RDP 2018-03: The Effect of Zoning on Housing Prices 5. Zoning Effect Estimates

We now use our estimates to show that zoning restrictions are a major contributor to the cost of housing, and the largest proximate cause of cross-city differences in average house prices.

Table 3 shows the decomposition of average house prices into the three components. The structure and land split from Table 1 is repeated in the first two rows. The third row shows the contribution of land value due to physical scarcity, calculated by multiplying our baseline (large equation) hedonic estimates of the marginal value of land by the average lot size. The fourth row shows the zoning effect, which is the difference between the average land value and the physical land value. These are the paper's central results.

Table 3: Average House Price Decomposition
$′000 (per cent of total), 2016
  Perth Brisbane Melbourne Sydney
Dwelling structure 242 (41) 267 (49) 268 (34) 395 (34)
Land 346 (59) 275 (51) 524 (66) 765 (66)
Physical land 140 (24) 116 (21) 201 (25) 276 (24)
Zoning effect 206 (35) 159 (29) 324 (41) 489 (42)
Total 588 (100) 542 (100) 793 (100) 1,160 (100)
Zoning effect as a percentage of physical input costs 54 42 69 73

Sources: Authors' calculations; CoreLogic

The zoning effect can be expressed as a percentage of the costs of the physical inputs or of the total price. We alternate between the two, depending on the context. We estimate that zoning restrictions raise the price of the average house in Sydney by 73 per cent above the value of the physical inputs (structure and physical land) required to provide it. Corresponding effects are 69 per cent for Melbourne, 42 per cent for Brisbane and 54 per cent for Perth. As a share of total property value, the contribution from zoning is 42 per cent in Sydney, 41 per cent in Melbourne, 29 per cent in Brisbane and 35 per cent in Perth. There are substantial differences between cities, especially in dollar terms, with zoning restrictions adding only $159,000 to the cost of the average house in Brisbane, compared with $489,000 in Sydney.

Estimates of the zoning effect in other countries span a wide range. The effect is estimated to be large in cities with tight restrictions and negligible where building is unrestricted. Our estimates are similar to estimated effects in cities with tight restrictions. For example, Glaeser et al (2005) estimate zoning effects, as a share of total price, for 1998–99 of 53 per cent in San Francisco, 47 per cent in San Jose and 34 per cent in Los Angeles. Cheung et al (2009) estimate zoning effects for 2005 of around 40 per cent for Miami and Orlando, 30 per cent for Tampa and 25 per cent for Jacksonville. Lees (2018) estimates effects of 56 per cent in Auckland, 48 per cent in Wellington and 32 per cent in Christchurch.

In important ways, our estimates are based on better data than was available to these earlier studies. Where other estimates often rely on small samples, our regressions have sample sizes in the tens of thousands. Where some other studies use self-assessed property values to estimate the hedonic value of land, we use actual sale prices. Where other studies only have industry construction cost estimates, we are able to use valuer general and ABS estimates also.

In contrast to the effect of zoning restrictions, differences in dwelling structure values across cities account for relatively little of the variation in average sale prices. Our estimates of average structure value for Australia are broadly similar (at 2017 exchange rates) to estimates for New Zealand in Lees (2018), but almost 50 per cent higher than estimates for the United States from Glaeser and Gyourko (2018) (see the Online Appendix, Section 1).

We estimate that the value of physical land for the average house in Sydney is higher than for Melbourne, and substantially higher than for Brisbane or Perth. However, as a share of average house prices these estimates are similar (between 21 and 25 per cent), and high compared with overseas estimates. Physical land accounted for between 4 and 9 per cent of average house value in Auckland, Wellington and Christchurch based on the numbers reported in Lees (2018). Estimates in Glaeser and Gyourko (2003) imply that physical land accounted for between 3 and 12 per cent of total land value in the six cities with the highest average house prices in their study. The relatively high physical land value that we find for Australian cities may reflect geographical constraints facing the expansion of these cities (e.g. coasts and national parks), combined with high demand reflecting population growth and low real interest rates.

In decomposing property values into the structure, physical land and zoning, some readers have asked whether ‘location’ – as, for example, measured by our suburb dummies – should also be allowed for, given that this is an attribute that buyers pay highly for. To the extent that land at a location is physically scarce, the value of location will already be captured in our physical land term. To the extent that land at a location is scarce because of administrative restrictions, it shows up in the zoning effect.

