Reserves, Panics, and Dumb Rules: George Selgin on the New Deal and Banking Crises

Contents

1. What this episode covers
2. Four promises, none delivered
3. A corridor needs no state switch
4. Opening the discount window is the second-best fix
5. A 12% loss rate, and the same experiment twice
6. ”It’s a model of the bank panic in Mary Poppins”
7. The chain that ran through harvest season
8. They say the tool works — how do I check?
9. A holding that’s down 50% — panic, or real damage?
10. The One Thing to Take Away

The people hunger because those above take too much in taxes; the people are hard to govern because those above do too much.

—— Laozi, chapter 75 (Spring and Autumn period; translated by the author)

On the October 5, 2026 episode of Macro Musings, host David Beckworth brought back George Selgin of the Cato Institute. The hardest number in the hour is the Fed’s pandemic-era Main Street Lending Facility: as of last June, realized losses had passed 12% of advances, and Selgin expects the final figure near 15%. The 1930s program everyone already treats as a failure lost about 3%. His reading is that banks were not overlooking creditworthy firms, so when the central bank stepped in to fill the gap, what it filled it with were the loans banks had already looked at and declined. Keep in mind that Selgin has criticized the current reserve regime for over a decade, and these are the numbers he chose to put forward.

What this episode covers

George Selgin studies monetary history and free banking. Around 2014 he wrote Floored!, the first book-length critique of the “ample reserves” system — the arrangement where the Fed keeps so many reserves in the banking system that it has to pin the market rate with a price (the interest it pays on reserves) rather than with the quantity of reserves. That regime started in October 2008, and the balance sheet went from a few hundred billion to several trillion along the way.

Four topics: the balance sheet and Kevin Warsh’s new task force on it, pandemic-era Main Street Lending, the standing of the Diamond-Dybvig bank run model, and the real causes of America’s late nineteenth-century banking crises. The first two are institutional assessments, the last two are historical revisions, and one thread runs through all four — when something blows up, go read the rules that were in force at the time.

Four promises, none delivered

Selgin calls interest on reserves a mistake from the start. It grew out of October 2007 thinking, when the Fed was working out how to keep rates from falling too fast; in hindsight the question was how to get rates lower, possibly below zero. He adds a legal problem: the 2006 statute authorizing the payment says the rate on reserves must be at or below the going market rate for short-term risk-free funds, and the Fed has run it at or above the policy rate ever since. He wrote to the Wall Street Journal objecting at the time.

The promise list stings more than the statute. The system was supposed to work with a couple hundred billion in reserves — it didn’t. It was supposed to simplify implementation and keep rates reliably controlled — instead came overnight reverse repos, a standing liquidity facility, a subfloor beneath the floor; Selgin calls it a Rube Goldberg contraption. It was supposed to save staff and resources — the New York Fed has more people on implementation than before. The reserves injected were supposed to come back out — and now nobody knows how to take them out.

A corridor needs no state switch

This was my favorite minute of the hour. The popular compromise is “run a corridor normally, switch to ample reserves in a downturn.” Selgin says the switch is redundant: once a corridor system hits the zero lower bound, reserves stop carrying an opportunity cost and the thing becomes a floor system on its own. Nobody has to announce anything. And away from the bound a corridor won’t let you do quantitative easing — which is exactly when you shouldn’t be doing it. Choosing the corridor builds the state contingency in.

Designs that produce the right behavior on their own outlast designs that add a decision rule, because a decision rule needs someone to make the decision, and people decide badly under pressure.

Opening the discount window is the second-best fix

The consensus now is to shrink the structural demand for reserves. One route is making the discount window ordinary business — count parked collateral toward liquidity requirements, kill the stigma — so banks don’t hoard reserves themselves. Beckworth favors this and gets accused of encouraging moral hazard.

