From theory to practice. Part 2 - page 114
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
Alas all. Everything is being swept clean now, manual scalping is dead, whoever is fastest feeds on everyone else. But that does not mean that one cannot act in the same way and compete with the best. But we need a different approach to research, not what many are doing, pushing Eurobucks in 100500 different classifiers in the hope for a miracle, or fitting in the tester, it's just a child's play.
Data, gentlemen, data...
I'm not ready to argue, because I'm not a hundred percent sure I'm right. I can only refer to Prado who says that it is necessary to check scalability of systems.
The sample mean MAY only be an EVALUATION of the MO, but only if a) the MO exists, b) the sample is independent, c) the sample is equally distributed. For a sinusoid, point (b) is exactly violated, because there is a rigid relationship between the elements in the sample.
In general, please do not call "expectation" that is called "the mean value of a function on the segment".
Like a frying pan...
Alpha
Thanks mate, that's interesting!
I understand mainly fundamental analysis is used, a stream of fresh data. Technical analysis of chart history is not particularly considered. Have I got it right?
I downloaded the brochure, I'm reading it.
Absolutely right. Oleg's matcad calculated a sample average. Because it can't calculate anything else for the sample. He (matcad, not Oleg) is dumb and does not know that it is a sine with MO=0. And the best estimate of MO for a sample is 0.
The mathematical expectation is a statistical characteristic of a random variable. A sine is not a random variable - it is a deterministic relationship.
I would probably agree with AK that thinning makes sense after all
I will provide the following screenshot as an example.
You can see with the naked eye how EAs are forced to make a wrong move:
I won't say what it is calculated, the main thing is the meaning. Although, in principle, it is from this data that the price is obtained ;)
It is a long and tedious wobbling of some price parameter in a wrong direction and gapping in order to jump where the market needs ;)
And only on М1 I got the calculation result that practically equals real and demo.
If the TF is higher, there is a fiasco.
For a sinusoid, point (b) is definitely violated, because there is a rigid relationship between the elements in the sample.
Alas all. Everything is being swept clean now, manual scalping is dead, whoever is fastest feeds on everyone else. But that does not mean that one cannot act in the same way and compete with the best. But we need a different approach to research, not what many are doing, pushing Eurobucks in 100500 different classifiers in the hope for a miracle, or fitting in the tester, it's just a child's play.
Data gentlemen, data...
Have you counted the costs of such goodies? Cost of equipment, cost of direct feeds, cost of collocation,etc.
Not to mention the cost of an FPGA specialist if you don't know anything about the technology yourself.
It's very expensive to maintain such a structure.
Of course, they are covered by the market, but the initial threshold of skills and technical support is very high.
Therefore, ordinary ordinary ordinary people, more or less understand where the fish is, use old tried and tested approaches.
Let it not huge profits, but it is there and stable. The point is not how you caught the fish, but that you caught it.
As far as satellite data is concerned, I studied this issue once. So the speed of this data is lower than Fibre optics.
Therefore satellite data is not suitable for HFT. For this reason, faster those on a short leg, or in the dark fiber.
On the alternative data, also thought about this subject, but did not look for an approach implementation, it is necessary to understand what we are looking for and from what.
This is a difficult task of data analysis, which again requires skills and not small. How many people have them? I doubt it.
Although there is an article on hubra as an example, on alternative data, to understand the process.
If Hearst is different from 0.5, you can make money.
For a"sine/cosine with a jumping or, to begin with, a stable period" Hearst will be close to 0 .
Have I answered your question?
I don't really need an answer. Almost all the answers were found thirty years ago :))) It's just a little straightening out. If you have to answer yourself, of course, so that it would be useful in the sense that the answer would then be applied in the market. That's actually what the question was for. I am not interested anymore :)))) (Hurst is in the wrong field, leave him alone).
A sample average MAY only be an EVALUATION of the MO.
Quite right. And this estimate has a variance which is inversely proportional to sample length. As the sample length increases, the ME tends to the ME of the population, i.e. in our case to zero.
Nobody forbids us to shuffle the sine values) the connection will disappear and the average will not change)
But it won't be a sinusoid anymore) And the violation of (c) will continue - the histogram will be different in different end segments if they don't coincide with the period. If you shuffle only inside periods, you will get just some generalized white noise)
PS. No, you can't get noise like that. And to understand what you get, you need a clear formalization of the mixing algorithm