Sunday, October 20, 2013

Can we measure risk tolerance?

One of the many interesting presentations here in Orlando at the Financial Planning Association's annual FPA Experience conference was "Changing Risk Tolerance" by Michael Guillemette of the University of Missouri and Michael Finke of Texas Tech University. 

An important question we all face is, can we measure risk tolerance? Joe Tomlinson's column in this week's Advisor Perspectives provides more background on this issue. 

In trying to measure risk tolerance, what we really want to know is: what is the most aggressive asset allocation someone can withstand before they will otherwise sell their position and leave the market after a drop, or will feel unable to sleep peacefully at night.

There is an active research literature on this topic, which I must admit I am not fully up to speed with the arguments and conclusions. One of the debates is about whether people have a constant underlying risk tolerance which is part of their individual personality, or instead that risk tolerance is fungible and may depend more on perceptions about what is currently happening in the market.

Michael Guillemette presented his research using the average risk tolerance as measured on a monthly basis for individuals with data collected from a well-respected company that focuses on measuring risk tolerance. He presented a figure which I thought was fascinating:





What this figure plots is the monthly measure of risk tolerance along with the level of the S&P 500 index since 2007. Note that the two variables track one another closely. Here is another relevant figure from the presentation:




This figure shows the P/E ratios of the stock market for different reported levels of risk tolerance. We can observe that stock markets are more overvalued at times when people reported the highest level of risk tolerance.

The implications of what is shown in these two pictures is important. It suggests that risk tolerance surveys could cause people to "buy high and sell low."  When stock markets are performing well, the risk tolerance survey will suggest that people are able to cope with more aggressive asset allocations. When stock markets are down, the survey will suggest that people are risk averse and need to use lower stock allocations. This is the exact opposite behavior of what would generally be expected to best serve investors.  

EDIT: Follow-up on the original blog post:
There has been some discussion of this on Twitter. Comments focus on the idea of risk tolerance vs. risk perception. This was also the matter that Michael Kitces was stressing during the Q&A at the conference. However, I'm struggling to understand how this is a defense of risk tolerance, as the measure is supposed to be about risk tolerance, not the perceptions of how much risk exist in the market. A risk tolerance measure should bypass perceptions to get to the underlying characteristics of the individual.

However, another argument I've heard which is more damning to the study described in the post is that the scales on the axes are grossly miscalibrated. The fluctuations in the risk tolerance measure are actually so small that they would have very little practical significance in their implications for asset allocation. Though the correlation to the S&P500 exists, the figures in the above blog post misrepresent that by making the fluctuations in risk tolerance seem much larger than they really are. This is an important point, but I'm not in a position to judge who is right without knowing more about how the risk tolerance measure is applied in practice to determine asset allocation.

32 comments:

  1. Wade, this research is wonderful.

    Maybe it will be the beginning of the end for two decades of diversionary foolishness about investment “education” flooding out from our universities and professors and the institutions that train and certify “fiduciary” investment-financial advisors!

    What advisors should be doing is educating and informing advisor and investors in pursuit of the investment purpose: preparing and showing clients comparisons of what investment choices appear to offer in possibilities and uncertainties for investors’ multi-year dollar plans and goals.

    Instead our universities and professors have misled us to misfocus on technical measures of percent return rate for the individual year with deceptive labels, and the Alice-in-Wonderland pseudo-psycho-science of risk tolerance questionnaires and “utility theory.”

    This could be the beginning of the Investment Education Enlightenment!

    Dick Purcell

    ReplyDelete
  2. Wade--

    As a psychologist, I'm so happy to see meaningful movement beyond the standard -- and pervasive -- risk tolerance surveys that ask about hypothetical sell decisions after large market drops. As Tomlinson notes and Dan Ariely previously observed, individuals in a "cold" (unemotional) state cannot accurately predict how they will respond in a "hot" (emotional) one. Judgments in emotional states are often dominated by primitive fight-flight and pain-avoidance responses as well as by visceral feedback (e.g., racing heart, jitters, excessive sweating, butterflies in stomach), not by rational left-hemispheric thinking.

