Understanding the Racial/Ethnic Gap in Bank Account Ownership among Older Adults

Abstract

The observed racial/ethnic gap in bank account ownership among older adults is substantial. We investigate socioeconomic, cognitive, and cultural barriers underling it. As additional potential barriers are accounted for, the residual gaps in financial inclusion with respect to Whites are reduced by 19% for blacks and 46% for Hispanics. We find that citizenship and “taste for privacy” play a limited role for both minority groups, while real asset ownership, health, cognitive ability, and cultural hurdles contribute substantially to the gap. For Hispanics, language barriers explain most of the gap, while neighborhood-level socioeconomic characteristics are more salient for blacks. We also examine how the racial/ethnic composition of couples influences financial decisions. We estimate a significantly smaller residual gap between “mixed” and white couples than between minority and white couples. We provide empirical evidence suggesting that, other things equal, mixed couples are less concerned with the cultural/psychological barriers facing minority couples.

Access to financial services has become an important policy issue in developed countries where, notably, a large proportion of the population lacks access to basic financial services. About 20% of the population are unbanked in the European Union and the United Kingdom (Datta 2009). According to the latest National Survey of Unbanked and Underbanked Households of the Federal Deposit Insurance Corporation (Federal Deposit Insurance Corporation 2016), 7% of households were unbanked in 2015 in the United States.

Being unbanked is associated with higher transactional costs and poor saving behavior. Individuals without a bank account are likely to save less and incur significant monetary costs for everyday transactions (Carbo Valverde, Gardener, and Molyneux 2005; Thaler 1999). Considering only the transaction fees when using nonbank payment services (thus excluding costs associated with the loss of interest earned on savings, lack of protection, etc.), it has been estimated that the cost of being unbanked in the United States represents around 4% of median household income (Caskey, Solo, and Duran 2006). Lack of participation in the financial sector is also related to the ability to effectively plan for the future. Lack of interaction with financial institutions, in fact, is likely to decrease individuals’ knowledge of existing financial products that can be used to save for retirement (Clarke and Ambrosio 2003; Lusardi and Mitchell 2011).¹

Minorities in the United States are more likely to be unbanked and underbanked (Hogarth, Anguelov, and Lee 2005). Latest data on the unbanked show that while the unbanked rate among non-Hispanic white households is 4%, Hispanic and black households exhibit the highest unbanked rates, with 20% and 21%, respectively (Federal Deposit Insurance Corporation 2016). Also, while the likelihood of having a checking or savings account increases steadily with age among non-Hispanic whites, minority older adults remain noticeably unbanked. Black and Hispanic individuals aged 65 and older are approximately two times less likely than their White counterparts to have a bank account (AARP Public Policy Institute 2010).

As of 2014, individuals 65 and older represent 14.5% of the US population and this fraction is expected to grow to 21.7% by 2060. Blacks aged 65 and older represent 9% of the older population in the United States and by 2060 this percentage is projected to grow to 12%. The proportion of older Hispanics is expected to grow more rapidly. As of 2014, Hispanics aged 65 and older represent 8% of the older population, a fraction that will rise to 22% by 2060 (Administration for Community Living 2014). In view of the growing presence of Hispanic and black minorities in the United States and the rapid aging of the population, these statistics constitute a cause of national concern as the financial health and economic security of more and more households is expected to be at risk in the near future. These statistics also call for a deeper understanding of the saving behavior of more vulnerable groups and the barriers they face toward participation in the formal financial sector. This knowledge is crucial for devising and implementing interventions that can effectively change financial decision making among households where resources are most scarce and cultural/psychological hurdles more difficult to overcome. In this paper, we take on this task by investigating what determines the racial/ethnic gap in bank account ownership among older adults in the United States and identify the main factors that are responsible for it.

We contribute to the existing literature in several ways. First, we use a representative sample of the US population over the age of 50, drawn from the Health and Retirement Study (HRS) over the period 2000–2012, and assess differences in financial inclusion of Hispanic and black households with respect to their non-Hispanic white counterparts.² While, for the reasons given above, minority older adults constitute an interesting case study, they have not been the focus of attention in previous work. Hogarth, Anguelov, and Lee (2005) consider the entire US population and analyze the socioeconomic characteristics, among them age and race, associated with bank account ownership. Lusardi (2005) examines the financial behavior of individuals between the ages of 40 and 65, arguing that high information and learning costs prevent minorities from accessing basic financial services and accumulating enough retirement wealth.

Second, exploiting the richness of information available in the HRS, we build on the empirical models adopted in previous studies and amend them with additional factors that are presumably related to the observed racial/ethnic gap in bank ownership. Specifically, we gauge the relative importance of (1) real asset holdings, (2) health status and cognitive ability, (3) being foreign born and citizenship status, (4) “taste for privacy,” and (5) language barriers in explaining the lack of financial inclusion among minorities, above and beyond the explanatory power of conventional socioeconomic characteristics, such as education, income, and wealth, that have been used in the previous work. Moreover, relying on restricted geographic identifiers, we condition our analysis on the characteristics of the neighborhood where households reside. To the best of our knowledge, such comprehensive analysis of what prevents access to basic financial services among older minorities has not been performed before.

Third, the design of the HRS allows us to explore how the racial and ethnic composition of couples is related to access to financial services. Specifically, we investigate whether the behavior of couples with both minority members is different from that of “mixed” couples, where only one member belongs to a minority group. Looking at possible spousal effects in terms of financial inclusion constitutes a major, innovative contribution of our study.

Finally, from a methodological point of view, we take full advantage of the long panel data set at our disposal and estimate both random-effects (RE) and correlated random-effects (CRE) models. In the latter case, we include averages of time-varying variables, which allow us to control more encompassingly for unobserved heterogeneity across households. Previous work has relied on RE models (e.g., Rhine and Greene 2006) only.

We document that minority older adults are significantly less likely to own a bank account than their white counterparts after controlling for a comprehensive set of socioeconomic variables and neighborhood characteristics. As we account for additional potential barriers to financial inclusion, the residual gap with respect to whites is reduced by 19% for blacks and by 46% for Hispanics. We find that nativity, citizenship, and “taste for privacy,” as measured by lack of consent to sharing social security number (SSN) with HRS play no important role in determining differences across racial/ethnic groups. Ownership of real assets, health, and cognitive ability accounts together for about 3 percentage points of the minority gap in ownership of a bank account. While socioeconomic characteristics at the individual and neighborhood level (especially income and wealth) are more likely to explain differences between blacks and whites, language is a critical determinant of the existing gap in bank account ownership between Hispanics and whites. This is consistent with the findings of the 2015 Federal Deposit Insurance Corporation’s biennial survey on unbanked and underbanked households which indicates that households that speak only Spanish are unbanked (defined as lacking a checking or savings account) at five times the rate of households where Spanish is not the only language spoken (Federal Deposit Insurance Corporation 2016). Plausibly, lack of English proficiency may also be the manifestation of other cultural and behavioral barriers associated with little assimilation/acculturation. Our results also show that minority couples are from 13 to 10 percentage points less likely than white couples and from 10 to 7 percentage points less likely than “mixed” couples to own a bank account. Notably, the gap in bank account ownership between “mixed” and white couples is small at around 3 percentage points across different specifications. We use different specifications to assess the contribution of assortative mating to our results and do not find it to be large. Yet, we remain cautious in interpreting our findings as evidence of spousal spillovers in financial decision making as we cannot completely rule out assortative mating on unobservable preferences/tastes for managing household finances and adopting financial instruments.

