Predicting House Price Bubbles in the United States

This study investigates new methods to predict house price bubbles in the United States. Using data from 1987 Q1 to 2023 Q2, the methods of Phillips, Shi and Yu (2015) [PSY] and Harvey, Leybourne and Whitehouse (2020) [HLW] are used to identify the timing of bubble episodes and construct a housing bubble indicator for the United States.
Both the PSY and HLW methods identify two episodes of housing market bubbles: one in the 2000s and another in the 2020s. Employing a binary bubble indicator as a dependent variable, this study investigates the relationship between housing bubbles and a number of macroeconomic and financial determinants using a Linear Probability Model as well as Probit and Logit estimation. We find evidence that housing market conditions, monetary policy, and stock market returns provide significant information on the probability of a housing bubble occurring. However, macroeconomic indicators, such as inflation, unemployment, and GDP growth, are not significant explanatory variables for housing bubbles across the full sample period.
In addition, subsample analysis reveals a structural break in the relationship between explanatory variables and the housing bubble indicator before and after the 2008 financial crisis. Moreover, extending the baseline model to include higher order lagged independent variables shows that the housing market reacts quickly to changes in economic conditions, as there is no improvement in in-sample forecast performance from variables lagged by half a year or more.
The study further assesses the out-of-sample forecasting performance of the three regression models using static, rolling, and recursive forecasts. It finds that static and rolling forecasts are inadequate for predicting housing bubbles, and the recursive model does not outperform a naïve random walk model in terms of forecast accuracy. These findings provide valuable insights into forecasting housing market bubbles and the impact of economic conditions.