    So what might work better than traditional approaches? At a minimum, first create a "hot" state before individuals are asked how they would respond to large market declines. In order to do this, they ideally need to feel the threat of some kind of real and meaningful monetary loss, not a hypothetical one on a paper-and-pencil survey or a game simulation without their having any skin (i.e., their own real money) in the game. You could, for instance, give clients $100 or $200 at the outset of a stock market simulation and then allow them to keep whatever's left (up to the original $) after a scenario involving a large market decline. There are other ways of inducing "hot" states, for instance by giving someone a large jolt of caffeine or deliberately creating a stressful (e.g., noisy, uncomfortable, etc.) environment when asking clients what they would do, etc. Years ago, Singer and Schachter used adrenaline injections in laboratory studies to induce emotional states.

    ReplyDelete
    Replies
    1. Very interesting. Thank you for sharing!

      Delete
  3. When I first heard the term 'risk tolerance' I expected the measure to be a well established psychophysics measure of some kind that has a great home in economics. Here, I find out it is a survey result. I am saddened by this.

    ReplyDelete
    Replies
    1. The measure is the first you mentioned.

      Delete
  4. Follow-up on the original blog post:

    There has been some discussion of this on Twitter. Comments focus on the idea of risk tolerance vs. risk perception. This was also the matter that Michael Kitces was stressing during the Q&A at the conference. However, I'm struggling to understand how this is a defense of risk tolerance, as the measure is supposed to be about risk tolerance, not the perceptions of how much risk exist in the market. A risk tolerance measure should bypass perceptions to get to the underlying characteristics of the individual.

    However, another argument I've heard which is more damning to the study described in the post is that the scales on the axes are grossly miscalibrated. The fluctuations in the risk tolerance measure are actually so small that they would have very little practical significance in their implications for asset allocation. Though the correlation to the S&P500 exists, the figures in the above blog post misrepresent that by making the fluctuations in risk tolerance seem much larger than they really are. This is an important point, but I'm not in a position to judge who is right without knowing more about how the risk tolerance measure is applied in practice to determine asset allocation.

    ReplyDelete
  5. It's a fascinating topic, but I can't track my own risk tolerance, let alone help someone else figure out theirs. Early in 2007, I was quite happy with risk. By March 2009, I'd had enough to last me for a while. I think how much risk you can afford is far more important than how much you can sleep with.

    ReplyDelete
    Replies
    1. Dirk, I do think that risk tolerance, risk capacity, and the need for risk are all relevant considerations. While it would be great to ignore the 'sleep factor,' people do need to sleep at night and they also need to avoid panicking and selling out their positions after the market dropped and it's too late. It seems that even if people can understand the rationality of an aggressive allocation in certain cases, their emotions can still get the best of them.

      Delete
  6. Wade - just to clear up the questions about the scale we used in the photo, both the S&P 500 and the risk tolerance scale are set to a mean of 0 and standard deviation of 1 so that one can make a valid comparison of two series with very different means. It is the most statistically defensible way to graphically illustrate correlation - we also measure the high raw correlation of the two series.

    ReplyDelete
  7. Anyone who uses the "perception" versus "risk tolerance" argument should read this paper:

    http://rof.oxfordjournals.org/content/17/3/847.abstract

    Wade's point about practical significance is very important. According to the company's website the changes we found would result in a 20% equity allocation shift.

    The risk tolerance questionnaire we used in our study is possibly one of the best out there. However, how risk tolerance questionnaires are constructed, and how a portfolio allocation is derived from a score, needs to be improved.

    ReplyDelete
    Replies
    1. Thank you Michael F. and Michael G., the study co-authors, for the comments.

      Your point about the 20% allocation shifts is very important and answers the question I was left with. It does sound like the movements in risk tolerance are translating into big differences in portfolio recommendations. Thank you for your work on this topic!

