Dependency statistics in quotes (information theory, correlation and other feature selection methods) - page 35
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
That's the kind of answer the question was meant to give. There is no noise.
The noise is the difference between the quotes in the terminal and the quotes from the provider. And there is a clear out-of-market.
There is no other noise.
1. And what will the detection of deviations between the vendor quotes and the quotes in the trading terminal do? They are there - the DCs have never concealed it. Will you take offence at the DC and go interbank with $1,000?
What is the question - go to yahoofinance, download quotes of any instrument and make sure that your TS is losing (or not) on any quotes.
2. I do not know any model that does not give deviations, ie with 100% accuracy determines the price of the instrument based on certain factors. Deviations are noise.
Noise is the difference between the quotes in the terminal and those coming from the supplier. And it's clearly off-market.
There's no other noise.
Thank you. (chuckles) So it's clear to me what we're talking about. But is your definition shared by the topstarter as well?
Then I will spell out a few platitudes, with your permission, to ask the following questions:
1. There is an exchange market - its quotes are common to all traders and the chart from the exchange is the same on all terminals. However, unauthorised quotes appear there too, because of failures. Which are removed with different speed - when almost immediately, when at the end of the trading session. So there can be no noise from your definition.
2. there is an OTC market - its quotes have differences, sometimes principled, sometimes not. "Principled" is, of course, relative and depends entirely on the methods of analysis and trading. Some people do not mind a dozen points added to the history retroactively, others do not mind an extra pip on the current bar. But this pricing is a natural characteristic of the OTC market. Therefore, it should be taken into account from the beginning.
It is a given.
For the second case, if you are an OTC market participant, it is not possible to understand and detect the difference between quotes. Firstly, "suspicious" quotes are not that common, and secondly, you must have a machine that is able to distinguish between "friend or foe" quotes in real-time data flow. Moreover, it is not enough to detect such a quote, you have to do something with it. And here the problem is different - you can not include the quote in your analysis, but you must consider that the counter-quotes (a brokerage company, for example) will take it into account independently of you and will execute an order or any other necessary action at this level.
Hence the question - what is the sense in catching a pair of "left" ticks out of millions of ticks a day? The second question - even if you have managed to catch two "left" quotes per day, what will be the use of them?
P.S. TheXpert questions, of course, are not addressed to you personally. You just answered and I kind of continued the dialogue with you;)
The top starter is worried that his topic has been ruined. Meanwhile, I, for one, wanted to understand his conclusions.
"I don't know which TFs are more noisy..."
Alexey, what is the physical meaning of noise on the chart? Can you describe the nature of it? On any chart, if you don't mind, can you show me where the noise is?
On your fingers, in general, to help me understand, if possible.
Well, sorry, I didn't see that question in your posts with the charts, and it wasn't there. But recently someone has been terribly trolling (by picking up old threads), but then the posts made no sense at all, while you still have the prediction results.
I really don't know where the more noise is. Noise is usually the difference between the deterministic component of the signal and the original signal. Noise can be calculated by applying a linear regression model or a non-linear regression model. And the noise level will always be different depending on the quality of the predictive model. Noise from linear autoregression will be the same, while noise from multilayer NS will be different. So, in essence, we are talking about model noise on the initial quotient data. Do you agree with this or not, if not, why not?
As for, "No other noise." (TheXpert), I cannot agree. Prove that there is no deterministic function in the quotes, or that there is and perfectly describes all data points, then yes, but otherwise it's a unsubstantiated claim.
Well, sorry, I didn't see that question in your posts with graphs, and it wasn't there. Someone recently trolled in this way (bringing up old topics), but then there was no point in the posts at all, while you have the results of forecasting.
I really don't know where the more noise is. Noise is usually the difference between the deterministic component of the signal and the original signal. Noise can be calculated by applying a linear regression model or a non-linear regression model. And the noise level will always be different depending on the quality of the predictive model. Do you agree with that?
As for, "No other noise." (TheXpert), I cannot agree. Prove that there is no deterministic function in the quotes, or that there is and perfectly describes all data points, then yes, but otherwise it's a unsubstantiated claim.
Noise is great. Noise is a profit. Noise should and should be traded. The only question is the choice of model.
If the model "tends" to the price reducing the deviation, then there will be no profit. If the price tends to the model - it's a grail. (Unless, of course, we are talking about pair or multiple trading - there one earns on "collapsing" spread no matter what moves)
That's the kind of answer the question was meant to give. There is no noise.
Noise is the difference between quotes in the terminal and quotes from a provider. And it's a clear out-of-market.
There is no other noise.
Noise=Signal - Useful_signal :)
In our case "signal" can be replaced by "quotes".
Since usefulness is a relative property, so is noise.
Well, sorry, I didn't see that question in your posts with graphs, and it wasn't there. Someone recently trolled in this way (bringing up old topics), but then there was no point in the posts at all, while you have the results of forecasting.
I really don't know where the more noise is. Noise is usually the difference between the deterministic component of the signal and the original signal. Noise can be calculated by applying a linear regression model or a non-linear regression model. And the noise level will always be different depending on the quality of the predictive model. Noise from linear autoregression will be the same, while noise from multilayer NS will be different. So, in essence, we are talking about model noise on the initial quotient data. Do you agree with this or not, if not, why not?
As for, "No other noise." (TheXpert), I cannot agree. Prove that there is no deterministic function in quotes, or that there is and perfectly describes all data points, then yes, but otherwise it's a unsubstantiated claim.
Alexey, let's forgive each other right away and move on;)
I've already heard about the trolls. I have shown in another thread that I do belong to "...". I have something to say and something to listen to.
On topic.
"Noise is usually the difference between the deterministic component of the signal and the original signal."
Here, for example. Let's harmonise the terms. The deterministic component is the trend? And the raw signal is ? Quote?
Here, for example. Let's agree on the terms. Is the deterministic component a trend? And the raw signal is ? Quote?
OK, peace and friendship and bubblegum.
Now to the discussion.
The initial signal is a quote, but we also roughen that down to timeframes, or try to squeeze something out of ticks (but that's the lot of the brave).
But the deterministic component in my book is not a trend. It is a predictive model, and its graph can be anything: straight or curve. And this model can also be non-linear and have an arbitrary level of complexity.
PS: I haven't heard what multipoint is :)
Noise=Signal - Useful_Signal :)
In our case, "signal" can be replaced by "quotes".
As utility is a relative property, so is noise.
Thank you.
Observe the down-trend:
MTS 60 min.
Can you show on the graph where the noise is?
)))))))))))))))))))
this guy really makes me happy - a childlike, direct view of the world!
Everyone was building models, doing calculations, but just draw a line from the ruler on the graph
+1000