Does financial wealth drive the share of risky assets in the portfolios of individual investors? Is the financial wealth elasticity of the risky share homogenous across investors or does it vary with their demographic, financial and portfolio characteristics? How does the aggregate demand for risky assets respond to changes in the wealth distribution? In portfolio choice theory, mechanisms such as habit formation, borrowing constraints, decreasing relative risk aversion, portfolio insurance, or a “capitalist” taste for wealth, all imply that richer households allocate a higher fraction of their financial wealth to risky investments. These theories also predict that the financial wealth elasticity of the risky share should vary with household characteristics, including financial wealth itself. Furthermore, a growing literature investigates how at the aggregate level, the demand for risky assets relates to the distribution of household preferences and characteristics.
The empirical household finance literature provides only partial evidence on these mechanisms. In cross-sections, richer and more educated investors are known to allocate a higher proportion of their financial wealth to risky assets than less sophisticated households (e.g. Campbell 2006; Calvet Campbell and Sodini, “CCS” 2007, 2009a, 2009b). In addition, the risky share has a negative cross-sectional relation to real estate holdings (Cocco 2005, Flavin and Yamashita 2002), leverage (Guiso Jappelli and Terlizzese 1996), and internal consumption habit (Lupton 2002). It is unclear, however, whether these variables directly impact portfolio choice, or simply proxy for latent traits such as ability, genes, risk aversion, or upbringing. Several recent papers suggest that panel data offer a possible solution to this identification problem when the characteristic of interest exhibits sufficient time variations (e.g. Brunnermeier and Nagel 2008, CCS 2009a, Chiappori and Paiella 2008). One difficulty with the dynamic panel approach is that the researcher needs to control for household inertia by using instruments, and the results are sensitive to the validity of the instruments.
In this paper, we consider an alternative estimation strategy based on the comparison of the financial portfolios held by twins. The analysis is made possible by a novel dataset containing the disaggregated portfolios and detailed characteristics of twins in Sweden. We observe the worldwide assets owned by each twin at the end of a tax year, including bank accounts, mutual funds and stocks but excluding retirement accounts. All holdings are reported at the asset level for the 1999-2002 period.
Our main results are the following. First, we estimate the average financial wealth elasticity of the risky share on the set of participants. As in the earlier literature, we begin by running pooled cross-sectional regressions of a household’s risky share on its financial wealth and yearly fixed effects. The elasticity of the risky share ranges from 21% in the absence of controls to 23% when a large set of a demographic and financial variables is included. It is an open question whether these cross-sectional estimates capture the direct impact of financial wealth on the risky share, or are instead driven by latent traits that are correlated with financial wealth.
We next consider linear panel specifications with yearly twin pair fixed effects, which is our main innovation. These specifications can be estimated by regressing twin differences in the risky share on twin differences in characteristics. The financial wealth elasticity of the risky share is measured at 20% in the absence of controls and at 22% in the presence of controls. A 10% proportional increase in a household’s financial wealth is therefore associated with a 2.0?2.2% proportional increase in its risky share. We report that the cross-sectional variance of the twin pair fixed effect is of the same magnitude as the variance of the predicted component obtained from characteristics. Moreover, the adjusted R2 coefficient is twice as high in twin panel regressions as in pooled cross-sections. Twin pair fixed effects are therefore important and explain a substantial fraction of the observed variation of the risky share.
Download
Twin Picks: Disentangling the Determinants of Risk-Taking in Household Portfolios
