The 1PM Echo: Treasury Auctions Are the Last Living Signal on the Half-Hour Clock

TL;DR — We cut the US regular session into its thirteen half-hour windows and tested every ordered pair — 78 predictor→target combinations — on SPY, QQQ and IWM across 1,370 full sessions (January 2021 – July 2026), full-tape 30-minute bars. The famous academic signal every intraday-momentum thread still quotes — first half-hour predicts the last half-hour — is not just dead, it is inverted: t = −2.3 on SPY, and trading it bled −0.55bp a day for five and a half years. The late-day momentum pocket (15:00→15:30 predicting the close) died in 2022. Exactly six of the 234 tests survive a cross-ETF significance screen, and the strongest cell on the entire clock — the 13:00–13:30 move predicting the 13:30–14:00 move, t ≈ +7 — turns out not to be a time-of-day anomaly at all. Split the sample on the US Treasury coupon-auction calendar (notes and bonds are auctioned at exactly 13:00 ET, results seconds later) and the cell falls apart cleanly: on the 513 auction days, t = +8.5 / +7.9 / +8.0 across the three ETFs; on the 857 days without an auction, t = +0.1 / −0.7 / +0.0. The signal does not exist unless the Treasury sells bonds that day. The bond tape itself shows almost no follow-through (TLT's own autocorrelation: t = +1.8) — bonds reprice in the print; equities echo for another half hour. On 10/20/30-year auction days, the naive rule — trade 13:30–14:00 in the direction of the 13:00–13:30 move — hit 61.4% on IWM at +5.5bp per half-hour; reading the signal from TLT's 13:00 bar instead made money in every year of the sample, including 2025, when the equity-bar version lost. The auction calendar is public weeks in advance. This is the rare intraday edge you can literally put in your diary.

Day traders build their entire day around three clocks: the 09:30 open, the 10:00 macro prints, and the 14:00 Fed. Almost nobody structures anything around 13:00 — lunch, the dead middle of the session, the stretch when volume dries up and desks rotate to coffee. But two or three times a week, at exactly 13:00 ET, the US government runs the largest scheduled risk-transfer event in financial markets: a coupon auction that moves $40–70 billion of duration in a single print, with results on the wire within about a minute. This study is about what that print does to the equity tape in the thirty minutes that follow — and about how it turned out to be the only thing left alive on the intraday clock.

The invention: test the whole clock, not the pet hypothesis

Most intraday-seasonality research — including the good academic kind — picks one pattern and tests it. We inverted the process. Take the regular session, cut it into its thirteen half-hour windows (09:30–10:00 through 15:30–16:00), and compute the predictive correlation of every earlier window's return with every later window's return: 78 ordered pairs per instrument, run identically on SPY, QQQ and IWM over 1,370 full sessions from January 4, 2021 to July 2, 2026 — full-tape (SIP) 30-minute bars, the ten early-close sessions excluded, and every half-hour return cross-checked against an independent source to a median difference of 0.01bp.

That gives 234 tests, and with 234 tests, a t-stat of 2 is noise — you'd expect a dozen of those by luck. So the screen is deliberately brutal: a cell only counts if |t| ≥ 2.5 on all three ETFs independently. Random noise doesn't replicate across three different instruments 1,370 times each.

Six cells survive. Not the six anyone would have guessed:

| Predictor → target | SPY | QQQ | IWM | What it is | |---|---|---|---|---| | 10:30 → 13:00 | −3.6 | −3.5 | −3.1 | Late-morning move mean-reverts into 13:00 | | 13:00 → 13:30 | +7.0 | +5.9 | +6.5 | The 1PM echo — this study | | 13:00 → 15:00 | +4.9 | +4.3 | +4.1 | The 13:00 window's reach extends into the afternoon | | 13:30 → 15:00 | +3.6 | +3.7 | +3.8 | Same regime, second leg | | 14:00 → 14:30 | −2.9 | −3.9 | −3.7 | The 14:00 push fades at 14:30 | | 15:00 → 15:30 | +4.4 | +4.5 | +3.3 | Late-day momentum — but see the obituary below |

Look at where the predictability lives: nothing before 10:30 predicts anything, and everything that works routes through 13:00. The morning — the part of the day retail treats as the information-rich session — is a predictive desert. The afternoon has a spine.

