Economcs Letters 76 (2002) 151 158 www.elsever.com/ locate/ econbase Gender dfferences n revealed rsk takng: evdence from mutual fund nvestors a b c, * Peggy D. Dwyer, James H. Glkeson, John A. Lst a Unversty of Central Florda, Orlando, FL, USA b Unversty of Central Florda, Chapman Unversty, Orlando, FL, USA c Department of Agrcultural & Resource Economcs, Unversty of Maryland, 220 Symons Hall, College Park, MD 20742-5535, USA Receved 6 May 2001; accepted 1 October 2001 Abstract Usng data from a natonal survey of nearly 2000 mutual fund nvestors, we nvestgate whether nvestor gender s related to rsk takng as revealed n mutual fund nvestment decsons. Consonant wth the receved lterature, we fnd that women exhbt less rsk-takng than men n ther most recent, largest, and rskest mutual fund nvestment decsons. More mportantly, we fnd that the mpact of gender on rsk takng s sgnfcantly weakened when nvestor knowledge of fnancal markets and nvestments s controlled n the regresson equaton. Ths result suggests that the greater level of rsk averson among women that s frequently documented n the lterature can be substantally, but not completely, explaned by knowledge dspartes. 2002 Elsever Scence B.V. All rghts reserved. Keywords: Rsk takng; Gender dfferences JEL classfcaton: D81; G11; J16 1. Introducton In recent years, the fnancal press has ssued frequent warnngs that women are ll-prepared for retrement years due n part to ther selected nvestment programs. Academc lterature has largely confrmed ths anecdotal evdence by suggestng that women are less lkely than men to nvest n rsker, but hgher returnng, assets (see, e.g., McDonald, 1997; Kahn, 1996; Rchardson, 1996). One partcularly nterestng lne of nqury addresses the relatonshp between gender and revealed *Correspondng author. Tel.: 11-301-405-1288; fax: 11-301-314-9091. E-mal addresses: jlst@arec.umd.edu (J.A. Lst), http:// www.arec.umd.edu/ jlst/ (J.A. Lst). 0165-1765/ 02/ $ see front matter 2002 Elsever Scence B.V. All rghts reserved. PII: S0165-1765(02)00045-9
152 P.D. Dwyer et al. / Economcs Letters 76 (2002) 151 158 fnancal rsk preferences usng large-scale survey data. For example, Janakoplos and Bernasek (1998) examne total holdngs of rsky assets, and fnd that sngle whte women generally hold lower proportons of rsky assets than other groups. Although these sorts of studes provde sgnfcant nsghts nto the nvestment decson across gender types, one crtcal shortcomng n the lterature s that large-scale survey studes offer poor control over potental gender dfferences n knowledge sets n fact, we are aware of no prevous large-scale study of actual nvestment practces that has measured and analyzed the effects of context-specfc knowledge on the nvestment decson. In ths paper, we have a unque opportunty to advance the lterature by emprcally examnng rsk-takng n mutual fund nvestments across gender types, whle controllng for nvestor-specfc fnancal nvestment knowledge. Usng mutual fund nvestment data s a natural extenson of the extant lterature gven that mutual fund nvestment s at record levels, contnues to grow, and s wdely dscussed n the popular press, makng t a relatvely unambguous decson context for both men and women. In addton, our data provde a natural test of the Hudgens and Fatkn (1985) conjecture that gender dfferences occur only n stuatons where the probablty of success s low. 2. Data In 1995 the Offce of the Comptroller of the Currency and the Securtes Exchange Commsson jontly conducted a survey of 2000 randomly selected mutual fund nvestors. In addton to basc demographc nformaton, the survey asked respondents about the types of mutual funds they owned and the channels through whch these funds were purchased. Respondents were also asked a seres of questons n order to determne ther understandng of basc fnancal concepts. Because one purpose of ths study s to examne whether the rsk-takng behavor of mutual fund nvestors s correlated wth gender, we focused on three peces of nformaton pertanng to the type of fund owned. Specfcally, we examned the types of mutual funds that respondents had purchased for ther LARGEST sngle nvestment, ther most recent (LAST) nvestment, and ther RISKIEST nvestment. The RISKIEST measure s a composte varable created by selectng the rskest mutual fund type reported across all nvestment channels. In order to examne the level of rsk wthn mutual fund selectons, the rskness of the fund type was coded usng an ordnal rankng system. Money market and muncpal money market funds were coded 0, muncpal bond funds were coded 1, bond funds were coded 2, mxed/balanced funds were coded 3, and stock funds were coded 4. The 0 4 rankngs correspond to the rsk level (typcally measured as the varance of returns) assocated wth each category, where 4 s consdered the rskest opton. The upper panel of Table 1 contans means and standard devatons for each mutual fund type, as well as the proporton of respondents that nvested n each fund type. The rght-most column of Table 1 contans large sample t-statstcs testng the null hypothess of homogenous nvestment decsons across men and women. For the LARGEST nvestment category, the bond fund response was not well-represented for men or women, leadng us to omt ths response from our analyss and truncate the remanng responses, resultng n a 0 3 rsk scale (.e. money market, muncpal bond, mxed, and stock) for ths nvestment category. A strkng fndng s that across the three nvestment categores women appear to take less rsk than men. Ths general observaton s supported va t-tests, whch n each case reject the null at the P, 0.01 level. We should note that nonparametrc sgn tests support all of the results of the parametrc t-tests.
