Market condition - flat or trend? Which dominates? - page 2

 
lna01:
I think if the market is 100% fractal then the trend/flat ratio should be the same for any timeframe (of course this may depend on the "measuring" technique). If the ratio turns out to depend on the timeframe, it may be one of the factors influencing the choice of the working timeframe. The methodology, imho, should be oriented to the intended game strategy. I myself have not done such research, the time horizon of the game is determined by other considerations.

But the results would be interesting.

I have come to the same conclusions (on an intuitive level), but I would like to confirm (or refute) my conclusions.

Moreover. I presume the possibility, means and methods of conducting such research. But not to start a vegetable garden, I brought up the subject to find out if anyone else has asked himself these questions and what answers and results he got.

I hope they are not only interesting but also useful.

 

to Neutron

Похоже мы с тобой пришли к одной цели совершенно разными путями! Я сейчас упорно работаю над Нейронными Сетями как универсальным инструментом для прогнозных целей. Забавная вещица - прогнозирует всё что только в ВР есть! Научился в маткаде писать НС с произвольной архитекртурой и нелинейностью. Занят её оптимизацией. Вобще, аж дух захватывает, когда видишь как она обучается и пытается применять полученные знания.


Not at all different :o). I've been dealing with NS for quite a long time, it's really amazing. Somebody proved, that with NS I can implement any algorithm, and the basis of mathematics of NS is very close to the principles of adaptive linear filtration in DSP. I use NS practically "every day", now I concentrate on searching of dependencies with NS instead of using it directly for forecasting (few people know, that NS does this kind of tasks better than image recognition for example). Roughly speaking, if a dependence exists, it will be detected, it is another matter whether it is possible to simply formalize it. Having said that, I am not a "fan" of including NS in experts, because it's simple - if there is no "valid law", no NS will help you, while learning - it's time to retrain again.

PS1: I have a question, (on the materials of a neighbouring thread) where can I read about the theory of optimal reinvestment? I think it will come in handy soon.

PS2: By the way, will have time - pay attention to models of NS based on the theory of information (the concept of information entropy is used). Maybe you will find their application to price time series interesting.


to Xadviser


If it's not a secret, what criterion was used to calculate and what was the result?

The parameters were collected for the model "http://www.metatrader4.com/ru/forum/6839/page107" post "grasn 11.01.07 16:16" and I'm not ready to describe them in details yet. Let's say, the predicted trend length was determined empirically ( linear regression was taken as a trend). All in all, it works.

4. Right, first of all we should decide on the criterion. That is why the question sounded calculations and criteria.

Looking forward to results of your experiments (it is interesting for me, but I can't spend time on it now). His criteria cited, Seryoga also gave quite a good criterion. If anything, ask.

PS:

Here I also came to the same conclusions (on an intuitive level), but I would like to confirm (or refute) their fabrications.

Moreover. I assume the possibility, means and methods for such research. But not to start a vegetable garden, I brought up the subject to find out if anyone else has asked himself these questions and what answers and results he got.

I hope they are not only interesting but also useful.

I, too, got 50/50. Maybe I shouldn't waste my time, otherwise I'll get 60/40 and you'll think it's closer to 50/50 or 90/10 :o)

 
There is, for example, a Hearst coefficient to estimate the flatness of the flux. Of course, if you measure it on the whole range, it will be 0.5. Like the average temperature in a hospital.
 
grasn:

PS1: I have a question, (from a neighbouring thread) where can I read about optimal reinvestment theory? I think it will come in handy soon.


I cannot cram the archives because of the large volume.

Look online: V.Zhizhilev "Optimal Strategies for Profit Extraction" and Ezhov A.A., Shumsky S.A. "Neurocomputing.

 
Neutron:
to grasn

You and I seem to have come to the same goal in very different ways! I am currently working hard on Neural Networks as a universal tool for predictive purposes. It's a funny thing - it predicts everything in BP! I've learned to write an arbitrary architecture and non-linearity neural network in Matcheda. Busy optimizing it. In general, it takes my breath away, when you see how it learns and tries to apply the knowledge I've got.


About Matcad and NS in it, it would be very interesting. I would appreciate it if you could share. Although I'm afraid I don't have time to touch, feel and examine everything. I used to program similar things about 20 years ago, maybe I am mistaken and my knowledge is outdated like that of a dinosaur.
 
Prival:
Neutron:
to grasn

You and I seem to have come to the same goal in very different ways! I am currently working hard on Neural Networks as a universal tool for predictive purposes. It's a funny thing - it predicts everything in BP! I've learned to write an arbitrary architecture and non-linearity neural network in Matcheda. Busy optimizing it. In general, it's breathtaking to see how it learns and tries to apply what it has learned.


I want to know how to use Matcad and VS in it, it would be very interesting. I would be grateful if you could share. Although I'm afraid there won't be enough time to touch, feel and examine everything. I've lost my confidence in them about 20 years ago, I used to program some similar things and may be mistaken, my knowledge is outdated like that of a dinosaur.
Hi Sergei!

