Factors that contribute to becoming unbanked.
The proportion of US families that are unbanked (i.e., have no type
of checking or savings account) has steadily declined for more than two
decades. Nonetheless, more than nine million families still do not
participate in the financial mainstream, and roughly half these unbanked
families previously held a traditional bank account. This study uses the
2004 longitudinal Survey of Income Program Participation to examine the
dynamic process within which changes in families’ circumstances
contribute to their becoming unbanked. Our findings suggest that
families are significantly more likely to become unbanked when there is
a decline in family income, loss of employment, or loss of health
insurance coverage. Race and ethnicity, level of education or family
income, and marital or housing status are also important determinants of
whether families participate in the financial mainstream or not. To our
knowledge, this is the first analysis of the dynamic process by which
families change bank status.
Participation in mainstream financial markets improves a
consumer’s ability to build assets and create wealth, protects them
from theft and discriminatory, predatory, or otherwise unsavory lending
practices, provides federal and state consumer financial protections,
and offers financial safety nets against unforeseen circumstances
(Rhine, Greene, and Toussaint-Comeau 2006). Consumers benefit from the
ability to deposit and cash checks, store and save funds, gain access to
cash when needed, pay bills, purchase money orders, make
account-to-account money transfers, and send or receive wire transfers
using their bank accounts. (1)
The proportion of unbanked families, those without either a
checking or a savings deposit account has steadily declined for more
than two decades. In 1989, close to 15% of families were unbanked
(Kennickell et al. 2000). By 2009, that number had fallen to 7.7%
(Bricker et al. 2011; FDIC 2009). This suggests that the number of
families participating in the financial mainstream has substantially
increased. Upon closer inspection, however, a striking countervailing
trend is revealed. For more than a decade, roughly half of unbanked
families once held a traditional bank account. Little is known about how
changes in a family’s attributes or circumstances influence their
bank status over time. Several studies have shown that the different
states of bank status are unequally distributed for families of
different racial/ethnic groups or by immigrant status. (2) For example,
Rhine and Greene (2006) found that minority or immigrant families were
significantly more likely to be unbanked than white or US-born families,
respectively, over a four-year period. However, detailed study of
changes in banking status remains lacking.
This study examines the dynamic process by which families’
circumstances contribute to changes in bank status from being banked in
one period to being unbanked in a subsequent period. Of particular
interest is how specific types of shocks may result in a family becoming
unbanked, thereby leaving the financial mainstream. We used the 2004
longitudinal Survey of Income Program Participation(SIPP) to estimate a
recursive bivariate probit model in which bank status in the later
period is conditioned on the bank status in the initial period. Our
findings suggest that a family’s shift away from the financial
mainstream (switch from banked to unbanked status) is significantly
influenced by declines in family income, by lost employment, and by a
loss of health insurance coverage. Race and ethnicity, level of
education and family income, marital and housing status, and geographic
location also play important roles in whether or not families exit the
To our knowledge, an analysis of the dynamic process by which
families change bank status has not been previously undertaken.
Documenting how changes in attributes or circumstances influence some
families to move out of the financial mainstream provides useful
information to policymakers and others interested in better
understanding why families become unbanked and move out of the financial
BANK STATUS OF FAMILIES
Consumer finance research describes the unbanked as those who do
not hold a checking or a savings deposit account. The family is
considered unbanked if neither the head of the family nor the spouse or
partner of the head has a checking or a savings account. Previous
research has consistently found that the unbanked are more likely to
have lower family income and net worth and to be less educated, younger,
members of certain minority groups, unmarried, or unemployed. (3)
Unbanked families are more vulnerable to financial disruptions caused by
natural or manmade disasters because their funds are not safely held in
financial institutions that provide important consumer protections and
financial safety nets (Cheney and Rhine 2006). (4)
Insights can be drawn from earlier studies about why families
switch bank status. For example, when asked why they either never had a
checking account or closed their account, more than one-third of the
unbanked surveyed said they did not believe they had enough money to
justify having an account (FDIC 2009). Other frequent reasons given for
not having a checking account include: don’t write enough checks to
be worthwhile, service charges are too high, and do not like dealing
with banks. In 2009, 9 million families were unbanked (FDIC 2009).
Changes in Bank Status
Changes in key socioeconomic and demographic conditions of the
family are expected to influence their bank status. The survey
respondent, representing the family, is either the head of the family or
the spouse or partner of the family head. (5) Job loss or fewer hours
worked by the head of the family or spouse or partner will lower family
income. Income losses will likely have a profound effect on the family
budget and its ability to purchase and maintain use of products and
services, including financial services provided by checking and savings
account ownership (e.g., Couch, Daly, and Gardiner 2011). (6) Under
these conditions, it may be that losses in family income will increase
the likelihood that a banked family becomes unbanked. Losses in income
that push a family under the poverty threshold also may increase the
likelihood of becoming unbanked.
