Measuring SPY overnight returns vs intraday returns means splitting every trading day into two pieces. The overnight piece runs from yesterday’s 4:00 p.m. close to this morning’s 9:30 a.m. open. The intraday piece runs from that open to today’s close. Every dollar of SPY’s price return since 1993 sits in one bucket or the other — and the two buckets look nothing alike.
We pulled daily SPY open and close prices from January 29, 1993 through August 26, 2026 — 8,450 overnight sessions — and compounded each bucket separately. The result is one of the more lopsided statistics in US equities, and it is also more fragile than the version you usually see quoted.
SPY overnight returns vs intraday: the headline number
| Measure | Overnight (close → open) | Intraday (open → close) |
|---|---|---|
| Compounded price return, 1993–2026 | +1,267% | +28% |
| Growth of $1 | $13.67 | $1.28 |
| Average session return | +3.3 bps | +0.8 bps |
| Median session return | +5.6 bps | +4.5 bps |
| Standard deviation per session | 0.67% | 0.96% |
| Sessions closing positive | 54.9% | 52.3% |
The two buckets multiply back to SPY’s full price return of roughly 17.4x over the period, which is the arithmetic check that the split is clean. And it means the overnight session did essentially all of the compounding while the regular session — the one everybody watches — added about a quarter of a dollar per dollar in thirty-three years.
Two methodology notes, because they cut in opposite directions. This is price return only, so dividends are excluded — and since SPY trades ex-dividend at the open, those distributions are actually subtracted from the overnight bucket. The overnight figure above is, if anything, understated. On the other side, nobody captures this cleanly: harvesting the overnight bucket means two round trips a day for 8,450 days, and no spread or commission is priced into the table.
Why the medians matter more than the totals
Look again at the median row. The typical overnight session returns +5.6 bps and the typical intraday session returns +4.5 bps. That is a rounding error apart. The enormous gap in cumulative return is not built out of a thousand small overnight wins — it is built out of the tails.
Both distributions have means well below their medians, meaning both are dragged down by large negative outliers. The intraday bucket is dragged much harder. Put plainly: the regular session is where the market does most of its violent selling, and the overnight session, on average, quietly walks it back. The three largest overnight moves in the sample were all downside — March 16, 2020 (-10.5%), October 24, 2008 (-8.3%), and September 17, 2001 (-8.2%) — so the overnight bucket is not a safe harbor. It just has a fatter left tail less often.
For anyone holding a position through the close, the frequency numbers are the practical ones: 27.2% of nights produced a gap larger than 0.5% in absolute terms, 8.7% produced a gap larger than 1%, and 1.8% produced a gap larger than 2%. Roughly one night in eleven moves your position more than a percent before you can touch it. That is the number that should drive how you size an overnight position around gap risk, not the average.
The decade-by-decade record is much lumpier
Here is where the popular version of this statistic falls apart. Split the sample into five-year blocks and the “overnight always wins” story becomes a story about a few specific stretches.
| Period | Sessions | Overnight | Intraday |
|---|---|---|---|
| 1993–1994 | 485 | +17.6% | -11.8% |
| 1995–1999 | 1,263 | +195.2% | +9.2% |
| 2000–2004 | 1,256 | +4.3% | -21.1% |
| 2005–2009 | 1,259 | +23.4% | -25.3% |
| 2010–2014 | 1,258 | +37.1% | +34.5% |
| 2015–2019 | 1,258 | +36.1% | +15.0% |
| 2020–2024 | 1,258 | +41.8% | +28.4% |
| 2025–2026 YTD | 413 | +15.6% | +13.0% |
The late 1990s did most of the heavy lifting. Strip out 1995–1999 and the overnight edge is real but far from a money machine. In 2010–2014 the two buckets were nearly identical. And the single-year record is noisier still: in 2022 the overnight bucket lost 14.7% while intraday lost only 5.6%; in 2023 intraday returned +19.2% against overnight’s +4.3%; then in 2024 it flipped hard the other way, +22.3% overnight against +0.8% intraday.
A statistic that holds over thirty-three years and reverses over eighteen months is a description of market structure, not a trade.
What the academic work adds — and where it pushes back
Bondarenko and Muravyev, publishing in the Journal of Financial and Quantitative Analysis in 2023, took the same question to tick-level E-mini S&P 500 futures from 2004 to 2018 and found something sharper than a night/day split. In their data, four hours around the Asian close and European open — roughly 11:30 p.m. to 3:30 a.m. ET — accounted for the entire average market return: 7.6% annualized with a Sharpe ratio of 1.67. The remaining twenty hours averaged about -0.8%. Their explanation is uncertainty resolution: European participants arrive and price the information that accumulated while the US slept.
The pushback is equally worth reading. Matthew Bartolini’s analysis of the SPY version of this strategy concluded that once bid-ask spreads and commissions on a twice-daily trade are counted, the statistical robustness of the outperformance is low enough that it behaves more like a random walk than a repeatable edge. Both things can be true: the drift is real in the data, and it is not something a retail trader extracts by mechanically buying every close.
Three things this actually changes about an overnight plan
- Direction is not your main variable. If the base rate for an overnight session closing positive is 54.9%, a directional coin-flip is barely a coin flip. Almost all of the outcome dispersion comes from magnitude, not sign.
- Your holding window has a built-in tilt, and it is small. Three basis points of average drift does not fund a position. It is a tailwind, not a thesis — and it disappears entirely in stretches like 2022.
- Costs scale with frequency, not with conviction. The academic critique of the overnight anomaly is a cost critique. Every additional night you trade for no reason moves you closer to the version of this that does not work.
None of this says the overnight session is where you should be. It says the overnight session is where the market’s price discovery has historically been concentrated — which is a useful thing to know before you decide whether a given night is worth holding through. Our nightly pressure readings are meant to help with exactly that decision, framing whether the current setup argues for holding or standing aside; they are decision support, not a signal to follow blindly. Structure matters more than any single reading, which is why it helps to understand what actually sets SPY’s opening price before you commit capital to it.
Where this fits with the gap data
Framing SPY overnight returns vs intraday returns answers a different question than gap-fill studies do. Fill studies ask what happens to a gap after the open; this one asks how much of the total return lives in the gap at all. Read together, they tell you the overnight window carries the drift and the intraday window carries the mean reversion — a framing worth holding alongside the gap-fill rates by size bucket we covered earlier this month.
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.
