Using Neural Networks in Trading. - page 7
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Alexei, am I even thinking correctly? Can you tell me something about it?
Artem, to be honest, I haven't tried to figure out how this relates to NS. I simply answered Sergei's question - purely theoretically. Of course, sigmoid is not erf(x) at all. But your general thought goes roughly in the right direction, since sigmoid is similar in form to erf(x).
Neutron, just do a model experiment to see for once that that nonsense is absolutely true. You don't even need MQL here, everything is done in grandpa's Excel, with an axe. I am attaching the archive. Explanation:
Initial normal with parameters K2, L2 - column A. Generation of normally distributed values is performed from the usual MSG using Excel statfunction NORMOBR(), the inverse of the normal distribution integral function (have you ever wondered what it's there for, by the way?). You don't have to believe that's the way to do it: I've drawn a histogram here that looks suspiciously like a Gaussian histogram. Spin the K2, L2 parameters and see how the histogram changes. I purposely generated more values, around 20 thousand, for the distribution to be smooth.
Then, I simply apply a straight normal distribution function with the same parameters to the same points in column A, and it builds a histogram again. The transformed values are in column E, the histogram is in a row.
Columns B, C and F, G are needed to build the histograms themselves.
P.S. If you need a real live experiment without any tricks with Excel functions, then try to find an array of normally distributed values in the grid (or some other way) and do the same.
Then I simply apply a straight normal distribution function with the same parameters to the same points in column A - and make a histogram again.
I don't get it :-(.
Something I do not do like you all. Well, I got the distribution function, then what? I should substitute the input vector in it as an argument and then I get a vector with a uniform probability density at the output?
P.S. Everything seems to work out for continuous VR (given analytically):
Here the original distribution (exponential) is shown in red, and the resulting uniform distribution is shown in blue. Indeed, if f(n) is the probability density of the input vector Y and F(n) is its distribution function, then to equalize the density, we have to construct F(y) and use it as the input.
Here, only for a discretely set value this trick doesn't work for me. And it's the discrete set point that we have to deal with.
By the way: long ago I wanted to ask - why should we consider price function to be continuous?
Good evening.
I understand that the members of this thread have their own hard-earned opinions on many issues. But I dare link to the foreword of Shiryaev's book "Fundamentals of Stochastic Financial Mathematics". He describes his thoughts on discreteness and/or continuity of prices.
Good evening.
I understand that the members of this thread have their own hard-won opinions on many issues. But I dare link to the foreword of Shiryaev's book "Fundamentals of Stochastic Financial Mathematics". There he describes his thoughts about discreteness and/or continuity of prices.
You can't put everything you've suffered and treasured into a book:) It is true that books have long since described everything you need in principle to adequately operate a network with time series. The problem arises, frankly speaking, with what people have here.
Only for a discrete value this trick does not work for me. And it's the discrete value we have to deal with.
Well, that link said so about continuity. To be honest, I didn't get into this nuance.
I don't really know what the problem is here.
I can't figure out exactly what Neutron's problem is either... I don't "rule" in ns yet...
In principle, as far as I understand it, price can be seen as both a discrete process and a continuous one... the question is: which is the right one?
To be honest, I haven't got to Shiryaev yet... left it for later as "the best part"...
Victor (renegate), could you please briefly formulate the conclusions Shiryaev comes to (I mean, what is he getting at - should the price be considered as a discrete or continuous value)?
I can't figure out exactly what Neutron's problem is either... I don't have a handle on it yet...
In principle, as far as I understand it, the price can be considered both a discrete and a continuous process... the question is, which is the right one?
to be honest, i haven't got to shiryaev yet... left it for later as "the best part"...
Victor (renegate), could you please briefly state what conclusions Shiryaev comes to (I mean what he's getting at - should we consider price as a discrete or continuous value)?
continuous processes exist only in mathematics
:) I think we're about to start arguing about whether an electron is a particle or a wave...
although in general I agree... in practice we are dealing with a discrete process of incoming ticks... but whether the objectively existing price in the market is discrete... is the question...
:) I think we are about to start arguing about whether an electron is a particle or a wave...
although in general I agree... in practice we are dealing with a discrete ticking process... but whether the price objectively existing in the market is discrete... is the question...
That's not the question. It is discrete and in a superposition of states.
It is not an issue. It is discrete and in a superposition of states.
and the arguments? why do you think so?