Two papers say the first five minutes of the US session predict the rest of the day. We rebuilt both rule sets exactly as published and ran them on 4,924 stocks, including 1,080 that were later delisted or renamed, plus QQQ and SPY, from 2016 through 2022, with a pre-registered test and realistic costs.
Nothing survived. The one variant that cleared our in-sample bar, QQQ with a stop at 5 percent of ATR, needed a median stop of 8.7 cents and lost money at 3 cents of slippage per side. On the stock universe, every version lost before a single cent of cost was charged: between 0.18 and 0.43 R per trade.
This is a measurement, not a verdict on your trading. It tells you what we found when we looked carefully. The rest of this page is how we looked.
What the two papers claim
The opening range breakout, or ORB, is an old day-trading idea: mark the high and low of the first few minutes, trade in the direction the session opened, cut the loss fast and let the winner run to the close. Two recent papers put numbers on it.
- Zarattini and Aziz (2023), QQQ. Buy at the open of the second 5-minute candle if the first candle closed up, sell short if it closed down, skip a doji. Stop at the far end of the first candle. Target ten times the risk, or the close. They report a 676 percent total return on QQQ from 2016 to early 2023, a 24 percent win rate and an average of +0.13 R per trade. A variant with a stop at 5 percent of the 14-day average true range and no target did much better in their search.
- Zarattini, Barbon and Aziz (2024), stocks in play. Every stock that opens above 5 dollars, trades at least a million shares a day and has a 14-day ATR above 50 cents. Place a stop order at the first candle’s high or low in the direction of that candle, stop-loss at 10 percent of ATR, exit at the close. The base version returned 29 percent over eight years. Filtering to the 20 stocks with the highest relative volume in the first five minutes, they report 1,637 percent, a Sharpe ratio of 2.81, and +0.38 R per trade when relative volume exceeded 30 times normal.
“R” is the risk unit: the distance from entry to stop. A trade that hits its stop is minus 1 R. A trade that makes three times its risk is +3 R. Both papers report in R, so we did the same.
Both papers assume no slippage. Both authors run day-trading education businesses. Neither fact makes the result wrong. They are reasons to check.
How we tested it
We wrote the rules down before running anything, with the windows, costs, statistic and pass criteria fixed in advance. Every change made afterward is dated in the pre-registration file. There were six: five were forced by a control that failed, and one records a number that was seen before the controls ran. The analysis was run with an AI assistant on our own data, from the studies’ own files.
- Data. Alpaca’s historical SIP feed, 5-minute bars, regular hours only, split-adjusted. For the stock universe we needed names that no longer exist, so the symbol list came from FINRA’s daily consolidated short-volume files from August 2018 onward and from archived copies of Nasdaq’s listing directories before that. 22,946 symbols went in. 4,924 were ever eligible under the paper’s own filters between 2016 and 2022, and 1,080 of those have price history that ends before 2023. A universe built from today’s listings never contains the stock that gapped up 40 percent on a press release in 2017 and was gone by 2019, and those are exactly the names an “in play” filter selects.
- Windows. 2016 through 2019 was the test. 2020 through 2022 was held back for any rule that passed. 2023 onward was locked and never opened. Nothing earned it.
- Costs. Commission as the papers state it: 0.05 cents per share on the ETF, 0.35 cents on stocks. On top of that, slippage of 1 cent per share per side on QQQ and SPY, 2 cents on stocks, plus 1 cent more on any stop exit. Every result is also reported at 0, 1, 2, 3 and 5 cents so you can pick your own number.
- The statistic. For every trade we also simulated the opposite direction with the same stop distance, entered at the same price. The difference between the two is what the first candle’s direction was worth, separate from costs and geometry. A rule passes when that difference beats a bar set by 2,000 random reshuffles across the whole family of rules, is positive in both halves of the test window, and is profitable after costs.
- The controls. Before reading a single real result we ran the engine on random walks with the same bar sizes as the real data. A random walk has no information in its first candle by definition, so the test should pass that check about 79 times in 80. Earlier versions of our statistic failed it four times. The final one, which measures the information from the close of the breakout bar rather than from the fill, passed 30 out of 30.
QQQ and SPY
| Rule | Trades | Win rate | Net R per trade at 1 cent | Passed the bar |
|---|---|---|---|---|
| QQQ, published rule, range stop, 10 R target | 990 | 24.5% | +0.08 | No |
| SPY, same rule | 989 | 20.3% | -0.06 | No |
| QQQ, 30-minute range instead of 5 | 999 | 36.0% | -0.04 | No |
| SPY, 30-minute range | 995 | 33.3% | -0.03 | No |
| QQQ, stop at 5% of ATR, no target | 975 | 12.4% | +0.36 | Yes, then failed validation |
The published QQQ rule carries a little information in the first candle. Not enough to clear the bar, and at 2 cents of slippage per side it loses money. An independent replication on different data found the same break-even, near 2.2 cents. The 30-minute version, which is what most retail tooling can actually run on a delayed feed, shows nothing on either ETF.
The ATR-stop variant is the interesting one, because it passed the first gate and then fell apart under inspection.
- Its median stop on QQQ was 8.7 cents. A typical 5-minute bar on QQQ in that period moved 13 cents from high to low. The stop sits inside the noise of a single bar.
- 88 percent of trades were stopped out, most within minutes.
- 87 percent of all profit in the test window came from the ten best days out of roughly a thousand.
- 2017 lost 84 R. In the validation window, 2020 through 2022, it made 0.14 R per trade and the test came back at a p-value of 0.064 against a 0.05 line.
- Slippage sensitivity, net R per trade: 0 cents +0.61, 1 cent +0.36, 2 cents +0.11, 3 cents -0.14.
Under the paper’s own sizing rules, 1 percent risk per trade and a 4x leverage cap, the cap binds on almost every trade at that stop size. 25,000 dollars became about 52,000 over four years with a 29 percent drawdown. The R multiples flatter the result by a factor of several.
Stocks in play
| Version | Trades | Win rate | Gross R per trade, no costs | Net R at 2 cents |
|---|---|---|---|---|
| Every eligible stock, 5-minute bars | 858,698 | 10.9% | -0.22 | -0.68 |
| Top 20 by relative volume each day | 19,610 | 5.8% | -0.34 | -0.80 |
| Relative volume above 30x, 5-minute bars | 624 | 5.8% | -0.43 | -0.94 |
| Relative volume above 30x, 1-minute bars | 600 | 8.2% | -0.18 | -0.68 |
Every version loses before costs. The direction of the first candle carried no measurable information about the rest of the day on any of them: the test statistics sat between -0.44 and +0.30 against a bar of 2.83. The 1-minute version matches the paper’s data resolution, and it is the one that loses least, which says the 5-minute results are a little pessimistic and the conclusion does not depend on it.
Three quarters of trades stopped out at 10 percent of ATR, which on these names is a median of 9 to 11 cents. The stop sits where the opening noise lives.
One number in the papers deserves a translation. Their “hit ratio” of 41 to 48 percent is the share of profitable days for a diversified portfolio of these trades, not the share of winning trades. Their own table shows 54.9 percent for a plain buy-and-hold of the S&P 500, which is the share of up days. A per-trade win rate near 10 percent is consistent with their figures, not a contradiction of them.
What we did not test, and where we could be wrong
- 2020 through 2023 for the stock rules. Our test window was 2016 through 2019 and nothing passed it, so the later window was never opened. The published profits in this literature are known to cluster in 2020 to 2022. If the effect only exists in that period, our result would not see it, and you should weigh that.
- A gap in the universe. NYSE and AMEX names delisted between January 2016 and May 2017 are missing, because FINRA’s public files start in August 2018 and the archived NYSE directory starts in mid-2017.
- Fill conventions. Both papers fill a stop order at the level. On real tape the first print through a level overshoots it. We kept their convention for the money results and labelled it optimistic. On random walks that convention alone was worth about a quarter of an R per trade, which is part of why the raw numbers in these papers look the way they do.
- One peek. Before the controls ran, one aggregate QQQ figure was seen during a mechanics check. Nothing was changed because of it. It is recorded in the pre-registration because that is the rule.
Why this matters if you trade around a job
The appeal of a breakout at the open is that it is done by 10 a.m. The cost of a breakout at the open is that the first five minutes are the most expensive five minutes of the day to trade: spreads are widest, fills are worst, and a stop sized to that window is a coin flip with a fee attached.
If you want to see what a stop that tight costs in dollars for your own position size, the position size calculator has an ATR mode. Clocky’s daily notes on his page show what the opening range actually did each session, with no action attached.
Frequently asked questions
Does this mean the opening range breakout never works?
It means the two published rule sets did not show a tradable edge on our data: 4,924 stocks and two ETFs, 2016 through 2022, after costs. A different rule, a different period or a different cost assumption could land elsewhere. We published the exact rules, windows and costs so you can judge that for yourself.
Why do your numbers differ from the papers?
Three reasons we can name. The papers assume no slippage, and stops of 8 to 12 cents turn a cent of slippage into a large fraction of the risk. The papers fill stop orders at the level, which on random data alone added about a quarter of an R per trade in our engine. And the stock paper's headline covers 2016 to 2023, while our test window ended in 2019 by design.
What is R?
The distance from your entry to your stop, in dollars per share. Results in R let you compare trades of different sizes. A trade that stops out is minus 1 R. A rule with a 10 percent win rate needs its winners to average well above 9 R just to break even before costs.
What is relative volume?
The volume in the first five minutes of today's session divided by the average of that same five-minute window over the previous 14 sessions. Thirty times normal means the stock opened on news.
Did you include stocks that went to zero?
Yes. The universe was built from historical listings, not today's, and 1,080 of the 4,924 eligible names have price history that ends before 2023. About 40 percent of the names that ever reached 30 times normal opening volume were later delisted or renamed.
Can I run this on Clock Out Capital?
No, and that is deliberate. The bots run on 30-minute and daily bars with a resting stop at the broker. We tested the 30-minute version of the idea so you would not have to, and it showed nothing on QQQ or SPY.
Sources
- Zarattini, C. and Aziz, A. (2023). Can Day Trading Really Be Profitable? SSRN 4416622; the authors’ summary.
- Zarattini, C., Barbon, A. and Aziz, A. (2024). A Profitable Day Trading Strategy For The U.S. Equity Market. SSRN 4729284.
- Independent replication of the QQQ rule with a slippage stress test: github.com/giovannibrusco/zarattini-2023-orb-qqq.
- The symbol universe: FINRA daily short sale volume files. Bars: Alpaca Market Data, historical bars.
Related
A Sharpe ratio of 2.81 is one of the claims above; what that number usually looks like in practice is in what counts as a good Sharpe ratio. For how a fixed threshold behaves across a whole market, see what RSI overbought and oversold actually mean. More lessons are on the learn page, and the about page explains who writes them and why.