Tag Archives: rent

Silly Housing Statistics 3: The CPI – Official silliness, or just misunderstood and misused data?

In previous posts I have criticised housing statistics produced by industry and the NFP sector. In this post I want to question the official housing statistics in the Consumer Price Index (CPI), which it turns out are not particularly useful in tracking housing costs. This is true even though the CPI includes specific data on housing and rent and is widely used for indexing and updating less frequent housing data.

CPI and Home Owners

The first thing to note about the CPI housing data is that, while it includes an item for house purchase prices, the actual CPI item is “New dwelling purchase by owner-occupiers”. In any given period, relatively few people purchase a house, and those purchasing a new dwelling is an even smaller subset, so the CPI figure here may not reflect prices for many home owners. But more importantly, the CPI figure for new dwelling purchases does not include land prices. In that sense, it measures the increase in the cost of building a new dwelling – which may be very different from changes in the market price of housing (particularly where there is significant speculation-based inflation).

Further, while the CPI includes this (limited) house sale price measure, it does not include mortgage payments (as they are not prices). Yet mortgage payments impact far more on weekly household budgets than house sale prices, and it has been mortgage increases which have driven much of what we now (mistakenly) call a “cost of living crisis”.

The absence of mortgage payments in the CPI is not the fault of the ABS or a problem in the CPI itself. It is a problem in the way the data is sometimes used. The CPI is about measuring price changes experienced by households – a measure of price inflation, rather than the cost of living. The inflation measure is important as a tool of economic management because of the (theoretical) macro-economic relation between production, money, aggregate price levels and jobs. It is particular measure for a particular purpose. In that sense, the statistical silliness is not in the CPI, but in its misuse as a cost of living measure.

The ABS recognises this and produces (a week after each quarterly CPI data release) Selected Living Cost Indexes, which include mortgage payments and are a much better reflection of the impact of housing costs on households. However, these living cost indexes are mostly ignored by the media (presumably out of ignorance) and economists (because their focus is on models rather than real households). Further, the Living Cost Indexes are produced for different household types (based on income-source) and don’t give a single headline figure which can be used conveniently for indexation in the way that CPI is used.

The bottom line is that using the CPI data is not useful in tracking housing prices.

CPI and Rent Prices

And it turns out that that the CPI rent data is not much more use in relation to rent. Unlike the commercial “asking rent” data discussed in my earlier post, the CPI does at least cover increases in existing rents not just new tenancies. However, the CPI significantly underestimates the increases in the market price of rentals (and so is less than helpful for indexing changes to other rent data). There are several reasons for this.

The CPI rent category includes public and community housing rents, which are income rather than market based. In South Australia, this accounts for around one-in-five rentals, but if these (predominantly Centrelink) incomes go up by less than private rents – as is likely in a tight rental market – the smaller increases in public housing rents will lower the overall CPI for rent.

Further, for those in the private rental market the data, the CPI is adjusted for increases in Commonwealth Rent Assistance (because the CPI is designed to capture the price paid by the consumer not the price charged in the market). The rent increases recorded in the CPI for tenants receiving CRA are therefore less than the market price increases.

This CRA adjustment to CPI is even more important because in recent years there have been two significant “above indexation” increases in the CRA. These were welcome increases for those struggling to pay increasing rents, but given that more than half of all tenants receive CRA it means that the CPI rent data further underestimated the actual rent price increases in the market. According to the ABS, without the changes to CRA, rents would have increased by 7.8% over the 12 months to the December 2024 quarter – as opposed to the 6.4% recorded in the CPI.

This is particularly important because it means that it is problematic to use the CPI to update other intermittent rent data. The graph below shows the census data on median weekly rent for the Greater Adelaide area. The census is one of the few sources which captures all rents paid (rather than just new rental agreements), and if the CPI accurately reflected increases in rent prices, the line would be a smooth increase from one census to the next. However, the sharp increases at each census point show that the actual rents actually increased quicker than CPI for rent.

Line graph showing median rents in the census data from 2006, 2011, 2016 and 2021, updated in between by CPI.

Conclusion

The result of all of the above is that, if we want to focus on all rents – rather than just the prices in new tenancy agreements – then we would need three different index series: one for public housing, one for tenants getting CRA, and one for unassisted private renters. While the CPI provides a weighted average index of these, it does not accurately reflect what is happening for each particular group. It exaggerates increases in public housing rents (if market rents are going up faster than income), but underestimates rent increases in the market overall (as shown above).

However, this shortcoming does not legitimise or provide a reason to use the “asking rent” data or the bond data for new tenancies as surrogates for actual rents or for measuring rent increases. As the graph below shows, there is a big difference between the rent increases in those data sets and the CPI. I suspect the real average rent increase is somewhere between these lines .

