How we tested this pattern
- Sample
- 483 occurrences on 4 markets (US stocks, ETFs, indices and commodities), 1973 to 2026.
- Detection
- Found by code, never picked by hand: every match of the rule on full price history, winners and losers alike.bullish island reversal pattern: After a downtrend: a gap down, a cluster of 1 to 10 bars isolated below it, then a gap up above the whole cluster, each gap at least half an ATR, every bar traded (high above low). The downtrend is measured up to the entry gap. Dated at the bar that gaps up.bearish island reversal pattern: After an uptrend: a gap up, a cluster of 1 to 10 bars isolated above it, then a gap down below the whole cluster, each gap at least half an ATR, every bar traded (high above low). The uptrend is measured up to the entry gap. Dated at the bar that gaps down.
- Benchmark
- Each occurrence is compared with the same assets on random bars: the move 10 bars later, and a 2:1 reward-to-risk trade.
- Checks
- Block bootstrap and a correction for the thousands of tests we run, plus a random-walk self-test that the bench passes: on random prices, no pattern should win.
- Updated
- Data updated November 5, 2026, code version c2a1683f. Read the methodology · bullish island reversal pattern data (JSON) · bearish island reversal pattern data (JSON)
At a glance: what textbooks say, what we measured
- Pattern
- Bullish island reversal
- Occurrences tested
- 221 occurrences, all markets, daily
- Textbook: price should rise after it
- Too close to call · +0.33% vs the market, 10 bars
- Its best market and timeframe, still negative, US stocks left out
- ETFs, daily: −0.14 R 2:1 trade vs random dates, not proven
- Verdict
- 10-bar move: No clear edge2:1 trade: No clear edge
- Pattern
- Bearish island reversal
- Occurrences tested
- 262 occurrences, all markets, daily
- Textbook: price should fall after it
- Too close to call · −0.19% vs the market, 10 bars
- Its best market and timeframe, still negative
- US stocks, daily: −0.05 R 2:1 trade vs random dates, not proven
- Verdict
- 10-bar move: No clear edge2:1 trade: No clear edge
On this page
- What is an island reversal pattern?
- Anatomy of the formation
- Bullish vs. bearish island reversals
- How to trade the island reversal pattern
- Reliability: what the data shows
- Advantages and limitations
- Common mistakes to avoid
- Real examples from our test bench
- Island reversal vs. similar candlestick patterns
- FAQ
- The bottom line
The island reversal pattern is a group of candles cut off from the rest of the chart by two price gaps in opposite directions: a gap in the direction of the trend, a short pause, then a second gap the other way. The isolated cluster looks like an island, and classical chart analysis reads it as a sign that the trend has ended.
This guide covers the structure, how to identify and trade the formation, and what happened after it in real markets. Our test bench detected 262 tops and 221 bottoms on daily charts of US stocks, ETFs, indices and commodities, and compared each one with random bars on the same assets.
What is an island reversal pattern?
The formation is made of three parts: an opening gap that extends the trend, a consolidation of one or more candlesticks, and a second gap in the opposite direction that leaves the cluster alone above or below the rest of the price action. Edwards and Magee described it in Technical Analysis of Stock Trends (1948), and classical chart analysis calls the first gap an exhaustion gap and the second a breakaway gap.
The classical reasoning is simple: traders who bought or sold in the cluster are now on the wrong side, with no trading in between to get out at their price.
Anatomy of the formation
- Prior downtrend
- Gap downthe next high stays below this low
- Islanda cluster of 1 to 10 bars isolated below the gap
- Gap upabove the whole cluster; dated at this bar
- Prior uptrend
- Gap upthe next low stays above this high
- Islanda cluster of 1 to 10 bars isolated above the gap
- Gap downbelow the whole cluster; dated at this bar
Identifying the two-gap structure
Our bench uses a rule a computer can apply without judgment. For the top version, after an uptrend (the close before the first gap at least one average true range above the close 10 bars earlier, in the upper half of the last 40 closes):
- a gap up: the low of a candle is at least half an average true range above the high of the previous one;
- a cluster of one to ten candlesticks that stays entirely above that gap;
- a second gap, down this time, at least half an average true range below the lowest point of the whole cluster.