The graveyard first

Consistent with how we published the Huddle Index and the First-Hour Fade, the dead signals get printed with full honors:

• The famous one is inverted. The best-known result in intraday-seasonality research — the first half-hour (including the overnight gap) predicts the last half-hour — was published on 1993–2013 data and still circulates in every momentum thread. On 2021–2026 data it is not merely gone: the correlation is negative (t = −2.3 SPY, −2.4 IWM). Trading it — long the close if the morning was green, short if red — returned −0.55bp per day on SPY for five and a half years, −26.6% cumulative on IWM. Everyone still quoting that paper is trading a ghost with the sign flipped.
• Late-day momentum died in 2022. The 15:00→15:30 cell looks alive in the full sample (t = +4.4) — but split it: 2021–2022, t = +5.4; 2023–2026, t = −0.2. Whatever drove it (the MOC-imbalance front-run was the usual story) has been arbitraged flat for four years. A full-sample t-stat can be a tombstone with good posture.
• The lunch reversal and the 14:30 fade are real but thin. Both negative cells replicate across all three ETFs, but at 1–2bp of expectancy they're context, not trades: the late-morning trend tends to give some back into 13:00, and the 14:00 push tends to stall at 14:30. Useful priors for exits; not entries.

Which leaves the 13:00 cell — twice the t-stat of anything else on the clock, in a window nobody watches.

The reveal: it's not a time. It's an event.

Our first cut at the 13:00 cell looked like a calendar oddity: the effect was enormous on Wednesdays (r = +0.45, t = +8.5 on SPY) and absent on Fridays. Day-of-week anomalies are where bad research goes to die, so we went looking for what actually happens at 13:00 on those days — and the answer is on TreasuryDirect: coupon auctions. Every 2, 3, 5, 7, 10, 20 and 30-year note and bond the US government sells is auctioned with a 13:00 ET close, results published seconds to a couple of minutes later. The long end — 10s, 20s, 30s — clusters on Wednesdays and Thursdays. Bills auction at 11:30 and don't move markets; the coupon calendar is the event.

So we split all 1,370 sessions on the official auction record — 513 coupon-auction days, 857 without — and the ambiguity ended:

On auction days: t = +8.5 (SPY), +7.9 (QQQ), +8.0 (IWM). On non-auction days: +0.1, −0.7, +0.0. The entire six-t-stat headline cell is auction days wearing a time-of-day costume. And the Wednesday "anomaly" executes perfectly on cue: Wednesdays without a coupon auction show r ≈ −0.09 on all three ETFs — indistinguishable from dead. There is no Wednesday effect. There is an auction effect, and auctions like Wednesdays.

This is the difference between the echo and everything in the graveyard above: the dead signals were statistical regularities — patterns with no anchor, which is why they rotted the moment enough capital found them. The echo is pinned to a recurring information event on a public schedule. It can weaken, but it can't quietly stop existing, because twice a week someone transfers tens of billions of dollars of duration risk at a price nobody knows until 13:01.

The mechanism: bonds print, equities echo

Why does the equity market trend for half an hour after a bond auction? Watch who reacts when:

• The bond tape absorbs the result almost instantly. TLT's own 13:00→13:30 autocorrelation on auction days is r = +0.08 (t = +1.8) — statistically nothing. The instrument that received the news is done repricing within the window that contains the print. That's what an efficient reaction looks like.
• The equity tape is the slow one. The equity move in the 13:00–13:30 window — the initial risk-on/risk-off read of the auction — continues into 13:30–14:00 at t ≈ +8. Equities take the rate shock secondhand: index-level flows, vol-control rebalancing, and cross-asset desks translating "weak auction, yields up" into "sell duration-sensitive equities" don't finish in one bar. The re-rating drips for another half hour.
• The cross-asset version proves the chain. Take the sign of TLT's 13:00–13:30 bar — the bond tape's verdict on the auction — and trade the equity 13:30–14:00 window in that direction. On the 501 measurable coupon-auction days: SPY 55.3% hit, +2.1bp; QQQ 55.1%, +2.8bp; IWM 54.1%, +3.3bp per half-hour. Now run the identical rule on the 813 non-auction days: it loses on all three (−0.6 to −1.3bp, hit rates 46–48%). Same signal, same windows, opposite result — because on a non-auction day TLT's 13:00 bar contains no information to echo, and you're just trading noise against the mild afternoon mean-reversion. A signal that only works when there's information in it is exactly what a real mechanism looks like.

Sort the auction days by what's being sold and the gradient confirms it. On the 83 long-end days (10, 20, 30-year) — the auctions with real duration risk — the naive own-bar rule sharpens to 53.7% / +2.6bp on SPY, 59.8% / +3.7bp on QQQ, and 61.4% / +5.5bp on IWM. Small caps, the most rate-hostage equity complex on the board, echo the loudest. And the effect is not an artifact of a few crash prints: bucket the IWM auction-day signal bars by size and all three terciles carry positive expectancy (+2.6, +2.6, +2.1bp).

November 9, 2023: the echo at full volume

The cleanest specimen in the sample is the infamous 30-year auction of November 9, 2023 — the day a ransomware attack had knocked the US clearing arm of ICBC, a major Treasury intermediary, partially offline, and the long-bond sale tailed more than five basis points, one of the ugliest auctions in years.

The tape tells the whole story in three acts. Before 13:00, both markets drift quietly — lunch. At 13:01 the result hits: TLT drops 0.88% inside the 13:00–13:30 window — the bond market's full verdict, delivered at once. IWM falls 0.53% in the same window: the first-read echo. Then the part that pays: from 13:30 to 14:00, TLT stabilizes (+0.15%) — the bond tape is done — while IWM falls another 0.40% and SPY another 0.22%. The equity market spent a further half hour digesting news the bond market had fully priced by 13:30. A trader who knew nothing except "30-year auction today, 13:00 bar red → stay short the echo window" captured that continuation with a defined 30-minute holding period. IWM closed the day −2.2% open-to-close.

What we won't oversell

• This is index-scale expectancy, not a get-rich half hour. +2 to +5.5bp per event, gross, on instruments with sub-basis-point spreads. On SPY or liquid index futures the edge clears costs; it will not survive a wide-spread instrument or sloppy routing. Its highest use on a discretionary desk is as a bias filter — knowing which way the afternoon tape is leaning, and not fighting it — layered onto whatever you already trade in the 13:00–14:00 window.
• The naive version had a losing year. In 2025's tariff-era chop the own-bar rule went 45–48% and bled 3.7–5.2bp per event, even though the underlying correlation was the strongest of the sample (r = +0.46 SPY). The resolution is instructive: the equity 13:00 bar got noisy, but the TLT-signed version stayed profitable in 2025 (+4.8bp SPY, +5.6bp IWM). Read the signal from the asset that received the news, not from the echo you're trying to trade.
• The correlation itself has been positive every single year — 2021 through 2026, r between +0.17 and +0.46 on SPY auction days — and it is strengthening: t = +4.5 in 2021–2023, +6.9 in 2024–2026, as auction sizes marched to records. More issuance, more duration to move, louder echo. That is also the honest risk disclosure: this edge is a child of the deficit era, and a genuinely boring Treasury market would quiet it.
• Multiple-testing discipline. We ran 234 tests and told you about all of them. The echo isn't the best cell of a fishing expedition — it's the only cell that survived a three-instrument replication screen and then produced a mechanism, an event calendar, and a cross-asset confirmation when interrogated.