Line graph showing the difference between increases in the CPI rent and in the SA government rental bond data for 2-bedroom units and 3-bedroom houses - a 40 point difference in less than four years.

This brings us back to where I started this series on silly housing statistics: there is actually no data that shows average rents in capital cities, and in this case, no way to unproblematically track or index rent increases. Yet we can’t just abandon the data, the housing crisis is too important. But we can try to understand and qualify the data to avoid using silly housing statistics.

Silly Housing Statistics 2: I am Priced Out of a renting a beachside house

In my last post I suggested that the claim that Melbourne rents were cheaper than Adelaide was a silly statistic – basically not true, and obviously so. That was the first of a series of posts on silly housing statistics and highlighted a problem with commercially-produced housing statistics. This week’s silly housing statistic comes from my world of not-for-profit advocacy for better housing.

A leading body in this NFP advocacy is the Everybody’s Home campaign which is a coalition claiming over 500 organisations, businesses and councils, and more than 43,000 individuals across Australia, aiming to tackle the systemic drivers of housing insecurity and inequality.

Last year they produced their Priced Out report which included a common, silly statistic used by our sector. I pick this report as an exemplar because of the organisation’s prominence and the importance of the recommendations in the report (which I support), but also because the methodology is repeated across other reports.

The Priced Out Statistics

The report uses SQM Research’s data on asking rents as its key statistics.[1] As I argued in my previous post, these are probably an inflated starting point for rent, and at times the Priced Out report slips into the language of representing these asking rents as all rents (rather than just new tenancies). But this is not the really silly statistic in the report. That comes in the key table (excerpted below) which compares the income of various low-income households with the SQM weekly rental data, and then calculates the percentage of income that would be required to pay the rent on a housing unit. All the results are well over the usual housing stress measure of rent above 30% of income.

Table 2 from the Priced Out report showing national comparison of various weekly incomes, average weekly rents and percentage of income required for rent.

These results would be concerning, but for the fact they are silly housing statistics. They are an asymmetric comparison of a very low income with a median rental price. It is an apples-to-oranges comparison resulting in the grand claim that people on low-income households can’t afford houses that are affordable for someone on twice their income.

Think about it: I can’t afford a beachside house, but I could if I had twice the income. That does not mean that I am at risk of homelessness or even in housing stress. It simply means that I need to have a house that is affordable on my income. The comparator to a market average does not give me data on that. Arguably, the relevant comparator for someone in the bottom 20% of the income spectrum is whether they can afford the cheapest 20% of houses, not the median market price.

It would be theoretically possible for all people in the lowest income bands to afford housing if the rental market spread reflected the income spread. Obviously it doesn’t, which is why the work of Everybody’s Home is so important – I just wish they wouldn’t use silly housing statistics to make otherwise good arguments.

Similar but better options

In my SACOSS role I have done similar income-rent comparisons, but the comparison is to median rent in the cheapest suburbs. In the September Quarter of 2024 the median rent for a 2-bedroom unit in the cheapest suburbs in Adelaide was $60 p.w. below the figure across the city as a whole. By my calculation, an Age Pensioner would require 60% of their income to rent a 2-bedroom unit in the cheaper suburbs of Adelaide, while the Everybody’s Home report puts the Adelaide figure at 74%. Similarly, the Priced Out estimate of 101% of a single JobSeeker’s income for a housing unit is reduced to 82% using the SACOSS methodology – still alarming, and grossly unaffordable, but a more accurate and defensible statistic.

However, the SACOSS methodology is arbitrary (defining the cheapest suburbs as the bottom half) and clunky (in that it is a median of suburb medians rather than a true median of cheapest suburbs).

A far better alternative is provided by the annual Anglicare Rental Affordability Snapshot. Its methodology is simple (but no doubt resource-intensive for them), but for me, it is probably the best of the NFP housing reports. It simply surveys all the rental properties listed online on a given weekend and compares them to the established income-types (Minimum wage, JobSeeker, Pension etc) to see how many are affordable (using the standard 30% of income measure). There remains the issue (discussed in my previous post) of non-advertised rentals, but there is no comparison in the Snapshot to a median price or to properties which would never be expected to be in the price range. It is a scrape to see if any rentals are affordable. Sadly, the answer is usually that there are none or next to none.

This is not a silly housing statistic – it is a damning one! And again, it shows why we need well researched advocacy for changes to the housing system.

