Why Treasuries Became Risky Again

Contents

1. 1790: bond markets began as war finance
2. The 1980s: Treasuries once behaved like stocks
3. 2021: a flattened reaction function
4. 2022 to 2023: the market learns from actions
5. Late September 2026: taking the yield apart
6. What to watch next
7. Going further
8. ”I bought bonds to diversify — so why did stocks and bonds fall together in 2022?”
9. ”The Fed hiked — why are yields still climbing?”
10. The one thing to take away

It may perish, it may perish — therefore bind it to the mulberry root.

—— Book of Changes, Hexagram Pi, line five (pre-Qin China; trans. by the author)

In the Bloomberg Odd Lots episode published on 5 October 2026, hosts Tracy Alloway and Joe Weisenthal talk to Carolyn Pfluger, associate professor at the University of Chicago’s Harris School of Public Policy and visiting scholar at the Chicago Fed. On the day of recording, 29 September, the US 30-year Treasury yield stood at 5.592% — its highest level since 2002. Pfluger’s research attributes the majority of the 2020–2025 rise in the 10-year yield to one mechanism: Treasuries began moving together with stocks, so investors started demanding a higher return to compensate for that risk. She also states the condition under which her conclusion flips. Returning to a 1980s-style bond market would take supply-side inflation shocks plus a Fed willing or forced to accept a recession, and that combination has not assembled yet.

1790: bond markets began as war finance

I did not expect an episode about yields to open with Alexander Hamilton. Pfluger cites his 1790 address on the public debt, where he calls American credit “the price of our liberty.” Her point is blunt: the development of bond markets was tied from the start to winning wars and paying for them.

The clearest case is Britain. Before the Napoleonic Wars it already had a deep bond market with low borrowing rates, which helped enormously during the fighting; winning then made it the undisputed military and financial power for decades. The US held a comparable position through the second half of the twentieth century — strongest militarily, cheapest borrower. The two advantages feed each other.

With a co-author at Columbia, Pfluger built a model with three regimes. When bond markets are tiny they do not matter: whoever holds an exogenous advantage — a moat, a mountaintop — is safest. When borrowing capacity is moderate, the country that starts with an advantage can save its way, little by little, into a combined financial and military lead, and the model has something like a tipping point: whichever side the system starts on, it rolls further that way. When borrowing capacity and debt-to-GDP are both high, causality can run the other direction. Markets decide one country is safer, offer it cheaper financing, the cheap financing funds the investment, and the investment makes the original expectation correct.

She flags that last case as hypothetical rather than a description of what has happened. The implication stayed with me anyway: a hegemonic transition that involves no war at all, settled by financial markets pricing their expectations.

Joe then asks a quiz question — what is the yield on China’s 30-year bond? Two percent, against 5.6% for the US that day. He notes the 10-year spread now exceeds 350 basis points, and that US yields sat below their Chinese counterparts until late 2020. Tracy adds the observation that made me laugh: the long downward trend in Chinese yields looks like it started around 2018–2019, when those bonds entered the global indices, so the benchmark index providers may be the future shapers of global hegemony.

The 1980s: Treasuries once behaved like stocks

The middle of the episode moves to Pfluger’s other research line, and she opens with a fact many people miss: Treasury bonds were not always safe. Through stretches of the 1970s, 80s and 90s they were viewed as risky — the era of bond market vigilantes and the inflation risk premium.

Her measure is how bonds and stocks move together. Before 2000, Treasuries looked stock-like. Stagflation was what everyone discussed, and stagflation is punishing if you hold nominal bonds: inflation eats into the $100 you get back at maturity, while the recession hits equities. Both fall together, and the bond is a risk asset. To decompose that period she has to use UK data, because Britain has issued inflation-linked bonds far longer; in those data, a large chunk of the bond risk sat on the inflation side.

After 2000 the character of recessions changed to the demand variety. Demand recessions come with lower inflation, nominal bonds benefit from lower inflation, the bond-stock correlation turns negative, and Treasuries become a hedge again. Her work with John Campbell and Luis Becerra argues this explains the flip in Treasury bond risk around 2000.