The paper is organized as follows. The next section presents a literature review on the determinants of access to financial services. Next we discuss the data and methodology used in this study. Then we present the results from the analysis and a conclusion.

DETERMINANTS OF ACCESS TO FINANCIAL SERVICES

Financial inclusion is clearly determined by both demand and supply factors. As far as the former, there is ample evidence that access to financial services is closely related to socioeconomic status (SES), with the rate of unbanked individuals being significantly higher among low-educated and low-income groups (Anderloni et al. 2008; Claessens 2006). Similarly, ethnic and racial minorities are less likely to own checking and savings accounts and other financial instruments (Atkinson 2006; Carbo Valverde, Gardener, and Molyneux 2005; Hogarth, Anguelov, and Lee 2005; Robles 2009). Bogan (2014) highlights that poor health and limited cognitive ability, such as numeracy and memory, directly contribute to the lack of participation in financial markets, especially among more vulnerable groups.

Language and cultural barriers may prevent some segments of the population from accessing financial services (Barcellos et al. 2016; Caskey, Solo, and Duran 2006; Collard, Kempson, and Whyley 2001; Rhine and Greene 2006; Robles 2014), alongside with legal status and long-term plans for residency (Datta 2011). Bogan (2014) documents that immigrants from countries with more exposure to financial services are more likely to participate in the financial system in the United States. Furthermore, Northwood and Rhine (2016) find that Mexicans and Latin Americans immigrants have a higher probability of using nonbank financial services (payday loans, pawn shops, remittances, among others) than native born with similar education and income levels.

There are also behavioral issues related to subjective perceptions of the costs and benefits of owning a bank account. Bertrand, Mullainathan, and Shafir (2004) argue that when studying individuals’ usage of financial services not only major factors, such as the cost of a bank account, financial education, and availability of financial information should be taken into consideration but also small situational barriers, like distance from a bank, hours of service, bank customer service, and concerns related to consumer privacy. Hogarth, Anguelov, and Lee (2004) investigate subjective reasons behind the decision of remaining unbanked. They observe that most individuals cite account features, such as minimum balance, service charges, and usefulness of checks, as reasons for not owning a bank account. Motivational (not having enough money) and institutional (not wanting to deal with banks) reasons are mentioned by around 20% of unbanked individuals. Interestingly, account features receive similar weights within SES and racial/ethnic groups.

Regarding the supply side, financial institutions tend to be scarcer in disadvantaged and diverse communities (Rubin 2007). Because of the lack of profitability, banks might not offer adequate services to certain groups, especially the most vulnerable, such as low SES individuals, the elderly, and minorities (Federal Deposit Insurance Corporation 2011). Straight (2002) analyzes asset accumulation differences between whites and blacks between 1995 and 1998 and argues that lower accumulation for blacks is partly due to the supply of financial services. Choudhury (2001) mentions that, by primarily targeting whites, financial institutions may have created a cultural bias against holding financial assets among minorities. Stevenson and Plath (2006) document that financial institutions have not fully developed strategies to reach Hispanics because they have not considered Hispanics’ preferences and trust toward institutions that understand their culture, particularly language. Community and neighborhood environments as well as peer effects, participation in the cash economy, and informal saving circles are critical aspects to consider when studying financial inclusion (Robles 2009, 2014). Bohn and Pearlman (2013) show that enclaves, defined as areas with high concentrations of immigrants from the same region, play an important role in determining racial/ethnic differences in bank use. Bogan (2014) observes that immigrants who live around other immigrants have less exposure and information about the financial system than immigrants that live in a community of native US citizens. A similar argument applies to ethnic concentration impacting negative participation in the financial system. Along the same lines, Vermilyea and Wilcox (2002), using data from a survey conducted in Los Angeles and New York in the late 1990s, document that the racial composition of the neighborhood affects the likelihood of owning a bank account and that individuals living in Census tracts with higher share of Hispanics or blacks are less likely to own a bank account.

Building on the existing literature, our study aims at comprehensively examining the determinants of the observed racial/ethnic gap in bank account ownership, with a specific focus on the population over the age of 50. Even though they are projected to constitute an increasing fraction of the US population and often face more difficult to overcome hurdles toward financial inclusion, older minorities have received relatively little attention in previous work on this topic.³ For the purposes of our analysis, we take advantage of the richness of information elicited by the HRS questionnaire and amend previously estimated models with both individual-level and neighborhood-level variables that are bound to influence the decision of being unbanked. These include, but are not limited to, household wealth composition, health status and cognitive ability, immigration and citizenship status, taste for privacy, language barriers, neighborhood SES status, and supply of formal financial institutions. We show that such a richer model, featuring an exhaustive range of both demand- and supply-side factors, can explain a significant fraction of the observed racial/ethnic gap in bank account ownership and identify the variables that contribute the most to explaining high unbanked rates among older minorities.

DATA

We use data from HRS for the period 2000–2012 and restrict our sample to individuals 51–90 years old. This group includes retirees, as well as workers on the verge of retirement, a time when sound financial decisions are critical to ensure future economic wellbeing and income security at older ages.⁴ We focus on individuals who are designated as the financial respondent (FR) of the household (e.g., the person mostly responsible of and knowledgeable about household financial matters). We exclude individuals over the age of 90, as they are at a higher risk of physical/cognitive impairments that could potentially preclude them from being able to personally manage their finances and, therefore, from having a bank account.⁵

The HRS is a multipurpose, longitudinal household survey representing the US population over the age of 50. Since 1992, it has surveyed age-eligible individuals and their spouses on a biannual basis to track transitions from work to retirement, to measure economic wellbeing and changes in financial situation over time, and to monitor individuals’ health as they age. The HRS oversamples blacks and Hispanics, making it an ideal data set to examine racial/ethnic disparities in bank account ownership. These two minority groups exhibit very similar recruitment and response rates at baseline and retention/attrition in follow-ups as whites (Ofstedal and Weir 2011). This is important for the purposes of our study as it does not introduce biases due to differential/selective participation in the survey among different racial/ethnic groups.

Our dependent variable is a dichotomous indicator for access to financial services (i.e., financial inclusion), taking value 1 if the individual owns a checking, savings, or money market account. We assign individuals to three mutually exclusive groups: white non-Hispanic, black non-Hispanic, and Hispanic (we drop respondents who report being of other races).⁶

Our baseline specification mimics empirical models used in previous studies. It includes the following set of individual- and household-level characteristics: gender (male), a quadratic polynomial in age, education, an indicator for couple household (age-eligible respondent and spouse/partner are both interviewed), household size, work status, indicator for whether the respondent or the spouse/partner receives any social security or pension income, household income, and wealth tertiles. Year dummies and state dummies are also included to account for general trends in bank account ownership over the observation period and to control for state characteristics that can be correlated with the rate of unbanked individuals within different segments of the population.