      Delete
  8. Hi all,

    Risk tolerance, as used by economists and decision scientists, is a trade-off measure. How much return do I require to bear a given level of risk (or how much excess utility above zero do I expect by bearing it, if the returns are good). For a given decision, it is combined with expected returns & expected risk to reach an investment decision.

    There are therefore three reasons someone may decrease their equity allocation:
    1. Their expectation of benefit (returns) has gone down.
    2. Their expectation of risk has gone up.
    3. Their tradeoff measure (risk tolerance) has gone down.

    The point is that a good measure of risk tolerance doesn't move much in response to market changes. By definition it shouldn't.

    However, a measure of how risky the individual *perceives* the stock market to be does. If I expect the same return, but a higher level of risk than I previously did, it may make sense to reduce my equity allocation. Whether or not my risk perceptions are too sensitive to recent events is a different (though related & relevant) question, and may relate to observations of volatility clustering in returns.

    Best,
    Dan

    ReplyDelete
  9. Q: Can risk tolerance be measured?
    A: Absolutely.

    Q: Are all risk tolerance surveys equal?
    A: Absolutely not.

    I have been tracking my personal risk tolerance via Riskalyze for approximately 5 years (through financial crisis, market rally and through 'today's all time stock market highs) and my risk tolerance has not changed with the markets.

    Risk tolerance will likely change over time if/as one's financial position changes (loss job, inheritance, winning lottery ticket, etc), but typically not by a meaningful degree.

    Using subjective/psychologically weighted risk surveys does not accurately identify risk tolerance; thus any analysis using this weak data is suspect. I have used at least a dozen subjective risk surveys and 90% of them spit out terrible suggested allocations (far to risky for me).

    Riskalyze's patented quantitative/math based approach accurately identifies investor risk tolerance.

    ReplyDelete
  10. Wade,
    The question of the scale of the risk tolerance changes was what drove my first question in the Q&A - I was specifically trying to address this. The chart on this page (http://www.kitces.com/blog/markets-may-be-volatile-but-research-shows-risk-tolerance-isnt/) shows what I believe is the exact same FinaMetrica data on average risk tolerance that they used, but instead of graphing market indexes and risk tolerance on their "standardized" scale, I left each to be shown on its own scale. As this scaling reveals, while the changes may be statistically significantly different, they are still miniscule relative to the measure and would not lead to a material change in results. I don't know how to reconcile this data with what Finke and Guillemette are claiming, that the risk tolerance changes are a full 1 standard deviation through the financial crisis.

    Either way, it's ultimately still not clear to me whether this is evidence that risk tolerance is unstable, or simply that FinaMetrica's data does not sufficiently screen out risk perception. The difference is extremely important, as stable risk tolerance with unstable perceptions is an education issue, while unstable risk tolerance is a planning issue. So understanding which is moving and which is not is critically important to everything from crafting the proper financial planning solution, to planning around market volatility, to designing the proper tool to measure the constructs in the first place.

    But again, the bottom line is that I really don't know how to reconcile the data I'm showing in my blog post, and what Finke and Guillemette produced, except to imply that their standardized scale is overstating the actual amount of risk tolerance volatility that occurred.
    - Michael

    ReplyDelete
    Replies
    1. Michael, thanks. About trying to reconcile the two versions of the chart, Michael G.'s point that the fluctuations translate into 20% stock allocation differences seems to be the only quantification of scale I've seen either way, thus far. That gets at the practical significance of the volatility in the risk tolerance measure.

      Thank you for the clarification about risk perception. Now I understand your point, which would be the underlying question about whether risk tolerance is stable and it is just that the measure used is not adequate. I thought you were using the risk perception argument to defend the measure of risk tolerance used in the study.

      Delete
    2. Wade,
      I don't know how to reconcile Michael G's comments and the scale here. I'm very familiar with Finametrica - we use it in practice with our clients as well - and the depth of changes shown on the chart from my blog do NOT translate into 20% stock allocation differences.

