Machine learning in trading: theory, models, practice and algo-trading - page 3216

 
fxsaber #:

I have a terminological misunderstanding, unfortunately.

Well, we live in the modern age of chatgpts :)

Bootstrap sampling is a statistical analysis technique that is used to estimate sample parameters by creating multiple subsamples from the original sample. This method estimates the variance and mean of a parameter, and constructs a confidence interval for the parameter. Bootstrap sampling can be useful when it is not possible to obtain a large sample or when the original sample is not representative of the entire population.

 

Almost always replaces forum conversations with inadequates. Bit dumbed down at the end though, perhaps insufficient context.


 
Further confirmation that we only have a small fraction of the data to analyse, which is essentially noise.

 
Forester #:
Further confirmation that we only have a small fraction of the data to analyse, which is essentially noise.

In essence, the point is that probabilistic uncertainty poorly describes real market uncertainty. This has long been no secret to economists, which was one of the reasons for the emergence and development of game theory. The problem is that game theory is still poorly developed compared to probability theory. Moreover, the lag is in the ideological part of the theory.

And the contrast between finance and industry in the video is a complete rubbish, of course. And "the imminent inevitable ruin of America" is a complete rubbish.

 
Aleksey Nikolayev #:

In essence, the point is that probabilistic uncertainty poorly describes real market uncertainty. This has long been no secret for economists, which was one of the reasons for the emergence and development of game theory. The problem is that game theory is still poorly developed compared to probability theory. Moreover, the lag is in the ideological part of the theory.

And the contrast between finance and industry in the video is a complete rubbish, of course. And "the imminent inevitable ruin of America" is a complete rubbish.

Back in Soviet science, apart from deterministic processes, stationary and non-stationary random processes, uncertain processes were considered - these are such random processes in which a person takes part. The most striking example is the random flow of passengers in the underground. Usually everything is perfectly well described by the theory of mass service, but if you puncture a balloon and shout "bomb", all stationarity flies into tatters.

All processes in economics belong to the class of uncertainty, and all attempts to even account for non-stationarity will ALWAYS fall back on the human factor, which in economics is called politics, which is known to be "the concentrated expression of economics".

I don't think that game theory can take into account the influence of politics on the economy in such a way to model an uncertain economic process.

 
СанСаныч Фоменко #:

Back in Soviet science, in addition to deterministic processes, stationary and non-stationary random processes, uncertain processes were considered - these are such random processes in which a person takes part. The most vivid example is a random flow of passengers in the underground. Usually everything is perfectly described by the theory of mass service, but if you puncture a balloon and shout "bomb", all stationarity flies into tartarus.

All processes in economics belong to the class of uncertain processes, and all attempts to even take non-stationarity into account will ALWAYS fall back on the human factor, which in economics is called politics, which, as we know, is "a concentrated expression of economics".

I don't think that game theory can take into account the influence of politics on the economy in such a way to model an uncertain economic process.

Uncertainty itself is an informal term from ordinary human language. Mathematics can only operate with some formal models of it. At the moment there are two such models - probabilistic uncertainty and game-theoretic uncertainty. Deterministic, chaotic and similar models of uncertainty are special cases of probabilistic uncertainty. In turn, probabilistic uncertainty is often considered as a special case of game-theoretic uncertainty, calling it "playing with nature". But game neoprstvo is bad and difficult to represent already at the level of basic notions - it is one thing to play a game and quite another to describe the same game formally. Perhaps, it is beyond human mind at all. Therefore, everything is usually reduced mathematically to probabilistic uncertainty (Nash equilibrium in mixed strategies, for example) or even determinism (minimaxes, etc.).

The current level of development of game theory does not allow to achieve too much in economics or political science, but in fact this theory has long been the basis, the "matan" of these sciences.

Of course, game theory has some practical successes too, for example, in organising auctions. But in our field, IMHO, its application so far is nothing more than games with terminology)

 

Forum on trading, automated trading systems and testing of trading strategies

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

Forester, 2023.08.19 09:41 AM

I think the difference is the seriality or repetitiveness of consecutive bars/ticks. During a trend most of them are in one direction, the randomiser makes them on average in 1.

Tried different options to account for seriality. They have the opposite effect. If seriality is divided into states {+1, -1, +1, -1, ....}, then after randomisation we get "trends" of seriality. Eventually several consecutive randomisations create just a straight line.


The symbol becomes super-trending if you take a small ZigZag as the seriality. Any such randomisation adds trendiness - long series in one direction.

Accordingly, if we take a large ZigZag, the same flat scalper does not merge (even earns something there). But this is for the reason that the flat spots are bypassed by the randomiser.


In general, there is no way to generate an earning cvr. Except in reverse time or in increments. If it makes sense to use increments, then only to check that the mathematically correct TS.

"Правильные" и "обобщённо правильные" по fxsaber`у ТС
"Правильные" и "обобщённо правильные" по fxsaber`у ТС
  • 2020.03.08
  • www.mql5.com
Здесь приведены некоторые соображения по поводу этой ветки. Формальное определение. Введём обозначения: r - ряд цен, s - система, e - эквити Подаём цены на вход системы и получаем на выходе эквити: r
 
fxsaber #:

Tried various options for seriality accounting. There is an opposite effect. If seriality is divided into states {+1, -1, +1, -1, ....}, then after randomisation "trends" of seriality are obtained. Eventually several consecutive randomisations create just a straight line.


The symbol becomes super-trending if you take a small ZigZag as the seriality. Any such randomisation adds trendiness - long series to one side.

Accordingly, if we take a large ZigZag, the same flat scalper does not merge (even earns something there). But this is due to the fact that the flat areas are bypassed by the randomiser.


In general, there is no way to generate an earning cvr. Except in reverse time or in increments. If it makes sense to use increments, then only to check that the mathematically correct TS.

About checking...

The main mathematical tool in trading is a family of various GARCH models (more than 100), which are fed ONLY price increments.

 
СанСаныч Фоменко #:

On the subject of testing...

The basic mathematical tool in trading is a family of various GARCH models (more than 100), which ONLY increments of price are fed to the input.

These models do not create an earned generated symbol from an earned original symbol. Yes and the idea itself is somewhat naive.

Forum on trading, automated trading systems and testing trading strategies

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

fxsaber, 2023.08.19 09:19 PM

What they do:

  1. Find several (let it be 100) stat characteristics in bar history.
  2. They generate a series of bars so that these 100 statistical characteristics coincide.

It is absurd that 100 values can describe an original series of millions of values! It seems to be a tool of theorists but not of practitioners.

 

Advanced resampling in the face of GMM other generative models are good at this task.

I obtained synthetic feature values from the original ones, trained the model on them, and it worked on the original data.

Продвинутый ресемплинг и выбор CatBoost моделей брутфорс методом
Продвинутый ресемплинг и выбор CatBoost моделей брутфорс методом
  • www.mql5.com
В данной статье описан один из возможных подходов к трансформации данных для улучшения обобщающей способности модели, а также рассмотрен перебор моделей CatBoost и выбор лучшей из них.