Record
I used yesterday's data to predict this morning's gap — the correlation was 0.000
41.6% of a trading day's movement happens while I cannot see a screen. And that stretch does not remember yesterday.
There is a fork in the road every options trader reaches early on.
The close is coming, the position has not gone your way yet — do you hold it overnight?
Yesterday I took the "hold" branch. Today I worked out what that branch cost, and while I was at it I checked what kind of thing the overnight session actually is, using 2026 data.
The numbers turned out clearer than I expected, so I am writing them down.
First, the decision I made
The index spent the whole session grinding in a narrow band. I was holding two next-day-expiry long puts on a downside read, and as the close approached they had not gone my way.
The sentence that showed up in my head was this one.
"If I close now the loss is locked in. It looks like it should drop tomorrow — let me give it one more day."
That sentence is the whole point of this post. It was not analysis. It was hope. And I was filing it under analysis.
The next morning the index opened gap up. Both went to zero — the entire premium was tuition.
The day's own trading → slightly positive
The two carried overnight → 106% of the day's total loss
I traded well that day. I had a limit order sitting there in advance and it filled for a solid gain. The result was still negative. It had already been decided the night before.
So I measured the overnight session
The question was simple. How much support does "it looks like it should drop tomorrow" actually have?
I checked 161 trading days of S&P 500 daily bars, 5 January through 25 August 2026.
gap = today's open − yesterday's close
Everything that happened while I could not look at a screen.
⚠️ I checked the data first. Not long ago a 15-year backtest went from 85.6% to 53.8% when I reran it. The cause was not the market — the free data source had been filling in synthetic "open" prices before 2016. This calculation uses 2026 only, so it does not touch that stretch. The median gap here (0.282%) sits in the same range as the corrected figure in that post (0.346%).
1. The gap does not remember yesterday
The first thing I wanted to know. Can yesterday's information tell me anything about tonight?
| Test | correlation r | t | direction match |
|---|---|---|---|
| Yesterday's return → today's gap | −0.056 | −0.71 | 48.1% |
| Yesterday's gap → today's gap | +0.000 | +0.00 | 45.0% |
The correlation is 0.000.
Not 0.1, not 0.05. Zero to three decimal places. Direction match comes in at 45–48%, which is right around a coin flip.
💡 "It fell hard yesterday so it should fall again" or "it gapped up yesterday so tonight it gaps down" — neither one shows up as support in this sample.
2. 41.6% of daily movement happens overnight
| Session | mean absolute move | standard deviation |
|---|---|---|
| Overnight (gap) | 0.352% | 0.471% |
| Regular hours (open→close) | 0.495% | 0.650% |
Put total daily movement at 100 and the overnight session takes 41.6 of it.
A trading day's movement
├─ overnight 41.6% ← no screen
└─ session 58.4% ← I can respond
During the session a wrong read is fixable. I can close, I can roll, I can take the other side. Overnight I can do none of those.
3. Gaps are bigger than they feel
| frequency | |
|---|---|
| mean |gap| | 0.352% (median 0.282%) |
| |gap| ≥ 0.5% | 37 days (23%) — about one day in four |
| |gap| ≥ 1.0% | 9 days (6%) |
| largest | +2.08% (137 index points) |
At today's index level of 7,700 that translates to:
0.5% ≈ 38 points
1.0% ≈ 77 points
⚠️ The points in the table below are at the index level of that day. The +2.08% on 8 April was +137 points because the index was in the 6,600s then; the same percentage is about 160 points at today's level.
About one day in four opens at least 38 points away from where it closed. For a next-day-expiry option, 38 points is not a rounding error.
| Date | gap | index points |
|---|---|---|
| 8 Apr | +2.08% | +137 |
| 23 Jun | −1.42% | −106 |
| 3 Mar | −1.18% | −81 |
| 15 Jun | +1.15% | +85 |
| 17 Jul | −1.14% | −86 |
If I hold a position overnight, I take that number as it comes — in my favor or against me.
4. The gap does not tell you the day's direction either
| gap direction == session direction | 55.3% |
| correlation | +0.146 |
| gap filled (touched prior close intraday) | 55.3% |
| gap-up days where the open was the day's low | 15.5% |
| gap-down days where the open was the day's high | 4.8% |
55.3%. Five percentage points above a coin flip.
The last two rows are the interesting ones. On gap-up days, only 15.5% went straight on without giving anything back. The other 84.5% traded below the open at some point.
💡 A gap is positional information, not directional information. It tells you where the day starts. It does not tell you where it goes.
The list of things I cannot control
Once you look at the numbers the reason is straightforward. Nothing that happens overnight is something I can influence.
