ABSTRACT
Regional housing prices may be dynamically linked and converge to a long-run equilibrium. The level of connectedness can be used to describe systemic risk in the housing market. This paper uses a recently developed vector autoregressive-based time-series approach to focus on the short-run dynamics in urban housing prices in China. The empirical results show that housing prices across cities have increasingly connected and consequentially associated with higher systemic risk. The empirical results are consistent with the implications of an information spillover mechanism documented in the recent literature.
DISCLOSURE STATEMENT
No potential conflict of interest was reported by the authors.
Notes
1. The price per square metre for an apartment in the city centre as of 2018 is US$17,131 in Beijing and US$13,685 in New York, according to Numbeo (https://www.numbeo.com/property-investment/).
2. The ‘half-life’ represents the speed of adjustment in the error-correction mechanism. It refers to the time needed to eliminate half the initial deviation.
3. See http://www.stats.gov.cn/tjsj/.
4. An impulse-response function is often used to demonstrate the estimated VAR system and also to show how shocks to one variable may affect others. This paper, however, aims to show how systemic risk (connectivity) changes across different groups and over time in China’s urban housing prices. The variance decomposition approach and its repackaging strategy proposed by Diebold and Yilmaz (Citation2009, Citation2012) are therefore more appropriate and better serve the present research purpose. The authors appreciate an anonymous referee’s suggestion to clarify this point for readers.
5. Ranking information is based on www.cbnweek.com and is available at Baidu Baike (www.baike.baidu.com). To avoid repeated sampling, we exclude cities in the ‘big cities’ group in tier 1.
6. For a discussion of the sensitivity of the connectedness measures to forecasting horizons, see Diebold and Yilmaz (Citation2012).
7. The Jarque–Bera tests show that the 100 random draw results follow a normal distribution, so no additional sampling is used. The descriptive statistics may vary in a different random draw. The mean and SD are generally stable.
8. The proportion of the causal relationship is 61.95% if a 10% level of significance is used to test for causality.