Debt Puzzle

Review of Finance, 2018, 2109–2137
doi: 10.1093/rof/rfx020
Advance Access Publication Date: 14 July 2017
The Credit Card Debt Puzzle and

Noncognitive Ability*
Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Hwan-sik Choi1 and Ron A. Laschever2
1
Binghamton University and 2Compass Lexecon
Abstract
Many households concurrently hold low-yield liquid assets while incurring costly
credit card debt. In our sample, more than 80% of households with credit card debt
also have low-yield liquid assets. Using data from the Health and Retirement Study
(N ¼ 30,517), we examine the role of noncognitive skills as well as the economic,
financial, and demographic factors that affect the likelihood of co-holding. We find
that the “Big Five” personality traits have a statistically significant and economically
important effect: households with a more agreeable, introvert, and less conscien-
tious head of household are more likely to co-hold. We also examine the role of
intra-household dynamics.
JEL classification: D1, D03, D12, D14

Keywords: Personality, Household finance, Liquidity, Self-Control, Intra-household dynamics,
Financial literacy
Received May 30, 2016; accepted March 12, 2017 by Editor Burton Hollifield.
1. Introduction
In the US, almost 40% of all households carry a credit card balance with an average interest
rate of 13% (Bricker et al., 2012). Most households with credit card debt also hold consid-
erable amounts of low-yield liquid assets such as checking and savings account balances
that have a negligible return. Gross and Souleles (2002) report that among households with
credit card debt, 95% have positive net wealth and almost 70% have positive home equity
that can be used to get lower-cost home equity loans to pay down their credit card loans.
Among the households in our sample that have a credit card balance not paid in full (22%
of households), more than 84% simultaneously have a positive checking and/or savings
* We thank an anonymous referee, David Gross, Sol Polachek, colleagues at Binghamton and
Purdue Universities, and seminar participants at University of Seoul, Korea University, Seoul
National University, Ehwa Woman’s University, and KAIST (Korea Advanced Institute of Science
and Technology) for helpful discussions and comments. The opinions expressed in this article are
the authors’ own and do not reflect the views of Compass Lexecon.
C The Authors 2017. Published by Oxford University Press on behalf of the European Finance Association.
V
All rights reserved. For permissions, please email: journals.permissions@oup.com
2110 H.-s. Choi and R. A. Laschever
account balance. This financial phenomenon is seemingly at odds with a no-arbitrage con-
dition and has been referred to as a “puzzle” in the literature (e.g., Gross and Souleles,
2002; Bertaut, Haliassos, and Reiter, 2009; Telyukova, 2013).
In this article, we study the role of noncognitive skills in explaining the credit card debt
puzzle using data for 12,976 households from the Health and Retirement Study (HRS).
Our focus on noncognitive skills is motivated in part by a growing literature in economics
that examines the role of cognitive limitations (Simon, 1955) and other psychological fac-
tors in explaining empirical anomalies in consumption and savings (Rabin, 1998), accumu-
lation of wealth (Ameriks, Caplin, and Leahy, 2003), portfolio choice (Barberis and Thaler,

2003), and labor market outcomes (Heckman, Stixrud, and Urzua, 2006). The HRS data
contain detailed longitudinal information on financial, economic, health, and psychosocial
measures, which allows us to investigate the role of noncognitive factors while controlling
for a host of financial and demographic variables.
We define the “puzzle group” as households with a positive credit card balance carried
over to the next billing cycle (commonly referred to as “revolvers”) and $500 or more in
low-yield liquid assets (checking, savings, and money market account balances). As in
Telyukova (2013), our preferred specifications use $500 as the threshold for low-yield
liquid assets, as such assets may be more convenient or necessary for certain types of
expenses.
There are two main components at the heart of our empirical identification strategy for
estimating the effect of the noncognitive factors. First, we exploit the longitudinal nature of
our data to overcome the inherent simultaneity between spending and saving decisions, as
well as other financial decisions. This also allows us to address other nonfinancial factors
such as a change in family composition or health shocks. Second, we exploit the constancy
of the noncognitive measures we use during adulthood. Though this assumption is used by
many, our data allow us to examine the validity of this assumption.
The finding of a credit card debt puzzle by Gross and Souleles (2002) has led to several
proposed explanations. Lehnert and Maki (2002) examine whether people strategically
increase credit card debt prior to filing for bankruptcy. Zinman (2007) finds a high pre-
mium on holding liquid assets. Becker and Shabani (2010) calculate that some households
would be better off redeeming their debt using their equity holdings. Fulford (2015) focuses
on the role of uncertainty in future credit availability that may lead households to not pay
down their debt. Telyukova (2013) also examines the demand for low-yield liquid assets by
developing a structural model in which credit card borrowers need low-yield liquid assets
for certain types of transactions for which credit cards cannot be used, such as rent or mort-
gage payments. Others have focused on the role of nonfinancial factors. Bertaut, Haliassos,
and Reiter (2009) construct an accountant-shopper model where high credit card debt is
used as a way to exert self- (or spousal-) control. The study by Gathergood and Weber
(2014) is the only one, to the best of our knowledge, to empirically examine the role of a
noncognitive skill (self-control) in explaining the puzzle. It studies the role of self-control
and financial literacy, and concludes that the former, rather than the latter, affects the like-
lihood of co-holding low-yield liquid assets and credit card debt, using a cross-section of
British households.
This article makes several important contributions to the existing literature. First, we
identify the factors that play a role in explaining the credit card debt puzzle by investigating
a much wider range of noncognitive skills than self-control. Second, using the rich data set
at our disposal our results complement the other types of explanations suggested in the
Credit Card Debt and Noncognitive Ability 2111
literature. For example, we control for the need for liquidity (e.g., Telyukova, 2013), and
self-control and financial literacy (e.g., Gathergood and Weber, 2014). Finally, our article
is the first to examine the effect of noncognitive abilities among (intra) household members
on households’ co-holding behavior.
To capture a broad and comprehensive range of noncognitive skills, we employ the “Big
Five” personality traits (McCrae and Costa, 1987, 1999). Personality traits are also referred
to by some as character skills, soft skills, or noncognitive abilities (see discussion in
Heckman and Kautz, 2012, p. 452). We follow the prevalent naming convention and also
refer to the traits as noncognitive skills. The five traits are Openness (O), Conscientiousness

(C), Extraversion (E), Agreeableness (A), and Neuroticism (N). For example, John,
Naumann, and Soto (2008, p. 120) describe Conscientiousness as “socially prescribed
impulse control . . . such as thinking before acting, delaying gratification, following rules,
planning, organizing, and prioritizing tasks”; Agreeableness is conceptually defined as
“prosocial communal orientation toward others . . . such as altruism, tender-mindedness,
trust and modesty.”1 We further discuss the measurement of the Big Five in Section 3.1.
The Big Five personality traits are by far the most commonly used in the field of psychol-
ogy and have been widely studied over the last couple of decades. The Big Five personality
traits provide several major advantages for our setting. First, they cover a very broad
domain of noncognitive abilities. Second, they have been extensively studied in other set-
tings, and have been shown to be a tractable set of measures for describing variation across
people in types of personality. Third, the measures have been shown to be relatively rank-
stable among adults, reducing the threat of endogeneity in our study.
Although psychologists have been studying correlations between various personality
traits and financial outcomes (e.g., income, debt, consumption, and saving) for several
decades,2 in recent years there has been a growing interest among economists in incorpo-
rating personality traits such as self-control, perseverance, and grit. Borghans et al.
(2008) and Almlund et al. (2011) provide an introduction to the recent developments in
the intersection of psychology and economics. Personality traits, including the Big Five,
have been shown to be an important complement to more traditionally economic meas-
ures of human capital in explaining education attainment, labor market outcomes,
wealth, etc.3
We build on several previous papers in the economics literature that have either: (i) con-
sidered only a narrow facet of noncognitive skills and its effect on savings, borrowing, or
1 Openness “describes the breadth, depth, originality, and complexity of an individual’s mental and
experiential life [with a behavioral example] of tak[ing] time to learn something simply for the joy of
learning.” Extraversion “implies an energetic approach toward the social and material world . . .
such as sociability, activity, assertiveness, and positive emotionality.” Neuroticism “contrasts emo-
tional stability and even-temperedness with negative emotionality, such as feeling anxious, nerv-
ous, sad, and tense.” Ibid. John and Srivastava (1999) provide an overview of the traits as well as a
historical account of the last several decades.
2 See Livingstone and Lunt (1992), Nyhus and Webley (2001), Norvilitis et al. (2006), Rabinovich and
Webley (2007), and Conti and Heckman (2014) for some examples.
3 Examples include earnings (Bowles, Gintis, and Osborne, 2001; Nyhus and Pons, 2005; and Mueller
and Plug, 2006); household finances (Brown and Taylor, 2014); educational attainment (Lundberg,
2013); and academic achievements (Heckman, Stixrud, and Urzua, 2006; and Heckman and Kautz,
2012).
the propensity to co-hold assets and debt (e.g., Ameriks, Caplin, and Leahy, 2003; Laibson,
Repetto, and Tobacman, 2003; Gathergood and Weber, 2014, respectively); or (ii) used a
neo-classical framework and do not consider noncognitive skills to explain the co-holding
of assets and debt (e.g., Bertaut, Haliassos, and Reiter, 2009; Telyukova, 2013; Fulford
2015). We combine these previous studies and their proposed mechanisms, and hypothesize
the channels through which personality traits operate. We hypothesize that four of the Big
Five personality traits might play a role in the likelihood that a household is in the puzzle
group: Conscientiousness, Extraversion, Agreeableness, and Openness might play a role for
the dimension of spending and/or borrowing; and Conscientiousness, Extraversion, and

Openness might play a role in the dimension of (not) using low-yield liquid assets to pay
down debt.
In Section 2, we consider three main channels through which co-holding may occur.
First, certain personality traits may help (hinder) a decision maker when dealing with their
finances. For example, those with higher levels of Conscientiousness might be more likely
to notice that they have sufficient low-yield liquid assets to pay down their debt. The sec-
ond channel we consider is precautionary saving for expected or unexpected liquidity
demand. The third channel focuses on the role of personality traits in intra-household (or
dual-self) dynamics. We find that even after controlling for differences in age and education
levels among couples, the personality of both partners explains some of the observed co-
holding patterns in the data.
We first implement a reduced-form approach. We find that the effects of
Conscientiousness, Extraversion, and Agreeableness tend to be the most persistent of the
Big Five personality traits across our various specifications. In our preferred specification a
one standard-deviation increase in Conscientiousness, Extraversion, and Agreeableness
changes the likelihood of being in the puzzle group by 0.54, 1.09, and 1.62 percentage
points, respectively. We find that this result holds after we additionally control for meas-
ures suggested in previous studies, such as liquidity demand, self-control, and financial
sophistication. In Section 4.1 we take into account the simultaneity between spending and
saving, and examine the role of personality in borrowing, and holding low-yield assets con-
ditional on borrowing separately.
Taken together, our findings suggest that regulatory policies, personal debt default
options, debt counseling, and educational programs are all domains that can be made more
cost effective by taking into account the role of noncognitive abilities. To illustrate the eco-
nomic significance of our results, we estimate that in the US, among those 50 years and
older, even a reduction of 1% in the number of households co-holding low-yield liquid
assets and credit card debt would translate into an annual decrease of $327 million in inter-
est payments while maintaining the same level of consumption.4
The rest of the article is organized as follows. We first describe our empirical framework
and source of identification in Section 2. The HRS data and our construction of the person-
ality measures are described in Section 3. The results are in Section 4. Section 5 concludes
and discusses some potential areas for future research.
4 We calculate the annual interest cost of co-holding per household using the average credit card
debt amount that could have been paid down after a month’s income is set aside and the average
interest rate of 14%. When extrapolating to the US population, we apply the annual interest cost
estimate to 1% of households in the US with a head over 50 (Table H2, U.S. Census Bureau, 2015).
2. Empirical Framework
The decision of how much to consume and save (or borrow) has long been studied, and
often modeled using the neo-classical expected life-cycle utility maximization framework.
There is also a large literature in economics and finance examining asset allocation across
types of assets and across time. Given that the focus of our article is on the role of personal-
ity traits, and not on the calculation of inter-temporal substitution rates or elasticity meas-
ures, we implement our empirical strategy using a reduced-form examination of the
decision of how much low-yield liquid assets and credit card debt to concurrently hold.

Our approach has two main advantages. First, we require far fewer assumptions by not
estimating a structural model. Second, our examination of the credit card debt puzzle
avoids the need to address the inherent simultaneity in the decision of consumption and
saving that consequently determine asset and debt accumulation. In Section 4.1, however,
we do examine the underlying mechanisms of our findings by studying the relationship
between asset holding and debt utilization.
We focus on a definition of the puzzle group in which holding more than $500 in low-
yield liquid assets, that is, checking, savings, and money market accounts, with positive
revolving credit card debt is considered a puzzle. However, we have also examined alterna-
tive definitions that allow households to have different levels of low-yield liquid assets for
liquidity purposes or different levels of credit card debt, and have found that our results are
robust to using alternative definitions.5 We estimate the probability of being in the puzzle
group using the linear-probability model (OLS). We have also used a logit model, and
obtained very similar qualitative and quantitative results. Our base reduced-form specifica-
tion can be written as:
Yit ¼ b0 þ x0iðt2Þ b þ it ; (1)
where x0iðt2Þ is a vector of the time invariant and (2-year lagged) control variables and we
assume that Eðit jxiðt2Þ Þ ¼ 0. Our empirical strategy exploits the panel nature of our data,
thereby allowing us to address the potential simultaneity inherent in the financial and dem-
ographic measures we examine. For example, a health shock could affect the need for credit
(due to large medical bills), uncertainty in future earnings, and one’s employment (requiring
someone to retire earlier than planned). Our preferred specifications therefore use 2-year
lags of financial measures.
Financial measures such as income and wealth are, of course, crucial for one’s saving
and borrowing decisions as they affect both the need for saving or borrowing and the
returns or costs (as different borrowers would face different interest rates). Personality
measures may cause two households with the same demographic and financial measures to
have a different need for liquid assets and debt. For example, those with higher levels of
Conscientiousness may be able to better interpret and more accurately perceive their
5 In other definitions, we consider households to not be in the puzzle group if they have up to $500 in
credit card debt as in Telyukova (2013); $1,200 or one-half of monthly income, whichever is larger,
in checking or savings accounts (following Bertaut, Haliassos, and Reiter, 2009); and one month’s
income in checking or savings accounts. We have also examined continuous measures that can
be interpreted as the cost of being in the puzzle group such as minflnðAÞ; lnðDÞg; lnðAÞ1fD > 0g,
and lnðAÞ0:5 lnðDÞ0:5 , where A and D are the amounts of low-yield liquid assets and credit card
debt, respectively, in excess of certain thresholds such as 0 or $500.
financial situation. More extravert people may be able to better negotiate and leverage their
financial situation when restructuring their debt with a lender, etc.
To examine the role of personality, we augment the model in Equation (1) by adding
the (5 1) vector pi of the Big Five personality traits. We include these measures additively,
and allow them in some of the specifications to have an interactive effect with another char-
acteristic zit:
Yit ¼ b0 þ x0iðt2Þ b þ p0i c þ zit p0i d þ it : (2)

Equation (2) is useful in demonstrating how noncognitive ability, such as personality, might
affect a household’s financial decision to co-hold low-yield liquid assets and credit card
debt.
Researchers have previously proposed various explanations (see Section 1) for the credit
card puzzle. An advantage of our reduced-form model is that it allows us to succinctly con-
trol for those proposed explanatory factors. We use financial controls (such as income, vari-
ous assets and debts), education levels, and demographic controls (such as age, and marital
status) that have been suggested in the literature as important in determining the decision to
save and borrow. In addition to the very detailed financial data at our disposal, we are also
able to control for other demographic variables that are likely to affect household financials
such as health status (both self-reported, and by controlling for medical expenditures) or
changes in family composition (due to death, marriage, or divorce).
The reduced-form specification examines the overall effect of a household’s characteris-
tics. Therefore, the specification in Equation (2) does not separate out the decision to be
a revolver, and the factors that affect the likelihood of being in the puzzle group
(i.e., become a revolver and hold a low-yield liquid assets balance simultaneously). As such,
our findings potentially encompass several channels or mechanisms at work. We further
examine the decomposition of the effect of personality to understand the relative impor-
tance of the potential mechanisms at play. For example, the overall reduced-form effect of
Conscientiousness might be zero. However, this might be because the underlying effects
nullify each other. Conscientious individuals might be more likely to qualify for or have
access to credit, but at the same time might be less likely to borrow and hold large amounts
of cash at the same time, since they carefully examine their monthly statements, or consider
the cost of debt.
To examine the decomposition, we consider two necessary conditions for being in the
puzzle group through which personality traits may operate. First, a necessary condition to
be included in the puzzle group is to be a revolver. Second, conditional on being a revolver,
one may or may not be in the puzzle group depending on whether one holds low-yield
liquid assets that are not used to pay down debt. We therefore separately examine the pro-
pensity to be a revolver, and the propensity of revolvers to hold low-yield liquid assets and
not pay one’s debt down.6 One could think of the two conditions as being related to two
dimensions: consumption and financial management of the household’s accounts for a
given level of consumption.
For the dimension of incurrence of debt, several personality traits are likely to affect
levels of spending and/or borrowing, and financial terms (such as interest rates and credit
6 This sequential framework is, of course, for exposition purposes only. An alternative would involve
the decision to hold low-yield liquid assets, and conditional on holding those assets, the decision
to incur debt instead of using one’s available assets.
limits) that affect debt levels. Self-control (Laibson, Repetto, and Tobacman, 2003;
Bertaut, Haliassos, and Reiter, 2009), impulse spending (Gathergood and Weber, 2014),
and the propensity to plan (Ameriks, Caplin, and Leahy, 2003) have been shown to be
related to incurrence of debt and wealth accumulation. These traits are all captured by
Conscientiousness, and the effect on debt is likely to be negative. Agreeableness may lead to
higher levels of spending and debt because agreeable people tend to spend more on others,
and might be more susceptible to marketing campaigns.7 A large literature has documented
lower incomes among those with higher levels of Agreeableness (e.g., Judge et al., 1999;
Babcock and Laschever, 2003; Mueller and Plug, 2006). A similar trade-off (less financial

gains in return for less conflict or an increased preference for others’ utility) is likely to play
a role in this instance as well. Extravert people may acquire financial advice from their
peers, or may be able to better negotiate and leverage their financial situation when restruc-
turing their debt with a lender. Those with higher levels of Openness may be more likely to
consume and spend more leading to higher levels of debt.8 The direction of the effect of
Neuroticism is ambiguous. Higher levels of Neuroticism would lower the likelihood of bor-
rowing due to the increased psychological cost of worrying about the future ability to
repay. On the other hand, lower levels of Neuroticism have been found to be associated
with more discretionary savings (e.g., Brandst€atter, 2005, p. 70). For example, Wang, Lu,
and Malhotra (2011) find a negative relationship between revolving credit use and meas-
ures related to low levels of Neuroticism. Donnelly, Iyer, and Howell (2012) show that
Neuroticism is positively related to compulsive buying.
For the dimension of co-holding low-yield liquid assets and credit card debt, we consider
three main channels through which personality may operate: the management of household
finances; liquidity demand; and intra-household dynamics. For the first channel, conscien-
tious individuals are more likely to notice they have sufficient low-yield liquid assets to pay
down their debt. Extraverts may be more likely to discuss their finances and solicit possible
solutions from others on how to pay down their credit card debt.9 The effect of
Neuroticism is a priori ambiguous. Those with higher levels of Neuroticism might be con-
stantly worried about their finances or missing a payment, thereby having a heightened
awareness of their ability to pay down credit card debt. On the other hand, people with
low levels of Neuroticism may make financial decisions in a calm and deliberate manner.10
For the second channel of precautionary saving, conscientious individuals might be
more likely to hold low-yield liquid assets even when they have debt, as they are more likely
to plan ahead. The effect of Neuroticism is a priori ambiguous. More neurotic individuals
7Bernerth et al. (2012) note that “the trusting, submissive, and accommodating tendencies of agreeable
individuals can put them in precarious positions as they sacrifice personal resources for others.”
8 For example, Brown and Taylor (2014) find that Openness is positively correlated with having
credit card debt. Matz, Gladstone, and Stillwell (2016) find that higher levels of Openness are
associated with higher levels of spending on entertainment, eating out, pubs, and tourism.
9 For example, in the context of coping and coping effectiveness under stress or constraints,
McCrae and Costa (1986) find that “Extraversion is correlated with rational action, positive think-
ing, substitution, and restraint.” Related, Carver and Connor-Smith (2010) find that “Extraversion
predicted more problem solving, use of social support, and cognitive restructuring.” See also
Connor-Smith and Flachsbart (2007).
10 For example, Donnelly, Iyer, and Howell (2012) find that Neuroticism is negatively related to the
management of personal finances.
might have higher demand for liquidity because they worry about their uncertain future.
On the other hand, they may worry about being burdened with debt and prefer to pay
down as much of it as they can.
The third channel we consider is intra-household (or dual-self) dynamics. For example,
as suggested by Bertaut, Haliassos, and Reiter (2009), an “accountant” may choose to
maintain high levels of credit utilization to control the spending temptation of their
“shopper” spouse. The personality traits we consider readily translate into those two types.
For example, an “accountant” is likely to have a high level of Conscientiousness, whereas a
“shopper” may have a low level of Conscientiousness and a high level of Agreeableness for

the aforementioned reasons.
Because a household might be a revolver for reasons correlated with the likelihood of
being in the puzzle group, we must find an exclusion restriction that would predict being a
revolver, but would not affect a household’s likelihood of being in the puzzle group. In
Section 4.1, we employ an exclusion restriction strategy and examine whether personality
has a differential effect on checking/savings balance among debt holders and those with no
debt. Our identification strategy is akin to using the 2-year lag of revolving behavior to pre-
dict current revolving behavior.
Finally, our specification also allows us to test whether personality might also interact
with a spouse’s personality, or a proxy for the household’s power structure. Here our iden-
tification strategy uses single households’ co-holding decisions to examine decisions among
couple households, and separate out the contribution of each family member to the overall
household decision.
3. Data and the Big Five Personality Traits