We amend the baseline specification with potential determinants of the racial/ethnic differential in financial inclusion that, because of data availability, have not featured models previously estimated in the literature. First, we take into account the ownership of real assets, such as home and vehicles, as well as debt exposure. Second, we construct and control for indexes of overall health status and cognitive ability. Third, we include in the model indicators for nativity and citizenship, which may capture legal barriers to financial inclusion. Fourth, we consider a proxy for individual “taste for privacy” measured by the respondent’s consent to share their SSN with the HRS for administrative data linkage. Fifth, we control for whether the survey was answered in Spanish, indicating lack of English proficiency. Summary statistics for all the individual- and household-level variables used in the analysis are shown in Table 1.

TABLE 1.

Individual- and Household-Level Variables: Summary Statistics

	Percentage/Threshold	Mean	SD	Minimum	Maximum
Ownership of bank account (no=0, yes=1)		0.840	0.366	0	1
Race and ethnicity, FR
White, non-Hispanic (no=0, yes=1)	75.21
Black, non-Hispanic (no=0, yes=1)	17.50
Hispanic (no=0, yes=1)	7.28
Race and ethnicity, couple households
White–White (no=0, yes=1)	74.32
White–minority (no=0, yes=1)	1.48
Minority–minority (no=0, yes=1)	24.18
Gender (female=0, male=1)		0.429	0.495	0	1
Age		68.439	9.924	51	90
Education
Less than high school (no=0, yes=1)	24.68
High school graduate (no=0, yes=1)	31.29
Some college or more (no=0, yes=1)	44.02
Couple household (no=0, yes=1)		0.499	0.500	0	1
Household size		2.024	1.141	1	18
Works (no=0, yes=1)		0.358	0.479	0	1
Receiving social security (no=0, yes=1)		0.632	0.482	0	1
Receiving private pension (no=0, yes=1)		0.308	0.461	0	1
Household annual income (2012 US$)		55,821	248,075	0	60,000,000
Household annual income, tertiles
Tertile 1	≤19,176
Tertile 2	19,177–47,935
Tertile 3	>47,936
Household wealth (2012 US$)		404,115	1,126,671	−2,245,500	90,700,000
Household wealth, tertiles		2.045	0.810	1	3
Tertile 1	≤48,000
Tertile 2	48,001–262,300
Tertile 3	>262,300
Home ownership
No home ownership (no=0, yes=1)	23.51
Owns home with mortgage (no=0, yes=1)	48.29
Owns home without mortgage (no=0, yes=1)	28.20
Owns vehicles (no=0, yes=1)		0.829	0.376	0	1
Has debt (no=0, yes=1)		0.301	0.459	0	1
Overall health index (standardized)		0.021	0.985	−3.966	1.724
Cognitive ability index (standardized)		0.022	0.973	−2.841	2.253
Cognitive ability index, quartiles
Quartile 1	≤−0.665
Quartile 2	−0.666 to 0.103
Quartile 3	0.104–0.765
Quartile 4	>0.765
Nativity, five regions
United States	94.57
Mexico	1.16
Central and South America, Cuba	1.30
Caribbean	0.63
Other	2.35
Citizenship (no=0, yes=1)		0.951	0.216	0	1
SSN consent
Missing (no=0, yes=1)	21.35
Refuse (no=0, yes=1)	24.46
Provide (no=0, yes=1)	54.20
Spanish survey (no=0, yes=1)		0.028	0.166	0	1

Notes: N =74,945. All individual-level variables refer to the household Financial Respondent (FR). N =37,454 for percentages of race/ethnicity for couple households. Data on nativity was obtained from restricted version of HRS. Source: HRS (2017), waves 2000, 2002, 2004, 2006, 2008, 2010, and 2012.

We further expand our empirical specification with measures of neighborhood environments, such as socioeconomic characteristics of residents and number of existing formal financial institutions. As mentioned above, these variables account for supply-side factors and peer/community effects that are bound to influence individuals’ access to formal financial services. For this purpose, we rely on restricted HRS geocodes to obtain the census tract where respondents reside at the time of the interview. Following Diez Roux et al. (2001), we construct a neighborhood SES index at the census tract level. Additionally, we create a variable recording the number of financial institutions in the census tract where HRS respondents reside at the time of the interview. We also consider in our analysis the racial/ethnic composition and primary language spoken by residents at the census tract level. Summary statistics for neighborhood-level variables are present in Table 2. Full details for individual-, household-, and neighborhood-level variable construction and sources are available in Table A1 online. Table A2 (online) presents the distribution of race and ethnicity in our sample by wave, while Table A3 (online) shows summary statistics by race/ethnicity for selected variables.

TABLE 2.

Neighborhood-Level Variables: Summary Statistics

	Mean	SD	Minimum	Maximum
NSES index	−0.08	0.94	−3.83	3.54
Number of financial institutions (weighted by population)	0.02	0.06	0	3.45
Blacks residents (%)	16.52	26.23	0	100
Hispanics residents (%)	12.65	19.89	0	100
Whites residents (%)	73.33	27.73	0	100
English language (%)	82.44	18.56	0	100
Spanish language (%)	10.95	17.13	0	100

Notes: N =74,945, unless stated otherwise. NSES index is the neighborhood socioeconomic status index constructed as the principal component of median household income, median household value, percentage of households with interest income, percentage of residents with high school, percentage of residents with college, and percentage of residents in managerial positions (we follow the methodology of Diez Roux et al. 2001). All variables correspond to the census tract where HRS households reside at the time of the HRS interview. Source: Components of NSES index and the percentage of blacks, Hispanics, and whites residents, and English and Spanish languages are at the census tract level and taken from the United States Census of 2000 and the American Community Survey (ACS) of 2009, 2010, and 2012. Number of financial institutions at the census tract level was provided by the Federal Deposit Insurance Corporation, to the authors in 2015. We use population benchmarks from Census and ACS for constructing the weighted number of financial institutions.

METHODOLOGY

The decision to own a bank account can conveniently be given a latent-variable interpretation. Indicate the net utility of owning a bank account for individual i at time t with

$$u_{it}^{*}={\beta }_0+{\beta }_1{z}_i^{\prime }+{\beta }_2{x}_{it}^{\prime }+{\eta }_i+{\epsilon }_{it},$$ (1)

where z represents time-invariant characteristics, among others the individual’s racial/ethnic group, and the vector x includes time-varying individualand household-level variables as well as environmental factors, such as SES neighborhood characteristics and availability of financial institutions in the place of residence. The term η_i represents unobserved, time-invariant individual preference/taste for financial inclusion while ε_it is an idiosyncratic shock assumed to be orthogonal to the explanatory variables.

Utility-maximizing individuals choose to own a bank account if and only if the net utility they get is greater than the one they would obtain under the alternative scenario of being unbanked. Denoting our indicator of financial inclusion with y, we would therefore observe

$$y_{it}=\{\begin{array}{l}1\text{if}{u}_{it}^{*}>0\\ 0\text{if}{u}_{it}^{*}\le 0\end{array},$$ (2)

where normalizing the net utility of being unbanked to 0 is without loss of generality as long as an intercept is included in equation 1. Assuming that

$$E\left[{\eta }_i|z_i,x_{it}\right]=0,$$ (3)

the binary, latent-variable model described by equations 1 and 2

$$y_{it}={\beta }_0+{\beta }_1{z}_i^{\prime }+{\beta }_2{x}_{it}^{\prime }+{\eta }_i+{\epsilon }_{it}$$ (4)

can be estimated using a RE linear probability model.

