BUILDING APPROVALS AS A LEADING INDICATOR OF PROPERTY SECTOR INVESTMENT

Overall, building approvals for new houses (BANHs) are viewed by most economic analysts/commentators as a leading indicator of property investment due to the importance of this sector to the whole economy and employment. This study seeks shed some additional light on modelling this seasonal behaviour of BANHs by: (i) establishing the presence of seasonality in Victorian BANHs; (ii) ascertaining it as to whether is deterministic or stochastic; (iii) estimating out-of-sample forecasting capabilities of the modelling specification; and (iv) speculating on possible interpretation of results. The study utilises a structural time series model of Harvey. Factors corresponding to June, April, December and November are found to be significant at five per cent level. The observed seasonality could be attributed to both the summer holidays and the end of financial year seasonal effects. Irrespective of partially incomplete nature of this research, the findings should be appealing to, among others, researchers, all levels of Government, construction industry and banking industry.


Introduction to Property Sector
This paper examines the impact of seasonal influences on Australian housing approvals, represented by the State of Victoria BANHs. All construction works in most countries have to be approved by the relevant government authority before building commences. In particular, BANH statistics are used to inform as to how many residential buildings are in the pipeline, that is, are expected to be constructed in the near future. In Australia, from July 1990, BANHs include all approved new residential building work valued at A$10,000 or more. Because BANHs provide timely estimates of future residential building work, BANHs statistics are viewed by many economic analysts/commentators as a leading indicator of property investment. Due to the importance of property sector to the whole economy, the general level of economic activity and employment, the data are relevance for the residential property sector as well as the property sector as whole. The demand for new housing fluctuates according to a range of determinants such as costs of building materials, finance costs (such as fees and interest rates) home buyer incentive schemes and employment opportunities. While non-residential building approvals can be viewed as an indicator of investment in a region, the local level residential building approvals are commonly viewed as an indicator of the availability of financial resources and commitment to live in an area.
In the years before the global financial crisis (GFC), Australian house prices rose faster than house prices in the United States of America (USA) and the United Kingdom (UK), implying that the then house price is a bubble. 1 House price bubble can be defined as situation when the behaviour of house prices satisfies at least one of the following four conditions: (i) house prices are significantly above its long-term average; (ii) house prices are significantly above comparable property prices in other economies; (iii) house price to house rent ratio is significantly above its long-term average, and (iv) house price to household income ratio is significantly above its long-term average.  of the initial stage of the fiscal strategy was to quickly increase household spending. This was achieved by making two sets of payments directly to households. The following stages of the fiscal strategy were characterised with investment in infrastructure and skill development, aiming to ensure ongoing fiscal stimulus once the initial boost from the payments directly to households abates. in explaining business cycles. 4 Typically, reducing housing prices tend to impose additional pressure on the banking sector. This happens not only because of increases in bad debts for mortgage loans, but also because of deterioration in the balance sheets of corporate borrowers that rely on property as collateral. Not surprisingly, fluctuation of housing prices and the extent to which they interact with the financial sector and the whole economy are very much of interest, among others, to the Government, the reserve bank and other financial regulators.
The conventional literature recognises for a long time that investment in housing and consumer durables lead non-residential business fixed investment over the business cycle (e.g. Burns and Mitchell 1946). Among others, this is corroborated by Fisher (2006), who observed that in seven of the ten post-war recessions in the USA, household investment achieved its peak and trough before business investment. Ball and Wood (1999) conducted comparative structural time series analysis of housing investment in advanced world economies. They looked at the impact of housing investment on the economy and concluded that housing investment fluctuations after 1960s become a destabilizing factor. This finding highlighted the significance of this category of investment and further accentuated the relevance of studies focusing on better understanding housing investment volatility.
As previously stipulated, the focus of this study is to examine the impact of seasonal influences on Australian housing approvals (represented by Victorian BANHs). 5 Victoria is geographically the second smallest state in Australia. It is also the second most populous state in Australia. Victoria has been selected as a test case because of its geographical homogeneity and economic relevance. Victoria is Australia's most urbanized state: nearly 90 per cent of residents living in cities and towns, it is the most densely populated state (22 people on square kilometre), and has a highly centralised population, with almost 75 per cent of Victorians living in the state capital and largest city, Melbourne. At the same time, the state of Victoria is the second largest economy in Australia, only after New South Wales, accounting for almost a quarter of the nation's gross domestic product (GDP). 6 All other Australian states are either geographically dispersed (cover a wide geographical region across different time and climatic zones) or economically insignificant. 7 This is very relevant to the focus of this research, as seasonal impacts would be much harder to isolate if the data include different time and climatic zones.
Furthermore, it is specifically important to note that the focus of the study is not on modelling the behaviour of time series in terms of explanatory variables (the conventional modelling approach). The conventional modelling approach assumes that the behaviour of the trend and seasonality can be effectively captured by a conventional regression equation that assumes deterministic trend and seasonality. Instead, the aim is to use a univariate structural time series modelling approach (allows modelling both stochastic and deterministic trend and seasonality) and show that conventional assumptions of deterministic trend and seasonality are not always applicable.
Specifically, the study seeks to cast some additional light on BANHs by: (i) establishing the presence, or otherwise, of seasonality in Victorian BANHs, (ii) if present, ascertaining is it deterministic or stochastic, (iii) determining out of sample forecasting capabilities of the considered modelling specifications and (iv) speculating on possible interpretation of results. To do so the study utilises a basic structural time series model of Harwey (1989). Compared to the conventional procedure, Harvey's (1989) structural time series model involves an explicit modelling of seasonality as an unobserved component.
Empirical evidence of seasonal variations in property related variables is relatively limited. Studies come from a range of different perspectives and employ a number of modelling techniques. Harris (1989) provided empirical evidence of strong second and third quarter seasonality in the USA house prices. Ma and Goebel (1991) established the presence of January seasonal effect for securitised mortgage markets, while Friday and Peterson (1997) and Colwell and Park (1990)  The paper is organised in 4 sections. The following section (section 2) of this paper outlines the methodology used. Section 3 provides data specification, presents modelling test results and interprets the modelling results. Finally, in section 4, the paper concludes.

