IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 


What is the relation between the two?
A quick dose of Googling tells me something along the lines of "white noise is the derivative of a Wiener process". Could someone shed some light and maybe point to some better references?
Thanks Eric 




IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 


Follow up...
Consider
dS/S = mu*dt + sigma*dW
with
S(t) = S(0)*exp(mu'*t + sigma*W)
so that
log[S(t)/S(0)] = mu'*t + sigma*W.
Has anyone ever looked at modelling this as
log[S(t)/S(0)] = mu'*t + sigma*W',
where W' is Gaussian white noise? Just curious.
Here is kind of a neat thing:
g(t)*W(t) = int_0^t g'(tau)*W(tau) dtau + int_0^t g(tau)*W'(tau) dtau
where W(t) is a Wiener process and W'(t) is Gaussian white noise.
I got that from this paper. 



Anthis

It's all Greek to me

Total Posts: 1180 
Joined: Jul 2004 


What is the relation between the two?
Just a quick answer before i go to bed... In discrete time econometrics, the white noise is called the error term with constant variance. If the variance is time varying then we have a random walk, and a random walk is the discrete time version of a Wiener process.
HTH

Αίεν Υψικράτειν/Τύχη μη πίστευε/Άνδρα Αρχή Δείκνυσι/Νόησις Αρχή Επιστήμης //Σε ενα κλουβί γραφείο σαν αγρίμι παίζω ατέλειωτο βουβό ταξίμι




IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 

 

Johnny

Founding Member

Total Posts: 4331 
Joined: May 2004 


White light, white heat
White noise is the signal generated by combining sine waves of all frequencies in equal proportions. It is the same as Brownian motion. Consider the defining properties of Brownian motion and check that they also correspond to white noise:
1. Continuity: Wt is continuous in t (check, it's the sum of sine waves) and Wo = 0 (check, set all sine waves to "start" at t=0). 2. Normality: Wt is distributed N(0,t) 3. Normally distributed independent increments: Ws  Wt is distributed N(0, st) for s>t
So far, so good. But enquiring minds probably want to think about (a) differentiability and (b) stochasticity vs deterministicity.
I can never remember the exact definition of a Wiener process, but isn't it (something like) any process of the form dXt = f(Xt, t) dt + g(Xt, t) dWt where dWt is the increment of a Brownian motion.
I'm not sure this is what you're looking for. 
The sound of one bear, uh, in the woods 



Nonius

Founding Member Nonius Unbound

Total Posts: 11665 
Joined: Mar 2004 


Johnny's first comments are correct. But, it is not the same as Brownian Motion. It is the "derivative" of Brownian Motion. In this sense, Eric's intuition is correct. 
If you look hard you will see NONIUS in CMB data. Muhauaha. 


Johnny

Founding Member

Total Posts: 4331 
Joined: May 2004 


Nonius is correct and I was wrong. But let us not say that white noise is the derivative of Brownian motion but instead say that the integral of white noise is the same as Brownian motion. The integral of white noise is known (for it's connection with Brownian motion) as Brown noise. And many thanks to all those people that pointed this out to me in private. 
The sound of one bear, uh, in the woods 



Cheng


Total Posts: 2585 
Joined: Feb 2005 


Eric,
white noise may be viewed as the derivate of a Wiener process. Unfortunately white noise has infinite variance and the nifty techniques one usually uses don't work here. The derivative exists only in the sense of distributions, a kind of generalized derivatives. If white noise were an integrator in the LebesgueStieltjessense you could rewrite any stochastic Itô integral as
\int_a^b f(s) dW_s = \int_a^b f(s)W'(s) ds,
W'(s) being the white noise.
I can dig up my old scripts and books at home if you want to know more. Hope this helps a bit building an intuition.
Regards 
"Don't try to run, don't try to hide. Believe me, the hammer's gonna make it right !" 


Nonius

Founding Member Nonius Unbound

Total Posts: 11665 
Joined: Mar 2004 


It is also in the sense of Eric's NCG calculus that one can view white noise as the noncommutative derivative of brownian motion. 
If you look hard you will see NONIUS in CMB data. Muhauaha. 



