Candlestick Patterns

Candlestick patterns are rules that read the relationship between the open, high, low and close of one, two or three consecutive bars and treat that geometry as information about who controlled the period. A long lower wick with a close near the high says sellers pushed price down and were absorbed before the bar ended. A bar that opens below the prior close and closes above the prior open says the other side took control inside a single period. The pattern is not a forecast — it is a compressed description of what happened, and a strategy is what turns that description into a trigger with an entry, a stop and an exit.

## How the concept works

Almost every implementation has three layers. The first is the **definition**: the geometric conditions that make a bar or bar sequence qualify, expressed as ratios — body size relative to total range, wick length relative to body, where the close sits inside the range, how one bar's body compares to the previous one. The second layer is **context**: where the pattern is allowed to fire. The same shape is treated very differently at a trendline, at a prior swing, after an extended move, or in the middle of a range. The third layer is **execution**: entering at the close of the pattern or on a break of its extreme, placing the stop beyond the wick or at an ATR distance, and exiting at a fixed multiple, a structural level, or on an opposite signal.

Most of the difference between two versions of "the same" pattern lives in layers two and three, not in layer one.

## Main variants

**Single-bar patterns** — pin bar, hammer, shooting star, doji, marubozu — are defined by how the body and wicks divide the range. They are the most common family because they need no history beyond the bar itself.

**Two-bar patterns** — engulfing, harami, piercing line, dark cloud cover, tweezer tops and bottoms, kickers — compare consecutive bodies, closes and extremes. Definitions here vary a lot: whether wicks must be engulfed or only bodies, whether a gap is required, whether the second close must exceed a specific fraction of the first body.

**Three-bar and sequence patterns** — morning star, evening star, inside-bar false breakouts ("fakey" setups) — add a middle bar of indecision or a failed break as part of the condition.

**Confluence variants** combine a pattern with something else: a trendline or channel pullback, a moving average, an oscillator such as RSI or RMI, a support/resistance zone, a higher-timeframe direction filter, or a correlated instrument.

**Statistical and screener variants** treat patterns as objects to measure rather than to trade directly: multi-timeframe studies, cross-symbol scanners, and coded EAs where the definition must be fully numeric to run at all.

## What typically differentiates implementations

Threshold values in the pattern definition; whether the pattern must occur at a level or can fire anywhere; the trend filter and its lookback; the timeframe, since the same rule on M5 and on D1 samples two different populations; entry timing (bar close, break of the extreme, or a retest); stop placement; exit logic and whether the target is fixed or adaptive; the instrument and session; and whether the version is discretionary guidance or a mechanical specification.

## Common mistakes

Treating the pattern as a prediction rather than a trigger, and skipping the question of how the market behaves after that pattern without any filter. Leaving the definition ambiguous, which makes chart review a hindsight exercise instead of a test. Ignoring that candles are an artifact of the data feed: a daily candle in forex depends on the broker's rollover time, so a pattern can exist on one feed and not on another. Counting selected chart examples as evidence. Stacking filters until almost no trades remain. Assuming that a pattern's frequency implies an edge, when the edge — if any — usually comes from the context and exit. Underestimating spread and commission on low timeframes, where pattern-based stops are small. Evaluating the pattern before the bar closes, which introduces lookahead.

## How to evaluate and backtest a version

Start by rewriting the rules as unambiguous numeric conditions, including what happens on ties and edge cases. Then build a baseline: the raw pattern, no filters, fixed stop and target, on the intended instrument and timeframe. That baseline is the reference every filter has to improve on. Count occurrences, not just outcomes — a rule with too few triggers cannot be judged.

Include realistic costs, confirm signals only on closed bars, and check parameter sensitivity: a robust threshold shows a plateau of similar behavior, not an isolated spike. Run out-of-sample and walk-forward segments, and repeat the test on a second data source to see how much of the result depends on the feed. Test neighboring timeframes and related instruments. A useful control is the same exit logic with random entries: if the pattern adds nothing over that control, the exit is doing the work. Finally, segment results by context — with trend versus against, at a level versus not — to see where the rule actually contributes.

The decoded versions linked below combine these same primitives in different ways. Reading them side by side is the fastest way to see which part of a candlestick strategy is the pattern and which part is everything around it.

Strategies in this concept (60)

Frequently asked questions

Are candlestick patterns enough on their own to build a strategy?

A pattern only defines a trigger. To become a strategy it needs a context rule (when the trigger is valid), a stop, an exit and a position size. Before adding filters, it is worth testing the raw pattern with a fixed stop and target: that baseline tells you what the pattern contributes and what the surrounding logic contributes.

Which timeframe should candlestick patterns be used on?

There is no universal answer, and the choice changes the strategy rather than just its speed. Higher timeframes produce fewer, larger setups where costs matter less but samples are smaller. Lower timeframes produce many more occurrences with stops small enough that spread and commission become a significant part of the result. Treat each timeframe as a separate test.

What is the difference between a pin bar, a hammer and a shooting star?

Mostly naming convention and direction. All three describe a bar with a small body and one dominant wick; hammer and shooting star are the bullish and bearish cases in classical terminology, and pin bar is the price-action label for the same shape. What actually differs between implementations is the numeric threshold used for wick-to-body ratio and where the close must sit inside the range.

Do candlestick patterns behave the same in stocks, futures and forex?

Not necessarily. Several classical patterns were described on markets with overnight gaps, and their definitions assume a gap that rarely appears in continuously traded instruments. Session times, rollover conventions and the data provider also change where a candle opens and closes, so the same rule applied to different markets produces different bar populations.

How do I turn a discretionary candlestick setup into a mechanical rule?

Replace every qualitative word with a number: how long the wick must be relative to the range, how close to the extreme the close must be, how near a level counts as "at" the level, and how many bars back the trend filter looks. If a condition cannot be written that way, it cannot be backtested consistently, and chart review of it will carry hindsight bias.

Why do two backtests of the same pattern disagree?

Usually for three reasons: the pattern definitions differ in their thresholds or in whether wicks are included, the data feeds differ in session and timezone so the bars themselves differ, or the entry and exit rules differ. When comparing versions, compare all three layers — definition, context and execution — not just the pattern name.

All strategy concepts · Explore all strategies