Opening range breakout on SPY shows no edge after costs

Kaspar Dolan, Trading-platform writer (AI-assisted) ·

The opening range breakout made no money after costs in our own test. A 30-minute range on SPY, traded across 59 sessions from 28 August to 20 November 2026, produced 58 trades and a t-statistic of 0.38 — far below the level that would suggest a real edge [1]. The average profit of 2.50 basis points per trade after costs is statistically indistinguishable from zero [1].

What the rule traded and what it made

The test, run on SPY 5-minute bars from 28 August to 20 November 2026, used a classic 30-minute opening range breakout rule: mark the high and low from 9:30 to 10:00 New York time, then, after 10:00, take the first breakout above or below as a signal to go long or short at the range boundary. Only one trade per day was taken. The stop was set at the opposite side of the range, and if still open, the position closed at the end of the 15:55 bar [1].

This structure matches what most ORB literature describes: the opening range defines the day's key levels, and a breakout beyond them is meant to catch early directional momentum [2, 3]. The 30-minute window, compared to shorter intervals, produces fewer signals and wider stops but is intended to filter out early noise [3].

Across 59 sessions, the simulation generated 58 trades: 28 longs and 30 shorts [1]. Twenty-seven trades were stopped out, while 24 trades finished positive, giving a win rate of 41% [1]. Before transaction costs, the total net was $12.83 per share, averaging 3.02 basis points per trade with a standard deviation of 50.59 basis points [1]. The table below summarizes the key figures:

MetricValue
Trades58
Winning trades24 (41%)
Total before costs$12.83/share
Average per trade3.02 bps
Std. deviation/trade50.59 bps

All results are from the site's own simulation run on 21 November 2026 [1].

What costs did to the result

Costs changed the picture. The test assumed a round trip of 2 cents per share for spread, then a second run at 4 cents per share — spread plus 1 cent of slippage on each side [1]. After the 4-cent assumption, the total profit fell from 12.83 dollars per share to 10.51 dollars per share, and the average to 2.50 basis points per trade [1].

The table shows the full result with costs included.

FigureValue
Sessions59
Trades58
Winning trades24 (41%)
Total, gross12.83 dollars per share
Total after 4 cents per trade10.51 dollars per share
Average after costs2.50 basis points per trade
Standard deviation50.59 basis points
t-statistic0.38

Figures from the site owner's own test data, 21 November 2026 [1].

The t-statistic is the number that matters. It measures how far the average result sits from zero, in units of its own variability. A t-statistic of 2.0 is the conventional threshold for statistical significance. The result here is 0.38 — about one-fifth of the way to that line [1]. The average is positive, but the variation between trades is so large that the positive average is easily explained by chance.

How many trades it would take to know

With the site's own calculated average of 2.50 basis points per trade and a standard deviation of 50.59 basis points (after costs, on SPY, from 58 trades in late 2026), the t-statistic lands at just 0.38—nowhere near a level that would distinguish the result from noise [1]. The arithmetic behind this is straightforward for practitioners: the t-statistic is the mean divided by the standard error (standard deviation divided by the square root of the sample size). To reach a t-statistic of 2.0 with the same mean and spread, the sample size must rise to about 1,638 trades—roughly 27 times the initial 58 [1].

For a rule that trades once per US trading day, that means collecting data for over six years before a modest edge like this one would reliably clear the noise floor by standard statistical measures [1]. In practice, this means that short backtests—like the three-month, 58-trade run here—will always be dominated by variance, not signal, if the mean result is small and the spread is wide. For MetaTrader users, this is a clear warning: even with tick-perfect fills and historical spreads, no amount of scripting or optimization can make a short test of a low-edge, high-variance rule statistically meaningful. The edge, if it exists, is buried deep in the noise, and only a large sample can dig it out. [1]

Where the test is optimistic

The test made assumptions that flatter the strategy. Fills are assumed exactly at the range boundary, but a real breakout order often fills worse when price jumps through the level, so the result is optimistic [1]. When a bar both breaks out and crosses the stop, the order within the bar is unknown; the test takes the breakout and checks the stop from the next bar [1]. Five-minute bars hide moves inside the bar, and tick data would be more exact [1].

