A Quarter-End Funding Spike Looks Like Stress. The Calendar Usually Explains It.

A jump in overnight funding rates on the last business day of a quarter is, in the base case, an artifact of financial reporting dates rather than evidence that cash has gone missing. Across the six quarter-ends from March 2025 through June 2026, the Secured Overnight Financing Rate rose between 5 and 16 basis points against the prior business day, the quarter-end print was the highest of its observed window every time, and the following session printed lower every time. What the data does not show is a clean snap back: one business day after the turn, SOFR still sat 2 to 7 basis points above the pre-turn session in all six cases.

That is the reading in full. A quarter-end funding spike is a scheduled, bounded, decaying event with a shape specific enough to test. It becomes information about stress only when the shape breaks.

The next turn is Wednesday, 30 September 2026. The prior business day is Tuesday the 29th, and the first decay session is Thursday, 1 October. Anyone carrying financed positions, holding short-dated rate exposure, or using funding spreads as a regime input will see a print that, read cold, looks like deterioration.

What SOFR Is Actually Measuring

SOFR is transaction data, not a survey and not a policy setting. The Federal Reserve Bank of New York's documentation on its reference rates describes it as a rate that "provides a broad measure of the general cost of financing Treasury securities overnight," and it is built from trades rather than quotes: "The TGCR, BGCR, and SOFR are each calculated as a volume-weighted median, which is the rate associated with transactions at the 50th percentile of transaction volume."

Two features of that construction matter for reading a turn.

• The headline is a median, so it is insensitive to a handful of expensive trades. When the headline moves, a large share of the market repriced. Somebody paying up in a corner does not move it.
• The distribution is published alongside it: "Also published alongside the volume-weighted median rate are the 1st, 25th, 75th, and 99th volume-weighted percentile rates and the transaction volume underlying the rate." The 99th percentile is where pressure shows up first, and it is free on the New York Fed's SOFR page.

One operational note that prevents a pointless argument the next morning: revisions are narrow. "Rate revisions will only occur if the change in the rate exceeds one basis point and only on the same day as initial publication." The quarter-end print seen yesterday is the print.

Why the Turn Moves at All

The mechanism is documented by the Federal Reserve Board. In a FEDS Note titled "What Happens on Quarter-Ends in the Repo Market," published 6 June 2025, Erik Bostrom, David Bowman, Amy Rose and Andy Xia write that "these dynamics arise because some dealers scale back part of their repo intermediation on quarter- or month-ends, which can cause the spreads they charge to widen and the level of repo rates to increase."

The reason dealers pull back is a reporting convention rather than a shortage: "Dealers and other financial institutions often engage in what is known as window dressing by altering their behavior around financial reporting dates, which are typically quarter- or month-ends, in order to minimize the effects of a variety of regulatory or accounting rules."

The cash does not vanish; it changes venue. "As seen in the top-left panel (panel A), borrowing in the triparty repo market has tended to decline at the end of the quarter (date 0 in the charts)," and "as dealers reduce their borrowing in triparty repo, money market funds and other entities that are active lenders in this segment also shift some of their lending to the centrally cleared sponsored repo market (panel B)."

Balance sheet rented less cheaply on one date is a different object from cash being unavailable at any price. That difference is the whole distinction between a calendar turn and a funding event. Figure 1 sets out the window it occurs in.

The same note records how far this has run before. At the September 2024 and year-end 2024 turns, "the spread between SOFR and the ON RRP rate rose as much as 25 basis points, with SOFR surpassing the rate paid on reserves and a significant volume of repo trades occurring at even higher rates (not plotted)." Turns larger than the recent band have happened, which is why what follows has an escalation path rather than a flat dismissal.

Six Turns, Measured

The measurements behind Figure 2, from the New York Fed's published SOFR reference rates. Each line gives the quarter-end median, the prior business day, the one-day change, and the 99th percentile on the turn date.

• 31 March 2025 — 4.41 percent against 4.34 percent on 28 March, a move of +7 bp. The 99th percentile printed 4.57 percent, 16 bp over the median.
• 30 June 2025 — 4.45 percent against 4.39 percent on 27 June, +6 bp. The 99th percentile printed 4.65 percent, 20 bp over the median and the widest tail in the sample.
• 30 September 2025 — 4.24 percent against 4.13 percent on 29 September, +11 bp. The 99th percentile printed 4.38 percent, 14 bp over.
• 31 December 2025 — 3.87 percent against 3.71 percent on 30 December, +16 bp, the largest of the six. The 99th percentile printed 4.00 percent, 13 bp over.
• 31 March 2026 — 3.68 percent against 3.63 percent on 30 March, +5 bp, the smallest. The 99th percentile printed 3.78 percent, 10 bp over.
• 30 June 2026 — 3.68 percent against 3.62 percent on 29 June, +6 bp. The 99th percentile printed 3.80 percent, 12 bp over.

The decay side, one business day after each turn: 4.39, 4.44, 4.20, 3.75, 3.65 and 3.66 percent. Lower than the turn in all six cases, by 1 to 12 basis points. Still above the pre-turn session in all six, by 2 to 7 basis points.

By the second business day, where the comparison is available, the gap against the pre-turn session was −1 bp on 5 January 2026, +2 bp on 2 July 2026, +3 bp on 2 April 2026 and +7 bp on 2 October 2025. The decay is real, but it is neither uniform nor complete, and in the October 2025 case the level simply settled higher rather than returning. A trader who assumes a full round trip by T+2 is assuming something the record does not support.

Year-end is the upside outlier in this sample. A December turn is not a template for a March or June turn, and a June turn is not a template for December.