Other factors that may trigger a shift in bank status include
changes in marital status, serious health shocks, and changes in housing
decisions. Married families are likely to face a greater need for
completing a larger number financial transactions than unmarried
persons. As such, married families may be more likely to be banked than
unmarried families. Moreover, families that include a larger number of
children may also require a greater number of financial transactions
than small families. Research by Love (2010) shows that divorce has a
significant influence on the couple’s post-divorce savings and
portfolio choices. A breakup between banked married persons may increase
the likelihood that at least one of the spouses or partners becomes
unbanked. As Lehrer (2003) describes, this may be particularly true for
women with children who undergo a significant decline in financial
well-being following divorce.
Health shocks can affect labor supply decisions and family income
(Coile 2004). Families faced with health shocks may deplete their
savings and checking account funds if pressures on the family’s
financial situation make it too difficult to keep accounts open. Whether
members of the family experienced serious health shocks is unobservable
in the SIPP longitudinal data. (7) What is observable is whether family
members are at least partially financially protected against health
shocks through insurance. Health insurance coverage serves as an
important way for families to protect themselves against health shocks.
(8) Possessing health insurance coverage has been shown to mitigate
effects of health shocks on labor supply and family income (Bradley,
Neumark, and Motika 2011). The number of individuals without health
insurance has steadily risen since the late 1990s because of a decline
in employer-sponsored health insurance coverage and the rising costs of
this insurance. (9) Banked families that lose or become priced out of
health insurance coverage are likely to become more vulnerable to
employment and income losses, possibly contributing to their becoming
Observed changes in housing decisions may be symptomatic of
financial or real estate shocks, changes in preferences, or life-cycle
influences. It has been well documented that wealth derived from home
values has substantially declined in recent years. (10) A drop in
housing wealth, possibly along with changes in financial circumstances,
preferences, or life-cycle effects, may have led some families to
reevaluate their housing asset price risks against the risks associated
with unstable rental costs. (11) As part of this process, it is possible
that banked families who turn away from homeownership in favor of rental
housing may become unbanked at the same time.
In spite of changes in socioeconomic or demographic conditions,
families may choose not to change their bank status. Stickiness could
occur because of habit persistence or a reluctance to change the status
quo. Under these circumstances, a family’s bank status could be
observed as either a “banked” or an “unbanked”
DATA AND DESCRIPTION OF SAMPLE
For this study, we employ the 2004 SIPP longitudinal survey. The
survey is a multistage-stratified sample of the US civilian
non-institutionalized population. It is a rotating panel made up of
twelve waves of data collected every four months. Sampling weights are
used in the analysis because the panel oversamples low-income families
that participate in income support programs. An assets and liabilities
topical module was administered in the third and sixth wave to collect
information about family financial assets, including deposit account
holdings along with other relevant socioeconomic and demographic
By design, the SIPP is particularly useful for studies of families
participating in income support programs. It has been suggested that
SIPP respondents are more likely not to answer certain questions related
to asset ownership. Main reasons given for not answering certain asset
ownership questions include concern about compliance with “asset
limits” enforced by social programs such as TANF cash assistance,
Supplemental Security Income (SSI), and food stamps (Chen and Lerman
2005) and a tendency to report only their most valuable assets (Czajka,
Jacobson, and Cody 2004). We find no evidence of respondent omission in
answering questions related to checking and savings account assets.
Overall, the SIPP longitudinal panel is particularly well suited for
asking how changes in specific attributes such as decreases in family
income, loss of employment, or loss of health insurance contribute to a
banked family becoming unbanked, especially families who are most
vulnerable to economic shocks.
For this analysis, there is a working sample of 21,521 primary
families that are representative of roughly sixty-seven million families
nationwide. Demographic indicators such as gender, age, and
race/ethnicity reflect information about the survey respondent, who is
either the head of the family or spouse/partner in a dual-headed
household. Respondents were at least 18 years of age. A description of
the socioeconomic and demographic characteristics used in the analysis
is provided in the Appendix.
The proportional means of these characteristics for both periods
are shown in Table 1. For descriptive purposes, drawing on the mean
attributes of the respondents in the first period, we find that the
majority of the sample is between 25 and 64 years of age, married,
white, has at least a high school or some college education, works (48%
work full time), owns a home, is a US citizen, and resides in a
metropolitan area. More than one in five families (21%) is unbanked.
(13) Table 1 also shows that 8% of families lost between 10% and 25% of
income and another 8% lost more than 50% of family income from the first
period to the second period. Similarly, 9% of families lost between 25%
and 50% of family income. Four percent of families that were above the
poverty line in the first period lost enough income to fall into poverty
by the second period. Six percent of families (head of household or
spouse or partner) became unemployed, while 3% of families experienced a
loss in health insurance coverage by either the head of the household or
spouse or partner. A small proportion of families (1%) went through
divorce or moved from a home to rental housing during the same time
ECONOMIC MODEL AND ECONOMETRIC FRAMEWORK
As in Rhine and Greene (2006), we consider a family’s bank
status from a consumer choice theoretical viewpoint. We define the net
utility for consumer i of holding a deposit account in period t as:
[y*.sub.it] = [beta]'[x.sub.it] + [[epsilon].sub.it] +
where [[epsilon].sub.it] is assumed to be unobserved effects that
may vary from period to period and [u.sub.i] is unobserved effects or
attributes of the consumer that are invariant from period to period,
both assumed to be normally distributed and uncorrelated with the
observed effects, [x.sub.it]. Being banked in period t is then
determined by the observation:
[y.sub.it] = 1 if [y*.sub.it] > 0 and 0 otherwise.
