From theory to practice - page 2

 

Yury Kirillov

How is it different:

...считывать значения цены с определенной частотой в независимости от того был ли это реальный тик или нет...

from

...выбираем константу t = 1 сек. и с этой частотой считываем тиковые котировки...

?

Alexander_K:


Every t=1 sec. we do not read the current price value, but a specific tick value. I.e., in 1 min=60 sec. you have the number of tick data for calculations, which will always be <=60 and will be floating, depending on trading intensity.


So it is not equivalent to : "read at 1 Hz?"

Is there some subtle difference that I don't see?

 
Yury Kirillov:

So it's not the same as "read at 1 Hz"?

Is there some subtle difference that I don't see?

We send a request at 1 Hz and only take the actual changing quote for the calculation.

So, when we receive all ticks we have in 60 sec. any amount of ticks data from 0 to infinity, when we receive data with frequency 1 Hz we have exactly 60 values per minute, and in our case we havein 60 sec. the amount of ticks data for the calculation <=60.

Nothing, Yuri - further will be harder, in fact, that is why I did not lay out all the text with calculations, and wrote gradually, so that people get used to the style of presentation.

 
Alexander_K:

We send a request with a frequency of 1 Hz, and for the calculation we take only the actually changed quote.

That is, when we receive all ticks we have in 60 sec. any amount of tick data from 0 to infinity, when we receive data with frequency 1 Hz we have exactly 60 values per minute, and in our case we havein 60 sec. the amount of tick data for calculations <=60.

That's all right, Yury - it will be harder to go further, that's actually why I do not put all the text with calculations at once, and I am writing gradually, so that people get used to the style of presentation.

I'm forced to return to the original text:

Alexander_K:
Уж сколько я тут начитался тем - как правильно организовать прием тиковых данных, важен ли каждый тик и т.д. и т.п. 

Вот если бы в формуле стояло просто W(x) - то да, надо было бы принимать каждый тик, 
а если бы W(x(t)) - то считывать значения цены с определенной частотой в независимости от того был ли это реальный тик или нет. 
Но, у нас именно W(x,t), а значит оба этих подхода к приему данных неверны. 
Верным будет следующий алгоритм считывания котировок - выбираем константу t = 1 сек. и с этой частотой считываем тиковые котировки.  Вот это будет правильно.

That is:

1. Taking every tick is not our method.

2. Readingprice values with a certain frequency, regardless of whether it was a real tick or not, isnot our method.

By the way, why not with a frequency of once per second?

3.Let's choose a constant t = 1 second and read tick quotes with this frequency - our method.


What is the difference between 2 and 3? What is the difference between "Readprice values at a certain frequency" and"Read tick quotes at that frequency"?

 
Yury Kirillov:

Forced to go back to the original text:

That is to say:

1. Taking every tick is not our method.

2. Readingprice values with a certain frequency, regardless of whether it was a real tick or not, isnot our method.

By the way, why not with a frequency of once per second?

3.Let's choose a constant t = 1 second and read tick quotes with this frequency - our method.


What is the difference between 2 and 3? What is the difference between "Readprice values with a certain frequency" and"Read tick quotes with this frequency"?

Point 3 is exactly that and nothing else.

All other tick reading methods certainly have the right to live, but they have nothing to do with solving the equation.

 

Let's check it out.

What the price will be on Friday on the cadjpy

 
Mickey Moose:

Let's check it out.

What the price will be on Friday on the cadjpy

:))) This is for the astrologers.

We can only talk about the probability of price being in a particular range within the sample volume. For cadjpy it is approximately 12000 ticks. That is approximately in the range of 4 hours.

 
Alexander_K:

:))) This is for astrologers.

We can only talk about the probability of price being in a particular range within the sample volume. For cadjpy it is approximately 12000 ticks. I.e. approximately in the range of 4 hours.


then what's the point of this modelling if it doesn't work?

Is it boring?

 
Alexander_K:

We send a request with a frequency of 1 Hz, and for the calculation we take only the actually changed quote.

I.e. if we receive all ticks we have in 60 sec. any amount of ticks data from 0 to infinity, if we receive data with frequency 1 Hz we have exactly 60 values per minute, and in our case we havein 60 sec. the amount of ticks data for the calculation <=60.

On the micro level, i.e. on tick quotes the amount of information noise is high. If we separate the really important data from the noise on tick quotes, the percentage of necessary data for the calculation is less than 50%, and it indicates the futility of operating with tick data.

And the question is not if this is necessary or not... the question is: how can we trust the data which is so % noisy? It's a difficult task to sift this noise from tick quotes...

One could stop at this option, if there were no simpler and more reliable options...

 
Alexander_K:

Let's continue.

So, it is very important for us to understand the physical and mathematical meaning of EVERY variable in the Fokker-Planck equation, which we, on top of that, have complicated with an additional integral term.

1. The probability density W(x,t) is the probability that at some point t the price MAY have a certain value . Moreover, for any volume of sample the values of price lie in the ranges defined by Chebyshev's inequality.

2. Price x is the value of Ask or Bid at a certain moment t. Moreover, it is exactly the tick quotes, i.e. if there were no deals, the price is not read over time and does not participate in calculations. Please note that ANY attempt to filter tick data does NOT destroy the non-mark sequence. so you can simply work with clean quotes and not complicate the task. It's complicated enough as it is :))

3. Time t. No, I would even say TIME t. This is the most important parameter! I've read so many topics here - how to properly arrange receipt of tick data, whether each tick is important, etc., etc. If the formula contained just W(x) - then yes, I'd have to receive every tick, and if W(x(t)) - then it would read the price with a certain frequency, regardless of whether it was a real tick or not. But, we have exactlyW(x,t), which means that both of these approaches to receiving data are wrong. The correct algorithm for reading quotes is as follows - we select a constant t = 1 second and read tick quotes with this frequency. This will be correct.

The left part of the equation is sorted out.

The question - where is the practice, where are all these lots, profits and money at the end of the day? The answer - it will be, it will definitely be. Be patient.

1) W(x,t) = 0

and there is no other way.

and it will never be greater than zero

for!

the market is struggling with the volume of purchases and sales.

2) the deviation of the price from some line is also an inapplicable notion

for!

there is a trend and a flat

 
Serqey Nikitin:

The question is not whether it is necessary or not... The question is: how can one trust data which is so % noisy? It's an arduous task to sift out this noise in tick quotes...

There is no difficulty at all in sifting out this noise. But I agree that this task (elimination of tick noise) is not necessary to solve at all.