She then asks the follow-up: luck or policy? Her answer is both, and the policy half shows up in the data with a specific shape — once monetary policy gained credibility, the policy rule priced into markets became more gradual and more inertial. The value of inertia only shows when a supply shock arrives: a gradual rule has a chance of sticking a soft landing, and if the soft landing gets priced in, stocks do not need to fall on the same day as bonds.

2021: a flattened reaction function

The most useful concept in this episode is the policy reaction function. Pfluger’s definition is plain: what markets expect the Fed to do in response to economic conditions. As a numerical example, if forecasters pair 2% inflation with a 4% policy rate and 4% inflation with a 6% policy rate, that is one-for-one, a coefficient of one.

How do you measure it? She and co-authors Michael Bauer and Adi Sundaram take two routes. The first uses survey forecasts: Blue Chip asks dozens of forecasters each month for their fed funds forecast, and also for the inflation and output assumptions behind it — the survey itself lays bare the rule each forecaster is plugging in, so you run the regression on that. The second looks at how yields jump on macro release days: when inflation surprises high, how much does the two-year rate move? The two methods often agree, which she says makes her more comfortable with either one.

Then she walks back to 2021. Rates sat at zero while inflation ran at 5% and 6%. The reaction function she measures for that period is flat — the perception was that the policy rate would stay at zero regardless of economic conditions. In May 2021 inflation surprised to the upside by almost half a percent, 6% annualized, and the two-year rate did not move at all.

One detail here stuck with me. At the time some people thought inflation was transitory and some did not. Her dataset lets her split the forecasters apart — and the ones expecting higher inflation also expected zero rates. The disagreement was not about inflation. On the question of whether the Fed would respond, both camps agreed.

She offers the mirror image too: in late 2011 the Fed gave date-based forward guidance, rates at zero at least until mid-2013, and every fed funds forecast collapsed to zero with no dispersion left.

2022 to 2023: the market learns from actions

Liftoff came in March 2022, first 25 basis points, then a run of very large hikes. In her data the perceived inflation response climbs from zero to one between early 2022 and late 2023. The increase arrived late, and it arrived after the Fed had started to act. She calls the mechanism learning from actions.

Tracy’s follow-up lands on the sore spot: if markets only update when the central bank actually moves, what happens to a central bank that wants to be gradual? Is there an accepted trade-off where you must demonstrate seriousness about inflation before markets will revise the rule in their heads, which then makes the inflation fight easier, at the cost of equities falling or financial stability cracking?

Pfluger does not dispute the trade-off; she connects it to the bond risk line. Part of why 2022 hurt so much is that bonds and stocks were positively correlated, so a portfolio holding both had nowhere to hide.

On whether a reaction function ought to be stable, her view is that it need not be. A mechanical rule would not require a central bank to execute it, conditions change, and the appropriate response differs across moments. She also ran a simple check: the perceived response to output tends to be steeper during tightening cycles — the periods when a Bernanke or a Yellen would say they are data dependent — while easings tend to arrive suddenly, after which little else is expected and the perceived response flattens.

Late September 2026: taking the yield apart

Back to that 5.592%. Joe asks how she decomposes the drivers over the past years, months, or even weeks, and she answers on a five-year horizon.

The big change over five years is that Treasuries got riskier, with bond-stock co-movement up a lot. What follows is basic investment logic: if an asset becomes riskier, investors should not be willing to pay as much for it, or equivalently they require a higher return as compensation. Prices and yields move inversely, so that compensation shows up as a higher yield.

With two co-authors she asked whether this risk is priced. The trick is to strip out the forecast error first: assume next year’s yield lands where informed forecasters expect it to land, and ask what return a 10-year Treasury offers today under that assumption. The expected return computed that way tracks bond-stock co-movement closely in the data. It holds for the US, across maturities, and across developed markets whose experiences sometimes differed considerably.

Her back-of-the-envelope estimate runs like this. Of the decline in the 10-year yield from the mid-80s to the 2010s, roughly a quarter came from Treasuries becoming better hedges, with falling inflation expectations among the rest. Of the increase from 2020 through 2025, the majority can be explained by bonds becoming more stock-like — because long-term inflation expectations have held steady over these years even though we lived through high inflation.