Selgin sits in between. He calls it a poor substitute for the thing it replaces: the interbank market, where a bank short on cash at the end of the day used to borrow from another bank with no moral hazard at all. That market is gone, and overnight price discovery went with it. He’d rather work on reviving unsecured interbank lending first. On whether open market operations are more market-friendly than discount lending, he says both are government intervention; the difference is that the window carries stigma while open market operations restrict eligible assets to Treasuries and direct participation to a small set of primary dealers. His own proposal, flexible open market operations, replaces that with multi-asset auctions any bank with eligible assets can bid into. He says it’s “about three-quarters baked, maybe half,” and he’d like someone to finish baking it.

A 12% loss rate, and the same experiment twice

Main Street Lending, he says, is an I-told-you-so story. The Fed normally lends only to financial institutions. In the 1930s a since-repealed authority, section 13(b), let it lend to ordinary businesses, on the same premise as 2020: commercial banks aren’t lending to all the worthy firms that need help, so the central bank must do it for them.

What happened in the 1930s was that the Fed couldn’t find many worthy borrowers, the program came in far below expectations, and the small book it did write lost about 3% of advances — more than three times what a banking system tolerates. The pandemic version replayed it. Uptake was tiny while standards held, so terms were loosened and repayment periods stretched. Volume came, and so did risk. Because principal and interest only fell due years later, the damage took this long to surface: over 12%, with provisions for future losses that Selgin expects to prove insufficient.

Two experiments, one answer: banks do a decent job of lending to sound businesses during a crisis, and the Fed sticking its mitts in burns resources. If the goal is to hand money to firms, that’s fiscal policy — let Congress own it and answer for it. The Fed’s case for lending to financial institutions rests on monetary stability, and ordinary businesses sit outside that logic.

”It’s a model of the bank panic in Mary Poppins”

Diamond-Dybvig is the textbook run model: depositors suspect each other, everyone rushes for the exit, a solvent bank fails, therefore we need deposit insurance and a lender of last resort. Selgin calls it a model of a legend — it starts from the belief that people simply panic, builds an elegant self-contained world out of that belief, and then gets cited as proof about the real one.

He jokes that it models the run in Mary Poppins, and that it doesn’t even fit It’s a Wonderful Life: that isn’t a bank, it’s a building and loan; there’s a genuine loss behind the trouble, since George Bailey’s brother misplaces the money; and the institution rides it out with no deposit insurance at all.

New evidence backs him. Stefan Luck and Emil Verner, with a coauthor, used large language models over 300 million articles to reconstruct the context of each bank panic, and found that what determines whether a commercial bank fails is fundamentals — capital, solvency — rather than runs on their own. Selgin’s line: “I love this research because I love all research that says I was right.” Then the honest footnote: economic historians had already pointed the same way with less exacting but serious work, and the large majority of Depression-era failures involved pre-run insolvency. What irritates him is the people who write “see Diamond-Dybvig” and consider the matter closed. What you’re citing doesn’t prove what you think it proves.

The chain that ran through harvest season

So why were the fundamentals bad? Dumb rules. The banks that failed in the 1930s were mostly undiversified unit banks in farm towns; one crop price falls and there isn’t enough diversification to absorb it. Canada, with branch banking and diversified assets, lost not a single bank during the Great Depression. Scotland went from the Ayr Bank collapse of 1772 to the early 1880s without an important bank failure. Selgin used to challenge his money-and-banking students: name a banking crisis and he’d name the dumb regulations without which the story doesn’t work. He says nobody ever beat the challenge.

The best stretch is his walk through 1893 and 1907. Everyone learns the Fed was created to provide an elastic currency; few learn why the currency was inelastic beforehand. The national banking system was built during the Civil War partly to create captive demand for Union bonds, so national banks could only issue notes against roughly 110% backing in specified Treasury securities, and state bank notes were taxed out of existence. Then the federal government ran surpluses for decades and retired debt, eligible bonds grew scarce, their premiums rose, and the note supply shrank with them — by 1890 the country had half the national bank notes it had in 1880, in a fast-growing economy.

