Why volatility falls after earnings, worked with Amazon’s July report

One day, a lot of variance

On an ordinary session a large stock moves for many small reasons: index flows, sector news, rates. On the session after an earnings release it moves mainly for one reason, the new information in the results, and that move is often several times the size of a normal day.

Volatility is measured in variance terms (the square of returns), so one large day weighs far more than its share of the calendar. A 15% day is not three times a 5% day in variance terms; it is nine times. That squaring is why a single earnings session can account for most of a month’s measured volatility.

The same logic is behind what options traders call the post-earnings volatility collapse. Options whose expiries span a scheduled release are priced to include the expected earnings-day move, so the volatility priced into those near-dated options typically rises into the event and falls once it has passed. We do not print option-derived volatility figures here; the realized numbers below show the same mechanism in data anyone can rebuild from daily closes.

Amazon on 30 and 31 July 2026

Amazon filed a Form 8-K under Item 2.02 (results of operations) on 30 July 2026. The SEC’s filing index shows it accepted at 16:06:23 ET, after the 16:00 close of the US cash session, so the first session that could trade on the results was 31 July.

Amazon closed at $235.50 on 30 July and at $271.58 on 31 July (Yahoo Finance daily closes). That is 271.58 ÷ 235.50 − 1 = +15.32%, or a log return of ln(271.58 ÷ 235.50) = 0.1425.

For the 20 sessions up to 30 July (returns from 2 July to 30 July, starting from the 1 July close), Amazon’s daily standard deviation was 1.82%, and its 20-day realized volatility was 1.82% × √252 = 28.84% annualized. Measured against that pace, the earnings-day log return was 0.1425 ÷ 0.0182 ≈ 7.85 standard deviations.

One session later, the 20-day window ending 31 July contains the earnings day, and Amazon’s 20-day realized volatility for that window was 58.69%, roughly double the figure a day earlier. Nineteen of the twenty returns in it are the same calm July sessions; one return changed the number.

The event-variance arithmetic

A common way to describe a window that contains a scheduled event splits its variance in two: total variance = normal daily variance × number of normal days + event variance. The split can be checked on Amazon’s numbers.

Normal daily variance, taken from the pre-release window, is 0.0182² ≈ 0.000330. The event variance is the square of the earnings-day log return: 0.1425² = 0.020319. The window ending 31 July has 19 normal days and one event day, so the split predicts a total of 19 × 0.000330 + 0.020319 = 0.006270 + 0.020319 = 0.026589.

The actual sum of the 20 squared log returns in that window was 0.026607, within 0.1% of the split. Dividing by 20 gives an average daily variance of 0.001330; its square root is 3.65%, and 3.65% × √252 ≈ 57.9% annualized. Our 58.69% figure is slightly higher because the standard-deviation formula subtracts the window’s average return and divides by 19 rather than 20.

Two ratios come out of this. The earnings day supplied 0.020319 ÷ 0.026607 = 76.4% of the window’s summed squared returns. And 0.020319 ÷ 0.000330 ≈ 62, so that one session carried as much variance as about 62 ordinary July sessions, roughly three months of trading at the pre-release pace.

Options pricing uses the same split in the other direction. Before a release, the event term is an expectation rather than a measured return, and it is part of the total variance priced into every expiry that spans the event. Once the release has passed, the event term drops out of the remaining life of those options, and only normal daily variance is left.

The cliff when the event leaves the window

A trailing 20-day measure keeps the earnings day for exactly 20 sessions. Amazon’s 20-day realized volatility stayed between 58.29% and 62.24% from 31 July to 27 August, even though the stock’s day-to-day moves in August were mostly ordinary.

On 28 August the 31 July return dropped out. The window ending 28 August runs from the 31 July close to the 28 August close and contains no earnings day; its realized volatility was 30.81%, against 58.29% one session earlier. Amazon itself rose 3.97% that day (266.43 ÷ 256.26 − 1), so the drop in the measure had nothing to do with a quiet session. It was the window, not the stock.

The post-release level of 30.81% is close to the 28.84% measured before the release. After the earnings day is set aside, Amazon moved at a similar pace in August as in July.

Volatility priced into near-dated options moves at the opposite end of the timeline. It falls when the event passes, because it looks forward; trailing realized volatility rises when the event passes and falls 20 sessions later, because it looks back.

Three more releases, three different sizes

The earnings-day move varies a great deal between companies and between quarters. The table measures each reaction-session move in units of the stock’s own daily standard deviation over the 20 sessions before it.

Apple and Alphabet followed Amazon’s pattern on a smaller scale. Apple’s 31 July move of −7.35% was 4.25 daily standard deviations; Apple’s 20-day realized volatility for the 20 sessions after it, from 3 August to 28 August, was 18.95%, below the 28.52% measured before the release. Alphabet’s 23 July move of −6.89% was 3.31 standard deviations, and its realized volatility over the following 20 sessions was 37.59%, above the 34.29% before, so August was not calmer for every stock.

Goldman Sachs released results before the US open on 13 October 2026 (its Item 2.02 8-K was accepted at 07:32:56 ET), so its reaction session is 13 October itself. The stock closed at $893.95 on 12 October and $891.91 on 13 October, a −0.23% move, or 0.16 daily standard deviations. In that quarter the release added almost nothing to Goldman’s measured volatility. A scheduled release only produces a large event term if the results move the price.

• “Before” is the 20 returns ending at the last close before the release; “after” is the 20 returns starting the session after the reaction session.
• Goldman’s “after” window has not completed: only five sessions had traded by 20 October 2026, so it shows “—”.
• Alphabet’s filing covers both share classes; the figures use the Class C shares (GOOG).

What the numbers do and do not say

Four releases are not a sample. They show the mechanism, not how large a typical earnings move is, and they say nothing about the direction of the next one. Amazon rose and Apple and Alphabet fell; the variance arithmetic treats all three the same, because squaring removes the sign.

Two cautions apply to the realized measures. A 20-day figure that contains an earnings day describes one large move plus nineteen normal ones, not a stock that moved 3.7% a day for a month. And an Item 2.02 filing is not always the scheduled quarterly report: companies also file under that item for preliminary results or other disclosures, so the filing date and time are worth matching to the actual release before treating a session as the earnings day.

How 20-day realized volatility is computed, and how the square-root-of-time rule turns a volatility into a price range, is covered at /articles/realized-volatility-and-expected-move-maths. How option traders use expected moves and IV rank in spread construction is covered at /articles/options-trading-expected-move-iv-rank-vertical-spreads.

Check the numbers yourself

Release dates and times come from SEC EDGAR. A company’s submissions file on data.sec.gov lists every 8-K with its items, and results are filed under Item 2.02; the filing index page for each 8-K shows the acceptance time in Eastern time. Daily closes come from Yahoo Finance’s history for each ticker.

In a spreadsheet, put the closes in a column, compute ln(close ÷ previous close) in the next, and apply STDEV.S to 20 of those returns, then multiply by SQRT(252), to reproduce each realized figure. Square the earnings-day return and divide it by the sum of the 20 squared returns in its window to get the event share. Our /earnings page lists upcoming release dates with their sources.