The Last 14 Minutes: Why We Stopped Looking at Options Closing Prices

What happened was the front leg's closing quote: it blew out to $4.70 wide — and dragged the computed mid far enough to nearly double the measured debit, and with it the relative volatility reading: 0.55% became 1.00%.

Two trading days later, on Monday, August 10, the close showed the other way a late snapshot drifts. RV always measures whichever strike is at-the-money at the moment of the snapshot — that's by design. In the final minutes the stock moved about half a point, so the closing snapshot's ATM landed one strike higher (275 instead of 270), where the calendar is structurally smaller. Fine so far. But then the closing quotes — more than $4 wide on the front leg — squeezed that calendar's measured value to $0.10, when the pre-close market had priced the same 275 calendar at $0.40. Stack the two effects and the close printed an RV of 0.10% against the pre-close reading of 0.50%. A jumpier anchor, measured through sleepier quotes.

None of these numbers was a bug — we reproduced all four from the raw quote records before publishing this. They are what the two snapshots honestly reported. The difference, in every case, was fourteen minutes.
 

Notice what the two distortions have in common besides their timestamp: one direction each way. The close didn't lean bullish or bearish on volatility — once too high, once effectively zero. And the Monday case generalizes: across the symbols we have re-verified so far, on about 1 in 18 entry days (5.6%) the closing snapshot anchored to a different at-the-money strike than the pre-close market did. Once you understand why the final minutes can do that, you understand something important about every backtest, every screener, and every "is this expensive?" judgment you have ever made from end-of-day options data. Note: This feature of “4pm compare” retires from the app after a couple of months.

What actually happens at 4:00pm

Most of us grew up treating the closing price as the official price — the number the newspaper printed, the mark your broker settles to. For stocks, that is roughly fair: the closing auction concentrates enormous volume into a single, hard-to-manipulate print.

Options are different. An option's "price" on an end-of-day file is usually the midpoint between the last bid and the last ask — and the final minutes of the session are precisely when that midpoint is least trustworthy. Market makers carry every open position overnight, through whatever news lands after the bell, so as the close approaches they protect themselves the only way they can: they widen. Quotes that sat a few cents apart at 2pm can be several times wider at 3:59. Attention shifts to the closing auction in the underlying shares. Some quotes go stale; a few cross or collapse to nonsense for a moment. The market is not wrong at 4:00pm — it is simply half-asleep, with one eye on the exit.

Measure a relative-volatility ratio, a spread-cost figure, or a backtest fill from that half-asleep market and the noise flows straight into your numbers — twice over. The wide quotes wobble every mid. And because "at the money" is defined by where the underlying sits at the snapshot, a late drift can hand the measurement a different anchor strike — correct by definition, but jumpier in the noisiest minutes of the day. Most of the time the distortion is modest. Sometimes — as our ADBE chart shows — it is enormous. And you cannot tell which kind of day you are looking at from the number alone.

This is not a novel observation; it is why professional options-data vendors have long sampled the market before the close rather than at it. The industry's quiet consensus is that a quote captured while market makers are still competing beats a quote captured while they are packing up.
 

When the close lies most: look at the calendar

Now look at when our two ADBE distortions happened, because the timing is not random. The August 6 spike was the Thursday close immediately before Friday morning's jobs report. The August 10 crater sat at the Monday close heading into CPI week. Both distortions landed at closes adjacent to major macro releases — which is precisely when you would expect the final minutes to be at their worst. A market maker going home flat-footed into a jobs number or a CPI print has every incentive to quote defensively into that bell; the overnight risk they are pricing is at its maximum. The quotes are widest, the mids wobbliest, exactly on the evenings when an earnings trader most needs a trustworthy reading — because those are also the evenings you are deciding whether tomorrow's setup is cheap.
 

This is why the RV charts draw the macro calendar directly onto the time axis — the JOBS, CPI, PPI, FOMC and PCE flags you see running along the top of every chart. A strange-looking reading next to a macro flag has a candidate explanation you can see at a glance; a strange reading in the middle of a quiet week deserves more suspicion. The pre-close snapshot removes most of the distortion. The macro overlay tells you where to be careful about what remains. The two features were built separately, but they answer the same question from two sides: can I trust this number, tonight?
 