The bottom version is the mirror image after a downtrend. The formation is dated on its last candle. A true gap leaves empty space between the two candlesticks, wicks included, with no overlap: an opening above the previous close that falls back into its range does not count.
The role of volume
Classical chart analysis wants heavy volume on the last candle, as a sign of conviction. We measured it, and the picture is mixed: the target rate did not rise, but the average result did. Top formations with volume above 1.5 times the average reached their 2:1 target in 10.4% of cases, against 11.3% on normal volume, for −0.27 R against −0.41 R per trade. On bottoms, 15.9% against 19.2%, for +0.35 R against +0.09 R. The groups hold 77 and 53 trades on tops, 44 and 52 on bottoms, and this breakdown is descriptive and not corrected for the number of tests: it is too short to conclude either way.
Bullish vs. bearish island reversals
Characteristics of a bullish island reversal
It forms after a downtrend: price gaps down, trades below the gap for a few sessions, then gaps up above the whole cluster. It is also called an island bottom.
Characteristics of a bearish island reversal
It forms after an uptrend: a gap up, a cluster of candles at the top, then a gap down below it. It is also called an island top. Both diagrams above are schematic drawings of the rule, not measured data.
How to trade the island reversal pattern
- Entry
- Buy stop at the pattern high
- Stop loss (SL)
- At the pattern low
- Target (TP)
- Twice the risk (2:1)
- Entry
- Sell stop at the pattern low
- Stop loss (SL)
- At the pattern high
- Target (TP)
- Twice the risk (2:1)
A clean case from our bench: ETFsChosen on its shape alone, never on the outcome. ETF prices are adjusted for splits and dividends, so they can differ from the prices quoted at the time.
A clean case from our bench: US stocksChosen on its shape alone, never on the outcome. Stock prices are adjusted for splits and dividends, so they can differ from the prices quoted at the time.
Entry strategies and signal confirmation
The aggressive approach enters at the close of the last candle. The conservative approach waits for confirmation: a break of the pattern low for a top, or of its high for a bottom. That is the trading strategy our bench measures: a stop order at the extreme of the pattern, valid 20 bars. For the trade, our bench takes the pattern as the 12 bars that end on the second gap: the island, both gaps and the bars before the first gap. Their highest high and lowest low set the entry and the stop, so these levels can sit beyond the island itself, as in the cases further on. The entry was triggered 52.3% of the time on tops and 44.8% on bottoms. The setup was canceled before any trade, because the opposite extreme broke first, in 19.1% of tops and 34.4% of bottoms; in the remaining cases, neither level was reached within the 20 bars and the order expired.
Where to place your stop loss
The natural stop is on the other side of the pattern: above its high for a top, below its low for a bottom. If price goes back beyond the island, the idea is wrong. This stop is wide: the median height of that 12-bar pattern area was 13.50% of price for tops (4.66 times the average true range) and 12.62% for bottoms (4.82 times). Size the position on that distance.
Setting profit targets for the reversal
Profit can be taken at the first gap, at prior support or resistance levels, or at a multiple of the risk. Our statistics use a fixed 2:1 reward-to-risk ratio. On daily bottoms, all markets combined, only 18 of 97 trades reached it (7 of 47 on US stocks), too few to give a typical duration.
Reliability: what the data shows
The island reversal is not a reliable trading signal on its own: on every daily market we tested, neither its average excess return over the rest of the market nor the average result of its trade (for example +0.23 R against +0.31 R per trade for bullish daily islands) survived our correction for multiple testing; only the 10-day move of ETF tops passed a simple test.
Moved the expected way 10 bars later, compared with random bars.
Measured on 221 daily bullish island reversal occurrences across 119 assets. Random bars on the same assets are the honest baseline.