Martingale

NP House Mouse

Total Posts: 2590 
Joined: Jun 2004 


it is still too early for me to think, but the white noise way of doing things are probably most done in engineering related stuff, but hey... you never know, here is something that might be of a little interest...
http://www.ma.ic.ac.uk/~pavl/ver3.pdf 



IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 


Nonius saw through my thinly veiled attempt to bring NCG into the picture
It is true, if someone could tell of any special algebraic relation for
dW' dW'
where W' is white noise, then I could reformulate white noise in terms of NCG, which might be kind of academically interesting.
The real point of this is, as I said, to try to understand the link (if any) between vector autoregression and multifactor stochastic differential equations. I'm just now learning the former and it is reminding me a lot of the latter.
Looking at VAR reminded me of my days back in grad EE and DSP, e.g. ztransforms, digital filters, impulse responses, etc., which made me think of deconvolutions, which made me think... (ad nauseum) 





this was precisely how the NYU math in finance "stochastic calculus" course taught by marco avellaneda last semester beganby pounding on the metaphor of white noise as a "derivative" of brownian motion. he's taken the homework files off the website, but i can send them to anyone interested in playing (numerically) with this stuff. it was very good for intuition.
disclaimer: i stopped attending this course after about three lectures, as avellaneda is a horrible lecturer. the psets are still decent, however. 



IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 


hammerbacher,
Sounds cool. I'd be interested in taking a look at that and any notes you might have.
Thanks 





in regards to vector autoregression, my understanding is this:
AR:
where that last epsilon is your white noise.
if phi<1, this series is covariancestationary, so you can take the expected value of both sides. E(epsilon)=0, so the expected value of the series is:
, or mu=c/(1phi).
In other words, it's a mean reverting series.
if phi==1, y_t is a Wiener process (a random walk), it has a 'unit root'. you can fix this by first differencing the series. since the derivative is white noise, you are reduced to the above case.
if phi>1, this time series is explosive, which is just no good.
VAR:
instead of phi, you have a matrix of coefficients, so you take the eigenvalues to see what effect each eigenvector is having. You check the effect in the same way as above: any lambda>1 is explosive and makes the series worthless. If all the lambdas are < 1 there is a static equilibrium to the system (kind of like meanreversion).
The difference with VAR is that some lambdas could be random walks and others stationary. That's where cointegration comes in (from that other thread)
t. 
the only reason it would be easier
to program in C is that you can't easily express complex problems
in C, so you don't. comp.lang.lisp 


IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 


Making some progress...
Here is kind of a neat thing:
g(t)*W(t) = int_0^t g'(tau)*W(tau) dtau + int_0^t g(tau)*W'(tau) dtau
where W(t) is a Wiener process and W'(t) is Gaussian white noise.
I got that from this paper.
In the above expression, the function g(t) must be smooth and have finite support. If we let g(t) = 1 up to some time T and then smoothly taper to zero beyond, then the above reduces to
W(t) = int_0^t W'(tau) dtau
for t < T. This expresses what others have said about Brownian motion being the integral of white noise.
If we approximate this integral as a Riemann sum, we get
W(t) ~= W(tdelt) + delta W'(t).
Comparing this to what tristan said for c = 0 and phi = 1, we have
y(t) = y(t1) + e(t).
This seems to support the statement that y(t) is Brownian motion for phi = 1. At least if you were to throw in a delt in there and take a limit as delt > 0. I can buy that
This is looking like it supports my suspicion that AR could be thought of as a finite difference approximation to a stochastic DE (I think).
Still some mysteries though. If anyone could shed some light, it'd be appreciated.
For example, the integral implies
dW = W' dt.
We know that
dW dW = dt
so that means that W' ~ 1/sqrt(dt) since
dW dW = (W')^2 dt dt = dt.
If we wave the wand of NCG, we convert this to a commutator
[dW, W] = dW W  W dW = W' dt W  W W' dt = W' W dt  W W' dt (W and dt commute) = dt.
The only way to satisfy this is if
[W, W'] = 1
which looks like a quantization rule to me. I'll try not to get pulled off onto that tangent right now (Note for anyone interested, the commutative relation [dW, W] = i*hbar*dt leads to the Schrodinger equation and in this case would lead to [W, W'] = i*hbar, which looks even MORE like quantization where white noise and Brownian motion play the role of conjugate variables like position and momentum.)
Back to something remotely practical...
If we want to solve
dy = mu*dt + sigma*dW
approximately we can rewrite it as
dy = mu*dt + sigma*dW = mu*dt + sigma*W'*dt.
The commutative relations suggest W' ~ 1/sqrt(delt) so this becomes
y(t)  y(tdelt) = mu*delt + sigma*W'(t)*delt = mu*delt + sigma*e(t)*sqrt(delt)
where I've set
W'(t) = e(t)/sqrt(delt)
and I really don't know exactly why other than to make it look like a simple Monte Carlo expression (and it seems to somehow relate to the dimension analysis from the commutative relations)
Rearranging terms gives
y(t) = mu*delt + y(tdelt) + sigma*e(t)*sqrt(delt).
Setting
c = mu*delt
and
epsilon(t) = sigma*e(t)*sqrt(t)
brings us back to something that formally looks like the AR expression.
Ok. As you can see, there are still some holes in my arguments so if anyone can help nail the final missing pieces, I'd appreciate it. That, or show me where I'm totally off the mark.
Just to summarize...
This whole thing is motivated by trying to understand the relation between ARs and SDEs. My hunch is that AR may be interpretted as an approximate numerical (finite difference/Monte Carlo) solution to a simple SDE.
This might be obvious to some, so help me see the light. I'm almost there
Cheers Eric 




Cheng


Total Posts: 2585 
Joined: Feb 2005 


It is still a bit early to think and maybe I am babbling big nonsense but
y(t) = mu*delt + y(tdelt) + sigma*e(t)*sqrt(delt)
boils down to the Euler scheme for numerical solution of SDEs if e(t) is some N(0,1) distributed rv. Then you get the increment
sigma*sqrt(delt)*N(0,1)
or alternatively
N(0,sigma^2*delt).
Under some technical assumptions this converges to the SDE for delt>0 with order O(delt^(1/2)). If mu and sigma are C^1 you can get convergence of order O(1/delt) by adding another term involving the first order derivatives (the Milstein scheme). Maybe this helps, if not feel free to trash it.
Regards 
"Don't try to run, don't try to hide. Believe me, the hammer's gonna make it right !" 


kr

Founding Member NP Raider

Total Posts: 3561 
Joined: Apr 2004 


eric  haven't read all the details, but my intuition is simply that dW should be thought of as a 'generalized function'  i.e. think dirac delta. That is, it is defined only to the extent that one can compute <g, dW> = int_0^t g(t) dW_t, and the value of this functional is mostly determined by a few key algebraic relations. In fact, writing <dW, dW> is not the right way to go, much better to have W = <1, dW> be thought of as a random variable, and then use the rules for <W, dW>.
Once that's said, it's clear that "dW'_t" does not operate on the correct space... it would have to be integrated against something else in order to produce a generalized function. Somewhere under all this is a deepdiscount graded algebra involving functions with existing derivatives. 
my bank got pwnd 



Cheng


Total Posts: 2585 
Joined: Feb 2005 


kr,
do you think along the lines of Sobolev spaces ? This was basically what I had in mind when I started mumbling about generalized derivatives.
Regards 
"Don't try to run, don't try to hide. Believe me, the hammer's gonna make it right !" 


