From the general results of the benchmark model in section 3.3, the expected sign of
the combined effect of age and age squared is negative, the occupation retired
dummy variable should have a positive sign (see Figure 2). For the gender dummy
variable, the expected sign could be negative due to risk-aversion differentials, and
expected labor income (growth rate) differentials according to gender. The expected
sign of low education is negative and of high education is positive, due to expected
labor income (growth rate) differentials according to education. Labor income and
net worth should both have zero coefficients, and the labor income/net worth should
be positive if the individual acts according to the CRRA benchmark. Otherwise, the
combined effect of the three variables should at least imply an upward-sloping curve
as shown in Figure 1 for CRRA-similar behavior.
Finally, U.S. investors should (due to the better risk-return trade-off of the risky
asset) for any values of the explanatory variables invest a higher share into the risky
asset than German investors.
5.3 Regression Results and Discussion
Table 3 displays the results of our regression analyses.
Table 3: Determinants for Share of Risky Assets (Tobit Regression); Dependent
Variable: Percent Risky 2 (for definition of variables, see Appendix B)
--- put Table 3 here ---
The age effect (age, age2) for the United States is hump-shaped, peaking at an age of
54; for Germany, the age and age2 coefficients imply a downward-sloping age
effect.17 Taken together with the positive occupation retired dummy variable for
17 Both coefficients are insignificant at the 10% level. Using only age or age2 in our
regression equation produces coefficients with negative signs that are significant at the
1% level. Thus, the overall effect of age on asset allocation is clearly negative. The
coefficients using both variables become insignificant due to the strong correlation
between them.
20