An investigation of regime shifts in UK commercial property returns: a time series analysis

Abstract

The random-walk hypothesis, vis-à-vis asset price, suggests that prices traded in a market cannot be predicted based on historical information. Employing unsecuritized UK commercial property returns, we analyse this hypothesis by investigating regime shifts or multiple changes in persistence in the series. Our results uncover regime shifts in both the aggregate and sector-specific data. Specifically, the shifts are less frequent in the Industrial sector, compared to the Office, Retail and Aggregate returns data. We highlight some implications for academics, practitioners and regulators.

Keywords:

• commercial real-estate
• property returns
• multiple changes in persistence
• fractional integration

JEL Classification:

• C22
• G11

Notes

¹ See among others, Cuñado et al. (2005) (NASDAQ composite dividend–price ratio); Pesaran et al. (2006) (US Treasury bills); Sollis (2006) and Navarro (2009) (the S&P composite dividend yield); Noriega and Ramos-Francia (2009) (US inflation rates) and Homm and Breitung (2012) (US stock market).

² For example, DTZ (2015), a global real estate adviser, estimates the current stock value invested globally in commercial real estate to be around USD13.6 trillion by the end of 2014.

³ For a good review of commercial real estate return distribution, see Lizieri and Ward (2000). Other related studies including Serrano and Hoesli (2012), Rehring and Sebastian (2011), Macgregor and Schwann (2003), Brown (2001) Lee and Ward (2000) investigate issues related to volatility, serial correlation, fractional cointegration in both securitized and unsecuritized commercial real estate indices, returns and prices for the UK and US markets.

⁴ IPD (2014).

⁵ The shape of the distribution of returns can vary with market conditions, e.g., when the markets suffered a major adjustment after the market crash in October 1987, this caused returns to be negatively skewed. This might also be captured by the real estate returns data on analyses, especially during the recent 2007–2009 financial crisis. Positive kurtosis suggests that probabilities of obtaining extreme values are higher than implied by the normal distribution. This could be a reflection of reality of the marketplace when large market surprises may tend to induce large movements in the markets and in property values.

⁶ Examples include Filardo and Gordon (1994), who use a Markov-switching model that allows for business cycle comovement to vary with the phase of the cycle. Gregory et al. (1997) use dynamic factor analyses for G7 countries, to establish that the importance of the common factor varies over the business cycle. Furthermore, Engle (2002) also proposes a class of multivariate models – dynamic conditional correlation models – which may be employed to assess comovement. We thank an anonymous referee for the suggestion.

⁷ Up to one-degree of freedom correction, the average of the comovements in each time period, as given by Equation 1, will be equal the Pearson correlation coefficient, i.e., $\frac{1}{T}\sum _{t=1}^T{\rho }_t^{ij}=\frac{T-1}{T}{\rho }^{ij}$, where the latter follows a Student’s t-distribution.

⁸ For robustness, we also applied more recent unit-roots tests including procedures developed by Elliot et al. (1996) and Ng and Perron (2001) on the individual series, and the I(0) condition is also confirmed (results available on request). We thank an anonymous referee for the suggestion.

⁹ The GPH estimator is inconsistent against d  > 1 alternatives. Hence, practically, under those circumstances, distinguishing unit-root behaviour from fractional integration may be problematic.

¹⁰ Phillips suggests removal of the deterministic trends from the series before application of the estimator. The test is performed using the data analysis and statistical software command ‘modlpr’. See Phillips (1999a, b) for a more detailed description.

¹¹ We note that the period between the end point of one I(0) regime and the start point of the next I(0) regime must represent an I(1) regime. See Table 6.

¹² We also perform the LKT tests, with a trend, as a robustness check. The results obtained are similar to the estimations with a linear trend. For parsimony and due to the similarities to dates reported, these estimates are not reported here, but are available upon request.