Two differences from the 1980s matter, and this is the part I would most want on record. First, the magnitude has not returned to 1980s levels. Second, a good chunk of the recent risk increase sits in the inflation-indexed bonds, so the driver differs from the inflation component that dominated the 1980s.

Joe presses: if market-priced long-run inflation expectations are stable after the Fed has missed its target for about five years, does that not imply the Fed is still seen as credible? Her answer is careful. How inflation expectations form is not her specialty, but what she observes is a stable point estimate for long-run inflation alongside more volatility, plus a de-linking between nominal and inflation-indexed bonds. Before 2000 those two were highly correlated because nobody was thinking about inflation; that correlation has changed. The average looks calm while the uncertainty underneath it has grown.

On supply her framework is deliberately simple: we are all taxpayers, and to the extent we also hold the bonds, issuance is one pocket to another — Ricardian equivalence. Reality violates it, but as a first approximation it remains a useful starting point. Joe then gives my favourite passage of the episode. When the first Trump administration passed tax cuts and everyone asked where the money would come from, his thought was that a tax cut had just told us: a bunch of people have more dollars in their accounts, maybe they buy a boat instead of Treasuries, then the boat seller buys stock, and the stockholder buys Treasuries. His footnote: it does not work in reality, it works in theory, and it has robust explanatory power.

What to watch next

Tracy asks the most direct question of the hour: how do we make bonds more bond-like again?

Pfluger splits it into luck and policy. Luck is the changing nature of shocks hitting the economy. Policy is getting a gradual response priced in — and for gradualism to work, the background requirement is central bank credibility, the trust that eventually this institution will do what is needed.

She sets the changing buyer base aside as outside her research, though she will commit to one judgment: the change in the safety of Treasuries themselves is a substantial component of the move in yields. A lower natural rate and demand for scarce safe assets may still matter, but they are not the whole story.

Going further

”I bought bonds to diversify — so why did stocks and bonds fall together in 2022?”

This is a hole I fell into myself. I treated hedging as a property of an asset, a label in my head: equities are risk, Treasuries are safety. What Pfluger’s whole research line is about is the parameter underneath that label — the bond-stock correlation — and nobody mails you a notice when it moves.

To make it usable, here is what I do before buying anything meant to diversify. Question one: which world does this thing help me in? Treasuries help in a demand recession — weak economy, low inflation, a central bank with room to cut. Question two: how does the probability of that world compare with the probability of the one I did not picture? A supply shock plus a central bank unwilling or unable to accept a recession is the other world, and in it bonds fall on the same day as stocks. Writing down answers to those two beats memorizing a 60/40 split.

”The Fed hiked — why are yields still climbing?”

Tracy names this tension at the top of the episode: there was a hike after Jackson Hole, yields kept soaring, and that leaves an open question about whether the hike had its intended effect on financial conditions.

The episode gives me two layers. The first is the learning lag: the rule in the market’s head does not change on announcement day, and the 2022–2023 update only completed after a sequence of hikes. So some of the gap between policy and market response is a learning cost rather than a policy failure.

The second layer matters more. A long-dated yield contains a risk compensation component, and that component is governed by how stock-like Treasuries are, not by one meeting’s decision. If you see the policy rate rising while long yields rise too, one reading is that the risk compensation is widening at the same time. To tell which you are looking at, check the inflation-indexed side — Pfluger notes a good chunk of the recent risk increase sits there, meaning the problem is not runaway inflation expectations. The comparison is a nuisance to run, and it keeps you from reading two different things as one.

The one thing to take away

Safety is a condition, not a property.

That is what four decades of Treasury history teaches. The same piece of paper is a risk asset in a world of supply shocks and a hawkish-by-necessity central bank, and a hedge in a world of demand recessions and a central bank with room. The paper did not change. The condition did. And our heads are built to drop the condition and keep the label.

Here is something I tried, offered rather than assigned. Take a sheet of paper and write down three things you feel certain will not break — a person, an income, a physical habit, a relationship. Then add one line to each: the specific condition under which it does break, specific enough that you could tell tomorrow whether that condition is happening. “My job is secure” might get “unless the division’s gross margin stays below X for two quarters.” The point is not to make yourself anxious; it is to move the condition out of your blind spot. The first time I did this, one of the three was a condition I could not write — and that was the one that broke.

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.