Candlestick Patterns and Volume Analysis

Core Point: Candlestick patterns and volume-price relationships are cornerstones of traditional technical analysis, but have limited predictive power when used alone. The correct approach is to use them as feature engineering inputs, not direct trading signals.

Why Understand These?

If you've used any trading software (TradingView, Thinkorswim, Interactive Brokers), you've certainly seen various candlestick pattern annotations and volume bar charts. Many traders rely on these visual patterns for decision-making.

As a quant practitioner, you need to understand these concepts for several reasons:

1. Communication language: Common vocabulary when talking with traders and analysts
2. Feature engineering: These patterns can be quantified as input features for ML models
3. Strategy evaluation: Understanding traditional methods' limitations helps you design better systems
4. Market psychology: Candlestick patterns reflect the battle between buyers and sellers

1. Candlestick Basics

1.1 Candlestick Components

A single candlestick contains four price points:

1.2 Information Content of Candlesticks

2. Common Single Candlestick Patterns

2.1 Doji

    │
   ─┼─  ← Open ≈ Close
    │

Characteristics: Very small or no body, upper and lower shadows can vary in length

Traditional interpretation: Signal of potential trend reversal Actual effectiveness: Prediction accuracy around 50-55% when used alone

2.2 Hammer and Hanging Man

   │
   █  ← Small body at top
   │
   │ ← Long lower shadow (at least 2x body length)

Difference is in context:

• Hammer: Appears at bottom of downtrend → Potential upward reversal
• Hanging Man: Appears at top of uptrend → Potential downward reversal

2.3 Shooting Star

   │ ← Long upper shadow
   │
   █  ← Small body at bottom

Meaning: Appears at top of uptrend, suggests strong resistance above, possible top

2.4 Large Bullish and Bearish Candles

Quantitative definition: Body length > 1.5× recent ATR

3. Common Multi-Candlestick Patterns

3.1 Engulfing Pattern

Bullish Engulfing (at downtrend):    Bearish Engulfing (at uptrend):

    █                                      │
    █  First small bearish               ███ First small bullish
   ┌─┐                                   ┌─┐
   │ │ Second large bullish              █ █ Second large bearish
   │ │ Completely covers first           █ █ Completely covers first
   └─┘                                   └─┘

Conditions:

1. Second candle's body completely covers the first
2. Two candles are opposite colors
3. Appears in a clear trend

3.2 Morning Star and Evening Star

Morning Star (bottom reversal):

█     ← First: Large bearish
█
█
  ┼   ← Second: Small body (doji or small candle), gaps down
 ┌┐
 ││   ← Third: Large bullish, closes well into first candle's body
 ││

Evening Star: Mirror image of Morning Star, appears at tops

3.3 Three Black Crows and Three White Soldiers

4. Volume Analysis Fundamentals

4.1 Meaning of Volume

Volume = Number of shares traded between buyers and sellers

Core principle: Price is direction, volume is momentum

4.2 Key Volume-Price Patterns

Volume Breakout

Price ─────────┐
              └──────── Breakout
Volume       ████████ ← Volume significantly increases (>1.5× average)

Meaning: Breakout confirmed by capital, high credibility

Low-Volume Pullback

Price    /\
       /  \  ← Pullback
      /
Volume ██  █ ← Volume shrinks during pullback

Meaning: Pullback is normal profit-taking, not trend reversal

Volume-Price Divergence

Price   /\  /\  /\ ← Price makes new highs
      /  \/  \/
Volume ██ █  ▪ ← Volume decreases

Meaning: Uptrend momentum exhausting, possible top (similar to MACD divergence)

4.3 Common Volume Indicators

5. Empirical Research on Pattern Recognition

5.1 Academic Research Conclusions

Multiple academic studies have tested the predictive power of candlestick patterns:

Summary:

• Pattern-only win rates typically 50-55%
• After trading costs, most pattern strategies aren't profitable
• Pattern effectiveness may decay over time (learned by market)

5.2 Why Patterns "Appear to Work"

1. Confirmation bias: People remember successes, forget failures
2. Hindsight bias: Looking at past charts, you can always find "perfect" patterns
3. Survivorship bias: Only successful traders share their stories
4. Vague definitions: What counts as a "standard" hammer? No precise definition

5.3 When Might Patterns Be Effective?

Research shows patterns perform better under these conditions:

• Combined with volume: Patterns with volume confirmation are more reliable
• At key levels: Appearing near support/resistance
• Market environment: In specific volatility regimes
• Timeframes: Longer timeframes (daily, weekly) more reliable than minute charts

6. Quantitative Implementation: Patterns as Features

6.1 Challenges in Quantifying Patterns

Traditional pattern analysis is "pattern matching" but with vague definitions:

• "Long lower shadow" - how long is long?
• "Small body" - how small is small?
• "In a trend" - how do you define trend?

