Mobile Edition
Print
Get the Mag
Weekly UpdatesABSTRACT. Prior research supports the proposition that house price diffusion shows a ripple effect along the spatial dimension. That is, house price changes in one region would reflect in subsequent house price changes in other regions, showing certain linkages among regions. Using the vector autoregression model and the impulse response function, this study investigates house price diffusion among Australia's state capital cities, examining the response of one market to the innovation of other markets and determining the lagged terms for the maximum absolute value of the other markets' responses. The results show that the most important subnational markets in Australia do not point to Sydney, rather towards Canberra and Hobart, while the Darwin market plays a role of buffer. The safest markets are Sydney and Melbourne. This study helps to predict house price movement trends in eight capital cities.
KEYWORDS: Regional house prices; House price diffusion; Vector autoregression model; Impulse response; Market efficiency
SANTRAUKA
Ankstesniu tyrimu duomenimis, nekilnojamojo turto kainq kitimas sukelia bangu efekta atsizvelgiant i erdvini matmenj. Tai yra nekilnojamojo turto kainu kitimus viename regione rodytu paskesnis nekilnojamojo turto kainu kitimas kituose regionuose. Taip ryskeja tam tikri glaudus rysiai tarp regionu. Taikant vektorini autoregresini modeli ir impulso perdavimo funkcija, sioje studijoje tiriama nekilnojamojo turto kainu kitimas tarp pagrindiniu Australijos miestu, nagrinejant vienos rinkos reakcija i kitu rinku naujoves bei nustatant uzdelstus terminus kitu rinku reakciju maksimaliai absoliutinei vertei. Rezultatai rodo, kad svarbiausios Australijos vidaus rinkos nera orientuotos i Sidneju, bet labiau i Kanberot ir Hobarta. Darvino rinka atlieka buferio vaidmeni. Saugiausios rinkos yra Sidnejus ir Melburnas. Si studija padeda numatyti nekilnojamojo turto kainu judejimo tendencijas astuoniuose pagrindiniuose Australijos miestuose.
1. INTRODUCTION
House prices in Australian main metropolitan areas displayed sharp increase trends from 1996 to late 2003 and early 2004 when the trends eased. Although the current Australian house prices movement trend does not exhibit any obvious recessionary signs, the housing market at the sub-national level, such as in Sydney, is taking the lead in experiencing a downturn after 2004. Housing prices in Sydney in the June quarter 2006 were still lower than in the December quarter 2003. On the other hand, Perth held its high rates of increase in the same period (ABS, 2008).
House prices in cities were influenced by their past house prices, house prices in other cities, mortgage rates, net migration, policy factors and others. The relationships between house prices and economic variables, and between house prices at the national level and a subnational level in Australia were tested by using the real estate data in the period 1989-1998 (Tu, 2000). Using the Granger causality test, two diffusion paths which formed a geographic diffusion pattern in the Australian housing market were determined: starting from Brisbane via Sydney ending at Melbourne, and starting from Brisbane via a national path and ending at Melbourne. A ripple effect which showed a diffusion pattern from north to south was detected in some capital cities. It was also found that house price indices were correlated statistically in several Australian capital cities (Abelson and Chung, 2004). It was suggested that a long-run relationship exit between house prices, house income and consumer index, while adjustment to equilibrium was found to be in significant lags, in the short run (Abelson et al., 2005). Moreover, Luo et al. (2007) studied the housing price diffusion pattern of Australian capital cities. The results supported that a 1-1-2-4 diffusion pattern exists.
This study investigated the dynamics of the house price diffusion in Australia's state capital cities, examining the response of one market to the innovation of other markets and determining the lagged terms for the maximum absolute value of the response. Using the eight capital cities' house price indices, the vector autoregression (VAR) model is constructed to investigate the impulse response function (IRF), which is utilized to analyse the sensitivity of one market to the shocks of others. The next section provides a review of related literature. Section 3 describes the data source and the investigation period with respect to the house price indices of eight capital cities in Australia. Section 4 presents the unit root tests and the results of the stationarity test on the data series. The VAR model and the impulse response function are described and used to measure the interregional housing markets' responses, respectively in the section 5 and the section 6. Finally section 7 provides conclusions.
2. LITERATURE REVIEW
Dynamic analysis of VAR model is carried out using the impulse response function (Sims, 1980). This approach is widely used in the real estate research. Using the Engle-Granger cointegration test and the vector autoregression Granger causality test, the relationships of regional housing markets were investigated in the South of England and, in the North and Midlands of England (Alexander and Barrow, 1994). Podlodowski and Ray (1997) examined regional repeat sales house prices from 1975 to 1994 in the USA. Using the VAR model, this study estimated the significance of a one lag order. The findings supported the notion that the market was inefficient and that contiguous regions release more influence than noncontiguous regions.
