income (other parameters at base values) would invest 100%, and a German investor
would invest around 61% into the risky asset.
After deriving these general theoretical results, we continue with a description of the
data used in our regressions.
4 Data Description and Sample Selection
For the United States, we use data from the 2004 wave of the Survey of Consumer
Finances (SCF). The data set contains detailed information on 4,519 households,
including household demographics, assets and liabilities, income, and other
characteristics.12 Data have been collected by a dual frame sample design. Data for
about 3,000 households are drawn from a representative sample of households in the
United States to reflect characteristics that are broadly distributed in the population,
such as home and vehicle ownership. The other set of 1,500 survey cases are drawn
from an oversampling of wealthy households (based on tax records) to represent
characteristics such as investment behavior, which might be disproportional in
wealth. Furthermore, missing values are systematically imputed by a multiple
imputation technique, so that the data set includes 22,595 records (i.e., 4,519 cases
times 5 implicates).
For Germany, we employ data from the 2003 wave of the Income and Expenditure
Survey (EVS). The available data set for scientific use also includes numerous data
on income, asset allocation, liabilities, and expenditures of 42,744 private
households.13
Because we are particularly interested in individual differences in investment
behavior, we only analyze data on persons that are neither married or live with a
partner. This assures that the decision observed was made by, for example, either a
12 See Bucks, Kennickell, and Moore (2006) for an overview of the SCF.
13 In the 2003 survey, 53,432 households were originally interviewed. The data set for
scientific use, though, was made anonymous, which has resulted in an exclusion of 20%
of the household data.
16