Candlestick patterns can describe a price reversal without proving a profitable trade. Thomas Bulkowski reports that the bullish kicking pattern reversed the prior trend 53% of the time, but that figure is not a strategy win rate. It does not tell you the entry, exit, stop, average winner, average loser, trading cost, or out-of-sample result.
Profitability starts with a fully defined trade and its expected value. Treat the pattern name as a hypothesis, then test the complete rule after realistic costs on data that was not used to design it.
The chart looks decisive: a black candle, an upward gap, then a tall white candle. A reversal percentage sits beside it. It is easy to read those facts as a trade recommendation.
That shortcut skips the most important question: What does that percentage actually measure?
For bullish kicking, Bulkowski is measuring how often the pattern acted as a bullish reversal of the existing price trend. He is not reporting the share of fully specified trades that ended with a profit after spreads, slippage, commissions, stops, and exits. Those are different measurements.
What Bulkowski’s Reversal Statistic Actually Measures
Bulkowski says bullish kicking acted as a bullish reversal 53% of the time. He calls that result “near random,” ranks the pattern 100th out of 103 for frequency, and gives it an overall performance rank of 96th out of 103. The best reported average move over ten days was 2.78% in a bull market after an upward breakout.
Primary source: Thomas Bulkowski, Bullish Kicking Candlestick. The page separates reversal behavior from post-breakout performance and notes that the pattern was rare.
| The statistic supports | The statistic does not establish |
|---|---|
| The pattern reversed the prior trend in 53% of tested occurrences. | A trader won the same share of completed trades. |
| The bullish outcome was only slightly more common than the alternative. | The average winning trade was larger than the average losing trade. |
| The pattern was rare relative to other candle types in the catalog. | The 4.7-million-bar research base produced a large bullish-kicking trade sample. |
| The shape can be studied as a defined pattern. | The result survived a specific entry, stop, exit, holding period, and cost model. |
A tradable strategy needs more than a pattern name. It needs a trigger, execution price, risk limit, exit rule, position size, and cost model. Change one of those rules and the win rate or payoff can change with it.
The same rule applies to other technical signals. A golden cross win rate or a simple moving average versus exponential moving average crossover result becomes useful only after the entry, exit, benchmark, and cost assumptions are clear.
Why Win Rate Alone Cannot Prove Profitability
Once the trade is defined, calculate expected value. The inputs must come from the same strategy and the same test window.
Consider a hypothetical strategy whose trade win rate has been measured separately at 53%. If winners and losers both average 2.5% and total round-trip cost is 0.20%, expected value is negative 0.05% per trade. Break-even is 54%.
Method in brief: Hypothetical trade inputs only; zero EV is the break-even benchmark. No candlestick sample, signal timing, or executable entry/exit price is modeled. The result illustrates trade economics, not a pattern edge.
Expected value per trade
EV = (p × W) − ((1 − p) × L) − C
- p = actual probability that the defined trade wins
- W = average gain on winning trades
- L = average loss on losing trades, entered as a positive number
- C = total round-trip cost as a percentage of capital committed to the trade
The break-even win probability is p* = (L + C) / (W + L). This model assumes each trade uses a comparable capital base. Taxes, changing position sizes, overnight gaps, and market impact need separate treatment.
This example does not use Bulkowski’s reported reversal statistic as a strategy win rate. It only shows why a trade hit rate cannot be judged without the size of winners, the size of losers, and trading costs.
| Scenario | Average win | Average loss | Cost | EV per trade | Break-even win rate |
|---|---|---|---|---|---|
| Equal payoff | 2.50% | 2.50% | 0.20% | −0.05% | 54% |
| Larger average winner | 3.00% | 2.00% | 0.20% | +0.45% | 44% |
| Larger average loser | 2.00% | 3.00% | 0.20% | −0.55% | 64% |
| Equal payoff, lower cost | 2.50% | 2.50% | 0.05% | +0.10% | 51% |
Independent calculation for the equal-payoff row: (0.53 × 0.025) − (0.47 × 0.025) − 0.002 = −0.0005, or −0.05% per trade.
The table shows why a hit rate slightly above 50% has no universal meaning. A strategy can make money below 50% when winners are much larger than losers. It can lose money above 50% when losses or costs are larger.
