Emotional Intelligence and Investor Behavior

Although gains and losses are a normal part of the economic cycle, most investors

do not respond equally to gains and losses (Kahneman and Tversky 1973, 1979).

Investors feel positive emotions from a realized gain but relatively stronger negative

emotions from a realized loss of the same size. As a result, some investors sell their

winners prematurely while hanging on to their losers (Shefrin and Statman 1985;

Barber and Odean 1999). Some trade too much, others, too little (Barber and Odean

2000). In the past, behavioral finance research attributed these kinds of mistakes

primarily to cognitive heuristics and biases (Gilovich, Griffin, and Kahneman

2002). Recently, psychologists and economists have shown increased interest in the

role of emotions in economic behavior and decision making (e.g., Hopfensitz and

Wranik 2008; Loewenstein 2000; Thaler 2000). Indeed, ample evidence now exists

that feelings significantly influence decision making, especially when the decision

involves risk and uncertainty (Schwarz 1990; Forgas 1995; Isen 2000; Loewenstein,

Weber, Hsee, and Welch 2001). Researchers still have much to learn, however,

about the influence of individual differences in these processes and the role these

differences and processes play in real financial investment decisions and behavior.

In the research reported here, we explored the relationship between investment

decisions and three psychological variables: emotional intelligence (a

measure of a person’s ability to perceive, understand, use, and manage emotional

signals), personality, and impulsiveness (the inclination to act on impulse instead

of careful reflection). We found important relationships among aspects of these

three psychological constructs and various investment behaviors.

Psychological Concepts

Experts have identified a number of personality and other individual differences

factors that may systematically influence investment decisions (see, for example,

Salovey 2001); however, there is still very little empirical evidence to determine the

impact and importance of these variables (Hopfensitz and Wranik 2008). Based on

past research and experience, we thus chose to focus on three psychological variables

expected to play a major role to our study: emotional intelligence, personality, and


What Is Emotional Intelligence? For our purposes, we use the term

“emotional intelligence” (EI) in a more scientific and specialized sense than the

concept popularized by such best-selling books as Emotional Intelligence (Goleman

1995). In the popular conception, EI comprises a broad range of personality traits,

social skills, and qualities, such as “character.” In our research, EI is a precisely

defined and measured capacity similar to traditional aspects of intelligence. Traditional

intelligence is a person’s ability to use observed information or data (language,

patterns, and spatial relationships) to think productively. Emotional intelligence is

a person’s ability to recognize and interpret emotions and to use and integrate them

productively for optimal reasoning and problem solving (Salovey and Mayer 1990;

Mayer and Salovey 1997). In this way, EI is similar to traditional intelligence, but

EI uses moods or emotions as data or information.

Emotional intelligence should be distinguished from simply “emotional.” An

emotional person may feel and/or act more intensely than others; an emotionally

intelligent person is one who is able to recognize and use emotions productively.

Research in the past decade has shown that moods and emotions play important

roles in reasoning, decision making, and social relationships. Moreover, and contrary

to popular beliefs, moods and emotions play not only the role of “culprit” in

these processes (and hence need to be eliminated or minimized) but often play the

role of “adviser” by containing valuable signals and clues that facilitate optimal

personal choices and decisions. The trick is to know how to use moods and emotions

in an advantageous manner. Those who are high in EI are able to use and integrate

their moods and emotions effectively. Those who are low in EI may ignore,

misinterpret, or be overwhelmed by their moods and emotions and thus may not

reap the potential benefits of these cues. Given the pervasiveness of moods and

emotions in all spheres of life (including financial decision making), the EI form of

intelligence is gaining in acceptance and the definitions, research, and measures of

EI are becoming more sophisticated over time (for a thorough review, see Mayer,

Roberts, and Barsade 2008).

Our EI research is based on Mayer and Salovey’s ability-based model of EI

(1997) and on an ability-based emotional intelligence test developed by Mayer,

Salovey, and Caruso (2002)—namely, the MSCEIT (Mayer–Salovey–Caruso

Emotional Intelligence Test). The model by Mayer and Salovey (1997) comprises

four distinct competencies:

• perceiving emotions—recognizing emotional signals in people’s faces and via

other communication channels,

• using emotions—using emotions to enhance thinking and problem solving (this

ability may involve such actions as harnessing disruptive feelings to assist

reasoning, problem solving, and decision making),

• understanding emotions—analyzing emotions, predicting how emotional states

will change over time, and evaluating the influence of emotions on an outcome

(this ability also includes using language to describe feelings and emotions), and

• managing emotions—understanding and regulating responses to emotional

stimuli in the context of a particular goal or social situation.

Momentary moods, especially stemming from negative feelings, such as sadness

or anger, influence real economic decisions; investors with the ability to use emotions

intelligently make investment decisions when they are in a positive frame of mind

(Lerner, Small, and Loewenstein 2004). Investors with the capacity to understand

and manage their emotions intelligently should be less influenced than other investors

by the tone of external information sources in making investment decisions.

Some of the most compelling—although indirect—evidence of the effect of

emotions on decisions comes from research in neuropsychology. In particular,

Bechara, Damasio, and Damasio (2000) and Bechara (2004) suggested that people

who have suffered damage to the ventromedial prefrontal cortex of the brain tend

to have cognitive capacities (as measured by the intelligence quotient, or IQ) that

fall into the normal or even above-average range but have problems experiencing,

understanding, expressing, and effectively using emotions.5 In other words, these

individuals have normal IQs but low EI, which tends to influence their decisionmaking

skills negatively (Bar-On, Tranel, Denburg, and Bechara 2003). In the

studies, low-EI individuals consistently made poor decisions and, contrary to

normal participants, showed an inability to learn from their previous mistakes. Most

importantly, these behaviors were especially strong when exact calculations of a

future outcome were not possible and choices had to be based on approximations,

which is usually the case with financial decision making.

Using the MSCEIT, one can measure EI within each of these four competency

categories and as a composite measure of a person’s ability in all areas. The research

we report here focused on investors’ abilities within each of the separate areas and

on the variety of influences that those abilities may have on actual investment

behavior. Although we were interested in all four areas of EI, we predicted that

skills in using and managing emotions would play a particularly large role in

“effective” investment decision making.

Personality Characteristics That Might Be Important. In addition

to emotional intelligence, we investigated how personality influences investment

decision making. Although many theories describe personality, one of the most

influential is the “Big Five” model. Evidence supporting the power of this theory to

characterize personality differences began with the research of Allport and Allport

(1921) and has been growing over the past 90 years. The work has been expanded

by, among others, Norman (1963), Eysenck (1970), Goldberg (1981), and McCrae

and Costa (1987, 1997). The Big Five are broad categories of personality traits

thought to be the most parsimonious set for describing interindividual variation in

behavioral propensities. Although a significant body of literature supports this fivefactor

model of personality, researchers do not always agree on the exact labels for

each dimension. The following five categories, however, are typical:

• extraversion—the tendency to be talkative, energetic, and assertive;

• agreeableness—the tendency to be kind, warm, and sympathetic;

• conscientiousness—the tendency to be efficient, organized, “planful,” and


• neuroticism/negative affectivity—the tendency to be moody, tense, and

anxious; and

• intellect/openness to experience—the dimension of having wide interests and

being imaginative, complex, and insightful.

We chose to measure personality by using the Big Five Inventory (BFI)

developed by John and Srivastava (1999) because it is the most reliable of the shorter

personality tests.6

Although personality and investment decisions probably have no direct or simple

relationship, just as corporate earnings and stock prices have no perfect relationship,

the data may contain trends or patterns. For example, past research has found that

introversion, lack of neuroticism, and lack of agreeableness determine higher levels

of household savings in the real population (Nyhus and Webley 2001) and that

conscientiousness and lack of neuroticism predict preretirement planning (Hershey

and Mowen 2000). Other research has shown that extraversion and lack of conscientiousness

are related to impulse buying (Verplanken and Herabadi 2001).

Impulsiveness. Impulsiveness is the immediate response to thoughts or

deeds without any consideration of the appropriateness or consequences. Studies

have linked impulsiveness to higher risks of smoking, drinking, and drug abuse and

to aggression, compulsive gambling, severe personality disorders, and attention

deficit problems. For our purpose, we were interested in the tendency of individuals

who are impulsive to make decisions faster than nonimpulsive individuals and often

to take higher risks (Zuckerman and Kuhlman 2000).

To understand impulsiveness in the financial domain, we find that differentiating

between “stimulating” and “instrumental” risk taking (Zaleskiewicz 2001) is

useful. On the one hand, the stimulating form of risk taking is motivated by hedonic

pleasure and high arousal. It tends to be rapid, effortless, and perhaps even

automatic. This form is important in such domains as impulse buying, gambling,

and extreme sports and is typically linked to the impulse trait known as “sensation

seeking.” The person who engages in instrumental risk taking, on the other hand,

is striving for a long-term future profit or benefit. This form of risk taking is

achievement and goal oriented and is related to the more complex functions in

information processing. For this research, we were interested primarily in instrumental

risk taking.

