Speculative bubbles, financial crises and convergence in global real estate investment trusts

Abstract

We test for price bubbles in 14 national real estate investment trust (REIT) markets, and measure the degree of their convergence towards a common trend. Our methodology consists of the recently developed test of Phillips, Shi and Yu (2014) for mildly explosive processes, and the Phillips and Sul (2007) method for modelling convergence among random variables. We find evidence of explosive behaviour in index levels of seven of the 14 markets. In contrast, explosive dynamics are found in only one price/dividend ratio. More than half of the episodes of explosive behaviour are date-stamped to periods prior to the 2007–2009 financial crisis. We also discover a number of periods over which the markets converge towards a common trend. Interestingly, all of the convergence intervals coincide either with periods of crisis, or with periods of market exuberance. For instance, evidence of convergence is found during the 2000 dot-com crash, the 2007–2009 subprime crisis and the 2010–2013 European sovereign debt crisis, as well as over the bubble period of 2004–2005.

Keywords:

• financial crisis
• price bubbles
• real estate investment trusts
• co-movement
• REITS

JEL Classification:

• G15
• F36
• C22

Acknowledgements

Australian Research Council Discovery Grant DP120102239.

Notes

¹ Noted by Stephen Schwarzman the chairman and CEO of the Blackstone Group on 11 March 2009.

² Rational bubbles occur when the price investors are willing to pay is higher than what is warranted by the market fundamentals, because they expect to be able to sell the asset at an even higher price in the future.

³ ${\beta }_{it}$ converges in probability to ${\beta }_i$ if, as $t\to \mathrm{\infty }$, the probability that ${\beta }_{it}$ differs from ${\beta }_i$ by more than some number $c$ can be made arbitrarily small for any $c$.

⁴ The function is ‘slowly varying and increasing’ in the sense that its value increases much slower than t.

⁵ Phillips and Sul (2007) show that a small fraction $r$ of the time series data should be discarded in order to emphasize what happens as the sample size gets larger.