Switching volatility and cross-market linkages in public property markets

Abstract

The primary contribution of this study is to examine the changes in cross-market relationship in international public property markets from a volatility regime switching perspective from January 1990 to January 2012. We find that global developed public property markets can be adequately characterised by a SWARCH model. In particular, most of the persistence in real estate stock price volatility can be attributed to the persistence of low-, medium- and high-volatility regimes in international developed public property markets. Moreover, there is a significant volatility increase during the crises periods for all markets examined. However, the identified high-volatility regime appears short-lived. Based on the SWARCH results, we find that the dynamic linkages among the markets are positively dependent on volatility regime. Specifically, the market correlations, foreign market influence, aggregate variance spillover index and variance–covariance matrix have intensified as market volatility increases during this period. Moreover, the evolution of the cross-market linkages among the sample public property markets is influenced significantly by both a time trend and a volatility regime factor that are independent of the influences of the global stock market and national stock markets. Our results imply that risk-reduction via international diversification in public property markets may only hold true in low-volatility periods. Consequently, portfolio managers need to understand and implement volatility state-dependent optimal asset allocation in order to better advise their clients.

Keywords:

• volatility regimes
• correlation
• market linkages
• variance decomposition
• public property markets

View correction statement:

Erratum

Notes

This article was originally published with errors. This version has been corrected. Please see Erratum (http://dx.doi.org/10.1080/09599916.2014.883218).

1. Henceforth, this term ‘public property market’ is used interchangeably with ‘real estate securities ‘market’, ‘securitised property market’ and ‘securitised real estate market’.

2. An alternative is to use a regime-switching GARCH model that advances the SWARCH specification in that it allows the past variance to be conditioned on regime changes. However the SWGARCH model is very difficult for estimation and is sensitive to the dataset. This is because the lagged variance term depends upon the entire prior path of the regimes, thus the likelihood value can only be approximated in some form. As further pointed out by Cai (1994), any data series with a sample size larger than 50 will consequently be difficult to estimate. Moreover direct maximum likelihood estimation via a nonlinear filter is also practically infeasible. Instead a SWARCH specification appears appropriate since the Markov switching process in ARCH equation can deal with the persistence itself, which addresses the concern of lack of memory in traditional ARCH models. In order not to add to the already computationally heavy SWARCH specification, we restrict the dynamic lag structure to an ARCH model and adopt a SWARCH model in the empirical analysis. We wish to thank a reviewer for raising this comment and sincerely hope our response is acceptable.

3. Hamilton and Susmel (1994) presented a three volatility regime model that adequately pictures the stock market in the United States.

4. Instead of using the individual SWARCH estimates to determine a common volatility regime setting for the sample markets, an alternative would be to stack the equations and conduct a vectorised version of this analysis in order to do more formal tests across countries. We fully agree with one reviewer’s comments that although the univariate SWARCH methodology is fine in characterising each country’s volatility behaviour and its persistence, the countries may need to be stacked with the between-country effects tested more formally within the context of more general specification when examining cross-market relationship. This means that we should use a multivariate extension of the SWARCH model to explore co-movements in real estate securities markets’ volatility across markets (Edwards & Susmel, 2001; Ramchand & Susmel, 1998). Since the multivariate SWARCH investigation is expected to be very computationally demanding in view of our SWARCH (3, 2) specification, we have included in a funded research project to explore in-detail the cross-country effects in volatility integration in the context of regime switching. We wish to thank the reviewer for raising this useful suggestion. Finally, 87.8% of all returns observations describe periods during which the majority of the markets are in a particular state. We sincerely hope the univariate methodology (which was also used by Jochum, 2001) and the associated results are acceptable to you and thanks again for your kind understanding on our response to your request.

5. See also Section 4 on a brief discussion regarding the SWARCH methodology.

6. Other stock market studies that have considered a three-regime SWARCH specification include Jochum (2001), Li and Lin (2003), Diamandis (2008) and Wang and Theobald (2008).

7. We estimate an AR (1) – GARCH (1, 1) model to examine whether there are significant ARCH effects in the real estate securities data. The results (not reported in order to conserve space) provide evidence that are significant ARCH effects for all series.

8. For each market, the volatility persistence value (first number) from the AR (1)-GARCH (1, 1) model is compared with that of the SWARCH (3, 2) model (second number). The results are: Australia (.968, .089), France (.964, .218), Germany (.993, .398), Hong Kong (.901, .064), Italy (.908, .286), Japan (.930, .091), Singapore (.940, .248), the UK (.906, .180) and the US (.901, .383). These results are in agreement with Lamoureux and Lastrapes (1990) and Hamilton and Susmel (1994) who argue that the observed high volatility persistence to the conditional variance indicates structural change in the statistical variance generating process. Since we further find the use of a SWARCH (3, 3) model only leads to a small reduction in the ARCH effects in some markets, we choose the optimal SWARCH (3, 2) specification for a parsimonious representation of the different possible regimes.

9. P (i, j) is the probability of moving from i to j.

10. Using the UK as a further illustration, the speculative attack on the pegged exchange rate in the European continent dragged the real estate securities market of the country from low to medium volatility states around 1992–1993. The rise of the smoothed probability for the medium volatility state in 2000 corresponds to the collapse of the dot-com bubble that influenced the public property market albeit at a smaller scale. During 2002–2003, the increase in the smoothed probability for the medium volatility state is attributed mainly to the Iraq War where the UK has played a major role. Then the global financial crisis from 2007 to 2009 and EU sovereign debt crisis from mid-2011 have significantly contributed to ‘high’ volatility phenomenon observed in the stock and public property markets.

11. We are grateful to one reviewer for highlighting this point.

12. Another alternative is to use the nine individual stock returns in lieu of the global stock market return factor. The global stock market is regarded as a good proxy for the underlying domestic stock markets.

13. We wish to thank a reviewer for raising this issue and the related arguments. We have revised the regression to include a weighted stock market liquidity factor (WLIQ). We use a popular measure of stock market liquidity: trading volume (collected from Bloomberg) since the bid-ask spread data (as suggested by the reviewer) are not available. Finally, Italy was not included in the analysis because her stock market trading volume data are not available.