Professional Documents
Culture Documents
We find that single women exhibit relatively more risk aversion in financial
decision making than single men. Using U.S.sample data, we examine household
holdings of risky assets to determine whether there are gender differences in f i -
nancial risk taking. As wealth increases, the proportion of wealth held as risky
assets is estimated to increase by a smaller amount for single women than for
single men. Gender dgerences in financial risk taking are also influenced by age,
race, and number of children. Greater financial risk aversion may provide an
explanation for women’s lower levels of wealth compared with men k. (JEL 516,
D81,(311) *
620
Economic Inquiry
(ISSN 0095-2583)
Vol. XXXV1;October 1998, 6 2 0 6 3 0 Owestern Economic Association International
JIANAKOPLOS & BERNASEK: ARE WOMEN MORE RISK AVERSE 62 I
TABLE I
Variable Means by Household Type for Households with Wealth Exceeding $1,000
Variable Definition Single Women Single Men Married Couples
RATIO (Ratio of Risky Assets to WEALTH) .40 .46 30
Human Capital 209,382 427,6 14 725,059
HUMAN (Ratio of Human Capital to WEALTH) 31.7 59.3 63.2
WEALTH
Mean 67,298 121,652 202.5 19
Median 18,750 17,200 32,050
AGE (Head of Household Age Category):
Less than 25 .05 .I4 .02
26-30 .03 .I5 .09
3 1-35 .08 .09 .12
36-40 .06 .I4 .15
41 4 .09 .07 .13
46-50 .06 .04 .08
5 1-55 .08 .05 .10
56-60 .08 .06 .06
61-65 .08 .07 .07
Over 65 .39 .I7 .17
WORK (Labor Force Status of Household Head):
Self-employed .06 .I5 .14
Employed by others .51 .54 .61
Retired .25 .I7 .19
Fanner .oo .02 .02
Unemployed or Not in the Labor Force .I7 .I2 .05
EDUCATION (Head of Household Years of Schooling):
Grade School or less .03 .06 .03
Some High School .I8 .I3 .18
High School Degree .33 .I9 .29
Some College .22 .24 .20
College Graduate .25 .37 .30
RACE (% Black) .I2 .06 .05
KIDS (Number of children in the household) .28 .I7 I .04
H O M E 0 WNER .60 .46 .77
Number of Observations 384 230 1,980
equal to 47% of the original investment.’ This many dollars in stocks as bonds. Systematic
difference is based only on the gender differ- differences in risk aversion by gender may
ence in the allocation of wealth between risky provide an explanation for women’s system-
and risk-free assets and increases to over 80% atically lower levels of income and wealth
of the initial investment if the gender differ- compared with men’s.
ences in the allocation of risky assets between This topic is related to two areas of previ-
stocks and bonds are taken into account. ous research. One line of investigation, con-
Women in the sample report holding approx- ducted in studies by Friend and Blume [ 19751,
imately equal amounts of stocks and bonds on Morin and Suarez [1983], Bellante and Saba
average, while men report holding twice as [1986] and Siege1 and Hoban [1982; 19911
among others, has focused on determining the
I . This example assumes that stocks are divided
coefficient of relative risk aversion, i.e.,
equally between large-company and small-company stocks whether the proportion of risky assets held in-
and that bonds are divided equally between long-term cor- creases, decreases, or remains constant as
porate and long-term government bonds. Furthermore, we wealth increases. While these studies provide
recognize that past returns may not be good indicators of
future returns. the theoretical and empirical framework for
622 ECONOMIC INQUIRY
where RATIO is the ratio of risky assets to As mentioned previously, estimates of rel-
WEALTH, RACE is a dummy variable equal ative risk aversion are sensitive to the defini-
to one if the respondent is black and zero oth- tion of wealth, particularly the inclusion of
erwise; KIDS is the number of people 18 years residential real estate and human capital. Sev-
or younger in the household; H O M E 0 WNER eral previous studies, such as Friend and
is a dummy variable indicating whether or not Blume [ 19751 and Siege1 and Hoban [ 19911
the respondent is a homeowner; HUMAN is have focused on how alternative specifica-
the ratio of human capital to WEALTH, AGEi tions of wealth alter estimates of the coeffi-
is a set of dummy variables indicating into cient of relative risk aversion. However, since
which of ten age categories the household the focus of this study is on gender differences
head falls; WORKj is a set of dummy variables in risk taking, for purposes of comparison we
indicating five categories of labor force status have chosen to focus on only one of the def-
of the household head, and EDUCATIOX is initions of wealth used in previous research.