The assumption that unobserved individual preference/taste for bank account ownership is uncorrelated with the explanatory variables of the model is undoubtedly restrictive. Yet, relaxing equation 3 implies that the parameters of the model cannot be consistently estimated unless the time-invariant individual heterogeneity represented by η_i is removed. This can be done via fixed-effects (FE) estimation, which, however, comes at the cost of not being able to identify the effect of time-invariant covariates on the dependent variable. For the purposes of our study, this is a major drawback as our interest lies in assessing the racial/ethnic gap in financial inclusion. Confronted with a similar situation, Rhine and Greene (2006) opt for maintaining assumption (3) and estimate the model in equation 4 by RE probit. We follow a different approach.

The RE and FE models can be unified by explicitly modeling the relationship between η_i and x_it, and therefore, by somehow relaxing the assumption in equation 3. Specifically, if ${\eta }_i=\gamma {\overline{x}}_i^{\prime }+v_i$, where ${\overline{x}}_i$ denotes individual-specific time averages and $E\left[v_i|{\overline{x}}_i\right]=0$, equation 4 can be rewritten as:

$$y_{it}={\beta }_0+{\beta }_1{z}_i^{\prime }+{\beta }_2{x}_{it}^{\prime }+\gamma {\overline{x}}_i^{\prime }+v_i+{\epsilon }_{it}.$$ (5)

This is known as a CRE model (Wooldridge 2010). As originally shown by Mundlak (1978), estimating β₂ in equation 5 by RE is equivalent to estimating β₂ in equation 4 by FE. Such result arises because ${\overline{x}}_i$ takes into account the correlation between the time-varying variables x_it and the time-invariant, unobserved individual preferences/tastes η_i. Moreover, since equation 5 is treated as a RE model, it is possible to estimate the effect of the time-invariant covariates z_i, which would be otherwise wiped out by a within/FE transformation. It should be noted that for the estimate of β₁ to be consistent, the condition $E\left[v_i|z_i,x_{it}\right]=0$ still needs to hold. While this may not be a serious concern for our main variable of interest—race/ethnicity—the assumption that individual heterogeneity is uncorrelated with other time-invariant variables in our data, like education, may appear too restrictive. For this reason, we must use caution when interpreting the estimate of β₁. Nonetheless, the presence of ${\overline{x}}_i$ ensures that the parameter estimates for β₁ are net of the correlation between η_i and x_it.

We carry out our empirical analysis using both the RE and the CRE estimation approaches, but focus our discussion on the CRE results. The RE estimation approach makes our results more directly comparable to those of previous studies on the topic that have commonly estimated RE models (Lusardi 2005; Rhine and Greene 2006). The CRE estimation approach allows specifying a less restrictive within-individual error structure and to obtain estimates of time-invariant variables that are corrected for individual heterogeneity as measured by across-individual differences in time-varying covariates. Moreover, by comparing the results of the RE and CRE estimations, we can assess the extent to which controlling more encompassingly for individual heterogeneity in our model affects the size of the estimated racial/ethnic gap in bank account ownership.

RESULTS

Bank Account Ownership by Race and Ethnicity of the Household FR

Table 3 shows the percentage of households who do not own a bank account by race and ethnicity of the FR and, separately, for each survey year. As can be seen, there exists a significant racial/ethnic gap in bank account ownership. During the 2000–2012 period, while only 9% of non-Hispanic white households do not own a bank account, 35% and 38% of black and Hispanic households are unbanked, respectively. Interestingly, the fraction of unbanked households has increased significantly for all groups around the time of the Great Recession (December 2007–June 2009) and remained somewhat higher than the precrisis level, especially among Hispanics. The unbanked rates in our sample, which only includes individuals aged 51–90, are higher than those for the adult population as a whole (at least 18 years of age). Indeed, according to the National Survey of the Unbanked and Underbanked Households, the unbanked rates among white, black, and Hispanic households in 2011 are 4, 20, and 21, respectively (Federal Deposit Insurance Corporation 2011), hence, much lower than those observed in the HRS and reported in Table 3 for either 2010 or 2012. In view of these differences, the results of our analysis should not be generalized to the entire adult population. They are, instead, specific of middle-age and older adults and illustrative of the barrier toward financial inclusion faced by minorities within this segment of the population. As argued above, given the increasing share of minorities in the United States and the rapid aging of the population, studying the banking situation of minority older adults is an important contribution of this paper, especially because minority older adults have not been the focus of previous work on the topic of financial inclusion. When looking at cross-wave transitions (Table A4, online), we observe that blacks and Hispanics are substantially more likely than whites to pass from having to not having a bank account (13 and 18 percentage points, respectively) and less likely to be banked if they were unbanked in the previous wave (29 and 31 percentage points, respectively).

TABLE 3.

Percentage of FRs 51 Years and Older Who Do Not Own a Bank Account at the Household Level by Race and Ethnicity

	2000	2002	2004	2006	2008	2010	2012	2000–2012
White (%)	10	7	7	8	7	9	11	9
Black (%)	39	34	32	30	30	37	38	35
Hispanic (%)	32	31	37	35	38	42	46	38
Observations	10,753	10,063	10,930	10,601	10,035	11,456	11,107	74,945

Notes: Ownership of a bank refers to whether the household owns a checking, saving, or money market account. Percentages are calculated using only the household FR. We use year-specific weights to compute percentages in each year. For the percentage during the period 2000–2012, we use a base weight, which is the weight of the individual when it entered the survey. Source: Data obtained from the HRS (2017) for the years 2000, 2002, 2004, 2006, 2008, 2010, and 2012.

We estimate the CRE model in equation 5 using seven specifications. Starting from a conventional baseline model, we sequentially add variables measuring economic status, health, cognitive ability, and cultural barriers to bank account ownership facing minority older adults. Table 4 shows the residual gap across progressively richer specifications, as measured by the coefficients on the indicators for whether the household FR is either black or Hispanic (whites are the reference group).⁷

TABLE 4.

