Forecasts and implications of the current housing crisis: switching regimes in a threshold framework

Abstract

This article aims to explore the implications of the recent housing crisis from a forward-looking perspective in the form of the two-regime switching phenomena under a Threshold Autoregressive (TAR) model. Three categories of thresholds, which are housing prices, housing volumes and price-to-income ratios, are adopted in an attempt to examine the US state-level housing market cycles in multiple dimensions. In general, while lagged differenced thresholds fail to capture the recent housing boom-and-bust regime switching, moving-average thresholds succeed in signalling a housing crisis in many states. The regime-switching autoregressive structures show that housing price cycle has remarkable series dependence, and the housing volume cycle tends to show a mean-reversion pattern. Most importantly, only four states, Arizona, California, Florida and Nevada, display the bust regime of housing price growth during 2006 to 2010. The price-over-income also indicates that less than five states have significant two switching regimes. Otherwise, most states exhibit low-growth regime of housing permits, implying a nationwide housing volume cycle during the current housing crisis. The forecasting system provides a helpful guideline for policy-making of the government as well as household decisions of consumption and investment.

Keywords:

• threshold autoregressive model
• housing crisis
• regime switching
• housing price cycle
• housing volume cycle

JEL Classification:

• C53
• E37

Notes

¹ As introduced in Franses and van Dijk (2000), the TAR model and the Markov switching model are regarded as the primary two families of regime-switching models which are used to capture the nonlinear behaviour of time series.

² The literature on exchange rate market includes Basci and Caner (2005), Sollis and Wohar (2006), Leon and Hajarian (2005) and Cerrato et al. (2010). The example study on the stock market includes McMillan (2001).

³ There are other variables which are used as the housing market fundamentals. Aside from personal income, GDP, employment, housing rent, population, construction cost, stock market wealth and interest rates are all important measures of fair values of houses. For example, Mikhed and Zemcik (2009) use various fundamental variables in their analysis of the recent US housing bubble.

⁴ Because we need to incorporate the time series lagged up to three periods to establish the thresholds, the analysed period lags the data period by three quarters.

⁵ The levels of price-over-income are also tried in this study, but the results fail to deliver useful information of housing market dynamics. Otherwise, the changes in the ratios, including CMHPI/income and FHFA/income, display important implications of a housing crisis for some state-level housing markets. Thus, this study uses the change, not the level, of the ratio as the threshold indicator.

⁶ They include California, District of Columbia, Louisiana, Mississippi, North Dakota, New York, South Dakota, Washington, West Virginia and Wyoming.