Dynamic periods for indicators - page 7

New comment
 
nikost >>:


0 вечером в ноябре - ночью минус.

))) And the ruler?

An indicator is an analytical tool, like the already mentioned thermometer and ruler. Of course, you can ascribe predictive functions to, say, a speedometer: by moving at its indicated speed, you can estimate when you will reach the goal. OK. But that is our interpretation - the speedometer does not know shit about it, just as we do not know about "whether we will make it")).

It's the same with the so called "ahead" indicators. It's like putting one and the same kilometrage into speedometer and as a result it shows us time to the target.

Anyway, if we're talking about indicators - it's worth talking about how accurately they reflect what you want to track.


===

!!!!!!!!!!!

Avals! Are you a telepath? (I am not, I have never noticed it myself.)) I mean the speedometer example... Eh. Now they will not believe that I made it up myself... )))))))) Buggahaha!)))

 
Svinozavr писал(а) >>

Avals! Are you a telepath? (I'm not, I've never noticed that.)) I mean the speedometer example... Eh. Now they will not believe that I made it up myself... )))))))) Buggahaha!)))

I propose to consider it co-authored :)

 
Avals >>:

предлагаю считать соавторством :)

I agree.)))

Shit! When I caught a glimpse of your post - so at first I thought I had posted in some insane way. I got worried about my own sanity.

When I read and understood that you wrote it, I got even more frightened... )))))))))))))

 
A little about the speedometer.
If you analyse the readings during the previous 1s, there will be little, if any, information about the future movement of the car.
If you analyse the readings for 1 hour, there is not much information either.
And if you analyze the history of speedometer readings for 2 years, you can assess the condition of the engine, tyres, brake pads, etc. with a very high probability, i.e. the "driving style" will be determined. And hence, one can conclude that if such and such acceleration is followed by braking, a chain of events over several time intervals (windows) is considered, and the more of them, the better.
I.e., it is impossible to adequately change dynamically the periods of indicators without having (and/or) analyzing the history of its readings.

HH This is just thinking out loud.
 
It seems the equivolume charts were not of much interest for me :)
Admittedly I didn't build them myself, but I thought it might work as an option for adaptation.
I understand correctly, the author's purpose is to see essential changes inside the candle:
if there is something interesting in a candle, we split it into N smaller ones and if there is no change, we merge N candles into one?


I myself tried as an adaptive indicator a bank of FIR low-pass filters. The first filter was built by N candles (the lowest frequency), the second by N\2 candles, the third by N\4, etc. up to 1 candle. For each filter I've calculated the standard deviation of the graph for its period. I chose a filter for the indicator at the moment that satisfies the RMS value and is the lowest frequency. I did not get anything useful :)

 
joo >>:
Немного про спидометр.
Если проанализировать показания в течении предыдущей 1с, информации о будущем движении авто будет немного, вернее вообще не будет.
Если проанализировать показания в течении 1ч, информации будет также немного.
А если проанализировать историю показаний спидометра лет 2-х, можно с очень большой вероятностью оценить состояние двигателя, покрышек, тормозных колодок и т.д., т.е будет определен "стиль езды". А значит, можно делать выводы, что если типа после такого то ускорения последует торможение, рассматриваются цепочки событий за несколько промежутков времени (окна), и чем их больше, тем лучше.
Т.е., невозможно адекватно менять динамически периоды индикаторов не имея (и/или) не анализируя историю его показаний.

ЗЫ Это так, мысли вслух.

And if we take dt, we get almost 100% predictor!!! ))

===

Yeah. You're not mixing odometer and speedometer into one?

 
Svinozavr >>:

А если мы возьмем dt, то мы получим практически 100% предсказатель!!! ))

===

Да. Вы не мешаете в одно одометр и спидометр?

Well, I don't really mean just the speedometer, but in general, every conceivable sensor that can be fitted to a car. :)

And about the dt, I didn't mean that. I wanted to say about the chain of events and the subsequent area of possible further movements of the car. It is absolutely impossible to predict the further trajectory with absolute certainty, but the probable area of trajectories is quite possible.

 
Svinozavr писал(а) >>

Tell me, what forecast does the thermometer give?

The indicator is indicating. Stochastic, for example, indicates that such is the case, the price on a given bar is so positioned between the max and min for the last %K bars. That is all.