We measured impulsiveness by using the UPPS Impulsive Behavior Scale

(Whiteside and Lynam 2001). This instrument measures four distinct traits related

to impulsiveness: (1) Urgency, (2) (lack of) Premeditation, (3) (lack of) Perseverance,

and (4) Sensation seeking. In this study, we used only the “lack of premeditation”

and “urgency” subscales because the third trait is similar to the conscientiousness trait

already measured by the BFI and the fourth trait is related to the stimulating form

of risk taking. The two traits we used are defined as follows:

• Urgency—difficulty in controlling or coping with urges to act in response to

unpleasant emotions. This trait is the component of impulsiveness most

strongly associated with problem gambling (Whiteside, Lynam, Miller, and

Reynolds 2005).

• Lack of premeditation—the tendency not to delay action until careful thinking

and planning can occur. Those who exhibit impulsiveness act on the spur of

the moment without regard to the consequences. Lack of premeditation, as

measured by the UPPS Scale, has been linked to disadvantageous decisions in

the Iowa Gambling Task (Zermatten, Van der Linden, d’Acremont, Jermann,

and Bechara 2005).7

Impulsiveness can have both positive and negative effects for investment

decisions. Impulsive investors may engage in more frequent trading than less

impulsive investors. Impulsive investors may not fully analyze the situations they

are in and, as a result, may make decisions too quickly. Being not impulsive can also

create problems for an investor, however, because hesitation or inaction can be a

liability over the long term.

Terminology. In the remainder of the book, when we discuss the psychological

test results, we use the following notational conventions. The four measures

of emotional intelligence are denoted EI-Perceiving, EI-Using, EI-Understanding,

and EI-Managing; the overall score is designated EI-Total. The five attributes

of personality as measured by the BFI are denoted BF-Agreeableness, BFConscientiousness,

BF-Extraversion, BF-Neuroticism, and BF-Openness. And

the two measures of impulsiveness from the UPPS IMP (Impulsive Behavior)

Scale are denoted IMP-Urgency and IMP-Premeditation.

The Survey Sample

We summarize the results of a recent survey of 2,595 investors at Vanguard. From

these investors, we collected demographic information, and we administered to them

the three psychological tests measuring (1) emotional intelligence, (2) personality,

and (3) impulsiveness. All sample members voluntarily responded to an e-mail

invitation from Vanguard to participate in this research by taking an online survey.

Invitations were sent to a selected sample of Vanguard clients who met a number of

conditions: All were born between 1946 and 1964 (i.e., were Baby Boomers); all

invitees had traditional IRA (individual retirement account), Roth IRA, or 401(k)

plan assets of at least $5,000, with at least $1,000 in two different mutual funds on

31 December 2005. All participants, obviously, had to have valid e-mail addresses.

Our final sampling universe was then randomly selected from the set of Vanguard

clients meeting all these restrictions who were still clients on 31 December 2006.

In addition, because one of the goals of the study was to examine transactional

activity in investors’ accounts and how it relates to emotional intelligence, we

oversampled investors with at least one transaction moving money from one fund

to another (we call this type of transaction an “exchange transaction”) in 2005. We

reweighted the overall sampling universe so that 75 percent of the invitations would

go to investors with at least one such transaction in 2005 and 25 percent would go

to those who had made no exchange transactions.8 Finally, we elected to sample

401(k) plan participants and retail IRA account holders who met the criteria already

mentioned on an equal-weighted basis; that is, half of the invitations went to IRA

investors and the other half, to 401(k) investors. For most of the analyses that follow,

we focus on the behavior of respondents in these two groups of investors separately.

We refer to the IRA account owners as “IRA investors” and the 401(k) participants

as “401(k) investors.”

Invitations were sent from Vanguard by e-mail in rolling weekly waves from 30

January through 5 March 2007.9 The invitation included an appeal to shareholders

to help further research in the field. As an incentive to participate, a copy of the

8In 2005, 28 percent of the Vanguard retail population made one transaction or more.

9A copy of the invitation is available in the online supplemental materials at www.cfapubs.org.

Emotional Intelligence and Investor Behavior

©2009 The Research Foundation of CFA Institute 7

summary research findings was promised to shareholders who completed the survey.

Each invitation provided the client with a link to a secure website where clients could

complete the three psychological tests and supply demographic information, such as

household income, age, and gender. The information we used was gathered from

individuals in five sections: an initial set of demographic questions, a section of

questions on impulsiveness, a section for the personality inventory, a section on the

EI-Using and EI-Managing aspects of emotional intelligence, and an optional

section on EI-Perceiving and EI-Understanding. Overall, filling out the entire test

took participants 30–40 minutes. Perhaps largely as a result of the length of the

survey, many individuals did not fill out the optional section; also, some attrition

appears to have occurred at each section break in the survey questionnaire.

The sampling strategy was to roll out new waves of invitations until we had

collected roughly 1,250 responses from IRA investors and 1,250 responses from

Response to the Survey. The availability of some demographic and

account information for the entire universe of invited participants enabled us to

analyze the relationship between various characteristics and the likelihood of

responding to the survey. Results of a basic probit regression of response (1 = response,

10The e-mail invitations were rolled out at a rate of approximately 2,500 a week over this time period,

in order of increasingly high ZIP Codes. We did not use our full sample of all qualifying Vanguard

clients before obtaining a full quota of survey responses, so our respondents are generally individuals

living in the northeastern United States.

aExcluded were duplicates (more than one response per client), respondents

outside the desired age range, and clients with unavailable account


0 = no response) on the characteristics available from accounting databases for all

sampled individuals are presented in Table 2.11

These results show that the impact of the three demographic variables on the

likelihood of responding was modest, with varying degrees of statistical significance,

for both the IRA and 401(k) samples. In general, age is negatively correlated with

responding, in the sense that the younger the client, the less likely a response. This

effect is slightly stronger in the IRA sample than in the 401(k) sample. The larger

the retirement account balance, the more likely a response, although the effect is

small, implying (roughly) that a 1 percent change in balance corresponds to a 0.01

percentage point difference in the response rate. The largest selection effect in both

samples is in the transaction variable: For investors who made at least one transaction

during 2005—arguably, a subset of investors who are more engaged in the

11The probit model is a commonly used statistical technique used to model binary-outcomes data

(yes/no or one/zero). The basic idea is to estimate the relative effect of each of a set of observed

characteristics on an index value that, in turn, given the assumption that the error in the index’s ability

to correctly predict outcomes is normally distributed, predicts the likelihood of a yes or a no response.

Table 2. Relationship of Demographic and Account Data to Response


IRA Investors 401(k) Investors



Coefficient p-Valuea




Coefficient p-Valuea



Intercept –1.9803 <0.0001 na –1.4350 <0.0001 na

ln(AcctBal)c 0.0721 <0.0001 1.58% 0.0083 0.5604 0.15%

Had transaction 0.2148 <0.0001 4.71 0.1971 <0.0001 3.66


42–44 years –0.2558 <0.0001 –5.60% –0.1361 0.0083 –2.53%

45–49 years –0.1088 0.0116 –2.38 –0.0641 0.1168 –1.19

50–54 years –1.1005 0.0152 –2.20 –0.0327 0.4283 –0.61

55–59 yearsd — — — — — —

60+ years 0.0395 0.6573 0.87 0.0969 0.3488 1.80

aThe p-value is the estimated probability that the coefficient in question is equal to zero.

bMarginal probability shows the estimated impact of a 1 unit change in the independent variable on the

probability that an individual would respond to the survey.

cAccount balances comprised rollover or Roth IRAs for IRA investors.

dThe age variables are dummy variables indicating membership in each age category. To identify the relative

effect of age on a response, the impact of age must be estimated as a deviation from the case in which an

investor is a member of an (arbitrarily chosen) reference group; we chose the 55–59-year-old age group as

the reference group. For example, a member of the 42–44 group was 5.6 percent less likely to respond than

a member of the 55–59 group.

management of their investment portfolios—the marginal probability of a response

was 3.66 percent or 4.71 percent higher than for, respectively, 401(k) investors or

retail investors not making transactions.12

Because the 401(k) sample was drawn from the database of plans for which

Vanguard not only managed at least some assets but also did the record keeping,

we had a much richer set of demographic data for members of the 401(k) client

sample than for members of the IRA population. Therefore, we were able to carry

out a probit analysis of response rates for the 401(k) sample in relation to an

expanded set of independent variables. The results are shown in Table 3.

• Investors who made at least one transaction in 2005 were 3.58 percent more

likely than others to have responded to our survey.

• Age and retirement portfolio balance had modest effects on the likelihood of

response, which is consistent with the results in Table 2.

• The effects of household income were mixed, with the highest marginal

probability of response being in the $75,000–$124,999 income range.

• The wealthiest clients in ZIP Codes where average household wealth exceeded

$1 million were most likely to have responded.

• Gender had no statistical significance.

The “engagement” of the investor as identified by “Status” also played an

important role in determining whether a client responded to the survey invitation.

For the “term-deferred” investors, the marginal probability of a response was nearly

7 percent lower. (Term-deferred investors would be plan participants who are no

longer employed by the plan sponsor but who have chosen, either by a conscious

decision or by inaction, to leave their 401(k) balances in the former employer’s plan.)

The issue of sample selection is important in the research design of any study.

The results in Tables 2 and 3 show systematic relationships between individual

demographic or financial variables and the likelihood of responding to our survey.

These effects suggest that our sample is not representative of the universe of

Vanguard clients invited to respond. Nevertheless, in general, the response effects

documented here are modest and intuitive. Therefore, we conclude that we have

successfully collected information from a large and diverse set of respondents

without skewing heavily toward or away from a particular subgroup (or subgroups)

other than intentionally oversampling individuals who made one trade or more in

their retirement account(s).

In addition to any selection effects that may have arisen as a result of voluntary

response to the survey, several other levels of selection effect may have influenced

the characteristics of the sample of investors. Our sampling strategy explicitly

Investors Only



Coefficient p-Value



Intercept –1.6003 <0.0001 na

ln[401(k) balance] 0.0227 0.1573 0.42%

Had transaction 0.1928 <0.0001 3.58


42–44 years –0.1334 0.0159 –2.48%

45–49 years –0.0477 0.2725 –0.89

50–54 years –0.0220 0.6173 –0.41

55–59 years — — —

60+ years 0.1006 0.3587 1.87

Household income

<$20,000 0.0851 0.3733 1.58%

$20,000–$49,999 –0.0156 0.8395 –0.29

$50,000–$74,999 0.1399 0.0016 2.60

$75,000–$124,999 0.0381 0.3590 0.71

$125,000+ — — —

Wealth rangesa

<$100,000 — — —

$100,000–$249,999 –0.1386 0.0222 –2.57%

$250,000–$499,999 0.1084 0.3037 2.01

$500,000–$999,999 –0.1217 0.4123 –2.26

$1 million+ 0.4833 0.0436 8.97


Maleb — — —

Female 0.0156 0.6529 0.29%


Activec — — —

Retired 0.3467 0.3037 6.44%

Term deferredd –0.3689 <0.0001 –6.85

Notes: See notes to Table 2.

na = not applicable.

aWealth ranges were estimated by IXI Corporation, an independent data

vendor that calculates wealth ranges based on the average wealth for households

in each ZIP Code.

bAll effects are relative to the male gender group.

cAll effects are relative to the active status.

dParticipants no longer employed by the plan sponsor but with 401(k)

balances in the former employer’s plan.

excluded investors with less than $5,000 in assets at Vanguard and those holding

only one mutual fund. In addition, we oversampled those with at least one

transaction in 2005. Clearly, selection effects may also be relevant when investors

choose Vanguard over other financial services providers; those who consciously

choose Vanguard tend to be cost conscious and interested in index fund investing.

This matter is not an issue for 401(k) participants, as it is for IRA investors, because

in the case of 401(k) funds, an investment committee—not the individual participant—

chose Vanguard. Nevertheless, we caution readers that we were unable to

control for a wide variety of selection effects that may differentiate this sample of

survey respondents from the overall broad population of investors.

Respondent Demographics. Definitions of demographic variables and

a description of how they were measured are in Exhibit B1 of Appendix B. Table 4

presents the averages of sample demographic data. Although the 401(k) and IRA

samples are similar in many respects, they have some interesting differences. Just

under 75 percent of retail IRA respondents were still working full-time, whereas

94 percent of 401(k) respondents were employed full-time. The groups were both

highly educated and had correspondingly high incomes and wealth. More retail

respondents, however, reported having a master’s degree or higher, household

income of higher than $100,000, and household financial assets of $500,000 or

more.13 As Table 5 shows, these differences translated into higher average account

balances for the IRA respondents.

According to Table 5, survey respondents had an average of 76 percent of their

IRA and 78 percent of their 401(k) allocated to stocks; the remainder was in fixedincome

assets. Differences in the use of index funds may have resulted from

individual choice/sample selection in the IRA group (retail investors may associate

Vanguard with index fund investing) and variations in 401(k) plan designs. The

menu of funds offered in 401(k) plans varies from one plan to another, and some

plans may not include index funds in the lineup offered to participants.

Note also that the distribution of equity exposure in our 401(k) survey sample

differed significantly from that in Vanguard’s overall defined-contribution plans.

Only two of our sample respondents had no equity exposure in their 401(k)s; 85

percent of respondents had more than 60 percent of their account assets in equities.

In Vanguard’s overall 401(k) client database, 13 percent of participants had no equity

and 61 percent had more than 60 percent in equity (Vanguard 2006).14 This clear

difference between our sample and the broad population of 401(k) investors is

probably a result, in part, of the voluntary nature of our survey and our oversampling of participants with transaction activity. In particular, our sample undoubtedly

excluded many individuals who tend to “ignore” their 401(k) plans. Such individuals

are likely to account for a large portion of the individuals in the participant

population that have little or no exposure to equity.

The last two rows of Table 5 present data on transaction activity by participants.

Because fund investors may move assets from one fund to another, from one

fund to several, or from several funds to one, resulting in multiple transactions

arising from a single investment decision on any one day, we aggregated all

transactions on a given trading day into a “transaction day” variable.15 Even though

our sample was constructed to oversample individuals with at least one transaction

in 2005, the degree of transaction activity we observed in our sample is modest.

Consistent with earlier research, retail respondents traded more often, on average,

than their 401(k) counterparts.16

Relationship of Psychological Tests Scores to Demographic

and Financial Variables. Although the subcomponents of each psychological

test measure a relatively different aspect of emotional intelligence, personality, or

impulsiveness, these subcomponents are not completely independent.17 The reason

is that psychological functioning, unlike chemical functioning, is not made up of

basic independent parts. Indeed, early personality research sought to describe and

compile such elements of personality in order to create a psychological “periodic table

of the elements.” Most statistical tools still treat personality and other individual

differences in this fashion, but decades of research have shown that complete

independence of psychological characteristics is neither possible nor desirable. The

value of modern research into individual differences (personality, emotional intelligence,

skills, intelligence, gender, etc.) is that researchers no longer try to demonstrate

independence of variables but, rather, try to determine how individual variation

along any one dimension helps predict behavior (Pervin and John 2001).

From Appendix A, the correlation coefficients between most of the demographic

and financial-activity variables tend to confirm the internal consistency of

the survey responses.18 For example, having a postgraduate degree correlates with higher incomes and higher portfolio balances. This outcome suggests that we

obtained reasonable responses. The data also reveal some less obvious relationships.

For example, being male is positively correlated with regular reading of financial

literature, with not seeking and acting on financial advice, with larger 401(k)

balances, with higher equity allocations, and with more trading. And investors with

larger portfolio balances tend to make changes to their accounts more often than

other investors in the sample and to seek and act on financial advice.

We emphasize that our research goal was not (and because of our survey’s

structure and small scale, could not be) to assess the EI of investors at large.19 Our

goal was to collect a significant amount of detailed data that could be used to

determine whether, within a group of Vanguard investors, variation in the components

of emotional intelligence and other variables representing individual

differences had a significant relationship with observed investor behavior.

Table 6 presents the mean and median scores and subscores for the two

Vanguard samples. As shown, the EI-Understanding score and, to a smaller extent,

EI-Using score are significantly higher among IRA clients than among the 401(k)

sample. In addition, the IRA sample appears to have slightly higher scores on the

IMP-Premeditation and IMP-Urgency scales. Given our coding convention in

which lower scores suggest greater impulsiveness, these two data items suggest that

401(k) sample members were characterized by a greater degree of impulsiveness.

Finally, Table 6 shows slight differences in BF-Agreeableness (IRA investors being

slightly less agreeable) and BF-Extraversion (IRA investors being less extraverted).

If one accepts that investors who have established an individual IRA account

are likely to have a higher “propensity to save” than an average 401(k) investor, these

psychological characteristics may play an important role in explaining who saves and

who does not. For example, the higher scores for understanding and using emotions

among IRA investors compared to 401(k) investors could have at least two explanations.

First, although managing emotions is considered the most important skill

for effective decision making (Salovey 2001), there is evidence that understanding

emotions may be the driving force behind effectively using and managing emotions

(Wranik, Barrett, and Salovey 2007). Therefore, the differences in scores could

reflect the fact that those who are more skilled in understanding and using emotions

are more likely to recognize the utility of saving for retirement. Second, understanding

emotions is related to higher verbal skills and to attaining a higher education

level (Lewis 2000). The differences may thus also reflect higher levels of education

and wealth and, therefore, the financial opportunity to open IRA accounts. Finally, the personality results for agreeableness are similar to those by Nyhus and Webley

(2001), who found that individuals scoring high on agreeableness had less in savings

and were more likely to borrow money than other individuals.

A comparison of mean test scores with demographic variables shows only a

few significant systematic relationships for gender, educational attainment, and

total assets:20

• Women had higher EI scores in both the IRA and 401(k) samples. This gender

effect has been found in most research in the domain of emotional intelligence.

