Expectation, volatility and liquidity in the housing market

Abstract

Measuring housing price volatility is fundamental to understanding the dynamics of housing price risk. This article aims to explore whether a liquidity factor plays a role in explaining the second moment (i.e. the volatility) of housing prices. Housing price volatility is measured as the conditional variance of a Generalized Auto Regressive Conditional Heteroscedasticity (GARCH) model under the Adaptive Expectations framework. The empirical evidence reveals that volatility transmits from smaller housing units to larger housing units, which indirectly supports the trade-up effect discussed in the literature. In addition, less liquid housing classes are more sensitive to unexpected liquidity shocks, and the starter housing class is extraordinarily sensitive to negative liquidity shocks. Consistent with friction search theory, pricing errors are alleviated as the trading volume increases, because the valuation price tends to be more accurate as more information is available.

Keywords:

• volatility
• liquidity
• adaptive expectations
• Hong Kong housing market

JEL Classification:

• C22
• G12
• C50

Acknowledgement

The author wishes to express appreciation to Prof. K.W. Chau and two anonymous reviewers for their valuable comments.

Notes

¹ For example, the 1997 Asian Financial Crisis; the 2008 US Subprime Mortgage Crisis.

² In contrast, the classical financial assets are homogeneous, thickly traded and the prices are determined by a market clearing mechanism.

³ The seller accepts the first bidder whose bidding price is equal to or above the reservation price.

⁴ See Andrew and Meen (2003) and Miller et al. (2011) for examples.

⁵ Chau et al. (2005) conclude that the Hong Kong Housing market is highly liquid for several reasons: (1) no capital gains tax; (2) highly mobile population with low cost of moving; (3) transparency and efficiency.

⁶ For example, Ortalo-Magne and Rady, 2006 note that households can also move between owning and renting.

⁷ For example, the annual turnover ratio (i.e. transaction units divided by total stock) of the housing market in 2010 was around 9%.

⁸ For expositional purposes and simplicity, we assume that there is no depreciation maintenance costs of housing properties and that the rental is net income.

⁹ Noting that return in this article denotes capital gain rather than total return.

¹⁰ It is unlikely that the discount rate would change significantly within a short interval, for example, when the monthly data are adopted.

¹¹ Miller et al. (2011) claim that housing turnover is a better indicator for measuring home sales than the number of housing transactions, because a large group (more housing stock) is more likely to have a larger magnitude of housing transactions.

¹² Baker and Stein (2004) suggest that turnover serves as an indicator of stock market ‘feeling’, and high turnover implies that prices are overvalued (see also Hui et al., 2013).

¹³ There are three alternative liquidity measurements in real estate market, i.e. trading volume, bid-ask spread and time on market (TOM). The turnover rate is a better indicator than trading volume, because a housing class with more housing flats generally has a larger trading volume (e.g. Miller et al., 2011; Zheng, 2013). Besides, the data of bid-ask spread and TOM are not available at the aggregate level. Hence, turnover rate is the best choice of this study.

¹⁴ It should be noted that public sector developments, including domestic units built under the Private Sector Participation Scheme for subsidized sale, and all units built under the Home Ownership, Buy or Rent Option, Mortgage Subsidy, Sandwich Class Housing, Urban Improvement and Flat-for-Sale Schemes, are not included. Both presale and primary sale transactions are also excluded. Source: Rating and Valuation Department of Hong Kong, online database at http://www.rvd.gov.hk/.

¹⁵ The EPRC database mirrors the property transaction records in the government’s Land Registry and has been widely adopted in previous literature (e.g. Chau et al., 2005; Yiu et al., 2008; Zheng, 2013). More information can be found at http://www.eprc.com.hk.

¹⁶ The results are available upon request.

¹⁷ To conserve space, we do not report the results of LM tests here; however, all the results are available upon request.

¹⁸ Engsted and Pedersen (2014) use similar approaches to decompose housing returns in 18 OECD countries.

¹⁹ Chan (2001) notes that households become ‘locked in’ if the mortgage balance is greater than around 90% of the total assets.

²⁰ This postulation is similar to the proposition of Clapp et al. (1995), who argue that greater the number of transactions lower the costs of information.

²¹ That is (9.58*48.4)/(3.76*2.2). This figure is simply for demonstrating purpose. In the real world, transactions in different classes within the same estates and/or districts are comparable to some extent.

²² Classes A, B, C, D and E account for 31.9%, 48.8%, 11.8%, 5.3% and 2.2% of the total stock, respectively.

²³ For example, the number of Class E houses only accounts for 2.2% of the total private housing stock.

Additional information

Funding

This study was supported by the National Natural Science Foundation of China [grant number 71473105], and the Fundamental Research Funds for the Central Universities [grant number 12JNYH002].