Machine learning in trading: theory, models, practice and algo-trading - page 3708

New comment
 
Maxim Dmitrievsky #:

Suppose, according to Mandelbrot, if H != 0.5, then there are some inefficiencies that can theoretically be transformed into TCs. But H is variable, and in a particular regime it may turn out to be constant. That is, so that at least something is constant. If multifractal spectra are calculated, there will be several components with different H.

An article on the subject

Crypto has the highest inefficiency rate on crypto. On dollar pairs, it is vanishingly small.

CONCLUSIONS This research utilizing the latest version of the Multifractal Detrended Fluctuation Analysis (MFDFA) provided a robust framework to uncover the fractal properties of these markets, offering deeper insights into their efficiency dynamics compared toearlier methods. The advanced capabilities of this method allowed for more accurate detection of multifractal behaviour. The result demonstrated that traditional asset classes such as stock indices (S&P 500, Nasdaq, Russell 2000, Hang Seng) , U.S. Treasury bonds (10-yearand 30-year), forex pairs (EUR/USD and GBP/USD), and commodities (gold and crude oil) exhibit characteristics of efficient markets. In contrast, cryptocurrencies like Bitcoin, Ethereum, USDT, and Binance Coin were found to be inefficient , highlighting a key distinction between emerging digital assets and traditional markets. The MFDFA method provided a robust framework to uncover the fractal properties of these markets, offering insights into their efficiency dynamics. These findings have importantimplicationsfor investors, indicating that traditional assets may provide more stable and efficient investment opportunities, while cryptocurrencies are subject to inefficiencies and higher volatility.
 
Maxim Dmitrievsky # :

Article on the topic

Cryptocurrency has the highest inefficiency rate. On dollar pairs, it is vanishingly small.

You have to look carefully at what is being counted. It is quite possible that it is essentially non-stationarity that is counted, and the fact that it is larger can only make things worse.

Roughly speaking, we need not only a difference from the SB, but also some constancy of this difference.

 
Aleksey Nikolayev #:

You have to look carefully at what is being counted. It is quite possible that it is essentially non-stationarity that is counted, and the fact that it is larger can only make things worse.

Roughly speaking, we need not only a difference from the SB, but also some constancy of this difference.

The type of randomness is considered (soft/bullish), what exactly is predicted and how, these methods don't tell you. Mandelbrot wrote that it's not for prediction, but for risk assessment :)

Like how big to build a dam or how big to put a stp-loss so that it doesn't drain :)

And to predict something concrete, you need to predict the continuation of eggs and chickens, in my opinion.
 
Eventually, the complexity of constructing all sorts of transition trajectories and phase spaces will cause one to want to use a convolutional NS instead of all that, which, as we know, does not lead to filling pockets either.
 
Aleksey Nikolayev #:
I don't understand anything, but it's very interesting. Some set of assertions that follow from nothing. As usual, though.
You're not speaking the same language.)
 
Aleksey Nikolayev #:

Well, then you'd have to cite the "told you before" references.


unlike Renat, it is easier for me to show again

TC, probably will forgive for some repetition of off-topic, in this topic I have already spammed with such pictures :-)

We were talking about "conditionally constant", so if we count statistics, we get this picture:

( I got a prettier visualisation with statistics, and it counts not only angles/rates but also bindings and it is easier to explain)

It is very easy to count with statistics - everyone knows how to build "horizontal profiles" ?
so if you build "diagonal profile" in the same way, it will show regular structures, almost sinusoids (the line on the screenshot is the "tops" of sinusoids).
Depending on the rait (angle) of the diagonal, more or less pronounced.
Modern capacities allow you to make almost a continuous enumeration and identify the most "outstanding" rait. By the way, it is one, and the rest are all kinds of its multiples and combinations.
Currencies do not shy away from anyone and just walk with this rate.
Statistics also makes bindings, objectively showing that the nodes of the grid exactly fall into the regular maximums and minimums of volatility.

It is also linked to volatility, the same angle can be calculated from tick volume.

(by the way, it is quite funny and unexpected that long movements usually start right in the morning, at the minimum of volatility, and not at all with reversals at the am session, which rather close the previous one.
Although it is possible that the observer's effect and just better noticeable).

We were talking about "catch-constant"? Here is a conditionally-constant pattern in quotes, closely related to the same conditionally-constant volatility.

I think that the presence of such a pattern, or rather it is not taken into account in your favourite methods ML,NN very much confuses them.

 
Maxim Kuznetsov #:

unlike Renate, it's easier for me to re-do it.

It would be curious to see an example of realisation by means of MO in the topic about MO, at least a sketch of what was meant.

 

Finished creating modes in python and exporting them to a pluggable library, then getting the current mode via a function like this.

I guess I need to think how to make it even more convenient, so that any MO-fit can connect to my bot :)

As far as I understand, the main problem is that MO is hard to do in MQL, and almost nobody knows python, just like nobody knows MO :)

so that it could be reduced to calling a single function.

int get_regime()
   {
      double features[];
      fill_arrays(features);
   
      double f[ArraySize(periods)];
   
      for(int i = 0; i < ArraySize(periods); i++)
        {
         f[i] = features[i];
        }
   
      static vector out(1);
   
      struct output
        {
         long           label[];
         float          proba[];
        };
   
      output signal[];
      
      OnnxRun(ExtHandle, ONNX_DEBUG_LOGS, f, out, signal);
   
      int sig = ArrayMaximum(signal[0].proba);
      return sig;
   }
 
Maxim Dmitrievsky #:

It would be curious to see an example of implementation through MO in the MO topic, at least an outline of what was meant.

and this is not MO, which I warned about in the header :-) you can consider it a branch of the dialogue about "conditionally constant" things.

What is on the screenshot - statistics, the method of obtaining it is popularly described. Those who are not lazy and interested will repeat and improve it - it's simple there

 
Maxim Kuznetsov #:

and this is not MO, which I warned about in the header :-) you can consider it a branch of the dialogue about "conditionally permanent" things.

What is on the screenshot is statistics, the method of obtaining it is popularly described. Those who are not lazy and interested will repeat and improve it - it's simple there

I don't even understand where to start :) then you need an article

1...370137023703370437053706370737083709
New comment