2
$\begingroup$

So I have the next problem:

In order to simulate the ruin of a risk process I need, of course, to simulate the risk process itself but in this case this process has some characteristics that make it particularly hard for me.

The risk process (Compound Poisson Process) is as follows:

\begin{equation} S_t=\sum_{i=0}^{N_t}X_i \end{equation}

Where $\{N_t:t \geq 0\}$ is a Poisson Process with intensity of $\lambda$ and $X_i \sim LogNormal(\mu,\sigma^2), \forall i$ where $X_i$ are independent of $N_t$.


First Idea

So, my first idea was to try to simulate this by using the Inverse Transform Technique: I would try to find the distribution of $S_t$, find its inverse function, and simulate it using values fromm a $U(0,1)$.

Following the above mentioned path I would have this:

\begin{equation} P\{S_t=k\}=\sum_{n=k}^{\infty}P\{S_t=k | N_t=n\}P\{N_t=n\} \end{equation}

And this is where the issues start to appear because $P\{S_t=k | N_t=n\}$ implicates knowing the distribution of the sum of $n$ independent $LogNormal(\mu,\sigma^2)$ random varibles and, as mentioned by Glen_b in https://stats.stackexchange.com/q/238566 and by the next answer (in descendant order of upvotes), the resulting distribution is a quite hard one to work with and even harder considering that I would still need to find the product of it with my poisson density and -possibly- reduct it to an expression from where I can obtain an inverse function.


Not that I don't find it interesting to follow the above mentioned path but I would rather either knowing a better way to simulate this process or something that helps me make the path that I suggested easier.

Please, help would be appreciated.


Second Idea

A new idea came to my mind, I could try to:

1.- Simulate a set $\{N_i\}_i$ of values from the underlying Poisson Process using that $N_t \sim Poisson(\lambda t)$ (based on the independence of it and the amounts $X_i$),

2.- Simulate $N_i$ values (for each $N_i$ obtained) from a $LogNormal(\mu,\sigma^2)$ random variable (based on the independence among the $X_i$),

3.- Sum the $N_i$ values obtained in 2 for each $N_i$ obtained in 1.

In the end, I would have $|\{N_i\}_i|$ (cardinality) simulations from the Compound Poisson Process.

All of this actually using the Inverse Transform Technique.

Am I right about this?


$\endgroup$
1
$\begingroup$

The second idea is the right one. I've already used it.

$\endgroup$

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.