P.D. Dwyer et al. / Economcs Letters 76 (2002) 151 158 153 Table 1 Descrptve statstcs Gender Mean (S.D.) Probabltes of each rsk category t-test Money Muncpal Bond Mxed Stock of means market P(0) P(1) P(2) P(3) P(4) LARGEST Female 2.09 (1.21) 0.19 0.11 0.11 0.59 23.56 Male 2.33 (1.13) 0.15 0.07 0.08 0.70 LAST Female 2.55 (1.63) 0.22 0.07 0.09 0.16 0.45 24.05 Male 2.85 (1.56) 0.18 0.05 0.06 0.16 0.55 RISKIEST Female 3.19 (1.35) 0.11 0.04 0.08 0.11 0.67 24.86 Male 3.47 (1.14) 0.06 0.03 0.05 0.08 0.78 Age Female 2.49 (1.26) 0.18 Male 2.48 (1.21) Educaton Female 4.30 (1.38) 23.66 Male 4.53 (1.39) Income Female 3.23 (1.16) 22.98 Male 3.38 (1.05) Investment knowledge Female 6.20 (2.22) 214.30 Male 7.67 (2.35) Means are for the ndvduals that nvested nonzero amounts n that category. Large sample t-statstcs presented n rght-most column. LARGEST, type of fund n whch respondents had the largest nvestment; LAST, type of fund n whch respondents made the most recent nvestment; RISKIEST, rskest type of fund n whch respondents held an nvestment. Investment knowledge, summed response to a 12-tem scale. Even though the descrptve statstcs n the upper panel of Table 1 suggest that women take less rsk than men, t s napproprate to draw such a concluson from uncondtonal dfferences. Theory and prevous emprcal fndngs suggest that other factors ncludng age, educaton, and ncome nfluence rsk takng. We also gather data on these attrbutes, and present descrptve statstcs n the 1 lower panel of Table 1. Large sample t-tests of means ndcate that populatons of men and women n 1 Age was coded 1 for 18 34 years, 2 for 35 44 years, 3 for 45 54 years, 4 for 55 64, and 5 for 65 and older. Educaton was coded 1 for some hgh school or less, 2 for completed hgh school, 3 for trade school past hgh school, 4 for some college, 5 for completed college, and 6 for attended graduate school. Income was coded 1 for less than $15,000 per year, 2 for $15,000 $35,000, 3 for $35,000 $75,000, 4 for $75,000 $150,000 and 5 for $150,000 and over. Gender s measured as a dchotomous varable, coded 1 for males and 0 for females. Of our subjects, 41.6% are female and 54.6% are college graduates (26.5% had some graduate school). Almost two-thrds (66.3%) had purchased ther frst mutual fund pror to 1993. The average respondent owned slghtly more than three dfferent funds, and 39.6% owned four or more funds whle only 23.3% owned a sngle fund.