We observe the consumer in two periods, denoted period 0 and 1 for
present purposes. Switching behavior may occur in either direction, so
the four possible outcomes, or cells, include: stable bank status
behavior (banked or unbanked in both periods) or switching from period 0
to period 1, either from unbanked status to banked, or the reverse. With
two periods of observation, the preceding random effects specification
defines a bivariate probit model in which the correlation across the two
periods is [rho] = [[sigma].sup.2.sub.u]/(1 + [[sigma].sup.2.sub.u]). In
this study, we are specifically interested in switching behavior. There
are two possible questions to be considered: (1) whether the consumer
switches between banked and unbanked status and (2) the direction of a
switch if one takes place. Let’s consider the second question
first. Based on the model suggested thus far, we might consider
analyzing the two cells ([y.sub.i0] = 1, [y.sub.i1] = 0) and ([y.sub.i0]
= 0, [y.sub.i1] = 1). This might be recast as a simple model for the
binary outcome only for those who do switch status:
[z.sub.i] = [y.sub.i1]|[y.sub.i0] [not equal to] [y.sub.i1] and 0
However, this neglects the dynamic aspects of the behavior.
Switching behavior will depend on the characteristics of the individual
and changes that might motivate a switch, such as family income,
employment status, or health insurance coverage. Thus, we consider a
dynamic specification for the bivariate probit model:
[y*.sub.i0] = [beta]'[x.sub.i0] + [[epsilon].sub.i0] +
[y*.sub.i1] = [beta]'[x.sub.i1] +
[alpha]'([DELTA][x.sub.i]) + [[epsilon].sub.i1] + [u.sub.i] (1b)
The fact that this is a two-period model makes it possible to
incorporate changes in characteristics that might help explain changes
in banking status. The generic term [DELTA][x.sub.i] indicates changes
in the subset of the measured characteristics whose changes might induce
a switch. We also note that preferences might change, which calls into
question the assumption that the coefficients are the same in the two
periods. Though the model now accommodates changes that might motivate
switching behavior, it does not account for habit persistence, or
reluctance to switch that is not otherwise explained. A dynamic model
that incorporates both of these ideas is:
[y*.sub.i0] = [[beta]’.sub.0][x.sub.i0] + [[epsilon].sub.i0] +
[y*.sub.i1] = [[beta]’.sub.1][x.sub.i1] +
[alpha]'([DELTA][x.sub.i]) + [[epsilon].sub.i1] + [delta][y.sub.i0]
+ [u.sub.i] (2b)
Although cast as a “panel data” (random effects) model,
with two periods observed, this is mathematically, if not logically
equivalent to a “recursive bivariate probit model.” (The
familiar recursive bivariate probit model would be atemporal.) We fit
the model using full information maximum likelihood (Greene 2012,
Based on the findings from previous studies, it is expected that
families are more likely to be banked if the family head
(spouse/partner) is more highly educated, including completed a high
school degree (EDUC12), completed some years at a college or university
(SOME COLLEGE), completed a degree from a college or university
(COLLEGE), or completed at least some years of training in a graduate
program at a college or university (GRADSCHOOL), than the family head
(spouse/partner) who does not complete high school (EDUCll). (14)
Similarly, married (MARRIED) families are more likely to be banked than
unmarried (UNMARRIED) families. (15) The greater the number of children
(NUMKIDS), the more likely it is that the family will be banked. Older
adult family members, including those between 25 and 45 years of age
(AGE45), between 46 and 64 years of age (AGE64), and 65 years of age or
older (AGE65), are more likely to be banked than younger adult family
members between 18 and 24 years of age (AGE24).
Consistent with other studies, it is expected that white (WHITE)
adult family members are more likely to be banked than black (BLACK),
Hispanic (HISPANIC), or Asian (ASIAN_OTHER) families. (16) In addition,
those who work full time (FULLTIME), part time (PARTTIME), or at varying
hours per week (WORK VARIES) are more likely to be banked than those who
are unemployed (UNEMPLOYED). It is also expected that families with
higher income (FAMINC) are more likely to be banked than those with
lower family income.
Considering the dynamic process of a family’s banking status,
we are interested in determining whether changes in certain
circumstances have a significant influence on the likelihood that a
banked family becomes unbanked. For instance, events or financial shocks
that lower family income (e.g., LOWINC1, LOWINC2, or LOWINC3) or push
family income under the poverty threshold (NOPOV_POV) are expected to
increase the likelihood that a banked family becomes unbanked. (17)
Banked families (i.e., respondent or spouse or partner) that become
unemployed (LOSTEMP) or lose health insurance coverage (LOSTINS) are
expected to be more likely to become unbanked than those that retain
employment or health insurance coverage. Health insurance is broadly
defined to include coverage provided by private insurers, VA/military,
Medicaid, or Medicare. If the respondent or spouse/partner loses health
insurance coverage, then there is a loss of health insurance coverage.
