Machine learning in trading: theory, models, practice and algo-trading - page 2718

 
Maxim Dmitrievsky #:
Macroeconomics can you handle it?)) pulling out from under the wheels of Econ?

Macro macro macro economics, just macro is too shallow)))))

 
Maxim Dmitrievsky #:
Can you get the macroeconomics out from under the wheels of the Econo?

Is there any way to do it through the back porch? Through ingenuity, invention, and artificial intelligence? No?

 
mytarmailS #:
Is it possible to create a successful algorithm without building a more or less working market theory or a market model, or a model of the local market situation?

The first two are not realistic yet) But local models, or rather models of price behaviour are quite possible, which is being done today, but again it can't relate to reality, because there are no connections, there is nothing to connect with. And inventing sort of logical connections is a finger and sky).

 
sibirqk #:

Is there any way to get in through the back porch? With ingenuity, invention, and artificial intelligence? No?

Well, that's what we do.

 
Maxim Dmitrievsky #:
Macroeconomics for intraday trading?) get the Econ out from under the wheels?

macroeconomics for intraday trading? genius!!!!

Valeriy Yastremskiy #:

The first 2 are not realistic yet)

There are people who do, I even know one... but he is a C.T.N. and has been researching the market and AI algorithms for 20 years....

Valeriy Yastremskiy #:

But local models, or rather models of price behaviour are quite good.

Well, that's where I stopped myself, otherwise combinatorial explosion.

Valeriy Y astremskiy #:

again, it can't have anything to do with reality, as there are no connections, nothing to connect with.

Well, if there are no links, then nothing will work and there is no need to do it, but where did this statement come from? Any research?

Valeriy Yastremskiy #:

And inventing sort of logical connections is a finger and sky).

And if it's a computer making it up?

It's not a finger in the sky, it's 100 fingers in the sky per minute.

 
Valeriy Yastremskiy #:

And coming up with sort of sort of logical connections is finger and sky)

then we can introduce a new event - dependence on the finger gesture.

100 times a minute to receive signals, until a favourable one falls out.

 
I suggested to focus on informative features, there is a method to obtain and select them from the time series (along with the targets). So far no one has responded
And, judging by the mood, people are busy with something very global and the meaning of the method will not be adequately understood, not even constructive criticism.
Oh, and nobody knows python.

Also no one has taken seriously the idea of filling the stationary series with time series information to get the most informative features.
 
mytarmailS #:

macroeconomics for intraday trading? genius!!!!

There are people who do it, I even know one... but he is a C.T.N. and has been doing market research and AI algorithms for 20 years....

Well, that's where I've stopped myself, otherwise it's a combinatorial explosion.

Well, if you don't, then nothing will work and you don't need to do it, but where does this statement come from? Was there any research?

What if a computer comes up with it?

It's not a finger in the sky, it's 100 fingers in the sky a minute.

Formalisation of events from real life must be understood, only then you can understand what is being modelled. A model is a formula data, and some comparisons of data to some events, factors, actions in real life. The model of minority game also gives a result similar to the price series. But it is definitely not a market model.)))))) In the market there are still a lot of variables for the model, these are states with laws, enterprises, commodities, traders, weather conditions. If there are formalisations of some variables, they are truncated models. They have been trying to model the harvest from the weather for a long time, but weather modelling is not very good even for a couple of weeks).

In general, the way of generating models and comparing them with the market can and will give the result that some model will behave similar to part of the market, has a place)))))) But it is too complicated and not guaranteed way. Informative attributes and their selection is more simple, maybe not quite right, but for today's capacities.

 
Valeriy Yastremskiy #:

In the market there are still many variables for the model, such as states with laws, enterprises, commodities, traders, weather conditions. If there are formalisations of some variables, they are truncated models. They have been trying to model the harvest from the weather for a long time, but weather modelling is still not very good even for a couple of weeks).

These are all long-term factors that do not affect the market every minute....

All we need to trade is to guess a good entry point, and this is time - seconds-minutes and the price itself....

So all of the above factors can be neglected, it's rubbish to study weekly reports, which come out a week late to open a pose for 10 minutes, leave this rubbish for the relevant specialists who have brains NOT turned off....

Valeriy Yastremskiy #:

In general, the way of generating models and comparing them with the market can and will give the result that some model will behave similar to a part of the market, has a place)))))

because the market is not a time series....

Remember I gave you the first one, levels in sequences? You need something like this, but 100 times more powerful in terms of complexity of rules.

 
Macroeconomics is not weekly reports, again the level of knowledge is below the plinth. You can look at both long-playing economic factors and intraday factors (e.g., connection with indices), combinations of factors. Big events do not affect at once, but have a duration, including speculators orientate on them to understand the basic dynamics.

You cannot, of course, omit the geopolitical component as a driver for the rest of the processes

Then you will finally be able to use the term Event. If you can single out a single event from the whole stream and describe it with indicators.

And the non-time series thing is just another bullshit without proof.