Elicitation of Risk Preferences: Complexity Versus Accuracy

Abstract

The answer to the question of how to elicit risk preferences is vital for predicting individual behaviour and the interpretation of experimental data. In this paper, we first present a current overview of the state of the art in the literature on the methods for eliciting risk preferences and categorise the literature in a systematic manner. Second, we conduct an experiment based on the midpoint chaining method (Krzysztofowicz Organ Behav Hum Perform 31(1):88–113, 1983) and evaluate this parameter-free elicitation method with different numbers of supporting data points in light of three popular parametric utility functions and data generated by an additional choice task in our experiment. We find that, at least for our choice problem with simple lotteries, less onerous methods are sufficient to predict decision behaviour.

Appendix

The parameter(s) of utility functions (6) to (8) on an aggregate level are reported in Table 6.

Table 6 Aggregated utility functions

The RSS value is very similar across the three parametric forms, with the exponential power utility function naturally possessing the best fit, as it includes both CARA and CRRA as special cases. The power and the power exponential utility functions are presented graphically in Fig. 3. It is apparent that the different parametric forms lead to very similar aggregate utility functions. The two utility functions are hardly distinguishable. In conformity with Ockham’s razor, of models with similar explanatory power, the simpler model should be favoured.

Fig. 3

Best fit of aggregated utility functions with constant relative risk aversion (CRRA) and the power exponential utility function (PEUF)