Families that experience a divorce (MAR_NOMAR) or move from home
ownership to rental housing (OWN_TO_RENT) also are more likely to become
unbanked than those who stay married or remain in their home,
The bivariate probit model assigns a probability to each of the
four possible combinations of banking status for the two periods. The
one of most interest in this study is the switch from being banked in
the first period to being unbanked in the second, that is, [y.sub.i0] =
1 and [y.sub.il] = 0. For families that are banked in the first period
the question of interest is what changes in their circumstances may
contribute to their becoming unbanked in the subsequent period.
Potentially important changes include a decline in family income, loss
of employment or health care coverage, divorce, or moving from
homeownership to rental housing.
The estimates of the model coefficients are reported in Table 2.
Column 1 relates to equation (2a) and shows the estimated index equation
for being banked in the first period. Column 2 is the index equation for
being unbanked in the second period and corresponds to equation (2b).
The estimated covariates in Column 1 include the socioeconomic and
demographic factors expected to influence the likelihood of being
banked. This specification mirrors analyses discussed earlier in the
literature review concerning the unbanked (see, e.g., Footnote 2).
Column 2 is an extension of the literature because it allows for a more
dynamic process where the specification includes not only covariates
expected to influence the likelihood of becoming unbanked but also
accounts for the banking status in the first period as well as how
changes in attributes or circumstances between periods contribute to a
change in bank status in period 2. Statistical significance of the
covariates is also reported in Table 2. The partial effects of all
covariates are reported in Table 3. (18) Statistically significant
covariates are discussed in the sections that follow.
Family Income and Income Losses
Family income has a significant small negative influence on the
likelihood that banked families become unbanked. As shown in Table 3, an
increase in monthly family income (FAMINC) by $100 lowers the likelihood
that banked families become unbanked by 0.02 percentage points. Although
declines in family income are expected to have a positive influence on
the likelihood that banked families become unbanked, we find that it
takes relatively large drops in income before banked families are
significantly more likely to become unbanked. (19) A loss in subsequent
income of between 25% and 50% (LOSTINC2) increases the likelihood of
becoming unbanked by almost 1.5 percentage points; while a loss in
income of 50% or more (LOSTINC3) increases the likelihood of becoming
unbanked by 2.8 percentage points. Smaller declines in family income
(e.g., 10% or less), however, are found to have an insignificant
influence on banked families becoming unbanked. This finding suggests
that there may be a “stickiness” to remaining banked even as
family income falls, in this case, between 10% and 25%, between the two
Race and ethnicity have large, significant influences on the
likelihood of changing bank status. Specifically, banked families that
are black, Hispanic, or Asian are significantly more likely to become
unbanked than white banked families. This finding is greatest for black
(BLACK) banked families, who are 12.7 percentage points more likely to
become unbanked than white banked families in the subsequent period. For
Hispanic (HISPANIC) banked families, the likelihood of becoming unbanked
is 11.5 percentage points higher than for white banked families. The
likelihood of being unbanked in the second period for Asian families
(ASIAN_OTHER) is almost 4.7 percentage points higher than for white
banked families. Married (MARRIED) banked families are 7.9 percentage
points less likely to become unbanked than unmarried families. In this
analysis, changes in marital status (MAR_NOMAR) or being a US citizen
(CITIZEN) did not significantly influence banked families toward
becoming unbanked. An increase in the number of children (NUMKIDS) has a
small (0.8 percentage point) positive significant influence on the
likelihood that a banked family becomes unbanked.
Higher levels of educational attainment have a significant negative
influence on the likelihood that banked families become unbanked. As
shown in Table 3, the partial effects become larger as the family’s
level of education is higher. For example, banked families who have a
high school degree (EDUC12) are 6.0 percentage points less likely to
become unbanked than banked families with less than a high school
education (EDUC 11). Similarly, banked families who have a college
degree (COLLEGE) are almost 14.0 percentage points less likely to become
unbanked than banked families with less than a high school education
Employment Status and Health Care Coverage
Being employed has a significant influence on whether banked
families changed bank status or not. For example, banked families
working full time (FULLTIME) are 2.5 percentage points less likely to
change bank status than are unemployed (UNEMPLOYED) banked families.
Similarly, banked families working part time (PARTTIME) are 2.8
percentage points less likely to become unbanked. Banked families (i.e.,
head of household or spouse or partner) that experience unemployment
(LOSTEMP) are almost 2.0 percentage points more likely to become
unbanked than those that remain employed. (21) Loss of health care
coverage is also found to have an important influence on the likelihood
of being unbanked. For banked families who lose health insurance
coverage (LOSTINS), the likelihood of becoming unbanked is 6.1
percentage points higher than for those who remain covered by health
Housing and Residential Location
A banked homeowner (OWNER) is almost 5.2 percentage points less
likely to become unbanked than a renter (RENTER). However, banked
homeowners who change their housing to rental (OWN_TO_RENT) are no more
likely to become unbanked than those homeowners who maintain their
housing status from the subsequent period. (22) Living in certain
regions also has a significant influence on the likelihood that a banked
family chooses to become unbanked. In particular, banked families
residing in either the northeast (NORTHEAST) or west (WEST) are less
likely to become unbanked than banked families residing in the south
(SOUTH). We find that banked families living in the northeast are 6.0
percentage points less likely to become unbanked; and banked families
residing in the west are 8.0 percentage points less likely to become
unbanked than those residing in the south (SOUTH).