Then harvest. Migrant workers bringing in the crops had no bank accounts, and with no branches there was nowhere to open one, so they had to be paid in cash. A farmer asks his bank for currency; the bank can’t afford the collateral to issue more of its own notes; the farmer takes gold or greenbacks instead — which means he’s drawing on the bank’s reserves. The bank calls its New York correspondent to draw down its balance, the New York banks tell the call money market there’s less credit today, borrowers get asked for more collateral, rates spike. That’s where the fires started. Canada’s currency supply over the same decades traces a sawtooth that follows harvest demand, and Canada had no central bank until 1935.

Hence his definition of the institution: the Fed is twelve banks exempt from the national bank collateral requirement. A national bank shows up with assets and asks, please, may I have some currency — the law won’t let me back notes with these myself, so discount them for me. And the Fed says sure, we’re exempt. That design won out because letting banks issue notes freely and branch would have killed the New York correspondent business.

One human detail: Selgin says that around 2017 and 2018, when he was in Washington, there was an internet club called the Corridor Club, consisting of everyone on the planet who thought the floor system was a bad idea. It got to about twelve people. Now a Fed chair has set up a task force and even defenders of ample reserves are discussing how to shrink the balance sheet. He calls that progress.

They say the tool works — how do I check?

The most useful part of this episode has nothing to do with monetary policy. You know the situation: on an earnings call management says the new product line will lift margins, the inventory problem clears next quarter, the acquisition pays back within a year. It sounds coherent, you neither believe nor disbelieve it, and you have no way to check.

Selgin’s method transfers directly. He doesn’t argue theory; he lists the four original promises — a couple hundred billion would do, implementation would simplify, staff would be saved, the injected reserves would come back out — and checks each against today’s facts. What those four share is that they were falsifiable when spoken: a number, a date, something observable. That shape of sentence is the thing worth writing down.

So on earnings calls I now sort what management says into two piles. “We remain optimistic about the long-term opportunity” goes in the discard pile, because it can never be wrong. Anything with a number and a date goes into a table with who said it and when, and I come back two quarters later to mark it. After three rounds you know what discount to apply to that management team, and the discount is more useful than any valuation model. Main Street Lending is the same move completed: Selgin wrote his prediction in 2020 and waited until principal and interest came due before checking the loss rate. Five years. He could wait.

A holding that’s down 50% — panic, or real damage?

The harder situation: something you own has halved, the forums have split into two camps, one saying the market has overreacted and quality assets are being thrown out, the other saying the fundamentals are already broken. Both read well, so you do nothing.

The Diamond-Dybvig segment gives this question a frame. What 300 million articles say is that banks fail because of solvency, and liquidity is the fuse that detonates the problem early. Separate the two and the question becomes answerable: is this company’s trouble that nobody will bid (liquidity), or that its ability to earn has changed (solvency)? The first repairs itself. The second doesn’t.

My own version is to write out the three reasons I bought it and ask, line by line, whether this quarter’s filing knocked one down. All three standing with only the multiple compressed is a liquidity problem; one knocked down is a solvency problem, and the two survivors don’t rescue it. I don’t run this well — when something drops I catch myself hunting for news that supports my original call, which is the same bug Selgin was laughing at in himself with “I love all research that says I was right.” He said it out loud, so I will too. The crude defense is writing down what would make you admit you were wrong before the drop, not after.

The One Thing to Take Away

When something goes wrong, go read the rules that were in force.

That’s the whole episode. The 1893 crisis happened because note issue was chained to an ever more expensive stock of government bonds, not because farmers got greedy. The banks that failed in the 1930s were forbidden by law to open branches, and that beats any story about timid depositors. The 12% loss on Main Street Lending came out of asking a central bank to do work the banks had already finished. Rules decide what can happen, and they’re invisible, because by the time anything blows up they’ve been sitting there for years.

Here’s something I’ve tried that works outside investing too. Pick one thing you’ve complained about twice this month — the chores argument at home, the report that forces overtime every week, the meltdown your kid has before leaving the house every morning. Write out the rules in force at the moment it happens, one per line, including who set them, when, and what problem they were meant to solve. Then cross one out and replay that morning in your head. Most of the time you’ll find those ten awful minutes are the shadow of a single rule, written to solve a problem that no longer exists.

Raw Markdown

This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.