Charles W. Calomiris, Stanley D. Longhofer, and William Miles
Final draft: April 2013
Critical Finance Review, vol. 2, no. 1, July 2013, pp. 49-93
Current estimates of housing wealth effects vary widely. We consider the role of omitted variables suggested by economic theory that have been absent in a number of prior studies. Our estimates take into account age composition and wealth distribution (using poverty rates as a proxy), as well as wealth shares (how much of total wealth is comprised of housing vs. stock wealth). We exploit cross-state variation in housing, stock wealth and other variables in a newly assembled panel data set and find that the impact of housing on consumer spending depends crucially on age composition, poverty rates, and the housing wealth share. In particular, states with more young people who are more likely to be credit-constrained, and older homeowners, likely to be “trading down” on their housing stock, experience the largest housing wealth effects, as suggested by theory. Also, as suggested by theory, housing wealth effects are higher in state-years with higher housing wealth shares, and in state-years with higher poverty rates (likely reflecting the greater importance of credit constraints for those observations). Overall, we estimate the average housing wealth effect to be approximately 8.1 cents per dollar. However, consistent with theory, demographic and wealth characteristics of the population cause this effect to vary widely across states and over time.
- DOWNLOAD PAPER (523 downloads)
- Social Science Research Network link: http://ssrn.com/abstract=1977353
The following files contain supplemental graphs and regression results referenced but not presented in the body copy of the paper
- Supplemental appendix with results from OLS regressions (432 downloads)
- Supplemental appendix with results using unemployment rate (435 downloads)
- Supplemental appendix with results using raw homeownership rate estimates (500 downloads)
- Supplemental appendix with weighted versions of Figure 5 (443 downloads)
- ASCII log file with complete regression output for the analysis in the text of the paper (501 downloads)
Stata and Excel data files containing the final data used in the analysis.