Measuring Risk Taking

How do you see yourself: “Are you generally a person who is fully prepared to take risks or do you try to avoid taking risks? Please tick a box on the scale, where the value 0 means: ‘unwilling to take risks’ and the value means: ‘fully prepared to take risk’.”

“What share of your lottery winnings would you be prepared to invest in this financially risky, yet lucrative investment?”

Evidence of heterogeneity found in survey across individuals shows that willingness to take risks is negatively related to age and being female, and positively related to height and parental education.(1) using standard lottery question to measure risk preference, found similar results regarding heterogeneity and determinants of risk preferences. The lottery question makes it possible to estimate the coefficient of relative risk aversion for each individual in the sample. Using five questions about willingness to take risks in specific domains — car driving, financial matters, sports and leisure, career, and health — the paper also studies the impact of context on risk attitudes, finding a strong but imperfect correlation across contexts. Using data on a collection of risky behaviors from different contexts, including traffic offenses, portfolio choice, smoking, occupational choice, participation in sports, and migration, the predictive power of all of the risk measures was compared.

The first question asks for attitude towards risk in general, allowing respondents to indicate their willingness to take risks on an eleven point scale, with zero indicating complete unwillingness to take risks, and ten indicating complete willingness to take risks.

The next five questions all use the same scale, and similar wording, but refer to risk attitudes in specific contexts: car driving, financial matters, leisure and sports, career, and health. All of these measures are characterized by ambiguity, in the sense that they leave it up to the respondent to imagine the typical probabilities, and stakes, involved in taking risks in a given domain.The last risk question is different, in that it corresponds more closely to the lottery measures used in previous studies. The question presents respondents with the following choice: Imagine you had won 100,000 Pounds in a lottery. Almost immediately after you collect, you receive the following financial offer from a reputable bank, the conditions of which are as follows: There is the chance to double the money within two years. It is equally possible that you could lose half of the amount invested.

Respondents are then asked what fraction of the 100,000 Pounds they would choose to invest, and are allowed six possible responses: 0, 20,000, 40,000, 60,000 80,000, or 100,000 Pounds.5 This measure shares the common feature of other lottery measures in that it presents respondents with explicit stakes and probabilities, and thus holds risk perceptions constant across individuals. Because beliefs are held constant, differences in responses are more clearly attributable to risk preference alone, as compared to the six measures above, which potentially incorporate both risk preference and risk perceptions.

“Willingness to invest in a hypothetical lottery with explicit stakes and probabilities”

We investigate the relationship between willingness to take risks and selected personal characteristics: gender, age, height, and parental background. We focus on these characteristics because they are plausibly exogenous and therefore allow causal interpretation. The analysis reveals several facts: (1) women are less willing to take risks than men, at all ages; (2) increasing age is associated with decreasing willingness to take risks; (3) taller individuals are more willing to take risks; (4) individuals with highly-educated parents are more willing to take risks. These effects are large and very robust, with the exception of parental education, which becomes insignificant in some specifications. This evidence on determinants has important implications. For example, differences in risk preferences could be one factor contributing to the well-known gender wage gap, gender-specific behavior in competitive environments, and gender differences in career choice. The impact of age implies increased financial conservatism in ageing societies, and the height result points to a possible mechanism behind the higher earnings potential of taller individuals. These four findings also suggest characteristics that can be used to partially control for risk attitudes in the absence of direct survey measures field experiments with a representative subject pool can be used to validate survey measures, in order to end up with both statistical power and confidence in the reliability of the measures. To test the validity of our survey measures, we conducted a field experiment in which participants had the opportunity to make risky choices with real money at stake; it was found that answers to the general risk question are good predictors of actual risk-taking behavior in the experiment.

A fundamental question surrounding the notion of risk attitudes is the relevance of context. In economics it is standard to assume that a single, underlying risk preference governs risk taking in all domains of life. In line with this assumption, economists typically use a lottery measure of risk preference, framed as a financial decision, as an indicator of risk attitudes in all other contexts, e.g., health. Some psychologists and economists, however, have questioned whether stable utility functions and risk preferences exist at all, given that risk attitudes appear to be highly malleable with respect to context in laboratory experiments. An alternative interpretation of this evidence, of course, is that a stable risk preference does exist, but that individuals believe the typical risk in one context is greater than in another, and indicate different willingness to take risks accordingly. Average willingness to take risks turns out to differ across contexts. However, the correlation across contexts is quite strong. Principal components analysis tells a similar story: one principal component explains the bulk of the variation, suggesting the presence of a single underlying trait, but each of the other components still explains a non-trivial amount of the variation.

