R package for the exact simulation of non-negative shot noise processes (with Bernoulli, Poisson, Gamma and inverse-Gaussian distributed jumps) and non-Gaussian OU processes, also known as Lévy-driven OU processes (in particular, OU-Poisson, OU-Gamma and OU-inverse Gaussian (OU-IG) processes) from the paper
[1] M. Tamborrino, P. Lansky, Shot noise, weak convergence and diffusion approximation, Physica D, 418, 132845, 2021, https://www.sciencedirect.com/science/article/abs/pii/S0167278921000038
The code was written by Massimiliano Tamborrino (firstname dot secondname at warwick.ac.uk) in Rcpp. Yan Qu provided us with the Matlab codes for the exact simulation of the OU-Gamma and OU-IG processes based on the papers
[2] Y. Qu, A. Dassios, H. Zhao, Exact simulation of Gamma-driven Ornstein– Uhlenbeck processes with finite and infinite activity jumps, J. Oper. Res. Soc., 471--484, 2019, http://dx.doi.org/10.1080/01605682.2019.1657368.
[3] Y. Qu, A. Dassios, H. Zhao, Exact simulation of Ornstein–Uhlenbeck tempered stable processes, J. Appl. Probab., 2021 (in press). Preprint at http://eprints.lse.ac.uk/106267/
Those Matlab algorithms have been then rewritten in Rcpp and included in the package.
Here we provide the code for
- the exact simulation of N values from a non-negative shot noise process
$X$ at time$t$ , solution of the stochastic differential equation (2) from [1]
dX(t)= -alpha X(t) dt + J dN(t),
where N(t) is a Poisson process with rate lambda>0 and J is the distribution of the jump amplitude. In the package, we consider J to be a Bernoulli distribution (shotnoise_JnBer), Poisson (shotnoise_JnP), chi-square (shotnoise_JnP with parameter scenario equal to 1), Gamma (shotnoise_JnG, with scenario equal to 2), inverse-Gaussian ( shotnoise_JnIG). The chosen parameters and the generating algorithm is described in [1].
-
the exact simulation of a trajectory of the shot noise process with Gamma and IG jump amplitudes (shotnoise_JnG_trajectory and shotnoise_JnIG_trajectory, respectively).
-
the exact simulation of N values from an OU-Poisson process at time t, solution of (13) of [1]
dY(t)= -delta Y(t) dt + rho dZ(t), (a)
with delta=alpha and Z(t) being a Poisson process with intensity mu>0, see Section 3 of [1]. R code: OU_Poisson
-
the exact simulation of N values from an OU-Gamma process at time t, solution of (a) above (i.e. (13) of [1]) with delta=alpha and Z(t) being a gamma process with Lévy measure \nu(dx)=tilde.alpha x^{-1}e^{-\beta x}dx. See Section 3.2 and Table 1 for the chosen parameters, and [2] for the description of the simulation algorithm. R code: OU_Gamma
-
the exact simulation of N values from an OU-IG process at time t, solution of (a) above with delta=alpha and Z(t) being an Inverse Gaussian process with Lévy measure nu(dx)=e^{-c^2 x/2}dx, with c^2=mu/tilde.sigma^2. See Section 3.2 and Table 1 for the chosen parameters, and [3] for the description of the simulation algorithm. R code: OU_IG
-
the simulation of a trajectory of the OU-Gamma process (OU_Gamma_trajectory) and the OU-IG process (OU_IG_trajectory)
-
the computation of integrated absolute errors (IAEs) between the shot noise process and the corresponding Gaussian OU or non-Gaussian OU process.
The simplest way is to do it via devtools, using devtools::install_github("massimilianotamborrino/shotnoise")
Use code_Figure1.r to generate two trajectories of the OU-Gamma and OU-IG processes, reproducing Fig. 1 of [1].
Use code_Figure2.r to compare the estimated probability density functions of the shot noise process, the Gaussian OU and the non-Gaussian OU, reproducing Fig. 2 of [1].
All parameters are chosen according to Table 1 of [1].
Digit ?nameroutine to find out what the routine 'nameroutine' is doing, its input parameters and its output. For example, ?OU_Gamma_trajectory describe what the function 'OU_Gamma_trajectory' is doing, describing each needed input entry and the returned output.