No clear edgeMoved the expected way 10 bars later, compared with random bars.
Measured on 262 daily bearish island reversal occurrences across 166 assets. Random bars on the same assets are the honest baseline.
No clear edgeBullish island reversal against random bars on the same assets, by market, on daily candles. Bars of almost the same length mean no edge.
- All markets
- US stocks
- ETFs
- All markets
- US stocks
- ETFs
Bearish island reversal against random bars on the same assets, by market, on daily candles. Bars of almost the same length mean no edge.
- All markets
- US stocks
- ETFs
- All markets
- US stocks
- ETFs
All statistics
Full results for the island reversal pattern, market by market
Every occurrence our rule finds on full price history, compared with what the same assets did on random bars. A pattern only earns an edge if it beats random entries after correcting for the thousands of tests we run.
Our verdict
Not a reliable signal on its own here. Use it as context, not as a trigger.
Bullish island reversal: 221 occurrences tested
Bullish pattern · All markets · Daily · 119 assets · 1973-09 to 2026-10
What this means. No measurable difference from the rest of its market over the next 10 bars.
Timeline
- Entry triggered, after the pattern
- 5 days · middle half 2 to 9 · on 99 triggered entries
- Stop order beyond the pattern's extreme. 2 of these entries are not among the 97 trades: still open at the end of our data, or left out by the trade rules.
- Time to the 2:1 target, when hit
- too few to measure · on 18 trades that hit it
- Time to the stop, when hit
- 25 days · middle half 14 to 46 · on 30 trades that hit it
- Time in the trade, all exits
- 100 days · middle half 41 to 100 · on 97 trades
Size and trade path
- Pattern range, high to low
- 12.6% · middle half 7.4% to 20.0% · on 221 occurrences
- Range vs average bar (ATR)
- 4.82× · middle half 4.14× to 5.60× · on 221 occurrences
- Best open profit during the trade
- 0.96 R · middle half 0.38 R to 1.62 R · on 97 trades
- Stopped trades that were first up 1 R
- 6.7% · on 30 stopped trades
- Trades a stop moved to breakeven at +1 R would have closed flat instead of at a loss
- Entry order triggered
- 44.8% · on 221 occurrences
Medians, each on the cases named on its line; "middle half" spans the 25th to 75th percentile; no value under 30 cases.
How we detect it. After a downtrend: a gap down, a cluster of 1 to 10 bars isolated below it, then a gap up above the whole cluster, each gap at least half an ATR, every bar traded (high above low). The downtrend is measured up to the entry gap. Dated at the bar that gaps up. Candlestick counts are not checked against planted examples, so they may not be exhaustive.
All measurements (5 rows)
10-bar move: did price go the expected way?
| Market | Occurrences | Pattern | Random bars | Excess | 10-bar move verdict |
|---|---|---|---|---|---|
| US stocks · Daily | 103 | 60.2% | 54.6% | +0.68% | No clear edge |
| ETFs · Daily | 101 | 56.4% | 55.5% | +0.27% | No clear edge |
| Indices · Daily | 13 | 53.8% | 56.0% | −0.76% | Not enough data |
| Commodities · Daily | 4 | 50.0% | 52.3% | −3.62% | Not enough data |
| All markets · Daily | 221 | 57.9% | 55.0% | +0.33% | No clear edge |
2:1 trade: target hit and average result
| Market | Trades | Target hit | Same trade, random dates | Per trade | 2:1 trade verdict |
|---|---|---|---|---|---|
| US stocks · Daily | 47 | 14.9% | 27.1% | +0.38 R vs +0.44 R | No clear edge |
| ETFs · Daily | 44 | 18.2% | 24.8% | −0.03 R vs +0.21 R | No clear edge |
| Indices · Daily | 4 | 75.0% | 25.0% | +1.64 R vs +0.49 R | Not enough data |
| Commodities · Daily | 2 | 0.0% | 0.0% | −0.39 R vs −1.00 R | Not enough data |
| All markets · Daily | 97 | 18.6% | 25.4% | +0.23 R vs +0.31 R | No clear edge |
Our verdict
Not a reliable signal on its own here. Use it as context, not as a trigger.
