A normal inverse Gaussian model for a risky asset with dependence (Q654485): Difference between revisions

Let \(\{T_t\}\) be a non-decreasing stochastic process. We consider the model for the option price \(P_t\) at time \(t\) \[ \log P_t = \log P_0 + \mu t + \theta T_t + \sigma B(T_t)\;, \] where \(B\) is a standard Brownian motion independent of \(\{T_t\}\). In a time unit, the increment of the log-price becomes \[ X_t = \mu + \theta (T_t - T_{t-1}) + \sigma [B(T_t) - B(T_{t-1})]\;. \] The latter can be constructed as \[ \mu + \theta (T_t - T_{t-1}) + \sigma \sqrt{T_t - T_{t-1}} \xi_t\;, \] where \(\xi_t\) are independent standard normally distributed random variables. By choosing the process \(\{T_t\}\) appropriately, a given distribution for the increments \(X_t\) can be obtained. For example, if the inverse Gaussian distribution \(\text{IG}(\delta, \gamma)\) is used for \(T_t - T_{t-1}\), then \(X_t\) has a normal inverse Gaussian distribution. The process \(\{T_t\}\) can be constructed as a mean reverting diffusion \[ \text{d} T_t = - \lambda (T_t-\mu)\;\text{d}t + \sqrt{v(T_t)}\;\text{d} W_t\;, \] for appropriate parameters \(\lambda,\mu\), function \(v\) and a Brownian motion \(W\), independent of \(B\). A nice property of the inverse Gaussian distribution is that a sum of independent \(\text{IG}(\delta_1, \gamma)\) and \(\text{IG}(\delta_2, \gamma)\) variables yields an \(\text{IG}(\delta_1+\delta_2, \gamma)\) distributed variable. It is therefore possible to add \(n\) independent processes \(\{T_t^{(k)}\}\) with \(\text{IG}(\delta_k, \gamma)\) marginal distributions. Since \(T_t = \sum_{k=1}^n T_t^{(k)}\) has an inverse Gaussian distribution, the increments \(X_t\) are normally inverse Gaussian distributed. Through the choice of the processes, different types of dependence can be constructed. For a finite superposition of diffusions, we always have short-range dependence. If an infinite number of diffusions are added, also long-range dependence is possible. Finally, it is shown how options can be priced in the model constructed.