Project Insights Report: Advanced Python Back testing for Custom SHIB Trading Strategies

1. Project Overview

Objective

This project I set out to design, backtest, and refine automated trading strategies on SHIB-USD using historical data. I sought to leverage Python’s powerful ecosystem of libraries to simulate real-world trading scenarios, enabling insights into profitability, risk, and trade optimization. This report summarizes the insights from three successful backtests, illustrating Python's capabilities in building and evaluating custom trading strategies.

Purpose and Audience

This report targets traders, data scientists, and developers interested in systematic trading strategies. It highlights how Python’s flexibility and depth allow for high-level customization, bridging the gap between traditional finance and algorithmic trading.

2. Tools and Languages Used

3. Project Workflow

Step 1: Data Collection

• Objective: Retrieve and preprocess SHIB-USD historical data.
• Data Source: Yahoo Finance, selected for its comprehensive coverage and convenient access.
• Details:Collected hourly price data for SHIB from 2023-01-01 to 2024-01-01.Stored the data in a Pandas DataFrame, which allowed for time-series analysis and custom transformations.

Step 2: Strategy Development and Testing

I developed three distinct strategies, each utilizing different approaches to capitalize on SHIB price movements. Here’s a detailed look at each:

Strategy 1: Dip-Rebound Grid Strategy

• Objective: Capture profits by buying at pre-determined dip levels and selling at calculated rebounds.
• Logic:We set up price grids for SHIB-USD, buying at each price dip and selling after a rebound of 5%.Dynamic threshold adjustment using the Average True Range (ATR) allowed for adaptability to SHIB’s volatility.
• Script Breakdown:Data Loading and Grid Setup: Created grid levels for predefined buy and sell signals.Signal Generation: Triggered buys at each dip in the grid and sold after a 5% rebound.Backtest and Analysis: Used VectorBT to evaluate the strategy’s returns and visualize the cumulative performance.

Strategy 2: Reinforcement Learning-Based Trading Agent

• Objective: Develop a self-learning agent capable of identifying optimal trading actions based on market patterns.
• Logic:Leveraged Stable-Baselines3’s PPO (Proximal Policy Optimization) algorithm within a custom Gym environment.The agent learned from historical data, identifying patterns to make buy, hold, or sell decisions dynamically.
• Script Breakdown:Custom Environment: The Gym environment was designed to simulate SHIB trading, with action options and rewards calculated based on changes in total asset value.Agent Training: Trained using a reinforcement learning loop to improve trade timing and threshold adjustment.Evaluation: The agent’s actions were backtested to examine cumulative returns, showcasing its self-learning improvements over time.

Strategy 3: XGBoost Classifier with ATR and Moving Averages

• Objective: Use machine learning to classify trading opportunities based on trend and volatility indicators.
• Logic:Features like moving averages (SMA), volatility (rolling standard deviation), and momentum were calculated for SHIB price.An XGBoost classifier was trained to predict whether the price would increase after each dip, based on technical indicators.
• Script Breakdown:Feature Engineering: Extracted ATR, moving averages, and volatility as predictive features.Training: Used XGBoost to train on historical price movements, aiming for high accuracy in trend prediction.Integration with Backtesting: Triggered buy/sell signals based on classifier predictions, with enhanced profitability from informed entries.

4. Results and Insights

Backtest Results Summary

Key Insights

1. Dip-Rebound Grid Strategy:
2. Reinforcement Learning Agent:
3. XGBoost Classifier Strategy:

5. Conclusion and Recommendations

Conclusion

Python, combined with powerful libraries, enables rapid strategy development, from data gathering to advanced backtesting. The strategies above illustrate how different techniques can yield varying results depending on market dynamics and model assumptions.

Recommendations for Future Work

1. Hybrid Strategies:
2. Optimize and Fine-Tune RL Agents:
3. Use of Live Data for Real-Time Testing:

6. Next Steps: Encouraging Further Exploration

I encourage data scientists and developers to explore Python’s capabilities in quantitative finance. The open-source libraries offer flexibility and power, suitable for crafting custom trading strategies adaptable to various market conditions.

Python enables full control over each step, making it a valuable tool for anyone looking to gain a deeper understanding of algorithmic trading and systematic strategy development.

This report demonstrates the power and flexibility Python offers to test, refine, and optimize trading ideas. As seen, a mix of conventional technical analysis, reinforcement learning, and machine learning models provided us with rich insights into SHIB trading opportunities. I hope this inspires others to leverage Python for backtesting and quantitative analysis.

This project was run in Google Colab: