Darrell Duffie on the Fed's Balance Sheet and the Changing Financial System

Darrell Duffie · Stanford Graduate School of Business

The Fed cannot fully unwind QE; a reserve ratchet keeps its balance sheet above prior levels.

Duffie builds on the Rajan and Acharya ratchet hypothesis from Jackson Hole 2017. When the Fed expands its balance sheet by creating reserves, commercial banks come to rely on those reserves, which pay market-like interest and solve payment and liquidity problems. They then restructure operations around that abundance. When the Fed tries to shrink, banks find reserves insufficient, market rates spike, and the Fed stops. The evidence he cites: the 2017-18 normalization got the balance sheet down to about $1.4 trillion, which was not enough, and the 2025 shrinkage only reached about $3 trillion, more than twice the earlier level. He calls the prediction 'batting a thousand' and says the effect interacts with post-crisis liquidity regulations.

“So this ratchet effect described by Acharya Rajan, it seems to be batting a thousand in terms of its prediction that when the Fed increases its balance sheet, it can't seem to get it back down as far as it was to begin with.”at 7:01

Darrell Duffie · Stanford Graduate School of Business

A liquidity savings mechanism at Fedwire could let the Fed hold fewer reserves while keeping rates stable.

Duffie proposes adding a liquidity savings mechanism (LSM) to Fedwire, the Fed's large-value payment system. Almost all large developed-market central banks already run such software. It matches banks' proposed outgoing and incoming payments so they can cancel each other, which reduces the reserves needed to settle. With Chuan Wang and Srisht Singh he found a simple matching method that is a dominant strategy for banks to use. Earlier work with Adam Copeland and Elaine Yang shows that when reserves are tight, banks hold back payments and rates spike. Duffie expects the LSM to reduce reserve needs by hundreds of billions, but declines to quantify it before further study. The Fed's own reserve-shrinking authority is not what he is advocating; he sees this as a tool to have available.

“We show that it's a dominant strategy for them to put their planned payments into the liquidity savings mechanism, get this cancellation, and avoid the use of so much reserve balances.”at 17:06

Darrell Duffie · Stanford Graduate School of Business

Interest on reserves was adopted to end scarce-reserve stress, not because of QE.

Duffie challenges a common view that the Fed started paying interest on reserves because of quantitative easing after the financial crisis. Before the crisis the Fed steered market rates by keeping reserves scarce, and the stress from that regime led it to seek congressional authority to pay interest on reserves well before 2008. The crisis only accelerated implementation. The original motive was to escape the scarce-reserve regime and its stress on the Fed and banks.

“That's actually not the case. It was the problems that you described with a scarce reserve system that caused the Fed to go to Congress much earlier than the financial crisis to get the authority to pay interest on reserves”at 22:27

Darrell Duffie · Stanford Graduate School of Business

Temporary open market operations are the easiest shrinkage tool but would cut reserves only modestly.

Duffie ranks his proposed tools by ease of implementation, which also tracks their impact on reserve needs. Temporary open market operations are easiest: the Fed already has the authority and has used them before, so the decision is mainly the FOMC's. Their impact is limited to tens of billions, perhaps 100 or 200 billion, far short of the roughly $3 trillion the Fed would like to reduce. Liquidity regulation changes are next, requiring supervisor coaching. The LSM would reduce reserves by hundreds of billions. Tiered interest on reserves is the most impactful and the hardest, because banks would resist it.

“I would start with temporary open market operations being the relatively easiest to implement. Fed already has the authority. It's used the tools in the past.”at 32:37

Darrell Duffie · Stanford Graduate School of Business

A two-tier reserve system is practical, not just theoretical, since the Fed used one briefly after Lehman.

Reserve tiering would be radical, but he argues it is not merely theoretical. After Lehman failed, the Fed ran a strict two-tier system, with a higher rate on required reserves and a lower rate on excess reserves. It collapsed into one tier when rates hit zero, and the Fed has not returned to two tiers since. The Reserve Bank of New Zealand also ran a two-tier system successfully from 2007, explicitly to shrink its balance sheet. The main obstacles are bank resistance to being paid differently on reserves and the difficulty of measuring required versus excess reserves.