Costs must also match the instrument and execution. Commission-free trading does not remove the bid-ask spread, slippage, financing charges, or market impact. A realistic technical analysis backtest should apply costs where the strategy could actually trade.
Why 4.7 Million Bars Can Still Mislead
Bulkowski says he examined more than 4.7 million price bars to identify and track 103 candlestick patterns. That shows the catalog was built from a broad research base.
Primary source: Thomas N. Bulkowski, The Eight Best-Performing Candles, methodology section.
But 4.7 million bars is the search universe across all 103 patterns. It is not the sample size for bullish kicking. Bulkowski says the pattern was rare and that he published only partial statistics because the sample was limited.
A large database can still produce a small number of observations for a rare event. What matters for the pattern-specific result is the number of qualifying cases, how those cases were spread across stocks and market periods, and whether the same rule worked on data that was not used to design it.
Independence matters too. Ten signals from closely related stocks during the same market shock are not the same as ten unrelated trials. Correlated trades can make a backtest look more certain than it is.
Use the large database as evidence that the search was broad, not as a substitute for the trade count and uncertainty around one rare pattern.
What the Research Says About Candlestick Patterns
The research is mixed, and the differences are useful. They show that a pattern name alone does not determine whether a trading rule works.
Caginalp and Laurent tested mathematically defined three-day patterns using daily open, high, low, and close data for S&P 500 stocks from 1992 through 1996. Their out-of-sample tests were statistically significant, and the authors reported a profit of almost 1% during a two-day holding period. That historical result shows that tightly defined price patterns can contain information in some samples. It does not establish a current net-of-cost return for a modern trading strategy.
Research source: G. Caginalp and H. Laurent, The Predictive Power of Price Patterns, Applied Mathematical Finance (1998).
Marshall, Young, and Rose reached a different result using Dow Jones Industrial Average stocks from 1992 through 2002. Their bootstrap tests found no consistent value from the candlestick rules. Lu, Chen, and Hsu later showed why trade design matters. On Dow Jones component data, eight three-day reversal patterns were profitable under the Caginalp–Laurent holding strategy after a 0.5% cost assumption. The result survived an adjustment for testing many rules. The same patterns were not profitable under the Marshall–Young–Rose holding strategy. The holding rule changed the result.
Research sources: Ben R. Marshall, Martin R. Young, and Lawrence C. Rose, Candlestick Technical Trading Strategies: Can They Create Value for Investors? (2006); Tsung-Hsun Lu, Yi-Chi Chen, and Yu-Chin Hsu, Trend Definition or Holding Strategy: What Determines the Profitability of Candlestick Charting? (2015).
Tharavanij, Siraprapasiri, and Rajchamaha tested reversal patterns on the 50 component stocks of Thailand’s SET50 large-cap index from July 2006 through June 2016. They used 1-, 3-, 5-, and 10-day holding periods and two exit methods. Most mean returns were not statistically different from zero, and the patterns did not reliably predict market direction. Filters based on Stochastics %D, the Relative Strength Index (RSI), or the Money Flow Index (MFI) generally did not improve profitability or prediction accuracy.
Research source: P. Tharavanij, V. Siraprapasiri, and K. Rajchamaha, Profitability of Candlestick Charting Patterns in the Stock Exchange of Thailand, SAGE Open (2017).
| Study choice | Why it matters |
|---|---|
| Market and period | Volatility, trading rules, and market structure change across countries and decades. |
| Pattern definition | Small changes in trend filters or candle rules change which events qualify. |
| Entry and exit | Different holding and exit rules create different wins, losses, and exposure. |
| Costs | A gross statistical edge can disappear after spreads, slippage, and turnover. |
| Test design | Fresh-sample tests and checks for repeated rule-searching reduce the chance of selecting a lucky rule. |
That is also why adding more indicators is not automatically an improvement. In the Thailand study, familiar filters usually did not rescue the tested candle rules. RSI or MFI may still help a different strategy, but the combined rule has to prove itself in a fresh test.
Support, resistance, and trend context need the same discipline. A chart annotation is not a repeatable filter until another person can apply the same rule to the same data and get the same signals. The same standard applies to support and resistance trading rules.
