1
$\begingroup$

Let $(X_t\colon \Omega\rightarrow E\times F)_{t\in T}$ be a Markov process, with $E,F$ topological spaces (or metric spaces, or whatever makes sense). I am wondering whether the induced process $(Y_t)_{t\in T}$ with $Y_t=\mathrm{pr_E}\circ X_t$ will still be a Markov process. I tried to prove this directly using the Markov property, but failed to do so. More specifically, if $\mathcal{G}_s=\sigma(Y_t~|~t\leq s)$ and $\mathcal{F}_s=\sigma(X_t~|~t\leq s)$ and $f$ is bounded and measurable on $E$:

\begin{align*} \mathbb{E}[f(Y_t)~|~\mathcal{G}_s]&=\mathbb{E}[\mathbb{E}[f(Y_t)~|~\mathcal{G}_s]~|~\mathcal{F_s}] \\ &=\mathbb{E}[\mathbb{E}[f(Y_t)~|~\mathcal{F}_s]~|~\mathcal{G}_s]\\ &=\mathbb{E}[\mathbb{E}[f(Y_t)~|~\sigma(X_s)]~|~\mathcal{G}_s]\\ &=\mathbb{E}[f(Y_t)~|~\sigma(Y_t)] \end{align*}

but the last equality does not seem to hold (this is related to this question). Am I trying to prove something that is not true at all?

$\endgroup$
2
$\begingroup$

A projection of a Markov process is not Markov in general. For example, a simple random walk on non-negative half-axes in $\mathbb Z^2$ is a Markov process, but its projection to any coordinate is not.

$\endgroup$

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.