Most traders size one position at a time. Check the gap history, set the risk budget, decide how many shares to carry, move on to the next idea. Do that five times and you have quietly made an assumption you never wrote down: that those five positions will gap independently of each other. They almost never do — and overnight gap risk is a portfolio-level number, not five position-level numbers that happen to share an account.
Right now that assumption is more dangerous than usual, because the options market is pricing stocks to move more independently than at almost any point in the last two decades.
What the options market is currently pricing
Cboe publishes an implied correlation index — the average correlation between S&P 500 constituents that option prices imply. As of the September 8, 2026 close, the three-month version (COR3M) read 9.9 against a long-run mean of 41, not far off the record low of 7.2 set on July 10, 2026. The companion measure, the spread between average single-stock implied volatility (VIXEQ, 37.9) and index implied volatility (VIX, 15.7), sat at 22.2 points — the 96th percentile since 2014.
Translated out of options language: the market expects individual stocks to swing hard and cancel each other out, leaving the index calm. That is a real regime, and it is why index-level volatility looks so subdued. It is also, by construction, a peacetime reading. Implied correlation prints its lows in melt-ups, never in panics.
The math on a five-position overnight book
Take a $100,000 account. You hold five overnight positions, each sized so that a typical adverse gap costs you about $1,000 — one percent of the account per name. That feels like disciplined risk management, and at the single-position level it is. The question is what the five of them add up to, because overnight gap risk does not add the way most traders assume it does.
The formula, in one line
For n equally sized positions each carrying gap risk σ, with average pairwise correlation ρ, the combined gap risk is σ × √(n + n(n−1)ρ). With n = 5 and σ = $1,000, the whole range looks like this:
| Average correlation (ρ) | Combined gap risk | % of account | vs. independent case |
|---|---|---|---|
| 0.00 (fully independent) | $2,236 | 2.24% | 1.00× |
| 0.10 | $2,646 | 2.65% | 1.18× |
| 0.25 | $3,162 | 3.16% | 1.41× |
| 0.40 | $3,606 | 3.61% | 1.61× |
| 0.70 | $4,359 | 4.36% | 1.95× |
| 1.00 (everything moves together) | $5,000 | 5.00% | 2.24× |
Two things jump out. First, the independent case — the one most traders implicitly assume — is the smallest number on the table, and it is less than half the fully correlated case. Second, the damage is front-loaded. Going from ρ = 0 to ρ = 0.25 costs you more added risk than going from 0.70 to 1.00. You do not need a crisis to lose most of the diversification you thought you had. You need a little correlation.
How fast ρ can move: the March 2026 example
This is not a hypothetical. In the second week of March 2026, a geopolitical escalation hit an equity market that had been happily pricing low correlation. Cboe’s one-month implied correlation index (COR1M) went from roughly 15 at the end of February to around 40 by late March. The dispersion trade built on low correlation had one of its worst months in over a decade — a JPMorgan index tracking the strategy fell 4.9%.
Map that onto the table. A book priced for ρ ≈ 0.10 found itself in a ρ ≈ 0.40 world in a matter of weeks. That is the second row moving to the fourth: combined gap risk up roughly 36%, with no change whatsoever to your position sizes. You did not add risk. The market added it for you.
The defining feature of a correlation shock is not the size of the decline. It is the speed at which the things you own stop behaving like separate decisions.
The asymmetry that makes it worse
Correlation does not rise symmetrically. Ang and Chen’s work on asymmetric correlations of equity portfolios (Journal of Financial Economics, 2002) found that correlations between US stocks and the broad market are substantially higher on downside moves than on upside moves — and that conditional on the downside, observed correlations exceeded what a normal distribution implies by 11.6%.
For an overnight trader that has a specific, uncomfortable meaning. Your winning nights will tend to be the ones where correlation stayed low and your positions did their own thing, which will feel like proof that your diversification is working. Your losing nights are disproportionately the ones where everything moved together. The good outcomes are spread out; the bad ones arrive in a single package. This is the same structural problem behind several of the swing trading mistakes that only show up in a trade log rather than in any individual trade thesis.
Three adjustments that follow from the table
- Set the risk budget at the account level first, then divide. If 3% of the account is the most you are willing to lose overnight, that is the ceiling for the book, not for each name. Under a plausible stress correlation of 0.4, five $1,000-risk positions already breach it. Either carry fewer names or cut the per-name number — the arithmetic does not offer a third option. This is the portfolio-level companion to sizing a single position off the gap rather than off the stop.
- Count your real positions, not your tickers. Four megacap technology names and one semiconductor ETF is not five positions. Ask what single piece of news would move all of them at once. If the answer comes easily, you hold one bet in five wrappers.
- Assume ρ is higher than it looks before an event. Scheduled macro catalysts are correlation events by definition — a CPI print or a policy decision reprices everything at once. Size those nights off the stressed row of the table, not the comfortable one, and treat the overnight gap risk of the whole book as the number that matters.
Structure is what makes this survivable, because none of it is intuitive at 3:58 p.m. with a position you like. A written account-level ceiling, decided before the close, is worth more than any single read on direction. SpyTomorrow’s nightly pressure readings are built to support that kind of pre-commitment — a structured input into the size-and-carry decision, not a signal to follow blindly, and never a substitute for knowing what your whole book does if the tape re-correlates.
What this analysis does not claim
Implied correlation is not a forecast. Extreme dispersion readings have not reliably preceded weak forward returns, and low correlation can persist for a very long time — record lows show up in long melt-ups far more often than at tops.
It is also worth being precise about what the index measures. COR3M and COR1M describe implied correlation among S&P 500 constituents over one- to three-month windows. They are a regime gauge, not a literal ρ for your particular five positions over one particular night, and the table above is a sensitivity analysis rather than a prediction. Use it the way you would use the six indicators that describe overnight risk better than the VIX: as context for how much size the current environment justifies, not as a timing tool.
The practical takeaway on overnight gap risk is narrow and durable. When correlation is priced near zero, the calm you are looking at is manufactured by offsetting single-name moves, not by an absence of risk. Your five positions are only five positions while that holds. Size as though it might not.
This article is for educational and informational purposes only. It does not constitute financial advice, investment recommendations, or a solicitation to buy or sell any securities. Past performance and historical data do not guarantee future results. All trading involves risk, including the potential loss of principal. Always do your own research and consult with a qualified financial advisor before making trading decisions.
