Quant · May 2025
GARCH Volatility Modeling and Stochastic Time Series
A GARCH(1,1) implementation applied to SPY and five single-name equities, validated using AIC, BIC, and Ljung-Box diagnostics. GARCH, LSTM, and rolling volatility baselines were benchmarked across the 2020 and 2022 stress periods through walk-forward error analysis.
SPY + 5 names
Universe
2020, 2022
Stress periods
AIC/BIC, Ljung-Box
Diagnostics
Problem
Volatility models are prone to overfitting. I sought a comparison that would withstand the 2020 and 2022 stress regimes and clearly expose the limitations of each approach.
Approach
I fit GARCH(1,1) to SPY and five single-name equities, validated the fits with AIC, BIC, and Ljung-Box residual diagnostics, and benchmarked GARCH, LSTM, and rolling-sigma baselines using walk-forward error analysis.
Results
The study characterized the limitations of each model under high-volatility regimes and identified conditions in which simpler statistical models remained more interpretable than the LSTM alternative.
Stack