|
Stochastic volatility models are a category of
stochastic
processes that have stochastic (random) second moments. Stated another
way, they have random volatility or are
conditionally heteroskedastic.
"Stochastic volatility model" is a technical term. While all stochastic volatility
models have stochastic second moments, not all models that have stochastic
second moments are called stochastic volatility models. In finance, two
categories of stochastic processes are widely used to model stochastic
second moments. One is stochastic volatility models. The other is
ARCH/GARCH models.
Both ARCH/GARCH and stochastic volatility models derive
their randomness from white noise processes. The
difference is that an ARCH/GARCH process depends on just one white noise
W. That white noise directly determines innovations in the
ARCH/GARCH process while also indirectly determining innovations in its
second moments. Stochastic volatility models generally depend on two white
noises, V and W. One directly determines
innovations in the stochastic volatility process. The other directly
determines innovations in its second moments.
Stochastic volatility models come in forms far more
diverse than those of ARCH or GARCH models. An example of a simple
univariate stochastic volatility model X is (see the
notation conventions documentation)
 |
[1] |
| [2] |
where V and W are independent
standard normal
Gaussian white noises.
Stochastic volatility models often employ logarithms as in [2] to ensure
conditional variances
are nonnegative.
While our example is of a
discrete-time
model,
continuous-time stochastic volatility models are widely used in
financial engineering.
|