Title: Individual Risk Attitudes New Evidence from a Large, Representative, ExperimentallyValidated Survey
1Individual Risk AttitudesNew Evidence from a
Large, Representative, Experimentally-Validated
Survey
- Thomas Dohmen, Armin Falk, David Huffman, Uwe
Sunde - (IZA, Bonn)
- Jürgen Schupp, Gert G. Wagner
- (DIW, Berlin, and IZA, Bonn)
2Motivation
- Risk and uncertainty are pervasive in economic
life. - As a consequence, individual attitudes towards
risk play a central role in economic theory. - Predicted to have a decisive impact on almost
every important economic decision. - The aim of this paper is to improve understanding
and measurement of individual risk attitudes.
3Data
- Large, representative sample 22,019 individuals.
- 2004 Wave of the Socioeconomic Panel (SOEP)
- Extensive socio-demographic information. New
survey measures. - New survey measures.
-
- Complementary field experiment 450 individuals.
- Subjects answer same survey measures as in SOEP.
- Also make risky choices for real money.
4Survey Measures
- General Risk Question
- Open-ended, useful benchmark for more specific
Qs. - Asks about Willingness to take risks, in
general - 11-point response scale completely unwilling to
completely willing. - Respondent free to think about curvature of
utility,but also imagine stakes and probabilities
involved in taking risks. - Context Questions
- Same scale, but ask about Car driving, health,
etc. - Hypothetical Investment Question
- Willingness to invest in a realistic investment
opportunity, with explicit stakes and
probabilities. - Holds beliefs constant, measures risk preference.
5Outline of the paper
- Distribution of risk attitudes.
- Determinants.
- Does the survey measure predict behavior?
- Alternative measure of risk preference.
- Impact of context on risk attitudes.
- Compare predictive power of different measures.
6Heterogeneity and DeterminantsGeneral Risk
Question
7Distribution of Willingness to Take Risks
8Gender and Willingness to Take Risks
9Age and Willingness to Take Risks
10Age and Willingness to Take Risks
11Parental Ed. and Willingness to Take Risks
12Height and Willingness to Take Risks
13Height and Willingness to Take Risks
14Determinants of Risk Attitudes (Probit)
15Determinants of Risk Attitudes (Probit)
16Determinants of Risk Attitudes (Probit)
17Can Survey QuestionsPredict Behavior?
18Experimental Design
- 450 subjects.
- Representative sample of adult population in
Germany. - Subjects answer the general risk question.
- Subjects also make 20 choices for real money
- Each choice is between a lottery and a safe
option - Lottery 300 Euros or 0 Euros, with equal
probability. - Safe option X Euros for sure, where X varies
across choices. - X0 in first choice, increases to 190 in 20th
choice (by 10S) - One choice is randomly selected and implemented.
- Measure of risk taking how high does X have to
be, to induce a switch from the lottery to the
safe option?
19Validation of Survey Risk Measure
20Alternative Measureof Risk Attitudes
21Survey measure
- Hypothetical Investment Scenario
- Choose how much of 100,000 Euros to invest in a
hypothetical asset. - doubles the investment, or returns half, with
equal probability. - Possible amounts 0, 20K, 40K, 60K, 80K, 100K
- Holds beliefs constant across individuals.
- Differences can be attributed to differences in
curvatures of utility. - Incorporates context of real financial decision.
22Investment in Hypothetical Asset
23Determinants of Investment (Probit)
24Determinants of Investment (Probit)
25Determinants of Investment (Probit)
26CRRA Coefficients (interior responses)
27General Risk Attitude and CRRA Coeff.
28Context and Risk Attitudes
29Correlation of Risk Attitudes Across Domains
30Determinants in Different Contexts
- Do the exogenous factors have the same impact
across domains?
31Determinants in Different Contexts
- Do the exogenous factors have the same impact
across domains? - Gender, Age, Parental Education, and Height
- Affect risk attitudes in most domains.
- Effects are qualitatively similar to general
risk. - But magnitudes differ across domains, e.g. gender
matters more in driving than in financial
matters.
32Primary Determinants of Risk Attitudes
33Comparing Predictive Powerof Alternative
MeasuresWithin and Across Contexts
34Predicting Behavior in Different Contexts
- Regress behavioral outcomes sequentially on each
risk measure separately. - General risk, 5 domain-specific, and hypothetical
investment choice. - Behavioral outcomes span the five contexts
identified in the SOEP - Real-world portfolio choice, participation in
sports , occupational choice, smoking, migration,
life satisfaction, traffic violations. - Compare predictive power of different measures,
within and across contexts.
35Determinants of Behavior in Different Domains
36Determinants of Behavior in Different Domains
37Traffic Offenses and Willingness to Take Risks
38Traffic Offenses and Willingness to Take Risks
39Comparing Predictive Power
- Most measures predict behavior across several
contexts. - Suggestive of a single risk trait.
- Provides additional evidence that the measures
are behaviorally relevant. - However, best predictor of behavior in a given
context is always corresponding domain-specific
question. - Measure of risk attitudes should incorporate
situation-specific context, in order to predict
specific behavior. - E.g. ask about willingness to take risks in
health matters, in order to predict risky health
behaviors, rather than using a measure of
willingness to take risks in financial matters.
40Conclusions
- Self-reported willingness to take risks varies
across individuals. - Exogenous factors explain individual differences
- Women and older individuals are more cautious.
- Children of highly-educated parents and taller
individuals take more risks. - Survey measures are behaviorally relevant
predict actual risk-taking behavior in a field
experiment.
41Conclusions
- Heterogeneity at least partly driven by
differences in risk preferences, rather than
subjective beliefs. - Exogenous factors affect risk preferences in a
similar way across domains, but magnitude varies. - CRRA coefficients consistent with range typically
assumed in economic models.
42Conclusions
- Risk attitudes are strongly but imperfectly
correlated across contexts. - All survey measures predict across several
contexts. - But best predictor of behavior in a specific
context is a question framed in terms of that
context.