Market condition - flat or trend? Which dominates?

 

Market condition flat or trend? Which prevails?

I would like to get a reliable and well-reasoned answer to this question.
There are a variety of opinions on this matter. Below are some of them in italics.

Trading strategy

Market prices are in constant motion. The market condition at any time can be

a strong unidirectional change (rise or fall of the price) of the price or a sharp fall of the price.

or flat - a sideways price movement with weak deviations from a certain average.

average. These market characteristics are arbitrary because there are no clear

There are no clear criteria according to which a trend or a flat may be identified.

For example there may be long periods of sideways price movement with strong deviations which

cannot be considered to be either a flat or a trend. It is generally accepted that in general the market

trends and that trends occur on the market approximately 15-20% of the time.

Fig. 110. Flat and trend in the market.

https://book.mql4.com/ru/samples/expert

Unfortunately, the author does not specify on the basis of which data the author has come to such a conclusion.

conclusion.

Rosh wrote:
That's why I asked. I think that the moments of a flat are precisely the unstable

states. The steady state for the market is movement. I think so.
12.10.2007 20:29
Alas, I probably can't formulate it. I'm reading Peters( once again) about the fractality of the market -

and agree with him that the normal state of any stable system is nonequilibrium. Here

and the self-similarity property of the fractal agrees with the presence of the investor on the horizon of any

duration, and non-linearity and asymmetry in decision making, and many other things.



It is a trend or a flat
What is a stable condition, a trend or a flat? In my humble opinion, it is only

just terminology and some agreement between the participants in their views. The authorities teach us

are taught that the market mostly sits in a flat and only spends a little time in the trend. After

my own, trivial experiments I've come to another conclusion: local (and there are no other

and trends exist in approximately equal proportion.

The author of the statement gives his arguments with calculations, but what criteria were

were used for the calculations are not stated.

Additional searches did not help to answer the question (there is always the variant that you did not

searched badly).

I would like to hear not just the opinion of the respected forum users but also the opinion backed up by the criteria

and calculations.

 

It is a question of "which came first, the chicken or the egg?

I've read about 50/50: the market is trends and the flat is a moment of volatility as it transitions from one trend to another, and the market is flat and the trades are just transitions from one flat to another. Both opinions are very valid.

 
timbo:

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

I've read about 50/50: the market is trends and the flat is a moment of volatility as one trend transitions to another, and the market is flat and the trades are only a transition from one flat to another. Both opinions are very credible.

That's another opinion to add to the box of judgement. I am not disputing it. But, "I would like to hear not just the opinion of respected forum members, but an opinion supported by criteria

And calculations."

 

to Xadviser

Автор высказывания приводит свои аргументы с вычислениями, однако какие критерии были приняты для расчетов не указано.

I told you, just in case:

Trend or Flat

What is a steady state, a trend or a flat? In my humble understanding, it is just terminology and some agreement between the participants in their views. The authorities teach us that the market mostly sits in a flat and spends very little time in a trend. After my own, trivial experiments I've come to another conclusion: local (and there are no others) flats and trends exist in approximately the same proportion. I am going to show you, as an occasion for reflections, the first segment of EURUSD (hours, (H+L)/2) that has come to hand. The algorithm for gathering statistics is simple, I "step with the times" in the tested interval, fix the length of the initial time series, look into the future at every interval and determine the duration of a sideways channel (flat) and a linear regression channel, of course, using the parameters of the same initial sample. This is what I got for the window of 600 samples:

  • Red - lifetime of the side channel (flat)
  • Blue - lifetime of the linear regression channel

The x-axis represents the samples of the analyzed range, and the y-axis represents the channel length scaled to the window size of the time series (i.e. lifetime - 2 means that the channel with initial parameters lives, two more initial lengths, 2*600). If we take the whole history, and go through the window lengths, we still get approximately the same picture (almost like in the figure). The average duration of the "flat" channels is slightly longer than the linear regression channels, but none of this is "significantly" what the authorities are writing about. Of course, the argument is indirect, but it has led me to some thoughts.

