Dependency statistics in quotes (information theory, correlation and other feature selection methods) - page 44
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
OK, let's not talk about SB here.
Returning to my topic, I was able to get positive forward tests on some financial series (DJI D1, I think it was) by calculating the mutual information between the quotes, but I cannot vouch for the fact that it was not random. And, in forex, I have not been able to get such results.
OK, let's not talk about SB here.
Too bad, it promises to be an interesting evening;)
But, you are the owner of the thread, it's up to you.
You trade the same break of an extremum. The model implies that stop losses will be triggered behind extrema as well as other momentum trades. These trades will be useful in terms of the model and all other trades will be noisy. Not only those that are against an open trade, but those opened for other motives than those included in the model. If the useful ones are very small compared to the noisy ones, then the effect will be a splash))) For other models these "noise" trades can be useful.
Noise is what a particular model does not account for. There is no such thing as absolute noise, but rather a lack of information on price formation processes. Who doesn't like the word noise, can call it "unaccounted for" for example))) But in essence it will act as noise in the background of a useful signal in physics.
Z.U. it's all about "noise" after entering a trade. The same is true before entering a trade.
Indeed. What a great company. Let's talk about interesting things. But not about the noise.
Yes, it's about time. I'd like that very much :)
I, by the way, want to go to Minsk in the next 2-3 months, just for fun. We can cross paths.
Those who have positive results on SB fall into three categories -- cheaters, dumbasses and freeloaders.
If you're itching for SB, you've got the wrong topic, buddy.
You're itching to bring up such topics, or jealous that you do not know how, or what, I do not know. Just where the topic is not mistaken, they disappear. You teoretyugamy just want to stick to the word all is clear that it means a pseudo-SB, although what is pseudo - you also can not explain. Simply in the nature idial SB does not exist, what then you with the theories name people dull-witted, though such are more you when attributing in calculations an abstract idyllic casual process to real conditions in which this casualty does not exist at all. You cover up your incompetence with insults.
Hello.
It's been stalled for a while, but it's been very interesting. I'll try to revive it.
Please don't kick me too much, I will give my thoughts on the subject and the issues raised.
About the noise:
As I'm a technician, I'm talking like a technician.
In radio engineering, noise is defined by the criterion of usefulness. All that is useful is the desired signal; all that is not is noise. Therefore the main task at the noise definition stage is to define the criterion of usefulness (excuse the tautology).
Examples are simple. You are sitting at a lecture and are trying to understand the material being presented, and your neighbour is sitting next to you listening to music. He is wearing headphones but you can hear the music and it constantly distracts you and prevents you from concentrating. To you it is noise. To your neighbour it is a useful signal and the teacher's chatter is noise.
Now we move smoothly to the criterion of noise assessment. In order to separate it from the useful signal, it is sufficient for the useful signal to exceed a certain value, which is called a threshold. The value of this utility is determined by the signal-to-noise ratio.
Anything above the spread is useful for trading. (At least in DC on forex). Therefore, the criterion of usefulness is the exceeding of the current quote value over the spread relative to the previous tick. When entering the position we will be charged the spread anyway. This is the noise threshold. Therefore, the task is to enter the position in such a way that this threshold barrier is exceeded at the next tick.
It is logical to go further and consider the investment horizon. The task then boils down to the following - on the investment horizon to gain:
the minimum task - break-even
The maximum task - the planned profit.
Everything is simple.
Now about the topic of the topic.
The theory of information was created as an application to the theory of signal transmission and was taught as "statistical theory of communication". Therefore, when reading the thread I often wondered - whether I have forgotten everything, or the participants are not quite understand what they want to dig up.
Statistical communication theory describes methods of coding information and recognition of this information on the receiving end. Recognition methods are optimized.
Joint partial problems - error recognition, error correction, encryption and decryption - are also solved. Various codes are used for error correction and recognition - BH (Bloch-Harkevich), Shannon, BCH (Bowes-Chowdhury-Hoekvingham) and many, many others. I think the number of them has grown exponentially since I graduated. So, all codes are built on the property of EXTREME. And this is the main thing in information theory. Exactly for description of redundancy the notion of entropy was introduced, which is equivalent to information quantity.
