Disclaimer: All work in this repository is my own. It was developed independently on personal time and does not reflect the views, proprietary methods, or intellectual property of any employer, past or present.
A collection of quantitative research projects spanning predictive modelling, systematic signal development, and deep learning on alternative data.
Forecasting hedge fund reporting stops and performance using XGBoost on 5,592 funds and 300k+ monthly observations. Published in the Journal of Forecasting (Wiley, 2020).
87% accuracy on reporting stop prediction · 75% on absolute performance · 74% on relative performance
XGBoost · TimeSeriesSplit CV · scikit-learn · pandas
→ Start here: src/main.py · Full README
Systematic intraday mean-reversion signal using DCC-GARCH dynamic correlations. Includes a reusable backtesting engine, GARCH volatility modelling, and markout analysis.
Sharpe 3.37 · In-sample backtest with grid search across signal types and parameters
GARCH · DCC · PCA · arch · NumPy · pandas
→ Start here: research/fx/dcc_signal.py · Full README
Predicting tweet engagement using 6 deep learning architectures — from bag-of-words baselines to BERT embeddings with metadata features. Final project for MIT Sloan's Hands-on Deep Learning (15.S04).
64.57% accuracy with BERT + metadata concatenated model
TensorFlow/Keras · BERT · GloVe · Twitter API · scikit-learn
→ Start here: notebooks/model_comparison.ipynb · Full README