Dependency statistics in quotes (information theory, correlation and other feature selection methods) - page 38
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
Yes what? Where's the answer?
Although it's about to descend into the usual dumb rubbish around models I can feel. Not interested.
I don't have a "noise" without a "model" because there is no model that 100% explains price behaviour. If you don't have a model, then either there is no noise, or everything is noise
Ah, well, then all the curwafitters go to hell. With their interpretations of noise together :)
Wouldn't you rather talk about signals than the model?
.... "trend" ("elusive" concept) which shows the "direction of travel" ....
)))))))))))))))))))
this guy really makes me happy - a childlike, direct view of the world!
Everyone was building models, doing calculations, and just drawing a line from the ruler on the graph
+1000
I have to give you a "warning". A smile and humour is a great rest for the brain when storming off, 'crossing the line' disrupts the 'train of thought'.
yes
That's not quite right. Not everything can be filtered out. It is not possible to include everything in the model, something has to be neglected. Accepting it as an error (or noise).For example, there is a sine wave Asin(Bx) + another signal. The signal may or may not be random. We are trying to estimate a model from the total signal, but only take the sine model for estimation. What we have neglected will give us an error in estimating the parameters of the sine model.
It's the same in trading. Price is a mixture of a bunch of different processes, some of which we neglect within a particular model. The neglected is noise.
Now on to the formulation of the question as I understand it ;)
"Not everything can be filtered out."
Or maybe nothing needs to be filtered? This is in terms of the alternative being put forward.
"For example, there is a sine wave Asin(Bx) + another signal. The signal may or may not be random."
Well, since "random", I beg to differ immediately here as well. What do you mean by accident? And explain on your fingers, where do the cause-effect relationships go? In this model they are unidentifiable and therefore... Or?
"Same in trading."
More correct in analysis, not trading, maybe?
"Price is a mixture of a bunch of different processes, some of which we neglect within a particular model. The neglected is noise."
Price as a stream of data, or a single quotation? How with the latter can you neglect the part? The quotation is indivisible. Or?
"For example, there is a sine wave Asin(Bx) + another signal. The signal may or may not be random."
Well, since "random", I beg to differ immediately here as well. What do you mean by random? And explain on your fingers where do cause and effect relationships go? In this model they are unidentifiable and therefore... Or?
"Same in trading."
More correct in analysis, not trading, maybe?
"Price is a mixture of a bunch of different processes, some of which we neglect within a particular model. The neglected is noise."
Price as a stream of data, or a single quotation? How with the latter can you neglect the part? The quotation is indivisible. Or?
"Or maybe you don't need to filter anything? This is in terms of an alternative.
Well, if you don't need to filter anything out within a particular model, you don't need to))
"Well, since 'random', then please let's get straight to the point here as well. What do you mean by random? And explain on your fingers where do cause and effect relationships go? They in this model are unstable and therefore... Or? "
Are random, so the observer does not know the law of formation
"Correct in analysis rather than trading, maybe?"
" in trading
" Price as a stream of data, or a single quotation? How can you neglect the part in the latter? The quotation is indivisible. Or? "
As a stream of data
Andrew, I understand roughly what you are talking about. Noise can be generated by imperfection of the communication channel, for example. In that sense, 99% of the quotes are probably pure signal, another 1% are distortions by filters on the side of the quote providers.
But it is possible that some very big players can be singled out and their behaviour can be taken as a determining factor, and then everyone else's fuss can be perceived as noise. For example, someone decided to increase the quote by 50 points, but in the process of increasing the price moved down by 1-5 points, i.e. small players were pulling it down. So, the signal from the big uncle distorted and as a result, the noise in the system was generated. So it is like this.
Aleksey, I understand your answers.
If you allow me to ask some more questions.
"Noise can be generated by imperfections in the communication channel, for example."
As a hypothesis, yes. But it needs confirmation to be accepted as an argument or as a component of the overall study. Right?
"In that sense, it's likely that 99% of the quotes are pure signal, another 1% are distortions by filters on the side of the quote providers."
I'm in this business. I know it from the inside. Assumption is possible, but not in the stated numbers, not everywhere, not by everyone and not everywhere.
I.e. if you want to accept it as noise, it has to be reliably captured, calculated and then factored into the specific and model and application environment.
"But, it's possible that you could single out some very big players and take their behaviour as a determinant, and then the fiddling of everyone else could be perceived as noise."
Maybe. But that would require proving that the "big players" influence price, then establishing in what proportions. Especially in terms of OTC markets this is unlikely to be possible. But as a hypothesis, sure. Just have to remember to prove it in order to apply ;) Otherwise it would also be impossible to "write off" anything to such players.
"For example, someone decides to raise the quote by 50 points..."
It's complicated... First of all, there are no actions on the market that reflect the phrase "someone decided to raise the price by 50 points". Yes, to buy on the upside. But that's not what you mean - you believe that markets are formed by someone else's desire to simply increase the price by 50 points. That requires proof. Preferably not similar to the various popular conspiracy theories of market makers, DTs and others.
