I am trying to solve this problem where we're asked to compute the quadratic variation of a process. I assume that it is necessary to apply Ito's formula but not sure how to get the right solution. Furthermore, I'm also not sure about how to apply Ito's formula to a function that includes integrals like this one. I am familiar with the basic Ito Formula and know how to apply it to simpler functions.

Let N be a (P,F)-Poisson process with parameter $\lambda$ > 0 and define the process: X = $(X)_t$

$$ X_t = 2 + \int_{0}^{t} \sqrt{s} dW_s + N_t $$

Compute $[X]_t$

is this attempt correct?

$$ X_t = 2 + \int_{0}^{t} \sqrt{s} dW_s + N_t $$

which we can write in differential form as:

$$ dX_t = \sqrt{t}dW_t + dN_t $$

then the quadratic variation is given as:

$$ d[X_t] = dX_t * dX_t = (\sqrt{t}dW_t + dN_t) * (\sqrt{t}dW_t + dN_t) = tdt + dNt $$

I assumed that the cross terms cancel out. Then we can then rewrite it in integral form to get $[X_t]$

$$ [X_t] = \int_{0}^{t} t dt + \int_{0}^{t} dN_t = \frac{t^2}{2} + N_t $$

If this attempt is correct, I am not quite sure why $dN_t * dN_t = dN_t$ would be true in the quadratic covariation step.


1 Answer 1


Your attempt is correct.

The quadratic variation for a Poisson process is: $$[N]_t=\lim_{\sup(t_{i+1}-t_i)\rightarrow0}\sum_{i:t_i\leq t}(N_{t_{i+1}}-N_{t_i})^2\tag{1}$$ for some partition $\Pi(t)=0\leq t_0\leq\dots\leq t_n\leq t$ of the segment $[0,t]$. Simply pick the partition such that the jump times $\tau_1\leq\dots\leq\tau_k$ of $N$ between $0$ and $t$ are included in $\Pi(t)$ then given $N_{\tau_j}-N_{\tau_j^-}=\Delta N_{\tau_j}$: $$[N]_t=\sum_{j:\tau_j\leq t}\Delta N_{\tau_j}^2$$ But $\Delta N_{\tau_j}$ is always equal to 1, and so is $\Delta N_{\tau_j}^2$, therefore the quadratic variation of the Poisson process jumps by $1$ whenever $N$ jumps. Consequently: $$\textrm{d}[N]_t=\textrm{d} N_t$$

Addendum. To justify picking the partition $\Pi(t)$ such that it contains the jump times $\tau_1\leq\dots\leq\tau_k$, consider the definition $(1)$ again. For each $i$, either there is no $j\in[1,k]$ such that $\tau_j\in[t_i,t_{i+1})$ in which case $\smash{\lim_{\sup(t_{i+1}-t_i)\rightarrow0}(N_{t_{i+1}}-N_{t_i})^2=0}$; else assuming without loss of generality that there is a single jump in the interval $[t_i,t_{i+1})$: \begin{align} N_{t_{i+1}}-N_{t_i} &=(N_{t_{i+1}}-N_{\tau_j})+(N_{\tau_j}-N_{t_i})\\ &=(N_{\tau_j}-N_{\tau_j})+(N_{\tau_j}-N_{\tau_{j-1}})\\ &=\Delta N_{\tau_j} \end{align} from which it entails the original argument is valid $-$ the argument generalizes to an arbitrary number of jumps within $[t_i,t_{i+1})$ at the cost of more involved notation.

  • $\begingroup$ thank you so much, this clears things up $\endgroup$
    – Niko
    May 19, 2023 at 1:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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