It probably reflects socialization experiences and cultural values (Bernet 1996;

Mayer, Salovey, and Caruso 2008).21 Therefore, to allow for an accurate

estimation of the effect of EI for the dependent variable, the norms for male

and female scores are frequently set by gender.

Table 6. Psychological Test Scores of Sample

IRA Investors 401(k) Investors

Test Mean Median Mean Median


EI-Perceiving 91.9 90.4 92.9 92.2

EI-Using 98.0 98.9 97.4 98.1

EI-Understanding 96.8 97.4 94.9 93.8

EI-Managing 96.0 96.7 96.6 97.5

EI-Total 94.4 94.1 64.4 94.5


IMP-Premeditation 3.20 3.18 3.13 3.09

IMP-Urgency 3.11 3.17 3.03 3.00


BF-Agreeableness 3.86 3.88 3.99 4.00

BF-Conscientiousness 4.17 4.22 4.16 4.22

BF-Extraversion 3.11 3.00 3.21 3.25

BF-Neuroticism 2.54 2.50 2.49 2.50

BF-Openness 3.65 3.70 3.63 3.60

Note: Because not every respondent answered every question, each group

contained missing observations.

• Women had higher scores for all personality traits except Openness in the BFI.

Gender differences in personality have not been systematically found by

researchers (Hyde 2005). The differences in our sample may reflect that we

had fewer women than men in our sample and that those women over 50 who

were active in investing are probably a unique group. Readers will want to note

the gender differences but should not place too much emphasis on them.

• Women had lower scores on the impulsiveness tests, which indicates that they

have the characteristics of less premeditation and more urgency than their male

counterparts. Gender differences in impulsiveness are not usually found

(Lynam and Miller 2004), although teenage girls may show slightly higher

urgency scores in some cultures (d’Acremont and Van der Linden 2005). Again,

these gender differences could reflect the unique nature of the women in our

sample. For example, impulsive women from the Baby Boom generation may

have been more likely than less impulsive women to forsake traditional values,

enter the workforce, and thus have accumulated investment accounts.

• The higher the educational attainment level, the higher the average EI-Total

score, BF-Openness score, and IMP-Urgency score. The positive relationships

between education and EI (Mayer et al. 2002) and between education and

openness to experience (Flynn, Smith, and Freese 2006) are normal; the

relationship with urgency is unclear.

• We found no systematic relationships between total assets and the EI or

personality characteristics.

Methodology for Examining Financial Behavior

Our analysis focused on five distinct aspects of investment behavior (or results) of

the group of individuals surveyed in the period 2004–2007:

1. asset allocation and overall exposure to stock market risk in retirement accounts,

2. frequency of trading or transaction activity in retirement accounts,

3. use of passive, index-based mutual funds as opposed to actively managed funds

as part of investment portfolios,

4. adoption of international equity investing, and

5. internal rate of return on investments in retirement accounts.

We first describe each dependent variable that captured the behavior of

interest, and we present summary statistics and a univariate analysis of the

relationship between these variables and the psychological characteristics that

we measured. For each dependent variable, we used a multivariate regression

framework to examine—while controlling for the influences of a large set of

demographic and other individual characteristics—the extent of the relationship

between the psychological characteristic and the outcome measure.

Our analysis of each of these areas of investment behavior is exploratory and

mainly descriptive in nature. Although we approached the data with hypotheses

and theories in mind that were connected to the psychological literature, we did not

have well-formed structural models of how the various psychological characteristics

might interact to affect observed behavior. For example, a piece of folk wisdom

among investment professionals is that trading on impulse leads to poor investment

results. What is not clear is whether those with a highly impulsive nature would

necessarily be poorer investors than others in the long term. If those who show

strong impulsiveness tend to possess any degree of “skill,” in the sense of identifying

unusually attractive or unattractive investment opportunities, impulsiveness may be

an element of their success.

Our empirical results allow several possible interpretations; a challenge for

future work will be identifying additional tests or analyses that would help narrow

the set of plausible interpretations and implications. In addition, the interaction of

various dimensions of personality and emotional profiles make it hard to cleanly

assess the independent impact of various characteristics for the entire population of

respondents. In some cases, we looked within specific subgroups of the population

to find a significant effect of a particular characteristic. For example, we examined

only respondents with high levels of neuroticism to find the significance of aspects

of emotional intelligence within this specific group.

The hypothesis we were most interested in testing is whether individuals who

demonstrate a high degree of EI demonstrate patterns in investment behavior, or

in investment results, that normatively appear to be better than other investors.

Asset Allocation. To model the asset allocation decision, we first formed

three groups of investors with different percentages of their retirement accounts

invested in stocks (0–49 percent, 50–90 percent, and 91–100 percent). These groups

were intended to represent individuals likely to be, respectively, (1) underexposed

to equity, (2) holding a portfolio that would be consistent with common practice

for retirement investing (the medium-equity group), and (3) those likely to be

overexposed to equity.

We then used an unordered multinomial probit model to isolate the psychological

and demographic characteristics that are related to membership in these three groups.22 Throughout this analysis, unless otherwise indicated in the discussion of

results in subsequent sections, we estimated standard errors by using procedures that

are robust to forms of heteroscedasticity within the population of respondents. We

chose to use an unordered multinomial procedure because our assumption was that

both low-equity investing and very high equity investing are less desirable investment

behaviors than medium-equity investing, although low- and high-equity holdings

are not necessarily better or worse relative to each other. Keep in mind that

psychological characteristics may have much to do with risk aversion but, here, we

are attempting to distinguish between three separate groups—one of which, we

maintain, is a group that has made the normatively best investment decision.

The assets on record at Vanguard may represent only a fraction of the overall

assets of the individuals we studied. Therefore, and because we are concerned that

an investor may devote less attention to an insignificant portion of his or her total

portfolio than to a more significant portion, we also performed a separate analysis

of only the subset of investors with a large percentage (more than 35 percent of their

total estimated financial wealth) in the retirement account we were examining.

A summary of the results of the regression analysis is presented in Exhibit 1.23

Examining the results obtained without controlling for various demographic characteristics

and focusing only on results with statistical significance at the 10 percent

level or better, we find that only the EI-Using score influenced membership in the

low-equity group; those with higher EI-Using scores had higher likelihoods of

being in this group. Membership in the high-equity group is related to aspects of

both personality and EI; higher BF-Neuroticism and higher EI-Managing are both

independently related to lower likelihood of being in the high-equity group.

When a set of controls was included for demographic and other individual and

household characteristics in the analysis, high BF-Agreeableness emerged as a

distinguishing characteristic of those in the low-equity group but high EI-Using

scores remain associated with greater likelihood of membership in the low-equity

group. Among those individuals who responded to the additional EI items,

EI-Perceiving is also negatively related to the likelihood of membership in the lowequity

group. Inclusion of the controls did not alter the uncontrolled results in the

analysis of high-equity-group membership.

The results of the controlled regression analyses show that higher age and

greater educational attainment or interest in math, finance, or statistics led to

membership in the low-equity group whereas higher account balances and having

children tended to lower the likelihood of being in this group. Older age, a higher

account balance, and having a pension—all decreased the likelihood of being in the

high-equity group.

Exhibit 1. Equity Allocations (unrestricted): Summary of Multinomial

Probit Results


IRA Investors 401(k) Investors

Less Than

50% Stock

More Than

90% Stock

Less Than

50% Stock

More Than

90% Stock

Uncontrolled EI-Using (+ +) BF-Neuroticism

(– –)


(+ +)


Controlled BF-Agreeableness


EI-Managing (– –) BF-Agreeableness

(+ +)


(+ +)

EI-Using (+ +) BF-Neuroticism

(– –)

EI-Using (+) EI-Using (+)


(+ +)

EI-Managing (–) EI-Understanding

(– –)

Demographics Age (+ +) Male (+ +) Age (+ +) Age (– –)

AcctBal (– –) Age (– –) MidEdu (+ +)

FinIntrst (+ +) AcctBal (– –)

Children (– –) Pension (–)

ReadsOften (+)

High assets at



(– –)

BF-Neuroticism (–) BF-Agreeableness (+) EI-Using (+)

Age (+ +) IMP-Premeditation

(– –)



EI-Managing (–)

EI-Using (+ +) EI-Managing (– –) EI-Using (+ +) EI-Understanding

(– –)

Male (+ +) EI-Managing (– –) Age (– –)

AcctBal (–) EI-Understanding

(– –)

Married (– –) Age (+ +)

ReadsOften (+ +) LowEdu (+ +)

Pension (–)

Notes: (+ +) means positive regression coefficient, significant at the 5 percent level. (– –) means negative

regression coefficient, significant at the 5 percent level. (+) means positive regression coefficient, significant

at the 10 percent level. (–) means negative regression coefficient, significant at the 10 percent level.

Interpretation of these results is partially intuitive and partially backed by

empirical findings from other areas. In terms of the psychological characteristics,

popular belief holds that individuals high in BF-Neuroticism should be risk averse.

