Modelling interactions among the housing market and key US sectors

Abstract

This study evaluates the dynamic interactions among the housing market and ten key US sectors including: consumer discretionary, consumer staples, energy, financials, industrial, technology, health care, materials, utility and telecommunications. Long-run results indicate that the housing market is integrated with each of the ten sector and that the degree of convergence has increased over time and especially after the onset of the most recent housing crisis. Moreover, the housing market contributes most heavily to the common trends indicating that the housing market is the ‘leader’ market that drives each sector towards the long-run equilibrium relationships. Short-run analyses indicate causal linkages emanating from the housing market to each sector with reciprocal feedback. Finally, impulse response function analysis reveal that shocks from each sector affect the housing market but that shocks from the housing market have a (comparatively) more profound and persistent impact on each sector.

Keywords:

• Domestic housing market
• US sectors
• global financial crisis
• recursive cointegration
• common trends
• impulse response functions
• portfolio diversification

JEL classification:

• C22
• C58
• G1
• G11

Notes

1. See, for instance Cheung and Ng (1998), Grammig, Melvin, and Schlag (2005), Baele (2005), Brada, Kutan, and Zhou (2005), Blitz and Vliet (2008) and Brunnermeier and Julliard (2008), Goyenko and Ukhov (2009), Panchenko and Wu (2009), Baur (2009) and Alter and Schüler (2012) among numerous others.

2. See, for instance Fadiga and Wang (2009), Clark and Coggin (2011), Aloui, Assa, and Nguyen (2011), Guo, Chen, and Huang (2011), Syllignakis and Kouretas (2011) and Miles (2014) among numerous others.

3. We thank two anonymous referees immensely for making this point.

4. See, for instance Greenwood and Hercowitz (1991), Bonnie (1998), Leung (2004), Davis and Heathcote (2005), and Leamer (2007) among others.

5. An important extension of this paper would be to examine the specific channels supporting each linkage. The author plans to explore this exciting topic in a future research endeavour. We thank an anonymous referee profusely for mentioning this important point.

6. See, for instance Siegel and Schwartz (2006), Chan, Lakonishok, and Swaminathan (2007), Vardharaj and Fabozzi (2007) and Anderson, Brooks, and Katsaris (2010) among various others.

7. Hurst and Lusardi (2004) and Brunnermeier and Julliard (2008) among others, use this same data-set.

8. Step-by-step illustration is available in Brooks (2014).

9. In cases where r > 1 i.e. r = 1, 2, 3, … the algebra and derivations become too lengthy/voluminous. Thus, they are not shown to preserve space.

10. The ADF and KPSS tests are conducted by including both a constant and time trend in the regressions. The inference regarding the nonstationary property of the data is similar. However for the purpose of brevity, only the results using constant in the regressions are reported. Details are available upon request.

11. In addition to the likelihood ratio test and for robustness purposes, the Akaike Information Criterion (AIC) and the Schwartz Bayesian (SBC) are also implemented to compute the lag length. Both the AIC and the SBC chose 1 lag, but when the residuals are tested, it is found that they are not free from serial correlation at the conventional levels. Details are available upon request.

12. We thank an anonymous referee for making this point.

13. Results are not reported to conserve space but available upon request.

14. In our analyses, we follow in the footsteps of several authors who also use the recursive cointegration technique and we normalise by the 10% critical value. See, for instance, Crowder and Wohar (1998), Phylaktis and Ravazzolo (2005), Yang (2005a), Yang (2005b), Gilmore et al. (2008) and Yunus (2013) among others. It is important to note that as explained in Hansen and Johansen (1999, p. 54), the scaling factor is not important and the analyses are not affected by the scaling factor since the primary objective of the recursive analyses is to visually understand the time paths of the trace statistics. Since using different scaling values would not affect/change the result or the implication of the results, we have continued our analyses and normalised by using the 10% critical value.

15. See, for instance, Gregory (1994), Richards (1996) and Gonzalo and Lee (2000) among others.

16. It is important to note that, the recursive cointegration technique has been used in an extensive number of published studies using different sample periods/sizes some of which are shorter than the twenty year (1991–2011) time span used in the current analyses. For instance, Crowder and Wohar (1998), Phylaktis and Ravazzolo (2005), Yang (2005a, 2005b), Shin, Yang, and Khan (2007), Gilmore et al. (2008) and Yunus (2013) among others use the recursive cointegration technique to understand the time-varying progression of linkages across various asset classes over time horizons which are less than the 20-year time period used in the current study.

17. We thank two anonymous referees profusely for making this important point.

18. Moreover, we use different types of deterministic specifications (no constant, constant and constant with trend) when conducting the recursive analyses to ensure that the results are robust over the sample period under investigation. The results and the implications of the results are very similar thus corroborating earlier findings. Results are available upon request.

19. The lag lengths of 2 is determined to be optimal using the Likelihood Ration tests (Sims, 1980) and ensuring that the residuals are white noise.

²⁰ I acknowledge S&P for supplying me with the data for the sectoral indexes.