I use the following design:

It is a grid with number of layers Layer, which can be changed from 1 to 3, number of inputs - In, number of neurons in each layer - n and one output, w - configurable weights. From publications it does not make sense to increase the number of layers further, as three-layer NS approximates an arbitrary surface on d-dimensional cube. The nonlinearity can be specified by an activation function f(x) or, even more abruptly, specified as a degree of x under the sum sign, in which case complete universality can be achieved even for a single layer NS, but there is a problem of applying back propagation of the error during training.

I use this grid to determine optimal architecture, number of inputs and training sample size. I feed price increments to the input of the NS and take a step forward forecast from the output.

 
grasn:

The parameters were collected under model "http://www.metatrader4.com/ru/forum/6839/page107" post "grasn 11.01.07 16:16" and I'm not ready to tell about them in details. Let's just say the predicted trend length was determined empirically (a linear regression was taken as a trend). All in all, it works.

1. Yes ... I've been familiar with this thread for a long time. And I noticed your post back then. But it's hard to grasp the immensity.

Looking forward to results of your experiments (it's interesting to me, but I can't spend time on it now). His criteria have given, Seryoga also gave quite a good criterion. If anything, ask.

2. Unfortunately, I do not have programming skills. Therefore speed of giving out results of experiments depends on speed of finding the interested person, having such skills. I will certainly give you the results.

Perhaps someone else will respond.


3. I would like to ask the other participants to have a constructive dialogue within the framework of the topic's title.

Respectfully.

 
Xadviser:
SK. wrote (a):
Yes, such a thought has a right to life. I've done it myself on occasion.
While editing my night post, you replied.

Interested in your opinion on this subject https://forum.mql4.com/ru/11661 I would appreciate it.

It overlaps somewhat with the topic of this thread. I.e. the use of some kind of price movement pattern in trading.
timbo:

It's a "which came first, the chicken or the egg?" level question.

I've read about 50/50: the market is trends, and the flat is a moment of volatility when moving from one trend to another, and the market is flat, and the trades are only a transition from one flat to another. Both views are very valid.


I am close to the point made by timbo. In my opinion, the answer to this question boils down to a conversation about concepts. Human beings are known to use many concepts, the approximate meaning of which in the overwhelming majority of cases is not defined. For definition of any concept it is necessary to find out exactly its "irreducible remainder", i.e. minimal set of attributes inherent to a concept. This set is not defined for most concepts. It is extremely difficult to find out the differences even for the simplest, ordinary objects. For example, what is the difference between a cup and a mug or a book and a magazine? Difficulties arise in relation to such notions as truth, regularity, law, etc.

Trend and flat are not exceptions. The boundaries of these concepts are blurred. In other words, they are rather expressions of a trend, determined on the basis of personal preferences. How can you use these concepts in practical work? It makes sense to me: at the first stage, it makes sense to set roughly the boundaries of these concepts. For example, in the form of the slope of the average regression line for a certain period, unit = points/time. The boundary can be defined arbitrarily, say, 10p/hour. If it is more than that, it is a trend, if it is less than that, it is a flat.

During the creation of a trading system, these concepts can be specified. If TS includes one algorithm for searching trade criteria for a trend and another one for a flat, then, by varying the period for regression line calculation and the angle threshold value, we can obtain the data set and select the best of them. As a result of such investigation, it will be possible to claim that the definitions for these concepts have been adopted for this TS. Within the framework of these definitions, a TS has been built that gives such-and-such results.

In any case, do not assume that the concepts of trend and fdat are something immutable and predetermined.
 
SK. писал (а):
The concepts of trend and flat are no exception. The boundaries of these concepts are blurred. In other words, these concepts rather express a trend determined on the basis of personal preference.

1. Thank you.

2. Completely agree, as I hold the same position. "... based on personal preferences".

3. Do you have an elaboration with personal preferences (criteria)? Essentially there were two questions: what are the criteria? (also preferably why they are) and what is the result under these criteria?

I.e. I'm not interested in the difference between trend and flat and not which of them is primary, but what is their relationship?

 
Xadviser:
SK. wrote (a):
The concepts of trend and flat are no exception. The boundaries of these concepts are blurred. In other words, these concepts rather express a trend determined on the basis of personal preference.

1. Thank you.

2. Totally agree, as I hold the same position. "... based on personal preference."

3. Do you have a track record with personal preferences (criteria)? Essentially there were two questions: what are the criteria? (preferably also why) and what is the result of these criteria?

I.e. we are not interested in the difference between the trend and the flat and not which of them is primary, but what is their relation?

You don't seem to have understood the leitmotif of my previous post.

Defining some numerical ratio in one way or another is defining a boundary, i.e. essentially defining a trend and a flat. In my previous post I tried to show that no immutable relation exists in principle. Specific definitions can be given only for a specific application to a specific trading system. I am not oriented on the notions of trend and flat in my developments. But if I had my own definitions, they would be suitable only for my system.