Our results suggest that BF-Neuroticism could be an asset, because these individuals

are neither risk seekers nor risk averse and are most likely to fall into the

advantageous medium-equity group. This conclusion receives added weight when

only those individuals who held a significant fraction of their financial wealth at

Vanguard are considered. In those regressions, the only robust effect of the

psychological and emotional measures is that of BF-Neuroticism, and it reduced

the likelihood of being in the high-equity group. Because moods and emotions can

be useful in decision making, individuals who experience more moods and emotions

than others may have more opportunity to integrate them effectively into their

decision making. Although more analysis is needed, these findings also highlight

the importance of not jumping to simple conclusions about how individual differences

such as neuroticism or “emotionality” influence real investment behavior.

Investors high in the ability to manage emotions should be less inclined to be

either risk averse or risk seeking (Salovey 2001), and as predicted, those with a

higher EI-Managing score were less likely to fall into the high-equity group. People

who feel comfortable dealing with their emotions should be competent enough to

avoid at least the largest emotional traps of the investment universe. The fact that

the EI-Using score strongly influenced the likelihood of belonging to the low-equity

group is less clear, and more analysis will be necessary to determine whether this

finding is meaningful.

Given our assumption that a middle level of equity in the investment portfolios

is the superior approach, the notion that those investors with greater backgrounds in

math or finance are more likely to have low levels of equity in their retirement

portfolios is somewhat puzzling. A possible explanation is that many of these

individuals work in the field of finance or in an area where their future earnings are

subject to financial risk related to the equity markets; they may tend to hold low-risk

assets in the their retirement or other portfolios to hedge these professional risks.24

Within the 401(k) sample in Exhibit 1, in the set of results without demographic

controls, only the BF-Agreeableness score had a strong positive influence on membership

in the low-equity group; none of the psychological or EI measures played a

significant role in differentiating members of the high-equity group from the

medium-equity group. Nyhus and Webley (2001) found that agreeable individuals had less savings and were more likely to borrow money than other individuals.25 This

result was unchanged by the inclusion of the set of demographic controls in the

regression framework.

Trading or Transaction Frequency. We examined transaction behavior

in terms of the number of days our sample investors (i.e., survey respondents)

made a transaction during 2004–2006.26 Like Barber and Odean (2001), we found

that within the group of respondents, men tended to trade more frequently than

women. Figure 1 shows the relationship for our 401(k) sample from the first quarter

(Q1) of 2005 to the second quarter (Q2) of 2007. The same relationship held for

the IRA respondents.27 Similarly, Figure 2 demonstrates a significant wealth effect

for 401(k) respondents: Investors with the higher reported financial wealth also

tended to trade more. We also found a positive relationship between the mutual

fund balance held at Vanguard and total transactions.28

To control for differences in these variables while assessing the importance of

emotional intelligence and other characteristics in explaining differences in transaction

activity, we used a mathematical model to predict the number of transactions

an investor would be expected to make based on his or her education, wealth,

income, gender, and psychological characteristics.29 The model allowed us to

estimate the effect of each characteristic, independent of the other characteristics,

on the “incidence rate” (i.e., likelihood of occurrence) of a transaction. We could

control for differences in education, wealth, and other variables when estimating

the relationship between EI and transaction activity.

An obvious extension of this work, which we will pursue in subsequent analyses, is to implement a

“hazard” model that would allow the probability of an event to vary potentially with the time since

the last event as well as with investor characteristics.

Figure 1. Transaction Days vs. Gender: 401(k) Investors

Figure 2. Transaction Days vs. Financial Wealth: 401(k) Investors

Average Number of Transaction Days per Quarter










Q1 05 Q2 05 Q3 05 Q4 05 Q1 06 Q2 06 Q3 06 Q4 06 Q1 07 Q2 07

Average Number of Transaction Days per Quarter










Q1 05 Q2 05 Q3 05 Q4 05 Q1 06 Q2 06 Q3 06 Q4 06 Q1 07 Q2 07

Because the subject of the analysis was explicitly transaction behavior, we had

to use sampling weights to reweight our sample to reflect the explicit oversampling

of individuals with transactions, or we had to treat the two groups for which we

altered the sampling frame separately. We had the information needed to reweight

the IRA sample by using broader population data, but this information was not

readily available for the 401(k) sample. Therefore, we treated the two groups of

sampled individuals differently in the 401(k) analysis.

Exhibit 2 summarizes the results of estimating the model parameters via a

regression analysis.30 These statistical results show strong effects of the reasoning

aspects of emotional intelligence (EI-Using and EI-Managing) and of impulsiveness

on the frequency of transaction behavior. (Recall that IMP-Urgency decreases

with the degree of impulsiveness; thus, the result in Exhibit 2 indicates that the

more impulsive individuals trade more.) The size of the effects from the regressions

can be quantified; in the case of EI-Managing, the coefficient estimates (not

reported in Exhibit 2) suggest that for each 1 unit increase in the EI-Managing

score, the likelihood of a transaction occurring decreased by about 1.5 percentage

points. Thus, a 10 percentage point difference in score on “managing emotions” led

to a 15 percentage point difference in the likelihood of a transaction. Such a 10

percentage point difference in EI scores was quite common in the survey data; the

standard deviation of the test score measure was roughly 10 percentage points.

30For brevity, Exhibit 2 reports only the estimates for statistically significant regressors; the full

regression results are reported in Tables B5 and B6 in the online supplemental material.

Exhibit 2. Frequency of Transaction Behavior: Summary of Poisson

Regression Results

401(k) Investors

Outcome IRA Investors Transactors Nontransactors

Uncontrolled EI-Managing (– –) BF-Agreeableness (– –) EI-Using (– –)

IMP-Urgency (– –)

Controlled IMP-Urgency (–) BF-Agreeableness (– –) IMP-Urgency (– –)

EI-Using (–) IMP-Urgency (– –) IMP-Premeditation (–)

EI-Managing (– –) EI-Using (– –)

Demographics Age (+ +) Male (+ +) Male (+)

AcctBal (+ +) Low assets (–) AcctBal (+)

%Stock (– –) Pension (–) %Stock (– –)

ReadsOften (+ +) Pension (–)

Children (+)

Note: See the notes to Exhibit 1.

Our finding that “managing emotion” has the strongest statistical significance

and largest quantitative effect on propensity to trade is similar to past findings.

Previous research has shown that the managing-emotions branch of EI has the best

predictive power for several behavioral outcomes, including everyday behavior of

young adults (Brackett, Mayer, and Warner 2004), quality of social interactions

(Lopes, Brackett, Nezlek, Schьtz, Sellin, and Salovey 2004), perceived quality of

social relationships (Lopes, Salovey, and Straus 2003), and such job-related variables

as performance, affect, attitudes at work, and leadership potential (Lopes,

Grewal, Kadis, Gall, and Salovey 2006).

Indexing vs. Active Investing. In addition to the asset allocation

decision, investors must decide whether to use index funds, active investment

management, or both for their mutual fund assets. Many factors may play a role in

this active versus passive decision, including the investor’s beliefs about market

efficiency, desire to minimize costs, and willingness to accept the risk in actively

managed funds (in exchange for positive expected risk-adjusted return). To our

knowledge, no empirical research has been carried out on what factors influence

investors’ choice of active versus passive management.

Our focus here is on the following question: Given that the investor has some

positive percentage in equities, what proportion of that equity allocation is indexed?

For IRA investors in our sample, approximately 25 percent had an all-active

portfolio whereas about 12 percent had an equity portfolio that was entirely indexed;

a fairly uniform distribution lay between these two endpoints. As was shown in

Table 5, the median allocation to indexed equities was 35.2 percent for IRA

investors and 24.5 percent for 401(k) investors.

To model the passive–active decision, we formed three groups of investors

with different degrees of indexed-equity exposure in their retirement accounts:

individuals whose equity funds were all actively managed (the 0 percent, or noindex,

group), those who held both active and passive funds (the 1–99 percent, or

some-index, group), and those who held only equity index funds (the 100 percent,

or all-index, group).

Again, we used an unordered multinomial probit model to isolate psychological

and demographic characteristics related to membership in these three groups. We

elected to use this procedure because our prior assumption was that both the noindex

and all-index investors had investment philosophies different from those of

the some-index investors. We do not assume that membership in any of these

groups implies better investment decision making; the purpose was simply to

distinguish and describe membership in the groups.

Summary details of the regression analyses for the no-index and all-index

investors in IRA and 401(k) groups are presented in Exhibit 3.31 Examining the

IRA results obtained without controls for various demographic characteristics, we

see that only the IMP-Premeditation score influenced membership in the no-index

group. Membership in the all-index group was related to two aspects of EI: Higher

EI-Using and lower EI-Perceiving scores were independently related to a higher

likelihood of being in the all-index group.

When we controlled for the demographic variables in the IRA investors group,

the IMP-Premeditation and EI results from the uncontrolled analysis remained. In

addition, we found that IRA investors with low EI-Perceiving scores are more likely

to have an all-active equity portfolio.