There has been evidence of differences in risk perception. In fact, risk perceptions are known to vary across individuals based on evidence from psychology.1 A number of studies have asked directly about risk perceptions and have documented a tendency for women to perceive dangerous events, such as nuclear war, industrial hazards, environmental degradation, and health problems due to alcohol abuse, as more likely to occur, in conditions where objective probabilities are difficult to determine.

Willingness to take risks in health matters is a better predictor of smoking than the hypothetical investment question, or the general risk question, or any other domain-specific question. Clearly, women are more likely to choose low values on the scale and men are more likely to choose high values. Clearly, the proportion of individuals who are relatively unwilling to take risks, i.e., choose low values on the scale, increases strongly with age. For men, age appears to cause a steady increase in the likelihood that an individual is unwilling to take risks. For women, there is some indication that unwillingness to take risks increases more rapidly from the late teens to age thirty, and then remains flat, until it begins to increase again from the mid-fifties onwards. It is important to note that this relationship could reflect a direct effect of age on risk preferences, but could also be driven by cohort effects, i.e., society wide changes in risk preferences over time, perhaps due to major historical events.

The difference in age patterns for men and women makes it less credible that the change in risk attitudes is attributable to cohort effects, because major historical events are likely to affect both men and women at the same time, but it is difficult to definitively disentangle the two explanations with the data available. The most important economic variables that need to be controlled for are measures for income and wealth. High income or wealth levels may increase the willingness to take risks because they cushion the impact of bad outcomes.

A variety of other personal and household characteristics were studied in relation to risk taking behaviours. These characteristics, which are all potentially endogenous, include among others: marital status, nationality, employment status (white collar, blue collar, private or public sector, selfemployed, non-participating), education, subjective health status, and religion.

Additional individual characteristics such as wealth, debt, household income, marital status, number of dependent children, country of residence before unification, foreigner status, schooling degree, employment status, occupational choice, employment rank, public and private sector employment, life satisfaction, general health status, smoking, and weight.
In economics it is standard to assume the existence of a single risk preference governing risk taking in all contexts. In line with this assumption, economists typically use a lottery measure of risk preference, framed as a financial decision, as an indicator of risk attitudes in all other contexts. By contrast, there is considerable controversy on this point in psychology. Based on laboratory experiments in which self-reported risk taking is only weakly correlated across different contexts, some studies conclude that a stable risk trait does not exist at all.

But little or no correlation would provide evidence against the standard assumption; a strong correlation would suggest the existence of a stable risk preference. Exploring the determinants of willingness to take risks in specific contexts, evidence indicates that the same factors determine risk attitudes across contexts would also lend support to the notion of a stable risk preference.

Another way of assessing the stability of risk attitudes is to check what fraction of individuals is relatively willing to take risks for all of the different measures. It turns out that 51 percent of individuals are willing to take risks in all domains under question (6 domains) and more than 1 third are willing in at least five domains. The relatively large correlation across contexts, and the stability of an individual’s disposition towards risk across domains strongly suggest the presence of a stable, underlying risk preference. The consistency across domains is not perfect, and could indicate some malleability of risk preferences, but it seems more likely that this variation reflects the risk perception component of the measures, e.g. a tendency for most people to view car driving as more risky than sports and thus state a relatively lower willingness to take risks in car driving.

Given that the questions ask about ”willingness to take risks,” it is possible that individuals could think of the same gamble, in utility terms, across contexts, in which case any variation at all in willingness to take risks would be inconsistent with stable risk preferences. It seems more likely, however, that individuals imagine the typical risk they expect to encounter in each context, based on their subjective beliefs, and state their willingness to take this risk. In this case, the pattern we observe would reflect a stable risk preference but varying risk perceptions.

Overall, having a high educated parent increases willingness to take risks. A more highly-educated mother is associated with a higher willingness to take risks in all domains, except for car driving and health. In summary, findings suggest the existence of a stable, underlying risk preference. One source of evidence is the strong correlation of risk attitudes across contexts, and the finding that a single principal component explains the bulk the variation in risk attitudes. Another piece of evidence is the fact that risk attitudes have similar determinants in all contexts, in the form of the four exogenous factors. There is some variation in risk attitudes with respect to context, but this seems likely to reflect variation in risk perceptions across contexts. Differences in risk perception could also potentially explain why the exogenous factors have effects of varying magnitudes across contexts, but a more detailed investigation of why, e.g., the gender effect is stronger in some contexts than others could be an interesting subject for future research.

