Simulation of M independent trajectories of a mixed stochastic differential equation (SDE) with linear drift and two random effects \((\alpha_j, \beta_j)\) \(dX_j(t)= (\alpha_j- \beta_j X_(t))dt + \sigma a(X_j(t)) dW_j(t)\), for \(j=1, ..., M\).
mixedsde.sim(M, T, N = 100, model, random, fixed = 0, density.phi,
param, sigma, t0 = 0, X0 = 0.01, invariant = 0, delta = T/N,
op.plot = 0, add.plot = FALSE)number of trajectories
horizon of simulation.
number of simulation steps, default Tx100.
name of the SDE: 'OU' (Ornstein-Uhlenbeck) or 'CIR' (Cox-Ingersoll-Ross).
random effects in the drift: 1 if one additive random effect, 2 if one multiplicative random effect or c(1,2) if 2 random effects.
fixed effects in the drift: value of the fixed effect when there is only one random effect, 0 otherwise. If random =2, fixed can be 0 but \(\beta\) has to be a non negative random variable for the estimation.
name of the density of the random effects.
vector of parameters of the distribution of the two random effects.
diffusion parameter
time origin, default 0.
initial value of the process, default X0=0.
1 if the initial value is simulated from the invariant distribution, default 0.01 and X0 is fixed.
time step of the simulation (T/N).
1 if a plot of the trajectories is required, default 0.
1 for add trajectories to an existing plot
matrix (M x (N+1)) of the M trajectories.
vector (or matrix) of the M simulated random effects.
Simulation of M independent trajectories of the SDE (the Brownian motions \(Wj\) are independent), with linear drift. Two diffusions are implemented, with one or two random effects:
If random = 1, \(\beta\) is a fixed effect: \(dX_j(t)= (\alpha_j- \beta X_j(t))dt + \sigma dW_j(t) \)
If random = 2, \(\alpha\) is a fixed effect: \(dX_j(t)= (\alpha - \beta_j X_j(t))dt + \sigma dW_j(t) \)
If random = c(1,2), \(dX_j(t)= (\alpha_j- \beta_j X_j(t))dt + \sigma dW_j(t) \)
If random = 1, \(\beta\) is a fixed effect: \(dX_j(t)= (\alpha_j- \beta X_j(t))dt + \sigma \sqrt{X_j(t)} dW_j(t) \)
If random = 2, \(\alpha\) is a fixed effect: \(dX_j(t)= (\alpha - \beta_j X_j(t))dt + \sigma \sqrt{X_j(t)} dW_j(t) \)
If random = c(1,2), \(dX_j(t)= (\alpha_j- \beta_j X_j(t))dt + \sigma \sqrt{X_j(t)} dW_j(t) \)
The initial value of each trajectory can be simulated from the invariant distribution of the process: Normal distribution with mean \(\alpha/\beta\) and variance \(\sigma^2/(2 \beta)\) for the OU, a gamma distribution \(\Gamma(2\alpha/\sigma^2, \sigma^2/(2\beta))\) for the C-I-R model.
Several densities are implemented for the random effects, depending on the number of random effects.
If two random effects, choice between
'normalnormal': Normal distributions for both \(\alpha\) \(\beta\) and param=c(mean_\(\alpha\), sd_\(\alpha\), mean_\(\beta\), sd_\(\beta\))
'gammagamma': Gamma distributions for both \(\alpha\) \(\beta\) and param=c(shape_\(\alpha\), scale_\(\alpha\), shape_\(\beta\), scale_\(\beta\))
'gammainvgamma': Gamma for \(\alpha\), Inverse Gamma for \(\beta\) and param=c(shape_\(\alpha\), scale_\(\alpha\), shape_\(\beta\), scale_\(\beta\))
'normalgamma': Normal for \(\alpha\), Gamma for \(\beta\) and param=c(mean_\(\alpha\), sd_\(\alpha\), shape_\(\beta\), scale_\(\beta\))
'normalinvgamma': Normal for \(\alpha\), Inverse Gamma for \(\beta\) and param=c(mean_\(\alpha\), sd_\(\alpha\), shape_\(\beta\), scale_\(\beta\))
'gammagamma2': Gamma \(+2 * \sigma^2\) for \(\alpha\), Gamma \(+ 1\) for \(\beta\) and param=c(shape_\(\alpha\), scale_\(\alpha\), shape_\(\beta\), scale_\(\beta\))
'gammainvgamma2': Gamma \(+2 * \sigma^2\) for \(\alpha\), Inverse Gamma for \(\beta\) and param=c(shape_\(\alpha\), scale_\(\alpha\), shape_\(\beta\), scale_\(\beta\))
If only \(\alpha\) is random, choice between
'normal': Normal distribution with param=c(mean, sd)
lognormal': logNormal distribution with param=c(mean, sd)
'mixture.normal': mixture of normal distributions \(p N(\mu1,\sigma1^2) + (1-p)N(\mu2, \sigma2^2)\) with param=c(p, \(\mu1, \sigma1, \mu2, \sigma2\))
'gamma': Gamma distribution with param=c(shape, scale)
'mixture.gamma': mixture of Gamma distribution \(p \Gamma(shape1,scale1) + (1-p)\Gamma(shape2,scale2)\) with param=c(p, shape1, scale1, shape2, scale2)
'gamma2': Gamma distribution \(+2 * \sigma^2\) with param=c(shape, scale)
'mixed.gamma2': mixture of Gamma distribution \(p \Gamma(shape1,scale1) + (1-p) \Gamma(shape2,scale2)\) + \(+2 * \sigma^2\) with param=c(p, shape1, scale1, shape2, scale2)
If only \(\beta\) is random, choice between 'normal': Normal distribution with param=c(mean, sd)
'gamma': Gamma distribution with param=c(shape, scale)
'mixture.gamma': mixture of Gamma distribution \(p \Gamma(shape1,scale1) + (1-p) \Gamma(shape2,scale2)\) with param=c(p, shape1, scale1, shape2, scale2)
This function mixedsde.sim is based on the package sde, function sde.sim. See Simulation and Inference for stochastic differential equation, S.Iacus, Springer Series in Statistics 2008 Chapter 2
# NOT RUN {
#Simulation of 5 trajectories of the OU SDE with random =1, and a Gamma distribution.
simuOU <- mixedsde.sim(M=5, T=10,N=1000,model='OU', random=1,fixed=0.5,
density.phi='gamma', param=c(1.8, 0.8) , sigma=0.1,op.plot=1)
X <- simuOU$X ;
phi <- simuOU$phi
hist(phi)
# }
Run the code above in your browser using DataLab