Machine learning in trading: theory, models, practice and algo-trading - page 2325
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
I looked a little more closely - I see that I was somewhat mistaken. They make a series of sliding nonlinear metrics from the original series (they write about fractal dimensionality and Lyapunov's indices). This new series they consider (based on practical observations) similar to SB. And so they multiply this series by Monte Carlo type method in the future and from the resulting set take a variant with the greatest closeness to the original one.
The secret is the type of specific transformation of the original series into a series of metrics and, more importantly, the inverse transformation.
In general, all this looks suspicious (first of all, the style of presentation of the results) and does not cause much desire to further study the question.
It also seems too beautiful and vague-something is not agreed upon. In addition, the series are taken similar statistical characteristics.
As funny as it is to you, that's exactly what you're doing...
Let me explain...
I meant that it would be good to create a "converter" of market data (not stationary) into a model (stationary, simplified, demonstrative, with preservation of the structure we need) and this model can be represented by sine waves
All scientists do it all over the world to understand a complicated process, build a model, study the model, predict the model but not the process itself, this is the world practice, everybody does it, except for the people with the lowest level of training who believe that AMO will do everything themselves...
It was a little clearer what you meant. Of course it did not become more useful.
For starters, just to say - that when you take quotes, ANY quotes, and do something with them, you are already dealing with a model, not with the very, as you put it, complex process.
Well about the rest of derivatives from quotes, needless to say)
Your derogatory comment about MO-scholars speaks more about your low level of training.
Of course, it didn't make it any more useful.
Neither have I, by the way.
For starters, let me just say that when you take quotes, ANY quotes, and do something with them, you're already dealing with a model, not, as you put it, the complex process itself.
Wow, thanks... Did anyone claim otherwise?
Your pejorative reference to MO-ers speaks more to your low level of training.
Right, right.....
============
What next, do you have something to say, or shall we state the obvious, and make subjective judgments?
What's next, do you have something to say, or are we going to state the obvious and make subjective judgments?
Well the obvious is that two or three sine waves can't even approximate the price normally over a long period. Okay, don't you agree?
Well, let's also talk about models - I have an idea - a model of the market through a network of players.
Roughly looks like this (I will try to use the context of the MO):
There are N subjects players, which are classified according to a number of features - volumes, frequency of deals, duration of position, tendency to buy-sell, access to information and speed,aggression, etc.
The market (instrument price) is modeled as the result of exchange between the subject players (simplified cup). There is an environment that is a provider of regular news and relatively random events to which players react, and the environment transmits information between players.
I hope it is generally clear. I don't remember similar research-publications in terms of model implementation, which is understandable, because practical results can only be achieved with access to big real data.
But as a model for research I think it is quite suitable. For interpretation in machine learning methods - room for creativity, obviously, simple architectures can't do here and we need to develop something special.
Well, the obvious thing is that two or three sine waves can't even approximate the price properly over a long period of time. Okay, don't you agree?
Well, let's also talk about models - I have an idea - a model of the market through a network of players.
Roughly looks like this (I will try to use the context of the MO):
There are N subjects players, which are classified according to a number of features - volumes, frequency of deals, duration of position, tendency to buy-sell, access to information and speed,aggression, etc.
The market (instrument price) is modeled as the result of exchange between the subject players (simplified cup). There is an environment that is a provider of regular news and relatively random events to which players react, and the environment transmits information between players.
I hope it is generally clear. I don't remember similar research-publications in terms of model implementation, which is understandable, because practical results can only be achieved with access to big real data.
But as a model for research I think it is quite suitable. For interpretation in methods of machine learning - room for creativity, obviously, there is no help here with simple architectures, and it is necessary to develop something special.
Agent-based models? There are a lot of such things in modern economic science. In my opinion, it is a good thing for the philosophical understanding of the market.
I am not sure about the possibility to derive any practical use from this approach (in the sense of creating trading strategies).
Agent-based models? Lots of them in modern economic science. In my opinion, a good thing for philosophical understanding of the market.
I am not sure about the possibility of extracting practical benefits from this approach (in the sense of creating trading strategies).
Yes, I remember scientific economic and (closely) sociological descriptions of such models long ago from the institute. With respect to trading in the light of the latest achievements of the MO it seems that the matter is not that it cannot be applied, and those who have resources - will not get the required benefit from it, they are good as it is. Enthusiasts have not yet reached, digest all sorts of GPT-3, and other breakthrough things, and maybe someone will reach some of the directions of development will indicate in this.
The difficulty is also that a lot of irrationality in the behavior of players, especially at key turning points in trends, which is difficult to reliably model with current models.
ap. Another thought - that it is not always correct to aim at predicting the price movement, it is so primitive. It is possible to receive information about the state of players by the price movement and make long-term conclusions from this constantly updated.
One promising approach is causal inference. This topic is quite actively developed by large IT companies. There are libraries.
There are articles on the topic
Making a calculator that will go through the options and find the best oneYes, I remember scientific economic and (closely) sociological descriptions of such models long ago from the institute. As for trading in the light of recent achievements of the MO, it seems that it is not that you can not apply, and those who have the resources - will not get from it the proper output, they are so far all right. Enthusiasts have not yet reached, digest all sorts of GPT-3, and other breakthrough things, and maybe someone will reach some of the directions of development will indicate in this.
The difficulty is also that a lot of irrationality in the behavior of players, especially at key turning points in trends, which is difficult to reliably model with current models.
ap. Another thought - that it is not always correct to aim at predicting the price movement, it is so primitive. It is possible to get information about the state of the players by the price movement, and already from here to make long-term conclusions, constantly updated.
In my opinion, the main problem lies in the choice of approach to describing the behavior of the biggest players in the market - the states. They (1) strongly influence the market, (2) their behaviour changes considerably with time, (3) their aims for action in the market often lie outside the market itself and are poorly known to us, (4) there are many states and they can interact with each other in very different (for the market) ways. Mathematically, the result is a complex, unsteady, and unclosed system.
The problem is not that it is impossible to come up with a model for such a system but that it is possible to come up with too many different ones and probably even contrary to each other in terms of conclusions.)
One promising approach is causal inference. This topic is quite actively developed by large IT companies. There are libraries.
There are articles on the topic
Make a calculator that goes through the options and finds the best onethat's the other side of the approach initially. RCT in medicine for anything and everything by the way killed medical techniques, not reproducible placebo)))
The task of behavior or condition to find a causal relationship)
ap. Another thought - that it is not always true to aim at predicting the price movement, it is so primitive. You can get information about the state of the players by the price movement, and from here you can make long term conclusions, constantly updated.
This is a good and correct idea. Only perhaps not the state of the players, but the state of the causes affecting the players. Although maybe this is just the next step.