February 25, 2021: the seven-year auction that shook Treasuries
A retrospective event study drawing on contemporaneous reporting, Treasury's original auction releases, and subsequent official research. The duration example illustrates risk; it does not report a trade or backtested return.
On February 25, 2021, a scheduled USD 62 billion auction of seven-year Treasury notes became a flashpoint in a bond market already selling off. Shortly after the 1 p.m. result, Treasury prices fell sharply and trading conditions deteriorated. The sharpest part of that price decline largely reversed within about an hour. The recovery in market depth took much longer.
That mismatch is what draws me to the episode. Reading the closing price would tell me one story. Watching the auction would tell me another. Trying to execute a large order during the worst minutes could have produced a third.
The evidence points to an interaction: investors were repricing the economic recovery, the auction delivered disappointing information about demand at prevailing prices, and liquidity providers struggled to absorb the resulting order flow. Describing the entire day as an auction failure would miss both the selloff that preceded it and the distinction between placing the bonds and trading them afterward.
The recovery trade was already under way
By early 2021, vaccine developments and expectations of additional fiscal support had improved the outlook for an economy still dealing with the pandemic. Investors had reasons to reconsider the very low yields established during 2020. They also had to absorb substantial Treasury issuance.
The BIS's March 2021 analysis of inflation compensation describes a rise in inflation expectations from the middle of 2020, alongside a larger measured inflation risk premium. It also examines how different supply conditions for nominal Treasuries and TIPS could influence that premium.
That provides context for the selloff, rather than a decomposition of the February 25 afternoon. A rise in breakeven inflation across several months cannot explain, by itself, why prices suddenly moved so far in a few minutes. It also cannot establish that every basis point of higher nominal yield represented a higher inflation forecast.
The Fed's public guidance remained accommodative. Its January 27 statement kept the federal funds target at 0–0.25% and committed to purchases of at least USD 80 billion of Treasuries and USD 40 billion of agency mortgage-backed securities each month. Purchases would continue until substantial further progress toward its employment and inflation goals.
In testimony on February 23, Powell said the recovery remained incomplete and that achieving that progress would take time. He also promised communication ahead of a change in the purchase pace.
Investors could accept that description of current policy and still demand a higher yield on a seven- or ten-year bond. Those securities span years of possible economic outcomes. The Fed's overnight target was an anchor for short rates, not a ceiling on longer yields.
What Treasury actually sold
The February auction offered a new seven-year note, CUSIP 91282CBP5, with a 1.125% coupon. It would be issued on March 1 and mature on February 29, 2028. The auction cleared at a high yield of 1.195%.
The previous month's result makes the change in participation easier to see:
| Auction measure | January 28, 2021 | February 25, 2021 |
|---|---|---|
| Public offering | USD 62 billion | USD 62 billion |
| High yield | 0.754% | 1.195% |
| Bid-to-cover | 2.30 | 2.04 |
| Primary dealers' share of competitive awards | 19.6% | 39.8% |
| Direct bidders' share | 16.3% | 22.1% |
| Indirect bidders' share | 64.1% | 38.1% |
Sources: Treasury's original January release and February release. Bidder shares use accepted competitive bids as the denominator, excluding noncompetitive awards and SOMA add-ons.
The offering size had not increased between these two auctions. Nevertheless, dealers' accepted competitive awards rose from about USD 12.16 billion to USD 24.68 billion, while indirect awards fell from USD 39.74 billion to USD 23.59 billion. The dealer share roughly doubled.
Those allocations show who received the securities at auction. They do not reveal how much risk dealers retained after hedging or selling to clients. Indirect bidding also describes a submission channel, not the nationality of the investor. The result supports a conclusion about the mix of auction participation; it does not identify a particular country withdrawing from Treasuries.
The high yield was 44.1 bp above January's. That measures the change between two monthly auction clearing yields. It is not the February auction's tail.
The auction surprise was about the price expected at the deadline
To measure a tail, the relevant comparison is with the when-issued yield immediately before the auction closes. When-issued trading lets market participants agree transactions in a security before issuance. An auction that clears at a higher yield than that benchmark has required a price concession beyond what the market anticipated at the deadline.
