I have seen in many books and sites that to find the mean of a process $X_t$ defined by

$dX_t = \mu X_t dt + \sigma X_t dW_t$

where $\mu$ and $\sigma$ are constants and $W_t$ is a standard Wiener process (or Brownian motion) you can simply integrate both sides

$X_t = X_0 + \int_{0}^{t} \mu X_s ds + \int_{0}^{t} X_s dW_s$

and then apply the expected value operator $\mathbb{E}[\cdot]$ to both sides to get

$\mathbb{E}[X_t] = X_0 + \int_{0}^{t} \mu \mathbb{E}[X_s]ds + \mathbb{E}\left[\int_{0}^{t} \sigma X_s dW_s\right]$

And then simplify by setting $\mathbb{E}\left[\int_{0}^{t} \sigma X_s dW_s \right] = 0$, thus turning the SDE into an ODE in terms of the expectation of $X_t$. However, how do we know that $\mathbb{E}\left[\int_{0}^{t} \sigma X_s dW_s \right] = 0$ is true, and is this valid for an arbitrary term $\sigma(X_t,t) dW_t$? For instance, what would be the expected value for the SDE

$dX_t = \mu X_t dt + \sigma X_t^2 dW_t$?

I appreciate any help on this topic.


The expectation of the stochastic integral vanishes because the stochastic integral is a martingale. Recall the following well-known result:

Let $f=f(s,\omega)$ be a progressively measurable function. If $$\mathbb{E} \left( \int_0^T f(s)^2 \, ds \right)<\infty \tag{1}$$ for any $T>0$, then $$M_t := \int_0^t f(s) \, dW_s$$ is a martingale.

Martingales have constant expectation, and therefore we obtain in particular

$$\mathbb{E} \left( \int_0^t f(s) \, dW_s \right) \stackrel{\text{def}}{=} \mathbb{E}(M_t) = \mathbb{E}(M_0)=0.$$

Since the SDE

$$dX_t = b(X_t) \, dt + \sigma(X_t) \, dW_t$$

has a unique square-integrable (in the sense of $(1)$) solution for any Lipschitz continuous coefficients $b$, $\sigma$, we find by applying the above result that

$$\mathbb{E}(X_t) = \mathbb{E}(X_0) + \mathbb{E} \left( \int_0^t b(X_s) \, ds \right).$$


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.