Matstat Econometrics Matan - page 20

 
Aleksey Nikolayev:

The standard approach in optimisation is to multiply the target by minus and the maximisation turns into a minimisation (and vice versa).

I already tried to explain to you that if errors are Gaussian distributed, then ISC==MLE. If the errors are distributed by Laplace, thenMNC==MLE==MLE method of least moduli. You can figure out for yourself the type of error distribution whenMLE==MLE by Huber.

In experiments, the type of error distribution is either known by some additional consideration, or it is chosen experimentally (usually in the form of a suitable loss function).

Apparently I didn't get it the first time, now I do ))
Thank you.

 
Aleksey Nikolayev:

Impressed by your knowledge. Are you making money in forex? Do you have a personal website? Do you take money into management?

 
pribludilsa:

Impressed by your knowledge. Are you making money in forex? Do you have a personal website? Do you take money for management?

Thanks, but the knowledge is so-so - only the basics, but more or less solid.

I do not know how to earn. Sometimes even forex)

I do not have pammers and signals, as I work alone (I prefer it that way). I am sure that it is practically impossible to create a system that scales well in terms of capital on my own.

 
Persistent or antipersistent increments can also be randoms, so Hurst doesn't say anything about predictability either. It makes no difference whether it is different from SB. SB is just a special case of randomness, "normal" randomness. In general, the shape of the distribution says nothing about predictability, I don't know what to look for there.
 
Roman:

To continue the topic.
A lot of people here mention data thinning.
There is a method called PCA (Principal Component Analysis), which isone of the main ways to reduce the dimensionalityof the data while losing the least amount of information.
Has anyone studied this method? Any conclusions about its applicability?
I know
that asset selection is thinned by this method. But I don't know if a dataset can be thinned by it without losing dimensionality.

As I see it, the main problem with thinning is dimensionality reduction. That is, the sample becomes a different size.
In a simple case, there are recommendations from the same lecturers of universities, not to throw out an element from a set, and replace it with an average value of neighboring elements for example.
At least this is how outliers are removed, in the simple approach. But with a caveat that there are other approaches, which are not explained.
Therefore PCA as a thinning idea, can be well investigated.

P.S. Clever site links, even finds articles on a similar topic
Oh how ))

A useless exercise, on new data the components will "jump" if it is not a sinusoidal

i.e. PSA is a way of fitting on a subsample, and a linear one at that.

it's not a way to find a pattern.

 
Maxim Dmitrievsky:

Futile exercise, with new data the components will "jump" if it is not a sine wave

So PSA is a way of fitting on a subsample, and a linear one at that.

it's not a way to find a pattern.

Maxim, I haven't delved into the method so far, I can't say anything.
I just watched a recorded seminar organized by Moscow Stock Exchange,
where brokers and all sorts of researchers like geeks, etc. shared their experiences, presentations, etc.
There I heard about this method, that it is used to select assets for further models.
He showed that this method works and gives some kind of growth.

I read his article as an idea, may be it is not working.
But anyone may be interested and find the profit.



Continue on youtube.

 
Roman:

Maxim, I haven't got into this method yet, so I can't say anything.
I just watched a recorded seminar organized by Moscow Stock Exchange,
where brokers and all kinds of researchers like geeks, etc. shared their experiences, presentations, etc.
There I heard about this method, that it is used to select assets for further models.
He showed that this method works and gives some kind of growth.

I also heard this method as an idea, maybe it is not working.
But who is interested, may find the sense of application.


Drimmer has an article here, applying PSA to build portfolios. But later he recommended everyone to go to the factory :)
 
Maxim Dmitrievsky:
Drimmer has an article here, applying PSA to build portfolios. But later he recommended everyone to go to the factory :)

Maybe the recommendation was because no one understood anything ?
;))

 
Roman:

Maybe the recommendation was because no one understood anything ?
;))

For quite objective reasons. A stationary portfolio only works in the moment, on new data things break down without the right skill
 
Roman:

Maxim, I haven't got into this method yet, so I can't say anything.
I just watched a recorded seminar organized by Moscow Stock Exchange,
where brokers and all kinds of researchers like geeks, etc. shared their experiences, presentations, etc.
There I heard about this method, that it is used to select assets for further models.
He showed that this method works and gives some kind of growth.

I read his article as an idea, may be it is not working.
But anyone may be interested and find the profit.



Continue on youtube.

Don't talk like that, bro.