نوع مقاله : مقاله علمی پژوهشی
نویسندگان
1 دانشکده مالی دانشگاه خوارزمی
2 دانشکده علوم مالی دانشگاه خوارزمی
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Forecasting price trends in financial markets—particularly in volatile and nonlinear environments such as cryptocurrency exchanges—has long posed a significant challenge for analysts, traders, and researchers. Among the tools commonly used for technical analysis, Japanese candlestick charts provide a powerful visual framework by depicting the behavior of prices over time. However, the interpretation of these candlestick patterns is often subjective and heuristic, lacking structured numerical criteria and reproducibility. This limitation reduces the reliability of traditional candlestick analysis. To address this issue, the present study introduces a fuzzy logic-based framework for modeling candlestick structures and predicting short-term price movements. The proposed model serves as a bridge between human intuition and algorithmic precision by leveraging fuzzy logic to quantify and formalize uncertain, linguistic concepts such as “short body” or “long upper shadow.”
The study analyzes hourly Bitcoin data from September 2017 to May 2025. Each candlestick is decomposed into body, upper shadow, and lower shadow, which are standardized and fuzzified using triangular membership functions tailored to Bitcoin’s volatility. Candles are then encoded into unique three-character fuzzy codes based on dominant membership values. These codes are linked to the next candle’s trend (bullish, bearish, or neutral), enabling the extraction of statistical trend tendencies. Two predictive models are built: a soft rule model, which predicts a trend if 3 out of 4 top-matching codes agree, and a hard rule model, which requires full agreement. Both models are tested on out-of-sample data and used to simulate hourly trading strategies.
Backtesting simulations are carried out under both fixed-capital and capital-adjusted strategies, assuming an initial investment of $100,000. In the soft rule model, over 10,891 trades were executed, resulting in a net profit of $190,419 with a win rate of 54.1% and an average profit per trade of $17.48. The hard rule model, with 6,761 trades, yielded a net profit of $187,979, a higher win rate of 55.8%, and a higher average profit per trade of $27.80. Both models significantly outperformed the benchmark buy-and-hold strategy.
To assess the statistical significance of the observed outperformance, a paired t-test was conducted using 1,000 randomly selected 14-day intervals (336 hourly candles each). The test compared the fuzzy model returns against the buy-and-hold returns for the same periods. The resulting p-values were <0.00001 for both soft and hard rule models, confirming the superiority of fuzzy strategies with more than 99.9% confidence. The comparison between the two fuzzy models yielded a p-value of 0.463, indicating no statistically significant difference in average returns between them—though the hard rule model demonstrated a higher Sharpe ratio and lower volatility.
In addition to performance evaluation, frequency analysis of fuzzy codes revealed recurring patterns associated with specific trend directions. These statistically dominant bullish and bearish fuzzy patterns could serve as a foundation for rule-based market forecasting systems. They also offer a promising opportunity for expanding the fuzzy knowledge base through semantic rule integration and asset-specific customization.
Overall, this study introduces a fully interpretable, standalone candlestick-based trading model that does not rely on external indicators or multi-candle sequences. The entire system—from feature extraction and fuzzification to decision-making—is transparent and reviewable, making it suitable for financial analysts seeking clarity and trustworthiness. Compared to black-box machine learning models, the fuzzy approach offers a balanced compromise between interpretability, adaptability, and predictive power.
کلیدواژهها [English]
)