The question is not one of programming, but rather of philosophy.
It is well known that noise is very good to measure when the signal characteristics are known. In our case it is not entirely clear what is noise and what is the signal.
For scalpers it is one thing, for intraday traders it is another thing, for short term traders it is another thing. But they catch different signals. And for these categories and the concept of noise is very different.
When we look at the chart, without any indicators we can see where the noise is and where the signal is. Especially on the history. The more experienced may understand even in the real time, where the signal exceeds the noise (I'm not saying that the deal at that moment will be successful, but the probability of such an outcome is higher).
I tried to do something in this direction. Say, regression lines, regression curves are the same lines, only at each point the derivative is equivalent to a line. And relative to them, count the noise. However, the human brain determines all this, if not faster, then much better. And where an automaton decides to enter a trade, it seems to a human not to do it at all, and vice versa, by the way, is also not uncommon. When testing, I usually look through most trades, including charts, and sometimes think - why did he go there? Although, yes, I told him so myself:).
So the thought came up that before we measure anything, it's a good idea to formulate what it is we're measuring. I suspect that when we look at a graph, we adaptively change our concept of noise every time, depending on the situation. That is, we intuitively think of several options, and choose the one that suits us at that particular moment. And we average each time according to a different criterion - we draw several lines in our head, and choose the one we like at the moment.
It is suggested to discuss what is noise and how to deal with it, in the sense of measuring its characteristics.
This is a slightly confusing question... The counter-question is why do you need to measure noise? In simplistic terms, anything that cannot be identified as a useful signal is noise. But again, why would you want to measure it?
Professional habit. All in all, I don't see anything confusing. The usual staged start to a discussion, if anyone is interested. At the same time we will find out if anyone is interested in the topic. You can't try, Pavrentiy Pavlovich (c) is right.
I build systems I adhere to the concept of noise and signals in the market. In brief, trades are executed when the signal level exceeds the noise level - a type of thresholding device common for control systems. It is for such systems that noise measurement and distinctive noise and signal criteria are needed.
If someone on the forum is doing a similar approach, why not discuss it.
The question is not one of programming, but rather of philosophy.
It is well known that noise is very good to measure when the signal characteristics are known. In our case it is not entirely clear what is noise and what is the signal.
For scalpers it is one thing, for intraday traders it is another thing, for short term traders it is another thing. But they catch different signals. And for these categories and the concept of noise is very different.
When we look at the chart, without any indicators we can see where the noise is and where the signal is. Especially on the history. The more experienced may understand even in the real time, where the signal exceeds the noise (I'm not saying that the deal at that moment will be successful, but the probability of such an outcome is higher).
I tried to do something in this direction. Say, regression lines, regression curves are the same lines, only at each point the derivative is equivalent to a line. And relative to them, count the noise. However, the human brain determines all this, if not faster, then much better. And where an automaton decides to enter a trade, it seems to a human not to do it at all, and vice versa, by the way, is also not uncommon. When testing, I usually look through most trades, including charts, and sometimes think - why did he go there? Although, yes, I told him so myself:).
So the thought came up that before we measure anything, it's a good idea to formulate what it is we're measuring. I suspect that when we look at a graph, we adaptively change our concept of noise every time, depending on the situation. That is, we intuitively think of several options, and choose the one that suits us at that particular moment. And we average each time according to a different criterion - we draw several lines in our head, and choose the one we like at the moment.
It is suggested to discuss what is noise and how to deal with it, in the sense of measuring its characteristics.
Not so much. There is actually no noise in the market. The only noise there is is sampling noise, which arises from the time sampling of the graph. If the data is not time sampled, then it becomes apparent to the eye that there is no noise, there are movements of different sizes. Small movements make up large ones and they all look alike.
The market is a complex dynamic system and it is impossible not to make noise. If we consider that this system contains positive feedback (information from the market to the trader), then the market input noise should also increase. If we place the market in an info-vacuum (no external influences, news, etc.), we see the system noise in its pure form. Let's say that flat is sort of the market noise. And this is the system output signal.
Market noise represents something similar to random walks, like a Wiener Random Process. At least many of the characteristics are the same. Self-similarity, among others.
The market as a complex dynamic system cannot help but make noise. If we consider that this system contains positive feedback (information from market to trader), then market input noise should also increase. If we place the market in an info-vacuum (no external influences, news, etc.), we see the system noise in its pure form. Let's say that flat is sort of the market noise. And this is the system output signal.
Market noise represents something similar to random walks, like a Wiener Random Process. At least many of the characteristics are the same. Self-similarity, among others.
If you put the market in a vacuum, the price will stop because no one will make any transactions and there will be no noise.