What is redundancy is very easy to understand. Everyone will understand the word "...ussian", although the first letter is missing. In general, language is redundant in its very essence, which is why it is easy to restore an original word or phrase in case of distortion.
About alphabet.
The alphabet is defined by information "patterns" consisting of primitives. In information theory these patterns are set automatically. For example, binary number system has two primitives. If we set an alphabet with length of 10 primitives, our alphabet will consist of 1k words (2 to the power of 10).
Now let's move on to the market. The task is to create an alphabet. So we need primitives - also patterns. So a logical question arises - what do we take for consideration.
Price?
OK. How do we create the alphabet for price? We take the largest price spread in the symbol's history and divide it by the smallest possible counts, i.e. points. Altogether we have two primitives (0 and 1) and the alphabet, the number of signs (states) of which will be two to the power equal to =Waves\number of points.
We introduce correction - we don't have tick history, only minutes. Then we look for the smallest minute in terms of its span, take its span value as one and get the alphabet again.
Time?
Time should be considered based on the understanding that every trader has an investment horizon.
We must conduct the analysis based on this horizon and consider everything that does not fit into this horizon as noise. Suppose we want to work on the watch. Then we need to adjust the price alphabet, based on the value of the minimum one-hour candle in the history. We should take the value of the interval as a unit of time. The question is how to determine how many intervals to take, because they will determine the "time alphabet capacity". The solution suggested by respected multipoints - consider intervals defined by price extrema. Again we have two primitives (0-beginning of the interval, 1-end of the interval) and the alphabet set by price extremums (time lags).
That's it.
Who is thinking clearly... ;)
Thank you unknown (or guided?) friend.
Let's talk about the topicstarter's first post. Did you find any incurable errors there?
The alphabet is defined by information "patterns" made up of primitives. In information theory, the patterns are set automatically. For example, binary number system - two primitives. An alphabet is 10 primitives long, so our alphabet consists of 1k words (2 to the power of 10).
Then we gradually pass to the market. The task is to create an alphabet. This means we need primitives, or patterns. So there is a logical question - what do we take into consideration?
Price?
Okay. How to form the alphabet for price? We take the largest swing in price in the history of the instrument and divide it by the smallest possible counts, i.e. points. Altogether we have two primitives (0 and 1) and the alphabet, the number of signs (states) of which will be two to the power equal to =Waves\number of points.
We introduce correction - we don't have tick history, only minutes. Then we look for the smallest range of a minute, take the value of its range as one and obtain the alphabet again.
Firstly, it is the price returns, not the price itself, that are investigated. This is the object of the study. Looking for additional arguments to show that price cannot be investigated in the same way.
Second, it is already done: we divide the distribution of returns into quantiles (the topicstarter did it differently). Each quantile is a letter of the alphabet. There can be anywhere from 2 to infinity, but more than 40 is probably impractical.
It is logical to go further and consider the investment horizon. The objective then boils down to the following - on the investment horizon to get:
1.a minimum objective - break-even
2. The maximum task - the planned profit.
The task is simple.
1. get a trading system in which the order will be at breakeven, say, 7 out of 10 times is not as difficult as it seems in these TC balance chart has a view of a series of rectangular triangles (earned, earned, lost all that earned, earned, ...... ). The minimum target is the minimum risk. It is better to consider real "task minimum", clearly defined profit/loss ratio ---> takeprofit/stop loss, otherwise another study_of_the_spherical_horse_in_the_vacuum
2. profit is good, but I have never met a FC, in which there is no series of losses, only profitable, the so-called "Grails" (I know, I have seen the grailers, but they do not have series of losses, but immediately lose their deposits ))))). ). I mean to say that the maximum target is to reduce the number of consecutive losing trades.
Everything is not simple at all, if it were simple - the Russian Pension Fund would earn on the markets, instead of losing ((