A Five-Part Audit Before Trading Candlestick Patterns
A candlestick pattern can be a useful observation tool. Before risking money, check whether the pattern has been turned into a complete trade rule that can survive costs and a fresh sample.
- Define the trade first. Specify the market, timeframe, pattern rule, trend filter, signal time, entry price, stop, exit, maximum holding period, and position size. Two people using the written rule should produce the same trades. Do not keep changing the definition after seeing which version made the most money.
- Measure the whole payoff. Record the trade count, win rate, average and median winner, average and median loser, largest loss, time in trade, and exposure. Calculate expected value from the actual trade list, then compare it with a simple benchmark over the same dates and exposure.
- Apply realistic costs. Use prices the strategy could have received, not an ideal midpoint or a closing price that was not tradable after the signal became known. Include spreads, slippage, commissions, borrowing or financing costs, and relevant taxes. Then test a range of cost assumptions.
- Protect a fresh test period. Keep data that was not used to choose the pattern, trend filter, stop, or holding period. If you tried many versions, account for that search. The best result among hundreds of attempts is more likely to contain luck.
- Check stability before sizing. Break the result down by stock, year, volatility, direction, and liquidity. A strategy should not depend on one ticker, one month, or one extreme trade. If a small change in cost, entry timing, or sample period destroys the result, it has not earned a large allocation.
Frequently Asked Questions
Do candlestick patterns work?
Some tightly defined candle rules have produced positive results in specific markets and periods, while other studies found little reliable value. The result depends on the pattern definition, trade rules, costs, market, sample period, and out-of-sample performance. A pattern name by itself is not enough.
Is a 53% reversal rate good?
That reversal rate is only slightly above an even split, and it is not automatically a trade win rate. To judge a strategy, you need the pattern-level sample count, exact trade rules, average win, average loss, costs, and out-of-sample results. A separate hypothetical 53% trade win rate can be profitable or unprofitable depending on those inputs.
What is the break-even win rate for a trading strategy?
With average win W, average loss L, and round-trip cost C, the break-even win rate is (L + C) / (W + L). When wins and losses are both 2.5% and cost is 0.20%, break-even is 54%. Replace those sample inputs with results from your own defined strategy.
Should candlestick patterns be combined with RSI?
RSI can add information in a specific strategy, but it should not be assumed to improve a candle signal. In the Thailand study, filters based on %D, RSI, and MFI generally did not improve profitability or prediction accuracy. Test the combined rule independently and account for the extra search.
How many trades are needed to validate a candlestick strategy?
There is no universal minimum that makes a strategy valid. The needed sample depends on the size and variability of the edge, dependence between trades, the number of rules tested, and the confidence you need. Report the exact trade count and uncertainty, then check whether the result survives fresh periods and different market conditions.
Bottom Line
Candlestick patterns compress price behavior into a visual label, but the label does not include the rules that make a trading system testable.
Bulkowski’s bullish kicking result says a rare pattern reversed the prior trend only slightly more often than it did not. It does not provide the trade definition or payoff data needed to claim a profitable edge. Research also shows that changing the holding rule, market, cost assumption, or test design can change the result.
Before trading a candle signal, define the rule, measure the payoff, subtract realistic costs, protect a fresh test period, and check stability across time and instruments. If those pieces are missing, the pattern has not earned capital.
YOUR TURN
Pick one candle setup you follow and write down its entry, exit, average win, average loss, and realistic cost. Which missing input changes your confidence the most?
AI tools assisted with source organization, calculation cross-checks, and editorial quality control. Danny Hwang reviewed the primary-source claims, formulas, scope limits, and final conclusions.
Update history
- Current 2026-09-29 CORRECTION
Separated Bulkowski’s reported reversal statistic from the hypothetical trade-win-rate example; tightened the research interpretation, added holding-strategy evidence, updated trust components and canonical links, and refreshed the SEO package.
- Archive 2026-07-21 UPDATE
Rebuilt the article around the distinction between reversal rate and trade win rate; removed the unsupported long-horizon portfolio-loss projection, mismatched calculator, and unverified pattern statistics; refreshed source links and methodology limits.