a set of dummy variables indicating which of Because the derivation of equation ( 1 ) as-
five levels of educational attainment the sumes that assets are infinitely divisible and
household head has reached. More detailed transactable, we have chosen a relatively nar-
definitions and summary statistics for the row definition of wealth which excludes in-
variables are presented in Table I. In previous vestments in residential housing and human
research, some investigators have used the capital. It is unclear the extent to which
logarithm of RATZO as the dependent variable. houses are owned for investment purposes in
Since that specification forces one to exclude addition to consumption purposes and human
households with no risky assets, we have cho- capital is certainly not infinitely divisible.
sen to use the level of the variable RATIO as However, holdings of residential real estate
the dependent variable. and human capital will certainly influence the
We estimate equation (2) using data from allocation of remaining wealth between risky
the SCF89. The SCF89 sample of 3,143 and risk-free assets. Consequently, we include
households was chosen to provide a compre- measures of these assets as explanatory vari-
hensive picture of the financial situation of all ables in the estimating equation.
U.S. households in 1989. Because wealth is WEALTH is measured as the sum of the net
highly skewed, the SCF89 oversampled high- value of risky and risk-free assets. The divi-
income households in order to obtain a more sion of wealth into risky and risk-free is by
accurate estimate of total household wealth. nature arbitrary, but we have tried to be con-
We make use of the SCF89 data employing sistent with previous studies. Risk-free assets
the multiple imputation procedure undertaken are defined to include dollar balances in
by researchers at the Board of Governors of checking, savings, and money market ac-
the Federal Reserve System. More informa- counts, certificates of deposit, U.S. savings
tion about these data can be obtained in bonds, IRA balances invested in certificates
Kennickell and Shack-Marquez [ 19921 and of deposit or bank accounts, and the cash
Weicher [ 19951. Because of the oversampling value of life insurance less policy loans out-
of high-income households, all statistics re- standing. Risky assets (which include mixed-
ported in this paper are sample weighted. risk assets) are defined as the sum of balances
Although these data are well-suited for our in IRAs not invested in bank deposits, stock
analysis, they do suffer from some limitations. holdings less margin loans outstanding,
For purposes of this study there are two im- bonds, trust assets, the net value of real estate
portant limitations of the data. First, no infor- owned other than residential housing, the net
mation regarding geographic location of value of businesses owned, and the net value
households is available in the public-use ver- of other miscellaneous assets (e.g. precious
sion of the data. Consequently, we are unable metal, futures contracts, art work) reported by
to construct estimates of household average the household.
tax rates since state of residence is unknown. Human capital is estimated by assuming
Second, for married households no informa- that 1989 wage, salary, and self-employment
tion is provided as to which spouse makes de- earnings of the household continue until re-
cisions regarding the allocation of assets. tirement. Earnings are calculated separately
624 ECONOMIC INQUIRY
for each spouse and summed for married cou- Since it is unclear whether financial port-
ples. The present discounted value of this folio-allocation decisions of married house-
earnings stream using a 2% discount rate is holds are made by men or women, we have
used as an estimate of human ~ a p i t a l Fol-
.~ also estimated the equation for married cou-
lowing Friend and Blume [1975] if the head ples. To the extent that males in the married
of household was under 65, retirement was households make the financial decisions,
assumed at age 65. If the head of household these estimates indicate whether there are
was working and aged 66 through 69, retire- gender differences in portfolio allocation be-
ment was assumed in four years; if aged 70 tween married men and single women. How-
through 74, three years; if aged 75 through ever, we recognize that these estimates are
79, two years; and if over 79, one year. Al- more likely to be clouded by our inability to
though this measure of human capital it at best determine the financial decision maker in
a very rough approximation, analysis by married households.
