RDP 2011-03: Urban Structure and Housing Prices: Some Evidence from Australian Cities 4. Empirical Evidence

4.1 Housing and Land Prices

There is relatively little published data on land prices in Australia's major cities, making it difficult to test some of the model's implications. One exception, however, is for Sydney where the NSW Land and Property Information (LPI) division publishes estimated land values for standard residential blocks in selected suburbs. Prices for 2010 are shown in Figure 9, converted into thousands of dollars per square metre, along with distance from the CBD. There is a clear tendency for higher values for land closer to the CBD, as illustrated by the fitted line based on a kernel density estimate. Indeed, the average land value for the five suburbs within 4 kilometres of the CBD in 2010 was around 16 times the average value for the six suburbs that are at least 50 kilometres from the CBD. This difference is consistent with the implications of the model, in which transport costs lead households to value proximity to the CBD.

Figure 9: Sydney Land Prices – 2010

We can also examine the structure of housing prices within the five largest Australian cities using using postcode-level data.[11] The analysis uses annual median prices for houses by financial year, subject to the requirement that there were more than ten transactions in a postcode in a year.

As is illustrated in Figure 10, house prices tend to be highest close to the centre of each city, consistent with the model's predictions about land and housing prices. Of course, distance from the CBD is not the only factor influencing the desirability of suburbs, and, consistent with Richards (2008), we find that house prices tend to be higher in waterfront suburbs. We illustrate the effect of proximity to the CBD and to the water in the city-level panels of Figure 10 by fitting regression lines with log prices in the five largest capitals as a function of distance to the CBD, the square of distance, and a dummy for waterfront suburbs.[12] The adjusted R-squareds indicate that distance to the CBD and water explain between 40 and 70 per cent of the variation in median prices in these five cities in 2009/10. The regressions indicate that the waterfront effect is largest (adding around 50 per cent to prices) in Sydney and Perth, and lowest in Brisbane and Adelaide. In results that are omitted to save space, we have found that proximity to the CBD and the water are also very significant in explaining the price of apartments in these five cities.

The bottom left panel of Figure 10 compares the fitted price lines for each of the five cities, with the line for each city being a weighted average of the waterfront and non-waterfront lines. The data show that prices tend to be highest in Sydney and Melbourne, the two largest cities in terms of population, and lowest in Adelaide which is the smallest city of this group. This is consistent with the model of Section 2, according to which – holding incomes, transportation costs and other factors constant – prices will be higher in cities with larger populations. While it is hard to be categorical, it appears that the slope of the price/distance relationship is less steep in the data than in the theoretical models, which might reflect employment opportunities in these cities being less monocentric than is assumed in the models.

Figure 10: Postcode-level House Prices – 2009/10

We also find evidence that prices have grown more rapidly in suburbs closer to the city centres as populations (and incomes and borrowing power) have risen. Table 1 shows the results of regressions of the growth in postcode-level median prices on distance to the CBD and a waterfront dummy over periods ranging from 17 years (Adelaide) to 29 years (Melbourne).[13] As in Richards (2008), we find that suburbs closer to the CBD have generally seen higher rates of price growth. Proximity to the water has also become more valued in many cases.

Table 1: Housing Price Growth Regressions – Major Cities
Starting year Constant Distance to CBD Waterfront Adjusted R2 Sample size
Sydney 1986/87 9.4*** −0.39*** 0.64*** 0.34 185
Melbourne 1980/81 10.1*** −0.42*** 1.09*** 0.40 220
Brisbane 1983/84 8.9*** −0.16* 1.03*** 0.25 89
Perth 1988/89 9.1*** −0.03 1.00*** 0.16 82
Adelaide 1992/93 8.2*** −0.23*** 0.38** 0.11 112
Sydney 1986/87 8.4*** −0.38*** 0.18 0.19 103
Melbourne 1980/81 9.4*** −0.36*** 0.17 0.16 120
Brisbane 1983/84 7.3*** −0.28** 0.32 0.09 48
Perth 1988/89 8.3*** −0.31** 0.58** 0.13 63
Adelaide 1992/93 7.8*** −0.40*** −0.09 0.07 75

Notes: ***, ** and * denote statistical significance at the 1, 5 and 10 per cent levels, respectively. Results from an OLS regression of annual average percentage growth to 2009/10 in postcode-level house and apartment prices on a constant, the distance to the CBD, and a dummy variable for whether the suburb is waterfront. The coefficient on the distance variable represents the difference in average annual percentage point growth associated with being 10 kilometres further from the CBD. Given that median price data can be quite noisy, some of the equations omit a few outlier observations.

Source: authors' calculations based on data from RP Data-Rismark

We can use the postcode-level housing price data to do preliminary analysis of developments in housing prices outside of the large state capitals as well. To do so, we calculate the average annual growth rate of median prices between 1992/93 and 2009/10 for all postcodes in Australia for which we have more than 10 transactions in both years. The results are shown in Table 2. For houses, we divide city postcodes into those that are waterfront, those that are non-waterfront but relatively close to the CBD, and those that are non-waterfront and further from the CBD.[14] As shown earlier, we find that prices in cities have grown more in waterfront and inner-city postcodes. We also find that prices have grown by around 1 percentage point faster per year in the capital cities than outside of the capitals. Interestingly, outside of the capital cities, prices in waterfront postcodes have grown more rapidly than in non-waterfront postcodes, just as has occurred inside the cities. In addition, in the subset of postcodes where we can calculate price growth for both houses and apartments, we find that house prices have grown by around 1.2 percentage points per annum faster than for apartment prices.

Table 2: Average Growth in Housing Prices for
Average annual price growth
Per cent
Capital cities (855) 8.4
Waterfront (217) 8.8
Non-waterfront (638) 8.2
Inner suburbs (213) 9.2
Outer suburbs (425) 7.9
Regional (744) 7.4
Waterfront (256) 7.9
Non-waterfront (488) 7.1
Postcodes with price data for both houses and apartments
Houses (821) 8.1
Apartments (821) 6.9
Memo items
Growth in CPI (including GST) 2.7
Growth in housing construction costs 3.7

Notes: The average annual growth rates are calculated as weighted averages of the growth in postcode-level median prices, with weights based on the geometric mean of the number of transactions in 1992/93 and 2009/10. Number of postcodes shown in parentheses.

Source: authors' calculations based on data from RP Data-Rismark

The differences in the rate of price growth identified here are likely to be the result of a large number of factors, some included and others excluded from the model discussed earlier. The fact that prices for houses (which have a relatively larger amount of land per dwelling) have grown faster than prices for apartments is consistent with the run-up in housing prices over recent decades being more a reflection of an increase in the price of land (the scarcer factor) rather than the housing structures per se.[15] The faster growth in prices in the capital cities relative to the growth outside the capitals seems consistent with issues on the supply side (such as those modelled in Section 3) that constrain the construction of new housing being more of a factor in the capital cities. The finding that prices have grown faster in coastal areas than in inland ones is likely to reflect some combination of changes in preferences and the effect of growth in incomes. Finally, the finding that housing prices have grown faster in inner suburbs than in outer ones could be a combination of many factors, including changes in population, incomes and preferences (the latter possibly linked to transport issues), as well as supply-side factors that constrain the increase in the supply of well-located housing.

4.2 Housing Density

Next, we turn to analyse housing density data based on the 2006 Census. As above, the analysis is done at the postcode level. As a measure of medium- and high-density living, we calculate the proportion of people in the postcode who lived in flats or apartments of three or more stories.[16]

Figure 11 shows scatter plots of the relationship between this measure of housing density and a suburb's distance from the CBD for the five largest state capitals. The panels also show a line capturing the typical level of density at each distance from the CBD, fitted using a kernel density estimate. There is a clear relationship between this measure of housing density and distance from the CBD. Furthermore, while the fitted curves are somewhat dependent upon the precise form of the kernel estimation, it is clear that density close to the CBD was highest in Sydney and Melbourne, the two largest Australian cities, and lowest in Adelaide, the smallest of the five cities. It is also noteworthy that density falls off quite quickly: at 10 kilometres from the city, the proportion of the population living in apartments of three or more stories in four of the five cities was very small in 2006. Sydney was the exception, with the fitted line indicating that the typical Sydney suburb 10 kilometres from the CBD had around 20 per cent of the population in such apartments.

Figure 11: Postcode-level Housing Density – 2006

However, the results also indicate significant variation in density in inner- and middle-ring suburbs, with some suburbs close to the CBD with very low density: development plans on council websites for these suburbs typically show relatively little land zoned for medium- and high-density use. In contrast, some middle-ring suburbs have significantly higher density, with some of these being public transport hubs or satellite business districts.

A comparison of housing structure in the 1996 and 2006 censuses provides information on how the structure of the housing stock has evolved and where population growth has occurred. Using data for census areas rather than postcodes, we first divide the census districts within the five largest cities into four groups (referred to as the inner, inner-mid, outer-mid, and outer suburbs), based on their approximate distance from the CBD. Each of the four groups contained approximately the same number of residents in 1996. We calculate how the structure of the housing stock changed for the four groups between 1996 and 2006 using the same metric as in the previous exercise, by comparing the proportion of the dwelling stock that are apartments of three stories or more.

The data, shown in Figure 12, indicate that in Sydney, Melbourne and Brisbane there was a material increase in the proportion of medium- and high-density dwellings in the inner suburbs, with increases of 4–5 percentage points. In Perth and Adelaide the initial proportion of apartments was much smaller, and the increase has also been smaller, at around 1 percentage point. Outside of the inner suburbs, there were mostly only quite small changes in the share of medium- and high-density housing. The exception was Sydney, where the proportion of such dwellings rose by about 2 percentage points in each of the ‘inner-mid’ and ‘outer-mid’ suburbs.

Figure 12: Housing Density Over Time

An alternative perspective can be gained from looking at growth in the overall number of dwellings in the different areas of these five cities. Figure 13 shows this growth between 1996 and 2006. The data indicate that growth in the number of dwellings has been strongest in the outermost suburbs in all five cities. Overall growth in the number of dwellings in the outer suburbs of the five cities was 24 per cent, 14 per cent in the ‘outer-mid’ suburbs, and around 8–9 per cent in the inner and ‘inner-mid’ suburbs. Together, the results in Figures 12 and 13 suggest that a shift towards high-density housing has been occurring in Australia's cities, especially in the three largest cities, although growth in the housing stock has clearly been strongest towards the fringes of the cities.[17]

Figure 13: Growth in Number of Dwellings by Location

We can also compare the density of the five largest Australian cities with large cities in other developed economies, with data shown in Figure 14. The measure of density here is the ratio of population to urban area, and the other cities are a sample of thirty-three cities in Europe and North America with populations of between 0.8 million and 6 million, selected to be broadly comparable in size to the five Australian cities with populations between 1.0 million and 4.3 million in 2006.[18]

Figure 14: Density and Population of International 

We regress the log of density (LDen) on the square of the log of population (LPopsq), with the squared term allowing for apparent nonlinearity in the relationship. Given that the data clearly indicate that North American cities are less dense than European ones, we include a dummy variable for North American (NorthAm) cities, in addition to the dummy variable for the Australian (Aust) cities. The estimated relationships for the three groups of cities are shown in Figure 14 and the equation is given below (with standard errors in parentheses):

The coefficient on the population is significant and takes the expected positive sign.[19] The finding that cities with larger populations tend to have higher density is not surprising and is consistent with the model discussed above. The results also indicate that, controlling for population, the density of US cities is significantly lower than European cities. The density of Australian cities is also significantly less than that of European cities, with the (log) regression coefficient suggesting density of around 43 per cent of European cities (and not significantly different to US cities). That is, notwithstanding the increase in housing density shown in Figures 12 and 13, the density of Australian cities in 2006 was still quite low relative to European cities, although broadly similar to US cities.

4.3 Land Use Policies

Unfortunately, there is little systematic quantitative data on land zoning or land usage in Australia's major cities, in part because some cities either do not have (or are only now moving towards) standardisation in zoning definitions across local government areas. However, in the case of Perth, the Department of Planning makes available some high-quality data on the location and zoning of land zoned for residential usage. This included a standardised measure of permitted housing density, namely the R-Code which represents the maximum number of dwellings per hectare allowed under the zoning: for example, the R20 code specifies no more than 20 dwellings per hectare, that is a minimum block size of 500 square metres (sqm) per dwelling. Codes from R5 to R25 (minimum block size of 2,000 sqm to 400 sqm) are classified low density and can be thought of as being for freestanding, single-family houses; codes from R30 to R60 (from 333 sqm to 167 sqm of land per dwelling) are classified as medium density and broadly correspond to townhouses; and codes of R80 or above (equivalent to less than 125 sqm land per dwelling) are classified higher density and correspond to apartments.

