The statistical tools described in this section are designed to model and forecast the conditional variance, or volatility, instead of the conditional mean of a variable. The analysis of the conditional variance may be useful for several reasons
such as pricing an option or improving the estimation of forecast intervals. The models described below assume that the conditional variance in time t
depends on past errors and variances. They are designed to model time varying volatility, in particular volatility clustering - a feature often displayed by financial market series. The variance at time t is expected to be higher when past
errors and variances were higher in the past and vice verse.
The phenomenon of time varying volatility is well known and generated a vast body of econometric literature following the seminal contributions by Engle (1982), Bollerslev (1986) and Taylor (1986) introducing the (generalized) autoregressive conditionally heteroskedastic ((G)ARCHi) process and the stochastic volatility model, respectively
NumXL has a complete set of tools for building on time-varying volatility models. The Add-in supports several variants of univariate GARCHi models, including standard ARCH/GARCH models, as well as asymmetric EGARCHi and GARCH in the mean (GARCH-Mi) models designed to capture leverage effects in asset returns