| The two essays contained in this dissertation focus on different aspects of the real estate market: housing market cycles and modeling mortgage prepayment regimes. The first essay entitled ‘Estimating the Housing Market Cycles’ describes a method for estimating the residential real estate cycle by relying on the cointegration between population and the total housing stock. The model applies the Beveridge-Nelson (1981) decomposition to the housing stock to estimate its permanent and transitory components. Using this analysis, I am able to show that the long-term housing market cycle peaked just before the recessions in the early 80s and has remained above trend during the 90s. This model is intended to help us better understand the typical boom-bust cycle experienced in residential real estate markets and identify periods during which the housing stock is overbuilt relative to population. The second essay, entitled ‘Modeling Mortgage Prepayment Regimes’ explores the efficacy of modeling mortgage prepayment using either unobserved or observed regime models. The competition that I have set up pits a 2-State, Markov-Switching with Time Varying Transition Probability model against three observed regime models: Logit, LSTAR, and Neural Net. To compare the efficacy of either approach, I compare the models' one step-ahead, out-of-sample forecasts. The results indicate that the observed regime models, particularly the LSTAR and the Neural Net, do a better job of modeling prepayment regimes than the unobserved regime model. I conclude that the process of securitization, which sufficiently homogenizes the MBS pool, makes the ‘blunt,’ unobserved prepayment regime model inappropriate for modeling prepayment phenomenon. |
