SDE¶
Miscellaneous¶
-
SDE.
syl
(a, b, c)¶ Solves the Sylvester equation
AX + XB = C
, whereC
is symmetric andA
and-B
have no common eigenvalues using (inefficient) algebraic approach via the Kronecker product, see http://en.wikipedia.org/wiki/Sylvester_equation
Stochastic Processes¶
-
SDE.
mu
(t, x, T, P)¶ Expectation \(E_(t,x)(X_{T})\)
-
SDE.
K
(t, T, P)¶ Covariance matrix \(Cov(X_{T}-x_t)\)
-
SDE.
H
(t, T, P)¶ Negative Hessian of \(\log p(t,x; T, v)\) as a function of
x
.
-
SDE.
r
(t, x, T, v, P)¶ Returns \(r(t,x) = \operatorname{grad}_x \log p(t,x; T, v)\) where
p
is the transition density of the processP
.
-
SDE.
bstar
(t, x, T, v, P::MvPro)¶ Returns the drift function of a vector linear process bridge which end at time T in point v.
-
SDE.
bcirc
(t, x, T, v, Pt::Union(MvLinPro, MvAffPro), P::MvPro)¶ Drift for guided proposal derived from a vector linear process bridge which end at time T in point v.
-
SDE.
lp
(t, x, T, y, P)¶ Returns \(log p(t,x; T, y)\), the log transition density of the process
P
-
SDE.
samplep
(t, x, T, P)¶ Samples from the transition density of the process
P
.
-
SDE.
exact
(u, tt, P)¶ Simulate process
P
starting in u on a discrete grid tt from its transition probability.
-
SDE.
ll
(X, P)¶ Compute log likelihood evaluated in B, beta and Lyapunov matrix lambda for a observed linear process on a discrete grid dt from its transition density.
-
SDE.
lp
(s, x, t, y, P) Returns \(log p(t,x; T, y)\), the log transition density
-
SDE.
euler
(u, W::CTPath, P::CTPro)¶ Multivariate euler scheme for
U
, starting inu
using the same time grid as the underlying Wiener processW
.
-
SDE.
llikeliXcirc
(t, T, Xcirc, b, a, B, beta, lambda)¶ Loglikelihood (log weights) of Xcirc with respect to Xstar.
t, T – timespan Xcirc – bridge proposal (drift Bcirc and diffusion coefficient sigma) b, sigma – diffusion coefficient sigma target B, beta – drift b(x) = Bx + beta of Xtilde lambda – solution of the lyapunov equation for Xtilde
-
SDE.
tofs
(s, tmin, T)¶ -
SDE.
soft
(t, tmin, T)¶ Time change mapping s in [0, T=t_2 - t_1] (U-time) to t in [t_1, t_2] (X-time), and inverse.
-
SDE.
Vs
(s, T, v, B, beta)¶ -
SDE.
dotVs
(s, T, v, B, beta)¶ Time changed V and time changed time derivative of V for generation of U
-
SDE.
XofU
(UU, tmin, T, v, P)¶ U is the scaled and time changed process
U(s)= exp(s/2.)*(v(s) - X(tofs(s)))XofU transforms entire process U sampled at time points ss to X at tt.
-
SDE.
stable
(Y, d, ep)¶ Return real stable d-dim matrix with real eigenvalues smaller than -ep parametrized with a vector of length d*d,
For maximum likelihood estimation we need to search the maximum over all stable matrices. These are matrices with eigenvalues with strictly negative real parts. We obtain a dxd stable matrix as difference of a antisymmetric matrix and a positive definite matrix.