← Research
Event studies Replication

What actually happens on an FOMC day

The volatility pattern on a Fed day is real and bigger than most people think. Which way the day closes is a coin flip, and the numbers everyone quotes to argue otherwise do not survive a confidence interval.

Atty Labs · · 13 min read
212 scheduled FOMC decision days measured, 2000–2026
3.1× volatility in the press-conference hour vs a normal day
0.6× volatility through the FOMC morning vs a normal morning
47% of the time the post-statement move survives to the close

Where this came from

This started as a video from StatsEdge Trading on Fed-day statistics. Rather than reprint it we rebuilt every number from source — the meeting calendar from the Federal Reserve, the policy action from the fed funds target on FRED, the price behaviour from our own SPY daily and 1-minute data — and ran the significance tests a chart of averages leaves out. On the headline figures we land almost exactly where they did. Where we differ is in what those figures are allowed to support.

FOMC days run about 20% wider

An FOMC day is a wider day. Measured high to low, the average scheduled decision day since 2000 moved 1.56% of price against 1.30% on every other session. Open to close it was 0.78% against 0.65%. Those are real differences — a permutation test on the daily means puts them at p = 0.001 and p = 0.017 — but the ratio is about 1.2×, not the order-of-magnitude event the calendar highlighting implies.

The overnight session runs the other way. The gap into an FOMC morning averages 0.38% against 0.45% on a normal night, which fits the intuition that nobody wants a position into a decision they can't handicap. We would flag that this is the weakest of the three results: at p = 0.068 it does not clear a conventional significance bar, and the 95% interval on the difference still touches zero. It is a lean, not a fact.

How big an FOMC day is, on three measuresSPY 2000-2026. FOMC days run wider than other days from the open to the close, but the overnight session ahead of a meeting is quieter.0.0%0.6%1.2%1.8%The whole day, FOMC day: 1.56% of price1.56%The whole day, every other day: 1.30% of price1.30%The whole dayhigh to low1.20xp = 0.001 · significantThe session, FOMC day: 0.78% of price0.78%The session, every other day: 0.65% of price0.65%The sessionopen to close1.19xp = 0.017 · significantOvernight, FOMC day: 0.38% of price0.38%Overnight, every other day: 0.45% of price0.45%Overnightprior close to open0.84xp = 0.068 · not significantFOMC dayevery other daySPY, 2000-01-03 to 2026-07-31. 212 scheduled FOMC days against 6472 other sessions. Mean moveas a percent of price. Significance from a 20,000-draw permutation test.
The whole day and the session are both wider on FOMC days. The overnight gap is the only one of the three that points the other way, and it is also the one that fails its significance test.

The morning is half a normal morning

This is where the averages start hiding something. Break the session into hours on 1-minute data and the shape is nothing like a normal day. From the open to 1:30pm, every single hour of an FOMC day is quieter than the same hour on an ordinary session — 0.7×, 0.6×, 0.7×, 0.6×. Then the 1:30 hour runs 1.8×, the 2:30 hour runs 3.1×, and the last half hour is still 1.7×.

The cleanest way to say it: on a normal session, 86% of the day's entire high-to-low range is already on the board by 2:00pm. On an FOMC day it is 49%. Half the day's range has not happened yet when most traders have already decided the session is a dud.

Where the volatility on an FOMC day actually isSPY 1-minute bars 2019-2026. The morning of an FOMC day is quieter than a normal morning; the hour after the press conference is three times louder.0.0%0.2%0.4%0.6%09:30-10:30, FOMC day: 0.219%09:30-10:30, every other day: 0.307%0.229:300.7x10:30-11:30, FOMC day: 0.136%10:30-11:30, every other day: 0.240%0.1410:300.6x11:30-12:30, FOMC day: 0.134%11:30-12:30, every other day: 0.196%0.1311:300.7x12:30-13:30, FOMC day: 0.114%12:30-13:30, every other day: 0.186%0.1112:300.6x13:30-14:30, FOMC day: 0.335%13:30-14:30, every other day: 0.185%0.331:301.8x14:30-15:30, FOMC day: 0.560%14:30-15:30, every other day: 0.183%0.562:303.1x15:30-16:00, FOMC day: 0.323%15:30-16:00, every other day: 0.194%0.323:301.7x2:00 statement2:30 press conferenceFOMC dayevery other dayratio, FOMC hour vs the same hour on a normal daySPY 1-minute bars, 2019-01-02 to 2026-07-21. 59 FOMC days against 1826 other sessions. Meanabsolute move within each hour, as a percent of price.
Every hour before the statement is quieter than normal; every hour after it is louder. 59 FOMC days against 1,826 other sessions on 1-minute bars.