154 P.D. Dwyer et al. / Economcs Letters 76 (2002) 151 158 the sample dd not dffer n age. However, men reported ncome and educaton levels that exceeded levels reported by women (P, 0.01). Whle we have a rough measure of general educaton va the Educaton varable, ntuton suggests that specfc knowledge of nvestment practce s related to nvestment rsk preferences. We measure nvestment knowledge by usng the summed responses to a 12-tem scale that had potental scores rangng from 0 to 12. Sx of the scale tems requred responses that could be compared to a known answer. An example of ths type of tem s: Can a stock mutual fund lose money? Correct answers to these sx tems were coded 1; ncorrect answers were coded 0. The remanng sx tems n the scale 2 measured self-reported understandng of the meanng of selected fnancal and nvestment terms. An example of ths type of tem s: Do you know what a redempton s? Yes answers to these tems were coded as 1; no answers were coded as 0. The lower panel of Table 1 ndcates that the average nvestment knowledge score s 6.20 and 7.67 for women and men. A t-test strongly suggests that ths dfference s sgnfcant (t 5214.30), mplyng that men and women have dfferent knowledge sets concernng nvestment decsons. Ths fndng suggests that the uncondtonal fndngs must be vewed wth cauton, snce subjects may be revealng dfferences n ther specfc nvestment knowledge rather than dsplayng any underlyng dfferences n rsk preferences. 3. The emprcal model To supplement our uncondtonal fndngs n Table 1, we use a smple emprcal model that controls 3 for other mportant factors that may affect nvestment choce. Gven that the survey responses are coded 0, 1, 2, 3, and 4 for those that nvested, a lnear regresson model s napproprate. Estmaton of the model va ordnary least squares would treat the dfference between 0 and 1 dentcal as that between 3 and 4. In fact, the responses represent a rankng and therefore one-unt changes are not drectly comparable n ths manner. To amend ths shortcomng, we buld a model around a latent regresson of the form: Y * 5 X9 b 1, (1) where Y * s unobserved, X s a vector of person-specfc exogenous varables, b s the estmated response coeffcent vector, and s the well-behaved random error component. Although we do not drectly observe Y *, we do observe an approxmaton of Y *: Y5 0fY * # 0; 5 1f0, Y * # f 1; 5 2ff 1, Y * # f 2; 5 3ff 2, Y * # f 3; 5 4ff, Y * # f. (2) 3 4 The f are unknown parameters that are estmated jontly wth b; Y * s unknown snce the questonnare requests the survey respondents to select the answer that most closely represents ther 2 In some cases, the psychology lterature has shown that there are gender dfferences n self-reports of knowledge and ablty, such that men tend to overestmate relatve to women. For ths reason, we performed a senstvty analyss usng an nvestment knowledge measure that contaned only the sx tems for whch there s a known answer. The results of these alternatve analyses were not qualtatvely dfferent than those presented n Tables 2 and 3. 3 Sample szes are generally less than 2000 due to ncomplete observatons.
P.D. Dwyer et al. / Economcs Letters 76 (2002) 151 158 155 true random varable value. As such, we obtan threshold levels of rsk takng by measurng how exogenous varable vector X, whch ncludes gender, age, ncome, educaton, and nvestment 4 knowledge, affects ranked responses, Y *. 4. Emprcal results Table 2 contans ordered probt estmates for each of the three dependent varables (LARGEST, LAST, and RISKIEST nvestment), calculated wth and wthout a control for nvestor knowledge. An 2 mportant frst fndng s that each of our models performs reasonably well: examnaton of the x statstcs suggests that the models explan a sgnfcant amount of the varaton n the dependent varable. Parameter estmates n Table 2 provde evdence of the control factors that affect rsk takng. Estmated coeffcents of ncome are postve and sgnfcantly dfferent from zero n four of the models, and the coeffcents of educaton are postve and sgnfcant n every model. Ths result Table 2 Ordered probt estmaton results Varable Model LARGEST LARGEST LAST LAST RISKIEST RISKIEST Constant 0.20 20.03 