Determinants of Ownership of Bank Account, Race, and Ethnicity for FR – CRE Model

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Black	−0.1481***	−0.1341***	−0.1189***	−0.1206***	−0.1196***	−0.1227***	−0.1202***
	(0.0052)	(0.0051)	(0.0052)	(0.0052)	(0.0053)	(0.0052)	(0.0053)
Hispanic	−0.1541***	−0.1389***	−0.1287***	−0.1177***	−0.1172***	−0.0847***	−0.0830***
	(0.0076)	(0.0075)	(0.0075)	(0.0083)	(0.0083)	(0.0088)	(0.0088)
Home owner, no mortgage		−0.0081	−0.0085	−0.0084	−0.0085	−0.0086	−0.0080
		(0.0064)	(0.0064)	(0.0064)	(0.0064)	(0.0064)	(0.0064)
Home owner, with mortgage		0.0024	0.0020	0.0021	0.0020	0.0019	0.0024
		(0.0067)	(0.0067)	(0.0067)	(0.0067)	(0.0067)	(0.0067)
Vehicle ownership		0.0648***	0.0640***	0.0640***	0.0639***	0.0640***	0.0641***
		(0.0055)	(0.0055)	(0.0055)	(0.0055)	(0.0055)	(0.0055)
Debt ownership		0.0356***	0.0359***	0.0359***	0.0359***	0.0358***	0.0359***
		(0.0033)	(0.0033)	(0.0033)	(0.0033)	(0.0033)	(0.0033)
Overall health index			0.0067***	0.0067***	0.0067***	0.0068***	0.0068***
			(0.0025)	(0.0025)	(0.0025)	(0.0025)	(0.0025)
Cognition, quartile 2			0.0109***	0.0109***	0.0109***	0.0109***	0.0109***
			(0.0042)	(0.0042)	(0.0042)	(0.0042)	(0.0042)
Cognition, quartile 3			0.0164***	0.0164***	0.0164***	0.0165***	0.0166***
			(0.0047)	(0.0047)	(0.0047)	(0.0047)	(0.0047)
Cognition, quartile 4			0.0193***	0.0193***	0.0193***	0.0196***	0.0195***
			(0.0055)	(0.0055)	(0.0055)	(0.0055)	(0.0055)
Foreign born, Mexico				−0.1492***	−0.1492***	−0.0525*	−0.0513*
				(0.0284)	(0.0284)	(0.0295)	(0.0295)
Foreign born, SA, CA, Cuba				−0.0269	−0.0271	0.0534*	0.0532*
				(0.0283)	(0.0283)	(0.0290)	(0.0290)
Foreign born, Caribbean				0.0246	0.0248	0.0599*	0.0601**
				(0.0306)	(0.0306)	(0.0306)	(0.0306)
Foreign born, other				−0.0313	−0.0303	−0.0256	−0.0255
				(0.0220)	(0.0220)	(0.0220)	(0.0220)
No citizenship				−0.0393*	−0.0388*	−0.0344	−0.0333
				(0.0232)	(0.0232)	(0.0231)	(0.0231)
SSN, refuse to provide					−0.0070	−0.0070	−0.0070
					(0.0054)	(0.0054)	(0.0054)
SSN, provided					−0.0021	−0.0021	−0.0022
					(0.0044)	(0.0044)	(0.0044)
Spanish survey						−0.1628***	−0.1625***
						(0.0141)	(0.0141)
NSES index							0.0046**
							(0.0021)
Ln(Finc. inst. Weig.)							0.0007
							(0.0008)
R2, overall	0.223	0.239	0.244	0.245	0.245	0.247	0.248

Notes: No. of obs.=74,945, No. of ind.=18,582. CA is Central America and SA is South America. Estimated coefficients for male, education, age, age squared, couple, household size, income and wealth tertiles, employment, receive social security benefits and pension income, and mean of all time variant variables (except NSES index and financial institutions), year and state dummies and constant are omitted in the interest of space. See Table A5 for full set of estimates.

^*Significance p <.1.

^**Significance p<.05.

^***Significance p<.01.

In our baseline specification, blacks and Hispanics are less likely to own a bank account than whites by 15 percentage points. As we account for additional potential barriers to bank account ownership, the residual gap for minorities is reduced, although to a different extent for blacks and Hispanics. In our full model, blacks and Hispanics are less likely to own a bank account than whites by 12 and 8 percentage points, respectively. Thus, compared to the baseline model, the estimated residual gap is cut by 19% for blacks and 46% for Hispanics.

Among the determinants of the gap, ownership of real assets accounts for about 1.5% age points for both minority groups. Together, physical and cognitive health account for an additional 1 percentage point of the racial/ethnic gap. Indicators of foreign nativity do not appear to explain the residual gap in financial inclusion for minority groups. Taste for privacy, proxied by the consent to share the SSN with HRS, does not correlate with bank account ownership. We acknowledge that this measure is imperfect, not only due to the large proportion of missing values but also because it may reflect reluctance of illegal immigrants to provide their SSN (if available). Lack of English proficiency, as measured by whether the HRS questionnaire was answered in Spanish, is the factor contributing the most to the gap in bank ownership for Hispanics (approximately 4 percentage points above and beyond the contribution of all the other variables described so far). This is in line with a recent report of the Federal Deposit Insurance Corporation, according to which Spanish-only speaking households are five times less likely to use a bank or credit union (Federal Deposit Insurance Corporation 2016). Lack of English proficiency may also more broadly be interpreted as lower cultural assimilation. This, in turn, may lead Hispanic immigrants to be unfamiliar with the US financial system and/or to mistrust financial institutions based on experiences in their countries of origin (Osili and Paulson 2008).

Characteristics of the neighborhood where the individual currently resides, such as a neighborhood SES index and number of financial institutions (weighted by population), do not explain much of the observed White/minority gap in financial inclusion beyond and above what is already accounted for by individual-level variables. It is worth noting that the contribution of the aforementioned variables to explaining the observed minority gap in financial inclusion is only marginally affected by the order in which the variables are entered in the model.

We obtain similar results as those of Rhine and Greene (2006), who, relying on a RE probit model, find that US-born blacks and Hispanics are 13 and 8 percentage points more likely to be unbanked than US born Whites, respectively. In our full model estimated by RE (Table A6 [online], column 7), the gap for blacks is 14 percentage points and the one for Hispanics is 10 percentage points. These estimated gaps are reduced to 12 and 8 percentage points, respectively, when estimation is carried out by CRE, which allows to better control for unobserved individual heterogeneity in financial decisions.

The CRE model also delivers FE estimated coefficients of time-varying variables, which help identify the factors that contribute to changes in the decision to own a bank account. Among them, health and cognition play an important role. One SD increase in the health index is associated with a 0.7 percentage-point increase in the likelihood of bank account ownership. This implies that the average individual moving from the top 5% to the bottom 5% of the health index distribution would be approximately 3 percentage points less likely to have a bank account. Cognition is also positively associated with financial inclusion. Compared to those in the first cognition index quartile, individuals in the second, third, and fourth quartiles are from 1 to 2 percentage points more likely to own a bank account. It is worth noting that minorities are about 30 percentage points more likely than whites to be in the bottom 25% of the cognition distribution and 15 percentage points less likely to be in the top 25% of the distribution. Consequently, they exhibit significantly lower propensity to be banked than Whites and, as discussed above, cognition is a significant contributor of this gap. There exist apparent income and wealth gradients. Compared to households in the first income tercile, those in the second and third are 2.7 and 3.4 percentage points more likely to hold a bank account, respectively. Similarly, compared to households in the first wealth tercile, those in the second and third are 4.6 and 7.2 percentage points more likely to be banked, respectively. Moving in and out of work also affects the probability of bank account ownership. Interestingly, receiving a pension or social security benefits has a positive impact on financial inclusion. This may be related to the fact that often electronic payments of benefits are strongly encouraged (for social security benefits, they became mandatory starting from March 2013).

