Quantitative Projects

Two Python-based quantitative finance projects showcasing volatility forecasting, risk management, and regime-based trading strategies using Bitcoin daily returns (2013–2024).

Projects

1. Rolling GARCH Volatility Forecasting & VaR Backtest

Walk-forward GARCH(1,1)/EGARCH(1,1) volatility forecasting with parametric Value-at-Risk backtesting on 4,069 daily BTC observations.

Key results:

Model	Out-of-Sample RMSE	MAE
EGARCH(1,1)	54.9	17.7
GARCH(1,1)	55.7	18.0
Constant Vol	62.0	37.6

99% VaR backtest: 1.63% breach rate (expected 1%), with all 50 breaches concentrated in stress regime (3.26% stress vs 0% calm). Kupiec test rejects at p=0.001.

• Tech: arch, scipy, numpy, pandas, matplotlib
• Directory: garch_var/

2. Regime-Based Position Sizing Signal

Markov-switching model identifies calm vs stress regimes, then constructs a position sizing signal that reduces exposure during high-volatility periods.

Key results:

Metric	Regime Strategy	Buy & Hold
Ann. Return	137.0%	109.8%
Ann. Volatility	58.1%	137.4%
Sharpe Ratio	2.36	0.80
Max Drawdown	-55.1%	-91.0%
Calmar Ratio	2.49	1.21

Stress regime accounts for only 13.4% of days. Sensitivity analysis (25 parameter combinations) shows Sharpe robust across grid (range: 1.77–2.98).

• Tech: statsmodels, numpy, pandas, matplotlib
• Directory: regime_signal/

Data

Bitcoin daily prices (Jan 2013 – Feb 2024, 4,069 observations), sourced from Investing.com via PhD research dataset.

Setup

pip install -r requirements.txt

# Project 1: ~15 min (walk-forward estimation)
python garch_var/rolling_garch.py

# Project 2: ~2 min
python regime_signal/regime_signal.py

Author

Hao Sun — PhD candidate in Economics, University of Bath