Forecasting House Prices in the United States with Multiple Structural Breaks

The boom-bust cycle in U.S. house prices has been a fundamental determinant of the recent financial crisis leading up to the Great Recession. The risky financial innovations in the housing market prior to the recent crisis fueled the speculative housing boom. In this backdrop, the main objectives of this empirical study are to i) detect the possibility of multiple structural breaks in the US house price data during 1995-2010, exhibiting very sharp upturns and downturns; ii) endogenously determine the break points and iii) conduct house price forecasting exercises to see how models with structural breaks fare with competing time series models – linear and nonlinear. Using a very general methodology (Bai-Perron, 1998, 2003), we found four break points in the trend in the S&P/Case-Shiller 10 city aggregate house-price index series. Next, we compared the forecasting performance of the model with structural breaks to four competing models – namely, Random Acceleration (RA), Autoregressive Moving Average (ARMA), Self- Exciting Threshold Autoregressive (SETAR), and Smooth Transition Autoregressive (STAR). Our findings suggest that house price series not only has undergone structural changes but also regime shifts. Hence, forecasting models that assume constant coefficients such as ARMA may not accurately capture house price dynamics.