Our data are based on the HRS (2012).11 The HRS is a biennial longitudinal survey that
collects detailed demographic, health, economic, and financial information from a nation-
ally representative sample of the population over age 50.12 The HRS has three main advan-
tages for our setting. First, the longitudinal nature of the data is crucial for our
identification strategy as explained in the previous section. Second, the data have high-
quality personality measures,13 as well as detailed financial information. Third, as
explained in Section 1, personality measures are more likely to be stable among older adults
thereby reducing the threat of validity to our results.14
11 Additional information can be found at http://hrsonline.isr.umich.edu/

12 The co-holding of credit card debt and low-yield liquid assets is prevalent among all age groups.
Telyukova (2013) reports that 27% of households with heads of age 25–64 years co-hold based on
the 2001 Survey of Consumer Finances (SCF). Using 1995 SCF data, Gross and Souleles (2002)
report in Table 6 that among bank card borrowers younger than 35 years, 95% hold positive liquid
assets and 25% hold more than one month’s income in liquid assets.
13 Other commonly used datasets may have a wider range of age groups but either lack any person-
ality measures (e.g., the SCF) or have less complete personality measures (e.g., The National
Longitudinal Survey of Youth).
14 While our preferred specification (column 5 in Table 2) uses all households, the magnitudes of the
personality effects are similar for households with a head 55 years or younger. However, we
acknowledge that our sample cannot be used to estimate the effect on those under 50 years old.
Our predictions for the entire population therefore require us to assume the effects are similar
among younger households.
The HRS contains both respondent-level and household-level data. Because most of the
financial measures are collected at the household level, and financial decisions depend on
and impact the entire household, our primary unit of analysis is at the household level. The
households in the data consist of singles and couples (we also control for the presence of
additional household members). For the households with couples, because our dependent
variables of interest are financial, we focus on the demographics and personality of the per-
son who has answered the survey questions related to household finances. In the HRS data,
this person is identified as the “financial respondent” of the household.15 Although we
model household behavior, we also use respondent-level information from financial

respondents with the assumption that the coordination within a household is not a signifi-
cant factor. However, as an extension, we relax this assumption in Section 4.2 and investi-
gate the effect of spouse characteristics using the data on couple-households.
Our main sample consists of 12,976 households between 2008 and 2012.16 Table 1
reports the summary statistics for the full sample and for the subgroup of households that
are revolvers. The average credit card balance among revolvers is $8,972. On average, 51%
of the households are a single household.
3.1 The Big Five Personality Traits

The Big Five personality traits have been measured in the HRS biennially since 2006. They
are measured as part of a questionnaire which is given to about half of the full sample in every
wave.17 As a consequence, we have personality traits for almost all individuals in either 2006/
2010 or 2008/2012. The HRS uses 26 personality survey items developed originally for the
Midlife in the United States Survey.18 The 26 variables are self-administered adjectival meas-
ures. Participants are asked to “Please indicate how well each of the following DESCRIBES
YOU” for 26 adjectives. Each adjective is coded from 1 (“not at all”) to 4 (“a lot”). The
adjectives are then grouped and averaged to create a score for each of the five traits. For
example, Conscientiousness is constructed from these five items (with “” indicating an
inverse coding): organized, responsible, hardworking, careless (), and thorough.19 The
stability of personality measures over time has been widely studied in the field of developmen-
tal psychology. For example, some studies have emphasized the hereditary and biological fac-
tors that shape traits (e.g., Bouchard Jr and Loehlin, 2001; Canli, 2006; DeYoung et al.,
15 Smith, McArdle, and Willis (2010) find that males and those with more years of education are
more likely to be the financial respondent of the household in the HRS survey. Our analysis con-
trols for both of these factors.
16 The attrition rate in the HRS is relatively low. For example, from 2010 to 2012, the attrition rate is
4.9% due to death, and 3.7% due to nonresponse. We also examined potential attrition bias by
testing whether membership in the puzzle group could predict attrition and find the effect to be
small and not statistically significant.
17 Although the core HRS survey is a biennial survey and participants are interviewed every two
years, some of the questionnaires, including those for personality, are only administrated every
4 years (alternating half of the sample every two years) to reduce the burden on survey partici-
pants (Juster and Suzman, 1995).
18 http://www.midus.wisc.edu/
19 The other measures are Openness (7 items): creative, imaginative, intelligent, curious, broad-
minded, sophisticated, adventurous; Extraversion (5 items): outgoing, friendly, lively, active, talka-
tive; Agreeableness (5 items): helpful, warm, caring, softhearted, sympathetic; and Neuroticism
(4 items): moody, worrying, nervous, calm ().
Table I. Summary statistics
Mean Standard deviation
Main Revolvers Main Revolvers

Variables sample only sample only
Observations 30,517 6,833

Couple household 49% 54% 50% 50%
Household-level variables

Revolver 22% 100% 42% 0%
Credit card debt $2,009 $8,972 $7,638 $14,076
Puzzle group
Revolver and low-yield liquid assets >$500 15% 68% 36% 47%
Assets and debts

Checking and savings $29,089 $11,142 $91,731 $42,253
No checking or savings 20% 16% 40% 36%
Financial assets $128,270 $40,175 $474,272 $158,769
Debts including credit card debt $4,430 $13,159 $31,472 $36,658
Value of business $38,617 $18,843 $345,359 $189,679
IRA balance $58,987 $29,660 $180,903 $100,089
Own home 75% 78% 43% 42%
Real estate $221,746 $198,198 $517,890 $486,374
Mortgages and home equity loans $34,019 $58,234 $85,027 $100,255
Income and medical expense

Income $58,061 $58,352 $355,317 $75,613
Received food stamp 8% 8% 27% 28%
Out-of-pocket medical expense $4,954 $4,786 $10,851 $8,443
Below the poverty line 13% 8% 33% 28%
Financial respondent-level variables

Personality traits (1–4)
Openness 2.92 2.97 0.54 0.52
Conscientiousness 3.35 3.36 0.46 0.45
Extraversion 3.18 3.19 0.53 0.52
Agreeableness 3.51 3.56 0.46 0.43
Neuroticism 2.02 2.06 0.59 0.59
Demographic and other variables

Age 70.41 66.25 10.66 9.09
White 79% 75% 41% 43%
Male 41% 39% 49% 49%
High school 54% 49% 50% 50%
Some post-secondary schooling 23% 29% 42% 45%
College (4 years) or more 22% 22% 42% 41%
Married 46% 50% 50% 50%
Separated/divorced 16% 19% 37% 39%
Widowed 30% 22% 46% 41%
Poor health (excellent (1)poor (5)) 2.90 2.91 1.09 1.06
Employed 32% 46% 47% 50%
Self-employed 8% 9% 26% 28%
Retired 60% 47% 49% 50%
2010). In financial and economic settings, many scholars assume that personality traits are
fixed among adults (e.g., Nyhus and Pons, 2005; Mueller and Plug, 2006; Heineck and
Anger, 2010). In recent years, an emerging view is that personality traits are influenced by
hereditary and biological factors, but can change over time and may be mutable by interven-
tion especially during early childhood. However, after early adulthood, the mean level
changes relatively less and the rank ordering of personality traits in a population becomes
increasingly more consistent (stable) as one ages (Roberts and DelVecchio, 2000).20
For the purpose of our study, the crucial issue is whether the measurement of personal-
ity is endogenous with respect to financial decisions. For example, Roberts, Walton, and

Viechtbauer (2006) find some mean-changes of personality traits over the life cycle.
However, we control for age and only examine adults later in life, so life-cycle patterns
are not a concern for our setting. Cobb-Clark and Schurer (2012) show that the mean
level of the Big Five personality traits is stable over a 4-year period among working-age
adults. Further, they show that intra-individual changes over time are not correlated with
life events in an economically significant way. Cobb-Clark and Schurer (2013) also show
that the changes of the mean level of locus of control, which is the personality trait of
their focus, are mild, and consistent rather than idiosyncratic. They argue that for
working-age adults the changes are economically insignificant.21
Taken together, the results suggest that our identification assumption regarding the exo-
geneity of personality measures is likely to hold. However, given the longitudinal data at
our disposal, we are further able to test the stability and inverse-causality of the Big Five
personality traits. We find that the personality traits are stable in our sample, and we do
not find any evidence of an inverse-causal relationship with our dependent variables of
interest after controlling for the relevant variables.
In our study, we use the personality measures calculated by the average of the personal-
ity traits over all available years for each of the Big Five personality traits. We then stand-
ardize the personality traits measures (Z-score) by subtracting the average and dividing by
the sample standard deviation calculated using the 2010 data.
4. Results
We first examine the reduced-form estimates in Table 2 corresponding to Equation (2)
without the interaction term using the pooled data from 2008 to 2012. In all columns, the
dependent variable is the binary indicator of whether a household is in the puzzle group
(i.e., not paying their credit card balance in full, and having low-yield liquid assets of $500
or more). Throughout, we report the linear-probability model results using the OLS method
with cluster-robust standard errors. For all regressions in this article, we use clusters defined
by the cross product of nine US geographical regions and three rural–urban groups based
on county population sizes (more than million, 250,000–1,000,000, and 250,000 or less).
This allows us to capture regional unobservable correlated shocks that may affect house-
holds in the same local economy.22
20 Borghans et al. (2008) and Roberts, Wood, and Caspi (2008) provide a review of this matter.
21 The average age in our sample is higher, but our results remain qualitatively and quantitatively the
same if we just focus on the younger working-age segment of our sample.
22 However, our results both in terms of the magnitude of the standard errors and statistical signifi-
cance are almost identical when we instead cluster at the household level.
Table II. The effect of personality on being in the puzzle group (reduced form)
Standard errors, in parentheses, are clustered at the region metro type (highly-urban, medium-
size, and rural). Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism are
standardized to have a mean of zero and a variance of one. All specifications control for race, mar-
ital status, employment status, whether self-employed, whether in nursing home, household size
(whether 2 or more), education (high-school, some college, and college or more dummies), and
region and metro type fixed effects. Columns 3–5 also include assets (transportation, housing),
whether underwater, and percent of household members employed. The sample includes house-
holds in 2008–2012 (with lagged measures in 2006–2010).

