To follow up - page 49

 
MetaDriver >>:

Не. Из жадности. ;)

Come on, what kind of greed is there. OK, let's talk about volatility. In principle it is not a bad choice at all for QC coordinate.

The main question is: which window to choose for volatility? In fact, this is a hidden parameter of the QC coordinate.

Second question: how to calculate volatility?

a) S.c.o.? I don't like it. It is not an estimate of volatility, because s.c.o. as an algorithm uniquely corresponds to the only distribution - normal. It is not, every youngster here knows that.

(b) Closer to reality would be something like the average of the modulus of returns: Bulashev sort of showed, without proof, that the distribution of the modulus of returns can be considered exponential, at least to a first approximation. And such a distribution is still closer to error estimates as the average of individual errors, rather than s.c.o.

c) ATR. Also not bad. Rather close ideologically to point b).

d) Percentile method of volatility estimation: choose a not insignificant window (about 100), define a number of percentile (say 50), and then look at the distribution of closing price moduli back 100 bars. The point corresponding to the returns dividing the distribution in half will be the volatility estimate.

Personally, I like this method the best. It is much more robust to the shape of the distribution returns (we are not making hypotheses about the distribution function). It is also more robust to the "volatility window".

That's pretty much all I think about volatility so far. If interested, I can show you what indicator d) looks like at different windows.

P.S. Yep, here comes the second volatility parameter as QC coordinates: we need to know some "boundary" volatility to compare with when making a trading decision.

 
Mathemat >>:
----------------------------------

Это пока почти все, что я думаю о волатильности. Если интересно, могу показать, как выглядит индикатор d) при разных окнах.

P.S. Ага, вот и всплыл второй параметр волатильности как координаты КК: нужно знать некую "граничную" волатильность, с которой будем сравнивать при принятии решения о торговле.

Show me. You need to choose a window and a TF that the volatility graph is close to a sine wave. I can't tell you which ones at a glance, but they are there.

 

No, no, Andrei, don't expect a nice sine wave. But there will be smoothness (although there is no smoothing in principle) - and, therefore, some predictability. A little later. There is an indicator, but it needs to be tweaked a bit.

 
Mathemat >>:

Не-нет, Андрей, не надейся, что будет красивая синусоида. Но будет плавность (хотя никакого сглаживания нет в принципе) - а, значит, и некая предсказуемость. Чуть попозже. Индюк есть, но его нужно слегка подкорректировать.

I did not mean a "beautiful" sine wave, but something similar with periodicity. Volatility in the form of a straight line is not needed, but on the other hand volatility calculated by a window of one bar will be very "phonetic".

 
Mathemat писал(а) >>

Come on, that's not greed. OK, let's talk about volatility. In principle it's not a bad choice at all for a QC coordinate.

The main question is: which window to choose for volatility? In fact, this is a hidden parameter of the QC coordinate.

Second question: how to calculate volatility?

a) S.c.o.? I don't like it. It is not an estimate of volatility, because s.c.o. as an algorithm uniquely corresponds to the only distribution - normal. It is not, every youngster here knows that.

(b) Closer to reality would be something like the average of the modulus of returns: Bulashev sort of showed, without proof, that the distribution of the modulus of returns can be considered exponential, at least to a first approximation. And such a distribution is still closer to error estimates as the average of individual errors, rather than s.c.o.

c) ATR. Also not bad. Rather close ideologically to point b).

d) Percentile method of volatility estimation: choose a not insignificant window (about 100), define a number of percentile (say 50), and then look at the distribution of closing price moduli back 100 bars. The point corresponding to the return dividing the distribution in half will be the volatility estimate.

Personally, I like this method the best. It's much more robust to the shape of the distribution (we're not making hypotheses about the distribution function). It's also more robust to the "volatility window".

That's pretty much all I think about volatility so far. If interested, I can show you what indicator d) looks like at different windows.

P.S. Aha, the second parameter of volatility as QC coordinates appeared: we need to know some "boundary" volatility, with which we will compare while deciding on trading.

1. Intraday volatility is very cyclical and if it is evaluated in context to intraday trading or any other trading with entries/exits based on TF below days, we need to take it into account. In fact each hour of the day has its own standard vol and non-standard vol too. I.e. you need to look at the relative deviations of volatility for each time of day, or take an averaging period multiple of a day. https://www.mql5.com/ru/forum/117000 A similar study of volatility is available in Shiryaev, I think in Volume 2 of "Fundamentals of Stochastic Financial Mathematics".

2. The natural measure of volatility is tick volume. If we take the total for sufficiently long areas, then its relative changes are almost independent of the AC. All the other methods are derived from it, a kind of filtration of some parameters.

Although, if we take long enough periods (with a period multiple of a day), then instead of the tick volum we can take the sum (High-Low) of all candlesticks for a quite low frame period. These values will change proportionally

 
Mathemat >>:

Главный вопрос: какое окно для волатильности выбрать? Фактически это скрытый параметр координаты КК.

Второй вопрос: как вычислять волатильность?

1. this is really the main issue.

2. By the logic of the term, it should not be calculated by Close, but by High and Low. For example, I did it this way

      pos2= pos+ tau;
      DDist = High[ pos]-Low[ pos2];
      UDist = High[ pos2]-Low[ pos];
      if ( UDist > DDist) DDist = UDist;

Further you may calculate either simple average (module that is) or RMS.

 
Avals писал(а) >>

Shiryaev has a similar study of volatility, I think in Volume 2 of Fundamentals of Stochastic Financial Mathematics.

appears to be in the first volume. The statistics of ticks, and volatility, are discussed there. Here is a picture for example)))

 
Mathemat:

Come on, that's not greed. OK, let's talk about volatility.

=========

the topic has stalled...

Lovins and Gadflies rule.

:(

 
Sorento:

=========

the topic is stalled...

Lovins and Gadflys rule.

:(


It is not even known if anyone has productively put the main ideas of the theme into practice. Pity.
 

She drowned and surfaced.

I guess the end of the world is nigh - dead women are rising from their graves. Oh, well. We'll have to live with that...

===

I won't write anything else. It's pointless and useless. Like a Russian revolt.

Were you trying to talk some sense into someone?
Mercy - no, of course not.
What the hell is there to feed?
The one who's stupid?
-

Fire away. After I'm married...