      Here is FinaMetrica's default scoring system:
      0% 0 - 22
      20% 23 - 38
      40% 39 - 51
      60% 52 - 63
      80% 64 - 81
      100% 82 - 100

      The magnitude of scoring changes on the chart are only on the order of about 2-3 points TOTAL. Perhaps the average score moved from 52.5 to 50.5, which would cross a 20% "threshold" here, but only because of an extremely arbitrary application of the FinaMetrica system. Yet as Finametrica notes in their own materials, these thresholds are not meant to be that arbitrary (they also have "marginal" ranges on either end where people may overlap, recognizing that these thresholds are not precisely concrete). Michael G seems to be implying that moving from 61 to 53 (an 8 point change) means nothing but moving from 53 to 51 is a 20% allocation change, which is NOT how FinaMetrica themselves direct the use of their system and scale.

      It would appear to me that if Michael G is suggesting the changes resulted in a 20% allocation change, it's merely because they are applying the scale in an arbitrarily concrete manner. If you assume the asset allocation scales linearly across each score range, the reality is that the changes they're talking about in their research are more like a 2%-4% allocation change, nowhere near a 20% allocation change. A FinaMetrica 53 score might have been a 62% equity allocation, and a FinaMetrica 51 score might be about 59% in equities; this is NOT a 20% allocation change! Arguably, it's so small as to be questionable about whether it's material in practice at all, as I don't think FinaMetrica would argue their scoring scale is meant to pinpoint equity allocations within +/- 2% in the first place!

      I was trying to be polite in the Q&A during the academic session, but this is exactly what I was trying to clarify in my questions in the first place. Though since they didn't state during the presentation they were using FinaMetrica, I didn't want to assume this was the case (I didn't find out for certain they were using FinaMetrica data until later). But it continues to appear to me that the results are drastically overstating the magnitude of risk tolerance changes due to their standardized scale, and they're overstating the magnitude of portfolio changes it would entail by applying small changes on a semi-arbitrary scoring scale where the shift purely coincidentally crosses the 52-score boundary between 60% and 40% equity allocations.

      Respectfully,
      - Michael

      Delete
    3. Michael,

      Thanks for the further insights. I'll make sure that the study co-authors (Michael G. and Michael F.) see your comments, as I haven't seen the actual study and also don't know their reasoning for the 20% allocation shift. If your explanation is right, then I would certainly agree with you that it wouldn't be fair to claim that the changing average score would imply a 20% allocation shift, but again I'm really on the sidelines on this one.

      The data source was supposed to be anonymous, but certainly the cat is out of the bag now. As I understand it, FinaMetrica is the most respected provider of risk tolerance analysis, so we are indeed talking about the cream of the crop in terms of risk tolerance assessments.

      Wade

      Delete
    4. Michael's point is exactly correct - and one that needs to be emphasized in the paper. Normalization was used to show the very high correlation between average changes in risk tolerance scores and changes in the market. But the magnitude of these scores in terms of practical significance on recommended portfolio allocation is not that great. Our primary objective was to provide evidence of time variation in risk tolerance and to test some theories about why this time variation occurs. It wasn't to judge the usefulness of the risk tolerance test - which is how some may have interpreted it. The normalized table was used to illustrate this pattern of correlation. Michael is correct in stating that the changes in risk tolerance score were far more stable than changes in the S&P.

      Delete
    5. Change in the S&P 500: Approximately 53%
      Change in the risk tolerance score: Approximately 7%
      (January 2007-May 2012)

      Assuming the risk tolerance score is used in a linear manner to construct a portfolio you are looking at a 4% equity shift. I provided some further context for my 20% comment below.

      Delete
    6. Michael & Michael,
      Understood on your point regarding the correlation, but to me it's still entirely unclear whether this is a demonstration that risk tolerance is varying with market returns (given the small magnitude of the changes, apparently not much) or if this simply makes the point that even FinaMetrica's questionnaire is not quite "perfect" in its measure of risk tolerance because it's still being partially tainted by perception.

      In other words, by just looking at the correlation between FinaMetrica scores and risk tolerance - without a separate construct and measure for risk perception - it's still not clear whether you're even showing a correlation between market volatility and risk tolerance, or a correlation between market volatility and risk perception (where risk perception is causing a slight impact to FinaMetrica's score).

