From theory to practice - page 412

New comment
 
Maxim Dmitrievsky:

Yes, I'll show you the results of the "memory" trading experiment later.

but it's not too soon, there's a lot of conditions to come up with... just for fun, but maybe something interesting.

No problem. Maybe Koldun and Alyoshenka will wake up. They'll help out.

 
Alexander_K2:

As a reminder, this histogram of the real tick flow:

suggests that market events (the appearance of tick quotes) also have a "memory". This goes nowhere. Diffusive Markov process theory then doesn't fit.

I need events to be without "memory". What's not to understand?

Bummer ....

You can't do that.

 
Renat Akhtyamov:

Bummer ....

That's not going to happen.

I got it, Rena. Well, I'm as stubborn as a sheep, so what's the big deal? At least I got as much out of the exponent as I could. It's not... I'm sitting on logarithms now. How could it get any more unmarked? It's beautiful! Schrodinger's cat and I have just begun our long journey down the spiral of time.

 

Н4H1

M15M5


M1 the diagrams are created with code like this...

      int size=10;
      for(int i=0;i<size;i++)
        {
         FileWrite(han,
         Array[i]/Array[10],
         Array[i+10]/Array[20],
         Array[i+20]/Array[30],
         Array[i+30]/Array[40],
         Array[i+40]/Array[50],
         Array[i+50]/Array[60],
         Array[i+60]/Array[70]
         );
        }
 
What's on the charts?

The distribution of increments between neighbouring bars on the highs (same for the lows), the distribution is sliced into 7 sections as follows:
the distribution up to 10 is normalized by the value of the distribution number 10 (i.e. the number of increments equal to 10)
the distribution from 10 to 20 is normalized to the distribution number 20, etc. up to 70
As you can see, we still see a certain law on M1, but the older the TF, the more and more chaos; on H4 the law in question is not observed at all.

That is what it means to have the right data, and it shows that the astronomical time is not relevant.

ZS In the code above the i counter is the increment module in pips.

 
Alexander_K2:

I took another look at that bar graph and converted my TS to logarithmic time intervals. For the first time! And straight to the real one.

And all in vain.

According to your histogram, Alexander, reading quotes in logarithmic time intervals is more deterministic than reading through the exponent. As for the sample itself, we should see the difference in kurtosis, skewness, dispersion and std deviation when reading different quotes. According to the idea the kurtosis should grow, the standard deviation should decrease, i.e. the process determination is growing, i.e. the process becomes less random. Also we need to look at histograms of tick increments for different readings. What is in them?

The more non-random the process, the lower its standard deviation, the narrower and higher the bell on the plot. Indeed, the spread of randomness relative to the mathematical expectation becomes more and more minimal.

Figures 25.3, 25.4

http://stratum.ac.ru/education/textbooks/modelir/lection25.html

PS. By the way,Dr. Trader pointed out exactly this.

 
Novaja:

On the subject of this:The more non-random a process is, the lower its standard deviation, the narrower and higher the bell on the graph. Indeed, the spread of randomness relative to the mathematical expectation becomes increasingly minimal.

In many cases this is true, for many processes from real life (not generated).

But not always, so it cannot be a general rule (law).

Take, for example, the cumulative sum of the GSF increments +1 and -1 (the infamous coin that Alexander is so afraid of). Get a random walk - the benchmark random process with no memory.

And its increments are two narrow peaks, no more)

I would recommend using these lectures with caution, they are somewhat "amateurish", very loose with the wording.
 


Alexander_K2:

I took another look at that bar graph and converted my TS to logarithmic time intervals. For the first time! And straight to the real.

And to hell with it.


Novaja:

In theory the kurtosis should increase, the std deviation should decrease, i.e. the determinacy of the process increases, i.e. the process becomes less and less random.

Alexander_K2, you should read quotations. First reading - once in 24 hours, second one - once an hour, next one - once in 24 hours, next one - once an hour ... )))
the process will be increasingly non-random.

what will the distribution be?

and most importantly, what will it give you?

 

It's quiet in my favourite thread....

Two experiences, these are the results. If anything is wrong please don't laugh as I have not done this before.


Experience 1


Experience 2


Histogram

In the first experience the ratio of stop size to profit size is 2/1 , in the second experience 1/2.

As I understand if increment is equal or lower (not observed) - 0.05 , then the success rate of positive increment is higher in the next step.

Except that -0.05 may appear several times in a row. Then we need to calculate when this value will appear with low probability.

 

(gasps) Okay!

Vivathe grail!


1...405406407408409410411412413414415416417418419...1981
New comment