Skip to content

Latest commit

 

History

History
87 lines (58 loc) · 6.2 KB

README.md

File metadata and controls

87 lines (58 loc) · 6.2 KB

TradingPatternScanner

Python CI

Author: Preetam Sharma

Overview

Trading Pattern Scanner Identifies complex patterns like head and shoulder, wedge and many more.

New Enhancements

Four new features for pattern detection have been added:

  1. Basic Head-Shoulder Detection: The initial unfiltered version of pattern detection. It uses a rolling window to track high and low points, then identifies Head and Shoulder and Inverse Head and Shoulder patterns.
  2. Head-Shoulder Detection with Savitzky-Golay Filter: This feature uses the Savitzky-Golay filter to reduce noise and improve pattern detection. It also considers the height of the "Head" or "Inverse Head" to avoid false pattern recognition.
  3. Head-Shoulder Detection with Kalman Filter: This feature utilizes the Kalman Filter, a recursive filter that estimates the state of a system in real time. It's particularly suitable for financial data due to its inherent noise and uncertainties.
  4. Head-Shoulder Detection with Wavelet Denoising: This final feature applies wavelet denoising to eliminate noise while preserving key features in the data. It makes pattern detection more robust and reliable, especially in the presence of high-frequency noise.

These enhancements provide more accurate pattern detection for your financial analysis needs.

Analysis

Each method has been rigorously tested and analysed on synthetic data that closely mirrors real-world financial data. However, it's important to note that synthetic data is not an exact representation of the real-world, and the performance of each algorithm may vary in a live setting. Therefore, users are encouraged to test each algorithm against their own datasets and pick the one that best suits their needs.

  • Accuracy for head_shoulder_pattern_window: 78.50%
  • Accuracy for head_shoulder_pattern_filter: 78.50%
  • Accuracy for head_shoulder_pattern_kf: 73.50%
  • Accuracy for head_shoulder_pattern_wavelet: 84.50%

Analysis

Heatmap Interpretation

  • For instance, let's consider a cell in the 2nd row and 2nd column. the score is 10, it means that a significant number of instances were correctly identified as "Head and Shoulder" pattern (abbreviated as HS).

  • On the contrary, a dark cell outside this diagonal indicates a high number of misclassifications. For example, a dark cell at the intersection of "HS" row and "I-HS" column would mean that a large number of instances were true "HS" but were incorrectly predicted as "Inverse Head and Shoulder" (abbreviated as I-HS) by the scanner.

Abbreviations

The abbreviations used in the heatmap and the code are as follows:

  • HS - Head and Shoulder pattern
  • I-HS - Inverse Head and Shoulder pattern

Installation / Usage

Install using pip:

$ pip install tradingpattern

TradingPatternScanner

Trading patterns:

  • Head and Shoulder and inverse Head and Shoulder: These patterns indicate a potential reversal in the market, with the "head" being the highest point, and the "shoulders" being the points on either side at a slightly lower level.
  • Multiple top and bottom: These patterns indicate a range-bound market, with multiple highs and lows forming a horizontal range.
  • Horizontal support and resistance: These patterns indicate key levels at which the market has previously struggled to break through.
  • Ascending and Descending Triangle pattern: These patterns indicate a potential breakout in the market, with the upper trendline being resistance and the lower trendline being support.
  • Wedge up and down: These patterns indicate a potential reversal in the market, with the trendlines converging towards each other.
  • Channel up and down: These patterns indicate a strong trend in the market, with price moving within a well-defined upper and lower trendline.
  • Double top and bottom: These patterns indicate a potential reversal in the market, with the market hitting a high or low twice and then reversing.
  • Trend line support and resistance: These patterns indicate key levels at which the market is likely to experience support or resistance based on historical price action.
  • Finding Higher-High and Lower-Low

Designed for fast performance:

  • Uses only Pandas as Numpy, no other external libraries: This approach helps to keep the library lightweight and fast.
  • Uses the concept of vectorization: This approach helps to improve performance by processing large amounts of data at once, rather than iterating over each individual data point.

New and Unique:

  • No other python library exists for such task currently: This library is new and unique, as it aims to provide an all-in-one solution for identifying various trading patterns.

Lets check if its works for simplicity I used finviz and checked the pattern with the respective stock.

  • Head and Shoulder: Head and Shoulder

We can see that it finds out that we have inverse head and shoulder pattern in the stock on 9th Januray 2023 in 1 day interval. Lets match with Finviz. Finviz

  • We can see that Finviz also detects on 9th Januray 2023 in 1 day interval.
  • You can adjust the window size to your liking. A smaller window size will be more sensitive to detecting patterns, but it will also increase the chances of false positives. A larger window size will be less sensitive to detecting patterns, but it will also decrease the chances of false positives.

Future add-ons:

  • Request your favourite pattern to get added in the list: The library is open for suggestions for adding new patterns.
  • Work on visualization and plotting: The library can be extended to include visualization and plotting features to help users better understand the patterns identified.
  • Add unit testing: The library can be extended to include unit testing to ensure that the code is working as expected and to catch any bugs early on.