The bivariate probit model gives us a glimpse into understanding
how a specific bank status in one period may influence the bank status
in another period. As shown in Table 2, Column 2, being banked (BANKED)
in the first period has a significant negative influence on the
likelihood that a family becomes unbanked, suggesting that families may
view being banked as providing value so that over time there is
resistance to change bank status.
This study provides valuable insights into how changes in
circumstances contribute to banked families becoming unbanked. Families
faced with falling income, loss of employment, and loss of health
insurance coverage are significantly more likely to become unbanked and
move out of the financial mainstream. Having higher levels of education,
being employed, earning higher income, and owning a home lower the
likelihood that a banked family becomes unbanked. After controlling for
these and other socioeconomic and demographic attributes, we find that
black, Hispanic, and Asian families are significantly more likely of
becoming unbanked than white families. Further research is needed to
better understand why this is the case.
To our knowledge, this is the first empirical analysis of the
dynamic process by which families change bank status. Becoming unbanked
exposes families to higher risks because their funds are no longer held
at an insured depository institution and their financial transactions
are unlikely to be covered by consumer protection laws and regulations.
As additional waves of SIPP become available, it will be possible to
improve our tracking of families’ bank status. Documenting how
changes in circumstances influence some families to move out of the
financial mainstream provides useful information to policymakers and
others interested in better understanding why families change bank
APPENDIX 1 Description of FatnilY Socioeconomic and Demographic Variables Name Description Age Groups Family respondent is: AGE24 between 18 and 24 years of age AGE45 between 25 and 45 years of age AGE64 between 46 and 64 years of age AGE65 65 years of age and older Gender Family respondent is: MALE male FEMALE female Marital Status Family respondent is: MARRIED married (spouse/partner present or not) UNMARRIED family respondent single, separated, divorced, or widowed Children in Family NUMKIDS Total number of children under 18 years of age Race/Ethnicity Family respondent is: WHITE white, non-Hispanic BLACK black HISPANIC Hispanic, non-white ASIAN OTHER Asian or other racial/ethnic groups not already specified Education Family respondent has: EDUC11 completed less than a high school degree EDUC12 completed a high school degree or equivalent SOME COLLEGE completed some years in college or university COLLEGE completed a degree at a college or university GRADSCHOOL completed at least some training at a graduate level college or university program (e.g., masters, PhD, MD, JD program) Work Status Family, respondent has: WORK either full- or part-time work in the relevant month of reference period FULLTIME at least a 35-hour work week in the relevant month of reference period PARTTIME less than a 35-hour work week in the relevant month of reference period WORK VARIES a varied work week schedule in the relevant month of reference period UNEMPLOYED no full- or part-time employment in the relevant month of reference period Familv Income Measure FAMINC Total family income for the relevant month of the reference period Housing Family respondent is: OWNER a home owner RENTER a renter Citizenship Family respondent is: US CITIZEN a US born citizen IMMIGRANT an immigrant born outside the US Geographic Location Family resides in: METRO a metropolitan area NORTHEAST the northeast region MIDWEST the midwest region WEST the western region SOUTH the southern region Bank Status BANKED Head of family and spouse/partner have either a single (alone) or joint checking and/or a savings deposit account UNBANKED Head of family and spouse/partner do not have either a single (alone) or joint checking or a savings deposit account Change Factors MAR NOMAR Respondent is married in period 0 and not married in period l NOPOV POV Family is not in poverty in the relevant month of reference period for period 0 and is in poverty in the relevant month of reference period for period 1 LOSTINS Respondent or spouse/partner has health insurance period 1 and has no health insurance period 2. Sources of health insurance coverage include private, VA/military. Medicaid, and Medicare. LOSTINCI Family income fell by 10% to 25% from period 0 to period 1 (family income in 2005 dollars) LOSTINC2 Family income fell by 25% to 50% from period 0 to period 1 (family income in 2005 dollars) LOSTINC3 Family income fell by 50% or more from period 0 to period 1 (family income in 2005 dollars) LOSTEMP Respondent or spouse/partner was employed in period 0 and unemployed in period 1 OWN TO RENT Respondent was homeowner in period 0 and renter in period 1
Bostic, Raphael, Stuart Gabriel, and Gary Painter. 2009. Housing
Wealth, Financial Wealth, and Consumption, New Evidence from Micro Data.
Regional Science and Urban Economics, 39 (1, January): 79-89.
