Let $X_t$ be a compound Poisson process, i.e.

$$X_t= \sum_{i=0}^{N(t)} Y_i,$$ where $N(t)$ is a Poisson process and $Y_i$ are iid.

I have seen how to calculate the expected value $E(X_t)$ which depends only on the expected value of the $Y$'s and the parameter of the Poisson process.

I am however interested in the time expectation/ time distribution of this process for a given threshold. Let $x>0$ and define the stopping time $T_x= \inf \{t| X_t \geq x\}$.

My question is: How is $T_x$ distributed and what is the expectation of $T_x$. For my application I can assume that I know the distribution function of $Y_i$. Furthermore, we can assume that $Y_i>0$ a.s.

EDIT: I searched a bit further and found that this problem is called the determination of the "first passage time". I found a few papers dealing with this situation, they all focussed on the determination of the distribution of $T_x$. This however seems to be rather complicated (one could relate the question above to the Cramer-Lundberg model).

As I will be working with empirical data I cannot make model assumptions on the distribution of $Y_i$. As the distribution of $T_x$ then seems to much to ask for I would settle with how to compute the expectation.


The basic idea in this setting is to condition on the (exponentially distributed) time $\tau$ of the first event in the Poisson process $N$ and on the value $Y_1$.

  • If $Y_1\geqslant x$ then $T_x=\tau$.
  • Otherwise, $Y_1=y$ for some $y\lt x$ and the compound Poisson process after time $\tau$ behaves like the original one shifted by level $y$ and time $\tau$, thus, $T_x=\tau+S_{x-y}$ where each $S_{x-y}$ is independent of $\tau$ and distributed like $T_{x-y}$.

In terms of Laplace transforms, this shows that, for every positive $u$, $$ E[\mathrm e^{-uT_x}]=E[\mathrm e^{-u\tau}]\,\left(P[Y_1\geqslant x]+\int_0^xE[\mathrm e^{-uT_{x-y}}]\mathrm dP_Y(y)\right). $$ This identity characterizes the distribution of every $T_x$. For example, considering the derivatives at $u=0$ (or going back to the decomposition which led to this identity), one sees that the function $\theta:x\mapsto E[T_x]$ solves the equation $$ \theta(x)=E[\tau]+\int_0^x\theta(x-y)\mathrm dP_Y(y). $$ An equivalent formulation is that $$ \theta(x)=E[\tau]\,\left(1+P[Y_1\leqslant x]+P[Y_1+Y_2\leqslant x]+P[Y_1+Y_2+Y_3\leqslant x]+\cdots\right). $$


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.