American Economic Association

Financial Decision-Making: Are Women Really More Risk-Averse?

Author(s): Renate Schubert, Martin Brown, Matthias Gysler and Hans Wolfgang
Brachinger
Source: The American Economic Review, Vol. 89, No. 2, Papers and Proceedings of the One
Hundred Eleventh Annual Meeting of the American Economic Association (May, 1999), pp.
381-385
Published by: American Economic Association
Stable URL: https://www.jstor.org/stable/117140
Accessed: 12-09-2018 17:40 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

American Economic Association is collaborating with JSTOR to digitize, preserve and extend
access to The American Economic Review

This content downloaded from 121.52.158.248 on Wed, 12 Sep 2018 17:40:09 UTC
All use subject to https://about.jstor.org/terms
 GENDER AND ECONOMIC TRANSACTIONS
Financial Decision-Making: Are Women Really
More Risk-Averse?
By RENATE SCHUBERT, MARTIN BROWN, MATrHIAS GYSLER,
AND HANS WOLFGANG BRACHINGER *
A widespread view concerning financial
decision-making is that women are more risk-
averse than men. A consequence of this ste-
reotype is statistical discrimination which
diminishes the success of women in financial
and labor markets. The perception that female
managers are less risk-prone than men has
been put forward as a major cause of "glass
ceilings" in corporate promotion ladders
(Johnnie E. V. Johnson and Philip L. Powell,
1994). Women are less trusted than men to
make the risky decisions that may be neces-
sary for a firm's success. Anecdotal evidence
from financial markets suggests that similar
stereotyping by investment brokers is possibly
to the disadvantage of female clients. Women
are expected to be more conservative investors
than men and are consequently offered invest-
ments with lower risks and therefore lower ex-
pected returns (Penelope Wang, 1994).
Given that preconceptions concerning the
risk propensities of men and women seem to
affect their economic success, it is important
to examine whether the above stereotype re-
flects actual economic behavior. There is
some evidence supporting the view that women
are more risk-averse than men in financial
decision-making. Recent survey data suggest
that wealth holdings of single women are less
risky than those of single men of equal eco-
nomic status (Nancy Ammon Jianakoplos and
* Schubert, Brown, and Gysler: Center for Economic
Research, Swiss Federal Institute of Technology, CH-
8092 Zurich, Switzerland; Brachinger: Statistics Seminar,
University of Fribourg, CH-1700 Fribourg, Switzerland.
We thank Armin Falk, Ernst Fehr, Simon Gachter, and
Patrik Schellenbauer for helpful comments. Reto Bleisch,
Hannes Egli, and Michael St6ve provided excellent re-
search assistance.
381
This content downloaded from 121.52.158.248 on Wed, 12 Sep 2018 17:40:09 UTC
All use subject to https://about.jstor.org/terms
Alexandra Bernasek, 1998; Annika E. Sunden
and Brian J. Surette, 1998). Also, when asked
about their attitudes toward financial risks,
women seem to report a lower risk propensity
than men (Robert B. Barsky et al., 1997). Fi-
nally, experimental evidence suggests that
women may be more risk-averse than men to-
ward gambles (Irwin P. Levin et al., 1988).
It is questionable, however, whether ste-
reotypic risk attitudes can be confirmed by the
above survey and experimental evidence.
First, in survey data, gender-specific risk atti-
tudes may be confounded with differences in
individual opportunity sets. Survey studies of
individual wealth composition provide only
weak control for gender-specific information on
choice options and gender-specific wealth con-
straints in the underlying financial decisions.
Second, behavior in abstract gambling ex-
periments may not correspond to risk behavior
in contextual decisions, as has been shown by
John Hershey and Paul J. H. Schoemaker
(1980). Thus, while abstract gambling exper-
iments provide strong control of the economic
environment surrounding risky decisions, their
data may not be adequate for drawing conclu-
sions on gender-specific risk attitudes of in-
vestors and managers.
The objective of the study reported here is
to circumvent the above problems in analyzing
gender-specific risk attitudes. We conducted
an experiment in which subjects not only made
abstract gambling decisions, but were also
confronted with financially motivated risky
decisions embedded in an investment or in-
surance context. The implementation of such
contextual decisions allowed us to examine
gender-specific risk propensities in contexts
which may be relevant to financial and labor
markets. At the same time, compared to survey
 382 AEA PAPERS AND
studies, the application of experimental meth-
ods provided stronger control of the economic
environment in which decisions were made.'
We elicited data which call into question the
prevalence of stereotypic risk attitudes in fi-
nancial decision-making. First, we find that the
comparative risk propensity of male and fe-
male subjects is strongly dependent on the fi-
nancial decision setting. Second, and more
important, we observe no gender differences
in risk propensity when subjects face contex-
tual decisions. Since in practice financial de-
cisions are always contextual, our results
suggest that the above gender stereotype may
not reflect true male and female attitudes to-
ward financial risks.
I. Gender-Specific Risk Behavior: An
Experimental Design
Our experiment was designed to examine
gender-specific risk propensity in decisions
relevant for investors and managers. Table 1
gives an overview of the experimental design.2
In the main treatment, we implemented risky
choices in the form of investment and insur-
ance decisions. Choice behavior in this treat-
ment (henceforth called the context treatment)
directly measures the risk behavior of male
and female subjects in contextual financial de-
cisions. We also ran a control treatment in
which the same risky choices were presented
as abstract gambling decisions. This control
treatment (henceforth called the abstract treat-
ment) allows us to validate the risk behavior
induced by our experimental procedure in
the light of the above-mentioned gambling
evidence.
In the context treatment, subjects were con-
fronted with risky choices in two different de-
cision contexts. Subjects first had to complete
a series of investment decisions. They were
then presented with a series of identical
choices, this time, however, framed as insur-
' Wealth effects from income differences outside the
laboratory may still affect risk behavior in our experiment.
They are controlled for in our model.
2Experimental instructions are available in German
from the authors or in English from our web page:
(www.wif.ethz.ch/publikationen.htm).
This content downloaded from 121.52.158.248 on Wed, 12 Sep 2018 17:40:09 UTC
All use subject to https://about.jstor.org/terms
PROCEEDINGS MAY 1999
TABLE 1-EXPERIMENTAL DESIGN
Treatment Subjects Decision frames
Contexta 36 males, 32 females investment
(gain domain)
insurance
(loss domain)
Abstractb 40 males, 33 females gain-gambling
(gain domain)
loss-gambling
(loss domain)
a Contextual frames.
Abstract gambling frames.
ance decisions. Each investment and insurance
decision incorporated a choice between a risky
lottery and a certain payoff. With the series of
choices implemented in the two frames we
elicited our subjects' certainty equivalents for
the same four risky lotteries (LI, L2, L3, L4)
in both contexts.:
We implemented two frames in the context
treatment in order to measure the risk propen-
sity of subjects toward both perceived gains
and perceived losses.4 In both frames, all pos
sible payoffs for each decision were positive.
In the investment frame these payoffs were
also presented as gains, while in the insurance
frame the same payoffs were presented as
losses relative to an initial endowment. There-
fore, elicited certainty equivalents in the in-
vestment frame measure subjects' risk
propensities in the gain domain, while cer-
tainty equivalents in the insurance frame mea-
sure risk propensity in the loss domain. If, in
contextual financial decisions, female subjects
are generally more risk-averse than men, our
results should display, ceteris paribus, lower
female certainty equivalents in both the in-
vestment and insurance frames.
'The four lotteries (LI, L2, L3, and L4) each had two
possible outcomes. Payoffs in Swiss francs (1 SFr =
$0.60) and their probabilities were (30 SFr, '/6; 10 SFr,
5/6), (30 SFr, 1/2; 10 SFr, 1/2), (30 SFr, 5/6; 10 SFr, 1/6), and
(50 SFr, 1/2; 20 SFr, I/2), respectively.
4Experimental evidence suggests that individual risk
propensity will vary systematically between the gain and
loss domain (Daniel Kahneman and Amos Tversky,
1979).
VOL. 89 NO. 2 GENDER AND ECONOMIC TRANSACTIONS 383

