← back to writing

Trading the Treasury auction concession

Independent strategy framework. The auction results are historical observations; the trade levels, durations and costs below are hypothetical.

My bias after a weak long-bond auction is to look for a temporary concession against neighboring Treasuries. The conditions matter: the auctioned sector must look cheap after accounting for the curve, selling pressure must stabilize, and the expected recovery must cover financing and execution costs. A disappointing auction result alone leaves that work unfinished.

I find auctions interesting because they give an unusually precise timestamp to a recurring market problem. Treasury needs to place duration. Investors name a price. Dealers help bridge the gap. The question for strategy is how much of the resulting price movement should reverse once that supply reaches willing holders.

November 2023: read the amounts as well as the percentages

On November 9, 2023, Treasury auctioned a new 30-year bond with a 4.75% coupon and a November 2053 maturity. Its high, or stop-out, yield was 4.769%, with a bid-to-cover ratio of 2.24.

The preceding October auction provides a useful comparison, with an important distinction: October was a reopening of the August 2053 bond, while November introduced a new security.

Published result or calculated share October 12, 2023 November 9, 2023 Direction
Public offering amount USD 20 billion USD 24 billion ▲ +20%
High yield 4.837% 4.769% ▼ −7 bp
Bid-to-cover 2.35 2.24 ▼ Lower
Primary dealers' share of competitive awards 18.2% 24.7% ▲ Higher
Indirect bidders' share of competitive awards 65.1% 60.1% ▼ Lower

Sources: Treasury's original October result and November result. Bidder shares divide each group's accepted competitive amount by total competitive awards. They exclude noncompetitive awards and the Federal Reserve's SOMA add-on.

Dealers took a larger share in November, and bid-to-cover fell. That combination warrants investigating how much supply other investors would absorb at prevailing prices. It does not reveal dealers' remaining inventory after client transactions and hedging.

There is another detail worth pausing over. Indirect bidders' accepted amount rose, from about USD 13.00 billion to USD 14.37 billion, even as their share declined. The offering had grown. A smaller percentage did not mean those bidders bought fewer dollars of bonds.

Nor does “indirect” identify a foreign buyer. Treasury's auction FAQ defines the category by how a bid reaches Treasury; it includes domestic and foreign customers. A claim about foreign demand needs investor-class evidence beyond that headline share.

A tail measures a pricing surprise

An auction tail compares the stop-out yield with the when-issued yield immediately before bidding closes. When-issued trading allows investors to agree transactions in the security before its issue date. If the auction clears at a higher yield than that contemporaneous benchmark, it tails; a lower yield represents a stop-through.

For example, a 4.80% stop-out against a 4.75% when-issued yield would be a 5 bp tail. Those are illustrative numbers, not reconstructed November quotes.

The timestamp and quote convention matter. Treasury's daily 30-year constant-maturity yield describes a fitted par curve at a different observation time. The auction's median accepted yield describes the distribution of successful bids. Neither supplies the missing when-issued quote. In November's release, subtracting the 4.650% median from the 4.769% high yield produces 11.9 bp, but that is not the market's auction tail.

A tail establishes that the auction cleared more cheaply than expected at the deadline. Relative value requires another comparison: the price now available against comparable securities and an explicit estimate of fair value.

There is evidence for a temporary supply effect

In New York Fed Staff Report 1188, published in March 2026 and revised in July, Michael Fleming, Weiling Liu and Giang Nguyen examine 33 years of intraday Treasury data. They document yields rising before auctions and reversing afterward, with order flow helping transmit dealer constraints into prices.

That gives the concession idea an economic mechanism. Intermediaries have limited capacity to absorb risk; other investors take time to adjust holdings. The authors also find that auction price pressure has diminished in recent years as non-dealer participants absorb more supply. Growing debt issuance does not, by itself, establish an ever-larger trading opportunity.

I take the paper as motivation to test a strategy. Its average pattern does not establish the recovery available after a particular weak auction, the probability of hitting a target, or the return left after costs. It also makes timing important: a concession that reverses within hours may have disappeared by the time a slower investor enters.

Express the view through relative duration

For a long-end concession, one candidate is a cash 20s30s flattener: long the 30-year Treasury and short a 20-year Treasury, matching dollar sensitivity to a one-basis-point yield move, or DV01. The position benefits when the 30-year yield falls relative to the 20-year yield.

Before entry, I would compare the 30-year sector with a fitted curve that excludes the issue being evaluated. A local 20s30s spread provides a transparent trading reference, but its normal level depends on curve shape, coupon and liquidity differences. A newly issued benchmark and an older bond need particular care. A large absolute yield is insufficient evidence of cheapness.

For illustration, assume the observed 20s30s spread sits 5 bp wider than its estimated fair level. That is an excess spread: the absolute 20s30s spread could still be negative. Enter only after executable prices confirm the concession and the relative selloff stops extending. Target a residual concession of 1 bp, reassess at 8 bp, and exit after five trading days if convergence has not occurred. Hold the fair reference fixed for the arithmetic below.

Assume a long market value of USD 1.25 million with modified duration 16 and a short market value of USD 1.60 million with modified duration 12.5. Using market value times modified duration times 0.0001, each leg has DV01 of USD 2,000 per bp.

For a change in the spread s=y30−y20s=y_{30}-y_{20}, expressed in basis points:

Πgross≈−2,000Δsbp.\Pi_{\mathrm{gross}}\approx-2{,}000\Delta s_{\mathrm{bp}}.
Hypothetical outcome Gross price P&L Net after USD 1,000 costs Magnitude Risk
Concession compresses from 5 bp to 1 bp +USD 8,000 ▲ **+USD 7,000** ███████░░░ Low
Concession stays at 5 bp USD 0 −USD 1,000 ▏ Med
Concession widens from 5 bp to 8 bp −USD 6,000 ▼ −USD 7,000 ███████░░░ High

The cost allowance represents assumed net financing and round-trip execution costs over the horizon. It is not a market quote. The target and stop describe scenario payoffs, not expected returns; the probability of each outcome remains to be estimated.

Matching DV01 cancels a small parallel yield move to first order. It leaves exposure to 20s30s curve changes, unequal convexity, changing sensitivities and liquidity differences. This is a curve trade, with a supply-based thesis. Its two legs also have different market values, so financing requires the actual cash balances rather than an assumption that the short automatically funds the long.

Decide what would invalidate the thesis

The catalyst is the absorption of auction supply and subsequent normalization of relative prices. The argument weakens if cheapness spreads across the long end, the reference curve shifts with new macro information, or the next supply event arrives before the position recovers. In those cases, a persistent change in required compensation may explain the price better than temporary inventory pressure.

Execution can invalidate an attractive spreadsheet result as well. The short must be available, both legs need security-specific repo terms, and a trade spanning issuance must handle when-issued settlement dates. Stop levels are review and exit rules; gaps or poor liquidity can produce larger losses than the table.

To turn this into a tested signal, I would use timestamped quotes and auction results available at entry, compare auctions of the same tenor and distinguish new issues from reopenings. The test would include nearby macro releases, realistic bid-offer costs, financing, unsuccessful signals and the time required for convergence. A recovery measured from an untradeable pre-result price would overstate the opportunity.

The strategy conclusion is conditional: buy a measurable, funded relative-value concession when post-auction prices support the supply-absorption thesis. Without that confirmation, keep the result on the research watchlist. The part I want to understand next is how often an apparently generous concession survives long enough for an investor to capture it.

Related frameworks: Treasury carry after the funding bill and two ways the Treasury curve can steepen.