Bank Account Ownership by Race and Ethnicity of the Couple

We delve more into the household financial decision making process and explore potential spousal effects on the choice of being unbanked. For this purpose, we restrict attention to couple households and create groups interacting race/ethnicity of the household FR and his/her spouse or partner. Specifically, we separate the sample into (1) white couples, where both members are white; (2) white-minority couples, where one member is either black or Hispanic (“mixed” couples); and (iii) minority couples, where both members are either black or Hispanic (Table A7, online, shows summary statistics for race/ethnicity at the couple level).

Table 5 presents the estimated coefficients for the white-minority and minority couple indicators (white couples are the reference group) across our seven specifications. The results show that minority couples are from 10 to 7 percentage points less likely than “mixed” couples to own a bank account. In our baseline specification, minority couples are 13 percentage points less likely to own a bank account than white couples. The residual gap decreases steadily as more hurdles to financial inclusion are considered in the regression equation. Compared to the baseline specification, the residual gap is about 3 percentage points (26% decrease) lower in the full model. Again, ownership of real assets, health, cognitive ability, and language barriers are the factors contributing the most to the reduction of the gap. In the last column of Table 5, the effect of language barriers is less pronounced than the one observed in Table 4 as minority couples are not just Hispanic couples for whom lack of English proficiency is most relevant. Notably, the gap in bank account ownership between “mixed” and White couples is small and remains virtually constant across model specifications (3 percentage points less than White couples).

TABLE 5.

Determinants of Ownership of Bank Account, Race, and Ethnicity for FR and Spouse (S) – CRE Model

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Mixed couple	−0.0335***	−0.0303**	−0.0283**	−0.0286**	−0.0287**	−0.0298**	−0.0288**
	(0.0123)	(0.0121)	(0.0120)	(0.0120)	(0.0120)	(0.0120)	(0.0120)
Minority couple	−0.1333***	−0.1195***	−0.1085***	−0.1074***	−0.1063***	−0.1010***	−0.0985***
	(0.0064)	(0.0063)	(0.0065)	(0.0067)	(0.0067)	(0.0067)	(0.0068)
Home owner, no mortgage		−0.0108	−0.0117	−0.0117	−0.0118	−0.0122	−0.0117
		(0.0102)	(0.0102)	(0.0102)	(0.0102)	(0.0102)	(0.0102)
Home owner, with mortgage		−0.0021	−0.0028	−0.0028	−0.0030	−0.0033	−0.0029
		(0.0102)	(0.0102)	(0.0102)	(0.0102)	(0.0102)	(0.0102)
Vehicle ownership		0.0748***	0.0740***	0.0740***	0.0740***	0.0741***	0.0742***
		(0.0095)	(0.0095)	(0.0095)	(0.0095)	(0.0095)	(0.0095)
Debt ownership		0.0349***	0.0352***	0.0352***	0.0351***	0.0351***	0.0351***
		(0.0042)	(0.0042)	(0.0042)	(0.0042)	(0.0042)	(0.0042)
Overall health index			0.0068*	0.0068*	0.0069*	0.0069*	0.0069*
			(0.0035)	(0.0035)	(0.0035)	(0.0035)	(0.0035)
Cognition, quartile 2			0.0184***	0.0184***	0.0184***	0.0183***	0.0183***
			(0.0059)	(0.0059)	(0.0059)	(0.0059)	(0.0059)
Cognition, quartile 3			0.0228***	0.0228***	0.0228***	0.0228***	0.0229***
			(0.0064)	(0.0064)	(0.0064)	(0.0064)	(0.0064)
Cognition, quartile 4			0.0245***	0.0245***	0.0245***	0.0246***	0.0246***
			(0.0073)	(0.0073)	(0.0073)	(0.0073)	(0.0073)
Foreign born, Mexico				−0.1542***	−0.1530***	−0.0586	−0.0584
				(0.0356)	(0.0356)	(0.0373)	(0.0373)
Foreign born, SA, CA, Cuba				−0.0059	−0.0047	0.0855**	0.0837**
				(0.0353)	(0.0353)	(0.0369)	(0.0369)
Foreign born, Caribbean				0.0048	0.0065	0.0394	0.0393
				(0.0411)	(0.0411)	(0.0412)	(0.0412)
Foreign born, other				−0.0506*	−0.0483*	−0.0468*	−0.0471*
				(0.0283)	(0.0283)	(0.0282)	(0.0282)
No citizenship				−0.0431	−0.0409	−0.0401	−0.0399
				(0.0302)	(0.0302)	(0.0302)	(0.0302)
SSN, refuse to provide					−0.0052	−0.0053	−0.0053
					(0.0073)	(0.0073)	(0.0073)
SSN, provided					−0.0009	−0.0009	−0.0009
					(0.0060)	(0.0060)	(0.0060)
Spanish survey						−0.1409***	−0.1409***
						(0.0168)	(0.0168)
NSES index							0.0042
							(0.0026)
Ln(Finc. inst. Weig.)							0.0013
							(0.0010)
R2, overall	.189	.202	.205	.207	.207	.208	.209

Notes: No. of obs.=37,454, No. of ind.=9,850. Coefficients for mixed and minority couple indicators are reported (White couples are the reference group). Mixed couples are defined as those where one member is black or Hispanic and the other person is White; minority couples are defined as those where both members are either black or Hispanic; and White couples are defined as those where both members are White. CA is Central America and SA is South America. Estimated coefficients for male, education, age, age squared, couple, household size, wealth and income tertiles, employment, receive social security benefits and pension income, mean of all time variant variables (except NSES index and financial institutions), year and state dummies and constant are omitted in the interest of space.

^*Significance p<.1.

^**Significance p<.05.

^***Significance p<.01.

Since in our regressions we do not control for spouses’ specific characteristics (e.g., education and cognitive ability), we cannot rule out the possibility that these results are determined by assortative mating. For instance, conditional on education of the FR, spouses in white, mixed, and minority couples may have different levels of education or financial literacy, leading to the observed differences in the adoption of bank accounts (as long as, at least to some extent, financial decisions are made jointly within the household). To shed further light on this issue, we repeat the exercise in Table 5 restricting attention only to couples where both members have the same level of education. The results of these regressions (reported in Table A8) are qualitatively and quantitatively unchanged. For minority couples, the gap in bank ownership relative to White couples is about 14 percentage points in the most parsimonious specification and it is reduced to 11% age points in the richest specification. In comparison, the gap between “mixed” and White couples is about 3 percentage points across all models. This suggests that assortative mating, at least on the basis of education, does not play a major role in driving the findings in Table 5.⁸

A potential mechanism for the lower gap of mixed couples estimated in Table 5 is that the spouse from the dominant race is also the one making financial decisions. To gain insight on this point, we distinguish between mixed couples with a white FR and mixed couples with a minority FR. The results of this additional analysis (presented in Table A9, online) show that the gap in bank account ownership relative to White couples is only 2 percentage points and not statistically significant for mixed couples with a White financial decision maker. For mixed couples with a minority financial decision maker the gap is 4 percentage points and statistically significant, but still 6 percentage points lower than that of minority couples. Together with the empirical evidence described above, this suggests that spillovers within mixed couples, by which the minority financial decision maker assimilates the financial culture and/or learns financial literacy from the spouse, may help reduce barriers toward financial inclusion faced by only minority households.

