Skip to content
The Math and the Mechanics on Kevin Warsh’s Smaller Fed

The Math and the Mechanics on Kevin Warsh’s Smaller Fed

What Warsh can actually do with the Fed’s $6.7 trillion.
Kyla Scanlon /
A partnership announcement banner featuring the U.S. Chamber of Commerce logo with the text "Unlocked in partnership with."

You’re reading Dispatch Markets, a weekly newsletter on economics featuring Scott Lincicome, Kyla Scanlon, Karl Smith, and Marian Tupy. To access more Dispatch reporting and analysis, become a member today.


Want to continue to receive our Dispatch Markets newsletter in your inbox every week? Sign up here.

Welcome back to Dispatch Markets! The incoming Fed chair has been remarkably clear that he wants a smaller central bank, but the plumbing of the financial system may not let him have one. 

Kevin Warsh wants a smaller Federal Reserve. He has been consistent about this for over 15 years, going back to his 2011 resignation from the Fed’s Board of Governors, through op-eds, speeches, and now his April 21 confirmation hearing, where he told the Senate Banking Committee that the Federal Reserve’s $6.7 trillion in financial holdings have “done quite a bit of harm” to the institution’s credibility.

If the full Senate confirms him in time to take over from Jerome Powell on May 15, the reorientation of those holdings could become the defining policy choice of his tenure.

He’s been wrong before. In 2010, with unemployment at almost 10 percent, Warsh’s view of a smaller central bank meant arguing the Fed should sit on its hands and force Congress to act. It led to some incoherence, as Wall Street Journal reporter Nick Timiraos noted, as in that same year, Warsh voted for quantitative easing and then days later published an op-ed questioning the decision.

But he’s also been directionally right. In June 2021, when the Fed was still dismissing inflation as transitory (a word Chair Powell retired later on) Warsh wrote in the Wall Street Journal that the central bank should “stop buying mortgage securities immediately” and “slow its Treasury purchases” before inflation took hold. 

Warsh has been a hawk in every type of weather. The question is if the current moment is the one his worldview was built for, or if he’s about to misread it again. He is walking into a job former Fed Chair William McChesney Martin Jr. once described as taking away the “punch bowl just as the party gets going.” 

Warsh’s challenge is that the party never wants the bowl taken away—and that gap between the rhetorical clarity of a “smaller and simpler” Federal Reserve and the operational mechanics of getting there is where his first year as chair could actually live. The smaller, simpler Fed he wants may turn out to be more selective than smaller.

Advertisement

Stay ahead of the policies shaping free enterprise and impacting American business.

Get the U.S. Chamber’s free newsletter for insights on the economic policy, workforce trends, and regulatory landscape that affect businesses and markets.

By subscribing you agree to receive communications from the U.S. Chamber of Commerce.

The Fed’s balance sheet.

The Fed owns two big categories of bonds. The first is U.S. Treasury debt—IOUs issued by the federal government, of which the Fed currently holds about $4.5 trillion. The second is mortgage-backed securities, which are bundles of home loans that have been packaged into bonds, of which the Fed holds just under $2 trillion. Add a few hundred billion in miscellaneous assets, and you get to the $6.7 trillion total.

# Alt Text Federal Reserve total assets trended upward from approximately $1 million in July 2007 to a peak of roughly $9 million in 2022, followed by a decline to approximately $7 million by April 2026.
Chart by Amanda Swinghamer Henderson.

The Fed buys bonds because buying them puts downward pressure on long-term interest rates. When the Fed shows up as a buyer of treasuries and mortgage bonds, demand for those bonds goes up, so prices go up and yields go down. 

Lower yields mean cheaper mortgages and cheaper business borrowing. The reverse is also true: When the Fed pulls back, demand falls, prices fall, yields rise, and credit gets more expensive. 