Bearish island reversal: 262 occurrences tested
Bearish pattern · All markets · Daily · 166 assets · 1978-09 to 2026-09
What this means. No measurable difference from the rest of its market over the next 10 bars.
Timeline
- Entry triggered, after the pattern
- 5 days · middle half 2 to 9 · on 137 triggered entries
- Stop order beyond the pattern's extreme. 6 of these entries are not among the 131 trades: still open at the end of our data, or left out by the trade rules.
- Time to the 2:1 target, when hit
- too few to measure · on 14 trades that hit it
- Time to the stop, when hit
- 39 days · middle half 20 to 63 · on 79 trades that hit it
- Time in the trade, all exits
- 59 days · middle half 25 to 100 · on 131 trades
Size and trade path
- Pattern range, high to low
- 13.5% · middle half 7.5% to 21.6% · on 262 occurrences
- Range vs average bar (ATR)
- 4.66× · middle half 4.22× to 5.44× · on 262 occurrences
- Best open profit during the trade
- 0.60 R · middle half 0.23 R to 1.15 R · on 131 trades
- Stopped trades that were first up 1 R
- 15.2% · on 79 stopped trades
- Trades a stop moved to breakeven at +1 R would have closed flat instead of at a loss
- Entry order triggered
- 52.3% · on 262 occurrences
Medians, each on the cases named on its line; "middle half" spans the 25th to 75th percentile; no value under 30 cases.
How we detect it. After an uptrend: a gap up, a cluster of 1 to 10 bars isolated above it, then a gap down below the whole cluster, each gap at least half an ATR, every bar traded (high above low). The uptrend is measured up to the entry gap. Dated at the bar that gaps down. Candlestick counts are not checked against planted examples, so they may not be exhaustive.
All measurements (6 rows)
10-bar move: did price go the expected way?
| Market | Occurrences | Pattern | Random bars | Excess | 10-bar move verdict |
|---|---|---|---|---|---|
| US stocks · Daily | 155 | 41.3% | 45.7% | −0.90% | No clear edge |
| ETFs · Daily | 93 | 45.2% | 44.4% | +0.88% | Promising |
| Indices · Daily | 7 | 28.6% | 41.4% | +0.14% | Not enough data |
| Commodities · Daily | 7 | 85.7% | 46.1% | +0.90% | Not enough data |
| All markets · Daily | 262 | 43.5% | 45.1% | −0.19% | No clear edge |
| US stocks · Weekly | 1 | 0.0% | 48.5% | −41.97% | Not enough data |
2:1 trade: target hit and average result
| Market | Trades | Target hit | Same trade, random dates | Per trade | 2:1 trade verdict |
|---|---|---|---|---|---|
| US stocks · Daily | 76 | 9.2% | 8.0% | −0.27 R vs −0.21 R | No clear edge |
| ETFs · Daily | 49 | 14.3% | 24.6% | −0.35 R vs −0.09 R | No clear edge |
| Indices · Daily | 3 | 0.0% | 33.3% | −1.03 R vs −0.01 R | Not enough data |
| Commodities · Daily | 3 | 0.0% | 0.0% | −1.01 R vs −1.00 R | Not enough data |
| All markets · Daily | 131 | 10.7% | 14.3% | −0.33 R vs −0.18 R | No clear edge |
| US stocks · Weekly | 0 | n/a | n/a | n/a vs n/a | Not enough data |
Data updated November 5, 2026 · code version c2a1683f · Read the methodology. No occurrence on crypto or forex in our data. We measure every pattern on crypto and forex (1-hour, 4-hour, daily and weekly charts) and US stocks, ETFs, indices and commodities (daily and weekly charts). A market or timeframe from that list missing from the tables above had no occurrence.
