Remembering veterans: Box and Jenkins - page 5

 
faa1947: (2) Create a TS. Calculate the residual and check for unit root. If the residue is non-stationary, we need a new TS, using the current one we have - it is hopeless and no positive tests, including forward ones, do not matter. It is bound to fail.


So you want to say that the TS that uses mathematical formulas must also be non-stationary? Non-stationary like the market? That is, the non-stationarity of the market must be compensated by non-stationarity of TS? In the residue we obtain a stationary function to trade by? How do you imagine it? Some kind of "synchronisation" with the market?

And what mathematical formula can give non-stationarity similar to market non-stationarity?

 
faa1947: What you are talking about is every TC's tactic, a matter of taste maybe.


But then how do you calculate the difference between a TS and a cotier, unless you predict on every bar?
 
LeoV:


So you are saying that a TS that uses mathematical formulas must also be non-stationary? Non-stationary like the market? That is, the non-stationarity of the market must be compensated by non-stationarity of the TS? In the residue we obtain a stationary function to trade by? How do you imagine it? Some kind of "synchronisation" with the market?

And what mathematical formula can give non-stationarity similar to market non-stationarity?


The formula has no non-stationarity - it is a property of a random variable.
 

faa1947: Формула нестационарностью не обладает - это свойство случайной величины.



I agree. But then, what you suggest is not possible )))
 
LeoV:

But then how do you calculate the difference between a TS and a quotient if you don't predict on every bar?
We are talking about history. Let's take the TS - the intersection of two bars. Remember, in the indicator buffers these are sets of numbers. On the chart they are smooth curves. We subtract the indicator buffers from the cotier and obtain the residue. This is what we are talking about. The wads are regular curves, randomness is out of the question. So the randomness is left in the residual, as the sum of the two dabs and the residual should give the quotient. This randomness left in the residual can be stationary or non-stationary.
 
LeoV:

I agree. But then, what you propose to do is not possible ))))
And talk?
 
faa1947: This randomness that is left in the residual can be stationary or non-stationary.

Agreed. But how do we use this in trading if we are trading in the future and not in the past. In the past it may be all good. How do we determine that the future will also be fine if the market is non-stationary?
 

paukas: А поговорить?

Talk is sacred ))))
 
LeoV:

I agree. But then, what you propose to do is not possible ))))

It's a matter of opinion.

In the Econometrics branch I gave tables with simulation results. There is an "ARCH" column there. Sometimes it is zero, sometimes it is not. But the residual is always stationary with respect to the available tests. They only reveal certain types of non-stationarity. It is reasonable to assume that there are some that we do not know about.

In any case. Two alternatives. Don't engage with non-stationarity and believe the forward test. Engaging and realising that a number of problems in this area have been solved.

 
LeoV:
I agree. But how do we use this in trading if we are trading in the future and not in the past. In the past, things can be good. How do we determine that the future will also be fine if the market is non-stationary?

Already answered. The TS works out the non-stationarity, that's how amazing the TS is.