The criterion is very simple - it is the length of the "trend"/"flat" and their comparison. Linear regression was taken as the definition of "trend", the average value and the deviation from it as "flat". Neither rule is strict, if only for the simple reason that reliable trend criteria for such series simply do not exist.

to timbo

This is a "what came first, the chicken or the egg?" level question.

It's a matter of "how to count" or "which one is the chicken and which one is the egg". In my practice, in one enterprise the boss set a condition - to double production output. The production is of a "process" class. So, the boss's instruction was carried out within 20 minutes, using another correct methodology as a basis.

 

to grasn

Hi Sergei, what are you working on at the moment?

to Xadviser.

Regarding the question raised by the author of the branch.

In order to answer correctly the question you need to decide on the TimeFrame we will work with. Look at the fig. The figure represents a sinusoid with period T. Now let's imagine that it is a series of quotes. It is intuitively clear that if we work on TF much less than period T, then this market will be a trend market. And vice versa, if we work on TFs much larger than T - it will be flat.


In this statement it is possible to give a qualitative estimation of the current market state. For this purpose, a time series (TSR) should be divided into equal timeframes equal to the selected TF and price increments should be recorded in each section. The trend market is characterized by a directed movement, i.e. the price increments have the same sign. For a flat market - different. That is why the sum of products of adjacent increments will point to a trend or a flat:


If r is close to 1, we have a pronounced trend. If r is close to -1, we have a pronounced flat. If r is close to 0, we have a chaotic market.

Of course, the question remains about the stability of this criterion, the value of "close to..." and the choice of T.



 
grasn:
... Here is the first available segment of EURUSD (hours, (H+L)/2). The algorithm for statistical data collection is simple - I "step with the times" fix the length of the initial time series in the tested period, look into the future at every interval and determine the duration of a side channel (flat) and a linear regression channel using parameters of one and the same initial sample, of course. This is what I got for the window of 600 samples:

The average duration of the "flat" channels is slightly longer than that of the linear regression channels, but none of this is "significantly" what the authorities write about.

Of course, the argument is circumstantial, but it has led me to some thoughts.


1. Your argument is the only sufficiently reasoned and computationally supported opinion I've been able to find on the forum. It is the one I deduced at the beginning of the post.

But for me it is not very clear the "first caught" criteria, how the original time series was fixed, why it came out 600 counts (or not, but was so taken), what are the counts (are they hourly candles?) why were hours taken for calculation, what are the measurements on other TFs?

There are no reliable criteria, here I agree, but clear (rigid) ones can be set

2. How much is a little?

3. it has been nudging me for a long time, that's why I want to calculate

 
Neutron:

In order to answer the question correctly, we need to decide on the Timeframe on which we are going to work.

The trend market is characterized by the directional movement, i.e. the price increments have the same sign. For a flat market they have different signs. Therefore, the sum of products of adjacent increments will point to a trend or a flat

You are correct in what you say. I agree that probably the measured values will change depending on the selected timeframe, or maybe they will statistically coincide. You have indicated one possible calculation and estimation. Are the results of those measurements available?


The question was not "how to measure", but what are the measurement results for such and such criteria (optionally, those suggested by you). I.e. I'm not interested in what is the market condition at any given moment, but the correlation of time intervals in the corresponding conditions (trend, flat) during a long period of time. So far nobody, except grasnay, has presented calculations. But we have some questions for him as well (I stated above).

 

to Neutron

Привет, Серёга! Над чем сейчас работаешь?

