Investing versus gambling: experimental evidence of multi-domain risk attitudes

Abstract

Economists have noted the empirical regularity that an individual’s attitude towards risk is not constant across elicitation settings. Such a pattern is conceptually consistent with the argument in psychology that risk is domain specific. To explore this view, we frame a common risk elicitation tool in laboratory experiments as both a gambling game and an investment opportunity. A survey instrument used to measure one’s attitudes towards risks on different domains is found to explain partially within-subject variation in observed behaviour between the two tasks. The evidence presented in this article demonstrates the need to consider more nuanced approaches to modelling risk.

Keywords:

• multi-domain risk attitudes
• framing effects
• laboratory experiments

JEL Classification:

• C91
• D81

Acknowledgements

The experiments reported in this paper were conducted while Deck was a visiting researcher at the Economic Science Institute (ESI) at Chapman University. The authors wish to thank the ESI for technical support. This research was also supported in part by NIH grant R21AG030184. Jungmin Lee acknowledges that his work was supported by the Sogang Research Frontier.

Funding

The authors wish to thank the ESI for financial support. This research was also supported in part by NIH grant R21AG030184.

Notes

¹ An alternative approach to direct risk elicitation techniques like H&L is to infer risk attitudes from behaviour in auctions. Frustratingly, there is little consistency between elicitation techniques (e.g. Isaac and James, 2000; Berg et al., 2005).

² The complete DOSPERT battery also considers a respondent’s likelihood of engaging in an activity and his perceived benefits and risks of doing so. For this research, we focus on the likelihood of engaging in an activity.

³ Copies of the directions and the survey are available in an online appendix.

⁴ Eleven subjects made choices inconsistent with a single switching point from A to B in the gambling frame. Of these 11, 7 also made inconsistent choices in the financial frame, but no one was inconsistent in only financial frame. This rate of inconsistency is not dissimilar from that reported by Holt and Laury (2002) and other researchers using this procedure. The results reported in this section do not change substantively if these inconsistent observations are omitted. 

⁵ The results are qualitatively the same when we use Poisson models.

⁶ We also measured the attitudes as dummy variables that equal 1 if the subject expressed a greater propensity to undertake risks in that domain than the average person in the sample. This specification was used to account for the skewness present in gambling attitudes. The results are qualitatively the same.

⁷ Age is measured as a deviation from the sample average in order to avoid the misinterpretation of the results that might arise from a baseline age of 0.