Candlestick Pattern Recognition Enhanced MACD-ATR Strategy for Cryptocurrency Trading

Authors: Zhe Niu (M.S. FinTech),

Trading Strategy & Plan

This project presents a candlestick pattern-enhanced MACD and ATR strategy that detects market microstructure patterns and swing trends in the cryptocurrency market. We combine rule-based indicators (MACD, ATR, EMA) with a CNN classifier trained to recognize candlestick pattern windows. The goal is to improve entry/exit accuracy in BTC/USDT trading through deep learning-enhanced pattern recognition.

Candlestick Pattern System for Market Trend Recognition

Understanding and classifying Japanese candlestick patterns is critical for interpreting trader sentiment and anticipating future price movements. The chart below presents a comprehensive taxonomy of bullish, bearish, and neutral candlestick configurations widely used in technical analysis. These patterns serve as visual signals for trend reversals and continuations in volatile markets like cryptocurrency.

Recent research by Mersal et al. (2025) demonstrates that integrating 61 distinct candlestick patterns with CNN-based classification systems can drastically enhance the precision of market trend prediction. Their work applies a sliding window segmentation and TA-Lib pattern recognition to extract structured features from OHLC charts, followed by CNN training to classify bullish/bearish trends. Their approach achieved up to 99.3% accuracy, far exceeding traditional strategies.

This strategy draws inspiration from their methodology by using candlestick chart windows labeled with directional trends, validated with moving averages and additional technical indicators. These patterns—such as the Morning Star, Engulfing, Doji, and Shooting Star—provide reliable inputs for deep learning models, enabling the detection of microstructure signals across 15-minute intervals.

Core Strategy Concepts

Our strategy is based on the following key elements:

  • MACD Trend Signals: Using MACD histogram and signal line crossovers to confirm trend changes
  • EMA Price Position: Price location relative to EMA30 to determine overall trend direction
  • ATR Dynamic Stop-Loss: Adaptive adjustment of stop-loss levels based on market volatility
  • Visual Pattern Recognition: Using CNN deep learning models to identify candlestick patterns
  • Signal Confirmation Mechanism: Requiring multiple consecutive signals before executing trades to reduce false breakouts