Size and sign asymmetries in house price adjustments

ABSTRACT

Long-run mean-reversion in real house prices is determined by the relative strength of fundamental factors against the short-run influences. This article suggests that the adjustment towards the long-run trend in house prices could display non-linear behaviour due to some intrinsic characteristics of the housing market. Accordingly, sign and size asymmetries as well as possible structural breaks are taken into account in a unit root testing exercise for twenty-nine countries. Our results suggest that mean-reversion exists for seventy percent of the countries in our sample. Moreover, the out-of-sample forecasting performance of our non-linear models in predicting house prices is better than a simple auto-regressive benchmark for some countries.

KEYWORDS:

• House prices
• mean reversion
• structural breaks
• non-linearity
• unit root testing
• forecasting

JEL CLASSIFICATION:

• E31
• E37
• C2

Acknowledgments

We would like to thank to the anonymous referee and Christophe Andre for their helpful comments.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

¹ This ‘leaning against the wind’ idea is criticized by others arguing that the monetary policy should concentrate on the consequences of these fluctuations rather than dampening asset price booms. For a survey of these alternative views see Kuttner (2011).

² Himmelberg, Mayer, and Sinai (2005) suggest alternative metrics for pricing in the housing market. See Williams (2013) for a discussion of house price bubbles in US. Also, see Gürkaynak (2008) for a survey on tests for asset price bubbles.

³ See Holly and Jones (1997), McCarthy and Peach (2004) and Girouard et al. (2006).

⁴ Moreover, individuals usually extrapolate past patterns into their forecasts which lead to a wedge between the fundamental value and the current price of assets (Barsky and De Long 1992).

⁵ For recent examples of studies that use econometric methodologies testing for house price bubbles, see Glindro et al. (2011) for Asia-Pacific countries; Ahuja et al. (2010) for Chinese housing market; Kivedal (2013) for US; Yiu, Yu, and Jin (2013) for Hong-Kong; Caspi (2016) for Israel; Engsted, Hviid, and Pedersen (2016) for a group of OECD countries; Eliasson (2017) for Iceland; Bourassa, Hoesli, and Oikarinen (2019) for Finland, Switzerland and US; and Coskun et al. (2017) for Turkey.

⁶ Testing for linearity is essential since non-linear models nest a linear model that is not identifiable under a linear data generating process as stressed by Teräsvirta (2006).

⁷ Also see Igan and Loungani (2012) for a cross-country analysis on house price dynamics covering 1970–2010 period.

⁸ A corrective reaction at point D for a seller would be postponing selling the house or sitting on the market for a longer period of time.

⁹ An alternative strand of the literature also explores these provincial differences under the concept of a ripple effect. As the argument goes, while the price levels in different regions might diverge in the short-run, they move together in the long-term. Hence, the ratio of regional to national price index reveals a stationary behaviour. This idea of spatial transmission of house prices shocks among different regions is tested for various countries using linear and non-linear unit root tests [See Cook (2003) and Balcilar et al. (2013) for non-linear examples]. Note that, existence of such a long-run convergence between provincial prices also supports our presumption of long-run mean reversion in the national index.

¹⁰ Engelhardt (2003) further shows the significant impact of such loss aversion on household mobility.

¹¹ Akdoğan (2015) employs a similar AESTAR framework to test such asymmetric reaction of the central banks from the deviations from inflation-targets for inflation-targeting countries.

¹² Tsai (2013) shows the asymmetric impact of monetary policy on housing prices for UK housing market. Chen et al. (2012) examines the impact of monetary policy on housing demand via the investment link and suggest that the asymmetries in house price movements is led by investment demand.

¹³ Sources: National sources, BIS residential property price database, www.bis.org/statistics/pp.htm. The countries for which at least 50 data points are available are selected in our exercise.

¹⁴ The combined identifiable seasonality test which covers three types of F-tests is used for the presence of seasonality. These three tests are parametric and non-parametric tests (both assuming stability) and test assuming moving seasonality.

¹⁵ Note that an ESTAR type adjustment is also employed in Christopoulos and León-Ledesma (2010) test.

¹⁶ The mean-reversion result for Korea also corroborates that of Glindro et al. (2011).