The quiet morning is not only a volatility story. Lucca and Moench documented that US equities earn large excess returns in the hours before a scheduled FOMC announcement — the pre-FOMC announcement drift — a result significant enough to run in the Journal of Finance.1 Worth knowing that Kurov, Wolfe and Gilbert later found the drift largely disappeared after the original paper was published,2 which is roughly what you would expect to happen to a published calendar effect. The low-volatility morning has survived; the free money in it has not.

The press conference is the real event

Splitting the afternoon finer separates two things that hourly buckets blur together. The statement drops at 2:00pm and the fifteen minutes after it run 3.2× a normal day. Then it goes quiet again — 2:15 to 2:30 is only 1.4× — while the market waits. The press conference starts at 2:30, and the half hour after it is 3.4×, with 3:00 to 3:30 still at 2.5×.

The biggest single slice of an FOMC day is not the decision. It is the Q&A. That matches the academic work directly: Narain and Sangani find market volatility more than three times higher during Chair Powell's press conferences than during his predecessors', with the post-2020 conferences responsible for most of it.3 Boguth, Grégoire and Martineau showed earlier that investors pay more attention to announcements accompanied by a press conference, and that the market risk premium is larger on those days.4 Since 2019 every meeting has one, which is why we cut the intraday sample there rather than running it back to 2000.

The statement and the press conference are two separate eventsSPY 1-minute bars 2019-2026, ratio of FOMC volatility to a normal day in each slice. The statement spikes at 2:00, goes quiet again by 2:15, and the press conference is the larger of the two.0x1x2x4xa normal day1:00–1:30: FOMC 0.076% vs 0.130% on a normal day0.6x1:00–1:301:30–2:00: FOMC 0.084% vs 0.125% on a normal day0.7x1:30–2:002:00–2:15: FOMC 0.311% vs 0.096% on a normal day3.2x2:00–2:152:15–2:30: FOMC 0.127% vs 0.090% on a normal day1.4x2:15–2:302:30–3:00: FOMC 0.430% vs 0.128% on a normal day3.4x2:30–3:003:00–3:30: FOMC 0.334% vs 0.133% on a normal day2.5x3:00–3:303:30–4:00: FOMC 0.323% vs 0.194% on a normal day1.7x3:30–4:00the statementthe press conferenceeverything elseSPY 1-minute bars, 2019-01-02 to 2026-07-21, 59 FOMC days. Mean absolute move in each slice,divided by the same slice on every other session.
Two separate events, thirty minutes apart, with a lull between them. Ratios are the FOMC-day move in each slice divided by the same slice on every other session.

The first move after the statement tells you nothing

There is a well-worn anecdote about the statement popping the market one way and the press conference taking it back. It is usually told with one chart of one meeting. We ran it across all 59.

Across those meetings, the 2:00–2:30 move and the 2:30–close move went the same direction 28 times out of 59 — 47%, with a 95% interval of 35% to 60%. A coin flip sits comfortably inside that. Meanwhile the second move averaged 0.71% against 0.33% for the first. The move you can see is less than half the size of the move you cannot, and its direction carries no information about what follows.

The first move after the statement does not tell you the closeAcross 59 meetings the 2:00-2:30 move and the 2:30-close move went the same way 47% of the time, and the second move was more than twice the size of the first.28 of 59 meetings continued31 of 59 meetings reversedcontinued 47%reversed 53%Did the rest of the day follow the first move after the statement?59 meetings, 2019–2026. 95% interval 35–60% — a coin flip sits inside it.and the second move is the bigger one2:00 → 2:302:00 → 2:30: mean move 0.33%0.33%2:30 → close2:30 → close: mean move 0.71%0.71%SPY 1-minute bars. Mean absolute move in each window, measured on scheduled FOMC days only.Direction compared sign to sign.
Reading the statement reaction as a signal for the rest of the day is a coin flip on a smaller coin.

