Deep Learning for Hedge Fund Precision

Project Overview

In this project, I explored how deep learning can enhance hedge fund strategies by building an AI-driven trading model to predict profitable trades on the SPY (S&P 500 ETF). The goal was to apply machine learning and financial engineering to derive meaningful insights from highly stochastic market data and improve swing trading decision-making.

Using 20 years of historical financial data, I designed a pipeline that performs extensive feature engineering, applies robust data labeling techniques, and tests multiple machine learning models to predict trade outcomes. The project balances both practical trading strategies and academic rigor, drawing on concepts from Advances in Financial Machine Learning by Marcos López de Prado.

Key Components

Data Collection

Engineered technical indicators (SMA, RSI, volatility) from financial data across multiple U.S. sectors to capture market patterns.

Triple Barrier Method

Implemented advanced labeling technique using profit targets and stop-loss thresholds based on volatility for more realistic trade classification.

ML/DL Models

Combined CatBoost for feature selection with LSTM neural networks to capture temporal patterns in market data.

Backtesting

Developed a Python-based backtesting engine to simulate trades and evaluate real-world performance metrics.

Results

0.54
ROC AUC (CatBoost)
0.138
MAE (LSTM)
0.004
Final Loss

The LSTM model proved significantly more effective than traditional approaches, demonstrating how neural networks can yield a meaningful predictive edge in algorithmic trading strategies.

Technical Skills

Python
Pandas/NumPy
Matplotlib/Seaborn
Scikit-learn
CatBoost
TensorFlow/Keras
LSTM Networks
Time Series Analysis
Triple Barrier Method
Feature Engineering
K-Fold Cross Validation
Backtesting