Evidence from prior research supports the proposition that house price shocks in one area are likely to spread to other areas (MacDonald and Taylor, 1993; Alexander and Barrow, 1994; Ashworth and Parker, 1997; Pollakowski and Ray, 1997; Meen, 1999; Tu, 2000; Stevenson, 2004; Cook, 2005). This is the so-called house price diffusion or ripple effect. The ripple effect or house price diffusion has been mentioned recently in literature describing the examination of UK regional house prices. It describes how house prices rose first in the South East and how this gradually spread out over the rest of the UK. In this case, two key elements should be focused on: diffusion paths and epicentre. Diffusion paths are certain kinds of relationships between regional housing markets.
It was demonstrated the concept of spatial dependence to explain the ripple effect (Meen, 1996). Spatial dependence refers to the linkages between regional markets. It was suggested that a single national housing market should be treated as a series of interregional linkages between housing markets (Meen, 1999). Both bidirectional and unilateral causalities between the regional housing markets illustrated a series of linkages between them. It was examined that the causal relationships between Irish regional housing markets (Stevenson, 2004). The results supported the view that Dublin had a lead effect with other markets. It was displayed that a causal relationship pattern and revealed the so-call "ripple down" effect in the UK regional housing markets (MacDonald and Taylor, 1993).
Market efficiency was also identified in this issue in some previous research. Tirtiroglu (1992) constructed two models, containing contemporaneous neighbouring and non-neighbouring markets; and lagged neighbouring and non-neighbouring markets, to test the speed of spatial diffusion. The results indicated that the market was inefficient. Clapp and Tirtiroglu (1994) tested the significance of a positive feedback hypothesis in Hartford, Connecticut. The study found that regional markets were affected not only by their own past values but also by neighbouring regions' past values.
[FIGURE 1 OMITTED]
3. DATA DESCRIPTION
The study focuses on house prices diffusion at the subnational level. House price indices for the eight state capital cities were collected from the publications of the Australian Bureau of Statistics (ABS). The period is from the December quarter 1989 to the June quarter 2007. The indices are based on the quarterly house prices for established and newly erected dwellings and each capital city's house price index based on 1989-90=100.
Figure 1 shows the house price movements in eight capital cities. The biggest change in house prices was in Darwin (+350.3%) during the investigated period, followed by Brisbane (+318.7%) and Adelaide (+286.9%). The Darwin housing market shows very different behaviour from the other seven markets. Except for Darwin, the other seven show a similar propensity during the investigated period. They all have a slow increase trend at first which is followed by a sharp increase. The start of the latest boom in Melbourne, Adelaide, Perth and Sydney led the other markets. Melbourne's boom started in the December quarter 1996 while the booms in Adelaide, Perth and Sydney started in the March quarter 1997, followed by Brisbane (June quarter, 2002), Canberra (June quarter 2000) and Hobart (June quarter, 2000). Darwin started its first sharp increase from the December quarter 1989 until the June quarter 1997, with an average change rate of 3.62% per quarter followed by a steady increase until the September quarter 2000. The latest sharp increase in Darwin started from the December quarter 2001. Melbourne, Sydney, Brisbane, Canberra, and Hobart both had an obvious hesitation in the December quarter 2003 and the March quarter 2004. However, Perth, Darwin ignored this strike and were experiencing their rapid increases.
4. STATIONARITY TEST FOR HOUSE PRICE INDICES
A stationary time series is significant to a regression analysis based on the time series, because useful information or characteristics are difficult to identify in a nonstationary time series. Therefore, a nonstationary time series would lead to a spurious regression. However, most economic time series are nonstationary in practice. Fortunately, time series can be made to be stationary after differencing. Useful information or characteristics can still be identified in the time series after differencing. Moreover, if two or more variables are nonstationary and have the same order of integration, they can be constructed in a cointegration model. Therefore, the stationarity test should be launched before the cointegration test. A time series is said to be stationary if its mean and variance are constant and, the covariances depend on upon the distance of two time periods. In order to indicate the difference from strict stationarity, the word "stationary" in the term "stationary time series" means weak stationarity or covariance stationarity in this study. In this step, the unit root test is used to test the variables' stationarity and the order of integration. The Dicky-Fuller unit root test (DF), Augmented Dicky-Fuller unit root test (ADF) (Dicky and Fuller, 1979) and the Phillips-Perron unit root test (PP) (Phillips and Perron, 1988) are often used to test stationarity. The ADF and PP tests were used in this study. There are 3 forms of the ADF and PP unit root test model.
FEED