Bradley, Cathy J., David Neumark, and Meryl I. Motika. 2011. The
Effects of Health Shocks on Employment and Health Insurance: The Role of
Employer-Provided Health Insurance. NBER Working Paper Series 17223
Bricker, Jesse, Brian Bucks, Arthur Kennickell, Traci Mach, and
Kevin Moore. 2011. Surveying the Aftermath of the Storm: Changes in
Family Finances from 2007 to 2009, Federal Reserve. Finance and
Economics Discussion Series (2011-17) (March).
Caskey, John. 1994. Fringe Banking: Check-Cashing Outlets.
Pawnshops. and the Poor. New York: Russell Sage Foundation.
–. 1997. Lower-Income Americans, Higher Cost Financial Services.
University of Wisconsin-Madison School of Business: Filene Research
Institute Center for Credit Union Research.
Chen, Henry and Robert I. Lerman. 2005. Do Asset Limits in Social
Programs Affect the Accumulation of Wealth’? The Urban Institute
(4) (August): 1-12.
Cheney, Julia S. and Sherrie L. W. Rhine. 2006. How Effective Were
the Financial Safety Nets in the Aftermath of Katrina? Federal Reserve
Bank of Philadelphia Payment Cards Center Discussion Paper (January).
Coile, Courtney C. 2004. Health Shocks and Couples’ Supply
Decisions. NBER Working Paper Series 10810 (September).
Couch, Kenneth A., Mary C. Daly, and Colin Gardiner. 2011.
Life-Cycle Shocks and Income. FRBSC Economic Letter 08 (March): 1-2.
Czajka, John L., Jonathan E. Jacobson. and Scott Cody. 2003/2004.
Survey Estimates of Wealth: A Comparative Analysis and Review of the
Survey of Income and Program Participation. Social Security Bulletin
Curran, Kelly. 2009. Home Prices Fall 13.8 Percent in 2009: Report.
HousingWire, (February). http://www.housingwire.com.
–.2009b. Household Net Worth Dives 18% in 2008. HousingWire,
(March). http://www. housingwire.com.
Cutler, David M. 2002. Employee Costs and the Decline in Health
Insurance Coverage. NBER Working Paper Series 9036 (July).
Federal Deposit Insurance Corporation. 2009. FDIC National Survey
of Unbanked and Underbanked Households (December). Washington, DC: FDIC.
Greene, William H. 2012. Econometric Analysis. 7th edition. Upper
Saddle River, N J: Prentice Hall.
Greene, William, Mark Harris, and Bruce Hollingsworth. 2012.
Inflated Responses in Measures of Self-Assessed Health, Department of
Economics and Business Working Papers, Curtin University (May).
Hogarth, Jeanne and Kevin O’Donnell. 1997. Being Accountable:
A Descriptive Study of Unbanked Households in the U.S. Proceedings of
the Association for Financial Counseling and Planning Education,
Holahan, John and Allison Cook. 2008. The U.S. Economy and Changes
in Health insurance Coverage, 2000-2006, Health Affairs, 27 (2, March):
Kennickell, Arthur, Martha Starr-McCluer, and Annika E. Sunden.
1997. Family Finances in the U.S.: Recent Evidence from the Survey of
Consumer Finances. Federal Reserve Bulletin, 83 (January): 1-24.
Kennickell, Arthur, Martha Starr-McCluer, and Brian J. Surette.
2000. Recent Changes in U.S. Family Finances: Results from the 1998
Survey of Consumer Finances. Federal Reserve Bulletin, (January): 1 29.
Kooce-Lewis, Joan, Roger Swagler, and John R. Burton. 1996.
Low-Income Consumers’ Use of the Alternative Financial Services
Sector. Consumer Interest Annual, 42:271-274.
Lehrer, Evelyn L. 2003. The Economics of Divorce, Chapter 3,
Marriage and the Economy, edited by Shoshana A. Grossbard-Shechtman.
Cambridge: Cambridge University Press.
Love, David A. 2010. The Effect of Marital Status and Children on
Savings and Portfolio Choice. The Review of Financial Studies, 23 (1):
National Academy of Sciences. 2002. Health Insurance Is a Family
Matter. Washington, DC: The National Academies Press.
Rhine, Sherrie L.W., William H. Greene, and Maude Toussaint-Comeau.
2006. The Importance of Check-Cashing Businesses to the Unbanked:
Racial/Ethnic Differences. The Review of Economics and Statistics, 88
(1, February): 146-157.
Rhine, Sherrie L.W. and William H. Greene. 2006. The Determinants
of Being Unbanked for U.S. Immigrants. Journal of Consumer Affairs, 40
(1, Summer): 21-40.
Sommer, Kamila, Paul Sullivan, and Randel Verbrugge. 2010. Run-up
in the House Price-Rent Ratio: How Much Can Be Explained by
Fundamentals’? NBER Working Paper Series 441 (July).
Taylor, John, Richard Fry, and Rakesyh Kochhar. 2011. Wealth Gaps
Rise to Record Highs Between Whites, Blacks, Hispanics. Washington DC:
Pew Research Center, Social & Demographic Trends.
