Machine learning in trading: theory, models, practice and algo-trading - page 3642

New comment
 
Aleksey Nikolayev #:

To begin with, it is worth to define what we are talking about - a specific analytically defined deterministic function, which we start to consider as a random process, or a time series, which we obtain from this function by taking its values at discrete moments of time and studying it by means of matstat.

In the first case there is no unambiguous stationarity due to time dependence.

In the second case, the answer about stationarity will depend on a) the time segment where discretisation is done, b) the frequency of discretisation.

Here I will conclude my participation in the discussion of such a fascinating and full of sense topic).

I am of course about sampling the data, we assume we don't know the generating process. I just generated the data with a single time step as originally plotted and calculated all the characteristics.
 
Maxim Dmitrievsky #:

Yes, but it is treated as a non-linear trend :) and series with a trend are non-stationary.

Right, the presence of a trend is evidence of nonstationarity. But our function is bounded on the unit segment and there is no trend in it.
If we add a trend to a series with seasonality, the series will be non-stationary, but not due to seasonality, but due to trend.
 

What are you guys discussing?

Stationarity, non-stationarity ... it's all about random processes.

Once again, Dick is just trolling.

Here's his quote from his original post:

So, the point of proper learning is to find the correct formula for the process, not its shadow, then it won't matter if the process is stationary or not. So what was wrong with the W1, W2, W3 example? - Or the opposite is true, but where does the golden key to neural network training lie?


Once again the hundred times discussed: signal + noise is being trolled.

And this is trolling, because for at least 10 years everyone has agreed that there is NO signal at all on financial markets.

And the thread has been led into nonsense again.

 
Evgeniy Chernish #:
Right, the presence of a trend is evidence of nonstationarity. But our function is bounded on the unit segment and there is no trend in it.
If we add a trend to a series with seasonality, the series will be non-stationary, but not due to seasonality, but due to trend.

The ph-ya is defined from - to + infinity, if you take more samples, then as a nonlinear trend in it is a sine wave.

We're talking purely about time series now. Subtract the sine wave, then we get a stationary series.
 
Maxim Dmitrievsky #:

The ph-ya is defined from - to + infinity, if we take more samples, then as a non-linear trend it has a sine wave.

We're talking purely about time series now. Subtract the sine wave, then we get a stationary series.
It's defined at -+ infinity, but it maps to a single segment. All I was talking about was sampling this function with sampling step = 1.
For me, the sampling of this function is a stationary process without any transformation.
 
Guys argue about what they cannot prove.

"We decided 10 years ago that there is no signal in prices".



Postulates are postulates, but you should start with the elementary: how does a price chart differ from a completely random chart with an infinite value on one side (+ infinity).

What are the technical limitations of pricing that give rise to serpentine/channel price movement in time (1-2-3-10 points of average deviation from the last price 99% of the time), and does it give rise to regularities.

What does it give rise toat the macro level, why does geometry suddenly work (Wolf Waves), why do flat systems have more sets when optimised in the strategy tester and are better retrained. The cause of these phenomena, the cause of the causes and so on.

Maybe somewhere in these areas to connect AI, MO, NS, LLM, etc.


And you are all about signal/noise in a non-physical environment, and the first 10 values of RSI shove to the input and think that its "stationarity" will take out the model with the fear of "not too many inputs, suddenly it will overtrain.

 
Evgeniy Chernish #:
It is defined at -+infinity, but maps to the unit segment. All I said was about sampling of this function with sampling step = 1.
For me, the sampling of this function is a stationary process without any transformation. 


import numpy as np
from statsmodels.tsa.stattools import adfuller
import matplotlib.pyplot as plt

# Создаем данные
x = np.linspace(0, 10, 1000)
y = np.sin(x)/4 + np.cos(x**2)/4 + 1/2

# Выполняем тест на стационарность (тест Дики-Фуллера)
result = adfuller(y)

print('ADF Statistic:', result[0])
print('p-value:', result[1])
print('Critical values:')
for key, value in result[4].items():
    print('\t%s: %.3f' % (key, value))

# Визуализация ряда
plt.figure(figsize=(12,6))
plt.plot(x, y)
plt.title('Временной ряд')
plt.xlabel('x')
plt.ylabel('y')
plt.grid(True)
plt.show()
It works like this.
 

and so.

# you code here

import numpy as np
from statsmodels.tsa.stattools import adfuller
import matplotlib.pyplot as plt

# Создаем данные
x = np.linspace(0, 5000, 5000)
y = np.sin(x)/4 + np.cos(x**2)/4 + 1/2

# Выполняем тест на стационарность (тест Дики-Фуллера)
result = adfuller(y)

print('ADF Statistic:', result[0])
print('p-value:', result[1])
print('Critical values:')
for key, value in result[4].items():
    print('\t%s: %.3f' % (key, value))

# Визуализация ряда
plt.figure(figsize=(12,6))
plt.plot(x, y)
plt.title('Временной ряд')
plt.xlabel('x')
plt.ylabel('y')
plt.grid(True)
plt.show()
#

I told you, I took a function with a sampling step equal to 1


DF

 
Evgeniy Chernish #:

and so

I told you, I took a function with a sampling step equal to 1


So yes, because the cycles are filtered )

 

Regarding sine plus noise. It's a stationary process with zero MOI.

And in the example you gave, the MO at a particular point of the process is considered. Naturally, at point t = 1, for example sin(t) = 0.841.... + noise with zero MO, so we have Mo at this particular point = 0.841.

But we observe the process at different moments of time t=1,2,3,4.... and we count Mo over the whole trajectory of the process for this period of time, not at a particular point.

I wonder who is the author of the textbook ?

1...363536363637363836393640364136423643364436453646364736483649...3741
New comment