The question is raised whether the survey instruments reliably predict risky behavior, despite the fact that they are not incentive compatible and are therefore potentially behaviorally irrelevant? The answer to this question is of great importance both from a methodological and a practical point of view. Second, how do the different risk measures compare in terms of predictive power? In particular, how do the alternative measures fare, compared to the more standard measure of risk preference, and how do context-specific measures perform within and outside of their corresponding context? For example, is smoking best predicted by a health related risk question or is it equally well explained by a general risk or hypothetical lottery question? In the past, economists have typically used only a single question, most often a hypothetical lottery question to predict risk taking behavior in all contexts. To address our questions, we use a collection of behaviors which includes portfolio choices, participation in sports, occupational choice, smoking, migration, life satisfaction and traffic offenses.

In summary the survey found that each one of our seven risk measures predicts several behaviors. We can therefore reject the hypothesis that the measures are behaviorally irrelevant. This is especially true for the general risk question, which is the only measure to predict all of the behaviors. The fact that this measure is capable of predicting risky behaviors across very different domains of life suggests once more the existence of an underlying risk trait that is not specific to a particular domain. Interestingly, the general risk question seems to capture this trait much better than the hypothetical risk question. This latter measure not only fails to predict important behaviors but in some cases appears to make the wrong prediction. In this sense, conclusions made are qualified that a significant correlation between a set of behaviors similar to the ones we study and a risk measure similar to our hypothetical investment question. Even though we think our results support the assumption of a stable underlying risk preference, our analysis also shows that individual risk perceptions vary significantly across domains. In order to predict domain-specific risk taking behavior, it is therefore indispensable to use domain-specific risk questions. Using, e.g., simple lottery questions can only be considered an inadequate substitute for measures using situation-appropriate context.

To summarize the results of survey, the first finding is that the distribution of willingness to take risks exhibits substantial heterogeneity across individuals. Second, these individual differences are partially explained by differences in four exogenous factors: willingness to take risks is negatively related to age and being female and positively related to and height and parental education. A third important finding follows from the main methodological contribution of the paper: the survey measures are shown to be behaviorally relevant, in the sense that they predict actual risk-taking behavior in our field experiment. Fourth, estimates of the coefficient of relative risk aversion for the sample provide support for the range of parameter values typically assumed in economic models. A fifth finding is that risk attitudes are strongly but imperfectly correlated across different life contexts. This provides some support for the standard assumption of a single underlying trait, but also points to a value-added from asking context-specific questions, in order to capture variation in risk perceptions. The sixth finding is that gender, age, parental education, and height have a qualitatively similar impact on risk attitudes in most contexts, but that the magnitude differs across contexts. A seventh finding is that the survey measures can predict a wide range of important behavioral outcomes, including portfolio choice, occupational choice, smoking, and migration. An eighth finding is that the general risk question is the best allaround predictor of these behaviors, outperforming a lottery measure or domain-specific measures. Ninth, the best predictor of behavior within a given context is typically a question incorporating the corresponding context, as opposed to a lottery measure or measures incorporating other contexts.

Demographic changes leading to a large population of elderly are predicted to lead to a more conservative pool of investors and voters, which could substantially influence macroeconomic performance and political outcomes, increase the resistance to reforms, and delay necessary but risky policy adjustments. In addition to adding to knowledge about risk attitudes, some of these findings have potentially important policy implications. A robust and pervasive gender difference in risk attitudes could play some role in explaining different labor market outcomes, and investment behavior, observed for men and women.

An age profile for risk attitudes was found to have important ramifications, at the macroeconomic level. Although it is found that risk preferences are relatively stable across situations, an age profile also raises questions about the stability of risk preferences over time. A role for parental education in shaping the risk attitudes of children highlights a potentially important role of education policy.



Extracted from:
Fowler, F. (1988): Survey Research Methods. Newbury Park, London.
Frey, B. S., and A. Stutzer (2002): Happiness and Economics: How the Economy and Institutions Affect Well-Being. Princeton University Press, Princeton and Oxford,
1st edn.36
Van Praag, B. M. S., and A. Ferrer-i Carbonell (2004): Happiness Quantified - A Satisfaction Calculus Approach. Oxford University Press, Oxford.
Weber, E., A. R. Blais, and N. Betz (2002): “A Domain-Specific Risk-Attitude Scale: Measuring Risk Perceptions and Risk Behaviors,” Journal of Behavioral Decision Making, 15, 263–290.
Byrnes, J., D. Miller, and W. Schafer (1999): “Gender Differences in Risk Taking: A Meta-Analysis,” Psychological Bulletin, 125, 367–383.
Camerer, C., and R. Hogarth (1999): “The Effects of Financial Incentives in Experiments: A Review and Capital-Labor-Production Framework,” Journal of Risk and
Uncertainty, 19(1), 7–42.