Thornton, Rodgers, and Brookshire [I9971
suggests that the assumption of a constant IV. EMPIRICAL RESULTS
growth rate of earnings, which is implicit in
our choice of a discount rate, rather than a Table I1 presents the estimated coefficients
traditional inverted u-shaped pattern, is sup- for equation (2) when it is estimated sepa-
ported by longitudinal data. rately for single women, single men, and mar-
Because the dependent variable RATIO can ried couples. Likelihood ratio tests indicate
only take on values between zero and one, we that the estimated equations are significantly
use a maximum likelihood tobit regression different between single women and both sin-
procedure, which allows for both an upper and gle men and married couples. Equation (2)
lower bound on the dependent variable, to es- was also estimated jointly for single women
timate equation (2). Consistent with previous and single men including a dummy variable
research, only households with WEALTH FEMALE, which e q u a l s o n e for s i n g l e
greater than $1,000 are included in the sam- women, interacted with each explanatory vari-
ples. The most direct test of gender differ- able in order to ascertain which coefficients
ences in portfolio allocation is between house- are statistically different between single
holds headed by never married females and women and single men. The coefficients
never married males. However, the sample found to be statistically different by gender in
sizes of these subgroups are too small to allow this manner are also indicated in Table 11. The
for valid inferences, particularly regarding the same exercise was conducted for single
influence of the socio-demographic variables women and married couples and the statisti-
included in equation (2). Consequently, we es- cally different coefficients also are indicated
timate equation (2) separately for single in Table 11.
women and single men, where single includes The estimated coefficient on In WEALTH
widowed and divorced as well as never mar- provides an estimate of the inverse of the co-
ried.4 To the extent that widows andlor di- efficient of relative risk aversion (Ck)up to a
vorced women continue the investment strat- positive multiplicative constant. A positive
egies begun by their late or ex-husbands, we (negative) coefficient indicates decreasing
have biased our results against finding any (increasing) relative risk aversion. For all
gender differences in financial risk taking. three household types, the estimated coeffi-
cient on In WEALTH is positive and signifi-
cantly different from zero, indicating decreas-
3. Friend and Blume [I9751 assume that wages in- ing relative risk aversion. This is consistent
crease at a 4% rate and use a 10% discount rate, which with previous findings of decreasing relative
together are equivalent to assuming constant earnings and risk aversion found by Friend and Blume
a 6% discount rate. Bellante and Saba [ 19861 assume con-
stant earnings and a 10% discount rate. We have chosen a [ 19751, Moran and Suarez [ 19831, Bellante
2% discount rate as one which more closely equals the and Saba [1986], and Riley and Chow [ 19921,
long-run growth rate of real GDP.
when similar definitions of wealth and risky
4. Of the 384 single women in the sample, 46% are assets were used.
widowed, 38% divorced, and 16% never married. In con-
trast, among the 230 single men in the sample, 18% are The estimated coefficient on In WEALTH
widowed, 41% are divorced, and 41% are never married. for single women is significantly less than for
JIANAKOPLOS & BERNASEK: ARE WOMEN MORE RISK AVERSE 625
single men. Thus, while both single men and characteristics of the base case. These predic-
women exhibit decreasing relative risk aver- tions are graphed in Figure 1. For single
sion, single women do so to a lesser degree. women the predicted proportion of risky as-
This suggests that single women are relatively sets approximates an inverted u-shape-iw
more risk averse than single men. This result creasing from the youngest cohorts, reaching
is robust when estimated over the sample of a peak with the 3 6 4 0 year old cohort, then
never-married women and men, although the declining in older cohorts (a negative pre-
set of explanatory variables used is more re- dicted value can be interpreted as the willing-
stricted as required by the smaller sample ness to buy insurance to avoid any risks). For
sizes. We find no significant difference in rel- most age cohorts, the predicted proportion of
ative risk aversion exhibited by married cou- risky assets held by single women tends to be
ples versus single females. More conclusive below that for both single men and married
results, however, require information regard- couples, indicating that regardless of age-co-
ing the gender of the financial decision maker hort, single women tend to be more risk averse
in married households. than comparable single men and married cou-
To illustrate the impact of this difference ples. Other studies, such as those by Moran
in relative risk aversion, Table I11 presents the and Suarez [ 19831, Bellante and Saba [ 19861
predicted share of risky assets in the portfolio and Riley and Chow [1992], have found evi-
of each household type based on the equations dence of increasing relative risk aversion with
estimated in Table 11. The base case is one set age, but these studies did not examine how
of household characteristics which, when ap- this relationship might differ by gender.
