RDP 2019-01: A Model of the Australian Housing Market 2. Previous Research

Most empirical research into the Australian housing market has focused on individual equations. For example, Bourassa and Hendershott (1995), Abelson et al (2005) and Tumbarello and Wang (2010) examine housing prices; McLaughlin (2011) and Gitelman and Otto (2012) examine housing construction. The IMF (2018) estimate and discuss equations for both variables separately. Our approach differs in that we put these and other equations together. We also re-examine this earlier work in the light of recent research and data.

Another strand of research has examined the housing market as part of structural macroeconometric models. Examples include Jovanoski, Stoney and Downes (1997), Powell and Murphy (1997) or the Reserve Bank of Australia's (RBA's) new MARTIN model (Cusbert and Kendall 2018). Our approach differs in that we look at the housing market in more detail, including variables that these models often exclude, such as building approvals, completions or the vacancy rate. The macroeconomic models are designed to settle down to an explicit steady state with simple properties, such as constant relative prices or expenditure shares. That facilitates the models' application to a wide range of questions. However, in discussing housing-specific issues, it is not desirable to have important results driven by assumptions for which evidence is weak. We do not constrain the steady state of our model except when the data suggest this is realistic.

These alternative approaches have advantages relative to our approach. Larger models allow more variables to be endogenous, allow for feedback and can answer a wider range of questions. Smaller models are less reliant on chains of causation, which can be as fragile as their weakest link. And more focused models allow examination of specific estimates in more detail. These different approaches are complementary. When a range of different approaches support similar results, we have more confidence in the conclusions. Accordingly, we discuss specific points of agreement and disagreement where they arise in the discussion.

In common with most previous Australian research on housing markets, and most macroeconomic forecasting, we focus on single-equation least squares estimates using aggregate quarterly data. Identification is typically through lags and a priori reasoning. For example, rents are explained by lagged vacancies and contemporaneous income, on the assumption that the right-hand side variables are weakly exogenous. Given that the future does not determine the past, that strikes us as plausible for most lagged variables. This argument is less compelling when the right-hand side variables are forward-looking, such as investment or asset prices. However, we are not aware of clear evidence or strong arguments that weak exogeneity fails in the relevant equations or of useful instruments that might rectify this. One alternative to our approach would be to assume model-consistent expectations, but that seems unrealistic. We recognise that finding ‘X regularly precedes Y’ does not prove that ‘X causes Y’. However, in the absence of reasonable arguments to the contrary, the latter statement seems a natural hypothesis to maintain pending more definitive tests. This approach – essentially that of structural macroeconometric modelling – rests on methodological assumptions that are controversial, but that debate is outside the scope of this paper.