This is not a quirk of our sample. Narain and Sangani report that since the start of COVID, markets have tended to move in the opposite direction during the press conference to their reaction to the statement, reversing the pre-2020 pattern where the conference reinforced it.3 Boguth, Fisher, Grégoire and Martineau go further, showing that announcement-window returns reverse significantly by the end of the announcement cycle and that prices immediately after an FOMC release are less informative about future prices than those before it — their reading is that a meaningful share of the announcement move is liquidity demand rather than information.5

Everyone reads the direction numbers backwards

Here is the finding that circulates most, and the one we would push back on. Sorted by what the Fed actually did that day, SPY closed green on 55% of hike days, 57% of holds and 48% of cuts. The usual reading is that this is counterintuitive and therefore interesting — cuts are supposed to be the good news.

Put confidence intervals on it and there is nothing to read. Hike days are 40% to 69%. Cut days are 31% to 66%. Even holds, with 145 observations, run 49% to 65%. Every one of those intervals contains 50%. Tested against a coin flip directly, the p-values are 0.64, 0.10 and 1.00. Tested against each other, no pair is distinguishable: hike versus hold p = 0.80, hike versus cut p = 0.58, hold versus cut p = 0.38.

Whether the day closes green, by what the Fed didSPY 2000-2026. Hike 55%, hold 57%, cut 48% — but every one of the 95% confidence intervals contains a coin flip, so none of the three is distinguishable from chance.0%20%40%60%80%coin fliphike: 22 of 40 days closed green55%HIKEn = 4095% CI 40–69%hold: 83 of 145 days closed green57%HOLDn = 14595% CI 49–65%cut: 13 of 27 days closed green48%CUTn = 2795% CI 31–66%share of days closing above the prior close, with 95% confidence intervalsSPY close to close on 212 scheduled FOMC days, 2000-01-03 to 2026-07-31. Error bars are 95%Wilson score intervals.
The three bars replicate the numbers that get quoted. The error bars are the reason they should not be.

Why 27 cut days can't answer this question

With n = 27, a true 50% coin would produce an observed rate between 31% and 69% most of the time purely by chance. Reading 48% as "cuts are bearish" is reading the sampling error. This is the single most common failure mode we see in published calendar statistics: the sample is large enough to compute an average and far too small to support the conclusion drawn from it.

Cut days are the wide ones

One difference in this data holds up. Cut days ran an average high-to-low range of 2.14% against 1.45% on hold days — a gap that clears significance at p = 0.009 on a permutation test. Hike days sit at 1.58%, and neither of the comparisons involving hikes is significant. The same ordering shows up in the tail: 30% of cut days moved more than 1.5%, against 17% of hold days.

How wide the day runs, by what the Fed didSPY 2000-2026 mean high-to-low range. Cut days are the widest, and the gap between cut days and hold days is the one difference in this chart that survives a significance test.0.0%0.8%1.6%2.4%hike: mean high-low range 1.58% across 40 days1.58%HIKEn = 40hold: mean high-low range 1.45% across 145 days1.45%HOLDn = 145cut: mean high-low range 2.14% across 27 days2.14%CUTn = 27mean high-to-low range as a percent of pricecut days vs hold days: p = 0.009 — the one difference here that is not noiseSPY, 2000-01-03 to 2026-07-31. Permutation test on the mean absolute move.
Cut days are the wide ones. The cut-versus-hold gap is the only comparison in the policy-action data that passes a significance test.

The mechanism is almost certainly not the cut. Bernanke and Kuttner established that it is the unanticipated component of a policy move that prices equities — a surprise 25 basis point cut is worth about 1% on the index, while the anticipated part is worth essentially nothing.6 Cuts arrive in conditions where the path is genuinely uncertain and the economy is deteriorating. Hikes arrive in expansions that everyone has watched coming for months. The cut is not causing the wide day; both are downstream of the same regime.