plied to each of the three estimated equations, It is tempting to draw conclusions about
generates approximately the average propor- how financial risk taking varies over the life
tion of risky assets held by that household cycle based on these predictions, but that
type. Thus, in the base case, where wealth would be incorrect. These predictions are
equals $20,000, single women are predicted based on a cross section of data that show
to hold 43% of their wealth in risky assets, different people at the same moment, while
compared to 51% by single men. If wealth the life cycle refers to changes as the same
increases to $100,000, holding other house- person ages. Inferences about life-cycle
hold characteristics constant, the share held in changes in risk taking require longitudinal
risky assets is predicted to increase for both data, or at least several cross sections. Life-
single women and men (evidence of decreas- cycle inferences are beyond the scope of this
ing relative risk aversion); however, the share paper, but are a topic for future research.
increases by 19 percentage points for women, The categorical variables indicating labor
compared to 28 percentage points for men force status are all significantly negative for
(evidence of greater risk aversion by women). single women, indicating that single women
As illustrated earlier, even small percentage in all categories take fewer financial risks
differences in the allocation of wealth be- than self-employed women, the omitted cate-
tween risky and risk-free assets can result in gory. However, only single females catego-
substantial differences in portfolio returns, es- rized as farmers have a significantly different
pecially when compounded over a number of portfolio allocation than single males based
years. on labor force status and holding other factors
The estimates indicate that age has a very constant. All of the labor-force-status coeffi-
different impact on portfolio allocation for cients for married couples are significantly
single women versus single men and married different from those of single women, but
households. All of the estimated coefficients there is no consistent pattern of differences in
associated with the age categories are signif- predicted portfolios. Self-employed and re-
icantly different for single women versus sin- tired single women hold relatively more risky
gle men. Every coefficient except that for the assets than married couples, while married
over-65 age group is also different between couples who are employed by others, farmers,
single women and married couples. Table 111 or out of the labor force (including the unem-
indicates the predicted proportion of risky as- ployed, but excluding the retired) hold signif-
sets for each age category for each of the three icantly more risky assets than single females,
household types, holding constant the other holding other factors constant.
62 6 ECONOMIC INQUIRY
TABLE I1
Tobit Regression Results-Dependent Variable: RATIO
(standard errors in parentheses)
Single Women Single Men Married Couples
In WEALTH .117*** .170***t .128***
(.009) (.010) (.003)
A G E 26-30 .044 -.137***t .004t
(.079) (.047) (.033)
AGE 3 1-35 .127** -. 342** * t -.090***t
(.063) (.059) (.032)
AGE 36-40 .381*** -.263***t -.014t
(.066) (.049) (.032)
AGE 4 1-45 .241*** -.252***t .018t
(.062) (.059) (.032)
AGE 46-50 .325*** -.295***t -.oss*t
(.066) (.078) (.034)
AGE 5 1-55 .130** -.350***t -.059*t
(.064) (.074) (.033)
AGE 56-60 .065 -.322***t -.078**t
(.065) (.065) (.035)
AGE 61-65 .250*** -.540***t .OlSt
(.066) (.077) (.036)
AGE over 65 -.232*** -.668***t -.122***
(.061) (.069) (.035)
Employed by others -.262*** -.I 13*** -.069***t
(.044) (.043) (.013)
Retired -. 18 I*** -.040 -.241***t
(.049) (.057) (.019)
Fanner -.431*** .120t .065**t
(. 165) (.lo@ (.030)
Not in the labor force -.395*** -.222*** -.032t
(.051) (.053) (.023)
Some high school -.3 1O*** -.300*** .ooot
(.066) (.067) (.029)
High school degree -.336*** -.274*** -.O 14t
(.064) (.067) (.029)
Some college -.254*** .325***t -.037t
(.067) (.070) (.029)
College graduate -.339*** -.244**' -.002t
(.067) (.068) (.029)
RACE .154*** -.024t -.065***t
(.035) (.066) (.019)
KIDS -.030* .034t .Ol3***t
(.016) (.027) (.OW
H O M E 0 WNER -.068*** .070**t -.037***t
(.024) (.03 1) (.011)
HUMAN -.002* * * -.ooot -.000***t
(.OOO) (.OOO) (.OOO)
constant -.244** -.668***t -.704***t
(.121) (.126) (.054)
JIANAKOPLOS & BERNASEK: ARE WOMEN MORE RISK AVERSE 627
TABLE I1 continued
Tobit Regression Results-Dependent Variable: RATIO
(standard errors in parentheses)
Slngie Women Single Men Married Coupler
sigma hat .415 .401 .398
log likelihood -1 177.399*** -679.521*+* -5891.676+++
Number of Observations 384 230 I980
pseudo R2 .248 .3 12 .228
Number left-censored 123 65 450
Number right-censored 2 6 17
?Significantly different from the single-female coefficients at the 10%significance level or lower.
