Find the joint density of sum process $f_{S_k, S_{k+1}}(s_k, s_{k+1})$

47 Views Asked by At

Given two sum process:

$S_k = X_1 + \ldots + X_k$

$S_{k +1} = X_1 + \ldots + X_{k+1}$

where all $X_i$ are iid random variables

I am required to find $f_{S_k, S_{k+1}}(s_k, s_{k+1})$ e.g. their joint PDF


Attempt:

We know that through independent and stationary increment properties of the sum process, we have that the CDF $F_{S_k, S_{k+1}}$ is given by: \begin{align} F_{S_k, S_{k+1}}(s_k, s_{k+1}) & = \Pr[S_{k} \leq s_k, S_{k+1} \leq s_{k+1}]\\ &= \Pr[S_{k} \leq s_k, S_{k + 1} - S_{k} \leq s_{k+1} - s_k]\\ &= \Pr[S_{k} \leq s_k] \Pr[S_{k + 1} - S_{k} \leq s_{k+1} - s_k]\\ &= \Pr[S_{k} \leq s_k] \Pr[S_{1} \leq s_{k+1} - s_k] \\& = F_{S_k}(s_k)F_{S_1}(s_{k+1}-s_k)\end{align}

Then the PDF is going to be the double partial derivative wrt $s_k, s_{k+1}$

\begin{align} \dfrac{\partial^2 F_{S_k, S_{k+1}}(s_k, s_{k+1})}{\partial s_{k+1} \partial s_{k}} & \triangleq f_{S_k, S_{k+1}}(s_k, s_{k+1})\\ & = \dfrac{\partial ^2 F_{S_k}(s_k)F_{S_1}(s_{k+1}-s_k)}{\partial s_{k+1} \partial s_k}\\ & = \dfrac{\partial f_{S_k}(s_k)F_{S_1}(s_{k+1}-s_k) - F_{S_k}(s_k)f_{S_1}(s_{k+1}-s_k)}{\partial s_{k+1} } \\ & = f_{S_k}(s_k)f_{S_1}(s_{k+1}-s_k) - \dfrac{\partial F_{S_k}(s_k) f_{S_1}(s_{k+1}-s_k)}{\partial s_{k+1} }\end{align}

Where the last two steps are computed using the chain rule.

The desired expression is :

$f_{S_k, S_{k+1}}(s_k, s_{k+1}) = f_{S_k}(s_k)f_{S_1}(s_{k+1}-s_k)$

However, observe that I have a bizarre term. $\dfrac{\partial F_{S_k}(s_k) f_{S_1}(s_{k+1}-s_k)}{\partial s_{k+1} }$

(what is the partial derivative of a PDF again?)

Where did I go wrong and how can I fix my derivation?

1

There are 1 best solutions below

0
On BEST ANSWER

What you did wrong:

$$S_k\leq s_k, S_{k+1}\leq s_{k+1} \not\Rightarrow S_{k+1}-S_{k} \leq s_{k+1}-s_{k}$$

My solution:

$$f_{S_k,S_{k+1}}(a,b)=f_{S_k,X_{k+1}}(a,b-a)=f_{S_k}(a)f_{X_{k+1}}(b-a)$$

Last inequality follows from independence of $S_k$ and $X_{k+1}$. Finally, use iid of $X_i$’s and $S_1=X_1$.