The clearest evidence for that sits just outside this sample. The two unscheduled 2020 emergency cuts are excluded from the 212 above because they were not scheduled meetings. Their market reactions were −2.86% on 3 March and −10.94% on 16 March. Those are the two most violent Fed days in the modern record and both were cuts — which tells you about March 2020, not about cuts.

Most FOMC days are unremarkable

Most FOMC days are unremarkable. The distribution of close-to-close moves piles up in the middle, with a mean of +0.23%, a median of +0.09% and a standard deviation of 1.21%. 19% of FOMC days move more than 1.5% in either direction, against 14% of ordinary days. So roughly one Fed day in five is genuinely eventful, and four in five are a normal session with a loud afternoon.

The distribution of an FOMC daySPY close-to-close on 212 FOMC days, 2000-2026. Most land in the middle; 19% move more than 1.5% either way.08162431-3.00% to -2.75%: 2 days-2.75% to -2.50%: 3 days-2.50% to -2.25%: 1 days-2.25% to -2.00%: 1 days-2.00% to -1.75%: 1 days-1.75% to -1.50%: 6 days-1.50% to -1.25%: 4 days-1.25% to -1.00%: 7 days-1.00% to -0.75%: 10 days-0.75% to -0.50%: 16 days-0.50% to -0.25%: 15 days-0.25% to +0.00%: 28 days+0.00% to +0.25%: 23 days+0.25% to +0.50%: 19 days+0.50% to +0.75%: 15 days+0.75% to +1.00%: 13 days+1.00% to +1.25%: 11 days+1.25% to +1.50%: 11 days+1.50% to +1.75%: 9 days+1.75% to +2.00%: 3 days+2.00% to +2.25%: 5 days+2.50% to +2.75%: 3 days+2.75% to +3.00%: 1 days+3.00% to +3.25%: 1 days+3.25% to +3.50%: 1 days+3.75% to +4.00%: 3 days-3%-2%-1%0+1%+2%+3%19% of FOMC days land outside the dashed lines — on every other day it is 14%SPY close to close, 212 scheduled FOMC days, 2000-01-03 to 2026-07-31. Bars are 0.25% wide andthe tails are clipped at ±4%. Count of days on the vertical axis.
212 FOMC days sorted by how much SPY moved. The fat middle is the base case; the coloured tails are the days people remember.

The first hike after a long pause

A statistic doing the rounds ahead of a possible first hike: the market has historically gone up on it. On our construction — the first hike following 250 or more sessions without one, 2000 to 2026 — there are four such days. Three closed green and the average was +0.85%.

Four observations is not a finding. The 95% interval on a 3-of-4 win rate runs from 30% to 95%, which is to say the data is consistent with anything from a losing bet to a certainty. We are including the chart because the number is circulating, not because it supports a position.

One methodological note, since it changes the count. This statistic is often quoted with five cases, adding February 2000. February 2000 does not meet the criterion: the prior hike was 16 November 1999, roughly 53 trading sessions earlier, nowhere near a year-long pause. Including it makes the sample look 25% larger than the definition allows.

The first hike after a long pause4 occurrences since 2000. Three of four closed green, but four observations cannot separate a real effect from chance.-2%-1%0+1%+2%2004-06-30: +0.54%, 1034 sessions since the prior hike+0.54%2004-06-301034 sessions2015-12-16: +1.46%, 2383 sessions since the prior hike+1.46%2015-12-162383 sessions2016-12-14: -0.83%, 251 sessions since the prior hike-0.83%2016-12-14251 sessions2022-03-16: +2.22%, 815 sessions since the prior hike+2.22%2022-03-16815 sessions3 of 4 closed green, mean +0.85% — but the 95% interval on that win rate runs 30%to 95%, which is every answer from a coin flip to a certainty.SPY close to close on the first hike following 250 or more sessions without one, 2000–2026.Wilson score interval.
Every qualifying case since 2000. Four observations cannot tell you what the fifth one does.

What to do with this

Nothing in this is a trade. It is a description of a session shape, and the honest use of it is defensive.