*Significant at 10%. **Significant at 5%. ***Significant at 1%.
†
Measures lagged by 2 years (previous wave) in columns 4–5.
Linear probability model; Dependent variable: In the Puzzle Group
Explanatory Variables (1) (2) (3) (4) (5)
Openness 0.0076** 0.0053* 0.0051* 0.0050*

(0.0030) (0.0027) (0.0027) (0.0027)
Conscientiousness 0.0090*** 0.0050* 0.0054* 0.0054*
(0.0029) (0.0029) (0.0028) (0.0029)
Extraversion 0.0130*** 0.0112*** 0.0110*** 0.0109***
(0.0033) (0.0028) (0.0027) (0.0027)
Agreeableness 0.0210*** 0.0153*** 0.0163*** 0.0162***
(0.0031) (0.0028) (0.0027) (0.0027)
Neuroticism 0.0012 0.0003 0.0005 0.0003
(0.0030) (0.0026) (0.0028) (0.0028)
Age 0.0104*** 0.0105*** 0.0111*** 0.0114*** 0.0115***

(0.0027) (0.0028) (0.0022) (0.0023) (0.0023)
Age-squared (divided by 100) 0.0100*** 0.0100*** 0.0094*** 0.0096*** 0.0097***
(0.0018) (0.0018) (0.0015) (0.0016) (0.0016)
Is male 0.0247*** 0.0177** 0.0160** 0.0174** 0.0169**
(0.0067) (0.0074) (0.0070) (0.0069) (0.0068)
In poor health 0.0034 0.0024 0.0042** 0.0040** 0.0042**
(0.0021) (0.0020) (0.0018) (0.0018) (0.0018)
Ln(financial assets excluding 0.0266*** 0.0239*** 0.0238***
low-yield liquid assets)† (0.0023) (0.0021) (0.0020)
Ln(retirement assets)† 0.0246*** 0.0238*** 0.0237***
(0.0026) (0.0029) (0.0029)
Is home owner† 0.0287** 0.0333** 0.0330**
(0.0138) (0.0157) (0.0155)
Ln(income)† 0.0096*** 0.0079*** 0.0083***
(0.0019) (0.0015) (0.0015)
Ln(medical spending)† 0.0056*** 0.0059*** 0.0061***
(0.0009) (0.0009) (0.0009)
Below poverty line† 0.0463*** 0.0345*** 0.0329***
(0.0061) (0.0059) (0.0062)
Ln(mortgageþHELOC)† 0.0118*** 0.0107*** 0.0107***
(0.0009) (0.0008) (0.0008)
2-year change in welfare and 0.0183**
food stamps assistance (0.0079)
2-year change in household size 0.0257***
(0.0091)
R2 0.05 0.06 0.11 0.10 0.10
Number of households 12,976 12,976 12,976 12,976 12,976
Observations 30,517 30,517 30,517 30,517 30,517
We control for household size with a dummy indicator for having more than one mem-
ber (including dependents), as well as marital status. For reasons explained in the previous
section, in the case of households with couples, we use the personal measures of the finan-
cial respondent (i.e., age, race, education, and personality traits). However, for couples, our
analysis shows that the personal characteristics of the financial respondent are a sufficient
control, as our results remain very similar if we additionally include some key spousal
measures. In Section 4.2, we further examine the effect of within-couple income and per-
sonality differences.
The first column in Table 2 includes basic demographic controls and employment sta-

tus. The Big Five personality traits Z-scores (standardized to have a mean of zero and stand-
ard deviation equal to one) are added in column 2. Conscientiousness and Extraversion are
shown to have a negative effect (statistically significant at the 1% level) whereas Openness
and Agreeableness increase the likelihood of being in the puzzle group (statistically signifi-
cant at the 5% and 1% level, respectively). Neuroticism is almost never statistically signifi-
cant across the various specifications.
We then include financial measures in column 3, such as financial assets (excluding
checking/savings), housing debt, income, and medical spending. To alleviate concerns of
simultaneity, we lag the financial measures by two years in column 4. We add controls for
changes to a household’s size and food-stamp usage, in column 5.23 The coefficients of the
personality traits remain qualitatively the same. The coefficients for Openness,
Conscientiousness, Extraversion, and Agreeableness are similar in columns 2–5, and are
statistically significant at the 10% level or better throughout.
In summary, the effects of the personality measures remain similar across our different
specifications, even after controlling for a wide range of financial measures. For example,
in column 5 which is our preferred specification, a one standard-deviation increase in
Conscientiousness and Extraversion, all else equal, decreases the propensity to be in the
puzzle group by 0.54 and 1.09 percentage points, respectively.24 For Openness and
Agreeableness, the probability increases by 0.50 and 1.62 percentage points, respectively.
In regards to demographic variables, we find an inverse U-shape effect of education that
is robust across our specifications. This could be due to the fact that those with the lowest
levels of education have less access to credit. The effects of the financial measures have the
expected signs. For example, households with higher financial assets such as stock, bonds,
certificate of deposits, real estate, or IRA accounts are less likely to be in the puzzle group.
We also include dummy indicator variables for zero assets and IRA balances to account for
the non-linearity of these factors.
23 Our results remain the same when we additionally examine shocks to health and employment
status.
24 To illustrate the effect of the personality traits, one can translate a personality effect into
the equivalent effect of a financial variable. An increase of $25,591 in financial assets (or 0.2269
in logs) would decrease the likelihood of being in the puzzle group by 0.54 percentage points
(column 5). This is the same decrease in the likelihood of being in the puzzle group that
would occur if Conscientiousness were to increase by one standard deviation (as the coefficient
on Conscientiousness is 0.0054). Hence, an increase of $25,591 in financial assets has the same
effect on the likelihood of being in the puzzle group as an one-standard-deviation increase in
Conscientiousness.
Generally, we find that household income tends to have a positive effect on the likeli-
hood of being in the puzzle group (significant at the 1% level in Table 2). Similarly, we find
that households with an income below the poverty line are less likely to be in the puzzle
group. These findings are consistent with low-income households having more difficulty in
qualifying for credit cards. Our controls for employment status, health status, etc. in the
reduced-form estimates capture the differential access to credit cards among households.
We also find that households with high mortgage debt and negative home equity are more
likely to be in the puzzle group. This is consistent with those having less access to cheaper
forms of credit (such as home equity loans and mortgages) are more likely to have to resort

to more expensive forms of credit, such as credit cards.
Our measures of personality are multiple-period averages. As a further robustness test,
we examine individual fixed effects specifications where we allow personality to vary
yearly. We find that none of the yearly components of the personality measures are statisti-
cally significant, nor are they jointly significant (p-value of 0.81).25 This finding is consis-
tent with the yearly within-individual variation in the personality measures being
idiosyncratic.
We have also examined Table 2’s specifications using lagged personality traits measured
in the past survey waves instead of our preferred multiple-period averages. This further
reduces the concern of endogeneity of the personality traits, as discussed in Section 3.1. We
find that our results are robust to the use of the lagged Big Five personality measures from
the most recent previous survey wave. The only exception is that in some specifications,
Conscientiousness is no longer significant at the 10% level.
To address the possible concern of reverse causality, that an individual’s personality
measures are influenced by their unobservable propensity to be in the puzzle group, we
have further examined both the stability of our personality traits, and the personality traits’
response to changes in financial measures, by exploiting the longitudinal nature of our
data. Within individuals, the overall change in personality over time is quite small. The
average change is between 0.07 and 0.03, and the sample medians are zero for all traits.
In addition, we find that being in the puzzle group two years earlier does not have a statisti-
cally significant effect on the level of changes for each of the Big Five personality traits. In
summary, controlling for a wide range of demographic, financial, health, and location
measures, we find a persistent effect of personality traits on the likelihood of being in the
puzzle group.
4.1 Decomposing the Effects on Credit Card Debt and Checking / Savings
Balance
The reduced-form results in the previous section examine measures that are functions of
both checking/savings balance and credit card debt, and as such circumvent the need to
address the inherent simultaneity in the decision of allocating assets. However, as explained
in Section 2, an aggregate effect of personality on co-holding may mask two opposing
effects, that is, one on checking/savings and the other, with an opposite sign, on credit card
25 We also examined a random-effects model that would allow for the inclusion of fixed-over-time
personality measures. The results for the personality traits are of the same sign and magnitude of
our preferred specifications, and Extraversion and Agreeableness remain statistically significant
at the 5% level.
debt. Those effects may in turn cancel each other out. Therefore, in this section, we first
examine whether the personality traits are affecting credit card debt, and then study check-
ing/savings balances conditional on credit card debt.
We first examine the likelihood that a household is a revolver, using the entire sample of
households.26 We estimate the determinants of credit card debt revolving using a dichoto-
mous variable for the existence of credit card debt, and the results are shown in Table 3.
Throughout the table, Neuroticism is never statistically significant at conventional levels,
whereas Openness, Conscientiousness, Extraversion, and Agreeableness are almost always
statistically significant at the 5% level or lower. The effect of Conscientiousness and

