The role of real estate uncertainty in predicting US home sales growth: evidence from a quantiles-based Bayesian model averaging approach

ABSTRACT

This paper investigates the role of real estate-specific uncertainty in predicting the conditional distribution of US home sales growth over the monthly period of 1970:07 to 2017:12, based on Bayesian Model Averaging (BMA) to account for model uncertainty. After controlling for standard predictors of home sales (housing price, mortgage rate, personal disposable income, unemployment rate, building permits, and housing starts), and macroeconomic and financial uncertainties, our results from the quantile BMA (QBMA) model show that real estate uncertainty has predictive content for the lower and upper quantiles of the conditional distribution of home sales growth.

KEYWORDS:

• Home sales
• real estate uncertainty
• quantile regression
• Bayesian model Averaging

JEL CLASSIFICATION:

• C11
• C22
• C53
• R31

Acknowledgments

We would like to thank an anonymous referee for many helpful comments. However, any remaining errors are solely ours.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ Based on the Financial Accounts data of the US corresponding to the fourth quarter of 2018 (https://www.federalreserve.gov/releases/z1/20190307/z1.pdf), residential estate represents about 82.39% of total household non-financial assets, 26.30% of total household net worth and 22.79% of household total asset.

² In fact, Strobel, Nguyen-Thanh, and Lee (2017) had shown that aggregate macroeconomic uncertainty actually does not affect home sales.

³ Note that, unlike the Markov-switching and the smooth threshold models, we do not need to specify number of regimes of home sales growth in an ad hoc fashion with the quantiles-based approach. At the same time quantile regression estimates are known to be more robust against outliers in the dependent variable.

⁴ The MU and FU indices are available for download from: https://www.sydneyludvigson.com/data-and-appendixes.

⁵ The REU index is downloadable from: https://sites.google.com/site/johannespstrobel/.

⁶ While REU1 and REU3 were weakly significant at the 10% level, REU12 was not significant even at the 10% level. Complete details on the OLS estimation results are available upon request from the authors.

⁷ Note that, we also applied the bivariate causality-in-quantiles test of Jeong, Härdle, and Song (2012), and found that REU1, REU3, and REU12 predicted home sales growth over its entire conditional distribution. But this causality test being based on a bivariate framework is likely to suffer from omitted variables bias, and hence, we cannot put too much reliance on these results, complete details of which are however, available upon request from the authors.

⁸ To validate this point, we analysed the cross-quantilogram (as developed by Han et al. (2016)) between home sales growth and the three real estate measures of uncertainty in turn. Note that, the cross-quantilogram measures quantile dependence and is a model-free test of directional predictability between two time series involved in the system. Based on the cross-quantilogram, we observed that changes in real estate uncertainty from its lower quantiles tend to have a stronger positive impact on home sales growth around the upper end of its conditional distribution, when compared to the same changes of real estate uncertainty from its upper quantiles. Complete details of these results are available upon request from the authors.

⁹ Ludvigson, Ma, and Ng (2019) has developed a narrower index of macroeconomic uncertainty based on 73 variables related to real activity, which they call real uncertainty (RU), and is downloadable from the link in Footnote 4. Replacing the MU index at horizons of one-, three-, and twelve-month-ahead, with RU1, RU3, and RU12, we re-estimated the QBMA three models. We found that using this sub-index of macroeconomic uncertainty, the coverage of quantiles for which predictability is observed for home sales growth, due to REU1, REU3 and REU12, increases, though the pattern of the sign remains the same. These findings clearly suggest that the information content regarding macroeconomic uncertainty in the MUs are, not surprisingly, greater than in the RUs. Complete details of these results are available upon request from the authors.

¹⁰ We thank an anonymous referee for pointing this out to us.