Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $X_t$ be a compound Poisson process with positive drift $$ X_t = t + \sum_{i=1}^{N_t}Y_i, $$ where $N_t \sim$Poisson$(\lambda t)$ and the $Y_i$ are i.i.d. with pdf $f$.
With $M_t$ the process $X$ reflected from its maximum, $$ M_t = \bar X_t - X_t, $$ where $\bar X_t = \sup_{s\le t}X_s$, let $L_t$ be the local time at zero for $M_t$.
Let $T$ be the hitting time of $M_t$ for $c>0$, $$ T = \inf_{t\ge0}\{M_t \ge c\}. $$ Find the distribution of $L_T$.

Since $X_t$ is a compound Poisson process, then $$ L_t = \int_0^t 1_{\{M_s=0\}}ds. $$ Whence, by conditioning on $T$, $$ E\left( e^{i \theta L_T }\right) =E\left( E \left( \left. e^{i \theta \int_0^T 1_{\{M_s=0\}}ds }\right|T\right). \right) $$ I can't go further.
Is it the right way to approach this problem?
I know a few results and technics on Levy processes, but I don't see which one applies here.

share|cite|improve this question
I think you can solve this problem for a birth and death chain, i.e. occupation time at zero before some passage time, and maybe the answer is opaque enough to discourage you. I think this can be done a couple of ways, but one is using a brownian embedding and quoting the ray-knight theorem. – mike Feb 28 '13 at 21:46
maybe it's not so bad. I suppose $M_t$ is markov, and the occupation time at 0 up to the first etc. must be exponentially distributed. – mike Mar 1 '13 at 13:22
up vote 1 down vote accepted

Assume that $T=\inf\{t\geqslant0\mid M_t\geqslant c\}$ (I do not understand the formula in the post). The process $(M_t)_{t\geqslant0}$ is a Markov process on $[0,+\infty)$ starting from $M_0=0$ with the following transition probabilities: if $M_t=x$ with $x\geqslant0$, then $M_{t+\mathrm dt}=0$ with probability $1-\mathbb P(Y\lt x)\lambda\mathrm dt+o(\mathrm dt)$ and $M_{t+\mathrm dt}$ is distributed like $x-Y$ conditionally on $Y\lt x$ with probability $\mathbb P(Y\lt x)\lambda\mathrm dt+o(\mathrm dt)$.

This implies that, once at $0$, $M$ stays at $0$ for an exponential time with parameter $a=\lambda\mathbb P(Y\lt0)$, and, once it leaves $0$, $M$ reaches $c$ before coming back to $0$ with some probability $p$. Thus, $L_T$ is the time spent at $0$ before the first heads in a sequence of heads and tails with probability of heads $p$, each tails counting for an exponential time with parameter $a$. Thus, $L_T$ is exponential with parameter $ap$.

Conditioning on the size of the first jump, one sees that $$ p\,\mathbb P(Y\lt0)=\mathbb P(Y\lt-c)+\int_0^cp_x\,\mathbb P(-Y\in\mathrm dx), $$ where, for every $0\lt x\lt c$, $p_x$ is the probability that $M$ starting from $x$ hits $c$ before $0$.

From which one might be able to deduce the value of $p$...

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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