Why this matters more for earnings traders than anyone else

Earnings trades live and die on small edges measured over short windows. When we ask "is this straddle cheap against its own history?", we are comparing today's price to a median built from dozens of past snapshots. If a meaningful fraction of those snapshots were taken from a distorted market, the median itself is polluted — and a setup can look cheap or rich for no reason except closing-quote noise on a handful of historical days.
 

It also matters at the moment of action. Members who use the Daily Screener know the workflow: the screener surfaces the day's candidates, you pick your structure, and then you have to actually get filled. Under 4:00pm data, there was always an honest gap between the reference price on the screen and the market you would meet the next morning — the printed mid sat inside a spread you knew had been artificially wide. In practice, many of us (myself included) compensated by scaling in: several limit orders at different prices, using the fills themselves to discover where the real market was. That works, but it is a workaround — you are burning patience and partial fills to rediscover information that a better measurement would have given you upfront.
 

A snapshot taken fourteen minutes earlier, while the quote competition is still alive, closes much of that gap. The reference mid is nearer to a price someone would actually trade with you. Spread-cost readings reflect the market you can participate in rather than the market's closing yawn. And the historical medians your judgment leans on are built from cleaner raw material, cycle after cycle.
 

To be precise about what this does not do: it does not make any strategy win more often, and it does not turn a mediocre setup into a good one. It makes the measurements honest. What you do with honest measurements is still up to you.
 

What we changed, and how you can audit it

Starting this cycle, EarningsStudy computes relative volatility from a 3:46pm pre-close snapshot instead of the 4:00pm close. The migration is deliberately gradual — current-cycle symbols with earnings in the next 30 days first, earlier years following after verification — because when you change the ruler, you re-measure everything carefully before you trust it.
 

And rather than swapping numbers quietly, we built the audit into the product. On the Calendar and Straddle pages you will find a 4:00pm compare checkbox: solid lines show the new pre-close data, dashed lines show the legacy close-based data for the same symbol, strategy, and cycle. On most names, most days, the two hug each other — reassurance that the history you have been using was broadly sound. And then there are days like ADBE's August 6 and August 10, where the dashed line spikes to a peak, then falls off a cliff, that the pre-close market never showed. Every one of those divergences is a day the old ruler would have misled you — and every one is now inspectable, down to the raw quotes, because we keep both snapshots until the verification is done.
 

One honest caveat while the migration runs: until the historical backfill completes, some today-versus-history comparisons mix a 3:46pm present with a 4:00pm past. The differences are usually small, but they are not zero — which is exactly why the compare view exists and the legacy snapshot is preserved until every cycle is re-verified. If you find a divergence that looks strange, the Ask button inside the app sends it straight to me with the page attached; several members' reports have already sharpened the verification.
 

The part I care most about

This upgrade exists because a member insisted on it. @Bhavan1986 made the case that the closing snapshot wasn't good enough — and he was right. When I brought it to the contributors, @Kim, @Yowster and @TrustyJules backed it without hesitation. The better data roughly doubles what we pay for market data, and we are absorbing that rather than passing it on: nothing changes about any member's current rate. That is what the SteadyOptions–EarningsStudy partnership is for — the tools this community trades with should keep getting more honest, not just bigger.
 

It sits in a line with the spread-cost column, the confirmed-date badge, and the per-cycle detail behind every median: the platform's job is to show you numbers you could actually have acted on, not numbers that flatter a backtest. Measuring the market while it is still awake is one more step in that direction.

SO members on the All-services bundle: full core access to EarningsStudy is free through the partnership — sign up at earningsstudy.com with your SteadyOptions email. The rest can subscribe at an incredible introductory price of $39.99.
 

EarningsStudy provides educational research and market information only — not investment, financial, or trading advice. Options trading involves substantial risk. Backtests and calculations may contain errors and should be independently verified.