 it estimated that around 60 per cent of the quarterly variation in median price changes came from the mix of sales rather than from price change.

What is the weakness of a hedonic index?

Model specification. The Reserve Bank's research notes that hedonic models "are only as good as the specifications used to derive them", and that omitting relevant variables can produce biased estimates of pure price change.

What is the weakness of a repeat-sales index?

It assumes the property did not change between sales, so renovation is read as price growth. It also uses only homes that have sold at least twice, which discards most of the market and risks a sample that is not representative.

Are the ABS and the private indices independent measurements?

Not in their source data. The ABS states that Australian residential property sales data is supplied to it by CoreLogic, drawn from state and territory land titles and Valuers-General offices together with agent data. The differences between published measures are largely processing and timing, not separate observations of the market.