- Geopolitical events
- Asian and European sessions
- Earnings and guidance from Korean and Taiwanese semiconductor names
- A central banker's remark
- Commodities and currencies
- After-the-bell announcements from large caps
Holding a position overnight is exposure, not prediction. It is not "I know the direction." It is "I have opened myself to that entire list."
Sometimes the exposure breaks your way and the payoff is large. That is not evidence of skill, though. The statistic says so — r = 0.000.
What my own record says — 144 trades with a day or more to run
The statistics above are about the market. I also counted what happened in my own account.
🚨 Same-day expiries are excluded. A 0DTE contract cannot be carried, so the comparison does not apply to it, and expiring worthless is its normal ending — leaving it in contaminates the same-day-close side.
I split the 144 trades that had a day or more to run on one question: was it closed the same day?
| trades | total loss | mean return | median | win rate | |
|---|---|---|---|---|---|
| Closed same day | 116 | 0.0% (none) | +19.5% | +11.5% | 65.5% |
| Carried over | 28 | 50.0% (14) | −41.9% | −98.6% | 32.1% |
Not one of the 116 same-day closes went to a total loss. Half of the 28 carried ones did.
For reference, the 62 same-day-expiry trades I removed ran a 41.9% total-loss rate and −28.4% mean — that story is in buying a same-day expiry.
Splitting those 28 further
Outcomes inside "carried over" are not uniform. Three groups:
| trades | share | mean return | range | |
|---|---|---|---|---|
| A. Profit | 9 | 32.1% | +76.3% | +1.7% to +206.7% |
| B. Loss, partly recovered | 5 | 17.9% | −92.0% | −97.2% to −79.1% |
| C. Expired worthless | 14 | 50.0% | −100.0% | — |
Two things show up here.
1. When it works, it works big. Nine trades averaging +76.3%, topping out at +206.7%. That is what stays in memory. And it is why the next one gets sized larger.
2. But on the losing side, B and C are barely distinguishable. Even B — "partly recovered" — averages −92.0%, and its best case is −79.1%. All 19 that went against me lost 79% or more.
🔑 A carried position splits into "wins big" or "loses nearly everything." There is no middle. The three groups together come to an expected −41.9%.
⚠️ This is not a causal reading. Many of those 28 carried-over trades were already underwater, which is why they did not get closed. It may not be "carrying it made things worse" so much as "things were already worse, which is why it got carried." The sample is also small at 28.
Either way the conclusion lands in the same place — do not be standing there.
So here is the standard I set
At the end of this calculation I wrote one rule.
What I open today, I close today.
Not because the night frightens me. Because of the arithmetic.
1. Overnight movement is not predicted by yesterday's data (r = 0.000)
2. Overnight is 41.6% of daily movement
3. I can take no action during that stretch
──────────────────────────────────────────────────
→ unpredictable + large + unactionable
The rule has a side effect that may matter more than the rule itself.
It removes the occasion to say "it looks like it should drop tomorrow."
What I actually did yesterday was defer a decision, and the sentence was what the deferral needed. With a same-day-close rule, there is nowhere for that sentence to stand.
Three lines
1. "It looks like it goes up / down tomorrow" may not be analysis. If that sentence only arrives while a position is underwater, the test I use is whether the same conviction shows up when I am flat. The answer separates the two.
2. Overnight is 41% of the movement, and during it you are in the audience. In the session you are a player. Overnight you are watching.
3. The overnight trade that paid off big is the one to be most careful with. Winning once in a game where r = 0.000 is a sample of one. Memory files it as skill anyway.
What I do not know yet
This covers 161 trading days in 2026 and nothing more. Next:
- Whether gap size relates to that day's range — right now no clear relationship shows
(days with gap
<0.1%average 0.893% range vs 1.112% for gap>1%) - Whether the night before a data release distributes differently
- Whether r = 0.000 survives when the sample extends across more years — it may not
I will post what comes back.
How I choose expiries is written up separately in choosing an expiry for a long put.
This post does not recommend any security or strategy. It is a personal record of observations. Investment decisions and their outcomes rest with the investor.
How was this to follow?
Knowing where it got hard is what lets me fix the next one. No name, no email.
Get the One-Page Options Field Guide
A strategy selector for bull, bear, and sideways markets, a pre-order checklist, and a glossary.
If this was useful
Why gamma, call walls, and put walls behave the way they do comes down to the structure of the options market. These books lay out that structure in order.
- Strategic US Options Trading I: Fundamentals — Start here if options are new — from reading the chain
- Strategic US Options Trading II: Strategies — When you want to actually place the order
- Strategic US Options Trading III: Advanced Strategies — When you want cash flow in a sideways market
Earlier posts live on Tistory. I'm moving them here a few at a time. optiontrading.tistory.com
Questions?
There are no comments or a guestbook here. Email me with questions about a post or to report an error — I read and reply. I don't give individual investment advice.
hello@optionphoenix.com