Reference 07 October 2026 · 6 min

A plain-English guide to event windows: how to tell what a catalyst actually moved.

Written after the Wednesday October 7 close, using that session as the live example. Our plan scored hits on the dollar and gold after hawkish FOMC minutes, but from 2:00 PM to the close DXY moved 0.04 and gold about $0.30; the moves came earlier. Close-to-close tells you what happened, not why. An event window measures a fixed slice around a release, the method academic studies of Fed announcements use. How to choose the window, what Wednesday looks like measured three ways, what windows cannot see (slow reactions, priced-in news, leaks), and how we will report them from now on.

By Igor Komolov

Written after the Wednesday October 7 close, using that session as the live example. Today's analysis note graded a pre-registered plan that said hawkish FOMC minutes would lift the dollar and push gold down. Both happened: DXY closed up 0.42 at 102.25 and gold fell about $56 (Yahoo Finance). Both were scored as hits. But nearly all of both moves happened before the minutes were published at 2:00 PM ET. From 2:00 to 4:00 PM, DXY moved 0.04 and gold less than $1. The plan was right about direction, and the minutes did almost none of the work. This guide explains the tool that separates those two statements: the event window.

The problem: a day has many causes

"The dollar rose on hawkish minutes" is the kind of sentence written every week. It compares the close with the previous close and assigns the change to the biggest scheduled event of the day. That is a guess presented as a finding. In the 24 hours from one New York close to the next, the market absorbs Asian and European trading, overnight headlines, every scheduled release on both sides of the Atlantic, auction flows and month or quarter-end positioning. On Wednesday alone there were the European session, a petroleum inventory report, a 10-year auction and the minutes. Close-to-close tells you what happened. It does not tell you why.

What an event window is

An event window is a short, fixed slice of time around a scheduled release. You measure the price just before the release and again a fixed time after, and call the difference the reaction. The logic is simple: if nothing else of note happened in those minutes, most of the change in that slice belongs to the release.

This is standard practice in academic work on monetary policy. Gürkaynak, Sack and Swanson (2005) measured how bond yields responded to FOMC announcements over a 30-minute window, starting 10 minutes before the statement and ending 20 minutes after, precisely so that other news would not contaminate the measurement. The method needs intraday data, which is why it was rare outside banks until recently. Free sources such as Yahoo Finance's 15- and 30-minute bars now make a rough version available to anyone.

Choosing the window

Start just before the release. The "before" price should be the last clean price ahead of the event, not the previous close. On Wednesday the previous close for DXY was 101.83; the price at 1:45 PM, just before the minutes, was about 102.24. Using the close would have handed the minutes 0.41 that it did not earn.
End soon enough to exclude the next event. Wednesday's 10-year auction closed at 1:00 PM and the minutes landed at 2:00 PM. A window that runs 1:00 to 3:00 PM mixes the two. Use 1:00 to 1:30 PM for the auction and 2:00 to 2:30 PM for the minutes, and accept that anything after 2:30 PM is unattributed.
Long enough to let a document be read. A payrolls number is digested in seconds; a 9,000-word set of minutes takes longer. For text releases, a window of 30 minutes to an hour is a reasonable minimum. Write the window down before the event, not after, for the reason below.
Match the window to the bar size you actually have. Yahoo's 30-minute bars for the 10-year yield are stamped at :20 and :50 past the hour, so a "1:00 to 1:30" window really is "12:50 to 1:20." Say so.

Wednesday, measured three ways

Market	Close to close	Auction window (about 12:50 to 1:20 PM)	Minutes window (2:00 PM to the 4:00 PM grading bar)
10Y yield	+0.8 bp	about -4 bp	about -0.5 bp
DXY	+0.42	about -0.02	+0.04
Gold (GC=F)	about -56	about +5	about +0.3
USD/JPY	-0.02	about -0.03	+0.10

Source: Yahoo Finance query1, 15-, 30- and 60-minute bars for ^TNX, DX-Y.NYB, GC=F and USDJPY=X; close-to-close uses the same reference levels as the analysis note. Window figures are approximate because bar boundaries do not line up exactly with release times.

Read across the rows and a different day appears. The 10-year's biggest identifiable reaction was to the auction, and it was down, not up. The minutes, the event every preview focused on, produced the smallest moves on the board. The dollar and gold moved overnight and in the early New York morning, before either scheduled event.

What event windows cannot do

They miss slow reactions. Some news is absorbed over days as positioning adjusts. A narrow window understates it. That is a reason to report the window alongside the close, not instead of it.
They cannot see what was priced in. A flat reaction can mean the news did not matter, or that it was exactly what the market expected. Wednesday's minutes were close to what futures implied, so a flat window is the expected result. You need a measure of expectations, such as CME FedWatch or a survey consensus, to tell which.
They are vulnerable to leaks and pre-positioning. If traders bought ahead of a release, the move happened before the window opened.
They need a quiet window. If a headline lands inside the window, the reaction is shared and cannot be split cleanly.

How we use it from here

Our pre-registered plans are graded on the close against the previous close, and that rule stays, because changing a grading rule after seeing results is how a track record gets quietly flattered. What changes is the reporting. From Thursday October 8, each plan names a fixed window for each release before the event, and the write-up records the move in that window next to the close-to-close grade. A hit with a flat window will be reported as what it is: right direction, cause unproven.

A conditional way to use it

If a release moves a market in its event window by more than the market's typical move over a quiet window of the same length on the same day, then the release probably carried information the market had not priced. Wrong if similar-sized moves show up in windows with no news at all, in which case the "reaction" was ordinary noise. Wednesday's 4 basis point drop in the 10-year around the auction passes this test against the quiet late-morning hour, when the yield moved about 1 basis point; the 0.04 move in DXY around the minutes does not.

Where this shows up in practice

October 7: hawkish minutes, a strong 10-year auction, and yields that went nowhere (the live example)
Friday recap: NFP 29k, yields rose anyway (the opposite case: right in the first 30 minutes, wrong at the close)