      I have to admit give the rather small magnitude of the shift (does anyone even know what the standard error is on FinaMetrica's measurement tool?) it still appears to me much more likely you're showing the former outcome, rather than the latter.

      Notably, virtually all of the other studies that show this implied relationship use far more inferior measurements for risk "tolerance" in the first place as well. The fact that FinaMetrica's results show less variability to market volatility than other more-perception-biased measures would also imply that what we're really seeing is risk perception variability, not risk tolerance variability.

      Granted, there may actually BE some risk tolerance variability in this mix somewhere. I'm certainly not in the camp that risk tolerance "never" changes. But I find these questionnaire-market-volatility correlation measures to be highly ambiguous at best, and just an outright confounding of risk tolerance and perception at the worst.

      Perhaps we should all be sitting down to do a joint study to test for risk tolerance AND perception against market volatility and try to establish more clearly which is driving the outcome. I'd hypothesize that if you had a SEPARATE measure for risk perception, and THEN you tested risk tolerance, and then you used the perception measure to control for it, you'd find that the risk tolerance-ex-perception results are even more remarkably stable.
      - Michael

      Delete
    7. Michael,

      As Wade said, I think we have to assume the FinaMetrica questionnaire is measuring "risk tolerance," and not "risk perception", since that is how it is being marketed. There was a study published in the Financial Services Review earlier this year that used FinaMetrica data. The authors found a positive relation between risk tolerance and positive stock market expectations. They also found risk tolerance to be lower for people who perceived the stock market to be riskier than it was two years ago. So it appears risk tolerance and risk perception moved together around the time period of the Great Recession. The FSR is slow at getting articles online so I will send you a copy of the paper when I have a chance.

      Delete
    8. Gibson, R., Michayluk, D. and Van de Venter, G. (2013). Financial Risk Tolerance: An Analysis of Unexplored Factors. Financial Services Review, 22, 23-50.

      Delete
    9. Michael,
      Until we have a factor analysis that separates out risk tolerance from risk perception, I still don't see whether we definitively know whether risk tolerance is really moving AFTER controlling for risk perception, or if all you're doing is simply showing that FinaMetrica's measure is "close but not perfect" (and thus is why their estimate of risk tolerance moves a little but not a lot).
      - Michael

      Delete
  11. This strikes me as having a correlations with a common measurement in medical practice- the pain scale. Medical practitioners long ago figured out that self-reported pain scales were essentially useless in aiding their diagnostic efforts. Self-reported pain means nothing. I suspect self-reported risk tolerance means nothing either. I'm not surprised that people say they're more tolerant of risk when they haven't had to deal with risk actually showing up for some time. We probably ought to find something that actually objectively measures risk tolerance- like seeing if people failed to rebalance or worse, sold out of risky positions after a downturn, and use that in studies like this rather than a simple survey.

    ReplyDelete
  12. I should have proofread that first sentence. Instead of "having a correlations with" it should read "being similar to".

    ReplyDelete
  13. According to various academic studies (including this one - http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1990811) risk tolerance is a relatively stable psychological trait. Fluctuations do occur however on average they are minimal.

    Just like Michael says above I cannot reconcile the data above with the reports from various risk tolerance questionnaire providers. I can only conclude that risk perception is being confused with risk tolerance and just like Mike McDaniels says above not all risk questionnaires are created equally.

    ReplyDelete
  14. The Weber, Weber and Nosix, (2013) paper seems to be used to show how risk perception changes but that risk tolerance is stable over time:

    http://rof.oxfordjournals.org/content/17/3/847.abstract

    However there are some major issues with this paper. Risk tolerance was mainly measured by one’s likert scale response to the following question: “It is likely I would invest a significant sum in a high risk investment.” This type of question is not theoretically sound and is too subjective to draw valid inferences. For example, what constitutes a “significant sum” or a “high risk investment?” The authors also disclosed that their sample didn't have the behavioral characteristics that we see in the general population (for example, they were not myopic with information use).

