What Factors Explain Who Remains Unbanked in the United States?

Calem, Henderson, and Wang (2026) ask which household and state-level factors explain who remains unbanked in the United States. They analyze FDIC household surveys from 2015, 2017, and 2019, each with more than 32,000 respondents, combined with state-level measures. They find that unbanked rates fell from 7.0% in 2015 to 5.4% in 2019, but large disparities remained. In 2019, households earning under $15,000 were more than ten times less likely to be banked than those earning over $75,000. Black households were about two-fifths as likely as White households to be banked, while internet access tripled the likelihood of having an account.

Why This Article Was Selected for The Policy Scientist

Financial inclusion matters because access to basic banking affects households’ ability to manage income, save, absorb financial shocks, and participate in an increasingly digital economy. The topic is especially timely as banking services move online while meaningful disparities in account ownership persist. Calem, Henderson, and Wang make a useful contribution by extending earlier research on income, race, financial literacy, and banking access with large FDIC surveys covering more than 32,000 households per year. The findings have broad relevance across U.S. jurisdictions, although institutional differences limit international generalization. The multilevel and fixed-effects regressions are appropriate for identifying associations, but they do not establish causation. Future research using stronger causal-inference designs or randomized interventions would substantially strengthen conclusions about which policies increase banking participation. Published in AEA Papers and Proceedings, an American Economic Association publication, the study appears in a prominent outlet within economics.

Full Citation and Link to Article

Calem, P., Henderson, C., & Wang, J. (2026). Who remains unbanked in the United States and why? AEA Papers and Proceedings, 116, 294–298. https://www.aeaweb.org/articles?id=10.1257%2Fpandp.20261007

Central Research Question

The study examines who remains without a bank account in the United States and which household and state-level characteristics are most strongly associated with unbanked status. The central concern is not simply whether financial inclusion has improved over time, but why substantial differences remain even as the national unbanked rate has declined. The analysis focuses particularly on the roles of income, race and ethnicity, age, employment, disability, family structure, digital access, immigration status, financial literacy, health insurance coverage, payday lending regulation, and physical access to bank branches.

A related question is whether observed differences across demographic groups can be explained by socioeconomic characteristics or whether important gaps remain after those factors are statistically controlled. This is especially relevant for racial and ethnic differences in bank account ownership. The study also asks whether the factors associated with being unbanked changed between 2015 and 2019 and whether state-level conditions help explain geographic differences in financial inclusion.

Previous Literature

The study builds on several strands of research concerning financial inclusion and bank account ownership. Hayashi and Minhas (2018) provide an important foundation by examining household characteristics associated with unbanked status using 2015 data. Their analysis demonstrated that unbanked status could not be understood solely as a function of income. The present study extends that work by incorporating additional variables, examining multiple survey years, and estimating alternative statistical specifications.

Race and ethnicity have also been prominent in the literature. Cox, Whitten, and Yogo (2025), using administrative tax data, identify race as an important predictor of participation in the banking system. The present analysis examines whether these differences persist after controlling for a broad set of socioeconomic and household characteristics.

Financial literacy represents another major line of inquiry. Barcellos and Zamarro (2019) find that financial literacy is related to whether households maintain bank accounts and use alternative financial services. The current study revisits this relationship using more recent data and state-level measures of educational proficiency as a proxy for financial literacy.

Prior studies have also emphasized the costs and accessibility of banking. Akana and Santucci (2021) show that overdraft services may serve as a form of short-term liquidity insurance for some households, complicating the assumption that bank fees uniformly discourage account ownership. Dlugosz, Melzer, and Morgan (2023) similarly find that restrictions on overdraft fees can affect financial inclusion.

Technological change is another relevant development. Boel and Zimmerman (2022) argue that online banking, smartphones, and financial technology have reduced the importance of proximity to physical bank branches. Morgan, Pinkovskiy, and Perlman (2018) likewise question the importance of so-called banking deserts. These studies provide a basis for evaluating whether physical branch availability continues to explain differences in bank account ownership.

Data

The primary data come from the Federal Deposit Insurance Corporation Survey of Household Use of Banking and Financial Services for 2015, 2017, and 2019. Each survey includes more than 32,000 household respondents, producing a large nationally oriented dataset covering financial participation and a wide range of household characteristics.

The surveys show a clear decline in unbanked status over the study period. Approximately 7.0 percent of households were unbanked in 2015, compared with 6.5 percent in 2017 and 5.4 percent in 2019. This overall improvement provides the context for the study’s emphasis on identifying which groups remained disproportionately likely to be unbanked.

