Abstract
This paper explores the empirical question of whether Chinese stock and property markets are integrated or segmented. We find that, at the national level, investment returns in property and the A-share markets were co-integrated in the long run. In the short run, property price Granger caused A-share prices, but not vice versa. However, the B-share prices were negatively correlated with property prices. Furthermore, the linkage between city-level property prices and stock prices showed significant variations across the country. These findings reveal that property and stock markets were integrated at the national level but the property markets were reasonably segmented among cities. They suggest that investment portfolios pursuing risk diversification should include both A and B shares and properties from different cities.
Notes
1. B. Powell, China's Property: Bubble, Bubble, Toil and Trouble, Time Magazine, 22 March 2010.
2. T. Chen, 2009, “Housing Assets are the Largest Single Source of Household Wealth in China.” Southern Weekend (Nanfang zhoumo), 8 October 2009.
3. Thirty-five cities include: Beijing, Changchun, Changsha, Chengdu, Chongqing, Fuzhou, Guangzhou, Guiyang, Ha'erbin, Haikou, Hangzhou, Hefei, Huhehaote, Jinan, Kunming, Lanzhou, Nanchang, Nanjing, Nanning, Ningbo, Qingdao, Shanghai, Shenzhen, Shenyang, Shijiazhuang, Taiyuan, Tianjin, Wulumuqi, Wuhan, Xi'an, Xiamen, Yinchuan, Zhengzhou, Xining, and Dalian.
4. The city statistics include Beijing (henceforth BJ), Shanghai (SH), Hangzhou (HZ), Shenzhen (SZ), Tianjin (TJ), Wuhan (WH), Shenyang (SY), Nanjing (NJ), Guangzhou (GZ), Chongqing (CQ), Chengdu (CD), and Xi'an (XA).
5. Cubic spline interpolation is a powerful data analysis tool. Interpolation is used to estimate the value of a function between known data points without knowing the actual function. The higher the degree of the spline, the smoother the curve. In order to obtain a smooth curve, cubic splines are frequently recommended (Stoer and Bulirsch Citation2002).
6. We have tried other single-equation co-integrating regressions such as Canonical Cointegrating Regression and Dynamic Ordinary Least Square methods and the results are rather consistent. Thus we chose the most used one, namely FMOLS to report.