The worry that many have expressed, including Warsh, is that cheaper credit and higher asset prices push people and businesses to invest more—and if the economy can produce enough to meet that extra demand, prices stay stable. If it can’t, say because supply chains are broken or workers are scarce, prices rise. Too much support for too long, in a constrained economy, can produce inflation, which is part of what happened in 2021, alongside supply chain breakdowns and fiscal stimulus.

The Fed pays for all those bonds with newly created reserves—electronic credits to banks’ accounts at the Fed. When the Fed buys Treasuries, it doesn’t buy them directly from the U.S. Treasury (prohibited by the Federal Reserve Act). Instead it buys on the open market, from a small group of large banks and broker-dealers called primary dealers. The Fed pays them by crediting their accounts at the Fed with newly created reserves, which is how money enters the banking system. 

Quantitative tightening (QT), the shrinking of the balance sheet, reverses the mechanism. In 2022, the Fed stopped replacing its bonds when they matured. Rather than selling, it just stopped reinvesting the cash it got back when bonds expired. That shrank the Fed’s holdings by more than $2 trillion over the next three and a half years from a peak of about $8.9 trillion. 

When a Treasury bond matures, the U.S. Treasury has to pay the Fed back. It does that by issuing a new bond, which a bank or investor buys using its reserves. The reserves leave the banking system, flow to the Treasury, and from there, back to the Fed, which extinguishes them. 

But those reserves matter, and the Fed stepping back usually has a big impact. That’s because banks owe each other money at the end of every day, and they settle up by shifting reserves back and forth between their accounts at the Fed. When reserves get too thin, banks start hoarding cash and charging each other higher rates to borrow it overnight in the repo market—the short-term funding market that keeps the financial system moving. 

We saw this play out a few times last fall. By October, bank reserves had fallen enough that the banks ran short of the cash they needed to settle the day’s transactions. They turned to the Fed’s overnight lending backstop, subsequently borrowing record amounts on Halloween and New Year’s Eve. Both events told the Fed the same thing: The banking plumbing was starting to fight it.

So the Fed stopped shrinking. In December it officially ended the shrinking program and directed all maturing mortgage proceeds to be invested in short-term Treasury bills. But ending QT is more like closing the valve when what the system really needed was to be refilled. 

So a few weeks later, the Fed actually started buying short-term Treasury bills to keep cash circulating in the banking system. The Fed gave this a deliberately boring name—“Reserve Management Purchases”—to signal this was not the same as the big crisis-era bond-buying, it was just to make sure the pipes are working.

So if/when Warsh takes over, he is inheriting a Fed that just figured out it had shrunk too much, stopped, and started buying again. The punch bowl does taste pretty good.

What Warsh has actually said.

In his April hearing, Warsh leaned on language about coordination, saying he would work with the Treasury on the unwind and told senators that any reductions would happen “slowly and deliberatively.” When Sen. Chris Van Hollen pressed him on whether sharp rate cuts would push prices up, Warsh redirected to the balance sheet: “The Fed has two important monetary policy tools. One is interest rates and the other is a balance sheet. ... Those tools should be working in concert, not across purposes.”

Warsh believes the balance sheet, in its current oversized form, props up financial asset prices and concentrates the benefits of policy on people who already own them. Shrink it, and you remove that hidden accommodation. That clears space to cut interest rates—the tool he describes as fairer because it “gets in the cracks” of the broader economy and reaches everyone, not just asset holders.

In a Hoover Institution interview last May, aptly titled “Inflation is a Choice,” he put the substitution logic more bluntly: “If the printing press could be quiet, we could have lower interest rates because what we’re doing right now is we have all this money that’s being flooded into the system, which causes inflation to be above target.” Shrink the pile so you’re not pumping money into asset markets, and you create room to cut rates without overheating the broader economy.

He has also telegraphed a call for a new “Treasury-Fed Accord,” which is a public agreement between the Fed chair and the Treasury Department secretary on the composition of the Fed’s holdings. He frames this as a restoration of Fed independence. Powell-era officials would likely call it the opposite, that giving the Treasury a formal voice in monetary policy is more like compromising independence than rebuilding it. That institutional fight will probably define Warsh’s first year more than any specific balance sheet number.

Two things are missing from all of this. The first is a timetable. The second is a commitment to the only mechanism that could shrink the balance sheet meaningfully within his term, which is actually selling bonds into the market rather than waiting for them to expire.

How to shrink a balance sheet.

Mechanically, Warsh has two real options for shrinking the holdings, and both are constrained.

Option one: Wait for the bonds to expire. Bonds have maturity dates. When they reach maturity, the borrower pays the Fed back, and if the Fed doesn’t reinvest the cash, the pile shrinks. This is the standard tool. But of course, the problem, especially with mortgage-backed securities, is that this approach is slow. As of late April, the Fed held $1.99 trillion in mortgage paper, down from $2.7 trillion when the shrinking began in mid-2022. The runoff has been slow because, as Fed staff have documented, most of the mortgages backing the Fed's holdings carry rates so low that almost no one has a reason to refinance.

Mortgage-backed securities (MBS) don’t simply expire on a fixed date the way Treasury bonds do. They shrink as the underlying homeowners pay off their mortgages, either by selling the house or refinancing into a new loan. With 30-year mortgage rates parked above 6 percent for most of the past three years, very few people have been refinancing. Households sitting on 3 percent mortgages from 2020 and 2021 are also reluctant to move, because moving means giving up the cheap loan. The Fed’s MBS pile rolls off only as fast as the housing market churns, and there is not much churning going on. 

Waiting for bonds to expire eventually gets Warsh to a smaller Fed, but that eventually is measured in many years, not really in a four-year term.

Option two: Sell. This is what would actually move the needle on mortgage holdings within a few years. Sell the mortgage-backed securities into the market. But selling mortgage-backed securities does something waiting does not: It directly raises mortgage rates for new homebuyers, which would be very painful for an already constrained housing market. 

A paper presented at the Kansas City Fed’s Jackson Hole Economic Policy Symposium in 2024, Monetary Policy and the Mortgage Market, found that during the 2020-21 easing cycle, the Fed’s portion of mortgage purchases compressed mortgage spreads by about 40 basis points. Running that in reverse—selling instead of buying—could push rates the other way. The exact magnitude is uncertain. 

But the political problem is not uncertain. In January, Trump asked Fannie Mae and Freddie Mac to buy $200 billion of mortgage-backed securities to push mortgage rates down. The administration is already running a parallel intervention, so Warsh entering the market and selling would be a direct repudiation of that move. It probably would not go over well.

The other constraint.

There is a third problem. 

The Fed needs banks to keep a certain amount of cash sitting with it. Banks use that cash to settle transactions with each other every night: Say bank A owes bank B a few billion dollars at the end of the day, and they square up by moving cash between their accounts at the Fed. When there is plenty of cash sloshing around, this works smoothly. But when cash gets tight, banks start hoarding it. They charge each other higher rates to lend it out, and the whole system gets twitchy.

The Halloween episode last year told us roughly where the line is. Bank reserves had drifted down to about $2.8 trillion when the repo market started breaking—a level that, as J.P. Morgan Asset Management has described, “prompted the Fed to pause QT and rebuild reserves to support stability.” 

The repo market is where big banks and financial firms borrow cash overnight to fund themselves, but when banks get short on reserves they stop lending in repo and overnight rates spike. This creates all sorts of problems.  

The Fed’s mortgage pile is about $2 trillion. Bank reserves at the Fed are about $3 trillion. If Warsh wants to sell mortgage-backed securities without sucking the cash banks need out of the system at the same time, the Fed has to do something. It could use the cash that comes in from selling mortgages to buy short-term Treasury bills, which keeps the cash circulating instead of pulling it out.