For 401(k) investors, in the uncontrolled regressions, the EI-Understanding

score had a weak influence on the active–passive allocation. Those investors with

a low EI-Understanding score had a higher likelihood of an all-active equity

portfolio; those with a higher EI-Understanding score had a higher likelihood of

having not an all-indexed portfolio but an equity portfolio that had both actively

and passively managed funds. This result is consistent with the positive and

statistically significant correlation between percentage of equity allocation indexed

and the EI-Understanding score.32 When we controlled for the demographic

variables, we found no statistically significant relationships.

31The full regression results are in Tables B7 and B8 in the online supplemental material.

Exhibit 3. Percent of Equity Allocation Indexed: Summary of Multinomial

Probit Results

IRA Investors 401(k) Investors

Outcome 0% Indexed 100% Indexed 0% Indexed 100% Indexed

Uncontrolled IMP-Premeditation

(+ +)

EI-Perceiving (– –) EI-Understanding



EI-Using (+ +)

Controlled EI-Perceiving (–) EI-Perceiving (– –) None None

EI-Using (+ +)

Demographics AcctBal (– –) AcctBal (– –) AcctBal (– –) AcctBal (– –)

%Stock (– –) %Stock (– –) %Stock (– –) %Stock (– –)

Income <

$250,000 (+)

Note: See the notes to Exhibit 1.

Retirement account balances and the percentage allocation to equities demonstrate

the same tendency for both the IRA and 401(k) investors. Those with

higher balances and those with higher equity allocations had higher likelihoods of

owning both active and index equity funds. In addition, for 401(k) investors, those

with incomes less than $250,000 were more likely to use all-index funds for their

equity allocations.

In addition to these results, which were obtained via a multinomial probit

approach, we also considered a simpler probit specification in which the dependent

variable was set to 1 if all of the equity mutual funds owned in the IRA

accounts were invested in index funds and set to 0 otherwise. In these regressions,

we found that none of the personality or impulsiveness variables was statistically

significant (either with or without demographic controls). The coefficient on EIUnderstanding

was positive, relatively large, and statistically significant both with

and without controls. In terms of controls in this framework, we found that

relatively low education had a negative influence on being an all-index investor,

as did the size of the IRA account.

Interpretations of these active–passive results and their relationships with

impulsiveness and EI scores should be made with caution. In terms of the results

in the 401(k) sample, the opportunity to use index funds in the 401(k) may be

significantly influenced by the choices made by investors’ employers rather than by

the investors; that we found no strong results is perhaps not surprising. Within the

IRA universe, scores in IMP-Premeditation (recall that our scoring convention is

that higher IMP-Premeditation implies greater tendency to premeditate) suggest

that investors who have chosen an all-active portfolio may have put great care and

energy into selecting active managers. Higher scores in EI-Using reflect the ability

to use emotions to enhance thinking and problem solving. Thus, investors skilled

in this ability may believe a better use of emotional energy is to place money into

indexed funds rather than deal with the risks of active management. Why EIPerceiving

seems to work in the opposite direction is not clear.

Finally, that investors choose both active and passive funds as their account

balances and equity allocations rise may reflect a tendency for investors to spread

holdings over more and more funds as they gain investment experience.

Adoption of International Funds. Because mutual fund inflows have

been shown to be strongly correlated with past fund performance (Ippolito 1992;

Goetzmann, Massa, and Rouwenhorst 1999; Grinblatt and Keloharju 2001), we

investigated whether relationships existed between adding international funds and

the psychological and demographic characteristics of our IRA investors during the

period of our study—a time when international equities performed exceptionally

well on both an absolute and a relative basis.33 For the dependent categorical variable, we subdivided these investors into three groups: those who owned no

international fund(s) as of September 2007, those who had at least one international

fund as of either December 2003 or December 2004 (early adopters), and those

who had at least one international fund as of either December 2005, December

2006, or September 2007 (late adopters).34

From 31 December 2003 to 30 September 2007, the number of IRA investors

who had at least one international fund gradually increased. Only 25 percent owned

international funds at the end of 2003, but by the end of September 2007, 52 percent

of the IRA investors in the study had some international exposure.

To isolate the psychological and demographic characteristics related to membership

in the three adopter groups, we again used an unordered multinomial probit

model. In this case, our prior assumption was that adding international funds during

a strong performance period may be chasing performance and thus be suboptimal.

Investors who witnessed the subpar performance of many non-U.S. equity markets

in the 1990s and early part of this decade, however, may have been slow to revise

their beliefs that domestic equities offer the best risk-adjusted returns. Thus, the

outperformance of international equities witnessed since 2003 may have been a

wake-up call that a more diversified portfolio would be prudent.

Exhibit 4 summarizes the results of the regressions for the IRA investors.35

Examining the uncontrolled IRA results, we find that BF-Agreeableness has a

statistically significant negative coefficient for both the early adopter and late adopter groups; those with higher BF-Agreeableness scores had a higher likelihood

of having had no international exposure during 2004–2007. Those with high BFExtraversion

scores were more likely to have been early adopters, those with high

EI-Using scores were less likely to have been early adopters, and those with high

BF-Neuroticism scores were less likely to have been late adopters.

All of these significant results remained when we included the set of demographic

controls in the analysis. In addition, investors with high BF-Conscientiousness scores

were more likely to have had no international exposure. Married investors were more

likely to have had no international exposure. And those with higher account balances

were more likely to have been early adopters, whereas those with higher allocations

to equities were more likely to have been late adopters.

Note that the BF-Agreeableness and EI-Using variables were also significant

for the regressions involving percentage of equity in the investor’s portfolio (see

Exhibit 3). In the case of both variables, a higher score made it more likely that the

investor had a lower stock allocation (i.e., less than 50 percent). In Exhibit 4, those

IRA investors with higher BF-Agreeableness scores were less likely to have any

allocation to international equities and those with higher EI-Using scores were less

likely to have invested in international equities before 2005.

The positive relationship between BF-Extraversion and adoption of international

equity may relate to willingness among extraverts to try new things because

it shows up most strongly in the early adopters (the point estimate for late adopters

is of the same sign and magnitude but with no statistical significance). In addition,

BF-Neuroticism negatively affected the likelihood of being either an early or late

adopter, although it is not statistically significant in the case of early adopters.

International equity is widely perceived to be less familiar to U.S. investors (and

perhaps more risky) than domestic equity, and BF-Neuroticism here may reflect a

lower level of comfort with this kind of uncertainty.36

Internal Rate of Return. To analyze variations in internal rates of

return achieved by investors, we began by computing quarterly IRRs at the

account level for all survey respondents in both samples. Our data for IRA

investors cover Q1 2004 through Q1 2007, and for 401(k) investors, Q1 2005

through Q2 2007. In both the IRA and 401(k) samples, given that the IRR

calculation requires a beginning and an ending balance, we necessarily excluded

observations for which we had no beginning and/or ending asset balance for the quarter. In addition, in the 401(k) sample, we excluded observations for those

investors who had an outstanding loan during the quarter.37

We computed the account opening balance (treated as a positive cash flow),

IRRs from all net cash flows to the set of accounts monitored for each investor on

each day of each quarter, and the closing balance (treated as a negative cash flow).

The IRRs reflect both the returns generated by the particular set of funds that each

investor selected and held over each period and the impact of the timing of cash

flows in or out of the various accounts and funds. IRRs can be volatile, especially

when large amounts of money remain invested for a small number of days or when

cash flows are similar in magnitude to the overall invested balance.38

To estimate the relationship between the various psychological characteristics

and the attained rates of return, we used a “stacked” or panel regression technique

in which all quarterly observations of each member of the sample over the period

were included:


where the dependent variable, Rit, is the quarterly rate of return for individual i at

time t. The independent variables and controls include the vector of psychological

characteristics of interest, for each individual; a set of control variables for other

individual characteristics, X; and a set of period dummy variables, T, that captures

differences in the average returns achieved by sample members in each period.

We estimated several specifications of Equation 1. In Exhibit 5, we summarize

the results from a random-effects panel model in which we assumed that the error

term in the equation can be decomposed into two components—an investorspecific

random effect and a standard noise term.39 The results in Exhibit 5 show

that BF-Extraversion had a direct effect on the IRRs of IRA investors and that

IMP-Urgency and IMP-Premeditation played a role in the IRRs of 401(k)

investors, although the effects of these two variables tended to offset one another.

The fact that a psychological variable showed a direct effect on investment returns

in these regression analyses is impressive. Controlling for other characteristics

weakened the impact of IMP-Urgency, but the IMP-Premeditation result

remained. This outcome suggests that those with low levels of impulsiveness (high

IMP-Premeditation on our scale) receive lower returns on their investments.

Several explanations are possible. In the case of IRA investors, it could be that

extraverts gather more information from various sources and are thus better

equipped than their peers to make good decisions (Wanberg and Kammeyer-

Mueller 2000). In the case of 401(k) investors, investors high in openness are

probably also open to risk (Nicholson, Soane, Fenton-O’Creevy, and Willman

2005). Extraverts are also more willing to take risks (Nicholson et al. 2005). Finally,

Whiteside and Lynam (2001) and Zuckerman and Kuhlman (2000) documented

that individuals high in impulsiveness are more willing than others to take risk.