The November 2021 interagency staff report, prepared by staff from Treasury, the Federal Reserve agencies, the SEC and the CFTC, describes the February result as about 4 bp above market expectations. It also identifies the bid-to-cover ratio as the lowest since the seven-year maturity was reintroduced in 2009.
That is a more precise statement of weakness than saying investors stopped buying. Treasury allocated the offering. The price required to do so disappointed expectations, and the composition of demand had shifted.
A small-sounding yield surprise can involve meaningful duration risk. Assume a hypothetical existing position worth USD 10 million with modified duration 6.5. For a 4 bp yield increase:
That is approximately a 0.26% price decline, or USD 26,000 on the assumed position, before convexity and other effects. It is a sensitivity illustration, not a measured loss for auction winners: buyers at the auction acquired the security at the clearing price.
The result contains another accounting detail worth keeping straight. Treasury reported about USD 75.43 billion in total awards because that figure included a USD 13.43 billion SOMA add-on for the Federal Reserve. The public offering was USD 62 billion. The published 2.04 coverage ratio uses public tenders and public accepted amounts, excluding that add-on.
Mixing the totals would distort both the size of the supply investors were bidding for and the assessment of demand.
The afternoon had more than one phase
The Federal Reserve's subsequent study of the flash event, by Alex Aronovich, Dobrislav Dobrev and Andrew Meldrum, provides a transaction-level account. Their figures show price and volume changes across six Treasury maturities.
| Stage | Evidence from the record | Interpretation boundary |
|---|---|---|
| Before 1 p.m. | Treasury prices were already falling amid elevated trading volumes. | The auction did not begin the day's selloff. |
| At 1 p.m. | Treasury released the weak seven-year result. | The announcement supplied new information about demand. |
| Shortly afterward | Prices dropped, volumes jumped and displayed depth fell. | Price discovery and liquidity provision were interacting. |
| Within about an hour | Much of the sharp price spike reversed. | The broader downward price trend had not necessarily ended. |
| Following weeks | Trading spreads recovered faster than some depth measures. | Different measures gave different recovery dates. |
Contemporaneous Reuters reporting recorded the ten-year yield reaching an intraday high of 1.614%, with a subsequent quote in that report around 1.508%. Those are intraday market observations, not the Treasury's daily constant-maturity series or an assertion about the final closing yield.
I find the partial reversal important. It is consistent with a temporary liquidity amplification layered onto a wider repricing. It does not prove that the auction signal was irrelevant or that buying immediately after the result would have produced an easy profit.
The researchers describe the auction as the likely proximate trigger. They also emphasize the difficulty of establishing exactly why participants withdrew liquidity. The timestamp is unusually clear; the behavior behind every order is not.
A narrow spread can hide a difficult trade
Much of the most active interdealer Treasury trading occurs in central limit order books. Participants post prices and quantities, and incoming orders consume the available quotes. Two familiar liquidity measures capture different aspects of that process.
The bid-ask spread measures the gap between the best quoted purchase and sale prices. Depth measures the quantities offered at particular price levels. An investor placing a small order may care most about the first. An investor transferring substantial duration also needs to know how much can trade before reaching worse prices.
Depth alone leaves out another variable: how quickly participants replace quotes after a trade. A market maker can show a small quantity, execute, then replenish at a similar price. That sequence can accommodate substantial trading even though a snapshot of the order book looks thin.
On February 25, the Fed study found that executed orders in the ten-year generally remained within available posted depth after 1 p.m., and bid-ask spreads were fairly stable. In longer maturities, including the thirty-year, transactions more often exceeded depth at the top few levels. Spreads widened, tightened as quotes returned, then widened again.
This is why a statement such as “the spread looked normal” would have been incomplete. Its meaning depended on maturity, order size and the speed at which the next quote appeared. The volatility of transaction costs could matter alongside their average level.
The interagency report provides a striking example: for a five-minute period, there were no bids for the twenty-year security on the largest electronic interdealer platform. The venue qualification matters. It does not establish that nobody anywhere would buy the bond. It shows that a major channel for transferring risk had lost displayed buying interest at a critical moment.
Who could supply the next bid?
