Brownian Motion

Integrated Brownian Motion

Let \(W_t\) be a standard Wiener process and define \[ X_t=\int_0^t W_\tau\,d\tau. \] We want the distribution of \(X_t\) and whether \((X_t)_{t\ge0}\) is a martingale.

Interactive sample path

The blue curve is one Brownian path \(W_\tau\). The signed area under it up to \(t\) is \(X_t\).

2.00

Distribution

\(X_t\) is Gaussian because it is a linear functional of a Gaussian process.

Mean
0
Variance
2.667
\[ X_t\sim \mathcal N\!\left(0,\frac{t^3}{3}\right) \]
Interview shortcut: linear combination/integral of Gaussian random variables ⇒ Gaussian. Then compute only mean and variance.
Step 1

Mean

\[ \begin{aligned} \mathbb E[X_t] &=\mathbb E\!\left[\int_0^t W_\tau\,d\tau\right] \\ &=\int_0^t \mathbb E[W_\tau]\,d\tau \\ &=0. \end{aligned} \]
Step 2

Variance

Because the mean is zero,

\[ \begin{aligned} \operatorname{Var}(X_t) &=\mathbb E[X_t^2] \\ &=\int_0^t\int_0^t \mathbb E[W_sW_u]\,ds\,du. \end{aligned} \]

For Brownian motion,

\[ \mathbb E[W_sW_u]=\operatorname{Cov}(W_s,W_u)=\min(s,u). \]
\(\min(s,u)=s\) \(\min(s,u)=u\) \(s\) \(u\)
Split the square along \(s=u\). By symmetry, \[ \operatorname{Var}(X_t) =2\int_0^t\int_0^u s\,ds\,du. \]
Inner integral: \[ \int_0^u s\,ds=\frac{u^2}{2}. \]
Therefore, \[ 2\int_0^t\frac{u^2}{2}\,du =\int_0^t u^2\,du =\frac{t^3}{3}. \]
Hence \[ \boxed{X_t\sim\mathcal N\!\left(0,\frac{t^3}{3}\right)}. \]
Step 3

Is \(X_t\) a martingale?

For \(s

\[ X_t =\int_0^s W_\tau\,d\tau+\int_s^t W_\tau\,d\tau =X_s+\int_s^t W_\tau\,d\tau. \]
known at time \(s\)

For \(\tau>s\), write \[ W_\tau=W_s+(W_\tau-W_s). \] The increment \(W_\tau-W_s\) is independent of \(\mathcal F_s\) and has mean zero, so

\[ \mathbb E[W_\tau\mid\mathcal F_s]=W_s. \]

Therefore,

\[ \begin{aligned} \mathbb E[X_t\mid\mathcal F_s] &=X_s+\int_s^t \mathbb E[W_\tau\mid\mathcal F_s]\,d\tau \\ &=X_s+\int_s^t W_s\,d\tau \\ &=X_s+(t-s)W_s. \end{aligned} \]
Since \[ X_s+(t-s)W_s\neq X_s \] in general, \(X_t\) does not satisfy the martingale property.

Final result

\[ \boxed{X_t=\int_0^t W_\tau\,d\tau \sim \mathcal N\!\left(0,\frac{t^3}{3}\right)} \]
\[ \boxed{(X_t)_{t\ge0}\text{ is not a martingale.}} \]

The key reason is that future Brownian levels have conditional mean \(W_s\), not \(0\).