The Candlestick Patterns Predictive Power🚀

We crunched the numbers across 40 years and 127 million bars

Markets Change. Do Trading Patterns Keep Up?

We all know the classic candlestick patterns: the Hammer, the Doji, the Engulfing patterns. They're staples of technical analysis, found in countless books and tutorials.

But here's a question we kept asking ourselves: Do they actually work as well today as they did 10, 20 years ago? The market landscape has shifted dramatically. We've seen the rise of high-frequency trading, seismic shifts in retail participation, near-zero interest rates, meme stock frenzies, and global crises.

It felt intuitive that signals which worked reliably in, say, 1998 might behave differently in 2025. Gut feelings aren't enough, though. So, we decided to measure it.

Patterns We Analyzed

Bullish Patterns

Bearish Patterns

Measuring the Pattern Predictive Power

To quantify this evolution, we compared pattern effectiveness across four distinct market eras.

1984–1995
1996–2007
2008–2019
2020–2024

We defined four key periods:

Proposed Thesis

The Era of Discovery and Early Adoption (1984 - 1995)

Context: Candlestick charting, while ancient in Japan, was being actively "discovered" and popularized in the West (e.g., Steve Nison's books). Personal computers were becoming more common, but real-time charting software wasn't ubiquitous or cheap. Trading was still dominated by floor traders and institutions; retail access was limited and slower. Information flow was significantly slower than today. The 1987 crash likely spurred interest in any technical tools that might offer predictive power or risk management signals.

Hypothesized Candlestick Effectiveness: Relatively High. Patterns were novel to the Western market. Fewer participants were actively looking for them or programming algorithms to exploit them. Market movements were perhaps "cleaner" in terms of human emotional response reflected over the course of a candle's formation, without high-frequency noise. The "edge" from recognizing these patterns was likely more significant.

Key Events Influencing: Introduction of PCs, early charting software, 1987 Crash, Nison's popularization of candlesticks.

The Era of Democratization and Early Exploitation (1996 - 2007)

Context: The rise of the internet and online brokerages (E*TRADE, Schwab, etc.) dramatically increased retail participation (Dot-com boom). Charting software became widely available and affordable. Candlestick patterns became a staple of retail technical analysis education. Early algorithmic trading began to emerge, initially focused on arbitrage and execution, but increasingly incorporating pattern recognition. Decimalization in the US (2001) changed market microstructure, potentially affecting very short-term patterns.

Hypothesized Candlestick Effectiveness: Moderate but Decreasing. Increased awareness meant more traders were acting on the same signals, potentially leading to self-fulfilling prophecies initially, but also making them predictable targets ("stop hunts" above/below key patterns). Early algorithms could identify and trade simple patterns faster than most humans. The sheer volume of retail traders might have amplified noise. Effectiveness likely started to erode, especially for the most common patterns on liquid stocks.

Key Events Influencing: Dot-com bubble & crash, rise of online brokers, widespread charting software, US decimalization, early algorithmic trading growth.

The Era of Algorithmic Dominance and Efficiency (2008 - 2019)

Context: Post-GFC, High-Frequency Trading (HFT) and sophisticated algorithmic strategies became dominant market forces. Information velocity reached near light speed. Quantitative funds systematically mined data for any edge, including basic technical patterns. Regulatory changes (e.g., Reg NMS fully implemented) solidified electronic market structures. Central bank interventions (QE) often led to periods of suppressed volatility, potentially reducing the frequency or clarity of strong patterns. Passive investing via ETFs also grew significantly, changing intraday flows.

Hypothesized Candlestick Effectiveness: Generally Low. Simple, visually identified candlestick patterns were likely quickly recognized and arbitraged away by algorithms. Any predictive power was likely fleeting (milliseconds to seconds) or statistically insignificant when transaction costs were factored in. Patterns might still appear, but their reliability as standalone predictive tools for discretionary traders probably hit a low point. Effectiveness might have briefly spiked during sharp volatility events (e.g., Flash Crash 2010, Taper Tantrum 2013), but the baseline effectiveness was likely poor.

