The causes and economic consequences of rising regional housing prices in New Zealand

Abstract

This paper explores the causes and economic consequences of recent increases and divergences in regional house prices in New Zealand. It identifies large and increasing ‘wedges’ between house prices and underlying supply costs. These house price distortions arise from the collision of rising demand for housing with housing supply constraints, including zoning rules that limit new subdivision and redevelopment of existing sites. Regions with larger starting price distortions appear to have experienced larger increases in house prices and rents in response to migration shocks. This results in large economic costs due to misallocation of labour away from high-productivity regions in New Zealand and increased net migration to Australia. A calibrated spatial equilibrium model is used to investigate what would have happened if house price distortions had increased at a slower rate in recent decades due to relaxation of supply constraints. This model implies that comprehensively removing constraints to housing supply would have increased New Zealand’s total economic output by up to 8.4%, increased per-worker output by 0.9%, and eliminated recent net migration outflows to Australia. More plausible counterfactual scenarios would result in smaller, but still economically meaningful, gains on the order of one to five percent of gross domestic product.

KEYWORDS:

• Housing prices
• zoning
• housing supply
• migration
• urban growth

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 In New Zealand, cities with constraints on housing supply often have abundant business land. For instance, MBIE (2017) finds that industrially-zoned land is often valued at a discount from nearby residential land. A 2015 study also found that one-quarter of the industrially-zoned land in New Zealand’s upper North Island, which includes housing-constrained cities like Auckland and Tauranga, was vacant in 2011, implying significant ‘slack’ in business land markets (Sanderson, Williamson, Paling, Stokes, & Dixon, 2015).

2 For example, a city surrounded by avocado and kiwifruit orchards will have a higher opportunity cost to convert rural land than a city surrounded by sheep farms, as orchards are a higher-value rural use.

3 This is calculated at the mean, but it could in principle be calculated at different points in the distribution of dwellings, as in Cutter and Franco's (2012) analysis of the impact of minimum parking requirements in Los Angeles.

4 I included the following variables: natural log of building floor area (lnF_i); indicator variables for decade of construction (A_i); indicator variables for Census area unit (CAU_i), which act as proxies for localised amenities/disamenities and accessibility to productive jobs; indicators for sale year (Y_i); indicators for building construction, building condition, roof construction, and roof condition; and indicator variables for whether the property has a view of water or a view of land (V_i).

5 In Auckland, Greenaway-McGrevy et al. (2018) find that properties with different intensity ratios have followed different price paths in recent years, which they interpret as evidence of a changing redevelopment premium for low-intensity sites. Hence including this variable should control (to an extent) for the degree to which property prices reflect expected future redevelopment profits.

6 Territorial authorities were based on 2006 boundaries, i.e. they split out the Auckland region into seven territorial authorities rather than grouping it into one region. There were a total of 72 TAs. To create the sales dataset, I excluded all properties not matching the description ‘Single Unit excluding Bach’, properties with missing or zero sales prices or rating valuations, properties with zero land area (which are predominantly cross-lease sites or misclassified apartments), properties with land area over 1 hectare (likely to be large lifestyle blocks or misclassified farm properties), properties with building coverage larger than site area or building floor area more than five times building coverage (likely to be misclassified apartment buildings or data entry errors), and properties with building floor area under 50m² or over 500m². I then filtered out the top and bottom 0.5% of the distribution of sales price to rating valuation ratios. For properties that were missing data on views, building or roof construction or condition, I created an indicator variable for missing values. For all properties, I restated sale prices and improvement values in real 2017 Q1 New Zealand dollars using Statistics New Zealand's (2018b) Consumer Price Index.

7 Coefficients on the intensity ratio variable vary in sign between regions and over time; in general, the price premium (discount) for less-developed sites appears to have increased (decreased) over time. In the online appendix, I present and briefly discuss coefficients for the intensity ratio variable.

8 This correlation coefficient is based on model 1 results; model 2 results show a similar degree of correlation.

9 This captures both changing demands for land and changing lot sizes.

10 Unlike in the previous section, this analysis takes place at the level of the territorial authorities defined at the 2013 Census, of which there are 66. In 2010, seven Auckland TAs were aggregated into a single Auckland Council. Unfortunately, several key data series were only available for the newly aggregated Auckland Council area.

11 Following Saiz (2007) and Glaeser, Kolko and Saiz (2001), I included the 2001 share of the population aged 15+ that had a bachelor’s degree or postgraduate degree, reasoning that places with a higher educational attainment may be more likely to benefit from broad structural economic changes (‘knowledge economy’), and also the 1972–2013 average annual sunshine hours, reasoning that places with a better climate may be more attractive in general. Grimes et al. (2016) find that higher human capital and a better climate positively affected regional growth over the 1926–2006 period.

12 Interestingly, coefficients on controls for local amenities, incomes and unemployment rates, which are not reported, are not statistically significant in most model specifications, and at times have counterintuitive signs (e.g. a negative sign on sunshine hours).

13 Strictly speaking, not all of this effect is due to migrant arrivals. At the national level, increased migrant arrivals tend to coincide with reduced departures of New Zealanders, which will also affect local housing demand. Hence these estimates may capture the effect of both outcomes.

14 Available online at: http://nzdotstat.stats.govt.nz/wbos/Index.aspx?DataSetCode=TABLECODE7037

15 New Zealanders do not necessarily earn the national average wage, or buy the average house, as they may cluster in certain industries or cities. However, as Australian wages and prices are treated as exogenous, choosing different values will not affect the model results.

16 However, large cities in Australia will tend to be more productive.

17 By comparison, Statistics New Zealand data suggests that labour productivity in the measured sector rose by 22% over this period, while real GDP (production measure) rose by 59%.

18 Preston et al also find that coastal, sunny, and scenic locations tend to have higher amenity levels, and that Auckland’s relative amenity levels have risen over the 2001–2013 period. However, their analysis suggests that Auckland offers relatively lower quality of life.

19 Both specifications assume full employment, with no labour supply response from people entering the labour force.

20 Coates, Oram, Bertram, and Hazledine (2016) critique the modelling assumptions underpinning this estimate and argue that the true economic gains may be significantly lower. Their critique reinforces my point that the gains from improving housing supply are likely to outweigh the gains from further trade liberalisation.

21 See Nunns (2018) for a further description of this measure. I also interact this measure with a dummy variable for whether the TA in question is a city or district council, as city council boundaries tend to be drawn more tightly around the existing urban area.

22 In the UK, Hilber and Vermeulen (2016) demonstrate that more stringent requirements to process consents within statutory timeframes has caused councils to reject developments faster to comply with the letter, but not the intent, of the law. Anecdotally, the response to the 2009 law changes in New Zealand has been more benign: councils use requests for further information from applicants in order to ‘stop the clock’ on processing time.

23 The Environment Court hears appeals on decisions taken under the Resource Management Act, which is the framework legislation for urban planning in New Zealand. This data is available from the last quarter of 1996 to the end of 2017.

24 A further issue is that councils tend to receive more appeals in the wake of major district plan reviews. In recent years, plan reviews have often (but not always) resulted in increased options to build new housing, e.g. via reduced minimum lot sizes or greenfield rezoning. This may create problems for interpreting this measure.

25 Further testing suggests that these impacts did not differ between city and district councils. City council boundaries are typically drawn tightly around urbanised areas, while district councils typically include a larger amount of rural land. This suggests that this measure is unlikely to suffer from endogeneity due to how it has been defined.

26 A share of migrants leave their destination unstated. This share does not appear to have trended up or down over the 2001–2016 period and hence I disregard it. However, I note that it will mean that my estimates of the effect of migration on housing prices are slightly over-stated.

27 I estimated individual incomes by taking the midpoint of the income band that Statistics New Zealand coded workers into.