What we’d test next

The direction question is closed at the daily level — 212 observations is as much sample as exists and it says nothing. The open question is whether the volatility result is tradeable after costs, which this study does not answer. Three cheap tests, in order:

  1. Does a straddle or strangle bought at 1:55pm and closed at 3:30pm cover the spread, given the implied-volatility collapse that follows the announcement? The move is real; the question is whether it is already paid for.
  2. Does the 49%-of-range-remaining figure hold conditionally — on high-VIX days, on days with a Summary of Economic Projections, on days with dissents?
  3. Does the same intraday shape appear in ES and NQ futures, where the overnight session is continuous and the cost structure is different?

Method

Meeting dates
Parsed from federalreserve.gov — the per-year historical calendars for 2000–2020 and the current calendar page for 2021–2026. Two-day meetings are dated to the decision day. The cancelled March 2020 meeting is dropped; the two unscheduled 2020 calls are held out of the main sample and reported separately.
Policy action
Derived, not hand-entered. The fed funds target comes from FRED (DFEDTAR through 2008, DFEDTARU after) and each decision day is classified by comparing the target in effect before the meeting with the target a week later. This yields 40 hikes, 145 holds and 27 cuts across 212 scheduled days.
Price data
SPY. Daily bars 2000-01-03 to 2026-07-31 for the day-level work (6,684 sessions). 1-minute bars 2019-01-02 to 2026-07-21 for the intraday work (1,885 full sessions, 59 of them FOMC days), restricted to the regular session and to full trading days. The intraday sample starts in 2019 because that is when every meeting began carrying a press conference; running it further back would average two different events together.
Statistics
Differences in means are tested with a 20,000-draw permutation test and reported with a bootstrap 95% interval. Proportions use Wilson score intervals and an exact binomial test against 0.5. No result in this article is adjusted for multiple comparisons, which if anything makes the significant findings look stronger than they are.
Reconciliation
Our meeting table produces 214 decided decision days. Removing the two unscheduled 2020 calls gives 212, which matches the count in the source material that prompted this, as do the hike/hold/cut splits of 40/145/27. The independent agreement is the reason we trust both.
Limits
One instrument, one index. Nothing here is transaction-cost aware, no strategy is backtested, and no result is out-of-sample tested because nothing here is a strategy. The intraday findings rest on 59 events, which is enough for a 3× volatility ratio and not enough for anything subtler.

References

  1. Lucca, David O., and Emanuel Moench. “The Pre-FOMC Announcement Drift.” The Journal of Finance 70, no. 1 (2015): 329–371. Link
  2. Kurov, Alexander, Marketa Wolfe, and Thomas Gilbert. “The Disappearing Pre-FOMC Announcement Drift.” Finance Research Letters 40 (2021): 101781. Link
  3. Narain, Namrata, and Kunal Sangani. “The Market Impact of Fed Communications: The Role of the Press Conference.” International Journal of Central Banking 22, no. 1 (January 2026): 313–. PDF
  4. Boguth, Oliver, Vincent Grégoire, and Charles Martineau. “Shaping Expectations and Coordinating Attention: The Unintended Consequences of FOMC Press Conferences.” Journal of Financial and Quantitative Analysis 54 (2019): 2327–2353. Link
  5. Boguth, Oliver, Adlai J. Fisher, Vincent Grégoire, and Charles Martineau. “Noisy FOMC Returns? Information, Price Pressure, and Post-Announcement Reversals.” Working paper, revise and resubmit at JFQA. Link
  6. Bernanke, Ben S., and Kenneth N. Kuttner. “What Explains the Stock Market's Reaction to Federal Reserve Policy?” The Journal of Finance 60, no. 3 (2005): 1221–1257. Link
  7. Cieslak, Anna, Adair Morse, and Annette Vissing-Jorgensen. “Stock Returns over the FOMC Cycle.” The Journal of Finance 74, no. 5 (2019): 2201–2248. Link
  8. StatsEdge Trading. FOMC-day statistics presentation, September 2026. The prompt for this study; all figures here are independently recomputed. statsedgetrading.com
Historical behaviour of an index ETF, not a strategy or a recommendation. Full disclaimer.