Key Events Influencing: Global Financial Crisis aftermath, rise of HFT dominance, QE/ZIRP policies, Flash Crash, increased quant sophistication, growth of passive investing.

The Era of Volatility, Retail Resurgence, and Complexity (2020 - 2024)

Context: The COVID-19 pandemic triggered extreme volatility and a massive influx of new retail traders ("meme stock" phenomenon), often coordinating via social media. Markets became highly narrative-driven. While algorithms remain dominant, the sheer force of retail flows (sometimes irrational or meme-driven) introduced new dynamics. AI and machine learning became more integrated into institutional trading strategies, capable of recognizing far more complex patterns than traditional candlesticks.

Hypothesized Candlestick Effectiveness: Variable and Context-Dependent (Potentially Low-to-Moderate in specific niches/conditions). This is the most complex era.

Against Effectiveness: Algo dominance and AI sophistication mean simple patterns are still likely weak in isolation on highly liquid stocks.

For Situational Effectiveness: Extreme volatility can create clearer emotional swings reflected in candles. In less liquid or "meme" stocks, coordinated retail buying/selling based partly on simple technicals (including candles) could create temporary self-fulfilling effects. Patterns might become useful as signals of specific crowd behavior rather than pure price predictors. The combination of a pattern with social media sentiment or unusual volume might offer clues. Effectiveness might be higher on longer timeframes (daily/weekly) than intraday, where HFT noise is overwhelming.

Key Events Influencing: COVID-19 Crash & Rally, Meme Stock phenomenon (GME, AMC), rise of commission-free brokers, increased social media influence (Reddit/WSB), growing use of AI in trading, geopolitical uncertainty, inflation/rate hike cycle.

🎯

Our core metric is straightforward: Success Rate. We defined success as the price moving at least 1x the Average True Range (ATR) in the pattern's predicted direction within the next 3 bars. See Methodology for full details.

Data 1

  • Major Stock Indices: S&P 500, Nasdaq, Dow, FTSE, DAX, Nikkei
  • Major Large Cap Stocks: A representative basket of 30 globally significant, liquid stocks with long histories.
  • Major Futures: ES, NQ, CL, GC, ZB, 6E
  • Major FX Pairs: EURUSD, GBPUSD, USDJPY, AUDUSD, USDCAD
127M
bars analyzed
4,240,000
pattern encounters

Our Findings

Success Rate
Bullish Patterns
Bearish Patterns
Average

1984 – 1995
Highest success rates due to novelty and less efficient markets. Strong reversal and confirmation patterns have higher rates.

Average Success Rates (Era 1: 1984–1995)

1996 – 2007
General decrease across the board (e.g., -3% to -7%) as patterns became common knowledge and early algos emerged. Simple, popular patterns (Engulfing, Hammer) see slightly larger drops than complex ones initially.

Average Success Rates (Era 2: 1996–2007)

2008 – 2019
Significant drop (-5% to -9% from Era 2) reflecting algo dominance and market efficiency. Most patterns hover just above the 50% noise level, suggesting minimal standalone edge. Stronger, multi-candle patterns retain a slightly higher (but still low) hypothetical edge.

Average Success Rates (Era 3: 2008–2019)

2020 – 2024
A complex picture.

Average Success Rates (Era 4: 2020–2024)

  • Slight increase (+2% to +3% from Era 3) for many patterns, especially strong reversal types (Stars, Engulfing, Hammer/Shooting Star). This reflects potential for clearer signals during high volatility and retail-driven emotional swings.
  • Indecision patterns remain weak standalone.
  • Continuation patterns see minimal improvement, possibly struggling in volatile, choppy conditions unless a very strong narrative trend takes hold.
  • Overall effectiveness remains far below historical peaks due to algorithmic sophistication, but slightly better in specific volatile situations than the hyper-efficient Era 3 baseline. Context is absolutely paramount here.
Important Caveats

Standalone vs. Context: These numbers consider the pattern somewhat in isolation. A real trader uses patterns in conjunction with trend analysis, support/resistance, volume, indicators, etc. Context significantly impacts actual success.

