| In this dissertation we explore the econometric modeling of interest rate and housing-mortgage market and develop Bayesian estimation methods. To analyze mortgage prepayments, the first essay develops a Threshold Switching model which allows for heterogeneity in threshold regimes by introducing random effects to the threshold parameter. The Threshold Switching model has been widely used in time series econometrics. Our extension makes this model applicable to longitudinal data where different subjects may present different threshold values. A Gibbs sampler is developed for identifying the unknown threshold and for estimating the random effects.; The second essay examines the spatial dependence in metropolitan housing price growth rates. We incorporate spatial dependence in time series processes and, to quantify the neighbor relation, introduce a multi-dimensional metric of economic distance between two geographic areas. Bayesian model selection is performed among univariate AR(1) model, spatial-temporal autoregressive (STAR) model based on the conventional physical contiguity relation, and STAR model based on the economic distance. We find that the estimation of metropolitan housing price is greatly improved by modeling spatial dependence together with temporal dependence and that the proposed metric of economic distance is superior to the physical contiguity in measuring neighbor relation.; The third essay develops Bayesian estimation methods of short-term interest rate models. For the model with white noise errors, we obtain the Bayesian inference of the parameters and compare them with MLE and GMM estimators. We further consider an interest rate model with ARMA (p, q)-GARCH (r, s) errors and derive a Markov Chain Monte Carlo method with Metropolis-Hastings algorithm for parameter estimation. As illustration, we estimate the models of Japanese overnight call rate. |
