Underbanked

Underbanked describes a household that has a checking or savings account at a bank or credit union but, in the past 12 months, also used at least one of a defined list of nonbank financial services. The classification is the FDIC's, and its point is that account ownership and being adequately served are two different things: a household can hold an account and still be paying a storefront or an app for something the account was supposed to do.

The list is closed and it is worth reading rather than summarizing. For 2023 it contained eight services, in two groups. Three are transaction services: nonbank money orders, check cashing and international remittances. Five are alternatives to mainstream credit: rent-to-own services, and payday, pawn shop, auto title and tax refund anticipation loans. A household with an account that used none of the eight is classified fully banked.

Two boundaries prevent most of the confusion around the word. First, underbanked and unbanked are mutually exclusive, not a scale. An unbanked household has no account at all, which is a different classification with a different definition and its own page. In 2023 the three categories divided the whole population: 4.2 percent unbanked, 14.2 percent underbanked, 81.6 percent fully banked. Second, the classification says nothing about how much someone used a nonbank service or why. One nonbank money order bought once in a year puts a household in the same bucket as one that runs its entire financial life through a check casher.

The most important thing to know about the headline number is that it cannot be compared with the previous one, and the FDIC says so in its own footnote. Footnote 11 of the 2023 report reads: "Underbanked and fully banked rates in 2023 are not directly comparable to those in 2021 because the question on tax refund anticipation loans was broadened to better match credit products available on the market. … Excluding the use of tax refund anticipation loans, the underbanked rate would have been 13.6 percent in 2021 and 12.7 percent in 2023. The decline between 2021 and 2023 was statistically significant."

So there are two defensible sentences about the trend and one indefensible one. It is correct to say that on the FDIC's own consistent basis the underbanked rate fell from 13.6 percent in 2021 to 12.7 percent in 2023, and that the fall was statistically significant. It is correct to say that 14.2 percent of households met the 2023 definition. It is not correct to place 14.2 percent beside the 2021 published figure and describe the difference as a change in behavior, because part of it is a change in the questionnaire.

What actually changed, since a broadened question is a vague explanation. The report prints both wordings. In 2019 and 2021 the survey asked, "did you or anyone in your household take out a tax refund anticipation loan? This is a way to receive your tax refund faster than the IRS would provide it." In 2023 it asked instead, "did you or anyone in your household pay a fee to get your tax refund faster than the IRS would provide it? This includes products like refund anticipation loans and refund advances." The narrower question caught 0.8 percent of households in 2019 and 1.0 percent in 2021. The broader one caught 2.5 percent of all households in 2023, and 2.4 percent of banked households. Since only a banked household can be underbanked, it is that second figure that flows into the composite, and it is more than a percentage point above what the old question was finding. The FDIC also publishes the counterfactual for the rest of the list: excluding tax refund products, the share of households using at least one of the other four credit alternatives was 4.2 percent in 2019, 3.6 percent in 2021 and 3.8 percent in 2023.

What is deliberately outside the composite, and why that matters more each year. The 2023 survey added questions on Buy Now, Pay Later and on crypto and did not add either to the underbanked definition. Prepaid card use is measured too and also sits outside it. Nonbank online payment services are measured and outside it as well, and they are by a wide margin the largest of the nonbank categories the survey measures: 49.7 percent of all households were using them in 2023, up from 46.4 percent in 2021, several times the share using any of the eight counted services. That is the composite's real weakness as a measure of the thing it is named for. If a household pays its rent through an app rather than with a nonbank money order, its behavior has barely changed and its classification has, because only one of those two is on the list.

The FDIC does measure that substitution, and its own figures make the point sharply. Among underbanked households using online payment services, 44.2 percent used them for at least one of paying bills, receiving income, or saving and keeping money safe, against 31.4 percent of fully banked households and 71.5 percent of unbanked households. Among underbanked households with prepaid cards the equivalent figure was 58.8 percent, against 39.7 percent of fully banked households. So the survey documents nonbank substitution running well beyond the eight services it counts.

What being underbanked correlates with, stated as correlation. In 2023, 5.8 percent of all households used a rent-to-own service or a payday, pawn shop, auto title, or tax refund anticipation loan, and use of those products ran higher among households with no mainstream credit, at 6.8 percent, than among households with mainstream credit. Most of the eight services have their own page here, each with its own cost structure, and reading those is more useful than reading the composite: the reason a household buys a nonbank money order is usually not the reason it takes a pawn shop loan.

How to use the classification well. It is a screening measure, built to tell a regulator where account ownership is failing to deliver the benefits of account ownership. It is not a diagnosis of a household and it is not a ranking. A household that used one nonbank service once is counted, and a household paying substantial fees to an app that is not on the list is not. Anyone quoting the figure should say which year's definition produced it, because the definition has changed inside the series and will change again.

A household that reports having a checking or savings account is asked whether anyone in it used each of the eight listed nonbank services in the past 12 months. One yes classifies the household as underbanked. All no classifies it as fully banked. Households with no account are classified unbanked and are not asked to sort into either of the other two.

A hypothetical example of how little it takes to change the label, and of why the 2023 number is not comparable with the 2021 one. The households are invented; the definitions and the survey figures are not.

The composite's sensitivity. The Okonkwos have a joint checking account, a savings account, a credit card and a mortgage. Once last year they bought a $40 money order at a convenience store to pay a landlord who would not take a card. That single purchase makes them underbanked. Had they bought the same money order at their own bank's teller window, they would be fully banked, because the service on the list is a nonbank money order. One $40 transaction, two different classifications, no difference in the household's financial condition.

The non-comparability, in numbers. Now take a hypothetical population of 1,000 households, and suppose that in both years exactly the same people did exactly the same things.

Under the 2021 question, 10 of the 1,000 answered yes to having taken out a tax refund anticipation loan, which is 1.0 percent, the published figure. Under the 2023 question, which asks about paying any fee to get a refund faster and expressly names refund advances, 25 answered yes, which is 2.5 percent, also the published figure. That is 25 minus 10, so 15 additional households, or 1.5 percentage points, and not one of them changed anything they did.

Whichever of those 15 households hold an account become newly underbanked, if the refund product was the only listed service they used. So a measured rate that rose by a little over a point could be produced entirely by the questionnaire. That is exactly why the FDIC published a like-for-like pair instead, putting the rate at 13.6 percent in 2021 and 12.7 percent in 2023, and why anyone comparing waves should use those two figures rather than the headline.

Pros (of the measure)

• It separates two things that get conflated, showing that opening accounts for people does not by itself stop them paying nonbank providers.
• The definition is a published, closed list, so the classification can be checked rather than argued about.
• The FDIC flagged its own break in comparability in a footnote and published a like-for-like pair alongside it, which is more than most statistical series do.
• The survey measures nonbank substitution well beyond the eight counted services, including online payment services and prepaid cards, so the broader picture is available to anyone who reads past the headline.

Cons (of the measure)

• The headline 2023 rate is not comparable with the 2021 one, and almost every secondary source that quotes both makes that comparison anyway.
• It is a binary flag with no intensity. One nonbank money order in a year counts the same as a household running entirely through a check casher.
• The list has not kept pace with how people actually move money. Buy Now, Pay Later, crypto, prepaid cards and nonbank payment apps are all measured and all excluded from the composite.
• Because the list can change between waves, the series is less reliable over time than the unbanked rate, which rests on one stable question.
• The label describes a household rather than a person, so it inherits the same limitation as the unbanked rate: one member's behavior can classify everyone in the house.