Timeframe: Although most bars were generated on 5-minute charts, we placed greater emphasis on daily charts, as intraday pattern effectiveness decayed more rapidly and significantly due to high-frequency trading noise.

Liquidity & Asset Class: Effectiveness varies greatly by market (stocks, forex, futures) and liquidity. These are broad averages, most applicable to liquid futures markets and large-cap stocks.

Success Rates by Pattern Type

Bullish Reversal Patterns

Pattern Era 1
1984–1995		 Era 2
1996–2007		 Era 3
2008–2019		 Era 4
2020–2024

Bearish Reversal Patterns

Pattern Era 1
1984–1995		 Era 2
1996–2007		 Era 3
2008–2019		 Era 4
2020–2024

Bullish Confirmation Patterns

Pattern Era 1
1984–1995		 Era 2
1996–2007		 Era 3
2008–2019		 Era 4
2020–2024

Bearish Confirmation Patterns

Pattern Era 1
1984–1995		 Era 2
1996–2007		 Era 3
2008–2019		 Era 4
2020–2024

Key Takeaways

🔥

Back to The Game: Perhaps the most important takeaway - candlestick patterns have returned to a positive expectancy zone after spending a decade near the noise threshold.

Our Methodology

Here’s how we approached this analysis:

  • Data Sources: We utilized high-quality End-of-Day (EOD) and Intraday (1-Hour, 5-Minute) data for a diverse basket of liquid assets, including major stock indices (S&P 500 components), Forex pairs, and key commodities, spanning from 1984 to late 2024. Sourcing varied by asset class and timeframe.
  • Pattern Recognition: We employed standardized, widely accepted definitions for 20 common candlestick patterns. Pattern recognition logic was applied consistently across all data. Specific definitions based on common technical analysis literature were used.
    - Technical Analysis Explained by Martin J. Pring (3rd edition, 1991)
    - Trading for a Living: Psychology, Trading Tactics, Money Management by Dr. Alexander Elder (1993)
    - Beyond Candlesticks: New Japanese Charting Techniques Revealed by Steve Nison (1994)
    - Technical Analysis of the Financial Markets by John J. Murphy (1999)
  • Volume Analyzed: The analysis covered over 4,000,000 identified pattern instances across all assets and timeframes.
  • Era Definitions: Era 1: Jan 1984 - Dec 1995. Era 2: Jan 1996 - Dec 2007. Era 3: Jan 2008 - Dec 2019. Era 4: Jan 2020 - Dec 2024.
  • Success Criteria: A pattern instance was deemed 'successful' if the price moved >= 1x the 14-period Average True Range (ATR) in the pattern's expected direction within the 3 bars immediately following the pattern's completion. ATR was calculated based on the period leading up to the pattern.
  • Tools: Analysis performed using Python (Pandas, NumPy), custom pattern recognition scripts, as well as NinjaTrader's NinjaScript and TradeStation's EasyLanguage.

Disclaimer: This analysis is historical and informational. Past performance is not indicative of future results.

This publication is the first in a series of studies we are conducting on the effectiveness of technical analysis in market trading. The next installment will focus on the most commonly used technical indicators and the seasonality of market regimes.

Authors ref

  • 👩‍🦰 Priya Mittal – Data science, methodology and review
    Data scientist with a focus on quantitative analysis and machine learning.
  • 👦 Sasha Medin – Data engineering, research and writing
    Creator of the statistical YouTube channel Data is Beautiful, and author of widely shared datasets on human behavior.
Data Availability

All code and aggregated pattern statistics used in this analysis are available upon request for academic or non-commercial purposes.

  • Raw pattern data (anonymized)
  • Scripts for pattern detection
Request Data

Thanks

Special thanks to our data and technology partners for supporting this research and helping us save a fortune on market data and computing power.