Hot and Cold Housing Markets

Hot and Cold Housing Markets: International Evidence
Jose Ceron and Javier Suarez

This paper examines the experience of fourteen developed countries for which there are about thirty years of quarterly inflation-adjusted housing price data. Price dynamics is modeled as a combination of a country-specific component and a cyclical component. The cyclical component is a two-state Markov switching process with parameters common to all countries. We find that the latent state variable captures previously undocumented changes in the volatility of real housing price increases. These volatility phases are quite persistent (about six years, on average) and occur with about the same unconditional frequency over time. In line with previous studies, the mean of real housing price increases can be predicted to be larger when lagged values of those increases are large, real GDP growth is high, unemployment falls, and interest rates are low or have declined. Our findings have important implications for risk management in regard to residential property markets.


The estimation yields three set of results. First, the dynamics of real housing prices is characterized by two rather persistent states that mostly differ in the volatility of price increases. Specifically, the variance of the unpredictable part of quarterly price increases in the high volatility state is almost four times as big as in the low volatility state. The low volatility state is associated with phases of higher growth, occurs with an unconditional probability of 47% and has an expected duration of 23 quarters. The high volatility state is associated with phases of lower price growth, occurs with a frequency of 53% and lasts, on average, about 26 quarters. Second, in addition to the latent state variable, a number of lagged macroeconomic variables have significant predictive power for the expected growth rate of real housing prices. Specifically, the prediction of quarterly growth rates depends positively on the lagged quarterly rate of real GDP growth, negatively on the lagged one-year variation in the unemployment rate and also negatively on the lagged long-term nominal interest rate. We find no evidence of the effect of these variables to be state dependent. Third, even after controlling for the effect of the latent state variable and the explanatory variables, the quarterly growth rate of real housing prices exhibits significant positive autocorrelation.



A problem with the multi-country approach is heterogeneity. Geographical, historical and institutional factors may make residential property prices evolve differently in different countries. Perhaps even simple methodological differences in the construction of each country’s price indices may make them show systematic differences in mean or variance. Thus, without properly controlling for the underlying heterogeneity, a regime-switching model estimated with pooled multi-country data, instead of capturing the common structure of the cyclical pattern, might end up associating the latent states with some rather cross-sectional partition of the data.