The market is a controlled dynamic system. - page 32

 
Nafany:
Which one?
Why should they have personal names? Ask Peter.
 
paukas:
Sure. At least three.

There's something about the big picture I don't understand. Increasing the sample size makes the estimate less biased. But only if the SV is normally distributed. See for 118 hours:

And now for 2000 hours:

The distribution has become non-normal at all!

But if you take two more differences from the difference in the private ACF a dependence appeared. Could it be a source of profit?


Note that we strictly reject the hypothesis of no dependence between rerurns(3).

 
faa1947:
.....

The distribution has become completely abnormal! ....


Compare it with normal. What is abnormality?
 
paukas:
Compare with normal. What is the non-normality?

Figures below: according to Jarque-Bera, the probability that it is normal is zero - we strictly reject the hypothesis that the distribution is normal. Also look at the bias and kurtosis.

Compare:

 
faa1947:
Figures below: according to Jarque-Bera the probability that the distribution is normal is equal to zero - strictly reject the hypothesis of distribution normality. Also see bias and kurtosis.

That's a very good picture.

Now take a moving average of say 100 hours and calculate separately for the bars above and below it. I wonder if the asymmetry will appear?

 
paukas:

That's a very good picture.

Now take a moving average of say 100 hours and calculate separately for the bars above and below it. I wonder if the asymmetry will appear?

What kind of moving average? From ARIMA? What do you mean "for the bars below and above them"?
 
faa1947:
What is the moving average from ARIMA? What do you mean "for the bars below and above them"?

The normal 100 - period average on bar opening prices.

Separately, we calculate the distribution for the bars opened above average and opened below average. We should get two images.

 
paukas:

Normal 100 - period moving average on bar opening prices.

Separately, we calculate the distribution for the bars that opened above the average and those that opened below the average. We should get two pictures.

Here is a picture from the terminal. Naturally there is a lag.

There is higher, there is lower. On a falling one it is lower, on a sideways one in half, on a rising one higher. What to count here is all well known.

 
faa1947:

Here's a picture from the terminal. Naturally there is a lag.

There is higher, there is lower. On a falling one it is lower, on a sideways one in half, on a rising one higher. What to count here is all well known.

No, you have to count over three years too. You can't see anything by eye.
 
paukas:
No, not in three years either.

H1 in three years does not fit in the terminal.

I can imagine the result theoretically. The market will be sideways, which means we will get an approximate equal number of openings and closings. If it's not so, then we'll get that the market has been growing (falling) for three years on the average. So what? Three years is for portfolio managers. I'm interested in the forecast one step forward. The conversation started with the fact that returns(1) is a random walk - a prediction is not possible. You need at least a random walk with drift. For returns(3) a dependence has been found - so a profit has emerged.