The characteristic function of Gaussian stochastic volatility models: an analytic expression

Eduardo Abi Jaber

DOI: https://doi.org/10.48550/arXiv.2009.10972

2022-05-09

Abstract:Stochastic volatility models based on Gaussian processes, like fractional Brownian motion, are able to reproduce important stylized facts of financial markets such as rich autocorrelation structures, persistence and roughness of sample paths. This is made possible by virtue of the flexibility introduced in the choice of the covariance function of the Gaussian process. The price to pay is that, in general, such models are no longer Markovian nor semimartingales, which limits their practical use. We derive, in two different ways, an explicit analytic expression for the joint characteristic function of the log-price and its integrated variance in general Gaussian stochastic volatility models. Such analytic expression can be approximated by closed form matrix expressions. This opens the door to fast approximation of the joint density and pricing of derivatives on both the stock and its realized variance using Fourier inversion techniques. In the context of rough volatility modeling, our results apply to the (rough) fractional Stein--Stein model and provide the first analytic formulae for option pricing known to date, generalizing that of Stein--Stein, Sch{ö}bel-Zhu and a special case of Heston.

Probability,Computational Finance

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is how to find the explicit analytical expression of the joint characteristic function of the log - price and its integrated variance in the Gaussian stochastic volatility model. Specifically, the author aims to solve the following problems:

1. **Non - Markovian and semi - martingale limitations**: Traditional stochastic volatility models (such as the Heston model and the Stein - Stein model) have analytical solutions, but the autocorrelation structures they generate are too simple to fit well with the complex characteristics in the actual financial market, such as persistence and the roughness of sample paths. Moreover, these models are usually Markov processes or semi - martingales, which limit their practical applications.

2. **Lack of analytical expressions**: For stochastic volatility models based on Gaussian processes such as fractional Brownian motion, although they can better capture the complex characteristics of the financial market, due to their non - Markovian and non - semi - martingale properties, it is difficult to obtain analytical solutions. This makes rapid pricing and calibration difficult.

To solve these problems, the author derives the explicit analytical expressions of the joint characteristic function of the log - price and its integrated variance under two different methods. These expressions can be approximated as closed - form matrix expressions, thus making it possible to use Fourier inversion techniques to quickly approximate the joint density and price derivatives.

### Specific problem description

By introducing the general Gaussian stochastic volatility model, especially by extending the Stein - Stein and Schöbel - Zhu models, the paper solves the following specific problems:

- **Derivation of analytical expressions**: The author derives the explicit analytical expression of the joint characteristic function of the log - price and its integrated variance. This expression is applicable to the general Gaussian stochastic volatility model and can be degenerated into known classical models (such as Stein - Stein, Schöbel - Zhu and Heston models) in some specific cases.

- **Numerical implementation**: By discretizing the operator into a finite - dimensional matrix, the author shows how to approximate the expression of the characteristic function with a closed - form solution. This enables the rapid calculation of option prices and model calibration by numerical methods in practical applications.

- **Theoretical contribution**: The author's work fills the gap of analytical solutions in the rough volatility model, especially providing the analytical formula for option pricing for the first time in the fractional Stein - Stein model.

### Mathematical expressions

The specific expressions of the characteristic function are as follows:
\[
E\left[\exp\left(u \log \frac{S_T}{S_t} + w \int_t^T X_s^2 \, ds \right) \mid \mathcal{F}_t \right] = \exp\left( \langle g_t, \Psi_t g_t \rangle_{L^2} \right) \det(\Phi_t)^{1/2},
\]
where:
- \( g_t(s) = 1_{t \leq s} E\left[X_s - \int_t^T K(s,r) \kappa X_r \, dr \mid \mathcal{F}_t \right] \)
- \( \Psi_t = (id - bK^*)^{-1} a (id - 2a \tilde{\Sigma}_t)^{-1} (id - bK)^{-1} \)
- \( \Phi_t = \begin{cases}
(id - bK)(id - 2a \tilde{\Sigma}_t)(id - bK^*) & \text{if } K \text{ is symmetric}, \\
id - 2a \tilde{\Sigma}_t & \text{if } K \text{ is Volterra}.
\end{cases} \)

These formulas provide a solid mathematical basis for understanding and applying complex Gaussian stochastic volatility models.