Extraversion on the likelihood of having credit card debt is negative, and the effect of
Openness and Agreeableness is positive.
Overall, the results in columns 1–5 show that personality traits predict the propensity
to have credit card debt, which is a necessary condition for being in the puzzle group. We
find that the magnitudes of the effects of personality traits are similar across our specifica-
tions. The results remain qualitatively similar even when we control for 2-year lag of
being a revolver (column 6). The effect of Agreeableness remains significant at the 1%
level.
Those with a high level of Conscientiousness are perhaps more likely to avoid credit card
borrowing because they are more likely to be self-controlled, plan ahead, and execute their
plan. This is consistent with the overall findings on the importance of Conscientiousness in
determining health, positive aging, and human capital (see Roberts et al., 2014 for a recent
survey). On the other hand, Agreeableness has a strong positive effect on the likelihood of
holding credit card debt. As discussed in Section 1, Agreeableness may lead to higher levels of
spending. In addition, as discussed in Section 2, several studies have found that Agreeableness
is negatively correlated with income. Similar preferences and mechanisms may be at work in
this case. Those with higher levels of Agreeableness prefer less conflict (with employer in the
case of wages, with self and friends in the case of incurring debt) over financial gains such as
higher wages or less debt.
Next, we examine the effect of personality on checking/savings balance, while control-
ling for credit card debt revolving status. Because revolving is endogenous to the decision of
how much to save, it is crucial to address this potential bias. One common empirical strat-
egy is to use an instrumental variables approach, or more generally, an exclusion restric-
tion. We use 2-year lagged revolving status as our exclusion restriction.
In our setting, we are interested in estimating the differential effect of the Big Five per-
sonality traits among those who are revolvers and those who are not. Given that there are
10 such effects (5 2), we do not directly estimate the model using two-stage least squares.
Instead, we consider a reduced-form regression and include the instruments directly into
the “second-stage” equation (Angrist and Pischke, 2008, p. 213). In other words, we esti-
mate the effect of our instruments, instead of their endogenous counterparts, on the amount
of checking/savings. For our exclusion to be valid, it must hold that 2-year lagged revolving
26 There is a large body of literature on personal debt focusing on both economic and financial fac-
tors, and on psychological factors. For example, in the psychology literature, Livingstone and Lunt
(1992), Wood (1998), Donnelly, Iyer, and Howell (2012), and Wilcox, Block, and Eisenstein (2011)
find effects that by and large are consistent with our findings.
Table III. The effect of personality on having revolving credit card debt
Standard errors, in parentheses, are clustered at the region metro type (highly-urban, medium-
size, and rural). Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism are
standardized to have a mean of zero and a variance of one. All specifications control for race, mar-
ital status, employment status, whether self-employed, whether in nursing home, household size
(whether two or more), education (high-school, some college, and college or more dummies),
and region and metro type fixed effects. Columns 3–6 also include assets (transportation, hous-
ing), whether underwater, and percent of household members employed. The sample includes
households in 2008–2012 (with lagged measures in 2006–2010). Column 6 only includes house-

holds in 2010–2012.

†
Measures lagged by two years (previous wave) in columns 4–6.
Linear probability model; Dependent variable: Have a Revolving Credit Card Balance
Explanatory Variables (1) (2) (3) (4) (5) (6)
Openness 0.0106** 0.0090** 0.0085** 0.0084** 0.0015

(0.0040) (0.0037) (0.0036) (0.0036) (0.0033)
Conscientiousness 0.0169*** 0.0091*** 0.0095*** 0.0095*** 0.0038
(0.0027) (0.0027) (0.0028) (0.0028) (0.0024)
Extraversion 0.0140*** 0.0123*** 0.0121*** 0.0120*** 0.0036
(0.0042) (0.0040) (0.0041) (0.0041) (0.0035)
Agreeableness 0.0298*** 0.0207*** 0.0221*** 0.0220*** 0.0137***
(0.0041) (0.0034) (0.0035) (0.0035) (0.0025)
Neuroticism 0.0048 0.0049 0.005 0.005 0.002
(0.0044) (0.0038) (0.0040) (0.0040) (0.0025)
Age 0.0108*** 0.0111*** 0.0155*** 0.0156*** 0.0157*** 0.0138***

(0.0032) (0.0032) (0.0029) (0.0030) (0.0029) (0.0026)
Age-squared (divided by 100) 0.0121*** 0.0123*** 0.0139*** 0.0140*** 0.0140*** 0.0108***
(0.0022) (0.0022) (0.0020) (0.0021) (0.0020) (0.0018)
Is male 0.0382*** 0.0274*** 0.0234*** 0.0256*** 0.0242*** 0.0174**
(0.0069) (0.0078) (0.0074) (0.0076) (0.0074) (0.0064)
In poor health 0.0187*** 0.0158*** 0.0113*** 0.0122*** 0.0117*** 0.0080**
(0.0031) (0.0032) (0.0028) (0.0028) (0.0028) (0.0031)
Ln(financial assets excluding 0.0289*** 0.0272*** 0.0273*** 0.0128***
liquid assets)† (0.0022) (0.0021) (0.0021) (0.0014)
Ln(income)† 0.0056** 0.0032 0.0037 0.0001
(0.0026) (0.0023) (0.0023) (0.0021)
Ln(medical spending)† 0.0079*** 0.0085*** 0.0092*** 0.0046***
(0.0013) (0.0012) (0.0012) (0.0011)
Below poverty line† 0.0659*** 0.0542*** 0.0575*** 0.0286***
(0.0111) (0.0119) (0.0122) (0.0098)
Ln(mortgage þ HELOC)† 0.0134*** 0.0125*** 0.0125*** 0.0053***
(0.0010) (0.0009) (0.0009) (0.0006)
2-year change in welfare and 0.0398*** 0.0394***
food stamps assistance (0.0119) (0.0094)
2-year change in household size 0.0353*** 0.0298**
(0.0098) (0.0122)
Previously a revolver 0.4663***
(0.0101)
R2 0.06 0.07 0.13 0.12 0.12 0.32
Observations 30,517 30,517 30,517 30,517 30,517 20,602
Table IV. The effect of personality on having a checking/savings account balance
Standard errors, in parentheses, are clustered at the region metro type (highly-urban,
medium-size, and rural). Openness, Conscientiousness, Extraversion, Agreeableness, and
Neuroticism are standardized to have a mean of zero and a variance of one. All specifications
control for all variables in column 5 of Table 2, including region and metro type fixed effects.
The sample includes households in 2008–2012 (with lagged measures in 2006–2010).
Linear probability model; Dependent variable: Have a Checking/Savings Account Balance Over $500

Explanatory Variables (1) (2) (3)
Openness 0.0097*** 0.0098*** 0.0096**

(0.0035) (0.0035) (0.0046)
Conscientiousness 0.0228*** 0.0228*** 0.0213***
(0.0048) (0.0048) (0.0054)
Extraversion 0.0216*** 0.0216*** 0.0207***
(0.0034) (0.0034) (0.0035)
Agreeableness 0.0051 0.0051 0.0072**
(0.0044) (0.0043) (0.0034)
Neuroticism 0.0164*** 0.0165*** 0.0148***
(0.0032) (0.0032) (0.0031)
Previously a revolver (2 years earlier) 0.0025 0.0031

(0.0083) (0.0082)
Lagged revolving status (revolved a credit card balance two years earlier)
interacted with personality traits
Was a revolver Openness 0.0009

(0.0105)
Was a revolver Conscientiousness 0.006
(0.0108)
Was a revolver Extraversion 0.0038
(0.0078)
Was a revolver Agreeableness 0.0088
(0.0088)
Was a revolver Neuroticism 0.0065
(0.0065)
R2 0.35 0.35 0.35
Observations 20,602 20,602 20,602
behavior, conditional on our other measures, is not correlated with the unobservable pro-
pensity to have a checking/savings balance 2 years later.27
27 To investigate the validity of the exclusion, we examined the typical measures associated with a
two-stage-least-squares framework using the specification in column 2 of Table 4. The first stage
R-squared is 0.32. The excluded instrument’s F-statistic (2126.95) suggests that there is no “weak”
instrument problem (Stock and Yogo, 2005). Furthermore, we could not reject the null that the
instruments are valid using the Sargan–Hansen test of overidentification (with 4-year lag revolving
status being the additional instrument).
The results in column 1 of Table 4 demonstrate that the Big Five personality traits have
a statistically significant effect on having a checking and savings balances of $500 or more
both for those who are revolvers and those who are not. Both for revolvers and non-
revolvers, Openness, Conscientiousness, Extraversion, and Neuroticism have a statistically
significant effect on checking and savings balance. The results remain very similar when we
add the lagged revolving status in column 2. In column 3, we add the interaction terms
between the personality traits and the lagged revolving status. Our results suggest that the
effect of personality on having low-yield liquid assets is similar between revolvers and non-
revolvers.

Taken together, the results in Tables 3–4 suggest that personality measures are impor-
tant predictors both for credit card debt and for holding low-yield liquid assets. We find
that Conscientiousness is important for both being a revolver and having a checking/savings
balance (but with opposite signs), whereas the effect of Agreeableness operates mainly on
the likelihood of being a revolver.
4.2 Intra-Household Dynamics

The effect of household dynamics on economic and financial outcomes has been widely
studied.28 These studies indicate that the financial behavior of a household could depend
on the inner dynamics within the household. In the context of credit card debt, Bertaut,
Haliassos, and Reiter (2009) derive a model where high credit card debt is used as a means
to curb the spending of a spouse (or one’s self) by committing to have less available funds
through a high level of existing debt. This is conceptually similar to the model considered in
Fudenberg and Levine (2006) who propose a dual-self model for impulse control.
In this section, we broaden our focus beyond that of the financial respondent’s personal-
ity. In our setting, the intra-household dynamics between the financial respondent and their
spouse may play an important role as each side may not be able to fully monitor or control
the spending and saving behavior of their partner. For example, each partner may have
their own credit card that can be used for spending without the others’ pre-approval.29
Personality differences (similarities) may exacerbate (alleviate) this dynamic. We examine
both a reduced-form specification, where household members are treated symmetrically
(columns 3 and 4), and specifications with some additional structure (columns 5 and 6)
where we proxy for power imbalance within a household using the income difference
(financial respondent’s income spouse’s income). Finally, we examine the role of the
financial respondent’s gender (column 7).
Column 1 of Table 5 demonstrates that the effects of the financial respondent personal-
ity traits remain largely the same when only couple households are examined. Next, to
allow for a more parsimonious examination, we define a “puzzle personality index” by esti-
mating the propensity to be in the puzzle group among single households.30 By examining
single households, we circumvent the added complication of intra-household dynamics.
28 For example, see Lundberg and Pollak (2007) for a recent review of some of the issues in the US
context.
29 Schaner (2015) argues that large differences in discount factors among couples can lead to hold-
ing individual bank accounts that have lower interest earnings than joint accounts.
30 To improve statistical power, we summarize the five personality traits with a single index that has
the most explanatory power. Otherwise, to examine both partners’ personality and an interaction
term would require as many as 20 variables.
Table V. The effect of intra-household dynamics on being in the puzzle group
Standard errors, in parentheses, are clustered at the region metro type (highly-urban, medium-size, and rural). Openness, Conscientiousness, Extraversion,
Agreeableness, and Neuroticism are standardized to have a mean of zero and a variance of one. All specifications control for all variables in column 5 of Table 2,
including region and metro type fixed effects. Age and education (years) differences measured as the absolute value of (financial respondent’sspouse’s). The
PPI aggregates the effect of the Big Five personality traits based on single household estimates. Couple power difference based on income difference between
couple members. The sample includes all couple households in 2008–2012 (with lagged measures in 2006–2010).
Explanatory Variables (1) (2) (3) (4) (5) (6) (7)
Puzzle personality (index) 0.0148*** 0.0141*** 0.0136*** 0.0152***

