David,
Ok, I can agree that the limit:
W_{t+dt} - W_t
exists, but the implication of writing:
dW_t = W_{t+dt} - W_t
could lead to misunderstanding. It would be rather tempting to divide both sides by dt, and write:
dW_t / dt = (W_{t+dt} - W_t)/dt
and what's wrong with this? The limit on the right hand side does not exist. While you are correct that W_{t+dt} - W_t --> 0 as dt -->0, once you divide by dt it will not converge.
This is because W_t has unbounded variation.
You can think of this intuitively by considering the fractal property of Brownian motion (it's fractual index is 2.0). You fix a point at t and plot another point at t+dt and draw a chord between them.
As dt decreases, the gradient of the chord changes violently, since each time you zoom in on the Brownian motion, more detail appears. The gradient never settles down and you will find that:
1/dt(W_{t_dt}-W_t) follows a Normal distribution, with mean 0 and variance 1/dt.
However, W_t does have bounded quadratic variation, which means we can define an integral operator, the "Ito integral".
Now, coming back to the original assertion:
dW_t = W_{t+dt} - W_t
This simply makes no sense once you remember that dW_t is an integral.
You can use the Riemann representation of the Ito integral to show this is questionable.
Set up t_i such that t_0 = 0, t_n = t and t_i - t_{i-1} = 1/n.
Then consider:
The right hand side is the Riemann representation of dW_t. Correct me if I am wrong, but I find it pretty difficult to believe that these will converge to the same value...
Now, I agree actuarial science is a branch of applied mathematics. But so is physics and so is quantative finance - and neither of these users of stochastic calculus would tolerate this.
If you posted the assertion:
dW_t = W_{t+dt} - W_t
to http://www.wilmott.com/index.cfm?NoCookies=Yes&forumid=1
(which is a quantative finance forum) you would instantly be told that this is nonsense.
If actuaries wish to gain the respect of the financial community then they need to show a proper understanding of stochastic calculus. This is well below the required standard in my opinion.
Last edited by a moderator: Feb 15, 2006