In the abstract treatment we confronted a Context Treatment

different group of subjects with exactly the
same risky decisions as those in the main treat- 2 E Investment Insurance
ment. However, as mentioned before, all
choices in this control treatment were framed 1W
U-
as abstract gambling decisions. Again, two
frames were implemented. In the gain-
gambling and loss-gambling frames we mea-
-2
sured the risk attitudes of subjects toward
LI L2 L3 LA LI L2 L3 LA
gambles in the gain and loss domain,
respectively.
Subjects in both treatments were told in ad-
Abstract Treatment]
vance that one of their choices would deter-
mine their experimental earnings.5 Further, all 2 2 Gain-Gambling Loss-Gambling
subjects completed a post-experimental ques-
tionnaire which yielded information on each
subject's disposable income. This information
0 -1
is necessary to exclude wealth effects due to
income differences outside the laboratory as
-2
an explanation of gender-specific choice
LI L2 L3 L4 LI L2 L3 L4
behavior.
FIGURE 1. DIFFERENCES IN MEAN CERTAINTY
II. Results EQUIVALENTS (CE) BETWEEN MALE AND FEMALE
SUBJECTS, BY FRAME

A. Descriptive Evidence Note: Certainty equivalents (and male-female differ-

ences) are measured in Swiss francs.