The Role of Household Financial Resources and Neighborhood Characteristics

We further investigate the role of household financial resources and neighborhood characteristics by estimating our model separately for those with income, wealth, and neighborhood socioeconomic status (NSES) index below and above the sample median. Table 6 presents the residual gap in bank ownership estimated using our richest specification separately for the aforementioned groups (Table A10, online, has the full set of estimates). As can be seen from panel A of Table 6, for blacks below median income, the probability of owning a bank account is 14 percentage points lower than for their white counterparts. This gap is reduced to 8 percentage points among older blacks above median income. The null hypothesis that the gap is the same for both income groups is rejected at 1% significance level in favor of the alternative that it is larger for blacks below median income. The differential gap in bank account ownership between Hispanics below and above median income is less pronounced and approximately equal to 2.5% age points, but still statistically significant.

TABLE 6.

Determinants of the Minority Gap in Bank Account Ownership by SES Group (Race/Ethnicity of the Household FR)

	Income (I)
	I<median	I≥median	χ2 test
Panel A
Black	−0.1411***	−0.0779***	70.7
	(0.0072)	(0.0066)	[0.00]
Hispanic	−0.0952***	−0.0697***	4.89
	(0.0121)	(0.0106)	[0.02]
	Wealth (W)
	W<median	W≥median	χ2 test
Panel B
Black	−0.1283***	−0.0624***	41.89
	(0.0071)	(0.0069)	[0.00]
Hispanic	−0.0880***	−0.0508***	4.65
	(0.0122)	(0.0103)	[0.03]
	NSES
	NSES<median	NSES≥median	χ2 test
Panel C
Black	−0.1166***	−0.1012***	5.26
	(0.0070)	(0.0077)	[0.02]
Hispanic	−0.0823***	−0.0812***	0.12
	(0.0124)	(0.0113)	[0.72]

Notes: No. of obs.=74,945, No. of ind.=18,582. Coefficients for black and Hispanic indicators are reported (whites are the reference group). They are estimated using the specification in column (7) of Table 4 and separately for the different groups indicated in the table. Standard errors are in parentheses. We report a χ² test (with p-value in brackets) for the equality of coefficients between groups. See Table A10 for full set of estimates.

^*Significance level p<.1.

^**Significance level p<.05.

^***Significance level p<.01.

In panel B of Table 6, we estimate sizable (4–6 percentage points) and statistically significant differential gaps by wealth for both minority groups. Blacks below median wealth are 13 percentage points less likely than whites to own a bank account. This difference shrinks to 6 percentage points among blacks above median wealth. Similarly, Hispanics with wealth below and above the median are 9% and 5% age points more likely to be unbanked than whites, respectively. The null hypothesis of no differential residual gap by wealth is rejected at 5% significance level against the alternative that the gap is larger for households below than above median wealth for both minority groups.

In panel C, we separate respondents living in low and high SES neighborhoods. For blacks residing in neighborhoods with below-median SES, the estimated bank ownership gap with respect to their white counterparts is approximately 1 percentage point larger than for those living in above-median SES neighborhoods, and this difference is statistically significant at the 5% significance level. There is, instead, no apparent difference for Hispanics. These findings suggest a limited role of community spillovers and supply-side factors in explaining the minority gap in bank account ownership, above and beyond the influence of individual-level SES and circumstances. At the same time, the NSES index may mask heterogeneity in language and cultural barriers faced by minority older adults, which appear to be the major drivers of lack of participation in the formal financial sector.

Therefore, we explore the role of specific neighborhood factors, such as the racial and ethnic composition as well as the primary language spoken at the census tract where sample households reside. Table 7 shows the results when we add to the full model the percent of blacks, Hispanics, and whites (columns 1 and 2) and the percent of those who primarily speak Spanish (columns 3 and 4) in the census tract. Holding these variables constant does not reduce the gap for blacks and only marginally affects the gap for Hispanics (by less than 1 percentage point). Notably, the main contributor to explaining the Hispanic-white gap is still an individual’s lack of English proficiency, which does not seem to be proxied by neighborhood indicators for race/ethnicity and language at the census tract level. Indeed, when (in columns 2 and 4 of Table 7) we control for neighborhood racial/ethnic composition and main language, and we do not take into account whether the HRS survey was answered in Spanish, the estimated gap for Hispanics is slightly above 10 percentage points. It is, instead, reduced to 7.5 percentage points when the Spanish survey indicator is added to the set of regressors.

TABLE 7.

Determinants of the Minority Gap in Bank Account Ownership Controlling for Race/Ethnicity and Language in Neighborhood (Race/Ethnicity of the FR)

	(1)	(2)	(3)	(4)
Black	−0.1176***	−0.1143***	−0.1202***	−0.1173***
	(0.0065)	(0.0065)	(0.0053)	(0.0053)
Hispanic	−0.0758***	−0.1026***	−0.0757***	−0.1015***
	(0.0092)	(0.0089)	(0.0092)	(0.0089)
Spanish survey	−0.1569***		−0.1561***
	(0.0142)		(0.0143)
Blacks (%)	−0.0003	−0.0004*
	(0.0002)	(0.0002)
Hispanics (%)	−0.0004***	−0.0006***
	(0.0001)	(0.0001)
Whites (%)	−0.0002	−0.0003
	(0.0002)	(0.0002)
English language (%)			−0.0003	−0.0004
			(0.0002)	(0.0002)
Spanish language (%)			−0.0007**	−0.0010***
			(0.0003)	(0.0003)
R2, overall	0.248	0.246	0.248	0.246

Notes: No. of obs.=74,945, No. of ind.=18,582. Other regressors are as in column (7) of Table 4. The full set of estimated coefficients is available upon request.

^*Significance p<.1.

^**Significance p<.05.

^***Significance p<.01.

While neighborhood racial composition and language seem to play a limited role in explaining the overall minority gap above and beyond household-level characteristics, one would still expect the gap to be wider in areas with larger concentrations of minorities, other things equal. To shed light on this point, we estimate our full model (including a control for whether the HRS questionnaire was answered in Spanish) for individuals who reside in census tracts with below and above median percentage of Hispanic residents (panel A of Table 8) as well as with below and above median percentage of Spanish speakers (panel B of Table 8). As can be seen, the observed gap between Hispanics and whites is driven by those households residing in neighborhoods with a relatively high representation of Hispanics and, therefore, Spanish-speaking individuals. Specifically, while there is virtually no noticeable residual gap in census tracts with below-median percentage of Spanish-speaking individuals, Hispanics are as much as 8 percentage points less likely than whites to own a bank account in census tracts with above-median percentage of Spanish-speaking individuals. As mentioned above, lack of English proficiency, which may hinder one’s ability to navigate the US financial system, is plausibly the manifestation with other cultural and behavioral barriers associated with little assimilation/acculturation. The results in Table 8 substantiate the expectation that such barriers are more easily formed and more difficult to overcome in relatively more segregated areas.

TABLE 8.