Seryoga hello! Good to see you in a good mood. I am working on the "trajectories". Let me remind the essence of the matter: I developed quite a good model a long time ago (I have several of them), which predicts, if I may say so, "the levels of high price concentration". Figuratively speaking (from an artistic point of view) - if you print the price chart with "thin dots" and step aside from the chart at some distance you can notice such "accumulations", such "black spots". Mathematically it is simple - it is a local process which conditionally can be called stationary (the price concentrates around a certain level, as a mean value). It turned out that such a "subprocess" can be confidently predicted. I got excellent quality of forecasts for each bar (I've already boasted about percentages of successful forecasts). But, unfortunately, very high drawdowns and great difficulty to calculate stops. I`ve already started to make some good results, so I`ve tried to trade some caps, but I`ve had no luck and asked Yuri to buy islands, so I lied and told him that all of them were sold out. :о))))))))


I was quite upset. But continuing researches I have understood, that the process of "local stationary sub-flows" formation is very much connected with zigzag, i.e. in many respects it defines "fluctuations" around these levels. In other words, I figured out that it is possible to go to ZZ prediction. I am working on it at the moment: I am collecting necessary statistics including those on ZZ and working on my model.


to Xadviser

In my opinion, I would classify your question as one of the fundamental ones, which largely determines the search for strategies with a statistical advantage. Understanding it the way I wrote - took some time to answer the question, at least for myself. But unfortunately, the search for solutions and answers lie rather in the philosophical plane, as there is no one strict criterion.

1. Your arguments are the only sufficiently reasoned and supported by calculations opinion that I could find on the forum. It is the one I deduced at the beginning of the post.

It is quite possible that colleagues have simply not posted their findings.

But for me it is not very clear about the "first caught" criteria, how the initial time series was recorded, why it turned out to be 600 counts (or not, but was accepted as such), what are the counts (are they hour candles?) why hours were taken for calculation, what are the measurements on other TFs?

The first segment I came across was given as a graphical illustration. Just the whole story will of course fit into the graph, but nothing will be visible. Of course, I studied the entire history available for the basic quotes, though I only used hours, I do not work with lesser or longer timeframes. I was going through the window length in increments of 10.

Addendum

1 The readout is 1 Bar.

Why the clock - it shows certain dependencies as much as possible


There are no reliable criteria, I agree here, but clear (rigid) criteria can be set

This will be hard related to your question "2. A little is how much?". It is important to use a criterion that will be used for modelling in one way or another. In fact, you will not be collecting statistics about trends and flotsam, but about the criterion. And if it works, then everything else (if the trend is on mars ...) is irrelevant. I got 10-15%, that's from memory and that's considering the clock - i.e. for Euro the length of history is around 50,000 counts. Well, the most important criterion, which was calculated together with what I got on the chart, I was silent about

"3. It's been nudging me for a long time, that's why I want to calculate"

I get that :o)

 

Xadviser писал (а):


why was the clock taken for the calculation, what are the measurements on other timeframes?


I think if the market is 100% fractal, then the trend/flat ratio should be the same for any timeframe (while of course it may depend on the "measurement" 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 did not do such research myself, the time horizon of the game is determined by other considerations. But the results would be interesting.
 
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.

 
grasn:

In my opinion, I would classify your question as one of the fundamental ones, which largely determines the search for strategies with a statistical advantage. Understanding it the way I wrote it - spent some time to answer it, at least for myself. But unfortunately, the search for solutions and answers lie more in the philosophical plane, as there is no one strict criterion.


1. The popular wisdom says that if you don't know your way around, don't go in the water. Without an answer to fundamental questions there is no point in getting into trading.

:-) What is Forex? What is a trend (or flat)? What affects the price change?

Just don't post the results of your research.

2. Perhaps. Probably, they have not had any studies, or have not decided on the criteria, or have kept silent about the results :-). That's why the question was raised.


The first segment I found was given as a graphic illustration. Just the whole story will of course fit into the graph, but you won't see anything. Of course, I was investigating using all available history for basic quotes, though I was using only hours, I was not working with lesser or longer timeframes. I was trying to change the window length in steps of 10.


3. Now I understand, thank you.

It is important to use the criterion that will be used for modelling in one way or another. In fact, you will not be collecting statistics about trends and flotsam, but about the criterion. And if it works, then everything else (if the trend is on mars ...) is irrelevant.

4. Right, you have to decide on the criterion first. That's why the question sounded calculations and criteria.


Well I was silent about the most important criterion, which was calculated along with what got on the chart


5. If it's not a secret, can you tell us which criterion was used to make these calculations and what the result was?