Our experiment was conducted with under-

graduates from different fields at the Univer-
sity of Zurich and the Swiss Federal Institute
of Technology (ETH). Figure 1 displays the
data from the context and abstract treatments.
The differences between average male and fe-
male certainty equivalents are shown for all
four lotteries in each decision frame.
In Figure 1 positive differences between
mean certainty equivalents of male and female
subjects imply lower average female certainty
equivalents and thus less risky female choice
behavior. The figure shows that, in the context
treatment, female subjects did not generally
make less risky choices than male subjects.
Average male and female certainty equivalents
do not display systematic differences in either
the investment or the insurance frame. How-
ever, the situation is completely different in
the abstract treatment. Here female subjects
display systematically lower certainty equiv-
5 On average, subjects earned 38 SFr (roughly $24)
including a 10 SFr show-up fee. Earnings were paid in
cash immediately after the experiment.
alents in the gain-gambling frame, while they
display more risk-seeking behavior in the loss-
gambling frame.
13. Gender-Specific Risk Attitudes
In the above results, gender differences in
risk attitudes are possibly confounded with
wealth effects due to gender-specific income
differences outside the laboratory. We there-
fore conducted a regression analysis to test for
significant gender differences in risk behavior
controlling for possible wealth effects. Sepa-
rate regressions were run for each frame in
both treatments. Individual certainty equiva-
lents, being our measure of individual risk pro-
pensity, were entered as the dependent
variable.
The data in each decision frame represent
repeated measures since certainty equiva-
lents of each subject were elicited for four
different lotteries. Consequently, in order to
allow for differences in behavior over ran-
domly selected subjects, a model with random

This content downloaded from 121.52.158.248 on Wed, 12 Sep 2018 17:40:09 UTC
All use subject to https://about.jstor.org/terms
 384 AEA PAPERS AND
intercepts varying over individuals was esti-
mated by generalized least-squares (GLS)
analysis.6 We further controlled for differ-
ences in behavior over lotteries by entering
three lottery dummies "L2," "L3," and
"L4" in each regression.
Gender differences in elicited certainty
equivalents were captured by the gender
dummy "Female." The coefficient of Female
measures, for any lottery, the average rise in
certainty equivalents from male to female sub-
jects. Finally, to control for wealth effects,
"Income" (measuring disposable income per
month in thousands of Swiss francs) was in-
troduced as an explanatory variable.?
The regression results for the context and
abstract treatments are reported in Table 2.
Our two main findings are the following:
(i) As suggested by Figure 1, male and fe-
male subjects do not differ in their risk
propensities toward contextual decisions.
In both the investment and insurance
frame the estimated coefficient for Fe-
male is not significant.8
(ii) In abstract gambling decisions, however,
gender differences in risk propensity do
arise. In the gain-gambling frame the es-
timated coefficient of Female is negative
and of borderline significance. In this
frame, female subjects are more risk-
averse than male subjects, confirming the
results of earlier gambling experiments.
Yet, this result is reversed in the loss-
gambling frame. Here the coefficient of
the gender dummy is significantly posi-
tive, suggesting that female subjects are
more risk-prone than male subjects.
Table 2 further indicates that disposable in-
come outside the laboratory was irrelevant for
6 A Lagrange multiplier test favors this random-effects
model over ordinary least-squares analysis in each frame.
'Preliminary analyses showed that interactions of L2,risk attitudes. Under controlled economic con-
L3, L4, and Income with the gender dummy did not con-
tribute to the explanatory power of the model. All relevant generally make less risky financial choices
gender differences in certainty equivalents are therefore
captured by Female.
8 This result is in stark contrast to evidence found by
Melanie Powell and David Ansic (1997) in experimental
insurance decisions.
This content downloaded from 121.52.158.248 on Wed, 12 Sep 2018 17:40:09 UTC
All use subject to https://about.jstor.org/terms
PROCEEDINGS MAY 1999
TABLE 2-GENDER-SPECIFIC RISK PROPENSITY,
CONTROLLING FOR WEALTH EFFECTS
Context treatment Abstract treatment
Independent Gain- Loss-
variable Investment Insurance gambling gambling
Intercept 16.09* 17.00* 14.23* 14.89*
(0.65) (0.84) (0.57) (0.72)
Female -0.37 -0.73 -o.9g 1.32*
(0.59) (0.78) (0.46) (0.6)
Income -0.74 -0.58 2.03* 1.30t
(0.5) (0.66) (0.59) (0.72)
L2 4.15* 4.8* 4.35* 4.8*
(0.42) (0.51) (0.41) (0.53)
L3 9.9* 10.45* 9.32* 8.73*
(0.42) (0.50) (0.41) (0.53)
L4 18.66* 18.95* 18.75* 17.97*
(0.44) (0.50) (0.42) (0.53)
R2: 0.83 0.77 0.86 0.78
Number of
observations: 254 248 269 260
Notes: Not all certainty equivalents of each participant were sub-
jected to regression analysis due to missing decisions and income
statements (8.6 percent of the data were excluded). Standard errors
are reported in parentheses.
Statistically significant at the 10-percent level.
* Statistically significant at the 5-percent level.
contextual decisions in our experiment. The
estimated coefficient on Income is insignifi-
cant in the context treatment. Interestingly,
wealth effects seem to matter in abstract gam-
bling decisions, with higher income increasing
risk tolerance. Finally, Table 2 shows that the
increase in expected payoffs from lottery LI
to lottery L4 led to a corresponding rise in av-
erage certainty equivalents in both treatments.
The coefficients of L2, L3, and L4, measuring
the rise in certainty equivalents for those lot-
teries over those for LI, are all significantly
positive and increasing from L2 through L4.
III. Discussion
In contrast to recent survey and experimen-
tal evidence, the results reported here question
the prevalence of stereotypic gender-specific
ditions, we find that female subjects do not
than male subjects.
The data reported here show that the com-
parative risk propensity of male and female
subjects in financial choices strongly de-
VOL. 89 NO. 2 GENDER AND ECONOMIC TRANSACTIONS 385

