The US Housing Market: Asset Pricing Forecasts Using Time Varying Coefficients

Abstract

The US housing market has experienced significant cyclical volatility over the last twenty-five years due to major structural changes and economic fluctuations. In addition, the housing market is generally considered to be weak form inefficient. Houses are relatively illiquid, exceptionally heterogeneous, and are associated with large transactions costs. As such, past research has shown that it is possible to predict, at least partially, the time path of housing prices. The ability to predict housing prices is important such that investors can make better asset allocation decisions, including the pricing and underwriting of mortgages. Most of the prior studies examining the US housing market have employed constant coefficient approaches to forecast house price movements. However, this approach is not optimal as an examination of data reveals substantial sub-sample parameter instability. To account for the parameter instability, we employ alternative estimation methodologies where the estimated parameters are allowed to vary over time. The results provide strong empirical evidence in favor of utilizing the rolling Generalized Autoregressive Conditional Heteroskedastic (GARCH) Model and the Kalman Filter with an Autoregressive Presentation (KAR) for the parameters’ time variation. Lastly, we provide out-of-sample forecasts and demonstrate the precision of our approach.