The IRRs in Exhibit 5 are raw returns that do not take volatility into account.

When assessing the skill of investors or the value of an investment in relation to

portfolio performance, however, controlling for risk or volatility of the return

sequence is usually important. Therefore, in addition to the regression summarized

in Exhibit 5, we examined a second specification in which we computed Sharpe

ratios for those investors for whom we could compute quarterly IRRs for all quarters

in the panel regressions of Equation 1. The Sharpe ratios were formed by subtracting

the average quarterly return for the Vanguard Prime Money Market Account

for each quarter from the investor’s quarterly IRRs, computing the average of these

“quarterly excess return” numbers for each investor, and dividing by its standard

deviation (again, for each investor). We then used this “individual Sharpe ratio” as

the dependent variable in a standard ordinary least-squares (OLS) regression. These

results are summarized in Exhibit 6.40

Exhibit 5. Internal Rates of Return: Summary of Panel

Regression Results

Outcome IRA Investors 401(k) Investors

Uncontrolled BF-Extraversion (+) IMP-Urgency (+ +)

EI-Using (–) IMP-Premeditation (– –)

Controlled None IMP-Premeditation (– –)

EI-Perceiving (+)

Demographics Age (– –) Low education (– –)

Low assets (– –) High AcctBal (– –)

%Stock (+ +) Low income (–)

%Stock (+ +)

High assets at Vanguard High AcctBal (–) IMP-Premeditation (–)

Low asset (– –) Low education (– –)

%Stock (+ +) Mid education (– –)

%Stock (+ +)

Note: See the notes to Exhibit 1.

These results show that in the overall population of respondents, we found

negative effects on IRRs (with marginal statistical significance) for BFAgreeableness,

IMP-Urgency, and EI-Using within the IRA sample and found

no significant effects within the 401(k) sample. In the cases of the impulsiveness

result and the EI result, however, our regression estimates provided offsetting

and quantitatively similar point estimates for other characteristics within the same

group of personality traits in the same regression. Specifically, in these regressions,

the magnitude of the coefficient on IMP-Premeditation is of nearly the

same size but of the opposite sign from the coefficient on IMP-Urgency, whereas

the coefficient on EI-Managing is of nearly the same size but of the opposite sign

from the coefficient on EI-Using. In both of these cases, formal F-tests of the

hypotheses that these pairs of coefficients are jointly zero could not be rejected

at the 10 percent level.

Finding relatively strong relationships between our psychological characteristics

and the investment behaviors we measured but finding only weak and unclear

relationships between these same characteristics and investor account IRRs suggests,

on the basis of our model, that no single character trait dramatically increases

Exhibit 6. Sharpe Ratios of Internal Rates of Return,

Summary of OLS Regression Results

Outcome IRA Investors 401(k) Investors

Uncontrolled BF-Agreeableness (–) None

IMP-Urgency (–)

EI-Using (–)

Controlled BF-Agreeableness (–) None

IMP-Urgency (–)

EI-Using (–)

Demographics Male (– –) No advice (– –)

Age (– –) FinIntrst (+ +)

AcctBal (+ +) Children (–)

Low assets (– –)

No advice (–)

FinIntrst (– –)

High assets at Vanguard,

controlled EI-Using (–) EI-Managing (–)

EI-Managing (+ +) Low education (– –)

Low assets (– –) No advice (– –)

Children (–) FinIntrst (+ +)

Note: See the notes to Exhibit 1.

or decreases IRRs. Rather, psychological characteristics only help predict specific

behavior and attitudes, which within specific situational constraints, may help predict

differences in IRR. Therefore, it makes sense to gain an understanding of which

characteristics are related to which behaviors before building complex economic

models of “investment skill” that may include specific traits.

A notable point about the demographic controls is that in the analysis of Sharpe

ratios for IRA respondents, the effect of the financial interest variable is statistically

significant and negative. In the previous set of results (which examined raw returns),

IRA investors with greater financial interest appeared to achieve higher returns, but

these investors did not receive higher returns than peers when volatility-adjusted

returns were considered.41 In the sample of 401(k) investors, however, Exhibit 6

shows that the effect of financial interest on IRRs remained positive despite the

volatility adjustment.

An important element in interpreting these results is recognizing that 401(k)

investors may have very different motives in forming their portfolios from the

motives of IRA investors. In particular, many IRA investors may hold only a portion

of their total retirement assets in IRAs at Vanguard. Thus, the Sharpe ratio we

computed would not include the diversifying effect of other holdings. A relatively

larger portion of 401(k) investors may have the bulk of their financial assets in that

plan and, therefore, be holding a balanced portfolio within their Vanguard account.

Indeed, when we examined the results for the Sharpe ratios among only those

investors who (in our estimate) held a significant portion of their reported wealth

at Vanguard, we found the estimated coefficient on financial interest lost statistical

significance; it was roughly one-half the size of the same coefficient in the overall

sample regression.

Summary and Discussion

Exhibits 7–9 summarize the results presented in previous sections, but the information

is organized by psychological construct rather than by investment behavior. In

each exhibit, we broke out the major areas of investment behavior/results that we

studied (risk taking, various aspects of the asset management process, and investment

returns) and show how each component of the three overarching psychological

constructs (emotional intelligence, personality, and impulsiveness) relate to each

specific behavior in uncontrolled and controlled regression specifications. We hope

that these summaries provide a clear picture of the roles that EI, personality, and

impulsiveness play in overall investment behavior.

Emotional Intelligence. In general, those high in EI were somewhat

more conservative and less aggressive in risk taking than those low in EI (see

Exhibit 7). In particular, they often held less than 50 percent of their assets in

stocks and were unlikely to hold more than 90 percent in stocks. They were also

less likely to trade or make changes to their portfolios; at the same time, they were

more likely to use index funds as a part of their portfolios. Together, these results

suggest that individuals high in EI are less likely to make extreme decisions and

prefer to pursue a more balanced investment approach. We found no systematic

relationship between EI and either IRRs or Sharpe ratios.

Although all EI groups contributed to the overall investor profile, the Using

group showed the strongest overall effect. We had originally hypothesized that the

Managing and Using branches would play the largest role in investment decisions,

which is the reason all participants were asked to fill out at least these two sections

of the EI survey (and had the option of stopping after these two sections). Indeed,

the Understanding branch did play a minor role and the Perceiving branch did play

an intermediate role.

The minor role played by Understanding emotions could point to the difference

between conceptual knowledge of emotions and practical, procedural knowledge.

Understanding emotions is useful in decision making only if the person also knows

how to perceive, use, and manage emotions. Thus, although understanding emotions

contributed to the overall role of EI in our measures of financial behavior, it

had only a minor main effect.

Finally, individual differences in the Perception of emotion are especially

important for the quality of social relationships but apparently play a minor role in

the types of financial decision-making behavior we measured.

Personality. Each of the Big Five dimensions played a somewhat different

role for investment decision making (see Exhibit 8). First, Openness did not play

any systematic role. Second, Conscientiousness played only a minor role; the only

result was that those who scored high on this trait were less likely to be early adopters

of new investment strategies. This finding probably reflects the reflective and

planning dimensions of this trait; conscientious investors may wait to see how things

evolve before making a decision. Third, those high in Extraversion were more likely

to be early adopters, which may reflect the excitement-seeking dimension of this

trait. This last result could also reflect the larger social network these individuals

might have, which could provide a greater opportunity to hear about new investment

possibilities. The fact that Extraversion is related to IRR could simply reflect the

market conditions during this period, which favored the decision to invest in

international stock and other forms of risk taking. (The effect of extraversion did

not survive risk adjustment of the return numbers.) Possibly, in periods in which

momentum investing and early adoption of new types of investments or investment

strategies pay off, extraverts will earn higher-than-average returns.

Exhibit 7. Summary of Results for Emotional Intelligence

Risk Taking Management Activities Returns

EI Metric Uncontrolled Controlled Uncontrolled Controlled Uncontrolled Controlled

Perceiving No effect (+ +) More likely

to allocate <50%

stock (IRA)

(– –) Less likely to

have 100% indexed

funds (IRA)

(– –) Less likely to

have 100% indexed

funds (IRA)

No effect (+) Positively related

to IRR [401(k)]

(–) Less likely to

have 0% indexed

funds (IRA)

Using (+ +) More likely

to allocate <50%

stock (IRA)

(+ +) More likely

to allocate <50%

stock (IRA)

(+ +) More likely to

allocate 100%

indexed funds (IRA)

(+ +) More likely to

allocate 100%

indexed funds (IRA)

(–) Negatively

related to IRR


No effect

(+) More likely to

allocate <50%

stock [401(k)]

(–) Less likely to be

an early adopter of

international funds

(pre-2005) (IRA)

(– –) Less likely to

trade in the nontransactors


(– –) Less likely to

trade in the nontransactors


(–) Less likely to

be an early adopter of

international funds

(pre-2005) (IRA)

(–) Negatively related

to transaction behavior


Understanding No effect No effect (–) Less likely to

have 0% indexed

funds [401(k)]

No effect No effect No effect

Managing (– –) Less likely to

allocate >90%

stocks (IRA)

No effect No effect (– –) Negatively related

to transaction behavior


No effect No effect

Note: See the notes to Exhibit 1.

