497
Views
32
CrossRef citations to date
0
Altmetric
PAPERS

Financial Crisis and Asian Real Estate Securities Market Interdependence: Some Additional Evidence

Pages 127-155 | Received 13 May 2008, Accepted 05 Nov 2008, Published online: 17 Dec 2008
 

Abstract

This paper investigates empirically the changes in long‐run relationship and short‐term linkage among the US, UK and eight Asian real estate securities markets before, during, and after the 1997–1998 Asian financial crisis as well as in the most recent period. Using a combination of Johansen linear cointegration, Bierens nonlinear cointegration, Granger causality tests, variance decomposition analysis and volatility spillover methodology, our results indicate that the degree of market interdependence in Asian real estate securities markets appears to have become stronger in the long run and short term since the Asian financial crisis. Further, this market interdependence seems to be on a rising trend ten years after the Asian financial crisis. This stronger market relationship between the Asian and US markets implies a portfolio combination of these markets is less likely to provide diversification benefit in the form of minimum risk. One important lesson to learn from our study is that portfolio managers should constantly review their international diversification models and strategies with respect to the constituent markets because of possible changes in market interdependence triggered by a major crisis.

Acknowledgement

The excellent research assistance provided by Ng Wee Hock is acknowledged.

Notes

1. See Section 2 below for a review on some studies in this area.

2. Following the finance literature, we interpret a greater degree of co‐movement in return and volatility as well as increased correlation between markets to reflect greater market interdependence.

3. In contrast, our literature review reveals there is abundant stock market work that examines the issue of market integration/segmentation in the context of Asian financial crisis using different market samples and different econometric methodologies over different sample periods. Examples of such work include Masih and Masih (Citation1999, 2001), Roca and Selvanathan (Citation2001), Chen et al. (Citation2002), Sheng and Tu (2002), Ratanapakorn and Sharma (Citation2002), Yang et al. (Citation2003), Narayan et al. (Citation2004), Wong et al. (Citation2004), Click and Plummer (Citation2005), Brailsford et al. (Citation2006), Cheng and Glascock (Citation2006) and Choudhry et al. (Citation2007).

4. Other non‐real estate empirical studies on non‐linear cointegration include Coakley and Furtes (Citation2001), Kanas (Citation2003), Chang et al. (Citation2005) and Davradakis (Citation2005).

5. To a certain extent, the countries' indexes are probably not strictly comparable because they have different types of constituent securities such as construction stocks; however, this is the only available international dataset that includes almost all Asian developing real estate markets. On further reflection, although the types of real‐estate related stocks might probably affect the risk‐return profile of the markets, this should probably have minimal impact on the market interdependence results based on a larger dataset.

6. It is probably debatable whether not adjusting for currency differences is appropriate.

7. Singapore and Indonesia are excluded from this period due to insufficient data.

8. Otero and Smith (Citation2000), however, offer some guidance by showing that higher frequency data do not improve the performance of cointegration tests whereas longer observation periods are important. One possible approach is to replicate the Monte Carlo simulation of Otero and Smith (Citation2000) using random variables to see if the cointegration methods are able to detect the relationship over the shorter 18‐month period. Basically, the idea is to create a 54‐month dataset of two variables that are cointegrated by construction. Then, test for cointegration over those series using the relevant statistics. If the cointegration over that interval is detected, move to the next shorter window (say 27 months, non‐overlapping intervals), then repeat for three non‐overlapping 18‐month intervals to check if the test statistics can detect the cointegration in each non‐overlapping window.

9. We wish to thank the two anonymous referees for raising this question. We are sorry for not being able to answer the question as to whether it is possible to have a long‐run relationship in a dataset that is only 1½ years in length. However, we sincerely hope that our responses are acceptable to the two referees that our study, which is consistent with the stock market literature, hopes to provide comparative evidence in international real estate regarding the influence of Asian financial crisis on market interdependence.

10. Before testing whether the price series are cointegrated, we first check that each univariate series for the four sample periods is non‐stationary, or I (1). The tests are necessary, as the finding of a unit root in any of the series indicates non‐stationarity, which has implications for modelling the relationship between any of the two series (bivariate) and all the series in the system (multivariate). Two standard procedures, the augmented Dickey–Fuller (ADF) test and the Phillips–Perron (PP) test, are applied to check the non‐stationarity of each individual series. The unit root test results (not included in order to conserve space) show that there is a unit root in each of the real estate prices in all four sample periods, but no unit root in their first logarithm differences at the 5% significance level.

11. Cointegration is indicated if the rank of r is between 0 and n (number of real estate price series). If n – r = n (r = 0), there are no stationary long‐run relationships among the variables; if n‐r >1 (reduced rank) this implies the existence of more than one common stochastic trend. Under this situation, while long‐run integration is not complete, the convergence process is underway with the number of independent stochastic trends reflecting the extent of the convergence and any diversification benefits. Finally, there is a single common stochastic trend and hence a single permanent force that creates the non‐stationary property of the data if n‐r = 1 (r = n‐1). The implication is that over long horizons national equity prices will be perfectly correlated.

12. According to Johansen and Juselius (Citation1990), the cointegration test results are stronger and more robust when there is more than one significant cointegrating vector.

13. As the constituent markets are operating at different time zones with different opening and closing times, testing for causality relationships between the US and the Asian markets presents a problem of data synchronization due to time zone shift differences. It is thus important to pair the closing market at time t + 1 with the later closing market at time t when testing for the causality relationships. For example, when testing whether the Singapore market Granger causes the US market, we set the Singapore market at t +1 and the US market at time t. The same adjustment method is adopted when testing for causality relationships between other markets.

14. Similar to the Granger causality analysis, a VAR model is appropriate for the pre‐crisis period while VECM methodology is employed for the crisis, post‐crisis and most recent periods. Orthogonalization is achieved by Choleski decomposition where the order of the variables is: US, UK, China, Hong Kong, Japan, Indonesia, Malaysia, Philippines, Singapore and Thailand.

15. The detailed variance decomposition results for the four studied periods are not included in order to conserve space.

16. Average ‘degree of exogeity’ is 70.38% (‘crisis’), 74.51% (‘post‐crisis’) and 63.73% (‘most recent’).

17. The ijth entry in the table is the estimated contribution to the forecast error variance of country i coming from innovations from country j. Therefore the ‘contribution to others’ (off‐diagonal column sums) or ‘contributions from others’ (off‐diagonal row sums), when total across countries, give the numerator of the spillover index. The denominator of the spillover index is given by the column sums or row sums (including diagonals) and totalled across countries.

18. We also compare the 30‐day ahead volatility forecast with 60‐day and 90‐day ahead forecast and find that the spillover estimates are of a similar magnitude. Hence the shocks from the volatility transmission are short‐lived. In addition, volatility spillover index is the highest for the ‘most recent’ period where the cointegration relationship is also the strongest.

Log in via your institution

Log in to Taylor & Francis Online

PDF download + Online access

  • 48 hours access to article PDF & online version
  • Article PDF can be downloaded
  • Article PDF can be printed
USD 53.00 Add to cart

Issue Purchase

  • 30 days online access to complete issue
  • Article PDFs can be downloaded
  • Article PDFs can be printed
USD 587.00 Add to cart

* Local tax will be added as applicable

Related Research

People also read lists articles that other readers of this article have read.

Recommended articles lists articles that we recommend and is powered by our AI driven recommendation engine.

Cited by lists all citing articles based on Crossref citations.
Articles with the Crossref icon will open in a new tab.