RDP 2006-03: Australian House Prices: A Comparison of Hedonic and Repeat-sales Measures 1. Introduction
May 2006
- Download the Paper 177KB
Movements in house prices can have important consequences for the Australian economy. Housing constitutes around 60 per cent of all household assets, and house prices can be prone to large swings or ‘boom-bust’ cycles. These swings can influence economic activity through both dwelling investment and consumption, with wealth effects influencing borrowing and spending decisions.
Since house prices are important for economic and financial developments, measuring their level and growth rate accurately is desirable. There are two key concerns which complicate this task: the prices of non-transacted houses, which are unobservable; and the difficulty in measuring the quality of houses that are heterogenous, especially if housing characteristics change through time. With only a subset of the population of houses sold in any given period and considerable heterogeneity across houses, the composition of houses sold can differ between periods. Differences in quality across houses at a given point in time can be difficult to control for, in part because of practical considerations regarding the available data, but also because housing is inherently heterogenous since the location of each house is unique. Quality also varies through time with changes to existing dwellings and the construction of new, typically higher-quality, housing. Hence, movements in house prices can reflect pure price changes, changes in the mix of houses sold and changes in the quality of houses.
Compositional and quality change can affect simple measures of house prices such as a median and, to a lesser extent, mix-adjusted measures (that adjust only for specific types of compositional change). To overcome the limitations of these measures, several regression-based approaches have been proposed in the literature.[1] These include: hedonic measures, which regress the log-level of prices against house attributes over time;[2] repeat-sales measures, which regress price changes of the same house over time (Bailey, Muth and Nourse 1963; Case and Shiller 1987); and hybrid measures, which combine the hedonic and repeat-sales approaches (Case and Quigley 1991; Quigley 1995).
Notwithstanding the extensive research on regression-based measures and other measures, there is still little consensus as to whether there is a superior approach to measuring house prices either on theoretical grounds or according to empirical comparisons, with many results conditional on the data used. One area that the literature has not focused on extensively is the volatility of alternative house price measures, particularly over a short-term horizon. Yet for researchers and policy-makers concerned with near-term movements in house prices, it is desirable to be able to distinguish between pure price changes, compositional changes, quality changes, and statistical noise associated with reporting error. This is particularly so around turning points in the data, where house price readings can be important in forming assessments of the housing market and the broader economy.
This paper makes two contributions. First, I compare two regression-based measures of house prices – specifically, hedonic and repeat-sales measures – to develop a regression-based approach to measuring house prices that controls for compositional change and, to some extent, quality change. This extends previous Australian research in this area, including Rossini, Kooymans and Kershaw (1995), Costello (1997) and Flaherty (2004), by using a wider range of measures and a larger sample of data, as well as placing more emphasis on the theoretical properties of alternative measures. Second, I gauge the performance of the regression-based measures by comparing them to some simpler measures of house prices, including a median and a mix-adjusted approach developed by Prasad and Richards (2006). This provides more general information about suitable measures in the Australian context, and whether the regression-based measures provide a better control for compositional and quality change. This work is complementary to Prasad and Richards, who investigate some of the key practical issues in house price measurement such as timeliness, seasonality, data availability, and the extent to which a mix-adjusted measure can help to control for compositional effects.
Estimates are based on unit record data on house sales for Australia's three largest cities – Sydney, Melbourne and Brisbane – from the March quarter 1993 to the September quarter 2005. The data are supplied by Australian Property Monitors (APM) (Sydney, Melbourne and Brisbane) and the Real Estate Institute of Victoria (REIV) (Melbourne). Data from APM are originally sourced from state land titles offices, while REIV data are a mixture of data from the land titles office and real estate agents. The data are comparable to that used by the Australian Bureau of Statistics (ABS) in producing its quarterly index on house prices, and cover around half of all house sales that occur in the Australian market.
The results suggest that regression-based methods can provide a useful control for compositional effects. In particular, hedonic and repeat-sales measures provide similar estimates of pure prices growth, suggesting that theoretical issues associated with choosing an appropriate specification may be less important for price measurement in practice. Estimates from these measures are also comparable to a simple mix-adjusted measure. In contrast, the median – which makes no adjustment for compositional change – displays considerable volatility and lower average price growth over the sample period. This implies that compositional change has mattered empirically.
The similarity between the hedonic, repeat-sales and mix-adjusted measures is somewhat surprising though, since the hedonic can, in principle, control for quality while the latter two cannot. This suggests that either the data available are not able to definitively capture quality change, or that quality change has been quite limited. In contrast, a repeat-sales regression with a constant (an alternative control for quality change) implied modest quality-related price increases over the sample period.
The remainder of the paper is organised as follows. Section 2 provides an overview of the concepts and definitions relevant to measuring house prices and discusses some of the theoretical difficulties involved. Alternative measures of house prices, both regression-based and non-regression-based, are examined in Section 3. Section 4 evaluates the various measures empirically. Conclusions are drawn in Section 5.