Risk Preference Elicitation and Financial Advice Taking

Abstract

Financial advisors rely on accurate measures of investor risk preferences. This study compares different risk elicitation methods (REMs) in terms of their perceived suitability and impact on financial advice taking. The results suggest that the perceived suitability of the suggested risk profile strongly predicts delegation to an advisory tool. REMs differ in terms of their perceived process similarity with the investor, which positively affects suitability (and thus, delegation) directly and through its positive effect on source credibility. Differences were also found with regards to the perceived complexity of the risk profiling task, which is positively related to suitability. In summary, the findings imply that applying suitable REMs matters not only because it avoids misrepresentation of an investor’s true risk preferences, but because it directly affects the propensity to delegate financial decision-making.

Keywords:

• Financial advice
• robo-advisory
• advice taking
• risk preference elicitation

JEL Classification:

• G11
• G41
• D14

Acknowledgments

I would like to thank Melinda Estelle (the Associate Managing Editor), an anonymous reviewer, Patrizia Altmayr, Jonathan Federle, Markus Glaser, Johannes Jaspersen, Rouven Litterscheidt, Alexander Rühr, Markus Witt, as well as participants of the Canadian Conference on Fintech and Insurtech 2021 for helpful comments. I would also like to thank the Munich Experimental Laboratory for Economic and Social Sciences (MELESSA) for providing laboratory resources. All remaining errors are mine.

Notes

1 Besides, robo-advisors may be more hesitant to pose a large number of questions to avoid alienating prospective investors in an excessive onboarding process, which is consistent with a lower number of risk profiling questions among digital providers as compared to traditional advisors (Tertilt and Scholz 2018).

2 While there is empirical evidence suggesting that differences in individual risk preferences are indeed associated with differences in the portfolio proportion invested in equity (e.g., Cardak and Wilkins 2009; Dohmen et al. 2011), it is likely not the only factor as aggregate risk aversion would have to be implausibly large to explain stock market risk premia (Campbell 2003).

3 For instance, the European Markets in Financial Instruments Directive (MiFID) suitability guidelines (ESMA 2018) specify that financial advisors must ‘understand the essential facts and characteristics about their clients’ (guideline 2(22)) and that this includes ‘the analysis of the client’s financial situation (including his ability to bear losses) or investment objectives (including his risk tolerance).’ (guideline 2(27)). In the United States, FINRA Rule 2111 institutes similar guidelines.

4 Another question that arises when eliciting risk preferences is whether investors can assess their preferences regarding changes in wealth without having experienced them. Evidence suggesting that risky choices differ when elicited from descriptions versus when sampled from experience imply they may not (Bradbury, Hens, and Zeisberger, 2015; Hertwig et al., 2004; Kaufmann, Weber, and Haisley, 2013).

5 http://www.financial-ombudsman.org.uk/publications/ar15/ar15.pdf

6 For comprehensive reviews of the most prominent REMs, the reader is referred to Charness, Gneezy, and Imas (2013), Frey et al. (2017), Menkhoff and Sakha (2017), and Pedroni et al. (2017).

7 Bonaccio and Dalal (2006) provide a comprehensive survey of the literature.

8 Decision-making processes are thought of as strategy vectors containing values in different decision-making dimensions. Process similarity can then be conceptualized as an inverse Euclidian distance measure. The closer two processes’ values in the different dimensions are, the lower the Euclidian distance measure (and the higher process similarity).

9 This definition refers to the expertise component of source credibility. The other components of source credibility are bias on the part of the advisor and bias on the part of the judge (Birnbaum and Stegner 1979).

10 Since the experiment was conducted online, participants were potentially able to google answers to the questionnaire. Since the questionnaire’s purpose is to induce effort and the initial budget is capped at €8, this is not considered an issue.

11 The fee has been calibrated using scenario and benchmark analyses and was shown to produce sufficient variation in delegation (Litterscheidt and Streich 2020).

12 Risk profile descriptions are based on typical industry standards.

13 ‘How confident are you that the assigned risk profile matches your preferences?’ [Not at all – 1 2 3 4 5 6 7 – very.] (cf. Budescu and Rantilla 2000).

14 The exact wording is ‘Are you generally a person who is fully prepared to take risks or do you try to avoid taking risk?’.

15 Consistent choices are enforced. Specifically, choosing the safe payoff of €0 in row 1 and switching between the two alternatives more than once is not permitted.

16 This is consistent with one participant’s post-experimental comment stating that the risk profile determined by the BRET was ‘overly risk-averse’.

17 Table B7 in the Appendix reports pairwise correlations for the four perception measures. Advice suitability is significantly positively correlated with task complexity, process similarity, and source credibility. Source credibility, in turn, is significantly positively correlated with task complexity and process similarity.

18 Computations were conducted using SmartPLS 3 (Ringle, Wende, and Becker 2015).

19 Table B9 in the Online-Appendix contains a decomposition of the effect in total, direct, and indirect effects.

20 In particular, average process similarity was 4.06 (SD = 1.37, N = 180) among participants with below-median output similarity and 3.89 (SD = 1.34, N = 81) among participants with above-median output similarity (z = 0.897).