RDP 2020-04: The Apartment Shortage Appendix D: Hedonic Regressions

As discussed in Section 7, we regress Sydney apartment prices on a wide range of hedonic controls and find that households do not have a strong preference for low-rise apartments (the ‘missing middle’) relative to high rises. Regression output is shown in Table D1. Most explanatory variables are discrete, with omitted categories denoted ‘--’. Coefficients are multiplied by 100 to be interpretable as approximate per cent changes. The first section of the table shows the value of architectural features (bedrooms, age, etc). The second section of the table shows the value of distance to nearby amenities.

The top rows show our key results. Being in a building with 10 or fewer dwellings adds 6.3 per cent to the value of an apartment, relative to being in a building with more than 100 dwellings, after controlling for apartment quality and spatial characteristics. Being in a building with 11 to 20 dwellings adds 0.3 per cent.

We do not include in our regression spatial variables whose coefficients are jointly insignificant. This includes train stations and large shopping centres. That surprised us given that real estate advertising and past research (Murray 2016; Pettit et al forthcoming) suggest that these locations are highly valued. We suspect that these variables are strongly correlated with other features (noise, parking, apartment quality) that are difficult to control for. Many other results from the regression are as expected. The number of bedrooms, number of bathrooms and proximity of the apartment to water were all associated with large and clear increases in price. Apartment age has large and clear effects. One interesting implication is that housing ‘filters down’ to lower price ranges as it ages. The large coefficients contrast with the small unconditional effects of age discussed in Appendices A and E. The difference may arise because new apartments are less expensive for other reasons – for example, if they are further from the city centre. For purposes of comparisons with the unconditional mean of supply costs the unconditional effect of age is relevant. For reasons of space we do not show coefficients on the approximately 650 suburb dummies, though these are collectively the most important influence on apartment prices. However, the general contour of suburb effects can be seen in Figure 4. Apartments near the city centre sell for several hundred thousand dollars more than those on the outskirts, other things equal.

Table D1: Hedonic Regression – Sydney
Dependent variable: log sale price; includes month and suburb fixed effects
  Coefficients (multiplied by 100)
Building density (baseline > 100)
≤10 dwellings 6.3***
11–20 dwellings 0.3
21–50 dwellings −0.5
51–100 dwellings −0.9
Bedrooms (baseline = 1)
Two beds 24.7***
Three beds 41.5***
Four beds 46.9***
Five+ beds 48.1***
Bathrooms (baseline = 1)
Two baths 11.5***
Three baths 24.8***
Four baths 34.0***
Five+ baths 34.3***
Bedroom/bathroom ratio 1.2
Parking spaces (baseline = 1)
Two spaces 8.2***
Three spaces 17.8***
Four spaces 15.9***
Five+ spaces 33.1***
Extra features
Swimming pool 0.7***
Air conditioned 2.5***
Ducted heating 1.8***
Scenic view 5.4***
Apartment age at sale
2–5 years −10.4***
5–10 years −17.1***
10–15 years −21.8***
15–20 years −22.4***
20–30 years −28.1***
30–40 years −30.8***
40–50 years −32.7***
50–60 years −32.9***
60+ years −15.8***
Arterial road/motorway (yes = 1) −4.0***
Log distance to CBD −10.7***
Spatial feature Distance from property
≤ 0.5 km 0.5–1 km 1–3 km > 3 km
Beach 13.8*** 8.3*** 3.9** --
Cemetery −5.5*** −3.2*** −2.7*** --
Club −2.4* −3.0** −3.2** --
Community facility 9.9* 9.5* 8.5 --
Education facility (TAFEs etc) −8.9*** −4.2** −0.3 --
Fire station −3.3** −0.7 1.2 --
Headland 20.3*** 7.3*** 0.5 --
Library 2.9** 1.9 1.9 --
Mountain 1.7 1.3 −2.3*** --
Light rail stop −6.1* −3.2 2.9** --
Sports centre 5.5* 5.2* 6.1* --
Swimming pool −3.5*** −3.1*** −2.5*** --
University 6.0*** 1.4 1.1 --
Combined school −3.1*** −1.9** 0.1 --
High school −0.8 −1.0 0.3 --
Sewage works −4.2 −3.5** 1.4 --
General hospital 1.7* 1.5* −0.6 --
Number of observations 553,275
R-squared 0.81
Root mean squared error 0.25

Notes: ***, ** and * denote statistically significantly different from zero at the 1, 5 and 10 per cent levels, respectively; -- denotes omitted category; standard errors (not shown) are clustered at buildings, there are 42,411 clusters; spatial categories are from Spatial Services (2019)

Sources: Authors' calculations; CoreLogic data; PSMA Australia; Spatial Services