Bitcoin Return Direction Prediction

I designed and built an end-to-end algorithmic trading framework using Logistic Regression, XGBoost, and Bidirectional LSTM models to forecast daily price movements. My work covered the entire quantitative pipeline: engineering complex technical and macroeconomic features, and writing the execution logic complete with fractional Kelly position sizing and Triple-Barrier risk management. To ensure the strategy's real-world viability, I validated the system through rigorous walk-forward backtesting, ultimately proving its statistical edge and outperformance against a standard buy-and-hold baseline.

0 Rating
0 Reviews
5 Students Enrolled

Wishlist

Free

Enroll Now

Tags:

Courselet Content

2 components

	Bitcoin Return Direction Prediction (pdf)	9.82 M
		17.48 M

Requirements

Numpy, Pandas, Scipy, Scikit-learn, TensorFlow, XGBoost, Optuna, yfinance, Matplotlib, Seaborn Data Sources (yfinance tickers): BTC-USD, ETH-USD, ^GSPC, SPY, DX-Y.NYB, ^VIX, GLD, GC=F, TLT, BITO

General Overview

Description

Project: Bitcoin Return Direction Prediction

Data & Feature Engineering: Built a robust pipeline to aggregate Bitcoin OHLCV data, macroeconomic indicators (S&P 500, DXY), and sentiment indices. Engineered 20+ advanced features, including volatility regimes, cross-asset correlations, and rolling momentum.
Machine Learning Architecture: Developed an L2-regularized logistic regression baseline, an XGBoost ensemble gated by ADX-based regime detection, and a bidirectional LSTM with 4-head attention to capture non-linear temporal dynamics.
Algorithmic Execution & Risk: Designed dynamic risk management using Triple-Barrier logic (ATR-based stop-loss and take-profit) alongside fractional Kelly position sizing. Programmed protective overlays, including a forced "loss cooldown" mechanism to protect equity during market reversals.
Backtesting & Validation: Executed strict walk-forward cross-validation with expanding windows and purge gaps to eliminate look-ahead bias. Successfully demonstrated positive alpha and superior risk-adjusted returns (Sharpe, Calmar) against a standard buy-and-hold benchmark.