Figure 15 provides an overview of the Perth zoning data as of early 2011, and the relationship between zoning and distance from the CBD, with distance broken into 2 kilometre bands. The figure shows the proportion of land zoned low, medium, and high density at each distance.[20] The data indicate that around 86 per cent of residential land in Perth is zoned for low density/single-family houses, with around 13 per cent for medium density/townhouses and 1 per cent for high density/apartments. Not surprisingly, the land that is zoned for medium- and high-density housing tends to be located closer to the CBD. However, even within 10 kilometres of the city, the majority of land is reserved for freestanding houses, with around 67 per cent of land zoned low density, and only 3 per cent of land zoned for high-density use.

Figure 15: Perth – Zoning of Residential Land

The data for Perth, together with a visual examination of zoning maps (where available) for local government areas in the other large cities, appear broadly consistent with the relatively low density of Australian cities illustrated in Figure 14. By themselves, however, these data say nothing about whether existing zoning may be a constraint on the development of the city along the lines of the effects illustrated in Figure 4. To draw conclusions on this issue would require evidence regarding whether land zoned for medium- and high-density housing use is or is not being used for this purpose, as well as evidence on land values, and whether land is valued significantly more highly when its zoning allows greater flexibility of usage and higher density development. There is, however, very limited hard evidence (as opposed to anecdotal) on these questions.

However, in the case of land at the city fringe, there is evidence from Melbourne of a very significant uplift in prices when land is included (or is considered for inclusion) in the urban growth boundary (UGB). Research by Charter Keck Cramer suggests an increase in value of $300,000 to $400,000 per hectare for land in the UGB, compared with an underlying value of $15,000 to $35,000 per hectare for farming land outside Melbourne.[21]

In the case of land within the city, there is some systematic evidence in a report by the Centre for International Economics (2011) which compares land valuations for blocks of land that are quite comparable in size and location, but differing in zoning. The median valuation differential across 43 Sydney local governments suggests a premium of around 15 per cent for land with medium-density zoning.[22] In addition, there are many anecdotes of very large increases in property values associated with changes in zoning.[23]

Finally, there is also some evidence on the broader question of how well the current dwelling stock in Australia's two largest cities matches the preferences of households. Kelly, Weidmann and Walsh (2011) study the surveyed housing preferences of Sydney and Melbourne households, in terms of location, type and size, taking into account incomes and the cost of various forms of housing. Based on this, the authors calculate the structure of the housing stock that would be implied by household preferences and compare it with the structure of the existing housing stock. Their analysis in this regard suggests that the population would prefer less detached housing in both cities, and more medium- and high-density housing, particularly semi-detached housing and apartments of four stories or more. The authors note that some mismatch between the existing and desired housing stock is always to be expected, given the long-lived nature of the housing stock. However, they note a series of supply-side factors – including many corresponding to the type of factors modelled in Section 3 – that have constrained housing construction. For example, in Sydney, they point to complexity of the planning process for infill development, and infrastructure charges and the price and supply of land on the city fringe. In the case of Melbourne, they note that there have been significant amounts of greenfield development on the fringe and of new inner-city high-rise apartments, but that development in established areas (outside the inner city) has been constrained by planning complexity and high construction costs for apartments.

Together, this evidence suggests that zoning and other planning rules are having a material effect on the use of land, and by implication must also be having some effect on the cost of housing, as per the results in Section 3. While the direction of these effects, and of the other policies modelled in Section 3, are fairly clear, it is difficult to make precise judgments of their magnitude. Indeed, there appears to be little empirical evidence from academics or planners in this area.[24] Given the importance of housing in household budgets and in broader social outcomes, this suggests significant scope for further data collection and empirical work on the pricing, usage and zoning of land in our major cities. There is also scope for studies on the extent to which policies on the supply-side of the housing market could be having greater effects on housing prices as the populations of our major cities have grown. Barker (2004), Bertaud and Malpezzi (2003) and others have argued that data on housing affordability, as well as on land zoning and land usage, should be important inputs into policies surrounding the planning process.


