Deposits stopped following people in 2022. Here is what replaced them.

For 15 years, the money followed the people

Deposits grew about six times as fast as people. Part of that is price, because these dollars are not adjusted for inflation. The better question is where the money went, and for most of this period the answer was the same places people went. From 2010 to 2021, counties that added people added deposits (rank correlation +0.52 across 1,001 counties).⁵

Then came the surge. Deposits jumped in 2020 and 2021, peaked in 2022 and sit 2.8% below that peak today. That makes your starting year part of the story. From 2016, deposits are up 41.6% and not one of the 200 large metros is down. From 2021, they are up 1.4% and 64 of those metros are down. Show your board both numbers.

National deposits inside the $500M office cap, June of each year, $tn

Ten years, June 2016 to June 2026

After the surge, the link between people and money broke. From 2021 to 2026 it halved to +0.26. Measure deposits from 2022 forward and the relationship disappears.

The examples are hard to miss. Waller County, Texas grew its population 60.3% and lost 21.1% of its deposits. Apache County, Arizona lost 11.1% of its people and grew deposits 37.9%. So what happened?

The break came from the banks, not the map

The obvious explanation is that the 2020 to 2021 deposit surge unwound. The data does not support it. What a county gained from 2019 to 2022 tells you nothing about what it did afterward.

The strongest predictor is who held the money. Counties where the four largest banks held a big share of deposits in 2021 saw the weakest deposit growth after 2022. That holds in rural, suburban and urban counties alike.

After March 2022, money left low-rate accounts at the largest banks for CDs, money market funds and Treasury bills. That movement landed on each county according to who banked there, not who lived there. The fastest-growing counties were hit hardest, because big banks held more of their deposits.

Smaller banks picked up the share. Sort every office by the size of its owner in 2022, and deposit growth rises at every step down the ladder. The 4,204 smallest banks grew nearly five times faster than the national base and took 1.35 points of share, most of what the four largest gave up (1.47).⁶

Change in deposits inside the cap, 2022 to 2026, by owner size

A shrinking county is not a shrinking market

Esri projects 1,477 counties to lose population by 2031. All but one of them gain household income anyway, about $450 billion between them.

Deposits already reflect this. From 2022 to 2026, the median county that lost population grew deposits 3.9%. The median county that gained population grew 3.5%.

The link between people and money is starting to re-form, but around where households are going rather than where they have been. Projected household growth to 2031 predicts recent deposit growth better than past population growth does.

Follow household growth, and be careful in the densest metros

Rank the 200 large metros by the share of national deposits they gained since 2021, then compare the top 40 with the bottom 40. The winners are growing faster, less dense and less wealthy. Projected household growth to 2031 is +5.3% against +2.2%. People per square mile are 3,041 against 5,788. Mean household income is $117K against $137K.⁷

Median five-year deposit growth inside the cap, 2021 to 2026, by density third of the 200 large metros

The West Coast is where the money left. Twelve of the 18 worst five-year markets are in California, Oregon or Washington. Texas, the Southeast and the Plains are still growing, mostly 6% to 9% over five years, with double digits in South Carolina, Arkansas, Kentucky, Tennessee, Nebraska and the Dakotas.

Density is the strongest drag, but it is a regional one. It matters in the West and Northeast and explains almost nothing in the South and Midwest. Household growth is the signal that holds in all four regions. Eleven of the 13 metros above 8,000 people per square mile lost deposits.

It is not an age story. The top 40 metros have a median age of 38.6 against 39.8 for the bottom 40, and that gap is not statistically significant. Income matters less than density. Younger, higher-income metros grew slowest of any group (median +1.2%). Younger, lower-income metros grew fastest (+5.9%).

So where should your next branch go?

Knowing where the money moved answers half the question. The other half is where a branch can still earn its keep. The next sections cover the network you are competing in, what makes a new branch work, and the places a branch cannot reach.

Branch closures are slowing but have not stopped

Banks are still pulling back. 190 of the 200 large metros lost branches over the decade, and California alone lost 1,539, 22% of its network.

The pace has eased: 4,191 closures in 2021, 2,463 in 2023 and 1,331 in 2025. But openings have run only 1,020 to 1,201 a year since 2021, so the count keeps falling.⁸

New branches are not safe either. Of 10,699 opened since July 2016, 1,199 (11.2%) were gone by the 2026 filing, and that is a floor. Among branches old enough to measure, 12.4% closed within five years. Plan for a smaller network, and put new sites where deposits are still growing.

Open new branches near a retail cluster

Where you put a new branch shows up in its deposits. Of 10,699 branches opened since July 2016, the median booked $6.7M in year one, $37M in year five and $57M in year ten. Averages run higher ($31M, $68M and $92M), pulled up by a few very large offices. Year-ten figures cover the 771 branches opened in 2016 and 2017 that are still open.⁹

Almost every new branch has a retail cluster within reach: 99% of urban openings, 96% of suburban and 80% of rural.

Hold market type, households, income, opening year and state constant, and a retail cluster within the Fed radius goes with about 23% higher year-five deposits. Sitting right next to the cluster goes with about 9% less than a site merely in reach, but those branches close less often: their odds of closing within five years are about a quarter lower, mostly in suburban markets. We did not control for operator or competition, so read this as association, not cause.

Median year-five deposits of full-service new branches, by distance to the nearest retail cluster (unadjusted; n is branches in each group)

Banking deserts will not fill on their own

The Fed defines a banking desert as a census tract with no bank or credit union branch within the Fed radius of its population center. The Fed counts 3,635 of them, home to 4.43 million households. We could place 3,625 on our block-group grid, and the comparisons below use those.¹⁰ Compared with well-branched tracts of the same type, they split three ways:

The stores are thin too. Inside the same radius, 87% of deserts have no national grocer, against 10% of well-branched tracts.

Few deserts can pay for a branch, so serve them another way

Credit union sites and new bank branches look alike on one test: a retail cluster sits within the Fed radius of 95% of each. Only 15% of deserts have one. The median suburban desert has 1,982 households within its radius. A new bank branch has 45,680.

Share with a retail cluster within the Fed radius

A third of new bank branches clear the median new-branch bar on all three tests: a retail cluster within the Fed radius, at least as many households within it, and at least the median predicted year-five deposits. Only 1 of 3,625 deserts does. Credit unions are not skipping these places. Almost none of them can carry a branch.

That moves the question from where to build to how to serve. Each kind of desert points to a different answer. This is our reading, not a tested result:

We did not test field of membership, charter limits, capital and cost, shared branching, ATMs or mobile service. Those are the options to price, and your field of membership decides which of these households you may serve.

What your credit union should do now

1. Stop using population growth alone to pick markets. Pair it with projected household growth and household income.
2. Look at who holds the deposits in your markets. Where big banks hold a high share is where money came loose after 2022.
3. Revisit counties you have written off. Losing people and losing a market are no longer the same thing.
4. Show two starting years. Put 2016 and 2021 side by side whenever you present deposit growth.
5. Cap first, then rank. Limit each office to $500M before you compare markets or peers.
6. Screen every site on three tests. Is a retail cluster within the Fed radius? Are there as many households within it as at new branches in similar markets? What deposits does the hex model predict?
7. Count credit union branches when you size a market. Counting banks alone can make a market look underbanked when your credit union is already its largest provider.
8. Price service options before you build. Where a market fails those tests, compare shared branching, ATMs, mobile service and digital enrollment with a branch before you commit capital.

Look up your neighborhood

Enter a ZIP code to see how households, population and income there changed, and how bank branch deposits in that ZIP are tracking against its metro.