[1]              The SQM methodology is more complicated than a simple data scrape, but still relies largely on real estate agent data (the limitations of which are in my last post). Curiously, they also claim their results closely align with ABS CPI rental data – which in the next post I will argue is a government-produced silly housing statistic!

Silly Housing Statistics: Is it more expensive to rent in Adelaide than in Melbourne?

This is the first of a series of articles on “silly housing statistics”, highlighting how bizarre claims are made with inaccurate or mis-used statistics. We begin with the claim made at the end of January that Adelaide rents had surpassed those in Melbourne for the first time. The story came from housing market analysts PropTrack and was covered in the Advertiser (see below) and elsewhere.

An example of a silly housing statistic - photo of the story from the Adelaide Advertiser, 31 January 2025, headed "Rent costs more here than in Melbourne"

Now I like a “killer stat” – a statistic which sums up an issue and grabs attention for a policy argument. I try to get such stats in most reports I write, but I also try to get the statistics right! So, is the claim about Adelaide and Melbourne rents right?

Turns out, the answer is complicated – and a bit of a primer on housing data.

A Silly Housing Statistic

The PropTrack data says that Adelaide’s median weekly rent in the December quarter of 2024 was $580, although this figure for the “asking” or advertised rent was considerably above the official government data on the median amounts actually agreed and paid – which was $550 per week. (The differences here are discussed further below).

Unfortunately, the official Victorian government data won’t come out until the middle of the year (which is why commercial data like PropTrack gets the headlines!). However, the PropTrack data suggests no change in the Melbourne median rent in the December Quarter – which if applied to the official data from the September Quarter would put the median rent at $560 p.w. – that is, above the Adelaide median.

We might also get a hint of the silliness of the “Adelaide more expensive than Melbourne” claim from another commercial data source. The SQM analysis for mid-quarter (12 November) showed Adelaide median rent at $609 per week, but Melbourne at $627 – both significantly higher than the PropTrack figures, but with Melbourne clearly higher than Adelaide.

But really, all these are silly statistics because the comparison is meaningless without also considering the mix of housing in the data. For instance, if rental houses are more expensive than units, then if one market has a greater proportion of houses than the other, its median rent price will be higher even if the comparable prices are lower. That is particularly relevant here because, at least in the September Quarter, only 40% of Melbourne rentals that quarter were houses by comparison with 55% of Adelaide rentals. That would make the overall median rental in Adelaide relatively higher, even if the rent for particular property types is lower.

Similarly, we would need to add location into the housing mix because if the majority of available rentals are in inner-city locations the median price would be higher than if the majority were in outer suburbs.

A further level of complexity here is provided by the government data for the September Quarter. It shows that Melbourne rental prices for units were significantly above Adelaide prices, but $25 a week lower for a 3-bedroom home. The housing mix clearly matters to the overall average – although it should be noted that the SQM data for September showed both Adelaide house and unit prices significantly below those in Melbourne.

A Silly Endeavour?

It is all a bit messy, and in that sense, the silliness of the Adelaide-Melbourne statistic is as much about making the simplistic comparison in the first place as it is about the conclusion of which jurisdiction is more expensive.

But even if you can sort through the comparison issues, there are also some pretty big questions about the usefulness of the base data.

The sad truth is that, despite a flood of housing data, there is actually no data that usefully gives average rents paid in each city, and no real basis from which we can say which city is more expensive – or more generally, what is happening “for renters”.

Navigating Rents, Rent Prices and Asking Rent Data

The first issue is that, all of the above data is in fact not comparing rents in Melbourne and Adelaide – it is comparing rent prices for new tenancies. But new tenancies are less than 10% of the total rental market (my calculation from SA bond data and 2021 census data). So it is simply a misrepresentation (and a silly housing statistic) to say, as in the Tiser headline, that these numbers refer to “rents”.

That said, there may be good reasons to focus on price of new tenancies. These prices represent the current market price faced by people looking to rent, and it is arguably a lead indicator which other rentals will follow. However, even with this focus, and even if what was being measured was accurately described (which it is usually not), there would still be problems.

There are two key data sources of data on rental prices for new tenancies: official state government data based on rental bonds paid, and the “asking rent” (i.e. advertised rent prices) data which is produced by a range of commercial firms (e.g. PropTrack, CoreLogic, SQM) that scrape real estate sites on the web to produce data and spruik their analytical tools.

As we saw above, there are data differences between the companies even on what the asking rent was, but the asking rents may also not be the price the property is rented at (which is what is captured in the official data). For instance, rent-bidding may mean actual prices go beyond the advertised price, but landlords may also wildly overestimate the rent they can get (the Tiser headline could also have read “Adelaide landlords more optimistic than Melbourne counterparts”!).