The household survey data include measures of income, educational attainment, marital status, number of dependents, employment, race and ethnicity, mobile-phone access, smartphone ownership, internet access, income variability, citizenship and immigration status, disability, age, and family structure.

The researchers supplement these household data with state-level information from several sources. These measures include Supplemental Nutrition Assistance Program participation, average credit scores, state laws governing check cashing and payday lending, educational proficiency measures used as proxies for financial literacy, and the percentage of state residents lacking health insurance. Combining household- and state-level information allows the analysis to distinguish individual characteristics from broader geographic conditions.

Methods

The statistical analysis proceeds in several stages. The researchers first estimate logistic regression models with state fixed effects. The dependent variable indicates whether a household has a bank account. Household-level characteristics serve as the principal independent variables, while state fixed effects control for unobserved characteristics that differ across states but are common to households within each state.

The researchers then estimate multilevel models that incorporate both household and state characteristics. In their preferred specification, state fixed effects are replaced by measures of educational proficiency, health insurance coverage, and payday lending regulation. These models allow the analysis to evaluate whether measurable state-level factors help explain geographic variation in banking participation.

Population weights supplied with the FDIC surveys are used so that the estimates more closely represent the national household population. Standard errors are clustered by state. Models are estimated separately for 2015, 2017, and 2019, permitting comparisons over time.

Additional analyses pool the 2015 and 2019 samples and allow coefficients to vary by survey year. This approach tests whether particular gaps widened or narrowed. A separate analysis of households in core-based statistical areas examines whether local bank branch accessibility predicts unbanked status.

These methods are appropriate for estimating conditional associations, but they do not provide a causal identification strategy. The regressions control for a large number of observable characteristics, yet unmeasured differences could still influence both the explanatory variables and bank account ownership.

Findings/Size Effects

Income is the strongest household-level differentiator. In 2019, households with annual incomes below $15,000 were more than ten times less likely to be banked than households earning more than $75,000. Although income-related differences narrowed between 2015 and 2019, they remained substantial.

Large racial and ethnic differences also persisted. Black households were approximately two-fifths as likely to be banked as otherwise comparable White households. Native American and Hispanic households were approximately one-half as likely to be banked. Importantly, racial and ethnic gaps did not materially narrow between 2015 and 2019. Additional controls reduced some of the estimated differences for Hispanic and Native American households, but they did not substantially explain the Black-White gap.

Age, employment, family structure, disability, and income stability were also associated with bank account ownership. Households headed by someone younger than age 55 were approximately one-third less likely to be banked than households headed by someone age 65 or older. Households experiencing unemployment or disability were roughly three-fourths as likely to be banked as their comparison groups. Households reporting highly variable income were about three-fifths as likely to have a bank account as households with stable income.

Digital access was particularly important. Households with internet access were approximately three times as likely to have a bank account as households without internet access. Households lacking a mobile phone were about two-fifths as likely to be banked as households with smartphones. These results suggest that technological access has become increasingly intertwined with financial inclusion.

Foreign-born noncitizens were also substantially less likely to participate in the banking system. They were approximately three-fifths as likely to have a bank account as U.S.-born citizens.

At the state level, greater financial literacy was consistently associated with lower unbanked rates. Health insurance coverage showed a more modest relationship. Payday lending restrictions were associated with higher unbanked rates in earlier years, but this relationship disappeared by 2019.

The study finds little evidence that physical branch accessibility explains remaining differences. Only 2.2 percent of unbanked respondents in 2019 identified inconvenient branch locations as their primary reason for lacking an account. The metropolitan-area analysis similarly found no significant relationship between branch accessibility and geographic variation in unbanked rates.

Conclusion

The study documents substantial progress in financial inclusion between 2015 and 2019 while showing that the remaining unbanked population is increasingly concentrated among households facing persistent socioeconomic and structural disadvantages. Income remains the most powerful predictor, but digital access, race and ethnicity, employment, disability, family structure, immigration status, and income instability also distinguish banked from unbanked households.

The results indicate that some traditional explanations have become less important. Physical proximity to bank branches appears to explain relatively little of contemporary unbanked status, while internet and mobile access have become major correlates. Differences associated with income, employment, and age narrowed over time, but racial and ethnic gaps remained largely unchanged.

banke

The study therefore shifts attention from aggregate progress toward the characteristics of the population that remains outside the banking system. It also identifies significant unanswered questions. In particular, the persistence of racial disparities after extensive statistical controls suggests that important explanatory factors remain unmeasured. The authors similarly note the need for additional research on disability and interstate differences. Future work using stronger causal designs could help determine which specific interventions or institutional changes actually increase bank account ownership rather than merely identifying the characteristics associated with it.