Projects

Scroll ←

LSTM Forecasting of Bitcoin Implied Volatility (DVOL)

🟦 Writing Phase
Master's Thesis – Economics (STEM Designated)
This thesis develops and validates a Long Short-Term Memory (LSTM) neural network model for forecasting Bitcoin implied volatility (DVOL), the Deribit 30-day volatility index. Using a unified framework of 17 models (13 linear/tree baselines + 4 LSTM variants), we demonstrate that LSTM models achieve competitive performance (R² = 0.9287) when properly evaluated.
Key Findings
  • LSTM with lagged volatility features achieves R² = 0.9287, competitive with best linear/tree models
  • 41,055 hourly samples with 17-model comparison framework
  • Lagged volatility features alone achieve near-optimal performance (7 features)
  • All models achieve ~50% directional accuracy - hourly DVOL direction is fundamentally unpredictable
  • VaR backtesting passes 95%/99% Kupiec tests - model does not underestimate tail risk
Model Performance
R²: 0.9287 | RMSE: 1.71 | MAE: 1.28 | Directional Accuracy: 50.1%
Benchmark Comparison
  • HAR-RV (3 features): R² = 0.9454
  • Random Forest (11 features): R² = 0.9492
  • OLS (11 features): R² = 0.9490
Academic Contributions
  • Multi-window normalization analysis: 72-hour window optimal for level prediction
  • Standard Lee-Mykland (2008) implementation verification
  • Pesaran-Timmermann (1992) directional accuracy methodology
  • DataParallel evaluation artifact identification (13.3% R² degradation fix)
  • Unified 17-model comparison framework
Technologies Used
Python, PyTorch, TensorFlow/Keras, Pandas, NumPy, Scikit-learn, SHAP
View on GitHub
Scroll ←

Credit Market Volatility Research

🟩 Completed
Quantitative Finance & Econometrics
This project investigates how credit market stress impacts the volatility of both Bitcoin and NASDAQ assets. Using a panel data framework and quantile regression, it explores nuanced, heterogeneous effects across different volatility regimes. The analysis highlights the role of credit spreads, Treasury yield curve, and implied volatility in driving asset volatility, with a focus on asset-specific risk transmission.
Objectives
  • Model the impact of credit market stress on asset volatility
  • Compare volatility dynamics between Bitcoin and NASDAQ
  • Visualize and interpret quantile regression results
Technologies Used
Stata (panel data prep, regression, quantile regression, table generation), Python (pandas, numpy, matplotlib for data visualization)
View on GitHub
Stats_of_Key_Variables.png Stats_by_Asset.png Quintile_Regression_Bitcoin.png Quintile_Regression_Nasdaq.png py_plot_L_baa_aaa_spread.png py_plot_L_treasury_spread.png py_plot_L_implied_vol.png py_plot_L_neg_log_ret.png
← Scroll →
Scroll ←

ES Futures VPOC Strategy Backtester

🟩 Completed
Algorithmic Trading & Machine Learning
Advanced algorithmic trading strategy for E-mini S&P 500 (ES) futures that combines Volume Point of Control (VPOC) analysis with statistical validation and machine learning. The strategy identifies high-probability trading opportunities by analyzing volume distribution patterns, value area migrations, and market microstructure. ML enhancements include feature engineering, neural network models, and distributed training for AMD GPUs.
Objectives
  • Develop and backtest VPOC-based trading strategies
  • Integrate machine learning for signal generation and risk management
  • Validate strategy performance with statistical and ML metrics
Technologies Used
Python (pandas, numpy, matplotlib, seaborn, scikit-learn, statsmodels, pandas-ta, torch, mpi4py), Jupyter Notebooks (strategy prototyping, analysis)
View on GitHub
strategy_comparison.png
Click Me ↑