5.1 Robustness of Our Hedonic Estimates

A central feature of our results is that the marginal value of land is less than the average value. Our method for estimating the contribution of zoning restrictions to house prices is based on this difference between marginal and average value, so sensitivity of our marginal land value estimates to the functional form of our hedonic regression would be a concern.

Our baseline logarithmic specification is simple, but fits our data well. It is fairly common in the literature, being used by previous hedonic house price applications, in both Australia (e.g. Hansen 2006; CoreLogic 2017) and overseas (e.g. Glaeser et al 2005). Our estimates of the log land area coefficient are not sensitive to the inclusion of control variables (as shown in Table 2 above) and are stable across time.

More importantly, our estimates of marginal land value are insensitive to a range of more flexible functional forms and with estimating regressions at a disaggregated level, although our baseline estimate for Melbourne is slightly lower than estimates from other methods. Table 4 shows some illustrative examples. See the Online Appendix (Section 2) for more detail on these alternatives. Cheung et al (2009) and Sunding and Swoboda (2010) also find that estimates of the zoning effect are robust to various regression specifications and levels of geographic aggregation.

Table 4: Physical Land Value Estimates from Alternative Specifications
Marginal value of land at mean sale price and lot size, 2016, $ per square metre
  Perth Brisbane Melbourne Sydney
Greater Capital City-level regressions
Baseline (log-log, all controls) 219 129 317 411
Including (log land area) squared 226 138 319 422
Replacing log land area with cubic land area terms 292 159 374 460
Weighted average from disaggregated regressions
LGA-level (log-log, all controls) 203 143 337 442
Clusters of SA2s (linear-linear, all controls) 235 102 364 404

Sources: ABS; Authors' calculations; CoreLogic

In contrast, some researchers have used more restrictive specifications, such as a linear-linear specification of sale prices and land area. However, this specification has the implausible implication that land is equally valued everywhere in the city (whereas in the log-log specification it is proportional to property value). We find even lower estimates for marginal land value using this functional form.[14]

Our hedonic estimates of the marginal value of land are similar to other hedonic housing studies of Australian cities.

Hansen (2006, p 16) reports regressions of log house prices using Australian Property Monitors data. His coefficient estimates on log land area are similar to ours for Sydney (0.28) and Melbourne (0.19). However, he reports a much larger coefficient of 0.39 for Brisbane. One cause of this difference could be that Hansen's regressions are run on only a sub-sample of 1 per cent of Brisbane sales due to missing values for a range of characteristics. In addition, Hansen's sample includes townhouses as well as detached houses, and controls for property zoning.

Syed, Hill and Melser (2008) construct hedonic house price indices for different regions of Sydney from 2001 to 2006 using regressions of logged sale prices on a range of characteristics including quadratic terms for lot size, and lot size interacted with a number of other variables. Their estimated quadratic lot size coefficients for 2006 (p 57) imply that a 1 per cent increase in lot size (for a property with a lot size of 635 square metres – the mean in our sample of sales in 2006) would increase sale prices by 0.21 per cent, similar to our estimate of 0.24.

Frino et al (2010) report regressions of log house prices in Australian cities for the period 2005 to 2009. They report coefficients on log land area that are substantially lower than our findings: 0.10 for Sydney, 0.08 for Melbourne, 0.13 for Brisbane, and 0.16 for Perth. This appears to be because they control for location at a much less granular level than we do (our coefficient estimates are similar if we include dummies for the ABS Statistical Area Level 4 of each property instead of the suburb).

5.2 Sensitivity of Our Zoning Effect Estimates to Inputs

Our zoning effect estimates are calculated indirectly, and so depend on our estimates of the other components of the cost of housing. We now consider how sensitive our zoning effect estimates are to changes in our estimates of structure value and physical land value. Overall, our baseline estimates are reasonably robust.

Table 5 reports the estimated zoning effect calculated using each of our methods for estimating structure value individually, instead of the average. Our sense is that although different approaches lead to noticeable variations in the precise estimate of the zoning effect, these are not large enough to alter our qualitative conclusions.