How the desk uses it

• Put the calendar in the diary. TreasuryDirect publishes the auction schedule weeks ahead; the quarterly refunding announcement maps the long-end supply a full quarter out. Circle every 10/20/30-year day. This is the only intraday edge we've published that you can schedule like a dentist appointment.
• The playbook is one sentence. On a coupon-auction day, do nothing until 13:30. Read TLT's 13:00–13:30 bar. Trade the 13:30–14:00 window in equities in that direction — or at minimum, don't fade it — with the position closed by 14:00. On long-end days, size up; on bills-only days, there is no trade.
• It stacks with the rest of the map. The 9:45 dead zone told you when morning breakouts lie; the First-Hour Fade told you when earnings gaps turn; this tells you what the early afternoon is for. The 13:00→15:00 cell (t ≈ +4–5) says the auction's directional read often colors the whole afternoon — a reason to let echo-aligned afternoon winners run toward 15:00.
• Know your clock. The echo window is 13:30–14:00 ET: early evening in London, 01:30 in Hong Kong (02:30 in winter). Unlike the 21:30 HKT open routine, this pocket belongs to the US and European sessions — or, given that the trigger is a single bar-close sign check with a fixed 30-minute hold, to automation. It is close to the most systematizable discretionary edge we've measured.
• Beware the collision days. CPI mornings and FOMC afternoons sometimes share a date with coupon auctions; on Fed days the 14:00 statement lands on the echo window's exit. The desk rule: the Fed outranks the echo — flat by 13:55, no exceptions.

The single sentence to take: every clock-based edge on the intraday tape has died or is dying, and the one still alive isn't a clock at all — it's the US Treasury selling bonds at 1PM, on a schedule it mails you in advance.

Related reads

The First-Hour Fade · The Night Shift · The 9:45 Dead Zone · The Huddle Index · VIX Term Structure for Day Traders.

Joining the desk

If your afternoon P&L is a coin flip because your afternoon information is a coin flip, the fix isn't more screens — it's knowing which half hours contain events and which contain noise. Vortex backs experienced US-equity day traders with DMA via Sterling Trader Pro, real-time cross-asset context on every scheduled macro and supply event, and a research desk that measures its edges before it trades them. The trader application takes about ten minutes.

---

Methodology: 1,370 full US regular sessions, 2021-01-04 to 2026-07-02, for SPY, QQQ, IWM and TLT. Prices are consolidated-tape (SIP) 30-minute bars, dividend/split-adjusted, DST-correct; the ten NYSE early-close sessions in the window are excluded; half-hour returns were cross-validated against an independent intraday source on 273 overlapping bars (median absolute difference 0.01bp, p95 0.41bp). Within-day half-hour returns are bar open→close; the legacy "first half-hour" test additionally includes the overnight gap (prior 16:00 close → 10:00), matching the original academic definition. The predictability matrix is the Pearson correlation of each earlier half-hour return with each later half-hour return, per ETF, with the cross-ETF screen requiring |t| ≥ 2.5 on all three; 78 ordered pairs × 3 ETFs = 234 tests, all reported. Auction dates and security types are from the official US Treasury auction record (2,358 auctions in the window; 515 coupon-auction dates, 513 in the equity sample after calendar filters; 83 long-end dates covering 10/20/30-year offerings). Backtested rules are gross of commissions and slippage; entries and exits are bar prints at 13:30 and 14:00 ET. The November 9, 2023 case study uses raw (unadjusted) 5-minute SIP bars, cross-checked bar-by-bar against the 30-minute series. One rate regime — 2021–2026, the highest-issuance era on record — is not a law of nature; nulls and the strategy's losing year are published above. Compiled from public market data — VCG Research.

#TreasuryAuctions #IntradayMomentum #MarketStructure #BondMarket #DayTrading #PropTrading #QuantResearch #CrossAsset #ExecutionIntelligence #VortexCapitalGroup