    The Gerrans paper has not been published: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1990811

    But here are a couple papers that have been published which show that risk tolerance changes with macroeconomic conditions:

    http://pfp.missouri.edu/documents/research/yao_domarketreturns.pdf
    http://www.federalreserve.gov/Pubs/Feds/2007/200766/revision/200766pap.pdf

    To provide some further context regarding my point about a 20% allocation shift, please refer to page 8 of the following document:

    http://www.riskprofiling.com/Downloads/User_Guide_To_Linking_Spreadsheet.pdf

    If a financial planner uses a standard set of allocations (0%, 20%, 40%, 60%, 80% and 100% growth) and wanted to stay within the "OK" risk range, then moving from a score of 51 to 55 might result in a large equity shift. However it is probably more likely that a planner would use the score in a more linear fashion. In that case you are still looking at a 4% shift in equities.

    ReplyDelete
  15. I have been slow to join this discussion because it started while I was in the last few days of a round the world business trip and since flying home to Sydney I have been dealing with a bad case of jet lag.

    As some will know, I am a FinaMetrica co-founder and am responsible for our research activities and our relationships with researchers around the world. It has always been our policy to support education and research by making our intellectual property and our data freely available to educators and researchers. We do ask that we see the results of any analysis at an early stage and an early draft of any material to be published. Our concern is that it is quite easy to misinterpret data and to draw false conclusions or express conclusions in a way that is easy to misinterpret. Unfortunately, having provided the researchers with the data, the first we knew of any analysis was when it was presented at Orlando.

    I am in contact with the researchers and hope to be able to post and publish a joint clarification in the near future. For the present I have three points to make.

    Firstly, any measurement system will include an error factor. A system which aims to measure a psychological trait will probably.almost certainly not give a pure measure. It is very important to understand the limitations of a score before drawing conclusions about the underlying trait. That did not happen in the present instance.

    Secondly, risk tolerance does not drive the asset allocation decision. Rather it is a constraint upon an investment strategy which would otherwise achieve the client's goals with the desired likelihood. It is a relatively loose constraint representing a reasonably broad comfort zone which shades either way through to either too much or too little risk.

    Thirdly,I would have thought that our explanation of how to use the results of our test in the advising process was clear, as was the explanation of how to link risk tolerance scores to asset allocations in our calculators and guide, but apparently not. So let me be clear now. It is completely ridiculous to suggest that a change of five points in a FinaMetrica risk tolerance score would result in a 20% change in an asset allocation.In fact, in all probability, it would not result in any change at all.

    In my absence, Michael Kitces has made comments very similar to those I would have made. Thank you Michael.

    To our knowledge the only other robust risk tolerance test that was used across the recent market turmoil was developed by the very impressive behavioural finance team at Barclays Wealth led by Greg Davies,the www.investmentphilosophy.com. They also report stable risk tolerance.

    Finally, the Gerrans paper to which Michael Guillemette refers has been accepted for publication by the Journal of Accounting and Finance. It reports stable risk tolerance in a longitudinal study across the recent market turmoil involving test/retest data for more than 3000 individuals controlling for demographic, social and economic and regional factors.

    ReplyDelete
  16. Geoff and all the Michaels,

    Thank you for all of the comments. It has been very instructive.

    I'd say that the way the results were presented at Orlando would suggest to listeners (or, at least, to me) that our ability to measure risk tolerance, or at least to distinquish risk tolerance from risk perception, is very limited. But I can see from this discussion that it isn't the case at all.

    It would be very alarming if the risk tolerance measure would suggest 20 percentage point changes in stock allocations over time, as that would feed into the notion of buying high and selling low, but it is clear from the discussion -- acknowledged now from both sides -- that this would not be a proper way to interpret the results.

    The discussion shows that the changes in stock allocation may be somewhere between 0-4%. Whether or not this is even statistically significant (Michael K. asked this, but it has been answered) may only be of interest to academics. The important point is that these results suggest that there is little practical significance to the fluctuations in the risk tolerance measure.