....*
'Significantly different from zero at the 10% significance level.
Significantly different from zero at the 5% significance level.
Significantly different from zero at the 1% significance level.
While one might expect more years of sion maker, compared to only 10% of the
schooling to increase household risk taking, women in white married couples. This sug-
this result is not found when other factors are gests that black women (albeit the survey re-
held constant. As Table 111 indicates, single fers to married women) are more involved in
women and single men with less than a sixth investing and perhaps more inclined to risk
grade education are predicted to hold portfo- taking than white women.
lios with much greater percentages of risky Additionally, risk taking is influenced by
assets (close to go%), compared with those the number of people under 18 in the house-
having more education, holding other factors hold. In most cases this is the influence of
constant. For every level of education except children, but by measuring the variable in this
for the 13-1 6 years of schooling category, sin- way, we are also allowing for the presence of
gle women are predicted to hold fewer risky grandchildren, younger brothers or sisters,
assets than single men. However, none of nieces, nephews, and other young people who
these differences are statistically significant. may be dependents in household. As the num-
Race appears to play a very different role ber of young dependents in a household in-
in financial risk taking for single women ver- creases, the proportion of risky assets held
sus single men and married couples. Single significantly decreases for single women, is
black women are estimated to hold signifi- unaffected for single men, and significantly
cantly more risky assets than single white increases for married couples.
women, but the reverse is estimated for single Finally, we recognize that residential hous-
men and married couples. Given base case ing and human capital have been treated as
household characteristics, and changing only risky assets in other studies, but because of
the racial characteristic, single black women indivisibility in the case of human capital and
are predicted to hold the largest proportion of the consumption component of investments in
risky assets-58%, compared to 49% and 42% residential housing, we prefer not to incorpo-
for single black men and married black cou- rate these assets directly into our measure of
ples, respectively. This result is consistent risky assets. Nevertheless, these assets may be
with other anecdotal evidence, such as survey expected to influence the allocation of other
findings reported by Smith [ 19971, showing wealth between risky and risk-free assets. In
that 2 1% of the women in African-American the case of human capital, we find that the
households were the primary financial deci- greater the value of human capital relative to
628 ECONOMIC INQUIRY
TABLE 111
Predicted Proportion of Risky Assets
~~ ~
FIGURE 1
Age Profile of Risky Assets
0.7
0.6
0.5
0.4
0.3
0.2
0.1
-0.1 1 , I I I
20 30 40 50 60 70 80
Age Range
-Single Women 1 1 ~ 1Single
1 Men ..A., Married Couples
wealth, the smaller the proportion of other held, comparing single women with single
risky assets held, holding other factors con- men and married couples.
stant. While this is a statistically significant On balance, we find that single women are
effect and differs for single women versus sin- relatively more risk averse than single men.
gle men and married households, quantita- We confirm previous studies which find that
tively the impact is not large. The impact of relative risk aversion decreases as household
home ownership is also to reduce holdings of wealth increases when wealth is measured ex-
risky assets for single women and married cluding residential housing and human capi-
couples, but to increase risky assets for single tal. However, we find that relative risk aver-
men. sion does not decrease as much for single
women as for single men, indicating that sin-
V. SUMMARY AND CONCLUSIONS
gle women are relatively more risk averse. We
also find that over most age ranges single
Women are significantly more risk averse women hold smaller proportions of risky as-
in financial decision making than men based sets than either single men or married couples,
on survey responses. In this paper we seek to holding other factors constant. In addition, we
empirically verify the stated and popularly find that, unlike single men or married cou-
perceived notion that there are gender differ- ples, single women reduce the proportion of
ences in risk taking. We estimate the impact risky assets they hold as the number of chil-
of household wealth and other socioeconomic dren in their household increases, holding
variables on the proportion of risky assets other factors constant. We also find evidence
that single black women are willing to hold a
630 ECONOMIC INQUIRY