Nonius

Founding Member Nonius Unbound

Total Posts: 11665 
Joined: Mar 2004 

 

Martingale

NP House Mouse

Total Posts: 2590 
Joined: Jun 2004 


The origional way of Wiener's way to construct the stochastic integral is kind of from white noise as generalized distribution ( throgh some L^2 isometry), but it appoved to be limited, that's why people start to consider this from some other point... I will try to dig some history literature on this 



urs


Total Posts: 2 
Joined: Feb 2006 


Hi Eric!
I thought I might just as well post a reply here.
You wrote
We know that
dW dW = dt
Ok.
so that means that W' ~ 1/sqrt(dt) since
dW dW = (W')^2 dt dt = dt.
This is making me a little nervous. But I realize my stochastic calculus is rusty.
If we wave the wand of NCG
Now I feel more at home...
[dW, W] = dW W  W dW = W' dt W  W W' dt = W' W dt  W W' dt (W and dt commute) = dt.
I understand
[dW,W] = dt .
But does it make sense to write dW = W' dt ??
Is there any discrete 0form W' with this property?
I wouldn't think so, but it has been a while since I thought about this stuff.
But consider a 2D diamond graph. Introduce "lightcone" variables A and B such that
t = A + B
x = A  B .
Then
dt = dA + dB
dx = dA  dB
and there is manifestly no 0form f such that
dx = f dt .
But in your example above W=x, if I understood correctly.
Do you agree? Maybe I am missing your point.
On the other hand, maybe you are right and it is useful to *introduce* new formal expressions like W' that are defined to satisfy relations like dW = W' dt. Haven't thought enough about that. 




IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 


Hi Urs!
Welcome to NP!
I'll take the liberty for a little introductory comment...
I had the great pleasure to coauthor a paper with Urs (the highlight of my academic life actually). After six years of flailing around in a corner on my own, I finally somehow finished my PhD and decided that reaching out to others wouldn't be such a bad thing. I knew Urs through the newsgroup sci.physics.research (of which he eventually became a moderator along with John Baez and others last time I checked). Urs had some interest in stochastic formulations of QM (by Nelson if memory serves) and some comments I made relating NCG and SC must have attracted some of his attention.
The basic idea is that any directed graph gives rise to a special algebra/calculus that is unique to that graph. I had also (re)discovered that a certain algebra leads to stochastic calculus. Since every directed graph gives rise to an algebra and a particular algebra gives rise to stochastic calculus, the question we asked ourselves one day is kind of an inverse problem, "Which directed graph gives rise to the algebra from which SC derives." Urs subsequently left for a two week bike ride across Europe and when he came back we had both answered the question independently. In a flash it all became almost obvious. (particularly in hind sight). The binary tree is the unique directed graph that gives rise to the algebra that leads to SC. That was enough to ignite an intense collaboration that was very exciting (for me at least) that lasted several months and culminated in this paper
Discrete Differential Geometry on Causal Graphs
Anyway, it is always a great pleasure to discuss things with Urs so I'm glad he made an appearance. So now back to the topic at hand! (which I think has some profound (although perhaps too academic) significance). That is the possible fact that Brownian motion and white noise are conjugate variables in the sense that position and momentum are in QM.
so that means that W' ~ 1/sqrt(dt) since
dW dW = (W')^2 dt dt = dt.
This is making me a little nervous. But I realize my stochastic calculus is rusty.
I don't blame you, it makes me nervous too Without making too much out of it, we should probably think of that as no more than dimensional analysis, i.e. getting the units to work out right. Although if anything I am saying has any validity, it must be more than that. That, or what I think is more likely, writing things like dW dW = dt is not really valid in the first place and the NCG approach of commutators is the proper way to proceed.
I understand
[dW,W] = dt .
But does it make sense to write dW = W' dt ??
I don't know Not sure if it helps, but it has been pointed out that W' should be thought of in the distributional sense. The reason for me feeling reasonably comfortable writing that down in the first place is the (almost) definition of white noise
W(t) = int_0^t W'(tau) dtau.
Since W(0) = 0, we can rewriting this as
W(t)  W(0) = int_0^t W'(tau) dtau
which can be rewritten
int_0^t dW = int_0^t W'(tau) dtau,
which I thought would allow us to write
dW = W' dt.
Is there any discrete 0form W' with this property?
The commutative relation
W W'  W' W = 1
would seem to indicate that perhaps neither W nor W' are discrete 0' forms (in the obvious sense) and maybe should be thought of as matrices or something. Not sure.
It might help to make the analogy with QM more clear. Since it makes me so happy, I will repeat the little computation
(using x in place of W now)
[dx,x] = dx x  x dx = x' dt x  x x' dt = x' x dt  x x' dt = dt
which implies
[x',x] = 1.
The QM version of this is
[dx,x] = dx x  x dx = x' dt x  x x' dt = x' x dt  x x' dt = i*hbar*dt
which implies
[x',x] = i*hbar.
(maybe should replace hbar with hbar/m)
Looking at this, we see that a Wiener process (Brownian motion) is analogous to momentum in QM.
It is then somewhat tempting to write something like
dW = W' dt = d/dW dt
so that
[W', W] = [d/dW, W]
and
[d/dW, W] psi = (d/dW W psi  W d/dW psi) = (psi + W d/dW psi  W d/dW psi) = 1 psi
so that
[d/dW, W] = 1.
Is it possible that if W is a Wiener process then white noise can be thought of as W' = d/dW?
That would be kind of wild and, if true, I'm pretty sure no one has ever observed that. We could maybe write up a page or two and throw it on the arxives
What do you think?
Cheers Eric 



