Building a trading system using digital low-pass filters - page 8

New comment
 
Mathemat:
Prival: stationary in both the narrow and broad sense. can=constant, sko=constant.
Wow. Privalych, you've made me happy. I'll sleep well now. Thank you, my dear. Of course you went too far, but the narrow one is enough for me. Let the MO, RMS and AF be constants (statistical), and the rest - to hell with it ...

P.S. And how did you determine that what came out is BGS (strictly)?


ACF + ACF spectrum.

And if I am right that it is a GSC, then it is also stationary in the broad sense, because all its parameters are determined by the two numbers MAJ and sko, and if they do not change over time (deviations must lie within the confidence interval of the estimates of MAJ and sko for the final sample), then the process is also stationary in the broad sense.

 
Wow, that's a hell of a thing. A halt, this is very serious and on a universal scale.

I don't care if the trend is getting killed. That's not what this is about. It's testing. It is stationarity in the narrow sense that matters to me (MO+SCO+ACF=const, and I don't care about the rest). Prival, this stationarity (narrow) has to be strictly proved. Not purely visually, but strictly.
 
Mathemat:
Wow, that's some serious shit. A halt, this is very serious. I don't care if the trend is getting killed. That's not what's useful here. In testing. It is stationarity in the narrow sense that is important to me (MO+SCO+ACF=const). Prival, this stationarity has to be strictly proved. Not purely visually, but strictly.

OK. Well, I will try to check tomorrow by Pearson's Neumann criterion. But I still do not understand how to do it without a trend ? Alexey the modelling methodology is not clear.
 
mql4-coding писал (а):
I.e. there is only one problem - correct spetroanalysis of a non-stationary series.

Absolutely right. One problem, but a very big one.


Guys, in fact (I've looked through everything written up to the end of the thread) the series of first differences is not really stationary, although it has many of the attributes of stationarity. The point is that it does have cyclicities. These cyclicities lead not only to changes in Mo, but also, what is sadder, to changes in the dispersion at the counting points. It is not easy to stop such a series.

 
NorthernWind:
mql4-coding wrote (a):
I.e. there is only one problem - correct spetroanalysis of non-stationary series.

Absolutely right. One problem, but a very big one.


Guys, in fact (I've looked at everything written up to the end of the thread) the first difference series is not really stationary, although it has many of the attributes of stationarity. The point is that it does have cyclicities. These cyclicities not only lead to changes in mo, but, what's worse, to changes in dispersion at reference points.


The cyclicities should show up in the spectrum, but they don't seem to be there visually. That's what I was building the spectrum for. Can you elaborate on how you determined the presence of cycles? I have to go to work now, I will try to post the check that I promised in the evening.
 
What do these pretty pictures and the stationarity criterion have to do with it? Prival has shown conclusively that the difference returns are reducible to GVH, and the process which is GVH is stationary.
 
While mo cyclically depends on the time of day and dispersion cyclically depends on the time of day. And this does not speak in favour of stationarity. If everything were stationary, we would see straight lines instead of these curved graphs. The variance would be fine if the distribution were normal, which is not fatal though not pleasant, but the distribution is abnormal.
 
NorthernWind:
PPPS. The red graph, EURGBP, seems that somewhere around 7-12 o'clock non-Moscow time there is a small mismatch between the price movement and the number of ticks. Why would that be?

You may have noticed that this is not the only "discrepancy" ....

I think (IMHO) it has to do with the fact that the European session is opening and people are trying to decide "where we'll trade today".

It means that both "bulls" and "bears" are trading and everyone is trying to move the market in their own direction.

Thus, the number of ticks exceeds the number of candlesticks movement.

After 12 o'clock the market has already moved in this direction.

 
Prival: Ok. I'll try tomorrow using Pearson's Neumann criterion. But I still don't understand how to model without a trend ? Alexey the methodology of modelling is not clear.
I described this method briefly here: https://forum.mql4.com/ru/9358/page6#51829 . It also says what I need it for.

And the trend still appears when integrating a series of returns (it recovers unambiguously).
 
NorthernWind:
While mo cyclically depends on the time of day and dispersion cyclically depends on the time of day. And this does not speak in favour of stationarity. If everything were stationary, we would see straight lines instead of these humped graphs.
Those pictures really say nothing about stationarity. They show very useful dependencies, but they are dependencies for averages. To check their stationarity, you either have to fit them and check the constancy of the fit parameters over a long time interval, or check the constancy of the MO and variance for each point in the graph. For each point, it means taking a chronological series of instantaneous values, say, for 7 hours, and checking this particular series for stationarity. And so for each hour. Or, if we are not interested in intraday details, we can check stationarity of the chronological series of the daily average.

P.S. Clarification. This will be less ambiguous, if not clearer.) "take a chronological series of instantaneous values, e.g. for a year, let it be, say, a seven-hourly point, and check this particular series for stationarity".
123456789101112131415...32
New comment