(0.0039) (0.0039) (0.0040) (0.0040)
Spouse’s puzzle personality (index) 0.0064* 0.0049 0.005 0.0056
Credit Card Debt and Noncognitive Ability
(0.0034) (0.0035) (0.0035) (0.0034)

Interaction of couple’s puzzle personalities 0.0086*** 0.0080*** 0.0080***
indices (0.0026) (0.0025) (0.0026)
Couple’s age difference 0.0001 0.0001 0.0002

(0.0006) (0.0006) (0.0006)
Couple’s years-of-schooling difference 0.0004 0.0005 0.0007
(0.0017) (0.0017) (0.0017)
Couple’s power difference (income) 0.0184* 0.0196*
(0.0106) (0.0105)
Openness 0.0127 0.0117

Couple’s power difference (income) (0.0125) (0.0122)
Conscientiousness 0.0082 0.0105
Extraversion 0.0071 0.0073
2127
(continued)

Table V. Continued
2128
Explanatory Variables (1) (2) (3) (4) (5) (6) (7)
Agreeableness 0.0177** 0.0190**

Neuroticism 0.0099 0.0112
Male’s puzzle personality (index) 0.0025

(0.0052)
Male’s puzzle personality (index) 0.0088
Financial respondent is male (0.0067)
Female’s puzzle personality (index) 0.0192**
(0.0072)
Female’s puzzle personality (index) 0.0096
Financial respondent is male (0.0079)
Openness 0.0061 0.006

(0.0048) (0.0048)
Conscientiousness 0.0071 0.0085
(0.0048) (0.0053)
Extraversion 0.0126** 0.0131**
(0.0051) (0.0053)
Agreeableness 0.0186*** 0.0195***
(0.0051) (0.0051)
Neuroticism 0.0018 0.0025
(0.0043) (0.0039)
R2 0.11 0.11 0.11 0.11 0.11 0.11 0.11
Observations 12,988 12,988 12,988 12,988 12,988 12,988 12,988
H.-s. Choi and R. A. Laschever

We first estimate the specification in column 5 of Table 2 with only single households.
Using the estimated coefficients for the personality traits from the single household esti-
mates, we create a puzzle personality index (PPI) as the Z-score of 0.007 Openness
0.003 Conscientiousness 0.009 Extraversion þ 0.012 Agreeableness 0.0006
Neuroticism. Column 2 of Table 5 includes the PPI only for the financial respondent. The
coefficient is significant at the 1% level, suggesting the index performs well in capturing the
likelihood of being in the puzzle group for the couple households. Next, we allow the per-
sonality of both members of the household to have an effect (column 3). The coefficient for
the financial respondent’s PPI remains statistically significant and of the same magnitude.

The spouse’s PPI is also statistically significant suggesting that both household members’
personality traits contribute to the likelihood of being in the puzzle group.31 In columns
4–6, we add an interaction effect between the couple’s PPIs. The interaction coefficient is
negative and statistically significant at the 1% level in columns 4–6, implying an attenuat-
ing effect of a spouse’s PPI on the marginal effect of the PPI of the financial respondent.
The specification in columns 5 and 6 of Table 5 examines the interaction effect of a
measure of power imbalance between a couple and the financial respondent’s personality
traits as well as intra-household differences (in absolute values) of age and schooling lev-
els.32 A positive value of this measure (the financial respondent earns more than their
spouse) could proxy for more power for the financial respondent within the relationship. In
column 5 we use the PPI, and in column 6 we instead use the Big Five personality traits for
the financial respondent and obtain similar results. In both cases, the likelihood of being in
the puzzle group increases when the nonfinancial respondent spouse earns more than the
financial respondent. Also, the interaction term between the financial respondent’s
Agreeableness and the income difference is negative and statistically significant at the 5%
level. The more disagreeable the financial respondent is, the more likely it is that the finan-
cial respondent would be able to impose his or her preference when facing a powerful
spouse. Our finding is consistent with high levels of disagreeableness (i.e., less willing to
accommodate or compromise) being a substitute for power.
In the last specification in Table 5 (column 7), we examine the role of the financial
respondent’s gender in addition to their spouse’s role. We control for the male’s PPI, the
female’s PPI, the financial respondent’s gender, and interactions between the male and
female’s PPI and the gender of the financial respondent. Our results suggest that the effect
of personality is larger for women, and that the effect of personality is the same regardless
of the financial respondent’s gender.
In contrast to the effects of personality, age and schooling differences among couple mem-
bers do not have a statistically significant effect on a household’s likelihood of being in the puz-
zle group, and the coefficients are small in magnitude. Overall, our analysis suggests that co-
31 This result is consistent with that of Schaner (2015), where the difference in the preferences of
household members leads to costly separate individual financial accounts rather than joint
accounts.
pffiffiffiffiffiffiffiffiffiffiffiffi
32 For the power imbalance measure, we use logðz þ 1 þ z 2 Þ, where z is the income difference
(financial respondent’sspouse’s). Our measure encompasses both unearned income (see
Lundberg and Pollak, 1996 for a discussion) and earned income. For example, Basu (2006) exam-
ines the effect of intra-couple power relationships, measured by income disparity, on household
decision making, and Ashraf (2009) uses an experimental design in the Philippines to examine the
role of spousal control in consumption and savings decisions.
holding behavior is influenced by the dynamics within a household. This highlights that poten-
tial policy interventions and future research should consider both members of a household.
4.3 Alternative Explanations

To examine whether our findings of the effect of personality measures using the Big Five
traits are just proxies for other measures, we examine the effects of other alternative meas-
ures, such as financial sophistication or self-control.
The first column in Table 6 adds a self-control/impulsiveness measure to our main speci-
fication in Table 2 as a proxy for self-control which Bertaut, Haliassos, and Reiter (2009)

and Gathergood (2012) suggested as an important factor for the puzzle. We find that con-
trolling for self-control/impulsiveness, three of the personality measures remain statistically
significant at the 5% level or lower. Column 2 in Table 6 examines the effect of internal
locus of control. Locus of control is the degree to which people believe in their own ability
to control events that influence them.33 People with a high internal locus of control believe
that outcomes are due to their own behavior, and people with a high external locus of con-
trol believe that events occur for reasons out of their control, or external factors. The coeffi-
cient for internal locus of control is statistically significant at the 5% level. However, four
of the personality traits remain statistically significant at the 10% level. In column 3, we
include cognitive functioning measures such as memory and mental status (which includes
numeracy).34 We also include the number series score which was developed for measuring
fluid intelligence. Though some of the cognitive measures have a statistically significant
effect, by-and-large, the effect of personality (both in magnitude and statistical significance)
remains the same or larger and our results are robust to various measures of cognitive
ability.
In column 4, we include a risk aversion measure, as those with higher risk aversion may
opt to keep a larger buffer in the form of less credit card debt or a higher checking account
balance. We use the 6-categories risk aversion measure from the HRS’s “income gamble”
scenario questions. None of the risk categories are statistically significant at the 10% level,
and overall the effect of the personality measures remains the same.
In column 5, we use the self-perception of how much a household has control over their
financial situation.35 We find that households who assess themselves as having more con-
trol over their finances are less likely to be in the puzzle group. To reduce the likelihood of
reverse causality, we use the lagged self-perception of financial control (though financial
problems may be persistent over longer horizons). We find that the effects of the Big Five
traits are quite similar to those in our base specifications.
Columns 6–7 of Table 6 include various measures of financial literacy (see Lusardi and
Mitchell, 2014) for a summary of the important role financial literacy has on economic
33 Locus of control has been shown to be an important factor in economic decision making. For
example, college attendance (Coleman and DeLeire, 2003); job search (Caliendo, Cobb-Clark, and
Uhlendorff, 2015); and loan delinquency (Kuhnen and Melzer, 2017).
34 For memory, we use the total of the immediate and delayed (5 min) word recall scores. For mental
status, we use the total of the serial 7’s test, counting backwards from 20 or 86, date recall, recall-
ing object names, and naming the President/Vice President.
35 We use the question: “Using a 0 to 10 scale where 0 means ‘no control at all’ and 10 means ‘very
much control,’ how would you rate the amount of control you have over your financial situation
these days?”
Table VI. Robustness checks with alternative explanations for the puzzle
Standard errors, in parentheses, are clustered at the region metro type (highly-urban, medium-size, and rural). Openness, Conscientiousness, Extraversion,
Agreeableness, and Neuroticism are standardized to have a mean of zero and a variance of one. All specifications control for all variables in column 5 of Table
2, including region and metro type fixed effects. In column 8, the excluded category is 7–14% interest rate.
Explanatory Variables (1) (2) (3) (4) (5) (6) (7) (8)
Openness 0.0076** 0.0058* 0.0140*** 0.0067 0.0082** 0.0151 0.0118 0.0106

(0.0031) (0.0028) (0.0046) (0.0096) (0.0034) (0.0229) (0.0206) (0.0264)
Conscientiousness 0.0021 0.0052* 0.0088** 0.0212** 0.0032 0.011 0.0065 0.0082
(0.0033) (0.0029) (0.0040) (0.0100) (0.0038) (0.0179) (0.0146) (0.0238)
Credit Card Debt and Noncognitive Ability
Extraversion 0.0122*** 0.0102*** 0.0124*** 0.0113 0.0124*** 0.0417 0.0498 0.045

(0.0031) (0.0028) (0.0036) (0.0099) (0.0034) (0.0366) (0.0310) (0.0473)
Agreeableness 0.0164*** 0.0162*** 0.0166*** 0.0302*** 0.0146*** 0.0428** 0.0461** 0.0621**
(0.0032) (0.0028) (0.0042) (0.0071) (0.0042) (0.0196) (0.0186) (0.0254)
Neuroticism 0.0011 0.0013 0.0028 0.0056 0.0036 0.0091 0.0018 0.0053
(0.0026) (0.0028) (0.0034) (0.0061) (0.0034) (0.0188) (0.0168) (0.0226)
Self-control/Impulsiveness 0.0068**
(0.0031)
Internal locus of control 0.0050**
(0.0022)
Total memory score (word recall) 0.0042***
(0.0011)
Total mental status score (numeracy) 0.0067***
(0.0019)
(continued)
2131

2132
Table VI. Continued

Explanatory Variables (1) (2) (3) (4) (5) (6) (7) (8)
Total number series score 0.0001

(fluid intelligence) (0.0001)
Risk aversion (6 levels) Included
Are any of the risk aversion levels No
statistically significant?
In financial control (lagged) 0.0028**
(0.0011)
Financial literacy 0.0144
(0.0461)
Credit card rates
0% interest rate 0.2315***
(0.0584)
1-6% interest rate 0.1609
(0.0998)
More than 15% interest rate 0.1174***
(0.0360)
Economic/financial understanding Included
(7 levels)
Are any of the of econ/financial Only the highest
understanding levels statistically level
significant?
R2 0.11 0.10 0.09 0.10 0.10 0.18 0.18 0.24
Observations 24,820 30,412 11,330 3,366 14,102 647 784 484
H.-s. Choi and R. A. Laschever

outcomes). We construct the financial literacy measure based on two questions (one on
compound interest and one on the effects of inflation). Column 6 in Table 6 uses a dummy
indicator for answering both questions correctly as the measure of financial literacy.
Column 7 uses a self-assessed measure of the degree of understanding of economics and
finance (7 levels). The results in columns 6–7 show that the effect of Agreeableness remains
statistically significant and the coefficients of the Big Five personality traits generally have
the same sign as those in our base specification in Table 2 even after controlling for finan-
cial literacy. This is consistent with the result of Gathergood and Weber (2014).
Our final specification controls for credit card interest rates on the card used most often.