Further, web sites like realestate.com.au and the other commercial sources are dominated by advertisements from real estate agents, but real estate agents only manage about two-thirds of Adelaide’s private rental properties. The other third may not be advertised or captured by the web-scraping data, and rents tend to be below the real estate agent’s “market price” (see Census GCP4GADE – Table 40).

The difference between asking and actual rents is clear in the table below. It compares rental bond data for Adelaide metropolitan area for each quarter last year with the SQM asking price data for the week in the middle of each quarter. The SQM data (from their National Vacancy Rate reports) is used because it is well-respected and accessible, but as the table shows, the asking price data is 6% to 17% higher than the median rents actually paid by new tenants.

Data table comparing SQM Adelaide rents with rental bond data for units and houses for each quarter of 2024.

Silly Housing Statistics

So, in summary, the data behind the fairly counter-intuitive claim that Adelaide median rents have surpassed those in Melbourne refers to only a fraction of rents being paid across the market, and is inaccurate even in relation to new rentals because it only relates to advertised prices rather than the rent prices actually agreed and paid.

This makes the headline claim about Adelaide prices surpassing Melbourne a particularly silly statistic – and the media’s uncritical reporting of it even sillier.

However, our beloved Tiser is not alone here. A lot of media and even so-called expert commentary, either blindly or lazily, blurs the differences or makes leaps from asking price to rent prices to rents in general. The result is analysis which is inaccurate, over-generalised or at least questionable given the limitations of the data.

Beware the silly housing statistic.

Massively Underestimating Inequality? The Problem of Housing Incomes

Just before Christmas the Australian Bureau of Statistics (ABS) published statistics on Personal Income in Australia for 2018-19. The data, based on income tax returns, provides a window on inequality – and unusually for the ABS, the data is individual rather than household-based.

The ABS summary data tells us that the ACT has the highest median income and Tasmania the lowest. The top 1% of income earners accounted for 9.5% of all personal income, while the top 10% took one-third of all income. As the graph below shows, it also tells us which local government areas have the highest median income.

Income Inequality by LGA

I was also interested to see some quirky South Australian patterns. Roxby Downs, with miners’ incomes and few pensioners, had a stand out median income ($95,196) – way more than the second highest local government area (Walkerville). SA and Tasmania also had fewer people in the highest income bracket (quartile), which partly accounts for the lower average income in those states.

Problems in the Data

However, despite such insights, this headline ABS data is incomplete and is arguably significantly underestimating inequality.

Most obviously the data is incomplete because it does not include those on very low incomes or government payments who don’t have to file a tax return. Accordingly, it is a very truncated income spectrum. Further, household income may look very different to personal income depending on whether there are one or more breadwinners. As the ABS notes, their Household Income and Wealth survey data is a much better base for analysis.

But there is a more fundamental problem (in both ABS data sets) that leads to underestimating inequality, and that is the treatment of housing income.

Housing is important in income terms because different housing tenures create differences in effective (actual purchasing power) household income. This flows through to very different living standards and chances of wealth accumulation. The two key issues, which are not captured in the basic income data, are imputed rents and capital gains.

Imputed Rent

Tenants pay rent to landlords for the use of a house (the provision of housing services) and this appears as income for the landlord. However, a home owner-occupier pays no rent, yet is clearly in receipt of a “housing service” – the provision of shelter and amenity. This issue has long been recognised as a problem in economic statistics. In national accounting, it is a problem because renting out an existing house would suddenly add income and “product” to the market, when in fact nothing more is produced – the house provides the same amount of housing service regardless of whether it is rented or owner-occupied.

The economist’s solution to this issue is to impute a value for rent – that is, assign a value of income as if the home-owner pays rent to themselves. This recognises the value of the housing service being enjoyed as in-kind income by the owner-occupier. In Australia’s national accounts, some 7.7% of Gross Domestic Product is the value of rent imputed to home owner-occupiers (my calculation from ABS data).

Theoretically, the accounts could go further and impute a value for the other services produced within that house (meals, cleaning, counselling, sex) – but that is another story (and a PhD).

However, in terms of housing income, the imputed value of the rent should be added to the income of home owner-occupiers to better reflect the differences in the purchasing power of income between renters and home-owners. This is similar to the common practice in poverty and inequality research of using “after-housing” income as the benchmark. In this case though, the adjustment happens on the income rather than consumption side of the household budget.

We will see below the potential difference this can make to inequality data.

Capital Gains

The second major problem in the basic income distribution data is that it does not include capital gains. Total income includes wages and all employment earnings, income from unincorporated businesses, and investment and superannuation earnings, but not the increase in the value of capital assets.