The sample itself is one market regime. Three months of SPY in late 2026 is a single stretch of conditions, and the same rule can look very different over years [1]. The ORB literature makes the same point from the other direction: most ORB strategies fail because they are overfit to a particular stretch of data, and robustness testing across regimes is essential [2].

A worked example

Suppose you apply the opening range breakout rule exactly as tested: one trade per session, with 100 shares per trade. Our own simulation on SPY, run over 59 sessions using 5-minute bars and including 4 cents per share in round-trip costs, recorded 58 trades in total [1]. The average return after costs was 2.50 basis points per trade, with a standard deviation of 50.59 basis points [1]. To put the volatility in context: the spread of outcomes across trades is over 20 times larger than the average gain.

If you ran the strategy for about a year—roughly 250 trades—the standard error of the mean would fall to the standard deviation divided by the square root of 250, but this still leaves the average well within the noise. As calculated, reaching a t-statistic of 2 would require about 1,638 trades at these same figures: more than six years of one trade per day, just to clear the bar for statistical significance [1].

Key figures from our 2026-11-21 simulation:

MetricValue (after costs)
Total trades58
Average per trade2.50 basis points
Standard deviation50.59 basis points
Trades needed for t=2~1,638

A single loss at the stop easily outweighs the average win, and in this test, the chance of distinguishing the edge from randomness after one year is negligible [1].

Questions readers ask

Does a 41% win rate mean the strategy loses?

Not by itself. The win rate is 41%, but the average result after costs is still positive at 2.50 basis points per trade because the winning trades are larger on average than the losing trades [1]. The problem is not that the average is negative; it is that the average is too small relative to the variation to be distinguished from zero.

Would a different range length change the result?

Possibly, but this test does not say. It tested one 30-minute range on one instrument over 59 sessions [1]. A 5- or 15-minute range triggers more often and fails more often, while a longer range is slower but cleaner [3]. Each choice needs its own test with costs included before it can be judged.

Should I stop trading opening range breakouts?

This result does not support that conclusion. It shows that one simple rule on SPY over one three-month period produced no statistically significant edge after costs [1]. The ORB family is large, with variations in entries, stops, filters and exits [2, 3], and a different setup or market could behave differently. What the result does show is that a small sample cannot confirm an edge — and most ORB tests are too short to know.

How the numbers work

Computed from the numbered sources and the assumptions named below; change an assumption and the results change with it.

FigureValueWhere it comes from
Number of trades (sessions with a trade)58source [1]
Number of winning trades24source [1]
Win rate0.41source [1]
Total profit before costs12.83 dollars per sharesource [1]
Round-trip cost per trade (spread plus slippage)0.04 dollars per sharesource [1]
Total profit after costs10.51 dollars per sharesource [1]
Mean profit after costs per trade0.025 dollars per sharesource [1]
Mean profit after costs per trade2.5 basis pointssource [1]
Standard deviation of results per trade50.59 basis pointssource [1]
Average profit per trade after costs (dollars per share)0.181 dollars per sharecomputed: Total profit after costs ÷ Number of trades (sessions with a trade)
t-statistic of the mean after costs0.376 $computed: Mean profit after costs per trade ÷ (Standard deviation of results per trade ÷ sqrt(Number of trades (sessions with a trade)))
Number of trades needed for t=2 (statistical significance)1,638computed: round((2 × Standard deviation of results per trade ÷ Mean profit after costs per trade) ^ 2)

Sources

  1. Our own data, 21 Nov 2026: SPY 30-minute opening range breakout, 59 sessions, every trade after costs
  2. Opening Range Breakout: Does It Still Work? 4 Setups Tested
  3. Opening Range Breakout (ORB) Strategy Guide | Velotrade

More articles