Table 5: Zoning Effect Estimates with Different Structure Values
Zoning effect as a per cent of average house prices
  Perth Brisbane Melbourne Sydney
Baseline estimate 35 29 41 42
State valuer general 44 40 43 33
ABS Building Activity Survey 35 34 44 49
Industry sources 26 14 35 44

Sources: ABS; Authors' calculations; CoreLogic; Department of Environment, Land, Water and Planning (Victoria); Department of Natural Resources and Mines (Queensland); Rawlinsons Group (2017); Rider Levett Bucknall (2017); State of New South Wales through the Office of the Valuer General; Western Australian Land Information Authority (Landgate)

Table 6 reports the estimated zoning effect using the estimates of physical land value from the alternative specifications reported in Table 4. Some of these specifications suggest a slightly higher physical value of land than our baseline estimate, but our zoning effect estimates are not very sensitive to these differences.

Table 6: Zoning Effect with Different Physical Land Estimates
Zoning effect as a per cent of average house prices
  Perth Brisbane Melbourne Sydney
Greater Capital City-level regressions
Baseline (log-log, all controls) 35 29 41 42
Including (log land area) squared 34 28 41 41
Replacing log land area with cubic land area terms 27 24 36 39
Weighted average from disaggregated regressions
LGA-level (log-log, all controls) 37 27 39 40
Clusters of SA2s (linear-linear, all controls) 33 34 37 43

Sources: ABS; Authors' calculations; CoreLogic

5.3 Other Issues

Our estimates take the value of physical land as exogenous. However, zoning restrictions that reserve land for industrial or agricultural uses increase the physical scarcity of residential land and hence raise its market value. This implies that our estimates of the zoning effect may understate the total impact of zoning restrictions on the cost of housing.

In contrast, any factor that increases the wedge between sale prices and our estimate of supply costs will tend to boost our estimate of the effect of zoning. One example is time delays between an increase in housing demand and when new housing supply comes online, which could temporarily boost prices. If these delays reflect regulatory policies, it is reasonable to interpret them as part of the zoning effect. However, the persistence of zoning effects documented in Section 6 suggests that simple delays are not as important as long-lasting obstacles. To the extent that delays reflect genuine physical/market constraints, considering estimates of the zoning effect over a multi-year period, as we do in Section 6, should mitigate this concern.

Another example is if developers are able to exert market power to earn supernormal profits, then sale prices of residential properties will be higher than the marginal costs that they face. However, the market for developing and building residential dwellings is generally considered to be highly competitive, with low barriers to entry and a large number of firms (Glaeser et al 2005; Minifie, Chisholm and Percival 2017; Lees 2018).[15] Glaeser and Gyourko (2018) show that in US cities without tight zoning restrictions, prices are usually close to costs.

Perhaps the most important concern about our estimates is the possibility that we have underestimated costs. As noted in Section 3, we exclude subdivision and infrastructure costs, which are substantial for greenfield developments but not for increasing housing density. We also exclude adjustment costs, specifically when demolition of old structures is necessary for a change in land use. When demolition is necessary, the cost of supply is not only the cost of land and new structures, but also includes the acquisition cost of partially-depreciated structures.

We have excluded adjustment costs for two reasons. First, they are highly variable and difficult to quantify. Second, house prices are determined on the margin, and marginal adjustment costs are often low. Australia's cities have low average density and heterogeneous housing stocks.[16] So underdeveloped blocks of land and heavily depreciated structures, for which adjustment costs are low, are widespread. Substantial increases in density – for example, from large-lot detached dwellings to high-rise apartments – reduce adjustment costs by spreading them across many new dwellings. Moreover, increasing the density of new developments does not incur a marginal adjustment cost. Adjustment costs are more likely to be material in already developed areas, and it is possible that this (along with the cost of consolidating land parcels for high-rise development) may further contribute to higher property values in inner-city areas, relative to the city fringe.


Some researchers include more complicated geospatial information in their hedonic housing models. We experimented with including various factors such as the distance to the CBD, schools, the coast, or railway stations in our regressions, but these had almost no effect on the estimated marginal value of land (likely due to the inclusion of suburb dummies). Specifications that controlled for the average sale price of nearby properties resulted in marginally lower estimates of the hedonic value of land. [14]

Concerns have been raised about the practice of ‘land banking’ pushing up the prices of potential greenfield development sites. However, there is no clear evidence of widespread delays caused by speculative behaviour, and reforming zoning arrangements would likely reduce the time taken for lots to be released for sale and increase competition (HSAR Working Party 2012). [15]

As noted by Ellis (2013), Australian cities are less dense than the cities of any other sizeable country. [16]