urs


Total Posts: 2 
Joined: Feb 2006 


I perfectly agree that the equation
[dx,x] = dt
or
[dx,x] = i hbar dt
morally encodes the commutator which you have in mind. What you point out is that if we could somehow divide both sides by dt, we would get something like a commutator of dx/dt with x.
This is certainly a good way to think about it. Personally, however, instead of wanting to rewrite this in the form [x',x] = 1, I would regard [dx,x]=dt as the better and more robust formula.
I also agree that there is a close relation to how white noise is morally like the derivative of the Wiener process. After all, one way to derive the commutation relation [x',x] = k 1 is to use the path integral formulation of the free particle. The rigorous version of this is nothing but an integral over paths using the Wiener measure.
Hence, many technically quite sophisticated facts are somehow quite neatly and very simply already encoded in an equation as simple as [dx,x]=dt .
Looking at this, we see that a Wiener process (Brownian motion) is analogous to momentum in QM.
Wait. Is that what you want to say?
I think the Wiener process is like the position observable in QM. White noise is like the velocity (~momentum) of the particle.
dW = W' dt = d/dW dt
The second equation looks a little strange. Wouldn't you want
W' dt = (dW/dt) dt ? 




IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 


Looking at this, we see that a Wiener process (Brownian motion) is analogous to momentum in QM.
Wait. Is that what you want to say?
I guess before we can settle this question, we'd have to settle the question of whether we can write
dW = W' dt.
I think we can, but there are enough moving pieces that I could be wrong about that. If we can write this, then
[dW,W] = dt
leads to
[W',W] = 1.
Looking at the first expression (dW = W' dt) makes you want to say
W' = dW/dt,
but looking at [W',W] = 1 makes me want to say
W' = d/dW
unless there is another representation of W' that satisfies that commutative relation. By a quirk in dimensions, at least the two have the same units, i.e.
[d/dW] = [dW/dt]
since
[W] = [sqrt(t)],
where I'm using brackets to denotes "units of", i.e. [W] = "Units of W".
I think there is something neat here, but as usually, I'm only smart enough to "sense" it and am struggling to dig it out
Oh! Let me clarify one point because it is important. The expression
[W',W] = 1
does not come from "dividing by dt". Rather, it comes from plugging
dW = W' dt
into
[dW,W] = dt
and using the fact that dW and dt commute. I carried out that computation (since it is so neat) a couple of times 



IAmEric

Phorgy Phynance Banned

Total Posts: 2961 
Joined: Oct 2004 


One more thing...
You said
I think the Wiener process is like the position observable in QM. White noise is like the velocity (~momentum) of the particle.
Yes. This is basically what I'm trying to say. I was being a little cavalier about constants and not making much effort to distinguish between velocity and momentum (although I made a comment about perhaps writing hbar/m at one point).
I would say it this way...
If we can write
dW = W' dt,
then the commutative relation
[dW,W] = dt
implies the Wiener process is like the position operator in QM and white noise is like the velocity (~momentum) operator in QM. 