Solution: Convert patterns to continuous features, not discrete signals

6.2 Feature Engineering Example

def calculate_candlestick_features(df):
    """
    Convert candlestick patterns to continuous features
    df needs: open, high, low, close, volume columns
    """
    # Basic calculations
    df['body'] = df['close'] - df['open']
    df['body_abs'] = df['body'].abs()
    df['upper_shadow'] = df['high'] - df[['open', 'close']].max(axis=1)
    df['lower_shadow'] = df[['open', 'close']].min(axis=1) - df['low']
    df['range'] = df['high'] - df['low']

    # Feature 1: Body ratio (0=doji, 1=no shadows)
    df['body_ratio'] = df['body_abs'] / df['range'].replace(0, np.nan)

    # Feature 2: Upper shadow ratio
    df['upper_shadow_ratio'] = df['upper_shadow'] / df['range'].replace(0, np.nan)

    # Feature 3: Lower shadow ratio
    df['lower_shadow_ratio'] = df['lower_shadow'] / df['range'].replace(0, np.nan)

    # Feature 4: Close position (where close sits in day's range)
    df['close_position'] = (df['close'] - df['low']) / df['range'].replace(0, np.nan)

    # Feature 5: Relative body size (compared to recent ATR)
    atr = calculate_atr(df, period=14)
    df['relative_body'] = df['body_abs'] / atr

    # Feature 6: Engulfing ratio (current body covers previous body)
    df['engulfing_ratio'] = df['body_abs'] / df['body_abs'].shift(1)

    return df

def calculate_volume_features(df, periods=[5, 20]):
    """
    Calculate volume-related features
    """
    for p in periods:
        # Volume ratio
        df[f'volume_ratio_{p}'] = df['volume'] / df['volume'].rolling(p).mean()

    # OBV (On-Balance Volume)
    df['obv'] = (np.sign(df['close'].diff()).fillna(0) * df['volume']).cumsum()

    # OBV trend (slope of OBV moving average)
    df['obv_slope'] = df['obv'].rolling(10).apply(
        lambda x: np.polyfit(range(len(x)), x, 1)[0]
    )

    # Price-volume correlation (recent price change vs volume correlation)
    df['price_volume_corr'] = df['close'].pct_change().rolling(20).corr(
        df['volume'].pct_change()
    )

    return df

6.3 Feature Usage Recommendations

7. Multi-Agent Perspective

In multi-agent systems, candlestick patterns and volume analysis can be positioned as:

Key principles:

• Pattern features are one input, not the only basis
• Combine with other features (MACD, RSI, fundamentals) for comprehensive judgment
• Agent outputs probabilities, not deterministic signals

8. Common Misconceptions

Misconception 1: Patterns are "Secret Codes"

"Learning to read candlestick patterns guarantees consistent profits"

Truth: If patterns were so effective, the more people use them, the less effective they become (reflexivity)

Misconception 2: Patterns Work Everywhere

"A hammer always signals a bottom"

Truth: The same pattern may perform completely differently across market environments and instruments

Misconception 3: Volume is Always Meaningful

"High volume means institutional buying"

Truth: High volume could be retail panic, algorithmic trading, index rebalancing, and many other reasons

Misconception 4: Patterns Can Predict Precisely

"An evening star means tomorrow definitely drops"

Truth: Patterns only provide probabilistic edge (if any), not deterministic predictions

Key Takeaways

• Candlestick patterns are visual representations of buyer-seller battles
• Pattern-only prediction power is limited (50-55% win rate)
• Volume confirmation can improve pattern reliability
• Correct usage is quantifying patterns as ML features, not direct signals
• In multi-agent systems, patterns are one of many inputs

Further Reading

• Lesson 04: The Real Role of Technical Indicators - MACD, RSI and other indicators explained
• Lesson 09: Supervised Learning in Quantitative Trading - How to input features into ML models
• Background: Feature Engineering Common Pitfalls - Avoid feature engineering mistakes

References

• Nison, S. (1991). Japanese Candlestick Charting Techniques
• Lo, A., Mamaysky, H., & Wang, J. (2000). Foundations of Technical Analysis. Journal of Finance
• Murphy, J. (1999). Technical Analysis of the Financial Markets