100 liquid US stocks / 2004–2026
Do most stock-market gains happen overnight?
Every trading day, we tested buying 100 liquid US stocks at the close and selling them at the next open. We repeated the test from 2004 through June 2026, then compared it with holding the same stocks.
Most stocks rose overnight. Buying every close and selling every open still loses to holding after realistic trading costs.
Without costs, $1,000 grew to $20,551. At 0.10% for each night's buy and sale, it fell to $71.68. Holding the same stocks grew to $3,154. We do not have actual broker fills for this test. A published experiment using 85,000 real retail market orders found costs from 0.07% to 0.46% per round trip, so 0.10% is plausible—but it is not measured for this exact trade.
01 / Did it work year after year?
Before costs, the overnight trade made money in 20 of 22 years. At 0.10% a night, it made money in only 2.
The overnight price rise was common. Repeating the trade every day made the cost compound until almost every year turned negative. Choose a cost below to compare the trade with holding the same stocks.
The overnight trade pays the selected cost every day. Holding pays it only when the quarterly stock list changes.
At no added costs, the overnight trade made money in 20 of 22 years.
At 0.02% each night, the overnight trade made money in 17 of 22 years.
At 0.05% each night, the overnight trade made money in 10 of 22 years.
At 0.10% each night, the overnight trade made money in 2 of 22 years.
At 0.15% each night, the overnight trade made money in 1 of 22 years.
At 0.20% each night, the overnight trade made money in 1 of 22 years.
How we checked whether the effect was widespread
We divided the history into five separate periods: 2004–2008, 2009–2013, 2014–2018, 2019–2023, and 2024–June 2026. For each stock, we calculated one overnight result for every year it was in the basket. In each period, 60.5% to 70.5% of the yearly stock results made money overnight.
The yearly comparison above treats the 100 stocks as one basket. The table below counts each company separately. Both checks use no trading costs because they measure whether the price pattern existed, not whether a trader could keep it.
| Period | Yearly stock results | Results that made money overnight |
|---|---|---|
| 2004–2008 | 527 | 60.5% |
| 2009–2013 | 526 | 70.5% |
| 2014–2018 | 524 | 70.2% |
| 2019–2023 | 511 | 63.8% |
| 2024–June 2026 | 284 | 68.7% |
02 / What drove the gain?
Remove the 20 best dates and the overnight trade loses to holding—even before costs.
If we add every individual stock's overnight gain and loss, the best 1% account for 98.1% of the net total. The pattern was widespread. The giant payoff depended on a few dates.
How we measured the effect of extreme nights
We ranked every individual stock trade by its gain from one close to the next open. We measured how much of the total gain came from the biggest 1%. We also recalculated the account after removing the best 1, 5, and 20 dates.
03 / Do trading costs wipe it out?
At 0.05% a night, the overnight trade already loses to holding.
At 0.02% a night, $1,000 grew to $6,629 and beat holding. At 0.05%, it grew to only $1,214 while holding reached $3,208. The overnight trade needed each night's total cost to stay below about 0.033%—33 cents for every $1,000 traded—to beat holding.
We do not have actual broker fills for this test. The closest useful published evidence is an experiment using 85,000 real retail market orders. It found round-trip costs from 0.07% to 0.46% across six brokerage accounts, before commissions. Those trades were placed during the day, not at our exact close and open. That makes our 0.10% assumption plausible, but not proven.
| Cost charged each night | Buy at close, sell at open | Buy and hold |
|---|---|---|
| 0.00% | $20,551 | $3,263 |
| 0.02% | $6,629 | $3,241 |
| 0.05% | $1,214 | $3,208 |
| 0.10% | $71.68 | $3,154 |
| 0.15% | $4.23 | $3,101 |
| 0.20% | $0.25 | $3,049 |
What the cost includes—and what we still do not know
The cost is not a commission. It stands in for the gap between the buy and sell quotes, price movement before each order fills, and any broker or exchange fees. The correct number changes with the stock, order size, broker, and time of day.
Published research gives us a realistic range for ordinary retail market orders. Our minute-price check below measures what happened near the times we wanted to trade. Neither one records what this exact strategy would have received in a real brokerage account.
04 / Does the daily trade still work with recent prices?
The overnight effect survived selling one minute after the open. Holding still made far more.
From August 2024 through August 2026, buying at 15:59 and selling at 09:31 turned $1,000 into $1,641 before added costs and $1,014 after our 0.10% assumption. Holding turned the same $1,000 into $1,812 before costs and $1,804 after costs.
These are real consolidated market trades, not broker fills. The move from published close/open prices to the 15:59 and 09:31 prices cost about 1.6 basis points per planned night. That measures the timing difference. It does not include the bid-ask spread or tell us what a broker would have filled.
| Prices used | No added costs | After 0.10% each night | Nights skipped |
|---|---|---|---|
| Published close to published open | $1,780 | $1,080 | 0 |
| Buy at 15:59, sell at 09:30 | $1,692 | $1,037 | 9 |
| Buy at 15:59, sell at 09:31 | $1,641 | $1,014 | 17 |
| Buy and hold the same stocks | $1,812 | $1,804 | 0 |
How we checked the timing—and the deliberately bad case
We used minute prices for 20 liquid stocks from August 2024 through August 2026. We compared the published close and open with buying in the 15:59 minute, selling in the 09:30 minute, and waiting until 09:31. Early closes without a 15:59 minute were skipped and counted below.
In a deliberately harsh check, we bought at each 15:59-minute high and sold at each 09:30-minute low. The account fell from $1,000 to $154 before added costs and $94.54 after 0.10%. We do not expect fills that bad. It shows how far a result can move when prices inside each minute are chosen against the trader.
05 / What does this prove?
We can say stocks often rose overnight. We cannot say trading every close and open beats holding.
The price pattern appeared across many stocks and survived a one-minute delay in the recent test. But a plausible retail cost made the daily trade lose to holding, and a few extreme dates produced most of the no-cost payoff.
What this test can and cannot prove
Most stocks made money overnight across the years we checked. SPY and QQQ also made money overnight. But rare jumps produced 98% of the net overnight gain added across individual stock nights. Realistic costs made the daily trade lose to holding.
To call this a usable strategy, we would need to write the exact rule before seeing the result, run it on untouched future data, and record real broker fills. We designed this test after the overnight effect was already known, so it cannot prove that the trade will keep working.
More details
How we got the numbers
Which stocks we used and how we handled missing prices
At the start of each quarter, we picked the 100 eligible US common stocks with the highest dollar trading volume over the previous 60 sessions. We kept companies that were later acquired, delisted, or failed. If a ticker was later reused by a different business, we treated it as a different company.
We measured the move from the close to the next open and from that open to the next close. Together, those two moves had to equal the full daily return after splits and dividends. If a stock disappeared before the next session and no price existed, our main result treated it as worthless. A separate check used its last closing price instead.
Where the prices came from and what they cannot show
We combined daily and minute market records from several sources and checked that overlapping prices matched. The full test contains 565,700 stock-nights. The recent minute check covered 41 tickers. Minute records show completed market trades, not bids, asks, or broker fills. They cannot prove what an account would have paid or received.
The test began with an X discussion about whether the MU result generalizes. The saved research is US-OVERNIGHT-EFFECT-001 at commit 89ee5288.
This is educational research, not personal investment advice or a recommendation to trade at the close or open.
The next test
See the next claim tested—not merely repeated.
Subscribe, then optionally tell us which investment claim or strategy you want checked next.