On Wednesday, the 10-year Treasury yield closed at 5.277% — the second consecutive daily close above the 5.25% level that bond strategists had flagged as a line in the sand. Thursday's intraday trade pushed to 5.34%, the highest since 2002. The 30-year closed at 5.62%, a post-2002 high. September was the worst month for Treasuries in four years, and the FT's summary of the mechanics has entered the lexicon: a "vicious loop" of selling.
And yet, on the same day, MarketWatch reported that retail investors are "aggressively piling into" bond ETFs as a contrarian bet. The FT ran an "optimist's guide to the bond market." Bianco Research called 5.25% a buying cushion. Germany's 10-year Bund yield eased off its 17-year peak of 3.642%.
Both things are true at once. That collision is the subject of this piece, because it contains the answer to the only question that matters for this regime: how does a bear steepener actually end?
The vicious loop, stated precisely
The FT's framing deserves to be taken seriously rather than treated as color. The loop runs roughly like this: rising yields force selling from leveraged holders (margin calls on basis trades, risk-parity deleveraging, convexity hedging from mortgage servicers), that forced selling pushes yields higher, which triggers the next round of forced selling. Price weakness creates its own supply. This is self-referential selling — the mechanics are well understood, and they were the transmission channel in 1994, 2003, and 2022.
Two features make the current loop distinctive. First, it is not oil-driven. Brent crude has oscillated around $100 for two weeks — reclaiming $100 on Thursday after Chinese refiners banned October fuel exports — but the long end kept rising even as oil eased. The repricing is inflation-expectation and term-premium driven, which means energy de-escalation does not rescue bonds. The FT has noted that the oil-Treasury coupling is the tightest since 1990, but the causality runs through inflation expectations, not through relief.
Second, the loop is global and synchronized. The OAT-Bund spread hit roughly 128bp this week — wider than Tuesday's 113bp print, which was itself the widest since 2012. France is preparing a record €340bn of 2027 issuance and responded with pension and state-salary austerity. Japan's 10-year JGB yield sits near multi-decade highs. There is no safe duration anywhere in the G7, so there is no flight-to-quality bid to catch the falling knife. TLT fell alongside equities again this week — the stagflation signature — because the thing being repriced is the risk-free rate itself.
The structural case: why this isn't over
The bear case for bonds rests on supply and demand arithmetic that hasn't changed:
- Record supply meeting shrinking official demand. France's €340bn 2027 program is the sharpest expression, but the US faces a heavy auction calendar and roughly a trillion dollars in bill supply. Foreign official holders — the marginal buyer of the last two decades — have been reducing. China's Treasury holdings sit at 2008 lows; Japan sold over $100bn in a single month this summer.
- A competing issuer with inelastic demand. AI-capex issuers are tapping IG markets at a pace measured in hundreds of billions, at spreads that still price them as safe. Every basis point of corporate demand is capital not available for sovereigns. Broadcom's residual-value guarantees and this week's reported $42bn Anthropic chip-lease are vendor-financing escalation — the supply of "safe-looking" AI credit is expanding exactly when sovereign duration needs friends.
- The Fed is not riding to the rescue. The central bank hiked in September and the December hike remains priced. Kashkari said Wednesday night that inflation is "still too high" even after the cooler PCE. The policy put under duration is absent — indeed, policy is the source of the repricing.
This case says the loop has further to run: 6% on the 30-year, which several strategists now discuss openly, is not a fringe scenario.
The exhaustion case: why it might be close
But the counter-evidence is no longer anecdotal, and bear steepeners do not die from supply arithmetic alone — they die when the marginal buyer changes:
- Retail is absorbing. The bond-ETF inflows reported this week are the classic capitulation-of-the-seller, arrival-of-the-buyer signature. Retail is wrong at tops and right at extremes; after the worst Treasury month in four years, this is at minimum an early absorption signal.
- Sentiment instrumentation is extreme. When an "optimist's guide to the bond market" and a "vicious loop" headline can coexist on the same day, the bearish consensus is no longer crowded — it is universal. Crowded short-duration positioning is fuel for the squeeze, not the trend.
- The first crack in the policy path appeared. Core PCE cooled to 3.0% from 3.3%, and October hike odds collapsed from roughly 70% to 35% within days. Williams says there is "no urgency"; Kashkari says inflation is still too high. That split is what the beginning of a policy-reversal branch looks like — messy, internal, and dismissible until it isn't.
- Bunds eased first. The German 10-year backed off its 3.642% peak. Core rates easing while periphery blows out is not repair — but it is the first divergence from the synchronized-rout pattern, and synchronization is what made this loop vicious.
- The real economy is carrying the yield. GDP was revised up to 2.2%, Micron beat and guided strong on 11x data-center revenue growth, and the demand leg of the AI buildout keeps printing. A 5.3% 10-year yield is sustainable if nominal growth is 5%+. The bear case requires not just high yields but failing nominal growth to force the policy reversal — and growth, for now, is holding.
How bear steepeners end: the two exits
The framework I have used all quarter says bear steepeners end one of two ways, and the evidence for each is now visible:
Exit 1 — Policy reversal. The Fed signals the hiking path is done (December de-prices; the dot path gets walked back), and the front end rallies hard, dragging the long end down by expectation rather than by demand. Falsifier: December hike odds rising again, or a hot NFP tomorrow combined with hawkish Fedspeak. Telltale: the Williams-vs-Kashkari split widening into official guidance.
Exit 2 — Demand repair. The loop's forced sellers exhaust while natural buyers (retail ETFs, pensions at these yields, insurance at 5.6% 30-year) scale in; auction tails shrink; basis-trade funding stabilizes. Falsifier: a failed auction with a large tail, or retail inflows reversing after a further leg higher. Telltale: this week's inflows continuing through a retest of the highs.
The honest read: the structural case is intact and the loop mechanics are real, but the convexity has flipped two-sided. The marginal buyer is changing before our eyes, and the most dangerous moment in a vicious loop is precisely when forced selling exhausts into universal pessimism.
Positioning implications
For duration shorts — my own TLT-short among them — the management rule writes itself: the trade is right, but the squeeze tail is now explicitly the demand-repair branch, so profit-taking discipline (trim on strength toward retrace levels, hard exit above the invalidation) matters more than conviction. For equities, a 5.3% risk-free rate competing against a 4%-ish earnings yield on the index is a valuation ceiling, and the breadth data (small caps at new lows, financials crushed, a handful of AI mega-caps holding the indices up) says the market agrees — Gundlach's "hollow tree" framing is the same observation. For gold, the real-yield constraint remains dominant; the debasement bid has a ceiling until the policy-reversal branch opens.
Friday's payrolls print is the next binary, and it cuts both ways: hot data feeds the loop (more hikes priced); weak-but-not-breaking data extends the stagflation mix; genuinely breaking data opens Exit 1. The OpEx pin sits at 7,625-7,650 on the S&P — inside the range the market has been grinding down through all week.
The vicious loop and the contrarian bid are not contradictory headlines. They are the two halves of an ending. Which half completes first is the trade.