The data for housing prices and distance to the CBD were provided by RP Data-Rismark. [11]

Waterfront postcodes are defined as those adjacent to the ocean or harbour, or on a river of significant width. For estimation of the fitted lines, we constrain the functional form so that house prices decline monotonically with distance from the CBD. This is a fairly basic example of a hedonic price model: see Hill and Melser (2007) for an example where price indices are estimated using individual sales data, and many more variables including distance to schools, public transport, etc. [12]

These, and the results in Table 2, do not control for the possibility of differential rates of quality change, for example the possibility that there has been a greater degree of renovation in particular types of postcodes or systematic differences in the rate of subdivision of existing blocks of land. However, we suspect that the results would be robust to any quality adjustments. [13]

For each capital city, the cut-off for non-waterfront to be considered close to the CBD was set to include one-third of these suburbs. [14]

Interestingly, data from LPI do not suggest that representative land prices in the 55 Sydney suburbs shown in Figure 9 have grown more rapidly over 1996–2010 than house prices for those suburbs. However, the LPI website notes a number of caveats, including the fact that estimated land values may not be comparable over time. [15]

We use this definition, rather than residents per square kilometre, to avoid the problem of obtaining estimates of the amount of urban land for housing in each postcode. In calculating this proportion, we add together those living in separate houses, semi-detached and terrace houses, townhouses, and flats, units or apartments of one or two stories, to compare with those living in flats, units and apartments of three or more stories. We omit six other (typically small) categories of dwellings which are harder to categorise. The analysis includes all postcodes with at least 500 residents. [16]

These data are for the period to 2006. It is likely that the 2011 Census will show some changes, for example, a relatively low rate of greenfield construction on Sydney's fringe. [17]

Data are from a range of sources, including from Infrastructure Australia, and are mostly for around 2006. Definitions of cities are for broader metropolitan regions and for urban land, consistent with the methodology suggested by Mees (2009). We have also obtained very similar results for a larger sample of 112 cities of comparable size in advanced economies in Asia, Australia, Europe and North America from the 2011 Demographia database. [18]

A variable for whether is a city is coastal was also tested. While it took the expected positive sign (i.e. land constraints should result in greater density), it was insignificant. [19]

The data are for the greater Perth area, specifically the Perth Metropolitan and Peel regions. The analysis of the raw data required some judgement. For example, some land in the database had an R-Code, although its zoned usage was something other than residential: in these cases we examined the relevant local planning scheme and included it in the analysis if residential was one of the allowed uses under the designated zoning. In other cases, land was zoned with multiple R-Codes, for example R20/40/80, which indicates that its primary use is for the lower density usage, but that development approval for a higher density use could be granted if certain conditions were met: in such cases we allocated it predominantly to the lower density usage, although the data shown in Figure 15 are not especially sensitive to this assumption. [20]

See GAIC Information Sheet, available at <http://www.gaa.vic.gov.au/gaic/>. Price differences in the case of London's fringe are far larger: Cheshire and Sheppard (2005) suggest differences of millions of pounds per hectare. [21]

There are, however, a few reasons to think that this could be an underestimate, in part due to imprecision in land prices used (based on state government valuations rather than market transactions) and imperfect matching of land (while blocks of land used in comparisons were within 100 metres of each other, it seems plausible that land zoned for apartments would be more likely to be closer to busier roads). In addition, if the typical zoning for apartments does not allow the type of development that that would result in an unconstrained market then the premium would also be understated relative to the unconstrained situation. [22]

For example, a recent article in The Australian Financial Review suggested an uplift of nearly 400 per cent for three blocks of land on Sydney's lower north shore following a rezoning to permit the construction of apartments (see Hurley (2011)). The article suggests three blocks in St Leonards were each valued at about $1 million with low-density residential zoning, but with rezoning to ‘mixed use’ were sold together for $14:5 million. [23]

For example, a search in EconLit with the terms ‘Australia’, ‘land prices’ and ‘zoning’ found no articles nor did ‘zoning + housing + prices + Australia’, while ‘zoning + Australia’ gave two marginally relevant results. In contrast, ‘zoning + United + States’ gave 34 results. [24]