“I just wanna emphasize this is not a... Only a theoretical possibility. The Fed did this for a short period of time, and it could do it again.”at 42:38

David Beckworth · Mercatus Center at George Mason University

A tiered reserve system would revive interbank lending, bringing price discovery and counterparty discipline.

Under tiered reserves, some banks would be short reserves and others long, so they would start lending to each other. He sees this as a wonderful development. It would create price discovery and counterparty discipline, and better allocation of reserves within the banking system. He does not want the federal funds market restored to its old glory days, but wants extra price signals. He cites the Norges Bank and other central banks that explicitly want interbank lending revived, after reserves stopped flowing between banks.

“some banks may be short one day reserves, some are long, and so they start lending to each other ... That'd be a wonderful development. There's price discovery. There's counterparty discipline.”at 44:37

Darrell Duffie · Stanford Graduate School of Business

Reviving the fed funds market is a good idea, though its importance is unclear.

The fed funds market is nearly inactive, at roughly $5 billion a day between US domestic banks, while the repo market runs at $8 to 10 trillion a day. Repo is mostly a large-bank business with setup costs, and many medium and smaller banks would use fed funds with enough incentive and lower frictions. Research with two Stanford PhD students suggests the spread between the upper and lower tier rates must be large enough to overcome reluctance from these frictions. Reviving fed funds would also counter the ratchet effect, in which banks lost operational skill after years of abundant reserves. He hedges that its importance is uncertain.

“Reinvigorating the federal funds market would, in my view, be a good idea.”at 47:22

Darrell Duffie · Stanford Graduate School of Business

Tokenization fragmentation will raise funding needs and likely enlarge central bank balance sheets, at least initially.

Once tokenization grows, money and securities will split into conventional and programmable forms that are not immediately fungible. Every large participant will then need two forms of everything. The cash and securities needed to start each day will rise, and funding costs will go up. Central banks will try to offset this by supplying more reserves, so their balance sheets will probably grow, at least at first. This is the cost of fragmenting between conventional and new payment and settlement systems. Multiple unconnected ledgers would add to these costs.

“Central bank will try to mitigate that by supplying more reserves ... and central bank balance sheets will, at least initially, probably get larger.”at 52:50

Darrell Duffie · Stanford Graduate School of Business

The Fed is unlikely to be an early mover on tokenized reserves; it is watching and waiting.

Central banks are not yet ready for tokenized finance, and the Fed has not moved. Some central banks are moving quickly, with the ECB and SNB supplying tokenized central bank deposits. The Fed appears to be watching and waiting on which technology to use and whether the private sector should lead. One option is to supply tokenized reserves directly, and another is special-purpose firms issuing tokenized deposits backed by central bank deposits. Duffie sees the second as a distinct possibility in the US, but expects the Fed to lag other central banks.

“But I don't expect the Fed to be an early mover on providing tokenized reserves.”at 57:20

Darrell Duffie · Stanford Graduate School of Business

Intraday liquidity rules, not the LCR, bind banks' reserve use and deter borrowing from the Fed.

Duffie argues the liquidity coverage ratio can be set aside for this discussion because it is a 30-day test. The evidence points to intraday reserve balances as the binding constraint, which shows up when interest rates spike and markets become volatile. Banks hold back lending and payments in the middle of the day. Reg YY, the internal liquidity stress tests (ILSTs), and CLAR resolution planning also play a role. Duffie notes that JPMorgan's Jamie Dimon, after the September 2019 crunch, described an informal red line at zero: going below zero would look like failing the self-sufficiency spirit of the liquidity rules, even though nothing in the rules forbids overdrafts.

“I think you can take the liquidity coverage ratio off the table. That's not the one. Because that's a thirty day test of sufficient liquidity. And what we're really seeing is evidence of intraday reserves balances being the binding constraint”at 9:46

Darrell Duffie · Stanford Graduate School of Business

The largest banks are starting to use Fed overdrafts and standing repo more freely, a shift still early.

Duffie sees early 'green shoots' in bank behaviour. At least one of the largest banks is now drawing on daylight overdrafts in the middle of some days. Standing repo operations drew about $75 billion at the end of the year. The discount window still sees almost no use from the largest banks because of heavy stigma. He expects the Fed to encourage the largest banks to use overdrafts and standing repo more liberally if it shrinks its balance sheet, and says it is 'still early days'.