Powell’s Fed is doing a passive version of this swap. That means that the total size of the Fed’s pile doesn’t really change. What changes is what’s in the pile, so fewer mortgage-backed securities and more government IOUs. Warsh’s revolution, if it happens, will probably follow this, more swap than shrinkage. 

Fed Gov. Chris Waller put it bluntly when CNBC asked about plans to shrink the balance sheet without addressing reserve demand: “To me, that’s just idiotic. But reducing reserve demand and then shrinking the balance sheet correspondingly—that’s something that you seriously could talk about.” The simple version of Warsh’s program—just sell more, faster—does not work, and the people who would have to vote for it know it does not work. The version that works is slower, harder, and depends on changing the rules that drive how much cash banks need in the first place.

What this means for borrowing costs.

Warsh’s pitch—shrink the balance sheet and you create room to cut rates without overheating the economy—only works if he can actually shrink the pile. The bond market doesn’t appear to be betting that he can. 

The 10-year Treasury bond—the one that influences mortgage rates, corporate borrowing costs, and how investors price stocks—has climbed from about 3.95 percent at the end of February to 4.44 percent in early May. The 30-year bond has gone from 4.61 percent to over 5 percent. 

Some of that is the Iran shock, and some of it is the federal government issuing more debt than the market wants to absorb (U.S. publicly held debt just topped 100 percent of GDP), and some is just the broader repricing of long-duration assets in a higher-rate world.

The problem is that the Fed cannot actually shrink the pile much further. The cash floor in the banking system blocks it. The rules that drive bank demand for cash block it. The politics of mortgage rates block it. Which means the easing Warsh wants to remove is mostly stuck in place—and the rate cuts Trump is quite loudly publicly demanding can’t really be justified by removing it. 

Something else has to give. Either inflation cooperates and Warsh gets to cut rates on the merits, or the balance sheet argument is mostly cover for cuts that exist for other, less comfortable reasons.

So the Warsh era at the Fed will be defined less by the size of the balance sheet than by what’s in it–and by the institutional fight over a Treasury-Fed accord, where a public agreement between the Fed and Treasury counts as restoring Fed independence in Warsh’s telling or compromising it. 

That is a real change. It’s just not the change Warsh’s speeches imply. The smaller, simpler Fed he wants is mostly, in the medium term, just a slower and more selective Fed, and the limits on how fast it can move are more about the stability of the plumbing of the banking system rather than the preferences of the new chair.

Markets FTW

AI companies need a lot of electricity. Way more than they can get right now. The big gas turbines that make most of that electricity are largely sold out through the end of the decade—GE Vernova is now actively selling 2030 slots, with roughly 10 GW left across 2029 and 2030 combined. New nuclear plants take years. Even just plugging into the regular grid is costly and slow, as Joseph Politano laid out recently.

So in response, the companies stopped saying, “We need more power!” and started saying, “We need power FASTER,” which ended up being a different problem with different answers.

Aligned Data Centers paid for a 31-megawatt battery next to one of their Pacific Northwest facilities so they didn’t have to wait on utility upgrades. Google cut deals with two utilities to dial down its data centers’ power use during the handful of peak-demand hours each year. A company called Terrestrial Energy said: We’ll give you gas power right now, and later when our nuclear reactor is ready, we’ll swap it in. The IEA says that many data center companies have signed handshake deals to buy power from small nuclear reactors that don’t exist yet—and those deals have nearly doubled in the last 16 months.Each company found a different smart trick, in partnership with the government. These deals sit on top of substantial public scaffolding—federal tax credits for nuclear and storage, fast-tracked permitting for reactor restarts, and various state subsidy programs. The market is doing what markets do when policy has loaded the dice toward a specific outcome: routing to it efficiently.

Chart of the Week

Everyone is talking about AI and jobs. From Jasmine Sun’s recent piece about the permanent underclass to Alex Imas’ work on what a job even means, everyone is trying to pinpoint exactly where the labor market is going.

