Using neural networks in trading - page 27
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
The book is definitely good, but it has nothing to do with making money in the financial markets. For the sake of self-education you can certainly read it, but again, it will not help to earn money in the financial markets )))).
Leonid, Robot asked: "I want to know more about neural networks...where to start", I think Vinin hit the bull's-eye with the answer to this very question. In the doc provided by Vinin it is exactly that - what paradigms there are and what they are for.
And "as applied to..." is the next "stage of evolution".
2 LeoV
You say all the right things, comrade.... I'm not arguing. The man wanted it, got it and said thank you. And we're getting hung up on God knows what.
I like what you said, Leonid. Let's talk about that - how to start with the multiplication table and end up with the right one.
There is a lot of talk about neural networks. I do not know whether it is possible to build AI using neural nets, but I think that we do not need it for a long time. That is, in other words, we don't need an intelligent analyser of "market reality".
If you strip away the labels and epithets from neural networks, all that is left is a powerful tool with two remarkable properties. The first one is non-linearity and the second one (and this is just cool) is auto-adaptivity )).
Two utterly defunct properties that we can't put "to good use". As I understand it, among the whole mass of neuro-workers there is a common opinion: I don't know, I don't understand, I don't see; no matter how I trade, I'm always in the red. Now I will make an artificial brain, I will show him what for, he will understand, see and evaluate everything, and will enter the market correctly and make lots of money. If we (attendees here) are doing the same thing, then we don't have to worry about anything - we will have something to do for many, many years. )) A paradoxical maxim was born - an endless dead-end road.
What have I personally tried? I tried two things. The first thing I described above and have already stopped doing it - building an artificial brain. )) The second thing - I tried preening.
I will now write a few controversial points, but these points are my personal conviction, acquired as a result of the work I have done and observations I have made. The first is that predicting and predicting are completely different things. The second is that predicting cannot be called "probable development". The third is that a prediction is nothing more than an extrapolation. And once that extrapolation is made, it ceases to have anything to do with the market. It is only a chart that sticks out beyond the zero bar in a certain way. And this "image" is defined only by the previous price movements. The market will move in an absolutely random direction. And as practice shows, not at all in the direction specified by the prediction. This suggests a non-trivial way to use the prediction - the prediction is the direction in which the market will never go.
A prediction, on the other hand, is another matter. If we simplify the situation to the maximum, the task may be formulated as follows: an hour bar has opened. Where will it close? Upper or lower? 1/2 probability already exists - either upper or lower (I deliberately excluded a doj from consideration, the doj is not dangerous). And how is the grid's ability to learn the multiplication table useful here?
Frankly speaking, I haven't even begun to solve this problem. I'm tempted to play with a probability grid. But this grid consists of two parts: the left part is a classifier (SOM), the right part is RBF (or the same pepper). It's the left part that presents the biggest challenge. I tried to load different information from the chart. But if you look at this information through the unit circle you get a mess. Scientifically speaking, it's chaos. There is simply nothing to classify. All SOM work on the principle of Euclidean metric. So how many classes you set, you get, and SOM will just scatter the cluster centres evenly. This isn't classification, it's nonsense. Consequently, calculating "current membership" makes no sense at all.
So, that's about it...
greatooooooo .... What are you going to feed to the input?
I wrote that no matter what you put in there, the result is... It's a bunch of bullshit.
Maybe you shouldn't put what you can in there after all. Maybe there should be some connection between what goes in and what comes out.
solar, I draw your attention to the fact that I hinted at the problem of classification. and this is the principle of learning without a teacher. i.e. there is no "out" concept here.
P.S. sorry, for some reason I had to change my nickname leksus to the current one.