From this point on it is your interpretation. What conclusion you draw and what decision you will make on the basis of this information is a matter of your own, your TS.



Peter, it's kind of sad to hear you say those "deep" thoughts. Who told you that using an indicator is all about using its last value only. Are you a fan of the zero bar? You look at the market only through its narrow window? Really, it's ridiculous.

The indicator by itself does not provide anything in the market. Don't you know this truism? But a good indicator in combination with the knowledge of its properties and capabilities, as well as a certain technology of its use can be the core of TS and the source of profit. That's why when I say "the target function of an indicator should be estimation of forecast reliability" it's obvious that this forecast is made not by the indicator, but out of it. And the algorithm for this prediction is the very interpretation that I made, without which nothing at all happens in the market: orders are not opened or closed, TP and SL are not estimated, no decisions about risks, pairs, timeframes, tactics, etc. are made. And it doesn't depend on whether I play with my hands or have formalised my interpretation in MTS.
And you write "your interpretation" with such intonation as if it is something completely derogatory for a true trader. :-)

 
joo >>:
Т.е., невозможно адекватно менять динамически периоды индикаторов не имея (и/или) не анализируя историю его показаний.

ЗЫ Это так, мысли вслух.

Very good idea! I thought it was implied "by itself". while we have no criteria for evaluating and changing an indicator period, we ourselves act as a "manager" selecting the indicator parameters. Of course, by applying different algorithms we should evaluate how accurately the new picture created by the indicator corresponds to our estimation of historical data. Here's just one small step to a normal neural network and genetic algorithms ;)

But I would not like to do it. As you know a neural network doesn't learn by itself. Selection of weights, activation functions of neurons, network topology is a separate task. And it must be done consciously. So back to the original question... And the example with the speedometer seems to me very good: no matter where you go, we should understand that the speedometer needle has inertia and the real speed after a forced boost it will show in a few seconds and maybe at some point it will wander further along the scale than the actual speed and then take it into account while driving. That is why I "urged" to analyze the question without reference to the profile of the vehicle - to investigate the thing in itself and to understand the limits of its possibilities and its controllability.

 
Here is probably something similar to what I was talking about. We take RSI and divide the chart into sections where it is above or below 50 (for clarity - subtract 50 to get a fair histogram). Now we count RSI within each section and choose the period length equal to the current number of bars in the current section. It seems that we did not see anything new, but the very first in the plot became "more pronounced" :) 

#property copyright "Copyright © 2006-201, Sergey Kravchuk"
#property link      "http://forextools.com.ua"

#property indicator_separate_window
#property indicator_buffers 3
#property indicator_maximum 55
#property indicator_minimum -55
#property indicator_color1 CLR_NONE
#property indicator_color2 CornflowerBlue
#property indicator_color3 Navy
#property indicator_style1 DRAW_NONE
#property indicator_style2 DRAW_HISTOGRAM
#property indicator_style3 STYLE_SOLID
#property indicator_width1 0
#property indicator_width2 10
#property indicator_width3 1

double LeftBars[];       
double Values[];       
double DynValues[];

extern int FixedPeriod  = 26;
extern int PriceType    = PRICE_CLOSE;
extern int MinPeriod    = 5;
extern int SmoothPeriod = 5; 

int init()
{
  IndicatorBuffers(3);
  SetIndexBuffer(0, LeftBars);
  SetIndexBuffer(1, Values);
  SetIndexBuffer(2, DynValues);
}


int start()
{
  int i, LeftBar;

  Values[Bars-1] = 0;
  for (i = Bars-2; i >= 0; i--)
  {
    Values[i] = iRSI (NULL, 0, FixedPeriod, PriceType, i) - 50;
    if ( (Values[i] > 0 && Values[i-1] <= 0) || (Values[i] < 0 && Values[i-1] >= 0) ) LeftBar = i;
    LeftBars[i] = LeftBar;
  }
  
  for (i = Bars-2; i >= 0; i--)
  {
    int DynPeriod = LeftBars[i] - i; 
    if ( DynPeriod < MinPeriod ) DynValues[i] = 0;
    else DynValues[i] = iRSI (NULL, 0, DynPeriod, PriceType, i) - 50;
  }
}
And here is the picture: the thin blue line is dynamic RSI.


Kind of if very conventionally: when we observe a big gap between "dynamics and statics" - it is "market change". when they start to get closer - it is "preparation for the market change".
12345678
New comment