Where the Money Moved

Three lines, ten years

Reported deposits are up 68.9%. Strip out the booking offices by capping every branch at $500m and the branch base is still up 41.6%. The branch count is down 17.4%.

Almost all of the deposit growth landed before June 2021. Since then the fenced base has moved +1.4% in five years.

Every June filing is on the chart now, and the flat ending is not flat. The fenced base peaked in 2022, fell 5.3% through 2024, and at 2026 it is still 2.8% below that peak. Reported deposits took the same shape, dropping 4.8% in 2023 before recovering.

On four snapshots this looked like a plateau. It is a surge, a retreat and a partial recovery, and only the annual series shows it.

The gap between the green and blue lines is money booked at offices no household ever walks into.

Over half of it is booked, not banked

Every office is capped at $500m before anything is measured. Nationally that cap removes $10.08tn of the $19.05tn filed, which is 52.9%. It comes from 2,590 offices out of 75,859.

The chart ranks metros by how much of what they filed the fence takes away. Sioux Falls files $1,128bn and keeps $13bn. Salt Lake City files $1,176bn and keeps $32bn.

These are not deposit markets. They are addresses on a charter.

At county grain it is starker: Lincoln County SD loses 99.4%, New York County $1,778bn.

And it is a short list of names

JPMorgan Chase carries $1,220bn above the fence, spread over 386 offices. Bank of America carries $1,204bn over 300.

Then it stops being about branch networks at all. Goldman Sachs books $472bn at two addresses. Morgan Stanley $268bn at one, State Street $242bn at one, Charles Schwab $225bn at one.

A single office holding a quarter of a trillion dollars is not a market signal. Left uncapped, it would decide the ranking of every market it sits in.

This is why the fence is set on office identity across all eleven years rather than year by year. Widening the panel changed the fenced national total by 0.00%.

Over ten years, almost everywhere grew

Every county shaded by the change in its fenced deposits since 2016. Not one of the 195 large metros is down over the decade. Only 10 of all 918 are.

Miami is up 39.4% and added $66bn. Dallas-Fort Worth added $79bn, Los Angeles $112bn, New York $137bn.

This is the baseline. The interesting question is what happened after 2021.

Where the dollars landed

California added the most, $293bn, then Texas $233bn and Florida $192bn. Together Florida and Texas are 16.3% of the national gain.

But the growth rates tell a different story. Georgia +60.8%, Tennessee +59.6%, South Carolina +67.4%, Idaho +73.3%. California grew +34.7% and New York +22.7%, the slowest of any large state.

California added the most money and closed 1,539 branches, 22% of its network.

In the newest year California is up just +0.63% against a national +1.41%, so it is still giving up share, but it has stopped shrinking.

By state, since 2021

The same five years, rolled to state level. Eleven states have lost fenced deposits and they form a block: Oregon −10.4%, Washington −7.8%, California −5.9%, plus Maryland, New York, Connecticut, Colorado, New Jersey, Arizona, Nevada and DC.

The growth is the interior and the Southeast. Arkansas +16.2%, Nebraska +15.0%, South Carolina +14.6%, North Dakota +14.3%, Kentucky +13.1%, Indiana +13.0%.

All fifty, ranked

Every state and DC on one axis. The spread runs from +16.2% to −10.4%, a 27 point gap over five years on the same measure.

Over the decade the picture inverts: every state is up, from Idaho at +73.3% to DC at +16.5%. Five years is where the divergence lives.

And the states are not uniform inside

Each county measured against its own state. Green means it beat its state, rust means it lagged. Nationally 1,481 of 2,392 counties beat their state, so the typical state is carried by a minority of large counties.

In Oregon only 4 of 29 counties grew deposits at all. In Nebraska, 58 of 60 did.

Which states are splitting

The share of each state's counties that grew fenced deposits since 2021. Nebraska 97%, Montana 95%, South Carolina 95%, Arkansas 94%.

At the other end Oregon 14%, Washington 27%, New Jersey 33%, California 36%, Arizona 42%, New York 47%.

States with at least eight counties holding five or more offices.

The last five years

The same counties, now shaded by change since 2021. The decade map was almost uniformly green. This one is not.

59 of the 195 large metros have lost fenced deposits since 2021, and 174 metros in all lost both deposits and branches.

Growing fastest

Myrtle Beach +25.7%, Fayetteville AR +24.1%, Sioux Falls +20.9%, Amarillo +19.8%, McAllen +18.9%, Lubbock +18.3%.

Texas takes four of the top ten. The rest is the Southeast and the smaller Plains markets. Not one is a coastal metro.

