Trusting Clients’ Financial Risk Tolerance Survey Scores

Abstract

We examine whether and to what extent financial advisers can trust financial risk tolerance scores derived from client survey responses. We propose using the standard deviation of standardized survey responses as a simple, practical measure for determining the reliability of client risk tolerance measures. Our findings suggest that advisers will better discharge their fiduciary responsibilities by reexamining a client’s survey results if there is substantial variation in that client’s standardized survey responses and resurveying such clients to better gauge their risk tolerance scores.

Dislosure: The authors report no conflicts of interest.

Editor’s Note

Submitted 26 December 2017

Accepted 15 January 2019 by Stephen J. Brown

Acknowledgments

We would like to thank the anonymous referees, who provided many suggestions that improved our article. We especially thank Professor Stephen J. Brown for his great patience, helpful guidance, and suggestions.

Notes

¹ Although a unique definition of financial risk tolerance does not exist, a very useful guide for our purposes is “the combination of psychological traits and emotional responses that determine the investor’s willingness to take on financial risk” (Klement 2015, 3).

² The prior literature tells us a lot in terms of basic demographic and socioeconomic “heuristic” financial risk tolerance associations. Generally, positive associations between financial risk tolerance and wealth, income, education, investment knowledge, self-employment, and ethnicity/culture have been reported, whereas there are negative associations between financial risk tolerance and age, gender (female), having a partner, and number of financial dependents (see, among others, Baker and Haslem 1974; Faff, Hallahan, and McKenzie 2009; Gerrans, Faff, and Hartnett 2015; Pålsson 1996; Powell and Ansic 1997; Santacruz 2009; Statman 2010; Yao, Gutter, and Hanna 2005).

³ The Financial Industry Regulatory Authority and the Certified Financial Planner Board of Standards, or CFA Board, are organisations in the United States. Similarly, in Australia, to meet a best-interests duty (Commonwealth of Australia 2001, sec. 961B), the regulator identifies the need to establish a client’s risk appetite or risk tolerance (Australian Securities and Investments Commission 2017). In the United Kingdom, the Financial Conduct Authority similarly requires that providers of advice establish a client’s risk profile to determine the suitability of advice.

⁴ The reader should note that we are not claiming that ours is necessarily the “best” approach (since it is impossible to know all the practical approaches applied by financial advisers); rather, we are claiming that ours is a significantly better approach than is commonly executed in the industry.

⁵ The identity of the commercial organisation providing the psychometrically validated survey product on which our sample is based is intentionally anonymized to comply with Financial Analysts Journal policy. The authors have no financial interest or conflict associated with the data provider.

⁶ This measure has also been called an “intra-individual variance” or “item-level variation” (Churchyard, Pine, Sharma, and Fletcher 2014), and more generally, such measures have been labeled “person reliability parameters” (LaHuis et al. 2017).

⁷ We urge survey providers to freely publicise the mean and standard deviation of the standardized response scores for surveys in their databases, because doing so would provide highly useful information for advisers, given that one standard deviation above the mean is a marker for the top 16% of cases.

⁸ The other five methods are personal or professional judgment, heuristics, objective assessments of revealed behavior, single-item questions, and mixed measures (Grable 2016, 23).

⁹ See, for example, Faff et al. (2009); Roszkowski and Cordell (2009); Santacruz (2009); Smith (2016).

¹⁰ For more details on the test instrument, see Bright and Adams (2000); Earl and Chew (2015).

¹¹ This approach is routine among personality scales. For example, see the interpersonal trust inventory of Rotter (1967) and the anxiety trait inventory of Spielberger (1983).

¹² Each person’s IRV is calculated as the standard deviation of item response scores around the mean item score. The formula used is ${\text{IRV}}_j=\sqrt{\frac{{\displaystyle {\sum }_{i=1}^S\left(x_{j,i}-{\overline{x}}_j\right)}}{S-1}},$ where S is the number of scale items and x_j is the mean score of client j on the S items in the scale.

¹³ However, Pearson correlations between our unadjusted IRV and FRT scores are 0.75 and 0.76 in the two surveys. This finding highlights the spurious effect discussed previously and suggests one contributor to the confounding results observed elsewhere in the literature in this context—for example, Roszkowski et al. (2009).

¹⁴ Nontabulated models reveal nonsignificant quadratic estimates for the two surveys’ associations between FRT and IRV.

¹⁵ Refer to Table S.1 in the online supplemental material, available at www.tandfonline.com/doi/suppl/10.1080/0015198X.2019.1575160, for complete details.

¹⁶ For robustness, we also estimated a regression that includes the level of each characteristic from Survey 1 and its square, in addition to the change in these characteristics already included. As expected, the levels of characteristics at Survey 1 are not informative regarding the change in FRT scores, and therefore, we retain the more parsimonious model reported in Table 3.

¹⁷ A lack of significance could alternatively occur if negligible change is observed in the distribution of independent variables. We investigated these distributions and confirmed that each variable’s distribution is similar to that of the overall 3,110-case sample shown in Table 3. This is also the case for the lower-quintile subsample. One-way ANOVAs (analyses of variance) and equivalent nonparametric Kruskal–Wallis/Mann–Whitney U tests confirm likely distribution homogeneity for all the variables included in the regressions for these two subsamples.

¹⁸ To investigate the robustness of the results, we reestimated both regressions in Tables 3 and 4 using a reduced set of items to form a second FRT score. This reduced scale included questions that are more clearly linked to “risk tolerance,” as opposed to other, possibly related concepts, such as overconfidence, regret, loss aversion, emotions, and sensation seeking. Specifically, we used a nine-question subset: Questions 1, 6, 10, 14, 16, 17, 20, 22, and 25. These alternative results are consistent with those of the full scale and are presented in Tables S.4 and S.5 in the online supplemental information, available at www.tandfonline.com/doi/suppl/10.1080/0015198X.2019.1575160.