Essays on Model Uncertainty and Real Estate Markets

Xiao, Hui
Journal Title
Journal ISSN
Volume Title
University of Guelph

Chapter 1 focuses on model selection and model averaging, both of which are approaches for handling modelling uncertainties. I aim to supplement the literature by studying the class of OLS post-selection estimators. Inspired by the shrinkage averaging estimator (SAE) and the Mallows model averaging (MMA) criterion, I further propose a shrinkage MMA (SMMA) estimator for averaging high-dimensional sparse models. The Monte Carlo design features an expanding sparse parameter space and further considers the effect of the effective sample size and the degree of model sparsity on estimators' finite sample performances. I find that the SMMA outperforms when averaging high-dimensional sparse models. In Chapter 2, the conventional perfect competition model is inadequate for the heterogeneous, illiquid, and decentralized housing market, which clears via multiple local time-varying equilibria. I first propose a spatial search model that caters to such market characteristics and provides theoretical micro-foundations to motivate the econometric model. Then, I introduce a nonlinear spatiotemporal autoregressive model with autoregressive disturbances (NLSTARAR) and augmented by local time-varying factors to unify the current hedonic pricing framework and uncover the real estate market structure by simultaneously identifying the spatiotemporal structure of the market's spatial dependence and its interaction with the housing market microstructure. To address model uncertainty, I propose both model selection and model averaging estimation strategies. Chapter 3 applies the methodologies developed in Chapter 2 to study the Greater Toronto Area (GTA) real estate market using a unique GTA dataset. The NLSTARAR model captures the effects of the local time-varying market microstructure besides the hedonic, demographic, and policy effects on the housing market. By model selection, I show that the real estate pricing is driven by a local time-varying market structure that effectively responds to the heterogeneity in assets consistent with existing theories. The local time-varying market microstructure dominates the spatial spillover effects with unexpected market shocks generating the market volatility. I further employ the rolling window approach to show that the uncovered real estate market structure captures the shifts in the market state, evolves as a market pricing mechanism, and better forecasts the real estate market out-of-sample.

real estate, hedonic pricing, cross-sectional dependence, nonlinear regression, model averaging, model selection, tuning parameter choice