Scribd
Upload
48 views3 pages

The Ornstein-Uhlenbeck Process: Parametrization

The Ornstein-Uhlenbeck process is defined by a stochastic differential equation that describes the evolution of a continuous-time process. It has a Markov property whereby the conditional distribution of the process at time ti only depends on its value at the previous time ti-1. The precision and rate parameters of the process are assigned log-normal priors and the model is specified for use in INLA by passing arguments to the f() function.

Uploaded by

Ariol Zere
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
48 views3 pages

The Ornstein-Uhlenbeck Process: Parametrization

The Ornstein-Uhlenbeck process is defined by a stochastic differential equation that describes the evolution of a continuous-time process. It has a Markov property whereby the conditional distribution of the process at time ti only depends on its value at the previous time ti-1. The precision and rate parameters of the process are assigned log-normal priors and the model is specified for use in INLA by passing arguments to the f() function.

Uploaded by

Ariol Zere
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

The Ornstein-Uhlenbeck process

Parametrization
The Ornstein-Uhlenbeck process is defined with (mean zero), as the SDE

dxt = −φxt + σdWt

where φ > 0 and Wt is the Wiener process. This is the continuous time analogue to the discrete time
AR(1) model.
The process has a Markov property. Let x = (x1 , x2 , . . . , xn ) be value of the process at increasing
time-points t = (t1 , t2 , . . . , tn ), then the conditional distribution

xi | x1 , . . . , xi−1 , i = 2, . . . , n,

is Gaussian with mean


xi−1 exp(−φδi )
and precision
τ (1 − exp(−2φδi ))−1
where
δi = ti − ti−1 , i = 2, . . . , n
and
τ = 2φ/σ 2 .
The marginal distribution for x1 is taken to be the stationary distribution, which is a zero mean
Gaussian with precision τ .

Hyperparameters
The precision parameter τ is represented as

θ1 = log(τ )

where τ is the marginal precision for the Ornstein-Uhlenbeck process given above.
The parameter φ is represented as
θ2 = log(φ)
and the prior is defined on θ = (θ1 , θ2 ).

Specification
The Ornstein-Uhlenbeck model is specified inside the f() function as

f(<whatever>, model="ou", values=<values>, hyper = <hyper>)

The optional argument values gives the time-points where the process is defined/observed on
(default is unique(sort(<whatever>))).

1
Hyperparameter specification and default values
doc The Ornstein-Uhlenbeck process

hyper

theta1
hyperid 16001
name log precision
short.name prec
prior loggamma
param 1 5e-05
initial 4
fixed FALSE
to.theta function(x) log(x)
from.theta function(x) exp(x)
theta2
hyperid 16002
name log phi
short.name phi
prior normal
param 0 0.2
initial -1
fixed FALSE
to.theta function(x) log(x)
from.theta function(x) exp(x)

constr FALSE

nrow.ncol FALSE

augmented FALSE

aug.factor 1

aug.constr

n.div.by

n.required FALSE

set.default.values FALSE

pdf ou

Example
## simulate an OU-process and estimate its parameters back.
phi = -log(0.95)
sigma = 1
marg.prec = 2*phi/sigma^2
n = 1000
locations = cumsum(sample(c(1, 2, 5, 20),n, replace=TRUE))

2
## do it sequentially and slow (for clarity)
x = numeric(n)
x[1] = rnorm(1, mean=0, sd = sqrt(1/marg.prec))
for(i in 2:n) {
delta = locations[i] - locations[i-1]
x[i] = x[i-1] * exp(-phi * delta) +
rnorm(1, mean=0, sd = sqrt(1/marg.prec * (1-exp(-2*phi*delta))))
}

## observe it with a little noise


y = 1 + x + rnorm(n, sd= 0.01)
plot(locations, x, type="l")

formula = y ~ 1 + f(locations, model="ou", values=locations)


r = inla(formula, data = data.frame(y, locations))
summary(r)

Notes
The Ornstein-Uhlenbeck process is the continuous-time analogue to the discrete AR(1) model (for
positive lag-one correlation only), but they are parameterised slightly different.

You might also like