economic agencies have devoted considerable efforts to forecasting housing
activity, especially in times of recession when attention is focused on the housing
market for signs of recovery. The commonly used measures of housing activity
are approvals, commencements and completions. In this paper, a forecasting
exercise is undertaken on these housing variables, recognising the need to test
for co-integrating relations among integrated variables.
Co-integration of economic variables is associated with "error-correction"
models, which allows the specification to capture, where necessary, long term
relations as well as the short term dynamics. Specifically, co-integrated systems
permit individual time series to be integrated of order 1,1(1), but certain linear
combination of these series can be stationary, integrated of order 0,1(0). Recent
empirical research in time series analysis has demonstrated that the relative
success of forecasting exercises depends importantly on the issue of integration
or stationarity of the variables in the systems investigated. For example, if an
1(1) system is co-integrated, then the commonly used vector autoregressive
(VAR) models in levels or first differences will be mis-specified. Granger (1986)
and Engle and Granger (1987) showed that, if two variables are co-integrated of
order (1, 1), then they can be modelled by an error correction model (ECM). The
forecasting ability of the ECMs, compared with the unrestricted VAR models,
for co-integrated systems was examined by Engle and Yoo (1987) and LeSage
(1990) using either a simulation approach or real data. They concluded that,
compared with the VAR models, the ECMs produce more accurate forecasts for
5 6
longer horizons.
The ECM derived by Engle and Granger (1987) for estimation of a cointegrated system is based on the assumption that every component in the system investigated is integrated of order 1 and co-integrated (designated as of order 1, 1 hereafter). In reality, there are many economic systems involving both integrated and stationary series. The co-integrating relations between these integrated series, if any exists, would also have important implications for forecasting the stationary series in the system if the integrated set and the stationary variables are related. As will be shown later, the housing model used in this study is such a system which involves both integrated and stationary series. While approvals, commencements and completions are found to be stationary over the sample period investigated, other financial and economic variables which influence housing activity, such as finance approvals for established dwellings and real aggregate income, are detected to be 1(1) and co-integrated. In this chapter, we derive the associated ECM-like representation for such a system and utilise the order selection procedure developed by Penm, Penm and Terrell (1992a) to identify the specification for estimation and forecasting.
While most traditional modelling uses finance approvals for new dwellings as the central component for forecasting the construction of new dwellings, see the Treasury's NIF-88 model, the results of the present analysis indicate that new construction is better forecast using finance approvals for established dwellings. Several competing models are compared with those determined by the proposed procedure using both the bootstrap technique and real data. The bootstrap results
5 7
indicate that the specifications determined by the proposed procedure exhibit superior forecasting characteristics to their alternatives. In the case of forecasting using real data, these specifications, compared with other competing models, bring significant gains to forecasts of new housing construction.
3.2. A Housing System for Forecasting
Modellers of Australian housing activity appear to have made different assumptions on the underlying relationships in the housing market. The Treasury's NIF-88 model uses an approach in which the variable, finance approvals for new dwellings, plays a major role in the forecasting of new construction. The ratio of loans for new dwellings to that for total dwellings, in that model, is postulated to be related (negatively) to the change in total finance approvals for both new and established dwellings (Bassanese, Horn and Simes 1989). This framework seems to suggest that, among other influences, changes in the market activity for established dwellings would influence new construction mainly via changes in the finance approvals for new dwellings. A different approach is to model the construction of new dwellings directly using variables representing the market conditions for established dwellings. Williams (1984) argues that the market price of established dwellings sets an upper bound on the price at which new dwellings may be sold, and builders will reduce or abandon activity if the total cost of construction is not sufficiently below the price of established dwellings. If this is true, then new construction activity could be better explained and forecast using information directly linked market activity for established dwellings such as the finance approvals for established dwellings.
5 8
Which approach will perform better in the forecasting of new construction is an interesting question and will be best investigated using real data.
In this study, an analysis is set up which focuses mainly on the activity of residential construction. A housing system is constructed which consists of seven housing, financial and economic variables - approvals, commencements, completions, numbers of finance approvals for new and established dwellings, real aggregate income and real construction costs. The underlying economic relationships between these variables appear complex. Clearly, there are relations between approvals, commencements and completions. If there are no leakages, all approvals will become commencements and all commencements will become completions. However, in reality, there are always leakages between these housing activity variables. These leakages may be related to changes in those economic factors which influence the housing market. In this system, the levels of finance approvals for new and for established dwellings are both used to explain housing demand as finance approvals will be influenced by changes in mortgage interest rates and house prices. We separate the finance approvals for new and established dwellings in an attempt to look at the relative importance of finance in each market on the construction of new dwellings. There may also be some feedback from housing activity to these finance approvals. For example, in a situation of over-supply of new dwellings, the price of new dwellings would become relatively cheaper in the short term compared with established dwellings, increasing the demand for new dwellings and hence the associated requests for finance approvals. Assuming an unchanged total demand for dwellings in the short term, the demand for established dwellings would fall as substitution