The market can indeed be thought of as a system covered by positive feedback, and it would be that simple if.... Imagine an amplifier that has one feedback with rigid parameters, what would happen? It would self-excited and harmonic oscillations would occur. Now imagine an amplifier with a million feedback circuits! with different parameters! And it is not always a million, now there may be 2 million, now there may be only 1000, and they all have different characteristics, different depth, different delay times, different transient times, and plus all these parameters float in time for each individual link. In this case there will never be a harmonic signal in the market, each movement can be strengthened or weakened in a completely unpredictable way, so it turns out that there is no noise in reality, it is all informative movements. You can't apply circuit theory or signal theory or filtering out noise, because the market consists of noise, and it is self-synchronizing on different reference intervals.
Why do we filter noise so easily in radio engineering? Because we know the signal that we need to isolate, all that is not a signal is noise. Our signal has quite definite characteristics and we know how to calculate it by which indirect or direct signs. In the market the noise itself is the signal, in fact there is no noise or signal, there is only fluctuation. There is nothing to filter.
Although some noise is present there, it is quantization noise, each deal is performed with finite exact volume, noises occur, but then these noises transform into movements, as I wrote above.
If you put the market in a vacuum, the price will stop because no one will make transactions and there will be no noise.
Positive feedback only at certain pvr parameters, leads to generation and self-excitation. Strictly speaking, we have only one feedback, not many, which goes to the input of the system, but the input filters are all different, and, yes, change over time. Besides, everyone has different filters for external influences, form and amplitude of response to these influences. I have been running such a model for a few years now. Mostly on a qualitative level.
So is there noise or not? If there is noise on the market, it is the signal, in fact there is no noise and no signal, just fluctuations. There is nothing to filter.
Let's say consolidated movements are not noise- the deviation from the average is negligible there. And the shape of the system's response is quite similar to the response of most systems with a slight overshoot. And in a flat, noise is quite similar to a tremor or wandering around the mean (yaw), maybe with drift. This response is also common to many control systems.
When we look at the chart, without any indicators, we can see where the noise is and where the signal is. Especially on history.
Let me ask you, then what are you doing here? Go to the major stock exchanges to crush the global financial system.
Let me ask you, then what are you doing here? Going to the major stock exchanges to crush the world's financial system.
The question is not one of programming, but rather of philosophy.
It is well known that noise is very good to measure when signal characteristics are known. In our case it is not very clear not only what is noise but also what is signal.
For scalpers it is one thing, for intraday traders it is another thing, for short term traders it is another thing. But they catch different signals. And for these categories and the concept of noise is very different.
When we look at the chart, without any indicators we can see where the noise is and where the signal is. Especially on the history. The more experienced may understand even in the real time, where the signal exceeds the noise (I'm not saying that the deal at that moment will be successful, but the probability of such an outcome is higher).
I tried to do something in this direction. Say, regression lines, regression curves are the same lines, only at each point the derivative is equivalent to a line. And relative to them, count the noise. However, the human brain determines all this, if not faster, then much better. And where an automaton decides to enter a trade, it seems to a human not to do it at all, and vice versa, by the way, is also not uncommon. I usually look through most trades during testing, including those on charts, and sometimes think - why did he go there? Although, yes, I told him so myself:).
So the thought came up that before we measure anything, it's a good idea to formulate what it is we're measuring. I suspect that when we look at a graph, we adaptively change our concept of noise every time, depending on the situation. That is, we intuitively think of several options, and choose the one that suits us at that particular moment. And we average each time according to a different criterion - we draw several lines in our head, and choose the one we like at the moment.
I suggest to discuss what is noise and how to deal with it, in the sense of measuring its characteristics.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
The question is not one of programming, but rather of philosophy.
It is well known that noise is very good to measure when the characteristics of the signal are known. In our case it is not very clear not only what is noise, but also what is a signal.
For scalpers it is one thing, for intraday traders it is another thing, for short term traders it is another thing. But they catch different signals. And the concept of noise is very different for these categories.
When we look at the chart, we can see where is noise and where is the signal, without any indicators. Especially on history. The more experienced may understand even in the real time, where the signal exceeds noise (I'm not saying that the deal at this point will be successful, but the probability of such an outcome is higher).
I tried to do something in this direction. Say, regression lines, regression curves are the same lines, only at each point the derivative is equivalent to the line. And relative to them, count the noise. However, the human brain determines all this, if not faster, then much better. And where an automaton decides to enter a trade, it seems to a human not to do it at all, and vice versa, by the way, is also not uncommon. When testing, I usually look through most trades, including charts, and sometimes think - why did he go there? Although, yes, I told him so myself:).
So the thought came up that before we measure anything, it's a good idea to formulate what it is we're measuring. I suspect that when we look at a graph, we adaptively change our concept of noise every time, depending on the situation. That is, we intuitively think of several options, and choose the one that suits us at that particular moment. And we average each time according to different criteria - we draw several lines in our head, and choose the one we like at the moment.
It is suggested that we discuss what noise is and how to deal with it, in the sense of measuring its characteristics.