Date of Award

2026

Keywords

wavelet analysis, empirical orthogonal function, spatiotemporal model, areal data, clusters, spatially varying covariates, Bayesian inference

Document Type

Thesis

Publisher

Edith Cowan University

Degree Name

Doctor of Philosophy

School

School of Science

First Supervisor

Ebenezer Afrifa-Yamoah

Second Supervisor

Johnny Lo

Third Supervisor

Steven Richardson

Fourth Supervisor

Ute Mueller

Abstract

This thesis conducts a comprehensive spatiotemporal analysis of housing prices across all eight Australian capital cities, with subsequent in-depth modelling focused on Sydney, Perth, and Canberra. While the first exploratory analysis considers all capitals to capture broad national housing dynamics, the restriction in scope to these three cities was occasioned by their distinctive economic and demographic characteristics, as well as the difficulty of obtaining comparable suburban-level housing data for the remaining capital cities. Drawing on advanced econometric, statistical, and spatiotemporal modelling techniques, the study presents three major findings arising from two exploratory analyses and a proposed Bayesian spatiotemporal modelling framework.

The first component investigates the long-term dynamics and short-term cyclical behaviour of housing prices for all eight Australian capital cities using wavelet-based methods applied to quarterly median prices from 1980 to 2023. Wavelet transforms are applied to decompose the temporal structure of price series, while wavelet coherence is used to examine interdependencies and evolving lead–lag relationships between cities. Information transmission mechanisms are further analysed using transfer entropy. The results reveal pronounced high-frequency variability across all cities, with Melbourne exhibiting the greatest volatility. Short-term fluctuations are particularly high in Perth and Brisbane, reflecting notable cyclical patterns. Coherence analyses reveal dynamic, time-varying interdependencies with leading roles alternating among cities, whereas transfer entropy reveals asymmetric, bidirectional information flows that highlight the complex interaction structure underpinning the national housing market.

The second component applies the rotated empirical orthogonal function (REOF) method to identify dominant spatiotemporal modes underlying suburban housing prices in Sydney, Perth, and Canberra using quarterly data from 2019 to 2024. REOF analysis is complemented with Moran’s I hotspot analysis to detect significant spatial clusters of elevated prices, and generalised additive model (GAM) is implemented to relate the extracted modes to socioeconomic and macroeconomic drivers. The dominant REOF mode explains approximately 80% of the total price variability, with persistent high-value clusters observed near central business districts and coastal areas across all three cities. GAM estimation reveals distinct city-specific sensitivities: unemployment shows strong nonlinear negative effects in Sydney and Perth, whereas Canberra’s market responds primarily to credit conditions, particularly mortgage interest rates.

The third component develops a Bayesian hierarchical spatiotemporal framework integrating spatial clustering with conditional autoregressive (CAR) modelling to analyse suburban housing prices in Sydney, Perth, and Canberra. Data-driven Voronoi tessellation clustering is constructed using spatial proximity and price evolution, and a novel specification of spatially varying covariate (SVC) effects is introduced to capture intra-cluster heterogeneity. This modelling strategy addresses key challenges including high dimensionality, spatial misalignment between suburb-level prices and coarse covariates, and irregular areal geometries. Comparative model assessment reveals the BYM2 specification of Riebler et al. (2016), augmented with cluster-wise SVC effects, consistently outperforms the Leroux CAR model and alternative approaches such as spatiotemporal autoregressive (STAR) and geographically and temporally weighted regression (GTWR), based on predictive accuracy metrics including RMSE, MAE, and MAPE. Population density exhibits consistently positive effects on housing prices, while mortgage interest rates exert a negative influence. Predictive maps reveal close alignment with observed price distributions, capturing persistent disparities between inner-city and outer suburbs as well as emerging growth in fringe areas. The model revealed out-of-sample MAPE of 8.5% which compares favourably with published housing price forecasting studies.

While the proposed framework advances spatial econometric modelling by effectively accommodating complex spatial structures and quantifying uncertainty, limitations associated with the resolution of covariates and the relatively short temporal window place some constraints on its immediate implications for policy design and urban planning.

Access Note

Access to this thesis is embargoed until 28th July 2028

Access to chapters 4, 5 & 6 of this thesis is not available 

Download

Available for download on Friday, July 28, 2028

Share

 
COinS
 

Link to publisher version (DOI)

10.25958/n0zd-hh49