2
$\begingroup$

Let $W(t)$ be a standard Brownian motion and let $I(t) = \int_0^t W(s) \, \mathrm{d}s$ denote its time integral. I'm interested in the joint distribution of $(W(t), \,I(t))$, but I can't seem to find any references for this. Is it unknown? I can't seem to find even a covariance calculation.

$\endgroup$
  • $\begingroup$ For each fixed $t$, the random vector $(W(t),I(t))$ is jointly normal with mean zero and the covariance matrix $$\begin{pmatrix} t & t^2/2 \\ t^2/2 & t^3/3 \end{pmatrix}. $$ Is this the kind of information that you are looking for? $\endgroup$ – Sangchul Lee Mar 8 '17 at 0:19
  • 1
    $\begingroup$ @SangchulLee Why jointly normal? $\endgroup$ – bcf Mar 8 '17 at 0:45
  • 1
    $\begingroup$ Because each $(W(t),I(t))$ is a linear transform of the gaussian family $(W(s))_{0\leqslant s\leqslant t}$. $\endgroup$ – Did Mar 9 '17 at 9:06
  • 1
    $\begingroup$ Re references, the keywords integrated Brownian motion should provide a wealth of information about the process $(W,I)$. $\endgroup$ – Did Mar 9 '17 at 9:08
5
$\begingroup$

Since the Brownian motion $(W_t)_{t \geq 0}$ is a Gaussian process, the random vector $(W_{t_1},\ldots,W_{t_n})$ is Gaussian for any $t_1,\ldots,t_n \geq 0$, $n \in \mathbb{N}$. This implies that the vector

$$X_n := \bigg( W_t, \sum_{j=1}^n W_{t_j} (t_j-t_{j-1}) \bigg)$$

is Gaussian for each $n \in \mathbb{N}$, $t>0$ where $t_j:= \frac{j}{n} t$. As $X_n$ converges pointwise to $(W(t),\int_0^t W(s) \, ds)$, we find that the latter is Gaussian as a pointwise limit of Gaussian random variables. Since Gaussian random vectors are uniquely determined by their mean vector and covariance matrix, it just remains to calculate $\mathbb{E}(W_t^2)$, $\mathbb{E}(I_t W_t)$, $\mathbb{E}(I_t^2)$ and to note that the mean vector equals $0$.

$\endgroup$
  • $\begingroup$ I know this is an old question. But I wonder how we can compute $\mathbb{E}[I_t W_t]$ ? $\endgroup$ – MathMen Oct 13 '18 at 11:03
  • 2
    $\begingroup$ @MathMen Apply Fubini's theorem: $$\mathbb{E}(I_t W_t) = \mathbb{E} \left( \int_0^t W_s W_t \, ds \right) = \int_0^t \mathbb{E}(W_s W_t) \, ds.$$ $\endgroup$ – saz Oct 13 '18 at 18:48
  • $\begingroup$ @saz How to deduce that the vector $X_n$ is Gaussian for each $n\in\mathbb{N}, t>0?$ $\endgroup$ – Idonknow Feb 14 at 15:08
  • $\begingroup$ @Idonknow Since Brownian motion is a Gaussian process, the vector $Y:=(W_{t_1},\ldots,W_{t_n})$ is Gaussian. Linear combinations of Gaussian random vectors are Gaussian, i.e. $A \cdot Y + b$ is Gaussian for any matrix $A$ and for any vector $b$. Pick suitable $A$ and $b$ which allows you to write $$X_n = AY+b$. $\endgroup$ – saz Feb 14 at 18:05

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.