Losing deposits outright

Santa Rosa −16.2%, Santa Barbara −15.1%, Lake Charles −13.7%, Tallahassee −12.9%, Boulder −12.6%, Portland −11.3%.

Thirteen of the eighteen worst are in California, Oregon or Washington. The decline is a West Coast event.

All of these are still up over the decade: they are giving back the 2020-21 surge. And in the newest filing that has largely stopped. 55 of 58 states are up on fenced deposits in the last year; only Washington, DC and Arizona are down.

In dollars, not percentages

Dallas-Fort Worth added $14.0bn since 2021, more than any other market. Indianapolis, Cincinnati, Nashville and Columbus follow.

At the other end, New York gave back $49.4bn, Los Angeles $27.8bn, San Francisco $20.4bn, Washington $10.8bn.

Over the full decade every one of those four is still strongly positive. New York added $137bn from 2016.

And the networks behind them

185 of the 195 large metros closed branches over the decade. The steepest are not the biggest: Bakersfield, Toledo and Youngstown are all down more than a quarter.

Only 28 markets added offices in the last five years, and they are the same Sunbelt names that lead the deposit growth.

Where the money sits

Los Angeles County holds $332bn inside the fence, twice Cook County. New York County holds $133bn and is down 15.8% in five years, the steepest fall of any large county.

Miami-Dade is the counterexample among the giants: +3.2% over five years and +$29bn over the decade.

Counties growing fastest

Madison County MS +30.7%, Delaware County OH +30.3%, Hendricks County IN +30.0%. All three are exurban rings with household growth above +6%.

Delaware County is up +141.9% over the decade, and Hendricks +125.7%.

Counties with 40 or more offices, so these are real markets rather than small-base effects.

Counties losing the most

Multnomah County OR −21.1%, Baltimore city −18.4%, Suffolk County MA −16.8%, Sonoma −16.2%, New York County −15.8%.

Baltimore city and New York County are the only two on this list that are also down over the full decade, at −1.4% and −4.1%.

Every metro is really its core county

Split each metro into its largest county by deposits and everything else. The ring beat the core in 91 of the 112 multi-county large metros, 81% of them.

Oklahoma City reads +15.4% as a metro. Its core, Oklahoma County, holds 64% of the deposits and grew +8.4%; the six suburban counties grew +30.9%.

Ranking metros and stopping there hides the part of the market that is actually growing.

The same thing, on the map

Each county measured against its own metro rather than the nation. Green beat its metro, rust lagged it.

The rings light up and the cores go dark, almost everywhere. Phoenix is the cleanest case: Maricopa is 97% of the metro at −1.5%, and Pinal next door grew +7.6% while adding offices.

Counties with three or more offices in 2021.

Households per branch, both charters

Divide each county's households by its branch count, then judge it against the country: 134.4m households across 96,712 branches is 1,390 per branch. The line is set at a round 1,400, the way it used to be a round 1,500. Below it the market is overbanked, above it underbanked.

Both charters are counted. 75,543 bank offices and 21,169 credit union branches: on bank offices alone the national ratio is 1,781, the line this analysis has used throughout, and counting only banks overstates how much room a market has by more than a quarter. Green on the map is underbanked, which for a credit union is the side with room in it.

Esri 2026 households against the June 30, 2026 FDIC filing and the August 2026 credit union branch file. Both sides are the 50 states and DC: the filing also covers 316 offices in Puerto Rico and the territories, which carry no household count here and are left out of the ratio rather than diluting it.

The count says one thing, the people say another

Seven in ten American counties are overbanked, and between them they hold under two fifths of American households.

At metro level: 105 of the 195 large metros are underbanked, 90 are over. Adding credit unions moved 25 metros and 20 large metros off the underbanked side.

The shape survives the change. Counting banks alone at the long-standing 1,781 line the split was 2,226 over and 867 under; adding 21,169 credit union branches moves 50 counties across the line and lowers the line itself to 1,400.

The thinnest coverage

On both charters, Lakeland runs 2,960 households per branch, Las Vegas 2,732, Riverside 2,611, Lake Havasu City 2,547, Tucson 2,484. All well above the 1,400 line.

At county grain, among counties with 20,000 households or more: Yuba County CA 4,259, Hoke County NC 4,173, Newton County GA 4,081, Bronx County NY 4,062.

Anchorage headed this list on bank offices at 4,348. Credit unions are 38% of its network, and it is no longer in the top ten.

And the thickest

Sioux Falls has 777 households per branch, Duluth 895, Davenport 897, Fargo 917, Portland ME 938. Several are on the fast-deposit-growth list from two acts ago.

Among counties with 20,000 households or more the thickest are Grant County WI at 512, Williamson County IL 641, Hall County NE 646, Lee County MS 665.