In 2016, Geoffrey Hinton, known as the “Godfather of AI,” told a machine learning conference that medical schools should stop training radiologists—within five years (or 10 at most), he said, deep learning would do their job better than they could. A decade later, the data has gone the other way. 

Per Apollo Academy, drawing on CMS provider data and Medscape compensation reports, there are roughly 5,000 more practicing radiologists in the U.S. than a decade ago, and average compensation is now north of $500,000 a year. In May 2025, Hinton walked the prediction back in the New York Times, clarifying he had only meant image analysis and that the future is radiologists working with AI rather than being replaced by it, with the technology making them “a whole lot more efficient.”

A graph illustrating the radiologist paradox displays two parallel rising lines from 2015 to 2025, tracking the number of US radiologists (left axis, green line) and their average income in thousands of dollars (right axis, orange line), both showing steady increases despite a 2016 notation about efforts to "stop training radiologists."

Reading a scan is a task, not a job, as explained by Alex Imas and Soumitra Shukla. When AI made the task cheaper, the job expanded because hospitals could now justify imaging that had previously been too expensive or too slow to order. This is Jevons paradox in a white coat: When something becomes more efficient, aggregate demand for it tends to rise.

Radiology is a generous version of this argument. A board-certified human still has to sign off on every read, so the regulatory moat is doing some of that work. 

But the same pattern shows up in other occupations. Citadel Securities’ analysis of Indeed data finds that since May 2025, job postings are up 18 percent for software engineers, 9 percent for customer service, 9 percent for banking and finance, and 18 percent for accountancy—the categories most often named as next in line for displacement.

A line graph tracking Indeed job postings for software engineers (light blue line, left axis) and overall job postings (dark blue line, right axis) from January 2024 to March 2026, showing both metrics declining through early 2025 before rising sharply, with an "Inflection in Software Hiring" marker noted around May 2025.
A line graph tracking two job posting indices from January 2024 to March 2026, showing overall indeed job postings and customer service positions declining sharply until March 2025 (marked as "Inflection in Software Hiring"), followed by a partial recovery with customer service postings remaining higher than overall postings.

Six months of posting data isn’t a decade-long experiment, and the next print could swing it the other way, but the direction of the data cleanly cuts against the displacement fears. 

History gives us a case where displacement did happen. Ernie Tedeschi recently looked at travel agents, an occupation that did get displaced, as employment is down 60 percent from the 2000 peak. But the displacement didn’t hit during the dot-com boom—it hit in the bust that followed, suggesting tech shocks may show up in labor markets only when paired with a recession. And the agents who survived moved upmarket, and their wages now match the overall private-sector average, up from 87 percent of it in 2000. 

We don’t know yet what AI is going to do to the labor market. The data we have right now leans toward a complement to people rather than a replacement, but Tedeschi’s point is important: We may not know anything until the economy really slows down. Adoption cycles for new technology tend to be slow and uneven, and labor effects can stay invisible for a long time. 

What I’m Watching

  • Tuesday, May 12: April consumer price index—whether shelter and services keep cooling will set the table for Warsh’s first meeting and tell us whether his “AI productivity will lower inflation” thesis has any near-term support in the data.
  • Thursday, May 14: Beth Hammack—Hammack was one of the people who dissented in the recent Fed meeting on language around an “easing bias,” along with Neel Kashkari and Lorie Logan. Her statement will help inform how the more hawkish members of the Fed are thinking about the economy. 
  • Friday, May 15: Powell’s Final Day—Powell’s term as Fed chair officially expires

Disclaimer: The opinions expressed above do not necessarily reflect those of the presenting sponsor.

Want to see more of The Dispatch in Google search?

Add Us as a Preferred Source

Newsletter selected

Click sign up to start receiving your newsletters.

More From Dispatch Markets

Newsletter selected

Click sign up to start receiving your newsletters.

Loading