In an earlier post, I noted a study by Lisa Adkins and others that argued that asset price inflation means that capital gains, capital income and inter-generational transfers are the preeminent drivers of inequality. Further, the authors argue that housing, more than any other asset class, is driving the dramatic increase in net wealth among high-income households. This is so important that they present a new categorisation of class in Australia based on asset ownership.

Of course, there are cash flow problems and costs/barriers to realising (spending) these non-cash incomes, but the fact remains that if renters and home owner-occupiers spent the exactly the same on non-housing items each week, the home-owner would be saving more and accumulating wealth which the renter would not. Their effective incomes are quite different.

Combining Income and Wealth Data

Again, we will see below the potential difference that including capital gains as income could make to inequality data, but including this (and imputed rent) also begins to address a key problem common in presentations which treat income and wealth inequality as different data sets. As I have previously posted, many low-income households do not have low wealth, and some high-income households do not have high wealth – and there are policy problems in shaping policy responses simply around low income.

The ABS did significant work on these issues in 2013 and developed a concept of low economic resource households (defined as those simultaneously in the lowest 40% of income and wealth) – and also included imputed rent in income calculations. However, it was a sideline to the main income data and the ABS has not really continued this work. Further, the concept is clunky to use and has not really been taken up by anti-poverty or inequality researchers (for instance, the World Inequality Database and the flagship ACOSS inequality report both largely present income and wealth data separately and do not include non-cash incomes).

That said, the ABS does still publish stand-alone estimates of imputed rent, and also household wealth from which capital gains can be calculated. Incorporating this non-cash housing income into income data may then be one step in integrating wealth and income inequality data into something more usable.

The Average Renter/Homeowner

The impact of the above can be seen in the following calculation, which is based on the baldest averages in the ABS Household Income and Wealth data for 2017-18.

The average household disposable income for homeowners (both mortgagees and full-owners[1]) was $1,972 per week, by comparison with $1,508 for all renters. In this standard comparison, the average homeowner earned 1.3 times more income than the average renter.

According to the ABS data, the average net rent imputation (that is, imputed rent minus housing costs) was $248 per week. This would take the average homeowner’s enhanced income to $2,220 per week – or 1.5 times that of the average renter.

The ABS data also shows that the average value of owner-occupied dwellings was $755,000 in 2017-18. Two years earlier, it was $673,000. Assuming that most of that $41,000 a year gain was price inflation rather than improvement in housing stock, that equates to $788 per week in capital gains income. This equates to 40% of the household income of the average home-owner (reinforcing the importance of asset price inflation).

The table below shows the impact of both the imputed rent and capital gains adjustments to the income of the average home owner-occupier. Total income for home-owners increases to $3008 per week, basically double that of the average renter.

Table showing Average Income + Net Imputed Rent + Capital Gain = Total Income

Clearly, the inclusion of non-cash housing income for owner-occupies significantly changes the relative incomes between the average home owner and average renter. It is another reason to argue that issues of housing affordability are felt most acutely by renters. While even on the raw ABS data it was possible to say that renters on average earned less than home-owners and spent proportionately more than home-owners on housing, the figures above show that the real situation is even more pronounced and more unequal.

Research Agenda

There is a caveat to the above because of the limitations of adding 3 separate averages (income, imputed rents, capital gains) and holding all other incomes/circumstances equal. It is order-of-magnitude approximation of the impact on inequality data, rather than a real estimation.

What is really required would be to use the ABS microdata to do this calculation for each household to produce a new total income data set. This could be made more complete by including asset price inflation of non-housing capital assets, and counter-balancing with an imputation for “social transfers in kind” – that is, the receipt of public services such as education, public health care, child care subsidies as well as a range of rebates and concessions.

From there a new income continuum could be constructed based on this total income, and then the standard income inequality measures could be applied to that. For instance, it would be possible to calculate a Gini coefficient, income percentile ratios and income shares for the top 1%, 5%, etc using the new income data.

Pretty clearly from the above, many home owner-occupiers would find themselves in much higher income brackets, while renters would be even more clustered in the lower income brackets. But I think that is the reality, that renters in effect have far lower incomes and much tougher cost of living struggles than home-owners (particularly non-mortgagee owners).

Obviously, the total spread of incomes would also be much greater, which I think would give a much truer picture of inequality in Australia.

Unfortunately though, while all this data is in the ABS Household Income and Wealth dataset, I don’t have the technical expertise to crunch the microdata numbers. Any volunteers?

[1]              The ABS presents this data separately, so I have combined them by multiplying the income by number of households in each category, and dividing the sum of both by the total number of households.