The Fed researchers identify a temporary deterioration in high-speed liquidity provision. They discuss the possibility that market makers became less confident in pricing relationships used by their algorithms as volatility and uncertainty rose.
That is a candidate mechanism, not a reconstruction of proprietary trading code. The authors could not observe each firm's algorithm, its inputs or the decision to switch it off. High-frequency liquidity provision also extends beyond specialist principal trading firms; banks and other participants use automated methods.
The interagency report adds a different constraint. Some participants reportedly reached trading limits imposed by interdealer platforms. Those limits managed exposures arising between a trade and settlement, with clearing arrangements affecting the protections available to the platform.
A firm could therefore see an apparent trading opportunity while lacking the permitted capacity to intermediate more flow. The report discusses those limits as a potential amplifier and notes that central-clearing membership was not the sole determinant of them. It does not establish a single explanation for the whole event.
For me, the useful insight is that liquidity requires more than a willing buyer's opinion about fair value. It also requires the ability to transact, finance, hedge and settle the position. A change in any of those conditions can interrupt the link between an attractive price and an executable trade.
Why ongoing Fed purchases did not prevent the episode
The Fed was already a substantial Treasury buyer. Its continued purchases could support market functioning and broader financial conditions while still leaving private participants to absorb price changes and order imbalances between operations.
Monthly purchase totals do not guarantee that a participant can sell a particular amount of a particular security at a chosen price at 1:05 p.m. They also do not eliminate uncertainty over future rates or the risk attached to dealer inventory.
The contrast with March 2020 is useful here. The Fed study finds that February's deterioration was less severe and shorter-lived, especially in its effect on quoted spreads. Treating the two episodes as interchangeable would obscure differences in scale, duration and the sources of cash demand.
Recovery still required more than the sharpest price move to reverse. Through early April, some depth measures remained depressed, with the five-year note standing out in the researchers' discussion. A portfolio manager looking only at a recovered spread could have underestimated how dependent execution remained on rapid quote replenishment.
What this would change in a strategy note
Immediately after the auction, I would want the public offering, stop-out yield, contemporaneous when-issued benchmark and bidder awards recorded on consistent bases. I would then compare the five-, seven- and ten-year sectors, along with longer maturities, to see whether the new cheapness was concentrated or spreading.
The next question would concern execution. A relative-value idea needs executable prices for its required size, evidence of two-way trading and a funding plan. A seven-year position hedged with five- and ten-year securities would require security-level risk calculations; matching aggregate DV01 alone could leave meaningful curve exposure.
That analysis should distinguish the economic view from the liquidity view. Improving growth expectations could justify higher yields over weeks or months while a temporary order imbalance pushed a particular price too far over minutes. Both could be true. A hedge designed for one mechanism might respond poorly to the other.
I would also keep the chronology of evidence intact. Treasury's release and market quotes were available that afternoon. The Fed's detailed study arrived in May, and the interagency report in November. Those later sources improve a retrospective explanation; they cannot be treated as information an analyst possessed at the auction deadline.
February 25 leaves me with a more demanding definition of a good trade idea. The forecast has to make sense, but so do the size, timing and route through the market. The question I would ask a trader after reading this episode is specific: how did the amount you could execute near the screen price change between the first bad auction headline and the first convincing recovery in buying interest?
That answer would tell me something a daily yield chart cannot.
Sources and calculation conventions
The auction comparison uses the original Treasury releases linked above. February's competitive awards were USD 24,678,990,000 to primary dealers, USD 13,717,900,000 to direct bidders and USD 23,592,880,000 to indirect bidders, totaling USD 61,989,770,000. Dividing each award by that total produces the displayed shares, rounded to one decimal place. Coverage ratios are Treasury's published, rounded figures.
The macroeconomic background uses Powell's February testimony, the January FOMC statement and the BIS's March review. The intraday reconstruction uses the May 14 FEDS Note and contemporaneous Reuters reporting. The platform-limit discussion uses the November 8 interagency staff report. These publications support different parts of the account; none supplies a complete measurement of every investor's motive.
For the broader frameworks behind the episode, see trading an auction concession, Treasury term premium, and the March 2020 dash for cash.