pends on the decision frame. Gender-specific of Losses: An Expected Utility Analysis."

risk propensities arise in abstract gambles, Journal of Risk and Insurance, March
with men being more risk-prone toward 1980, 47(1), pp. 111-32.
gains but women more risk-prone toward Jianakoplos, Nancy Ammon and Bernasek,
losses. Moreover, when identical decisions Alexandra. "Are Women More Risk
are presented as investment and insurance Averse?" Economic Inquiry, October 1998,
choices, no gender differences in risk atti- 36(4), pp. 620-30.
tudes are found. Johnson, Johnnie E. V. and Powell, Philip L.
Our findings suggest that gender-specific "Decision Making, Risk and Gender: Are
risk behavior found in previous survey data Managers Different?" British Journal of
may be due to differences in male and female Management, June 1994, 5 (2), pp. 123 -
opportunity sets rather than stereotypic risk at- 38.
titudes. Our results also suggest that abstract Kahneman, Daniel and Tversky, Amos. "Pros-
gambling experiments may not be adequate for pect Theory: An Analysis of Decision under
the analysis of gender-specific risk attitudes Risk." Econometrica, March 1979, 47(2),
toward financial decisions. pp. 263-91.
In practice, risky financial decisions are Levin, Irwin P.; Snyder, Mary A. and Chapman,
inherently contextual. Our findings on con- Daniel P. "The Interaction of Experiential
textual financial decisions suggest that pre- and Situational Factors and Gender in a
conceptions concerning the risk attitudes of Simulated Risky Decision-Making Task."
female investors and managers may be more Journal of Psychology, March 1988,
prejudice than fact. 122(2), pp. 173-81.
Powell, Melanie and Ansic, David. "Gender
Differences in Risk Behavior in Financial
REFERENCES Decision-Making: An Experimental Anal-
ysis." Journal of Economic Psychology,
Barsky, Robert B.; Juster, F. Thomas; Kimball, November 1997, 18(6), pp. 605-28.
Miles S. and Shapiro, Matthew D. "Prefer- Sunden, Annika E. and Surette, Brian J. "Gender
ence Parameters and Behavioral Hetero- Differences in the Allocation of Assets in
geneity: An Experimental Approach in the Retirement Savings Plans." American Eco-
Health and Retirement Study." Quarterly nomic Review, May 1998 (Papers and Pro-
Journal of Economics, May 1997, 112 (2), ceedings), 88(2), pp. 207-11.
pp. 537-79. Wang, Penelope. "Brokers Still Treat Men Bet-
Hershey, John and Schoemaker, Paul J. H. "Risk ter Than Women." Money, June 1994,
Taking and Problem Context in the Domain 23(6), pp. 108-10.

This content downloaded from 121.52.158.248 on Wed, 12 Sep 2018 17:40:09 UTC
All use subject to https://about.jstor.org/terms