“We're starting to see green shoots in terms of banks. The largest banks, or at least one of them is, I think, now drawing, overdrafting in the middle of some days. You saw at the end of the year 75,000,000,000 drawn from standing repo operations.”at 13:14

Darrell Duffie · Stanford Graduate School of Business

Largest dealer banks' overnight repo lending decides whether money markets function normally.

Duffie says the repo market is the most important money market rate by far, with roughly $10 trillion traded daily. Market-makers in repo are the largest dealer banks, not smaller banks, which are willing to overdraft. When those dealers lend overnight at rates close to the Fed's administered interest-on-reserves rate, markets are fine. The dealers' decisions therefore matter most for the money-market rates the Fed is targeting.

“it's the largest dealer banks that make markets in repos. And so when they're willing to lend funds overnight at rates close to the Fed's administered rate, which is called interest on reserves, then you know markets are fine.”at 14:19

Darrell Duffie · Stanford Graduate School of Business

Fed clarification that drawing against parked HLA counts as bank resources helps banks meet liquidity rules.

The liquidity rule, Reg YY, requires banks to meet daily liquidity needs from their own resources, and the question is whether Fed borrowing counts. The Fed has clarified that liquidity drawn against highly liquid assets parked at the Fed counts as the bank's own resources. This is a big plus if banks are convinced of it, because they can draw on HLA. Broadening this to other collateral, as Bill Nelson proposed, would make compliance easier. But the Fed has not done this, and such loans do not count as own resources. Coaching banks that supervisors will not penalize drawing on these resources would also help.

“the liquidity that it draws from the Fed against that collateral counts as the bank's own resources in meeting Reg YY and the other regulations.”at 36:46

Darrell Duffie · Stanford Graduate School of Business

Tokenized trading of traditional securities is coming; major banks and market infrastructure are preparing for it.

The back end of the financial system, where Treasury repos and large FX trades settle, is moving toward tokenized settlement. He cites the largest banks, the DTCC and international central banks as preparing for it. The new ledgers may not be literal blockchains but will share their features. These include smart contracts, trading conditioned on programmable rules, and trading without a central counterparty for each trade. He concludes the world is going that way, sooner or later, and that the story is not about Bitcoin.

“they're all sharpening their knives to prepare for the ability to trade traditional financial securities on programmable ledgers. ... So the world's going that way sooner or later.”at 49:05

Darrell Duffie · Stanford Graduate School of Business

No current stablecoin is safe enough to settle multitrillion-dollar daily markets; central bank money must be the unit of account.

Stablecoins are a programmable form of cash, but none is safe enough to settle very large markets. Any stablecoin he knows of trades close to a dollar but not exactly at it. In a billion-dollar repo, a counterparty would not accept cash priced even slightly below par. The unit of account for the largest financial markets must therefore be central bank money, a principle agreed before tokenization. Settling large market infrastructure on commercial or other money risks a sudden stop and a financial crisis if that money is not safe on a given day.

“You might say, well, why not stablecoins? That's a programmable form of cash, but there are no stablecoins safe enough to settle multitrillion dollar a day markets. ... Any stablecoin that I know about has a price that fluctuates close to a dollar, but it's not exactly a dollar.”at 50:18

Darrell Duffie · Stanford Graduate School of Business

Programmable finance will let firms fund several times a day, cutting the cash they need at each day's start.

Conventional funding such as overnight repo settles on a one-day cycle. Programmable finance will allow funding cycles multiple times a day. Broadridge already does about $8 trillion a month of intraday repo that settles within the day rather than the next. This lets banks manage with less liquidity at the start of each day. He presents this as the private sector's task, distinct from the central bank's job of supplying reserves.

“In the new world of programmable finance, you will be able to do funding cycles multiple times a day.”at 54:52

Darrell Duffie · Stanford Graduate School of Business

Major central banks are reluctant to keep huge balance sheets and will be less willing in future.

Duffie links the Fed's balance-sheet debate to moves at the Bank of England, the European Central Bank, and even the Bank of Japan, which is reducing its appetite for JGBs. Central banks are generally reluctant to keep very large balance sheets. They were willing to do so in the past but he expects them to be less willing in the future, and they all want to get smaller.

“central banks are generally reluctant to have such huge balance sheets. They're... In the past, we're willing to do that. I think they're gonna be less willing in the future.”at 3:59