Column 8 in Table 6 shows that interest rates have a statistically significant effect. Yet, the
magnitudes of the Big Five personality trait effects remain largely the same as those in col-
umn 6–8. Openness, Conscientiousness, and Extraversion lose their significance at the 10%
level, but Agreeableness remains statistically significant. While the sample size for this spec-
ification is rather small (only 1 in 20 of our households in 2010 were asked the question),
the results suggest that our findings capture “true” interest-paying revolvers and are not
solely driven by households that are taking advantage of 0% balance transfer offers.
In summary, we find that the personality effects are robust to controlling for self-
control, locus of control, cognitive abilities, risk aversion, financial literacy, and credit card
interest rates.
5. Conclusion
Using a rich longitudinal data set, we find that controlling for a host of demographic, finan-
cial, and economic factors, personality traits play a role in explaining the credit card puzzle.
We find that Conscientiousness, Extraversion, and Agreeableness have statistically signifi-
cant effects, and that the signs of the effects are consistent with findings in other domains.
The results complement other types of explanations suggested in the literature for the credit
card puzzle, as they hold after controlling for other suggested factors. Our findings that a
broad set of noncognitive measures are important, suggest that researchers should be cau-
tious of focusing on a single aspect or dimension of noncognitive ability when studying
financial decision making.
Our results also suggest that intra-household dynamics might play an important role in
financial decisions, and highlight the importance of both coordination and power dynamics
among household members. The effect of intra-household interactions on the economic
and financial well-being of families remains an important area for future research.
The findings in this article contribute to an emerging and growing set of economic and
financial outcomes (e.g., earnings, education, etc.) in which noncognitive skills play an
important role. It is, therefore, plausible that noncognitive skills would play a role in areas
of finance that have yet to be examined.
There are two types of polices to consider that could yield substantial returns: invest-
ments in noncognitive skills, and policies that target the noncognitive aspects of financial
decision-making. Investments in noncognitive skills, such as planning, may have large bene-
fits.36 However, personality is mostly/solely malleable at early childhood, so such
36 Some schools, such as NY City’s KIPP charter school, teach skills, such as grit, as part of their
curriculum. See Kautz et al. (2014) for a review of early intervention programs targeting noncogni-
tive skills.
investments require an early intervention and a long-term horizon. The second type of poli-
cies is related to the design of interventions or marketing campaigns to address debt. Both
the government and non-profit organizations have programs to assist people with managing
their finances.37 Participants in these programs could be asked to answer a short survey
that would assess their personality type, and could then receive an intervention that would
be tailored to their profile. Financial planners already ask their clients about their invest-
ment goals and risk tolerance. Similar assessments could be performed by debt counselors
or by consumers visiting websites.38 In addition, banks and credit unions routinely have
access to credit reports. They are, therefore, potentially positioned to combine information

on the availability of low-yield liquid assets and revolving credit card balances and alert
their clients to costly levels of co-holding.
Co-holding credit card debt and low-yield liquid assets exerts a non-negligible toll on
households. Extrapolating our results, in the US, even a 1-percentage-point decrease in the
fraction of households in the puzzle group, would generate interest payment savings of over
half-a-billion dollars per year, while maintaining the same level of consumption.
References
Almlund, M., Angela Lee, D., James, H., and Tim, K. (2011) Personality psychology and econom-
ics, in: Eric A Hanushek, Stephen J Machin, and Ludger Woessmann (eds.), Handbook of the
Economics of Education, Vol. 4, chap. 1, Elsevier B.V., North-Holland, pp. 1–181.
Ameriks, J., Caplin, A., and Leahy, J. (2003) Wealth accumulation and the propensity to plan,
Quarterly Journal of Economics 118, 1007–1047.
Angrist, J. D. and Pischke, J.-S. (2008) Mostly Harmless Econometrics: An Empiricist’s
Companion, Princeton University Press, Princeton, NJ.
Ashraf, N. (2009) Spousal control and intra-household decision making: an experimental study in
the Philippines, American Economic Review 99, 1245–1277.
Babcock, L. and Laschever S. (2003) Women Don’t Ask: Negotiation and The Gender Divide,
Princeton University Press, Princeton, NJ.
Barberis, N. and Thaler R. (2003) A survey of behavioral finance, in: G. M. Constantinides, M.
Harris, and R. Stulz (eds.), Handbook of the Economics of Finance, Vol. 1, chap. 18, Elsevier,
North-Holland, pp. 1053–1128.
Basu, K. (2006) Gender and say: a model of household behaviour with endogenously determined
balance of power, Economic Journal 116, 558–580.
Becker, T. A. and Shabani, R. (2010) Outstanding debt and the household portfolio, Review of
Financial Studies 23, 2900–2934.
Bernerth, J. B., Taylor, S. G., Jack Walker, H., and Whitman, D. S. (2012) An empirical investiga-
tion of dispositional antecedents and performance-related outcomes of credit scores, Journal of
Applied Psychology 97, 469–478.
Bertaut, C. C., Haliassos, M., and Reiter, M. (2009) Credit card debt puzzles and debt revolvers
for self control, Review of Finance 13, 657–692.
Borghans, L., Lee Duckworth, A., Heckman, J. J., and Ter Weel, B. (2008) The economics and
psychology of personality traits, Journal of Human Resources 43, 972–1059.
37 For example, the Consumer Financial Protection Bureau launched in 2015 a financial coaching ini-
tiative with 60 certified coaches all across the US (Consumer Financial Protection Bureau, 2015).
38 The National Foundation for Credit Counseling, the oldest and foremost nonprofit organization for
financial counseling, now accepts credit counseling requests online. See https://www.nfcc.org/
agency-locator/online-counseling/
Bouchard, T. J., Jr. and Loehlin, J. C. (2001) Genes, evolution, and personality, Behavior Genetics
31, 243–273.
Bowles, S., Gintis, H., and Osborne, M. (2001) Incentive-enhancing preferences: personality,
behavior, and earnings, American Economic Review 91, 155–158.
Brandst€atter, H. (2005) The personality roots of saving – uncovered from German and Dutch sur-
veys, in: Klaus Günter Grunert and John Thogersen (eds), Consumers, Policy and the
€
Environment A Tribute to Folke Olander, chap. 4, Springer, New York, NY, pp. 65–87.
Bricker, J., Kennickell, A. B., Moore, K. B., and Sabelhaus, J. (2012) Changes in US family finances
from 2007 to 2010: evidence from the Survey of Consumer Finances, Federal Reserve Bulletin

98, 65–68.
Brown, S. and Taylor, K. (2014) Household finances and the ‘Big Five’ personality traits, Journal
of Economic Psychology 45, 197–212.
Caliendo, M., Cobb-Clark, D., and Uhlendorff, A. (2015) Locus of control and job search strat-
egies, Review of Economics and Statistics 97, 88–103.
Canli, T. (2006) Biology of Personality and Individual Differences, Guilford Press, New York,
NY.
Carver, C. S. and Connor-Smith, J. (2010) Personality and coping, Annual Review of Psychology
61, 679–704.
Cobb-Clark, D. A. and Schurer, S. (2012) The stability of Big-Five personality traits, Economics
Letters 115, 11–15.
Cobb-Clark, D. A. and Schurer, S. (2013) Two economists’ musings on the stability of locus of
control, Economic Journal 123, F358–F400.
Coleman, M. and DeLeire, T. (2003) An economic model of locus of control and the human capi-
tal investment decision, Journal of Human Resources 38, 701–721.
Connor-Smith, J. K. and Flachsbart, C. (2007) Relations between personality and coping: a meta-
analysis, Journal of Personality and Social Psychology 93, 1080–1107.
Consumer Financial Protection Bureau (2015) Remarks of Richard Cordray. Aceessed May 20,
2015. https://www.consumerfinance.gov/about-us/newsroom/prepared-remarks-of-cfpb-direc
tor-richard-cordray-at-the-financial-coaching-launch/
Conti, G. and Heckman, J. J. (2014) Understanding Conscientiousness across the life course: an
economic perspective, Developmental Psychology 50, 1451–1459.
DeYoung, C. G., Hirsh, J. B., Shane, M. S., Papademetris, X., Rajeevan, N., and Gray, J. R.
(2010) Testing predictions from personality neuroscience brain structure and the Big Five,
Psychological Science 21, 820–828.
Donnelly, G., Iyer, R., and Howell, R. T. (2012) The Big Five personality traits, material values,
and financial well-being of self-described money managers, Journal of Economic Psychology 33,
1129–1142.
Fudenberg, D. and Levine, D. K. (2006) A dual-self model of impulse control, American
Economic Review 96, 1449–1476.
Fulford, S. L. (2015) How important is variability in consumer credit limits? Journal of Monetary
Economics 72, 42–63.
Gathergood, J. (2012) Self-control, financial literacy and consumer over-indebtedness, Journal of
Economic Psychology 33, 590–602.
Gathergood, J. and Weber, J. (2014) Self-control, financial literacy & the co-holding puzzle,
Journal of Economic Behavior & Organization 107, 455–469.
Gross, D. B. and Souleles, N. S. (2002) Do liquidity constraints and interest rates matter for con-
sumer behavior? Evidence from credit card data, Quarterly Journal of Economics 117,
149–185.
Heckman, J. J. and Kautz, T. (2012) Hard evidence on soft skills, Labour Economics 19,
451–464.
Heckman, J. J., Stixrud, J., and Urzua, S. (2006) The effects of cognitive and noncognitive abilities
on labor market outcomes and social behavior, Journal of Labor Economics 24, 411–482.
Heineck, G. and Anger, S. (2010) The returns to cognitive abilities and personality traits in
Germany, Labour Economics 17, 535–546.
HRS (2012) Health And Retirement Study, 2006-2012 Public Use Datasets, Produced and distrib-
uted by the University of Michigan with funding from the National Institute on Aging (grant
number NIA U01AG009740), Ann Arbor, MI.
John, O. P., Naumann, L. P., and Soto, L. P. (2008) Paradigm shift to the integrative Big Five trait
taxonomy: history, measurement, and conceptual issues, in: Oliver P. John, Richard W. Robins,