Where the June 2026 Turn Sat Against Policy Rates

"SOFR spiked" means nothing without the reference levels. Per the Federal Reserve's implementation note of 17 June 2026, the FOMC maintained "the federal funds rate in a target range of 3-1/2 to 3-3/4 percent," the Board voted to keep "the interest rate paid on reserve balances at 3.65 percent," and the Desk was directed to "conduct standing overnight repurchase agreement operations at a rate of 3.75 percent." The implementation note of 29 July 2026 left all three unchanged.

Measured against that, the 30 June 2026 median of 3.68 percent sat 3 bp above the interest rate on reserve balances and 7 bp below the standing repo rate. Inside the target range, and unremarkable. The 99th percentile of 3.80 percent, by contrast, printed above the top of the target range and above the standing repo facility rate.

The tail crossed the facility rate while the median did not come close. That is the signature of a distribution stretching on one date, not of a market short of cash. Reading only the headline would have missed it; reading only the tail would have overstated it.

Concrete Framework

Four tests, run in order, on the published print rather than on commentary about it.

1. Fix the baseline before the date arrives. Record the median SOFR and the 99th percentile for the five sessions ending on T−1. Doing this on 29 September removes the temptation to define "normal" after seeing the spike.
2. Size the one-day jump. T minus T−1, in basis points. Roughly 5 to 16 bp is the band the last six turns produced. A print inside the band is a non-event. A print far outside it is a question, not yet an answer.
3. Measure the tail, not only the median. The 99th percentile minus the median on the turn date; 10 to 20 bp is what the recent sample produced. Compare the tail to the standing repo facility rate rather than to yesterday's SOFR, because the facility rate is the reference that says whether paying up was avoidable.
4. Watch the decay, and give it two sessions. The test the last six turns pass is not a full retrace, which did not happen once. It is narrower: T+1 prints lower than T, and the residual against T−1 sits in the low single digits. If T+1 prints higher than T, the calendar explanation is finished and the other three tests stop mattering.

Two further inputs are observable and worth adding. The Desk runs standing overnight repo operations twice each business day, from 8:15 to 8:30 a.m. and 1:30 to 1:45 p.m. Eastern, against U.S. Treasuries, agency debt and agency mortgage-backed securities. Under the New York Fed's statement of 10 December 2025, effective 11 December, these run in a full allotment format with the previous aggregate limit removed, with each eligible counterparty able to "submit one proposition per security type during each of the twice daily operation times, with a maximum per-proposition amount of $40 billion." Operation results are published, and the facility's FAQ sets out who may participate. A turn where the tail widens but take-up stays negligible is telling a different story from one where it does not.

Second, collateral supply is the other half of this market, and Treasury issuance mechanics change what a like-for-like comparison even means, a point worked through in A Reopened Ten-Year Is Recorded as 9-Year 11-Month, and Comparisons Break There.

The position consequence follows from the decay test rather than from the spike. If the framework holds, financing costs on leveraged positions are elevated for one to three days and then are not, which is a carry question rather than an exit question. That is the same separation between a mechanical threshold and a discretionary decision that governs margin generally, set out in Reg T Sets 50 Percent. The Maintenance Call Lands at a 33 Percent Drop. If the decay test fails, position size moves ahead of the view, and the volatility-regime question covered in When Monthly Moves Go From 4 Percent to 42 Percent, Adjust Stops First becomes the live one.

What Would Invalidate This

The framework above is a conditional claim, and these are the observations that would break it. Any one of them is sufficient.

• Pressure that arrives early. If SOFR is already elevated on T−3 or T−2 and keeps climbing into the turn, the move is not being caused by a single reporting date. Window dressing is timed to the date; a funding shortage is not.
• A turn that prints far outside the band. The 5 to 16 bp range comes from six observations in one policy environment. The FEDS Note's own record of a 25 bp spread over the ON RRP rate in 2024 shows the band is not a law. A materially larger print means the sample was too short, not that the reading is wrong in kind — but it does mean the base case no longer applies.
• T+1 higher than T. This is the cleanest disconfirmation available and it requires no judgement. In the six turns examined it did not occur once.
• A residual that holds instead of fading. If SOFR is still 10 bp or more above the pre-turn session a week later, with no change in the target range or the interest rate on reserve balances to explain it, the level has shifted and the calendar did not do it.
• Tail rates far above the facility. A 99th percentile a few basis points through the standing repo rate is one thing. A tail well above it, sustained, says counterparties who could have gone to the Desk did not or could not, and that is a question about access rather than about accounting dates.
• A change in the policy framework itself. The band described here was measured under a specific configuration of the target range, the interest rate on reserve balances, the ON RRP offering rate and the standing repo rate. If the FOMC changes any of them, the comparison points move and the historical band has to be re-measured rather than reused.

Two Objections Worth Holding

The first is that six observations is a thin sample, and it is. The turns span roughly fifteen months over which the median fell from 4.41 to 3.68 percent, with the target range unchanged only across the June and July 2026 notes. A band built on six numbers should be treated as a rough prior, not as a distribution. The honest version of the framework's second test is that a print inside the band is unsurprising, not that a print outside it is alarming.

The second is that the reversal evidence is weaker than the jump evidence. The one-day jump is unambiguous in all six cases. The decay is not: the residual against the pre-turn session was still positive at T+1 every time, and at T+2 in three of the four cases where the comparison exists. Reading "it reverses" into this data is reading more than is there. What the data supports is narrower, and the framework is written to that narrower claim: the turn date is the local maximum, and the next session is lower.

Neither objection changes the practical conclusion for the coming turn on 30 September. It changes how much weight a single print should carry. A funding spike on a reporting date is the market's least surprising event, and treating it as news is a reliable way to trade a calendar rather than a condition.