Candlestick Pattern Detector

Generally, multi-candle patterns are more reliable than single-candle. Engulfing, morning/evening star, and three white soldiers/black crows have the highest reliability. However, context matters more than the pattern itself - any pattern at a key level with volume is stronger.

No. Start with the most common and reliable: hammer, shooting star, doji, engulfing, and morning/evening star. Master these before learning others. Quality of pattern identification matters more than quantity of patterns known.

Daily charts are most reliable - enough data per candle to be meaningful, less noise than lower timeframes. Weekly for longer-term positions. 4H and 1H for active trading. Lower timeframes have more noise and false patterns.

For single-candle patterns like doji, confirmation is essential. For strong patterns like engulfing at key levels, you can enter on the pattern itself. Confirmation increases probability but gives slightly worse entry. Match your approach to pattern strength.

Patterns fail due to: poor location (not at S/R), no prior trend (for reversal patterns), low volume (weak conviction), or external factors (news, broader market). This is why we use stops. Patterns are probabilities, not certainties.

Use patterns for timing within broader context. Example: weekly trend is up, daily at support, 4H shows hammer = high-probability long. Combine with volume, S/R, moving averages, RSI for confluence. More factors agreeing = higher probability.

Very important. Volume confirms conviction. High volume on signal candle shows commitment. Low volume patterns are weaker. Some add +10% to success rate for volume-confirmed patterns. Always check volume as a filter.

Engulfing: body engulfs prior body. Outside day: entire candle (body + wicks) contains prior candle's entire range. Outside day is stronger as it shows complete price rejection. Some count outside day as super-engulfing.

Patterns are less reliable in choppy/ranging markets. Options: 1) require stronger patterns (only engulfing, not doji), 2) require more confluence (at S/R), 3) reduce position size, 4) wait for clearer trend before trading patterns.

Reversal patterns attempt to catch trend changes, so yes, they trade against the current trend. BUT: they should only be traded after extended moves at major levels. Don't trade reversal patterns in the middle of a trend. Wait for exhaustion signs.

Steps: 1) Calculate candle metrics (body_ratio, wick_ratios, etc.), 2) Define pattern rules mathematically, 3) Add context assessment (trend, S/R distance), 4) Score pattern quality, 5) Filter by threshold. Implement in Python with pandas. Scan universe on each new candle.

Base rates without context: 50-55%. With proper context (S/R, trend, volume): 60-70%. Best setups (multiple confluence): 70%+. These are probabilities over many trades. Individual patterns can and will fail. Edge is in aggregate statistics.

Options: 1) Classification - features are candle metrics + context, target is forward return. 2) Pattern-specific models - predict success/failure of each pattern type. 3) CNN on chart images - learn visual patterns. Train on historical data, validate with walk-forward testing.

Track partial candles for awareness but only confirm patterns on candle close. A potential hammer at 3:55 PM might not be a hammer at 4:00 PM. For intraday, wait for candle completion. For daily, wait for market close before confirming.

Trading low-quality patterns degrades edge significantly. Base pattern without context might be 52% win rate. Adding trade friction (slippage, commissions), this becomes break-even or negative. Only trade high-quality patterns (65%+) where edge exceeds costs.