Machine learning in trading: theory, models, practice and algo-trading - page 1028
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
A little bit about the mechanics of the market...
I'm going to throw some pictures right away, and then as it goes...
1)
Forex broker "oanda" which publicly broadcasts open positions of its clients
This is the difference in the sum of net open positions of the broker buyers - sellers
2)
Long time ago in the 14 th year when we had our own trader's firm, I traded by the indicator which was built on the mullion, the cumulative sum of buyers and sellers was built for a certain period of time and delta was calculated, the indicator was built in Excel, here are the pictures from my diary, 2014
3)
Trained neural network, for two classes to buy to sell, but instead of a class like "11010010" it gives the probability of a class, again we do a cumulative sum of the probabilities of classes to buy and classes to sell and count the difference, this is the blue graph and the neural network work on the new data
Is it necessary to draw arrows or to do it yourself? )))
by the way this is page 40 of this thread
As you can see, different sources, different approaches, but the essence is the same...
The market rises when they sell and falls when they buy...
I wonder if anyone knows how to distinguish a randomly generated BP from a real price one, explain if anything...
You could try applying agreement criteria (Kolmogorov-Smirnov, for example) between samples of increments of the generated and real series. It is believed, for example, that the real prices give thicker tails and a sharper center than the Gaussian distribution.
As you can see, different sources, different approaches, but the essence is the same...
The market goes up when you sell and down when you buy...
that's true
one small and almost insignificant nuance
How can a robot sell when the price is rising?
oops!
Just think about it, ok?
I've already written above why that is, if anything...
I wonder if anyone knows how to distinguish a randomly generated BP from a real price one, explain if anything...
In real BP you can observe repetitive (24 hour period) at certain times of the day volatility increases associated with session openings. It does not exist in the random one.
one little and almost insignificant nuance
How can a robot sell when the price goes up?
oops!
I don't get it, what's the problem?
I don't understand, what's the discrepancy?
........
Do you have to draw the arrows or have you already done it yourself? )))
by the way this is page 40 of the same thread
Yeah, we need arrows.
This is exactly the same open-source indie I put in the forecasts thread back in 2012. I hope it's still there.
everyone was laughing there, because no one understood anything ;))))
of course the mirror kotir, depending on the situation, it is buying or selling
and again.
All this works fine, as well as any neural network, as long as the market is flat.
I generated a random (THIS IS NOT THE PRICE) series with a trend component and painted it with technical analysis figures, to show that post factum classical TA can be described even by Random)), and that even in Random this TA kind of exists, but it is not there, it is just a property of series with a trend component.Any reversal can be described by TA figure, you know? it will always be either head shoulders, or double or triple top, even in random, but at the same time it does not give any predictive properties to these figures
It seems to me that it's not a matter of whether or not these shapes are there in symmetric random walk. The more correct question is whether there is a statistically significant (and practically useful) difference in the behavior of a series of real prices around these figures from a version of the randomly generated ones. If we want to check it precisely, we will have problems with formal definition of figures, etc., etc. For example, gaps used to close noticeably faster than they should have with symmetric random walks (not that there's any money to be made on that).
Random symmetric rambling is quite good at "drawing" trends and cycles, and it can be shown by the theoretic methods. But of course it is impossible to make money on it.
I think that the random walk may be used as an alternative history for optimization of primitive trending systems that have no predictive properties, but are simply trend following.
For example, I can generate a 3000 year alternate history and optimize a trend robot on these data. I think that the robot will behave better in real trading using the new data than if it was optimized for the last several years of the real history.