Figures, facts.
"But in the process of growth the price also moved down by 1-5 points, that is, small players were pulling it down, thus the signal, sent by the big man, was distorted and as a result the noise in the system was generated".
In other words - the "small" players are noise? What then about the fact that the "bigs" can neither sell nor buy as long as the "smalls" don't take part in the transaction? There are no counterparties to the big small or medium players and nothing can be done, no deal.
Understand? You can't look at a quote in isolation from its execution, lest you get the above ... not an argument, but at least a premise.
Have you not seen the consequences of big bank traders' mistakes? A man sold 1 yard instead of buying, within seconds he was flooded, only a hairpin was left on the chart. Its length is the shorter the more liquid the market is. If you are not at the interbank terminal, you are not able to understand what happened. You can assume that the client drew (it happens), you can assume that the system failed (it happens), and you can assume everything else, but you can assume one thing and filter out another.
"Correct in analysis rather than trading, maybe?"
In trading
Alexei, I understand your answers.
If I may ask some more questions here.
"Noise could be generated by an imperfect communication channel, for example."
As a hypothesis, it's fine. But it needs confirmation to be accepted as an argument or as a component of the overall study. Right?
"In that sense, it's likely that 99% of the quotes are pure signal, another 1% are distortions by filters on the side of the quote providers."
I'm in this business. I know it from the inside. Assumption is possible, but not in the stated numbers, not everywhere, not by everyone and not everywhere.
I.e. if you want to accept it as noise, it has to be reliably captured, calculated and then factored into the specific and model and application environment.
"But, it's possible that you could single out some very big players and take their behaviour as a determinant, and then the fiddling of everyone else could be perceived as noise."
Maybe. But that would require proving that the "big players" influence price, then establishing in what proportions. Especially in terms of OTC markets this is unlikely to be possible. But as a hypothesis, sure. Just have to remember to prove it in order to apply ;) Otherwise it would also be impossible to "write off" anything to such players.
"For example, someone decided to increase a quote by 50 points, but in the process of increasing the price also moved down by 1-5 points, i.e. it was pulled out by small players. Thus, the signal sent by the big man distorted and as the result the noise in the system was generated. So there you go."
It's complicated... First of all, there is no action in the market reflecting the phrase "someone decided to raise". Yes, to buy in anticipation of a rise. But your point is different - you allow markets to be shaped by someone else's desire to simply raise the price by 50 points. That requires proof. Preferably not similar to the various popular conspiracy theories of market makers, DTs and others.
Figures, facts.
"But in the process of growth the price also moved down by 1-5 points, that is, small players were pulling it down, thus the signal, sent by the big man, was distorted and as a result the noise in the system was generated".
In other words - the "small" players are noise? What then about the fact that the "bigs" can neither sell nor buy as long as the "smalls" don't take part in the transaction? There are no counterparties to the big small or medium players and nothing can be done, no deal.
Understand? You can't look at a quote in isolation from its execution, lest you get the above ... Not an argument, but at least a premise.
Have you seen the consequences of mistakes made by traders of large banks? A man sold 1 yard instead of buying, within seconds he was flooded with it, only a hairpin remains on the chart. Its length is the shorter the more liquid the market is. If you are not at the interbank terminal, you are not able to understand what happened. You can assume that the client drew (it happens), you can assume that the system failed (it happens), and you can assume everything else, but you can assume one thing and filter out another.
Thank you, it's quite clear. I've actually never tried to quantify the influence of the big players directly affecting the market, the influence of the small players. It turns out, from your explanations, that the strong factors are made up of the influence of players of different sizes. It's more complicated, I don't even know how to describe such a process, although I've thought about stochastic modeling of impact of different players on price movement (but the idea did not go further than that).
Nevertheless, the problem of choosing the timeframe, where there is less noise, remains and needs to be researched. I just came across the clustering of volatility on intraday frames in this topic from the very beginning, and on larger ones as well. And that makes it very difficult to assess the impact of history on the current moment. I did not manage to devolatilize the process completely.
Thank you, that's quite clear. I actually never tried to quantify the impact of the big players directly affecting the market, but the impact of the smaller players. It turns out, from your explanations, that the strong factors are made up of the influence of players of different sizes. It's more complicated, I don't even know how to describe such a process, although I've thought about stochastic modeling of players of different sizes in price movement (but the idea did not go further than that).
Nevertheless, the problem of choosing the timeframe, where there is less noise, remains and needs to be researched. I just came across the clustering of volatility on intraday frames in this topic from the very beginning, and on larger ones as well. And that makes it very difficult to assess the impact of history on the current moment. I did not manage to devolatilize the process completely.
"Nevertheless, the problem of choosing a timeframe where there is less noise of any kind remains and needs to be researched."
Once the noise is identified, yes. Perhaps we should look for a timeframe or something. Before the "noise" is identified - it's a hypothesis. And you can't build a proof of one on the unprovenness of the other, I guess.
DDFedor says the same thing : The presence of "noise" depends on the research methods. You have to decide on the "method", and then probably talk about the "noise".