GARCH Models in Python
Chelsea Yang
Data Science Instructor
Auto Regressive Conditional Heteroskedasticity
Developed by Robert F. Engle (Nobel prize laureate 2003)
"Generalized" ARCH
Developed by Tim Bollerslev (Robert F. Engle's student)
White noise (z): Uncorrelated random variables with a zero mean and a finite variance
Residual = predicted value - observed value
Expected return: $$ \mu_t = Expected[r_t | I(t-1)] $$
Residual (prediction error): $$ r_t = \mu_t + \epsilon_t $$
Expected volatility: $$\sigma^2 = Expected[(r_t - \mu_t)^2 | I(t-1)]$$
Volatility is related to the residuals: $$ \epsilon_t = \sigma_t * \zeta (WhiteNoise)$$
Autoregressive: predict future behavior based on past behavior
Volatility as a weighted average of past information
To make the GARCH(1,1) process realistic, it requires:
$$\omega, \alpha, \beta >= 0 $$
$$\alpha + \beta <1$$
long-run variance: $$\omega / (1-\alpha - \beta)$$
GARCH Models in Python