What you often read
“The bullish island reversal is a reliable reversal signal.”
What our data shows
On 221 daily occurrences, it moved in the expected direction 57.9% of the time, against 55.0% for random bars: no clear edge, since the average excess return is not statistically significant.
What you often read
“The bearish island reversal is a reliable reversal signal.”
What our data shows
On 262 daily occurrences, it moved in the expected direction 43.5% of the time, against 45.1% for random bars: no clear edge, since the average excess return is not statistically significant.
All rows of this table are daily charts. The index rows rest on too few occurrences to be read: they are shown for completeness only.
Table quoted in the text (verdicts and every market: see All measurements)
| Version and market | Occurrences | Moved as expected 10 bars later | Random bars | Closed trades | 2:1 target hit | Same trade, random dates | Expectancy (random dates) |
|---|---|---|---|---|---|---|---|
| Top (bearish), all markets | 262 | 43.5% | 45.1% | 131 | 10.7% | 14.3% | −0.33 R (−0.18 R) |
| Top, ETFs | 93 | 45.2% | 44.4% | 49 | 14.3% | 24.6% | −0.35 R (−0.09 R) |
| Top, indices | 7 | 28.6% | 41.4% | 3 | 0.0% | 33.3% | −1.03 R (−0.01 R) |
| Bottom (bullish), all markets | 221 | 57.9% | 55.0% | 97 | 18.6% | 25.4% | +0.23 R (+0.31 R) |
| Bottom, ETFs | 101 | 56.4% | 55.5% | 44 | 18.2% | 24.8% | −0.03 R (+0.21 R) |
| Bottom, indices | 13 | 53.8% | 56.0% | 4 | 75.0% | 25.0% | +1.64 R (+0.49 R) |
Read the table in pairs of columns. 43.5% of tops were followed by a lower close 10 bars later, against 45.1% for random bars on the same assets. The excess return over the rest of the market was −0.19%, with a 95% confidence interval from −1.34% to +0.75%: zero is inside, so no measurable advantage. Bottoms look better in absolute terms (57.9% higher, +0.23 R per trade before costs) mostly because the same assets also rose on random dates: 55.0% of random days were higher 10 days later, and the same trade on random dates earned +0.31 R. Neither the average excess return (+0.33%) nor the gap in average result per trade is statistically significant.
ETF tops are the one encouraging line: 45.2% of them were followed by a lower close 10 days later, against 44.4% for random days, with an excess return of +0.88% (95% interval +0.26% to +1.66%). The 10-day move is rated Promising, on 93 occurrences: it does not survive our correction for multiple testing, so it is not a proven setup. The 2:1 trade on these tops rests on 49 trades and is rated No clear edge. On weekly charts, our strict rule found only 1 top and no bottom: too few to test.
How often does it form?
On US stocks, one top every 47,221 daily bars and one bottom every 71,061. Crypto trades around the clock and forex 24 hours on weekdays, so a new candle opens near the last close and leaves little room for a gap: with its strict rule, our bench found no island reversal on crypto or forex charts, so it has no result for these markets.
Advantages and limitations
- Advantage: the structure is objective, and the island gives a clear level where the idea is wrong.
- Limitation: the stop is wide, a median of 13.50% of price on tops and 12.62% on bottoms, and 19.1% of tops and 34.4% of bottoms were canceled before entry.
- Limitation: neither its average excess return over the rest of the market nor its average result per trade showed an edge that survives our correction on any market we tested; the one encouraging line, the 10-day move of ETF tops, rests on 93 occurrences.
Common mistakes to avoid
- Entering before the second gap: until it prints, there is no island, only a gap.
- Ignoring the trend: without a prior trend, two gaps do not make an island reversal under the rule, and our bench does not test them.
- Using a stop within the island, where ordinary price swings can reach it before the idea is proven wrong.
