Housing market cycles and duration dependence in the United States and Canada

Abstract

Housing wealth is a large component of households’ total wealth and plays an important role in aggregate business cycles. In this article, we explore data on real house price cycles at the aggregate level and city level for the US and Canada. Using a panel of 137 cities, we examine the duration and characteristics of housing market cycles in North America. We find that North American housing cycles are long, averaging 5 years of expansion and 4 years of contraction. We estimate a discrete time survival model with a probit specification for house price expansions and contractions. This model allows us to test for duration dependence. We find that US housing market expansions have positive duration dependence since their exit probabilities increase with duration, while contractions seem to have no duration dependence. Canadian house price cycles did not exhibit duration dependence. Standard determinants of house prices (interest rates, income and population growth) are included as controls.

Acknowledgements

We are grateful to Larry Schembri, Peter Kennedy, Robert Lafrance, Jeanine Bailliu, Don Colleti, Phillip Maier and Marc-Andre Gosselin and seminar participants for helpful comments. We also thank Ryan Felushko and Joan Teske for research assistance.

Notes

¹ The data are for metropolitan statistical areas (metros), but we refer to metros and cities interchangeably. The US population rankings are based on metropolitan statistical areas population data for 2004.

² Duration dependence may also be analysed using Markov switching models as in Durland and McCurdy (1994) and Lam (2004). Markov switching models are popular for analysis of macroeconomic time series. Since our dataset is organized as a panel with a large cross-sectional dimension, we employ a survival model framework for our analysis.

³ Jenkins (2005, p. 21).

⁴ Although many studies of duration dependence only include duration on the right-hand side of the regression, if we omit these fundamental variables then the estimated coefficients on the duration variable may be biased, as noted by Layton and Smith (2005).

⁵ The survival model is estimated separately for contraction and expansions since in a sample with both phases together the disturbance term may be conditionally heteroskedastic. It is not clear if robust estimation would control sufficiently for this. See Ohn et al. (2004, p. 533).

⁶ Although it would have been interesting to include household formation data in our set of covariates, this data is not available by metro level and thus has not been included.

⁷ Further discussion on the selection of dummy dates is provided in the working paper version of this article. Green and Wachter (2005) provide a thorough examination of the evolution of mortgage markets.

⁸ The results did not change substantially if we split the sample into somewhat different time periods. For example, just splitting the sample in 1990 or 1995, and both indicated the most recent decade has substantially different pattern of duration dependence. This is likely driven by the fact that most of the last decade has witnessed a long expansion in most cities. Similar findings to those in Table 10 were found with a time trend, rather than decade dummies.

⁹ See the BLS website at http://www.bls.gov/cpi/cpifaq.htm#Question_16 for further details on coverage, frequency and quality of CPI measures at the regional and metropolitan level.

¹⁰ Aside from interest rate variables, all regressors are metro level specific.

¹¹ Using one or two lags of population growth rather than the contemporaneous value, we find the same signs and the coefficients remain significant.

¹² City-level construction cost data, unfortunately were not available for inclusion in our study. Glaeser and Gyourko (2007) uses construction cost data from a private consulting firm, while the data used in Capozza et al. (2002) do not extend beyond 1995.

¹³ We thank an anonymous referee for suggesting the separate regression.

¹⁴ The literature on housing price bubbles and monetary policy examines this in more detail. See for example Herring and Wachter (2002).