(1.) Bank accounts include deposit or share accounts at financial
(2.) Examples of cross-sectional studies that examine the
determinants of being unbanked: FDIC (2009), Barr (2009), Rhine, Greene,
and Toussaint-Comeau (2006), Hogarth and O’Donnell (1997),
Kooce-Lewis, Swagler, and Burton (1996), and Caskey (1994, 1997). A
study by Rhine and Greene (2006) was longitudinal over a four-year
(3.) Barr (2009), Rhine, Greene, and Toussaint-Comeau (2006),
Hogarth and O’Donnell (1997), and Kooce-Lewis, Swagler, and Burton
(4.) Among the most important consumer protection laws and
regulations for transaction and savings account holders are: FD1C
insurance coverage; Electronic Fund Transfer Act and Regulation E,
Electronic Funds Transfer; Federal Truth in Lending Act and Regulation
Z, Truth in Lending; Expedited Funds Availability Act, Check Clearing
for the Twenty-First Century Act and Regulation CC, Availability of
Funds and Collection of Checks: and Truth in Savings Act and Regulation
DD, Truth in Saving.
(5.) For this study, single persons without children are also
considered a family unit.
(6.) This study describes how job loss, divorce, and disability can
have negative effects on individual and family income over time.
(7.) One health proxy considered was an analysis of the
respondent’s self-reported state of health. In each period, the
respondent reported his/her health status, which was measured on a
5-point Likert scale with 1 being poor health and 5 being excellent
health. A potential proxy for a health care shock could be a measurable
change in health status from one period to another. However, our
analysis found that there was very little movement along the scale
because the vast majority of respondents reported health status within a
very narrow band, typically at point 2 or point 3 along the 5-point
scale. This observation is common in analysis of self-reported health
measures, in many countries. See, for example, Greene, Harris and
(8.) A detailed discussion about the importance of considering
coverage obtained by the head of house and/or partner or spouse is given
in National Academy of Sciences (2002).
(9.) Cutler (2002) and Holahan and Cook (2008).
(10.) Bostic, Gabriel, and Painter (2005), Curran (2009b), and
Taylor, Fry, and Kochhar (2011).
(11.) Curran (2009a) and Sommer, Sullivan, and Verbrugge (2010).
(12.) The wave 3 Assets and Liabilities topical module was in the
field from October 2004 until January 2005 and the wave 6 Assets and
Liabilities topical module was in the field from October 2005 until
January 2006. More information about the U.S. Census SIPP program is
available at www.sipp.census.gov/sipp.
(13.) The 21% of unbanked households compares to 7.7% reported
using the 2009 Survey of Unbanked and Underbanked Households, a randomly
drawn, national survey sponsored by the Federal Deposit Insurance
Corporation. We attribute the differences to the greater emphasis on
lower-income households by the SIPP.
(14.) See, for example, Rhine, Greene, and Toussaint-Comeau (2006),
Rhine and Greene (2006), Hogarth and O’Donnell (1997) and
Kooce-Lewis, Swagler, and Burton (1996).
(15.) UNMARRIED status includes those who are single, separated,
divorced, or widowed.
(16.) ASIAN_OTHER includes Asians and/or other racial/ethnic groups
not already specified and too small in sample size to analyze
(17.) A change in poverty status is based on a monthly poverty
threshold measure, where a family is determined by the US Census to be
in poverty if the family’s monthly family income is lower than the
(18.) Table 3 contains the estimated partial effects for the
specific variables discussed. The probability of interest in our study
is Prob[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. Average
partial effects are obtained by averaging over the sample the
derivatives of this expression with respect to the variables in
question. The calculations are done using the PARTIALS procedure in
NLOGIT Version 5.0.
(19.) Family income for both periods is in 2005 dollars.
(20.) Other family income loss measures, such as losses measured in
income quartiles and quintiles were calculated and estimated to have
little, if any, influence on the likelihood that a banked family becomes
(21.) In separate analyses other measures of lost employment were
used to determine whether alternative changes in employment status have
a significant influence on the likelihood that a banked family chooses
to become unbanked. Other employment change measures included: full time
to part time, full time to working varying hours per week, and part time
to varying hours per week. None of these measures contributed
significantly to a banked family becoming unbanked. In part, this
outcome may be because the sample sizes for these measures were fairly
small or the window of observation was not long enough for these more
(22.) Rental housing includes families that reside in housing where
they do not pay rent. This category is less than 2% of the total sample
and has been nested within the rental housing variable.
Sherrie L.W. Rhine (firstname.lastname@example.org) is a Senior Economist in the
Division of Depositor and Consumer Protection at the Federal Deposit
Insurance Corporation. William H. Greene (email@example.com) is a
Professor of Economics at Stern School of Business, New York University.
The authors wish to thank the anonymous referees for their helpful
comments and suggestions to improve this article. The views expressed
here are those of the authors and may not represent the view of the
Federal Deposit Insurance Corporation.