20.09 20.37 0.31 20.03 (1.24) (20.20) (20.75) (22.96) (2.40) (20.20) Gender 0.23 0.14 0.18 0.07 0.27 0.14 (3.21) (1.93) (3.41) (1.34) (4.66) (2.33) Age 20.03 20.04 20.01 20.02 20.02 20.03 (21.17) (21.54) (20.50) (21.02) (20.89) (21.51) Educaton 0.14 0.11 0.09 0.04 0.14 0.07 (5.48) (3.82) (4.64) (2.02) (6.50) (3.50) Income 0.02 0.01 0.14 0.13 0.13 0.12 (0.54) (0.20) (5.76) (5.18) (4.79) (4.06) Investment 0.07 0.09 0.12 knowledge (4.32) (7.32) (8.13) 2 x (d.f.) 49.3 (4) 67.8 (5) 88.5 (4) 142.3 (5) 111.2 (4) 180.3 (5) n 1316 1316 1927 1927 1996 1996 Gender s a dchotomous varable that equals 1 for males, 0 for females. t-ratos are reported n parentheses beneath coeffcent estmates. Estmates of f are avalable upon request. LARGEST, type of mutual fund n whch respondents had the largest nvestment; LAST, type of mutual fund n whch respondents made the most recent nvestment; RISKIEST, rskest type of fund n whch respondents held an nvestment. Investment knowledge, summed responses to a 12-tem scale. 4 A few aspects of our estmaton procedure mert further consderaton. Frstly, snce the fs are free parameters, there s no sgnfcance to the unt dstance between the set of observed values of Y, thus avodng symmetrc treatment of one-unt changes n the dependent varable. Secondly, estmates of the margnal effects n the ordered probablty model are qute nvolved because there s no meanngful condtonal mean functon. We therefore compute the effects of changes n the covarates on the j probabltes: Prob[cell j]/ X 5 [ f(f 2 X9b ) 2 f(f 2 X9 j21 j b )]*b; where f(? ) s the standard normal densty, and other varables are defned above. By defnton, these effects must sum to zero snce the probabltes sum to one. Thrdly, our choce of regressors follows theory and prevous emprcal fndngs, whch suggest that age, educaton, nvestment knowledge, and ncome are assocated wth rsk takng.
156 P.D. Dwyer et al. / Economcs Letters 76 (2002) 151 158 mples that wealther, more educated nvestors tend to take on more rsk than ther less educated, less wealthy counterparts. When nvestor knowledge s ncluded, ts effect s postve and hghly sgnfcant for all three measures, ndcatng that t contrbutes to rsk takng above and beyond the contrbuton made by general educaton. Estmated coeffcents of the gender varable n Table 2 provde nterestng nformaton. We should frst note that, n the models that exclude the measure of nvestor knowledge, the coeffcent of the gender varable s postve and sgnfcantly dfferent from zero (P,0.05) n all cases. Ths suggests that men take on more rsk than women when selectng mutual funds. However, when nvestor knowledge s ncluded, gender s sgnfcant at P,0.05 for only the RISKIEST nvestment. In contrast, ts sgnfcance drops to P,0.054 n the LARGEST nvestment model and s below conventonal levels for the LAST nvestment model (P50.18). Interestngly, ths change n the level of sgnfcance s drven by the reducton of the coeffcent estmate rather than mprecson. Comparson of the gender coeffcents reveals that for each model type the estmate decreases by nearly 50% when nvestor knowledge s ncluded. Although these coeffcent estmates provde nsghts nto how the gender effect changes when an ndvdual-specfc knowledge regressor s ncluded, not much nformaton beyond ther statstcal sgnfcance can be used snce they are not margnal effects. To amend ths stuaton, we present margnal effects from the models that nclude the nvestment knowledge regressor. The estmates n Table 3, panel A, correspondng to respondents LARGEST mutual fund nvestments, can be read as follows: men are 5% more lkely than women to be n cell 4 (stock fund). Alternatvely, men are 3.4% less lkely to be n cell 0 (money market fund). Ths fndng s consstent across each of the nvestment categores and serves to enhance the results dscussed above. Gven that these effects are robust across nvestment category, t s nterestng to understand how they change when the nvestment knowledge regressor s excluded from the specfcaton. Margnal effects estmates from these models are drectly beneath the margnal effects estmates from the models that nclude the nvestment knowledge varable and support the general observatons of the coeffcent estmates n Table 2. For example, n the LARGEST nvestment category, men are 8.3% more lkely than women to be n cell 4 (stock fund) an ncrease of nearly 60% compared to the margnal effect estmates when nvestor knowledge s ncluded. Lkewse, the margnal effects ncrease by 140% and 95% when the nvestor knowledge varable s excluded from the other models. These results confrm our fndngs n Table 2 and suggest that the gender effect s greatly attenuated when one properly controls for nvestor knowledge. 