Big City Difference? Another Look at Factors Driving House Prices

Abstract

The purpose of this paper is to re‐examine the issue of whether inter‐urban housing markets can be modelled using a set of common economic fundamentals (such as economic growth, employment and the like). This is a timely analysis in view of the current widespread interest in housing markets as a result of the fall‐out from the housing sub‐prime crisis in the United States. House prices and economic fundamentals within each city are tested for cointegration and, in the event of a cointegrating relationship being found, restriction tests are applied to ascertain whether particular economic fundamentals can be excluded from the long‐run equilibrium house price model for that city, and whether the given fundamental contributes to speed of adjustment back to equilibrium once a disturbance has taken place. This allows a test of whether the given factor/s can be considered a long‐run driver/s of house prices in each city. The main finding is that there are clear differences across Australian state capitals in long‐run driving factors for house prices.

Keywords:

• House price drivers
• economic fundamentals
• cointegration
• restriction tests

Acknowledgement

The authors gratefully acknowledge the various suggestions offered by four referees that resulted in a substantially improved paper.

Notes

1. This important consideration was suggested by a referee. The number of transactions varied each quarter throughout the study period and, as an example of Capital City differences, the number of transactions in the final quarter of the study period for Sydney was just under 12,000 while the number of transactions in Darwin was just over 400. While the Australian Bureau of Statistics maintains a constant sample size across all cities in the construction of its price indices, drawing a sample from a smaller population of transactions may well have repercussions for the analysis. Regarding the price data ‘The price index for established houses covers transactions in detached residential dwellings on their own block of land regardless of age (i.e. including new houses sold as a house/land package as well as second‐hand houses). Price changes therefore relate to changes in the total price of dwelling and land’ (cf. ABS definitions on house price indices). As noted by a referee the inclusion of both new and established housing in the index could be a source of bias, but the ABS does produce an index to distinguish this.

2. This is not meant to imply that there was no longer interest in aggregate house price models in later decades (cf. Peek and Wilcox, 1991; Iacoviello, 2000; Iacoviello and Minetti, 2003; Abelson et al., 2005; Goodhart and Hofmann, 2008).

3. An excellent survey of the methodologies and results can be found in Ball (1973).

4. A ripple effect has also been tested in Australia (cf. Luo et al., 2007).

5. Two referees pointed to a potential source of bias here. The CPI includes a housing component. However, weighting data were not available to ‘wash this out’, so the full CPI was used for adjustment.

6. Note that the aim of the graphs is not to be able to visually identify each city but to observe differences and similarities of movement.

7. Also see the earlier discussion on UK research on ripple effects.

8. The Inoue (1999) procedure suffers from a test limit of five variables within any given model structure, whereas neither the Johansen (1988, 1991) nor the JMN (2000) procedures have this limitation.

9. A summary may also be found in Wilson and Zurbruegg (2003).

10. As it is powerful against all alternatives. Strictly speaking the Inoue (1999) procedure is used to test for cointegration in the presence of a possible break and is not a breakpoint test in itself.

11. Inoue (1999) suggested that, in the absence of information on the type of break, Model B is the preferred model. The Inoue (1999) model B indicated the following potential breakpoints: March 2000 in Sydney, Melbourne, Brisbane, Perth and Hobart; September 1998 in Canberra; March 1989 in Darwin; and September 2001 in Adelaide. These breakpoints were then used in the JMN procedure. As shown in Table 5 the outcome from the JMN procedure showed that the cointegrating rank was virtually the same as that resulting from a conventional Johansen procedure.