and Lawrence A. Pervin (eds.), Handbook of Personality: Theory and Research, 3rd edn,
Guilford Press, New York, NY, pp. 114–158.
John, O. P. and Srivastava S. (1999) The Big Five trait taxonomy: history, measurement, and theo-
retical perspectives, in: L. A. Pervin, and O. P. John (eds.), Handbook of Personality: Theory
and Research, chap. 4, 2nd edn, Guilford, New York, pp. 102–138.
Judge, T. A., Higgins, C. A., Thoresen, C. J., and Barrick, M. R. (1999) The Big Five personality traits,
general mental ability, and career success across the life span, Personnel Psychology 52, 621–652.
Juster, F. T., and Suzman, R. (1995) An overview of the Health and Retirement Study, Journal of
Human Resources 30, S7–S56.
Kautz, T., Heckman J. J., Diris R., Weel B. T., and Borghans L. (2014) Fostering and measuring
skills: improving cognitive and non-cognitive skills to promote lifetime success. Working Paper
No. 20749, National Bureau of Economic Research.
Kuhnen, C. M. and Melzer, B. (2017) Non-cognitive abilities and financial delinquency: the role
of self-efficacy in avoiding financial distress. Working Paper No. 23028, National Bureau of
Economic Research.
Laibson, D., Repetto, A., and Tobacman, J. (2003) A debt puzzle, in: Philippe Aghion, Roman
Frydman, Joseph Stiglitz, and Michael Woodford (eds.), Knowledge, Information, and
Expectations in Modern Economics: In Honor of Edmund S. Phelps, Princeton University Press,
Princeton, NJ, pp. 228–266.
Lehnert, A. and Maki D. M. (2002) Consumption, debt and portfolio choice: testing the effect of
bankruptcy law. Finance and Economics Discussion Series 2002-14, Board of Governors of the
Federal Reserve System (US).
Livingstone, S. M. and Lunt, P. K. (1992) Predicting personal debt and debt repayment: psycho-
logical, social and economic determinants, Journal of Economic Psychology 13, 111–134.
Lundberg, S. (2013) The college type: personality and educational inequality, Journal of Labor
Economics 31, 421–441.
Lundberg, S. and Pollak, R. A. (1996) Bargaining and distribution in marriage, Journal of
Economic Perspectives 10, 139–158.
Lundberg, S., and Pollak, R. A. (2007) The American family and family economics, Journal of
Economic Perspectives 21, 3–26.
Lusardi, A. and Mitchell, O. S. (2014) The economic importance of financial literacy: theory and
evidence, Journal of Economic Literature 52, 5–44.
Matz, S. C., Gladstone, J. J., and Stillwell, D. J. (2016) Money buys happiness when spending fits
our personality, Psychological Science 27, 715–725.
McCrae, R. R. and Costa, P. T. (1986) Personality, coping, and coping effectiveness in an adult
sample, Journal of Personality 54, 385–404.
McCrae, R. R. and Costa, P. T. (1987) Validation of the five-factor model of personality across
instruments and observers, Journal of Personality and Social Psychology 52, 81–90.
McCrae, R. R., and Costa P. T. (1999) A five-factor theory of personality, in: Lawrence A. Pervin
and Oliver P. John (eds.), Handbook of Personality: Theory and Research, 2nd edn, Guilford
Press, New York, NY, pp. 139–153.
Mueller, G. and Plug, E. (2006) Estimating the effect of personality on male and female earnings,
Industrial and Labor Relations Review 60, 3–22.
Norvilitis, J. M., Merwin, M. M., Osberg, T. M., Roehling, P. V., Young, P., and Kamas, M. M.
(2006) Personality factors, money attitudes, financial knowledge, and credit-card debt in college
students, Journal of Applied Social Psychology 36, 1395–1413.
Nyhus, E. K. and Pons, E. (2005) The effects of personality on earnings, Journal of Economic
Psychology 26, 363–384.
Nyhus, E. K. and Webley, P. (2001) The role of personality in household saving and borrowing
behaviour, European Journal of Personality 15, S85–S103.

Rabin, M. (1998) Psychology and economics, Journal of Economic Literature 36, 11–46.
Rabinovich, A. and Webley, P. (2007) Filling the gap between planning and doing: psychological
factors involved in the successful implementation of saving intention, Journal of Economic
Roberts, B. W. and DelVecchio, W. F. (2000) The rank-order consistency of personality traits
from childhood to old age: a quantitative review of longitudinal studies, Psychological Bulletin
126, 3–25.
Roberts, B. W., Lejuez, C., Krueger, R. F., Richards, J. M., and Hill, P. L. (2014) What is conscien-
tiousness and how can it be assessed? Developmental Psychology 50, 1315–1330.
Roberts, B. W., Walton, K. E., and Viechtbauer, W. (2006) Patterns of mean-level change in per-
sonality traits across the life course: a meta-analysis of longitudinal studies, Psychological
Bulletin 132, 1–25.
Roberts, B. W., Wood, D., and Caspi, A. (2008) The development of personality traits in adult-
hood, in: Oliver P. John, Richard W. Robins, and Lawrence A. Pervin (eds.), Handbook of
Personality: Theory and Research, 3rd edn, Guilford Press, New York, NY, pp. 375–398.
Schaner, S. (2015) Do opposites detract? Intrahousehold preference heterogeneity and inefficient
strategic savings, American Economic Journal: Applied Economics 7, 135–174.
Simon, H. A. (1955) A behavioral model of rational choice, Quarterly Journal of Economics 69,
99–118.
Smith, J. P., McArdle, J. J., and Willis, R. (2010) Financial decision making and cognition in a
family context, Economic Journal 120, F363–F380.
Stock, J. H. and Yogo M. (2005) Testing for weak instruments in linear IV regression, in: James
H. Stock Donald W. K. Andrews (eds.), Identification and Inference for Econometric Models:
Essays in Honor of Thomas J. Rothenberg, chap. 5, Cambridge University Press, New York,
NY, pp. 80–108.
Telyukova, I. A. (2013) Household need for liquidity and the credit card debt puzzle, Review of
Economic Studies 80, 1148–1177.
U.S. Census Bureau, 2015, Current Population Survey, 2014 annual social and economic supple-
ment, January.
Wang, L., Lu, W., and Malhotra, N. K. (2011) Demographics, attitude, personality and credit
card features correlate with credit card debt: a view from China, Journal of Economic
Wilcox, K., Block, L. G., and Eisenstein, E. M. (2011) Leave home without it? The effects of credit
card debt and available credit on spending, Journal of Marketing Research 48, S78–S90.
Wood, M. (1998) Socio-economic status, delay of gratification, and impulse buying, Journal of
Economic Psychology 19, 295–320.
Zinman, J. (2007) Household borrowing high and lending low under no-arbitrage. Working
paper, Dartmouth College.

Debt Puzzle

Uploaded by

Copyright:

Available Formats

Debt Puzzle

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Debt Puzzle

Uploaded by

Copyright:

Available Formats

Review of Finance, 2018, 2109–2137

The Credit Card Debt Puzzle and

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

JEL classification: D1, D03, D12, D14

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Yit ¼ b0 þ x0iðt2Þ b þ it ; (1)

Yit ¼ b0 þ x0iðt2Þ b þ p0i c þ zit p0i d þ it : (2)

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

3. Data and the Big Five Personality Traits

11 Additional information can be found at http://hrsonline.isr.umich.edu/

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

3.1 The Big Five Personality Traits

Table I. Summary statistics

Mean Standard deviation

Main Revolvers Main Revolvers

Observations 30,517 6,833

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Assets and debts

Income and medical expense

Financial respondent-level variables

Demographic and other variables

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Linear probability model; Dependent variable: In the Puzzle Group

Explanatory Variables (1) (2) (3) (4) (5)

Openness 0.0076** 0.0053* 0.0051* 0.0050*

Age 0.0104*** 0.0105*** 0.0111*** 0.0114*** 0.0115***

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

*Significant at 10%. **Significant at 5%. ***Significant at 1%.

Explanatory Variables (1) (2) (3) (4) (5) (6)

Openness 0.0106** 0.0090** 0.0085** 0.0084** 0.0015

Age 0.0108*** 0.0111*** 0.0155*** 0.0156*** 0.0157*** 0.0138***

Table IV. The effect of personality on having a checking/savings account balance

*Significant at 10%. **Significant at 5%. ***Significant at 1%.

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Openness 0.0097*** 0.0098*** 0.0096**

Previously a revolver (2 years earlier) 0.0025 0.0031

Was a revolver Openness 0.0009

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

4.2 Intra-Household Dynamics

*Significant at 10%. **Significant at 5%. ***Significant at 1%.

Linear probability model; Dependent variable: In the Puzzle Group

Explanatory Variables (1) (2) (3) (4) (5) (6) (7)

Puzzle personality (index) 0.0148*** 0.0141*** 0.0136*** 0.0152***

(0.0034) (0.0035) (0.0035) (0.0034)

Couple’s age difference 0.0001 0.0001 0.0002

Openness 0.0127 0.0117

Downloaded from https://academic.oup.com/rof/article/22/6/2109/3970879 by guest on 20 March 2024

Linear probability model; Dependent variable: In the Puzzle Group

Explanatory Variables (1) (2) (3) (4) (5) (6) (7)

Agreeableness 0.0177** 0.0190**

Male’s puzzle personality (index) 0.0025

Age 0.0104* 0.0105* 0.0111* 0.0114* 0.0115***

Significant at 10%. Significant at 5%. Significant at 1%.

Openness 0.0106 0.0090 0.0085 0.0084 0.0015

Age 0.0108* 0.0111* 0.0155* 0.0156* 0.0157* 0.0138*

Significant at 10%. Significant at 5%. Significant at 1%.

Openness 0.0097* 0.0098* 0.0096**

Significant at 10%. Significant at 5%. Significant at 1%.

Puzzle personality (index) 0.0148* 0.0141* 0.0136* 0.0152*

Agreeableness 0.0177 0.0190

Significant at 10%. Significant at 5%. Significant at 1%.

Openness 0.0076** 0.0058* 0.0140* 0.0067 0.0082 0.0151 0.0118 0.0106

Extraversion 0.0122* 0.0102* 0.0124* 0.0113 0.0124* 0.0417 0.0498 0.045