    To conclude: the title of this blog entry is asking the wrong question, as this study has nothing to do with whether we can measure risk tolerance. What we do learn, at least, is that investors are not being ill-served by any sorts of fluctuations in the risk tolerance measure which happen to be correlated with the S&P 500.

    Thank you all.

    ReplyDelete
  17. Hi Michael G,

    Thank you for providing some new citations - I'll check them out.

    WRT the Weber/Nosic paper, I was heavily involved in building the referenced risk tolerance scale, and the longitudinal survey methodology. We were quite thoughtful about both theoretical and methodological issues. A review paper on risk attitude elicitation resulted from our work, and can be found here:
    http://onlinelibrary.wiley.com/doi/10.1002/9780470400531.eorms0169/abstract
    It touches on standard psychological, economic, and even behavioral measurement of risk tolerance, and interpretability. Please let me know any thoughts you have.

    I have to admit, I'm not aware of any theories under which Likert scale elicitation is unsound. Perhaps you mean "can be mathematically converted into a risk aversion coefficient", which it is true, they cannot. Whether or not that matters really depends if a mathematical coefficient is necessary for your analysis. However, most standard economic elicitation methods which do directly yield risk aversion coefficients are so subject to framing and reliability issues that they are even less useful.

    And indeed, the whole point of using subjective terms such as a 'a significant amount' is that they make the question relevant to the individual's sense of risk. Methods that use either absolute numbers (say $50,000) or percentages of wealth (10%) mean different things depending on your starting wealth level, and how numerate you are.

    In comparison, Likert scales score highly on the standard desiderata for this type of measurement (discussion in the paper).

    As for the self-selection, I can only see that as correct and a virtue of the study. If a study of what makes sports gamblers choose a given team excluded people who don't follow the sport at all, would you criticize it for self selection? I think it's correct that relatively little weight be given to the investing behavior of people who don't invest.

    Kind regards,
    Dan Egan

    ReplyDelete
  18. As a long-time skeptic of "risk tolerance" questionnaires, I am intrigued by the apparent passion around this topic. That can be explained by the deep desire we all have to apply as much science as we can to the service of clients. Still, as I have spoken to on several public occasions,I am not persuaded that these questionnaires do more than cause advisors and their clients to actually talk about the 2 key concepts of risk...both the psychological tolerance for it and the financial capacity to withstand the negative risk events. We have very robust tools to model the latter. I don't believe we have any really reliable tools to measure the former.

    I say this from several decades experience in advising thousands of clients. Psychological risk tolerance is not a worthwhile a priori input. People routinely falsely report it or can't really self-assess. Further, it changes over time, due to various factors, including the client's age and other non-portfolio wealth factors. But, most importantly, it is coachable...at the outset and on an ongoing basis... to help make the client comfortable with and then persevere in the portfolio allocation most likely to achieve the client's objectives. It is a derived conclusion, not an input...in my experience. And thus arriving at it is necessarily more an art than a science.

    At the end of the day, I think such questionnaires have some value...but chiefly as an introduction to the conversation about matching the client's tolerance to their capacity...or need.

    Thanks, for the opportunity to comment,

    Tim Kochis

    ReplyDelete
  19. Dan,

    By theories I just meant expected utility theory or prospect theory. I don't see how the question, “It is likely I would invest a significant sum in a high risk investment.”, can be tied to either of those theories. Ideally we would like to be able to derive a risk aversion or loss aversion coefficient. However I realize that while "pension-gamble" questions theoretically measure risk aversion they are also too confusing for most people to understand. So we need risk tolerance questions based on EUT or prospect theory that are easier to understand (and the FinaMetrica survey includes some of those types of questions). As for the self-selection bias, my issue was that the sample appeared to include all non-myopic investors. The average retail investor is myopic, so I think having a sample with a mix of myopic and non-myopic investors (or all myopic investors) would have been more informative.

    All the best,
    Michael

    ReplyDelete