Machine learning in trading: theory, models, practice and algo-trading - page 3594

New comment
 
The problem with using macroeconomic indicators is about the same as with using price history - there are no clear enough correlations. They even come up with special names for the lack of clear correlations, like "PPP-puzzle".
[Deleted]  
Aleksey Nikolayev #:
I didn't see any success stories in the form of stats on their site).
Maybe it's a similar name. There was a popular one that opened pending orders before the news. It seemed to take into account the statistics of strong changes in the exchange rate.
 
Aleksey Nikolayev macroeconomic indicators is about the same as with using price history - there are no clear enough correlations. They even come up with special names for the lack of clear correlations, like "PPP-puzzle".
It seems to me that the problem is with the target for the market, not with the rest... that's why they don't find correlations with ZZ knee or candlestick colour etc.

The real target is most likely a multivariate function.
 
mytarmailS #:
It seems to me that the problem is with the target for the market, not with the rest... that's why they don't find dependencies with ZZ knee or candle colour, etc.

The real target is most likely a multivariate function.

I don't understand at all what is being discussed here....

For training you need a sample, at least 500. What period do they want to trade on? A day? Of course not, a week, no. A month? Well, maybe for a month you can find something from macroeconomic data. But you need 500 months to learn, more is better. But in reality there won't be 500 quality marks, but much less. I have about 10-20 times less marks than the sample. It's a real pain in the arse with the teacher.

Macroeconomics is rubbish.

 
mytarmailS #:
It seems to me that the problem is with the target for the market, not with the rest... that's why they don't find dependencies with ZZ knee or candle colour, etc.

The real target is most likely a multivariate function.
Imho, the real desired target is a probability distribution, which may well be multivariate. Formally, during classification (when the MO algorithm produces class probabilities) a probability distribution is constructed, but it is a discrete distribution that loses a lot of information about the probable price behaviour during a trade.
[Deleted]  
How to fix your target ones already wrote above. Good thing no one uses it but me :)

The easiest way to think of BP is as a set of different repetitive patterns. From the observer's point of view it is so.

The advantage of MO is that it's often not important to know what those patterns are. That is, it makes it easier to find them.
 
СанСаныч Фоменко #:

I have absolutely no idea what's being discussed here.....

You need a sample of at least 500 for training. What period do they want to trade on? A day? Of course not, a week, also not. A month? Well, maybe for a month you can find something from macroeconomic data. But you need 500 months to learn, more is better. But in reality there won't be 500 quality marks, but much less. I have about 10-20 times less marks than the sample. It's a real pain in the arse with the teacher...

Macroeconomics is rubbish.

The official science is just saying that macroeconomics is very bad at predicting exchange rates.

Imho, from a practical point of view, if macroeconomics improves the prediction even a little bit, it is worth using it in addition to prices. This is quite consistent with the idea of ensemble in the IO.

[Deleted]  
I counted about 7 sellers in the marketplace, who saw bots on articles about catbust. I wonder what is the total output from the market :)
 
Maxim Dmitrievsky #:
How to fix your target ones already wrote above. Good thing no one uses it but me :)

The easiest way to think of BP is as a set of different repetitive patterns. From the observer's point of view it is so.

The advantage of MO is that it is often not important to know what those patterns are. That is, it makes it easier to find them.

Well, everyone is going crazy in their own way here)

The approach closest to me is when it comes down to constructing probability distributions and looking for their deviations from what they would be if they were randomly wandering.

 
Maxim Dmitrievsky #:
I counted about 7 sellers in the marketplace, who saw bots on articles about catbust. I wonder what the total output from the market :)
It is necessary that they pay royalties to the author)
1...358735883589359035913592359335943595359635973598359936003601...3742
New comment