6
$\begingroup$

I was wondering when a stochastic process defined via a SDE is Markovian? The SDE may involved Ito integral, Lebesgue integral, jump component, and any other things. The reason I ask this question is that I don't understand when we can and cannot discuss properties of a Markov process, such as Kolmogorov equations, for a process defined by a SDE. Thanks and regards!

$\endgroup$
  • $\begingroup$ Hi, in Protter's book "Stochastic Integration and Differential Equations v 2.1", more precisely in Chapter V section 6 the matter is extensively treated in a semimartingale setting.Best regards $\endgroup$ – TheBridge Feb 15 '13 at 16:31
2
$\begingroup$

If SDEs are of the form $\mathrm dX_t = \sum_{i=1}^n f(X_t)\mathrm dY^i_t$ where $Y^i$ are Markov processes and $f$ are some functions, than unless $Y^i$ and $f$ are not "nice enough", $X$ will be a Markov process. Intuitively, this is the case since the future dynamics of $X_t$ is independent from its past: e.g. $$ \mathrm dX_t = X_t\mathrm dW_t $$ is a Markov process whereas $$ \mathrm dX_t = \left(\int_0^t X_s\mathrm ds\right)\mathrm dW_t $$ is not. I am not sure that there is a complete answer, which "nice" conditions $Y^i$ and $f$ have to verify in order for $X$ to be a (strong) Markov process - but I guess you are aware of many example of such conditions for jump diffusion.

$\endgroup$
  • $\begingroup$ Thanks! Actually, I am little aware of "example of such conditions for jump diffusion", although I hope to learn them. (I haven't figured out the definitions of diffusion processes and jump processes yet). $\endgroup$ – Tim Feb 15 '13 at 13:17
2
$\begingroup$

Consider the following SDE

$$ X_t ^{s,x} = x +\int _s^t \sigma(u,X_t ^{u,x}) ~ dW_u+\int _s^t b(u,X_t ^{u,x}) ~ du $$

satisfing the following hypothesis:

there is $C>0$ such that, for all $(x,y) \in \mathbb R ^p \times \mathbb R ^p$ and $u \in \mathbb R _+$

$$\left|\sigma(u,x) -\sigma(u,y) \right|+\left|b(u,x) -b(u,y) \right|\leq C\left|x-y \right|$$

for all $t\in \mathbb R _+$ and $x\in \mathbb R ^p$

$$ \int_s^t \left(\left| \sigma \right|^2(u,x)+\left|b \right|^2(u,x)\right)~du <+ \infty$$

If $b$ and $\sigma$ are time-homogeneous, ie, $\sigma(u,x) =\sigma(x)$ and $ b(u,x)= b(x)$, the simple Markov property applies to the solution of this SDE.

$\endgroup$
  • 1
    $\begingroup$ what is the reasoning ? $\endgroup$ – mastro Jul 29 '15 at 14:25
1
$\begingroup$

A good example of this is the time-integral of the Ornstein-Uhlenbeck process. On its own, this is not a Markov process. However, a two dimensional stochastic process, with one co-ordinate being the Ornstein-Uhlenbeck process and the other being its time-integral, will be a two dimensional Markov process. Similarly, a one-dimensional Markov process and its supremum process, comprise a two-dimensional Markov process.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.