These are not opportunities. They are the markets where consolidation still has work to do.

Twenty-five metros were never underbanked

Credit unions are 21.8% of the combined network, and they are not spread evenly. In 25 large metros a market that reads underbanked on bank offices alone sits below the line once both charters count.

Ogden runs 3,026 households per bank office and 1,376 per branch. Credit unions are 55% of the network there. Beaumont-Port Arthur 2,328 to 1,067, Kalamazoo 2,455 to 1,163, Provo 2,447 to 1,300.

Four of the eight are in Utah, Wisconsin or Michigan. Counting only banks would send a credit union into the market it is already the largest provider in.

Thin coverage follows growth

Across the 195 large metros, households per bank office correlates with projected household growth at +0.17 and with density at +0.24. Both significant, and both measured on bank offices, where the eleven-year panel lets the relationship be traced over time.

Branch networks lag population. The places adding households fastest are the places least served, because the offices were sited against a population that has since moved.

A handover at the top

The three largest branch networks are all shrinking and all losing share. Bank of America is down 617 offices and 1.16 points. Wells Fargo down 832 and 0.97. Truist down 634 and 0.28.

The gainers are building: JPMorgan +217 offices and +0.92 points, Fifth Third +403, PNC +255, Huntington +161, BMO Harris +444.

In the last year alone, 190 institutions left the filing, including Cadence, Comerica and Synovus. 2,533 offices changed owner.

Who just changed hands

190 institutions left the 2026 filing holding 2,533 offices. 144 of the 195 large metros had at least one owner disappear.

Detroit lost 111 offices to a single deal, Comerica, 14.9% of its network. Columbus GA is 32.7% on Synovus. Clarksville 17.4% on F&M. Providence 11.4% across four banks.

An acquisition reliably produces member churn for the following year or two. This is the shortest-dated opportunity in the whole dataset.

Network equity

network equity = deposit share of a market ÷ branch share of that market

Above 1.0 a bank pulls more deposits than its branch count entitles it to. The level is interesting; the change is the decision.

Rising

East West Bank in Houston: 0.98 to 2.61 on an unchanged branch count, and +1.11 points of market share. Pure deposit gathering.

HSBC in New York: 1.38 to 2.65 by closing 69 of its offices while giving up 1.75 points. Same ratio move, opposite strategy.

Falling

Wells Fargo in Chicago, 2.02 to 0.62 while adding 10 offices. The network grew and the deposits did not follow.

Frost Bank appears twice, Dallas and Austin, adding 43 offices between them and diluting on both.

Somebody else's definition, on purpose

This is the Fed Communities Banking Deserts Dashboard definition, not one of ours: a 2020 census tract measured from its population-weighted center, with a bank branch required within 2 miles urban, 5 suburban, 10 rural.

A potential desert is a tract with exactly one branch: a desert the day it closes. Together they are 8.5 million households.

Using the published standard rather than our own matters here. An earlier cut of this analysis used block groups and a two-mile test on suburban tracts, and counted more than twice as many deserts as actually exist.

Growing slowly, while closures collapse

Deserts grew 6.9% in six years. Meanwhile branch closures have collapsed: 4,191 in 2021, 2,463 in 2023, 1,331 in 2025, on FDIC structure events. Openings held near 1,056 a year throughout.

The branch contraction is ending. The deserts it created are not closing behind it.

Credit unions fill almost none of it

21,169 credit union branches on the same test. They reach 3.4% of banking deserts, against 86.9% of well-branched tracts. The median credit union is 8.2 miles from a desert center and 1.3 miles from everywhere else.

4.28m households have neither. Coverage by density: urban 43 of 725, suburban 61 of 2,409, rural 18 of 501.

Credit unions are a present-day overlay only. 41% of their branch records carry no opening date, so there is no 2015 network to compare against and no claim here about who created anything.

Not who you would expect

Deserts are less likely to be majority-minority and less likely to be low income than the average tract.

This is not in conflict with the finding that a third of branch closures land in low-income or majority-minority neighborhoods. The two measure different things. A closure in a dense neighborhood rarely creates a desert, because other branches remain. A desert needs the last branch to go, and that happens where there was never more than one.

And not relocations either

A branch that moves keeps the count intact and could still strand a tract. So it was worth checking. 8,400 relocations, 2016 to 2026, every charter class, measured from the filed coordinates at both ends:

The median relocation is 0.16 miles. A bank moving 850 feet cannot create a two-mile desert.

The tracts branches moved out of are less likely to be deserts today than tracts at large, 0.4% against 4.3%. Banks relocate inside healthy markets. The desert count is a closure story, not a churn artifact.

