woman or a man (among other characteristics of a respondent) and enables us to
extract individual-specific characteristics from the data.14 We exclude all self-
employed individuals from the analysis. For the United States, self-employed
individuals represent 7.9% (based on the weight given in the data set); for Germany,
4.3% of individuals in the data set are self-employed. This exclusion was necessary,
because, in the German data, the value of one’s own business—a major part of the
asset allocation among those who are self-employed—is not reported. Furthermore,
we excluded cases that are located below zero in the income and net worth
distribution.
Another restriction is given by the welfare analysis in section 6 that builds on the
intertemporal optimization model of section 3. The data and regression results serve
as input for section 6. The underlying optimization model does—as a standard
assumption—not allow for borrowing and short-selling. Deleting individuals having
debt would exclude the majority of U.S. respondents, which is not desirable. Instead,
we deducted debt from the amount of investments made into the risky asset—that is,
we treated debt as a negative asset. Thus, finally we had to exclude only individuals
for whom the resulting risky asset share variable used in our optimization (see
section 3.1) and regression (see section 5.2) models was below 0% (15.5% of
weighted cases for the United States and 5.9% for Germany).
Our final SCF data set reduces to a sample size of 894 cases (4,470 records divided
by 5 implicates), and the EVS data set contains 10,764 cases.
Table 2 shows country-specific descriptive statistics on demographic and financial
characteristics of our sample. The variables finally used in our regression analyses
are in italics. For the U.S. data, the statistics are based on the weight given in the data
set to account for the oversampling of wealthy.
14 Using households with more than one person would make inferences about individual
behavior almost impossible. Household decisions would reflect some kind of average
preferences and would implicitly incorporate numerous not directly observable
diversification effects (e.g., with respect to labor income risks or life span uncertainty).
17