EGARCH
Models log conditional variance and permits asymmetric responses to shocks.
How it works
EGARCH models the logarithm of conditional variance. This makes the variance positive after exponentiation and lets the equation respond differently to equally sized positive and negative standardised surprises. A common specification includes a term that changes with shock size and a separate term that changes sign with the shock.
Worked example
Suppose the signed response coefficient in a log-variance equation is โ0.1. All else equal, shocks of โ1 and +1 standard deviations give signed contributions of +0.1 and โ0.1. Holding the size term fixed, the next-day variances differ by a factor of exp(0.2), approximately 1.22.
Validation and limitations
Check the sign convention of the leverage parameter and benchmark long-horizon forecasts by simulation when necessary. Standardise past shocks consistently with the fitted conditional distribution. Assess volatility and tail calibration separately, especially if using an innovation distribution with heavy tails.