Abstract
This study estimates changes in the relationship between housing wealth and consumption among homeowners during the recent housing boom and bust in the United States, focusing on the period 2001-2007, during which house prices increased and financial innovations led to an increased availability of products enabling households to extract home equity; and on the period 2007-2013, during which house prices declined and home equity withdrawal products became largely unavailable. The estimated elasticity of consumption with regard to housing wealth increased in 2004 and 2007 (.06) relative to 2001 (.04). The estimated elasticities then decreased in 2010 and 2013 (to below .04). In addition, the increase was larger among borrowing constrained households than unconstrained households. No relationship between housing prices and consumption was found among renters. These additional tests for subpopulations support the hypothesis that the increase in consumption out of housing wealth occurred through the collateral channel.
Acknowledgments
The author thanks the editor and reviewers for their useful comments that contributed to improve this paper. The author also thanks Raphael Bostic, Richard Green and Gary Painter for their inputs on earlier versions of this paper. All remaining errors are my own.
Notes
1 A pairwise t-test of the difference in the estimates finds that the difference in the 2004 and 2010 estimates are marginally significant at the 10% level.
2 The differences in estimated elasticities and marginal propensity to consume come from wide variations in the types of data used, specifications, and definition. Of particular note, there are important variations in how housing wealth is defined: by overall house prices, individual house value, or net equity, and in whether only homeowner’s or the entire population’s consumption is considered. Differences also exist as to whether the level or change in wealth is considered. The estimated elasticities of .04 to .06 found in this paper are consistent with those found by Bostic et al. (Citation2009) and with other recent findings for the same period in the United States (Aladangady, Citation2017; Guren et al., Citation2018).
3 A number of mortgage systems have a large prepayment penalty if refinancing outside a move.
4 Since households are surveyed five times to collect complete information on expenditures over four quarters, the consumption period can vary somewhat depending on the households.
5 Despite the attempt to limit the number of dimensions, a number of cells are empty. A limit of the SCF data is that it does not include any geographic data, making it impossible to constrain matching on a location dimension that would match households within markets that have experienced similar house price trends.
6 In order to ensure that this match does not lead to spurious results, the sampling is repeated 100 times and each regression is then estimated 100 times with the different matches using a bootstrap procedure. The results presented are the average parameter values and the standard error of the parameter estimates over the 100 estimations.
7 A value of 1 is added to each individual’s financial wealth to account for households reporting 0 financial wealth.
8 An attempt to stratify the sample between households underwater (with a mortgage above their estimated house value) and not was unsuccessful due to the limited number of households in that situation in most years. Even in 2010, the year in which they are the most common, underwater households represent less than a quarter of the sample.
9 The imputation is based on the Barakova et al. (Citation2003) method to predict FICO score based on SCF variables.
10 As a robustness check, results in Panel B use only the CEX data (and therefore do not control for wealth). The overall results are similar in nature. This supports the fact that the matching procedure does not affect the results substantially.