Ah, I see now!
You're suggesting that the Wiener process Wt has a standard normal distribution
N(0,1) for all t, i.e., the process is stationary with respect to its distribution. This is a very important
distinction, so let me clarify the situation in detail.
Wiener Process: Standard Normal N(0,1) vs N(0,t)
1. Wiener Process Definition: The standard Wiener process Wt is typically defined such
that:
• W0=0 (it starts at 0),
• The increments Wt2−Wt1 are independent for any t1<t2,
• Wt∼N(0,t), meaning that the value of the process at time t is normally distributed
with mean 0 and variance t,
• The process has continuous paths.
So, the distribution of Wt is not N(0,1) for all t, but rather N(0,t), where the variance
increases with time. The variance grows linearly with time, which means the spread of the
process becomes larger as time progresses.
2. What Happens if Wt Were N(0,1) for All t? If you were to assume that Wt is always
distributed as N(0,1), then the variance would not increase over time. This would
fundamentally change the nature of the process and break the idea of the Wiener process as
typically defined.
• For example, with a process where Wt∼N(0,1) for all t, each increment Wt2−Wt1
would have a constant variance of 1, and the process would have a kind of
stationarity in terms of its distribution. However, the path of the process would no
longer exhibit the behavior of typical Brownian motion, because Brownian motion is
expected to have increasing variance over time.
3. Implication on Stationarity: The properties you're mentioning, such as:
• (i) Independence of Increments: The increments of Wt are independent,
• (ii) Stationarity: The joint distribution of {Wt1+t,…,Wtk+t} does not depend on t,
• (iii) Zero Mean: E[Wt]=0 for all t,
would hold only if you assume a process where the variance is constant over time (i.e.,
N(0,1) for all t). But this would contradict the idea of a Wiener process in the classical sense,
where the variance increases with time.
4. Does This Mean There Is No "Reasonable" Stochastic Process Satisfying (i) and (ii)?
Yes, and this is the crux of the statement you're asking about. In traditional stochastic
process theory, a standard Wiener process Wt does not have a constant variance (i.e.,
N(0,t)), which means strict stationarity does not hold. A process where the variance of each
Wt is constant (i.e., N(0,1) for all t) would not exhibit the typical behavior expected from
Brownian motion, especially with respect to the growing spread over time.
Therefore, the statement "there does not exist any ‘reasonable’ stochastic process satisfying
(i) and (ii)" refers to the traditional Wiener process and its characteristics. You can have a
process with independent increments (like N(0,1) for all t), but it would not be a Wiener
process in the conventional sense.
Conclusion:
You're correct that a constant normal distribution N(0,1) could describe a stationary process in the
sense that the distribution doesn't change over time. However, this would not describe the Wiener
process, where the variance grows with time. The Wiener process traditionally has N(0,t), and this
growing variance prevents it from being stationary in the strict sense (despite having stationary
increments).
Thus, a process with constant variance (such as N(0,1)) would indeed be stationary in distribution,
but it wouldn't have the typical properties of a Wiener process
