This section presents the results of the models proposed in Section 2. The first model to be
estimated is the Markowitz model. This model has the proportional budget endowments as a
dependent variable and the risk-return trade-off, the interaction of the risk-free rate with the
risk-return trade-off and the interaction of the respondent’s CRRA and the risk-return trade-off
as the independent variables. Estimation is done in LIMDEP, a statistical package suited for
estimating complex binary and multinomial logit models. While the first model is rather simple
in terms of number of explanatory variables, the fact that the three choices yield polytomous
data3 that constitute the dependent variable classifies the model as a multinomial one. Another
advantage is that the program allows inclusion of a specification for panel data. Since respondents
have performed eight subsequent choice tasks in the questionnaire, it is helpful to include a panel
specification in the model to increase model power by controlling for the repeated-measure effect.
The basic model is estimated including a constant, the risk-return trade-off as a main effect and
the two interactions of risk–rate and risk–risk aversion. Table 2 presents the results.
As we expected, the risk-return relationship has a significantly positive effect on the portfolio
choice. This relationship is moderated by two additional factors, the risk-free rate and the individual’s
CRRA. Again both effects are significant and have the expected signs. The coefficient for the
interaction of the risk-free rate with the risk-return trade-off is negative because an increase in the
risk-free rate decreases the utility of the risky asset and increases the utility of the risk-free asset.
Similarly, the coefficient on the interaction term of the risk-return trade-off with the risk aversion
coefficient is negative and significant, showing that a higher CRRA decreases the utility of the
risky asset. Although the ρ2(adj) is relatively low (8.30 percent), the model is significant based