Real examples from our test bench
The gallery in this section shows real island reversals picked by our bench (best and worst results, textbook shapes, most recent, random draws), with their actual outcome. This section also writes up three cases from our data on daily stock charts: a winner, a loser and a canceled setup.
Real occurrences from our test bench, each on its market's main timeframe. US stock examples come from liquid assets only (the most traded stocks). Best and Worst are the best and worst 2:1 trade results, Textbook the cleanest shapes (chosen on geometry alone, never on the outcome), Most recent the latest detections among the assets the gallery draws from, Typical a random draw. Shaded area: the pattern. Dashed lines: entry, stop loss (SL) and 2:1 target (TP) of the tested trade. Dotted line: 10 bars after the signal. A result above +2 R or below −1 R means a bar opened past the target or the stop, and the trade was closed at that open. Stock prices are adjusted for splits and dividends, so they can differ from the prices quoted at the time. We left out 4 cards with a shape that does not read as the definition on this page. They still count in the statistics, and are not replaced: our bench picks a fixed number of cards per category.
In one case from our data (prices adjusted for splits and dividends), Linde (LIN) printed a three-session island bottom: it gapped down on October 28, 2020, held below that gap through October 30, then gapped up on November 2. The buy stop at the top of the 12-bar pattern area, 218.15, a high set on October 23, before the island, was filled on November 4, with a stop at 197.17, the low of the island, and a 2:1 target at 260.12. Ten sessions after the pattern LIN was 11.8% higher, and the target was reached 97 sessions after entry: the wide stop gave the trade room, but it took months.
CRH printed a six-session island top from September 28 to October 5, 2017, between a gap up on September 28 and a gap down on October 6. The short entry came on October 19, when the stock opened at 28.25, below the low of the 12-bar pattern area. The stop went at 30.08, the high of the island. Ten sessions after the pattern CRH was 1.2% higher, and price later traded back above the island and hit the stop: a full 1 R loss.
Honeywell (HON) printed a three-session island bottom: a gap down on September 22, 2011, then a gap up on September 27. The buy stop at the top of the 12-bar pattern area, 32.78, a high set on September 20, above the island, was never reached: price broke the island low (28.52) on October 4 and the setup was canceled. Ten sessions after the pattern HON was 4.9% higher, a rise that came after the cancellation.
Island reversal vs. similar candlestick patterns
The abandoned baby is a one-candle version of the pattern where the isolated candle is a doji. The bullish version reached its 2:1 target in 41.2% of cases against 30.4% on random dates, on only 17 trades, too few to interpret. The key reversal bar and the morning star also mark turns, but without empty space on both sides.
- Abandoned BabyBullish or bearishBullish, 10-bar move, daily Not enough dataBearish, 10-bar move, daily Not enough data
- Key Reversal BarBullish or bearishBullish, 10-bar move, daily No clear edgeBearish, 10-bar move, daily No clear edge
- Morning StarBullish10-bar move, daily No clear edge
FAQ
Is an island reversal bullish or bearish?
Both exist: an island bottom after a downtrend is bullish, an island top after an uptrend is bearish.
Does the second gap have to be filled?
No. If price fills it and moves back through the cluster, the structure has failed.
The bottom line
The island reversal pattern is easy to define and rare enough to stand out, but on 262 daily tops and 221 daily bottoms its average excess return over the rest of the market was too small to separate from zero. Use it as a place to define risk, keep the stop beyond the pattern, and size each position small. Trading involves risk, and nothing here is investment advice.
About the author and the data. Written by Amaury, founder of PatternsWizard, who is passionate about finance, quantitative trading and data analysis. He built the PatternsWizard test bench, which detects each setup with fixed, published rules on full market history: crypto, forex, US stocks, ETFs, indices and commodities. Trades and win rates are compared with random bars on the same assets, excess returns with the rest of the market, with error margins and a correction for multiple testing, and the whole pipeline is checked on random-walk prices, where nothing can be predicted. Statistics on this page were generated on November 5, 2026. The full methodology explains every step.
