9

Other financial models: from ARMA to the GARCH family

The previous chapter dealt with stochastic processes, which consist of (returns) models involving a mixture of deterministic and stochastic components. By contrast, the models developed here present three major differences:

  • These models are deterministic; since they are aiming to model a non-deterministic variable such as a return, the difference between the model output and the actual observed value is a probabilistic error term.
  • By contrast with stochastic processes described by differential equations, these models are built in discrete time, in practice, the periodicity of the modeled return (daily, for example).
  • By contrast with usual Markovian stochastic processes, these models incorporate in the general case a limited number of previous return values, so that they are not Markovian.

For a time series of past observations on the variable x up to t − 1, all these processes are of the form

Unnumbered Display Equation

where f(.) is linear.

9.1 THE AUTOREGRESSIVE (AR) PROCESS

Let us consider a series of past returns {r0, …, rt−1} or, in short, {rt}, of 0 mean, such as:

Unnumbered Display Equation

where inlinet is the error term, also called “innovation”, 1 or “white noise”. In practice, ...

Get Mathematics of the Financial Markets: Financial Instruments and Derivatives Modelling, Valuation and Risk Issues now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.