The risk preference parameter (also referred to as “risk aversion” or “risk tolerance”)
estimated through Euler equations. Browning, Hansen, and Heckman (1999) survey the literature
and review the general framework.
Survey techniques measure risk preferences in a more direct way, with methods parallel
to the ones developed to measure time preference. The key difference between time and risk
preference is that time preference describes the devaluation of rewards as a function of their
delay, whereas risk preference describes the devaluation of rewards as a function of their
uncertainty. Of course, in the real world, risk and delay are inextricably confounded. A standard
result (Gorman, 1968) shows that, in a separable model, intertemporal substitution and risk
aversion are indistinguishable attributes of preference specifications. In addition, all deferred
rewards carry with them some risk that they will not in fact be received. All consequences
involving risk lie in the future. If they were immediate, there would be no uncertainty to
negotiate. Despite the common discounting process, however, Green and Myerson (2004) argue
that the two constructs belong to different underlying psychological processes. As evidence, they
point out that the two constructs react differently to the same effect: for example, an increase in
the size of reward generally decreases the discount on time but increase it on probability.54 This
is evidence against the standard intertemporally separable model of risk aversion.
Survey questions assessing risk preference usually pose a series of questions involving
the choice between a lottery and a certain outcome: “Which would you prefer: $100 dollars
today, or a 50 percent chance of receiving nothing and a 50 percent chance of receiving $200?”
Two recent studies have introduced measures of risk preferences in field experiments. Harrison,
Lau, and Rutstrom (2007) use real stakes to elicit risk preferences on a representative sample of
253 people in Denmark. Dohmen et al. (2005) use a lottery experiment with a representative
Socioeconomic Panel (SOEP). In their study, the general SOEP question about “willingness to
take risks, in general” answered on an 11-point scale predicts self-reported risky behavior in the
domains of health, driving, financial matters, leisure, and so on better than does a typical lottery
question. Also, such a general question is free from framing effects that shape behavior in
presence of risk (Kahneman 2003). The findings for general samples suggest that preferences are
best measured in ways that do not require a high level of numeracy. As we previously argued for
time preference, the effects of numeracy and intelligence may not simply constitute
methodological artifacts, but are root explanations for behavior in the face of uncertainty. As for
time preference, in fact, there appears to be an inverse relationship between cognitive ability and
risk aversion, where higher IQ people have higher risk tolerance (Benjamin, Brown, and Shapiro
2006; Dohmen et al. 2007).55
Risk preference also varies with socioeconomic characteristics. However, there is no
general consensus on the direction of such differences: some studies find a negative relationship
between education and risk aversion (Weiss 1972; Belzil and Leonardi 2006; Ferrer-i-Carbonell
2005; Binswanger 1980, 1981; Guiso and Paiella 2001; and Andersen et al. 2005), while Barsky
et al. (1997) find an inverse U-shaped relationship, with risk tolerance peaking at 12 years of
education. There are some consistent patterns of a negative relationship between unemployment
duration and risk aversion, (see Feinberg 1977, and Kohn and Schooler 1978). Most of the
studies find that women are more risk averse than men56 (see Barsky et al. 1997; Donkers,
Melenberg, and van Soest 2001; Hartog, Ferrer-i-Carbonell, and Jonker 2002; Jianakoplos and
Bernasek 1998; see Eckel and Grossman 2008 for a review), but a few do not find this gender
difference (see Andersen et al. 2005; Harbaugh, Krause, and Vesterlund 2002; and Holt and
among children with fewer siblings and first-born children (Dohmen et al. 2006). Risk aversion
dips sharply in adolescence (Steinberg, 2004, 2007) and then throughout adulthood seems to
increase with age (Dohmen et al. 2005).
The empirical findings summarized in this section assume that risk preference can be
modelled with a single parameter across situations. Yet, like time preference, risk preference
may be multidimensional rather than unitary. Weber (2001) shows that risk preference varies by
domain, and a scale that assesses risk taking in five different domains shows low correlations
across these domains (Weber, Blais, and Betz 2002). One can be quite risk-averse when it comes
to financial decisions but risk-loving when it comes to health decisions (Hanoch, Johnson, and
Wilke 2006). Weber’s risk-return model of risk taking (Weber and Milliman 1997; Weber and
Hsee 1998) finds that low correlations among risk taking preference across domains can be
explained by domain-specific perceptions of riskiness and return. Perhaps the relatively modest
predictive validity of risk aversion for actual risk-taking behaviour (for example, Barsky et al.
1997) might be improved considerably with a multidimensional and domain-specific approach to
its measurement.
A behavioral task and self-report measure from the psychology literature are of interest.
Lejuez and colleagues (2002, 2003) have developed a behavioral task for risk preference. Their
Balloon Analogue Risk Task (BART) is a computer game in which participants make repeated
choices between keeping a certain smaller monetary reward and taking a chance on an
incrementally larger reward. Scores on the BART correlate with real-world risk behaviors such
as smoking, stealing, and not wearing a seatbelt. BART scores also correlate with sensation
seeking, a trait proposed by Zuckerman (1994) and defined as “the tendency to seek novel,
sake of such experience.”57 More than 2000 published articles have incorporated sensation
seeking self-report questionnaires, and collectively these studies have established that sensation
seeking predicts risky driving, substance use and abuse, smoking, drinking, unprotected sex,
juvenile delinquency, and adult criminal behavior (see Zuckerman 2007 for a review). This
broadens the notion of risk aversion to include the enjoyment of risk, per se, at least over certain
ranges. Unfortunately, few, if any studies, have included typical risk preference propositions of
the sort relevant to economic decision making when sensation seeking is estimated.58 An
integration of these two endeavors would be useful.