5. Concludng remarks Usng data from nearly 2000 mutual fund nvestors, we fnd evdence that suggests women take less rsk than men n ther mutual fund nvestments. We fnd, however, that the observed dfference n rsk takng s sgnfcantly attenuated when we nclude a fnancal nvestment knowledge control varable n the regresson model, suggestng that the gender effect found n prevous studes that employ less specfc knowledge controls may be based upward. Our fndngs have several practcal mplcatons. Frstly, our results are contrary to Hudgens and Fatkn (1985), who conjecture that gender dfferences occur only n stuatons where the probablty of success s low. Accordngly, the prevalence of educatonal nvestment marketng efforts that target women s understandable. Secondly, our fndngs
P.D. Dwyer et al. / Economcs Letters 76 (2002) 151 158 157 Table 3 Margnal effects estmates Varable Money Muncpal Bond Mxed Stock market P(0) P(1) P(2) P(3) P(4) LARGEST Gender 20.034 20.010 20.007 0.052 w/ o knowledge 20.055 20.016 20.012 0.083 Age 0.011 0.003 0.023 20.016 Educaton 20.026 20.007 20.005 0.039 Income 20.002 20.005 20.001 0.003 Investment 20.017 20.005 20.004 0.026 knowledge LAST Gender 20.020 20.003 20.003 20.003 0.030 w/ o knowledge 20.050 20.008 20.007 20.007 0.072 Age 0.006 0.001 0.001 0.001 20.009 Educaton 20.011 20.002 20.002 20.002 0.017 Income 20.035 20.006 20.005 20.005 0.052 Investment 20.025 20.004 20.004 20.004 0.037 knowledge RISKIEST Gender 20.020 20.006 20.010 20.011 0.045 w/ o knowledge 20.037 20.012 20.019 20.020 0.088 Age 0.005 0.002 0.003 0.003 20.012 Educaton 20.010 20.003 20.006 20.006 0.024 Income 20.015 20.005 20.008 20.009 0.037 Investment 20.015 20.005 20.008 20.009 0.037 knowledge Gender s a dchotomous varable that equals 1 for males, 0 for females. Margnal effects are calculated as changes n the covarates on the j probabltes: Prob[cell j]/ X 5 [ f(f 2 X9b ) 2 f(f 2 X9 j21 j b )]*b. LARGEST, type of mutual fund n whch respondents had the largest nvestment; LAST, type of mutual fund n whch respondents made the most recent nvestment; RISKIEST, rskest type of fund n whch respondents held an nvestment. Investment knowledge, summed responses to a 12-tem scale. may be relevant to the current dscusson regardng the prvatzaton of the US socal securty system. Proponents of prvatzaton have suggested that women would beneft from the rght to manage ther own retrement nvestments (Anonymous, 1999). However, our fndngs rase the concern that prvatzaton could further magnfy exstng gaps between men s and women s retrement savngs. Fnally, our fndngs may help to explan the paucty of women n professons that requre a penchant for rsk-takng behavor (Chevaler and Ellson, 1999). Acknowledgements We thank Peter Ngro at the Offce of the Comptroller of the Currency for provdng the survey data examned n ths paper.
158 P.D. Dwyer et al. / Economcs Letters 76 (2002) 151 158 References Anonymous, 1999. Women and socal securty: are prvate accounts the answer? Journal of Accountancy 187 (6), 12 13. Chevaler, J., Ellson, G., 1999. Are some mutual fund managers better than others. Cross-sectonal patterns n behavor and performance. Journal of Fnance 54 (3), 875 899. Hudgens, G.A., Fatkn, L.T., 1985. Sex dfferences n rsk takng: repeated sessons on a computer smulated task. Journal of Psychology 119 (3), 197 206. Janakoplos, N.A., Bernasek, A., 1998. Are women more rsk averse? Economc Inqury 36, 620 630. Kahn, V.M., 1996. Learnng to love rsk. Workng Woman 21 (9), 24 27. McDonald, K.S., 1997. No guts, no glory. Workng Woman 22 (4), 42 46. Rchardson, P., 1996. Gong after the gender gap. Insttutonal Investor 30 (4), 141.