Using Neural Networks in Trading. - page 9
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
we have to use interpolation between discrete samples of the distribution function F(n)
Yes, it looks like that... although, in principle, you can do without interpolation... but it will completely ignore the tick volume...
there is another option: to use the interpolation mechanism of the MT tester itself... just take an .fxt-file... there is a sequence of ticks there... I don't remember exactly: is it technically possible to open the generated file and somehow pass this sequence to the network (or to a simple expert)... but I think we can work something out... we just have to generate a new .fxt-file every time, but if the sample volume is small, I think the speed will be acceptable...
but in general, Neutron, you'd better stay out of these ticks... why do we need such precision... one minute sampling is enough for us... we just need to "patch up" the holes in the data...
SZZ... and anyway, I still don't understand what we're talking about... if you've already got the distribution function F(n), I don't really understand what discrete reports you're talking about... (I'm talking about "my" thing - the price vs. time function :))
I don't understand what talk about discreteness of value can be when considering incremental series...
Example of discrete: the number of trades before a profit of 150 pips, i.e. as soon as you get >=150 pips, the count starts again. So in such a sample there may be numbers 1,2,3,4,...8,...100... but not 12,3 or 2,7.
If you look at the price range itself - hard to say if it's a discrete value or not, more likely discrete...
Neutron, can you please send me the sample you're working with, just 1 or 2 rows that need to be aligned? just so I can't understand what you're working with...
continuous time according to Shiryaev is simply a Markov chain with continuous time...
he did not raise the issue of price discreteness at all... i.e. in fact, the price should always be considered a continuous value!
Neutron, could you please send me the sample you are working with, just 1 or 2 rows that need to be aligned? Just so I can't understand what you are working with...
Please.
The file contains an exponentially distributed random variable. The task is to get a uniform density of distribution from it and show the way. You cannot stretch splines. All processing is in discrete form only.
Yes, does it make a difference where I got it? - I generated it in Matcad. That's not the point, the point is that I don't understand what's going on!
Look, we take a time series (TP) rounded to integer values (like price - discreteness of a whole pip) with exponential distribution (see file above) and build its probability density function (red circles, left picture), then we draw through these points, by least squares, the exponent of the form y(x)=A*exp{B*x}. Now, we construct the distribution function (PDF) for discrete densities and for the analytically defined one (middle figure). We have done it and now try to equalize the initial distribution by influencing it with discrete PDF and analytically given one (fig. on the right):
You can see that in both cases it was not possible to obtain a rectangular distribution. This is what I am struggling with.
However, if I set BP with the same distribution, but without rounding the values to integers (see file below), the picture changes:
Now, for the analytically approximated distribution we easily get the desired rectangular density distribution (Fig. right, blue circles), but for the discrete case it is still bad (red ones). So the method works only for analytically given density distribution of increments. Well, or, as usual, I'm missing something! In short, the distribution cannot be smoothed by an easy move, we have to pre-stretch the splines on the initial one, and it's already a headache.
In short, it's not possible to align the distribution with a light movement, you have to pre-stretch the splines on the initial one, and that's a headache.
>>Yes it's easy. Piecewise linear approximation of the distribution, then redistribute accordingly to the area.
listen, Neutron, i can't understand what you have on y-axis in distribution function? some 5000, 10000... what is it?
By definition, FR=integral(from PR). That's where the thousands come from, it's a commutative sum.
It's easy.
Go ahead and show the "easy" for an integer BP.
Yes, does it make a difference where I got it? - I generated it in Matcad. That's not the point, the point is that I don't understand what's going on!
Look, we take a time series (TP) rounded to integer values (like price - discreteness of a whole pip) with exponential distribution (see file above) and build its probability density function (red circles, left picture), then we draw through these points, by least squares, the exponent of the form y(x)=A*exp{B*x}. Now, we construct the distribution function (PDF) for discrete densities and for the analytically defined one (middle figure). We have done it and now try to equalize the initial distribution by influencing it with discrete PDF and analytically given one (fig. on the right):
You can see that in both cases it was not possible to obtain a rectangular distribution. This is what I am struggling with.
However, if I set BP with the same distribution, but without rounding the values to integers (see file below), the picture changes:
Now, for the analytically approximated distribution, we easily get the desired rectangular density distribution (Fig. right, blue circles), but for the discrete case it is still bad (red ones). So the method works only for analytically given density distribution of increments. Well, or, as usual, I'm missing something! In short, the distribution cannot be smoothed by an easy movement, we have to pre-stretch the splines on the initial one, and that's a headache.
I do not understand how you got uniform (Fig. 2, the file did not look) ...
And the analytical notation here is different, the law of distribution is different, most likely Poisson...
There is still a way to code a discrete value so that it was distributed uniformly, but can't do without headaches, I'll post ressues later ...
No, you can't do anything with a discrete, only continuous...