Let $S_n = X_1 + \dots + X_n$, where the increments $X_i$ are not i.i.d. Let $\tau$ be some random time (not necessarily a stopping time) and assume $$(\tau, X_1, \dots, X_\tau)\quad \text{ and } \quad (X_{\tau + 1}, X_{\tau + 2}, \dots) \qquad (1)$$ to be independent. I want to show with rigor that the process $(S_{\tau + n}- S_\tau)_{n \geq 1}$ is a Markov chain (satisfies the Markov property). For convenience, let us write $\tilde{S}_n = S_{\tau + n} - S_\tau$. As $\tilde{S}_n$ is independent of $\tau$ and $S_\tau$ it makes absolutely sense that the process is Markov but I would like to see the argument in more detail.

Edit: Possibly, the assumption (1) is not enough. Let us in addition to (1) assume that $$ (X_1, X_2, \dots) \overset{d}{=} (X_{\tau + 1}, X_{\tau +2}, \dots).$$ Maybe this is now enough to prove the claim that $\tilde{S}_n$ is a Markov chain.

Edit2: I am starting to think that the assumptions are not necessary. Simply by the structure of $S$ and in view of $\tilde{S}_n = X_{\tau + 1 } + \dots + X_{\tau + n}$ we have should have that $\tilde{S}_n$ conditioned on the values of $\tilde{S}_1, \dots , \tilde{S}_{n-1}$ only depends on $\tilde{S}_{n-1}$. Is this true / why is it not that easy?

  • $\begingroup$ Is the independence well defined? $X_{\tau+k}$ is a well defined random variable for $k\geq 0$ since it is just $X_{\tau+k}(\omega) = \sum_{i=0}^\infty X_{i+k}(\omega) \mathcal{X}_{\tau = i}(\omega) $. But what is hidden on the dots of $X_1, \cdots, X_\tau$? What is $X_{\tau-1}$ and how many variables are there? $\endgroup$ Aug 10, 2020 at 6:37
  • $\begingroup$ @LucasResende Maybe first of all, how to read assumption (1): The random variables up to the random time $\tau$, that is $X_1, \dots, X_\tau$, are independent from the random variables after the random time $\tau$, that is $X_{\tau+1}, X_{\tau +2}, \dots$. If I understand your question correctly: Hidden in the dots are the random variables $X_{\tau - i}$, where it is difficult to write them out precisely, because it is only well-defined for $i \leq \tau$ and $\tau$ is as well random. So, the number of variables that are in $(X_1, \dots, X_\tau)$ is $\tau$. I hope this helps. $\endgroup$
    – MMM
    Aug 10, 2020 at 14:29
  • $\begingroup$ But $\tau$ is a random variable. You can't say that $(X_1, \cdots, X_k)$ is independent of $(X_{k+1}, X_{k+2}, \cdots)$ where $\tau=k$. $\endgroup$ Aug 10, 2020 at 14:58
  • $\begingroup$ @LucasResende That is true. But I am afraid that I don't really understand the question. To me the independence stated in (1) makes sense. $\endgroup$
    – MMM
    Aug 10, 2020 at 16:31
  • $\begingroup$ My point is that your independence seems to be dependent on the value of $\tau$. You can't say that $(X_1, \cdots, X_k)$ is independent of $(X_{k+1}, \cdots)$ when $\tau=k$ and $(X_1, \cdots, X_j)$ is independent of $(X_{j+1}, \cdots)$ when $\tau=j$. You need to properly define what you mean by $X_{1}, \cdots, X_{\tau}$, because you can't change which variables are independent based on $\tau$. $\endgroup$ Aug 10, 2020 at 16:36

1 Answer 1


I will try to answer the question myself: (I would be grateful for feedback)

By assumption (1), $S_{\tau}$ and $\tau$ are independent from $S_{\tau+ k } - S_\tau$. Thus,

\begin{align} & P(S_{\tau + n } - S_\tau \in dx \vert S_{\tau + 1} - S_\tau \in dx_1, \dots , S_{\tau + n-1} - S_\tau \in dx_{n-1}) \\ & \overset{(1)}{=} P(S_{\tau + n } - S_\tau \in dx \vert S_{\tau + 1} - S_\tau \in dx_1, \dots , S_{\tau + n-1} - S_\tau \in dx_{n-1}, S_\tau = dy, \tau = m) \\ & = P(S_{m + n } \in dx+dy \vert S_{m + 1} \in dx_1 + dy, \dots , S_{m + n-1} \in dx_{n-1} + dy) \\ & = P(S_{m + n } \in dx+dy \vert S_{m + n-1} \in dx_{n-1} + dy) \\ & = P(S_{\tau + n } - S_\tau \in dx \vert S_{\tau + n-1} - S_\tau \in dx_{n-1}, S_\tau = dy, \tau = m) \\ & \overset{(1)}{=} P(S_{\tau + n } - S_\tau \in dx \vert S_{\tau + n-1} - S_\tau \in dx_{n-1}), \end{align} where I just used the definition of conditional probabilities when I wrote (1) over the equal sign. What do you think?

  • $\begingroup$ I want to ask what does $\overset{d}{=}$ mean in your equation? $\endgroup$ Aug 16, 2020 at 18:21
  • 1
    $\begingroup$ @GENIVI-LEARNER It means equal in distribution. So, $X \overset{d}{=} Y$, if $X$ and $Y$ follow the same distribution. $\endgroup$
    – MMM
    Aug 18, 2020 at 20:58

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.