Methodology
A structural time series framework approach used in this paper is in line with that promulgated by Harvey (1989). Such models can be interpreted as regressions on functions of time in which the parameters are time-varying. This makes the approach a natural vehicle for handling changing seasonality of a complex form. Once a suitable model has been fitted, the seasonal component can be extracted by a smoothing algorithm. Following Harvey (1989) and Harvey, Koopman and Riani (1997), the basic structural time series model is formulated in terms of a trend, seasonal and irregular components. All are assumed to be stochastic and driven by serially independent Gaussian disturbances that are mutually independent. If there are s seasons in the year, the model is where the trend, seasonal and irregular are denoted by t µ , t γ and t ε , respectively. The trend is specified as follows: µ is the level and t β is the slope. The disturbances t η and t ς are assumed to be mutually independent. Setting 0 2 = η σ gives a trend that is relatively smooth.
where ( ) . If the Kalman filter is initiated with a diffuse prior, as shown by De Jong (1991), an estimator of the state with a proper prior is effectively constructed from the first 1 + s observations.
On the other hand, if we choose to fix the seasonal pattern in (1), thus specifying a deterministic seasonal component, t γ , may be modelled as: Where s is the number of seasons and the dummy variable jt z is one in season j and zero otherwise. In order not to confound trend with seasonality, the coefficients, j γ , are constrained to sum to zero. The seasonal pattern may be allowed to change over time by letting the coefficients evolve as random walks as in Harrison and Stevens (1976). If t γ denotes the effect of season j at time t, then ( ) Although all s seasonal components are continually evolving, only one affects the observations at any particular point in time, that is jt t γ γ = when season j is prevailing at time t. The requirement that the seasonal components evolve in such a way that they always sum to zero is enforced by the restriction that the disturbances sum to zero at each point in time. This restriction is implemented by the correlation structure in ( ) ( ) , coupled with initial conditions requiring that the seasonal sum to zero at 0 = t . It can be seen from the equation above that In the basic structural model, t µ in (1) is the local linear trend of (2), the irregular component, t ε , is assumed to be random, and the disturbances in all three components are taken to be mutually uncorrelated. The signal-noise ratio associated with the seasonal, that is , 2 2 ε ω ω σ σ = q determines how rapidly the seasonal changes relative to the irregular.
An example of how the basic structural model successfully captures changing seasonality can be found in the study of alcoholic beverages by Lenten and Moosa (1999).

Modelling Results and Their Interpretation
The data are sourced from the ABS. For consistency, the sample for each variable is standardised to start with the first available July observation and end with the latest available June observation. The structural time series model represented by (1)   With respect to the goodness of fit, the modelling results are well defined. Overall, the diagnostic tests are also predominately passed. The only exception is the test for serial correlation (Q) which is slightly above the statistically acceptable level.   Table 1, factors corresponding to June ( 1 γ ), April ( 3 γ ), December ( 7 γ ) and November ( 8 γ ) are found to be significant at five per cent level. 9 The primary cause of this seasonal effect is that the Australian Taxation Office (ATO) accepts prepayment of the interest payable on investment properties. 11 In order to test the robustness of the models specified as well as to determine forecasting power of the three models considered, out-of-sample forecasting was undertaken. Firstly, the three models are estimated over the period 2000:6 -2005:5.
10According to Karamujic (2009) Figure 3, shows that in all cases, the variability in the actual data was difficult to predict with the exception of the specifications including the fixed seasonals. This is corroborated again in Table 2, which reports on the following two statistics that measure the forecasting power: the sum of absolute forecasting errors and the sum of squared forecasting errors.