Emotional Intelligence and Investor Behavior

©2009 The Research Foundation of CFA Institute 35 Exhibit 8. Summary of Results for Personality

Risk Taking Management Activities Returns

BFI Metric Uncontrolled Controlled Uncontrolled Controlled Uncontrolled Controlled

Agreeableness (– –) Less likely to

be an early adopter


(+) More likely to

allocate <50%

stock (IRA)

No effect (– –) Less likely to trade

in the transactors

group [401(k)]

No effect No effect

(– –) Less likely to

be a late adopter


(–) Less likely to be

an early adopter


(+ +) More likely to

allocate <50%

stock [401(k)]

(– –) Less likely to be

a late adopter (IRA)

(+ +) More likely to

allocate <50%

stock [401(k)]

(+ +) More likely to

allocate >90%

stock [401(k)]

Conscientiousness No effect No effect No effect (–) Less likely to be

an early adopter

(pre-2005) (IRA)

No effect No effect

Extraversion (+) More likely to be

an early adopter

(pre-2005) (IRA)

(+) More likely to be

an early adopter

(pre-2005) (IRA)

No effect No effect (+) Positively related

to IRR (IRA)

No effect

Neuroticism (– –) Less likely to

allocate >90%

stock (IRA)

(– –) Less likely to

allocate >90%

stock (IRA)

No effect No effect No effect No effect

(– –) Less likely to

be a late adopter

(post-2005) (IRA)

(– –) Less likely to

be a late adopter

(post-2005) (IRA)

Openness No effect No effect No effect No effect No effect No effect

Note: See the notes to Exhibit 1.

Perhaps, the most interesting relationships are in the realm of Agreeableness

and Neuroticism. At first glance, the findings for agreeable investors send mixed

messages. They were more likely to place less than 50 percent of their assets into

stocks but also more likely to place more than 90 percent of their assets into stocks.

These results indicate extreme behavior because 50–90 percent in equities is often

considered a reasonable, balanced level of exposure. They were less likely to trade

within their 401(k) accounts, however, which shows conservative behavior or

inertia. Finally, they were neither early adopters nor late adopters of international

investing but were somewhere in the middle. This result supports a conclusion that

they had passive or “wait and see” attitudes.

In summary, they were extreme (in choosing a level of risk) yet conservative (in

trading and adopting new trends). How can this result be explained? Taking note

of several studies on the impact of 401(k) plan design on investor behavior (Madrian

and Shea 2001; Utkus and Young 2004; Holden and VanDerhei 2005), we believe

part of the answer lies in the 401(k) structure, which is where we found agreeable

investors taking extreme levels of equity risk. Most companies control important

aspects of the design of their employees’ 401(k) plans. Some sponsors opt for a

“default” arrangement in the plan that can lead to high equity exposure (for example,

if the employer matches savings with company stock). Other sponsors use a default

investment strategy that involves low equity exposure—for example, defaulting

participants into a “guaranteed investment account” (sometimes called a “guaranteed

insurance contract”) or a money market account. If employees do not opt out

of the default plan, their assets will simply be placed in the default plan chosen by

the employer. Thus, agreeable people may be ones who are unlikely to make changes

to the default strategy and find themselves in whatever strategy the employer chose.

This explanation is supported by the fact that agreeable investors are least likely to

trade within their 401(k) accounts. If one accepts this explanation, then the rest of

the data make sense: Agreeable IRA investors are more likely to choose less than

50 percent equity and are neither late nor early adopters. Agreeable investors are

thus apparently likely to be relatively conservative and passive. A detailed analysis

of the relationship between Agreeableness and an investor’s susceptibility to accepting

default plans is an interesting and important topic for future research.

Popular belief holds that individuals high in Neuroticism are highly anxious or

emotional and that this trait is a disadvantage in investment decision making. Our

results provide a more balanced picture. Although we found that investors high in

Neuroticism were less likely to put more than 90 percent of their assets into stocks,

we also found they did not place less than 50 percent in stocks. They were

conservative and less likely to be early adopters, but they did not exhibit extreme

emotional, risk-averse, or “fear-based” behavior. Indeed, we found that being

anxious can have positive as well as negative effects. If anxious individuals worry

about their financial future, they may spend more time searching for information

and choosing the best options. A nonanxious person may falsely believe that things

will work out no matter what and thus not take the time to select the best investment

option. Of course, if an individual is too anxious, or does not know how to manage

the anxiety, the anxiety may lead to “freezing” behavior in which no decision is made.

Future research should examine the cumulative and interaction effects of

various psychological variables. For example, do individuals high in Neuroticism

and also high in EI make good decisions? In contrast, do individuals high in

Neuroticism and high in Impulsiveness make counterproductive decisions?

Impulsiveness. First, we found Impulsiveness, especially the measure of

Urgency, to be strongly related to high transaction activity in both IRA and 401(k)

accounts (see Exhibit 9). Second, those high in Premeditation were more likely to

have none of their assets (in the Vanguard accounts) in indexed funds. Because a

high volume of trading and a lack of indexed funds are considered less-than-optimal

behavior (Barber and Odean 2000; French 2008), this result would help explain

why both Impulsiveness dimensions were strongly and negatively related to IRR.

In summary, Impulsiveness is the only psychological construct that seems to have

clearly negative relationships with the chosen financial indicators.

Exhibit 9. Summary of Results for Impulsiveness



Risk Taking Management Activities Returns


Lack of


No effect No effect (+ +) Less likely to

have 0% indexed

funds (IRA)

No effect (– –)


related to



(– –)


related to



Urgency No effect No effect (– –) More likely

to trade in the

transactors group


(– ) More likely

to trade (IRA)

(+ +)


related to



No effect

(– –) More likely

to trade in the

transactors group


(– –) More likely

to trade in the


group [401(k)]

Notes: The impulsiveness metrics have values that decrease with greater impulsiveness, which is why the signs

of the effects (+/–) and the verbal interpretations of those effects are in opposite directions. See also the

notes to Exhibit 1.

U = Uncontrolled; C = Controlled.


We followed a conservative and robust strategy in analyzing the data for this report;

nevertheless, we found some strong and consistent relationships between our

psychological constructs and real financial behavior. As a result, we can conclude

that individual differences are not simply noise within economic models but, rather,

play a larger role than even we expected.

The psychological variables we examined played a larger role within IRA

accounts than within 401(k) accounts. This finding makes sense because the

decision to place money into an IRA account is an individual one whereas the

placement of assets in 401(k) accounts depends on the sponsor’s offerings. In

particular, the choice may depend on the default option. As a result of mental

accounting or similar phenomena, investors may view assets within the two types

of account very differently, and our results may reflect those differences.

We found indications of important relationships among EI, other psychological

characteristics, and investment behavior in several, but not all, areas that we

examined. The value of these findings, and of the growing body of similar research,

is that they underscore the importance of identifying the specific psychological

mechanisms that guide investment decisions. Although these early results are

suggestive, they are not the final word.

Applied to our data, this logic means that we should examine more complex

patterns of behaviors in relation to our psychological characteristics. For example,

we found a relatively strong effect of psychological variables on risk taking. But

individuals high in impulsiveness may demonstrate other types of risk-taking

behavior; for example, these individuals may be especially likely to have nonindexed

stocks and never ask for advice. People high in openness may invest in

indexed funds and have portfolios high in equity only if they also take advice and

read financial information.

In addition, we should examine interaction effects. For example, even though

men apparently trade more frequently than women, the reason could be that women

high in EI or high in impulsiveness trade more than women who are low in these

traits. Perhaps, women high in impulsiveness or in EI trade more than men who

are low in impulsiveness or in EI. Teasing out such subtle differences should shed

more light on the relative influence of different psychological characteristics for

investment behavior patterns.

As researchers pinpoint the sources of investor biases, particularly those that

lead to investor mistakes, the investment industry can use this information to

develop products and services that may help save investors from sabotaging their

financial futures. Further pursuit of this line of research might result in tailoring

asset allocation advice on the basis of information from in-depth interaction between the investor and financial adviser, which could even include simple

psychological tests. The advice would be based not only on an investor’s financial

goals and risk tolerance but also on the investor’s psychological characteristics

(including his or her emotional intelligence).

Our data suggest that simply identifying personality types might reveal particular

biases or predispositions that affect investment outcomes. A key question would

be whether investors would make the same choices or would alter their behavior

after having their biases revealed. New portfolio construction methods that combine

the best of mathematical finance with rigorously quantified psychological metrics

could be used to improve the models that practitioners use in giving financial advice

and could create portfolios that enhance investors’ likelihood of reaching their

financial goals.