Determinants of the Minority Gap in Bank Account Ownership by Race/Ethnicity and Language at the Census Tract Group (Race/Ethnicity of the FR)

	Hispanics in census tract, percentage (H)
	H<median	H≥median	χ2 test
Panel B
Black	−0.1225***	−0.1223***	0.01
	(0.0070)	(0.0074)	[0.90]
Hispanic	−0.0306	−0.0857***	1.99
	(0.0245)	(0.0096)	[0.15]
	Spanish language census tract, percentage (S)
	S<median	S≥median	χ2 test
Panel B
Black	−0.1249***	−0.1201***	0.03
	(0.0071)	(0.0071)	[0.86]
Hispanic	−0.0095	−0.0870***	5.07
	(0.0234)	(0.0097)	[0.02]

Notes: No. of obs.=74,945, No. of ind.=18,582. Coefficients for black and Hispanic indicators are reported (Whites are the reference group). They are estimated using the specification in column (7) of Table 4 and separately for the different groups indicated in the table. Standard errors are in parentheses. We report a χ² test (with p-value in brackets) for the equality of coefficients between groups.

^*Significance p<.1.

^**Significance p<.05.

^***Significance p<.01.

Among these barriers, legal citizenship could represent a significant hurdle toward financial inclusion. In neighborhoods with a large representation of Hispanics and where Spanish is the predominant language, the number of illegal immigrants is likely higher and this may contribute to the differences estimated in Table 8.⁹ To gauge the importance of this potential contribution, one would ideally run the same regressions as above for census tracts with relatively low and high percentage of illegal residents. But, obviously, there is no official count of undocumented individuals.¹⁰ It should be noted that legal status is not a necessary condition to be banked. For example, major US banks have been accepting the matricula consular (an ID card from the Mexican consulate issued to undocumented immigrants) to open an account. Legal status, however, might add to the demand-side factors preventing financial inclusion of minorities, as mistrust in banking institutions may keep undocumented individuals in the cash economy.

Unfortunately, we do not have data to investigate to what extent language barriers may also be a supply-side phenomenon stemming, for instance, from the shortfall of services in Spanish provided by financial institutions. Developments in the banking system in the last couple of decades, however, would suggest that this is unlikely to be the case. A 2006 report by the Comptroller of the Currency Administrator of National Banks (2006), which regulates and supervises national banks, found that many banks have progressively expanded their provision of multilingual services. The largest increase has been observed for services in Spanish. This is not surprising as Hispanics have been the fastest growing minority group in the United States and have greatly gained in terms of purchasing power, thereby representing a source of retail growth for financial institutions. Major national banks now offer services in Spanish both online and in most branches, while credit unions increasingly reach out to the Hispanic community via targeted advertising material, pamphlets, and posters in Spanish.

Limitations

Our analysis contributes to the existing literature by providing further insights onto the racial/ethnic gap in bank account ownership. To this end, we exploit richer data than previously available to other authors and rely on a CRE estimation approach that, at least partially, accounts for individual heterogeneity. We recognize the following limitations of our study. First, we cannot rule out potential biases stemming from unobservable characteristics not captured by individual-specific averages of time-varying variables. Hence, we must warn against considering our estimates as causal. Second, and as already mentioned above, we must be cautious in interpreting the observed differences among white, mixed, and minority couples as evidence of spousal/peer effects on bank account ownership. To the extent that the CRE model does not fully control for it, assortative mating on unobservable preferences/tastes for financial decision making might drive our results. For future research, we warrant including parental SES and early-life family background characteristics to proxy for individuals’ attitude toward managing personal finances. Third, we only look at the impact of current neighborhood characteristics on the likelihood that older adults own a bank account. Clearly, early-life environmental circumstances are bound to be important as well in this domain, as they might shape wealth accumulation patterns and determine cultural integration. Unfortunately, the HRS does not provide data on early-life residence. This information would constitute a very valuable addition to gauge the potential role of neighborhood effects more generally.

CONCLUSION

We investigate the determinants of the racial/ethnic gap in bank account ownership among older adults in the United States. Exploiting the wealth of information available in the HRS and its longitudinal dimension, we estimate CRE models for the probability of owning a bank account featuring a comprehensive range of household- and neighborhood-level explanatory variables. As additional potential barriers to financial inclusion are accounted for, the residual gaps between blacks and whites and between Hispanics and whites are reduced by 18% and 46%, respectively. We find that citizenship and “taste for privacy,” as measured by lack of consent to share SSN with HRS, play a limited role for both minority groups. On the other hand, real asset ownership, health, and cognitive ability contribute substantially to the racial/ethnic gap. Conditional on household-level SES measures, the effect of neighborhood socioeconomic characteristics, including the number of available financial institutions, is rather minor, although somewhat more pronounced for blacks. On the other hand, we find that the Hispanic-white gap is significantly larger (8 percentage points) in neighborhoods with above-median percentage of Spanish-speaking individuals and negligible (about 0) in neighborhoods with below-median percentage of Spanish-speaking individuals. This points at the lack of English proficiency, which plausibly is the manifestation of little acculturation/assimilation, as one of the most important barriers toward financial inclusion for older Hispanic households. As noted by a recent Consumer Financial Protection Bureau’s (2017, 9) report “there is a general understanding that language cannot always be separated from culture, and that offering information in languages other than English does not necessarily mean that all Limited English Proficient populations will be able to access, understand, or use the information provided.” The same report calls for more extensive educational campaign to reach out and engage underserved populations, especially minorities (Consumer Financial Protection Bureau 2017).

Our results also offer suggestive evidence of spousal effects in financial decisions. Indeed, we estimate a significantly smaller residual gap between “mixed” and white couples than between minority and white couples. We reach similar conclusions when we restrict attention to couples where both members have the same level of education to better control for assortative mating. Our findings may indicate that mixed couples are less concerned with the cultural/psychological barriers facing minority couples.

Our work aims at providing a deeper understanding of the financial behavior of more vulnerable groups, such as minority older adults. This knowledge is crucial for devising interventions that can effectively change financial decision making among households where resources are most scarce and cultural hurdles more difficult to overcome.

Supplementary Material

supplemental-material

Table A1. Description of variable construction and source.

Table A2. Distribution of race and ethnicity by wave (analytic sample, financial respondents only).

Table A3. Summary statistics by racial/ethnic groups, selected variables.

Table A4. Bank account ownership wave-transitions by race and ethnicity (percentages).

Table A5. Determinants of ownership of bank account, race, and ethnicity for financial respondent – correlated random effects (CRE) model.

Table A6. Determinants of ownership of bank account, race, and ethnicity for financial respondent – random effects (RE) model.

Table A7. Distribution of race and ethnicity for financial respondent (FR) and spouse (S) by waves.

Table A8. Determinants of ownership of bank account, race, and ethnicity for financial respondent (FR) and spouse (S) – correlated random effects (CRE) model – reducing sample for households where FR and S have same level of education.

Table A9. Determinants of ownership of bank account, race, and ethnicity for financial respondent (FR) and spouse (S) – correlated random effects (CRE) model – using four categories.

Table A10. Determinants of ownership of bank account, race, and ethnicity for financial respondent by income, wealth, and NSES index below or above or equal median groups – correlated random effects (CRE) model.