TABLE 1 Proportional Means for Family Socioeconomic and Demographic Variables Proportional Variable Name Means of Sample Levels factors Wave 3 Wave 6 AGE24 .04 .02 AGE45 .44 .44 AGE64 .38 .39 AGE65 .14 .15 MALE .46 .46 FEMALE .54 .54 MARRIED .78 .77 UNMARRIED .22 .23 NUMKIDS .98 .97 WHITE .71 .71 BLACK .11 .11 HISPANIC .12 .12 ASIAN_OTHER .06 .06 EDUC11 .10 .08 EDUC12 .26 .28 SOME COLLEGE .36 .36 COLLEGE .18 .18 GRADSCHOOL .10 .10 WORK .62 .61 WORK FULLTIME .48 .47 WORK PARTTIME .09 .09 WORK VARIES .05 .05 UNEMPLOYED .38 .39 FAMINC (2005 dollars) $5,536 $5,611 HOMEOWNER .76 .77 RENTER .24 .23 US CITIZEN .86 .88 NONCITIZEN .14 .12 METRO .79 .75 NORTHEAST .l8 .16 MIDWEST .23 .26 WEST .22 .20 SOUTH .37 .38 BANKED .79 .80 UNBANKED .21 .20 Change factors UNBANKED in the second period -- .08 conditional on being BANKED in the first period MAR_NOMAR -- .0l NOPOV_POV -- .04 LOSTINS -- .03 LOSTINC1 -- .08 LOSTINC2 -- .09 LOSTINC3 -- .08 LOSTEMP -- .06 OWN_TO_RENT -- .01 Sample Size = 21,521 TABLE 2 Bivariate Probit Model for Banking Status in Two Periods (a) Column 1: Index Equation for Banked Column 2: Index Equation (Period 0) for Unbanked (Period 1) Constant -0.823 *** (0.078) 0.925 *** (0.093) (b) AGE45 0.106 * (0.056) 0.072 (0.060) AGE64 0.120 ** (0.058) 0.169 * (0.062) AGE65 0.287 *** (0.064) 0.030 (0.074) MALE -0.053 ** (0.023) 0.037 (0.025) MARRIED 0.333 *** (0.026) -0.173 *** (0.034) NUMKIDS -0.067 *** (0.011) 0.044 *** (0.012) BLACK -0.426 *** (0.034) 0.321 *** (0.043) HISPANIC -0.432 *** (0.033) 0.263 *** (0.043) ASIAN-OTHER -0.201 *** (0.050) 0.105 * (0.058) EDUC12 0.289 *** (0.037) -0.147 *** (0.045) SOME COLLEGE 0.569 *** (0.037) -0.357 *** (0.052) COLLEGE 0.715 *** (0.044) -0.433 *** (0.062) GRADSCHOOL 0.801 *** (0.056) -0.537 *** (0.077) WORK FULLTIME 0.301 *** (0.026) -0.138 *** (0.033) WORK PARTTIME 0.264 *** (0.043) -0.158 *** (0.048) WORK VARIES 0.118 ** (0.050) -0.078 (0.066) FAMINC (divided 0.040 *** (0.002) -0.010 *** (0.002) by 1000) US CITIZEN 0.044 (0.032) -0.038 (0.035) HOMEOWNER 0.491 *** (0.026) -0.268 *** (0.038) Change factors MAR_NOMAR -- -0.026 (0.129) NOPOV_POV -- 0.008 (0.058) LOSTINC 1 -- 0.034 (0.042) LOSTINC2 -- 0.083 ** (0.041) LOSTINC3 -- 0.170 *** (0.048) LOSTEMP -- 0.103 * (0.055) LOSTINS -- 0.303 *** (0.069) OWN_TO_RENT -0.099 (0.105) Rho (1,2) -- 0.097 (0.084) Sample size = 21,521 (a) Equation (2a): Probability of being banked in period 0 and equation (2b): Probability of being unbanked in period 1 given banking status in period 0. (b) Robust standard errors in parentheses. ***, **, * denote significance at the 1%, 5%, and 10% level, respectively. TABLE 3 Partial Effects of the Probability of Being Unbanked in Period 1 Given Banked in Period 0 Covariates Partial Effects Age groups AGE45 0.0001 AGE64 0.016 AGE65 -0.028 Gender MALE 0.014 Marital status MARRIED -0.079 Children in family NUMKIDS 0.008 Race%thnicity BLACK 0.127 HISPANIC 0.115 ASIAN-OTHER 0.047 Education EDUC 12 -0.060 SOME COLLEGE -0.131 COLLEGE -0.140 GRADSCHOOL -0.149 Work status FULLTIME -0.025 PARTTIME -0.028 WORK VARIES -0.014 Family income measure FAMINC -0.002 Citizenship US CITIZEN -0.007 Housing OWNER -0.052 Geographic location WEST -0.080 NORTHEAST -0.060 MIDWEST -0.041 Change factors MAR_NOMAR -0.005 LOSTINC 1 0.006 LOSTINC2 0.015 LOSTINC3 0.032 LOSTEMP 0.019 LOSTINS 0.061 OWN_TO_RENT -0.018 NOPOV POV 0.001