Who moves branches

The moves are concentrated in the same handful of national networks that dominate the closures. Wells Fargo filed 240 relocations since 2016, PNC 213, Bank of America 188, JPMorgan Chase 167, Regions 166.

SunTrust filed 163 and no longer exists as a brand; Truist filed 110 more after the merger. BBVA USA filed 100 before being absorbed.

Relocation tracks integration work. The names on this list are the ones that merged, rebranded, or rationalized an overlapping network.

And where

Measured per 100 offices, so a large market does not simply win on size. Naples runs 16.7 moves per 100 offices, Huntington-Ashland 16.3, Portland 16.0, Madison 14.6.

On raw counts the order is just the size of the market: New York 348, Los Angeles 187, Miami 173, Chicago 169.

6,551 of the 8,400 relocations carry a usable destination county. Activity peaked at 1,179 in 2021 and has fallen every year since, to 493 in 2025.

But they do qualify for designation

A block group inside a banking desert is 9 points more likely to qualify for low-income designation. For a credit union that is the difference between entering a market and entering it with secondary capital, the member business lending exemption and non-member deposits.

The split holds the same way as everything else: rural deserts qualify at 64.7%, urban ones at 52.6%, below the national rate.

Twenty years of New Markets Tax Credit

The CDFI Fund's NMTC program has placed $71.0bn of qualified investment through Community Development Entities since 2001, across 5,074 census tracts. Eligibility runs on poverty and income, not on whether anyone can reach a bank.

The map shades each county by NMTC dollars per household. The overlap with the desert map is the question.

CDFI Fund public release FY2003 to FY2023, joined on 2020 census tract, the same key the Fed deserts file uses. 5,032 of 5,074 tracts matched.

A desert gets half the investment

A banking desert is less than half as likely to have received NMTC investment, and receives 46 cents on the dollar per household.

The program is not failing on its own terms. It is means-tested on income, and it goes where the poverty is. Banking deserts are simply not the same map.

The suburban hole

Suburban tracts are two thirds of all banking deserts and draw NMTC at a sixth the rural rate.

Put the whole thread together. Suburban deserts are economically ordinary, so they fail the income tests that drive NMTC. They are weaker on low-income designation for the same reason. Credit unions reach 61 of 2,409 of them. And they have no bank within five miles.

Rural poverty has instruments pointed at it. The largest group of banking deserts in the country has none.

What moves with the money

Across the 195 large metros, projected household growth predicts branch count change at +0.42 and deposit share at +0.36. Child growth, +0.39 and +0.37. Density is negative on both.

Branch siting tracks the demographics more tightly than the deposits do. Banks are acting on this before it shows in the balances.

The winners and losers are different kinds of place

Rank all 195 large metros by how much of the national deposit base they gained or lost since 2021. Take the top 40 and the bottom 40. Below is the median metro in each group, measure by measure.

Read it row by row. The winners are younger, faster growing, less dense and less wealthy, and they are keeping their branches: the losers closed offices four times faster.

Gaining or losing the national base is a relative test. A metro can grow its deposits and still sit in the bottom 40 if the country grew faster.

By household type it is a ladder

Weight every county by how many households of each LifeMode group live in it and both series line up with density, in order, with no exceptions.

Countryscapes, least dense at 1,088 per square mile, grew deposits +10.7% and lost 5.8% of its branches. Urban Harmony, at 28,739, lost 3.7% of deposits and 11.2% of branches.

One last measure: share

Everything above is growth. Share drift is the competitive version: a market's share of the national base at the end minus its share at the start, in points.

It is the harsher test. Miami grew 9.6% as filed and still lost share, because the country grew faster. 128 of the 195 markets grew deposits and 23 of those lost share anyway.

Rust on this map means losing ground, not losing money.

1. The base year decides the answer.From 2016 the fenced base is +41.6% and no large metro is down. From 2021 it is +1.4% and 59 are. The surge went into branches in 2020 and 2021 and nothing since has. Quote both or quote the decade.
2. Cap the booking offices.2,590 offices hold more than $500m each and 53% of all deposits. Uncapped, New York reads +12% over five years; capped, −6%.

1. The recent decline is regional, not general.Thirteen of the eighteen worst five year markets are in California, Oregon or Washington. Texas, the Southeast and the Plains are still growing deposits at double digits.
2. The network is the leading indicator.Branch count change correlates with household growth at +0.42, better than deposits at +0.36. Watch where banks are opening.
3. Density is the strongest negative signal there is.Rho −0.40 against deposit share and −0.88 across household types. Dense urban is a retention problem, not a growth market.