Quantitative research · Derivatives pricing · Market-data systems
MSc Data Analytics, Queen Mary University of London — September 2026
London, UK · Available full-time from September 2026
Forecasting realised volatility and pricing derivatives, with implementations written from first principles rather than assembled from libraries. Every model is benchmarked against a published baseline under walk-forward evaluation, and every numerical routine is verified against an independent reference before it is used.
| Repository | Summary | Result |
|---|---|---|
| chronos2-cpp | C++17 realised-volatility pipeline: Bloomberg BLPAPI → SQLite → feature engineering → Chronos-2 inference via ONNX Runtime | ADF implementation agrees with statsmodels to 12 decimal places; benchmarked against HAR-RV and persistence with Diebold–Mariano tests |
| chronos-volatility-forecasting | Transformer foundation model applied to realised-volatility forecasting under strict walk-forward evaluation | +55.7% skill over persistence; outperforms HAR-RV at p < 0.0001 across 30 years of data |
| temporal-neural-network | Temporal convolutional network built from scratch in NumPy — causal dilated convolutions, weight normalisation, GELU, AdamW, hand-derived backpropagation | ROC-AUC 0.728 on five-day-forward volatility regimes, +0.035 over HAR-RV across 20 years; gradients verified by finite differences |
| option-pricing-cpp | European option pricing in C++ — Black–Scholes closed form and Cox–Ross–Rubinstein binomial with all five Greeks, on a live market-data feed | Binomial converges to within 0.04% of closed form at 500 steps; put–call parity enforced by test |
| gaussian-process-classification-numpy | Laplace-approximation Gaussian process classifier with a mixed-type additive kernel and Gauss–Hermite quadrature | Test AUC 0.852 on 10,000 samples, with calibration and threshold-sensitivity analysis |
Languages — C++17, Python, SQL, MQL5
Quantitative — time-series forecasting, derivatives pricing, Bayesian inference, Gaussian processes, hypothesis testing, numerical methods
Libraries — NumPy, pandas, PyTorch, scikit-learn, statsmodels, Eigen, ONNX Runtime
Data & infrastructure — Bloomberg Terminal and BLPAPI (C++ and Python), Interactive Brokers TWS API, SQLite, Git, Linux
Practices — walk-forward evaluation, leakage-controlled ingestion, test-driven development, reproducible research, RAII
Top 9.8% worldwide, CME Trading Challenge.
