Do Urban House Prices Converge?

Abstract

We investigate two aspects of housing market price dynamics. Firstly, whether the spatial pattern of house prices in a metropolitan housing market converge or diverge over time and secondly, whether suburbs with relatively low (high) house prices 20 years ago continue to occupy the same relative position in the house price distribution. The empirical work uses a property transaction database for Melbourne to examine the changing distribution of suburban house prices over a nearly 20-year period (1990–2009) that spans two housing cycles. We focus on convergence measures that use Melbourne submarket-based repeat sale house price indexes as a unit of measurement. We find that house prices diverge, and so the gap between low-priced submarkets and high-priced submarkets is increasing. A second key result is that low-priced submarkets typically remain at the low end of the house price distribution, because their rates of appreciation fall short of those at the upper end of the house price distribution. The geography of house price dynamics suggests that the price gradient with respect to distance from the central business district is becoming steeper.

本文考察住房市场价格变化的两个方面。第一，大城市住房市场房价的空间分布随时间趋同还是分异； 第二，20年前房价偏低（高）的郊区在房价分布中是否依然占有同样的位置。研究利用墨尔本房产交易 数据，考察近20年（1990–2009）来郊区房价的变化分布，跨越两个住房周期。我们以墨尔本次级市场 重复销售指数为测量单位，测量趋同性，结果发现房价呈分异趋势，低价次级市场与高价次级市场的差 异扩大。本研究的第二个主要成果是，低价次级市场基本位于房价分布的低端，因为其升值率低于房价 分布的高端。房价变化的地理分布表明，与距商业中心距离相关的房价等级差距正在加大。

Keywords:

• House price convergence
• repeat sale house price indexes

Acknowledgements

The authors would like to thank Jessie Pomeroy for research assistance and Elizabeth Taylor who was responsible for assembling the data set. The authors would also like to thank Rachel Ong for constructive advice. All errors and omissions are the responsibility of the authors.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

^1. Yates (2012, p. 83) shows that between 1990 and 2010 housing wealth accounted for almost 60 per cent of total household wealth.

^2. In a similar vein house price divergence could signal widening socio-economic disparities between neighbourhoods. Fleishman (2013) shows that in Israel regional median house prices are closely correlated with composite indices of socio-economic status, and argues that they can be reliably used as a proxy for socio-economic status.

^3. If house prices in suburbs are viewed as randomly subject to various forces, some causing appreciation, others precipitating decline, relative price differentials will emerge as the product of these random forces. While this version of central place theory (Evans, 1985, p. 69) explains the existence of a price hierarchy, it does not permit change in the ranking of suburbs by price, and assumes that there are no systematic forces correlated with the level of prices. The following economic theories indicate that this assumption is clearly incorrect.

^4. Clark et al. (2003) use the Panel Survey of Income Dynamics to analyse the sequence of housing states defined in terms of tenure and the quality/price of dwellings that households in the USA occupy during the course of their housing careers. Of all careers with two or more states, 77 per cent show an ascending pattern in terms of tenure and/or quality (price) of dwellings occupied.

^5. In Australia (UK) over the period 2001–10, Ong et al. (2013) found that 56 per cent (46 per cent) of those trading on hold larger amounts of housing equity in their new purchase, and so rollover the capital gains accrued on their previous owner-occupied home.

^6. We are grateful to Elizabeth Taylor who was responsible for the original design and creation of the merged data set.

^7. Property characteristics have not been used in the approach to measurement of rates of price appreciation, as we estimate repeat sales models, but the location of each property has been critical in defining the 108 submarkets. See Wood and Cigdem (2012) for a more detailed discussion of data sources and how our two main data sets were merged.

^8. A complete list of suburbs and corresponding submarkets is listed in table AIII of Sommervoll and Wood (2011, p. 82).

^9. In 1990, skewness is estimated to be approximately 1.2, and this grows to be approximately 1.7 in 2009.

^10. The thick black line above every whisker denotes a set of outliers, that is, prices that are judged atypically high when compared to the rest of the price distribution. An outlier is judged as any value smaller (larger) than the following lower (upper) limits: [Median −  1.5 ×  IQR, Median + 1.5 ×  IQR], where IQR =  75^th percentile −  25^th percentile.

^11. Young et al. (2008) demonstrate that β convergence is a necessary but not sufficient condition for σ convergence. Hill et al. (2009) measure σ (but not β) convergence in the Sydney housing market over the period 2001–2006, and Cook (2012) estimates β (but not σ) convergence in UK regional house prices.

^12. The growth rate is calculated as (I_(T)/I₍₀₎ − 1)^*100 (where I₍₀₎ is the repeat sales price index in the first time period (0) and I_(T) is the price index in period T, the final time period).

^13. Barro and Sala-i-Martin (1992) estimate a different version of Equation (1) as they compare growth rates in per capita incomes across country data sets in which growth rates are measured over different time spans (see footnote 7 of Sala-i-Martin, 1996, p. 1334). The values of β estimated from a linear model will not allow a comparison of the speed of convergence when time spans differ, an issue that Barro and Sala-i-Martin (1992) are keen to address. Our analysis is city specific with each suburb's house price growth measured over exactly the same time span, and so the simple linear specification is sufficient for our purposes.

^14. Both the constant and β estimate are significant at the 1 per cent level. Note, we use Newly West Adjusted Standard Errors as results from the Durbin Watson test applied to residuals indicate autocorrelation is present. The autocorrelation probably reflects omitted variables such as the spatial distribution of income that is likely correlated with the starting period distribution of median house prices. However, since our aim is to detect whether the drivers of the temporal and spatial pattern of house prices result in convergence or divergence of metropolitan house prices, not the identification of these drivers, we retain the naive specification in Equation (1).

^15. The standard deviation divided by the median house price.

^16. Starting at 0.53 in 1990 it increases to 0.73 in 2007, before drifting back to 0.60 in 2009.

^17. Thin labour markets can impact on efficiency in other ways. A small geographically scattered pool of suitably skilled labour will deter investment in technologically sophisticated capital as it is riskier in such labour market settings (Acemoglu, 1997; Moretti, 2011).