Econometrics: let's discuss the CU balance sheet. - page 10
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
what does it mean to be stationary? How is it defined?
well, here we go - I have to repeat the first year curriculum...... And I thought all bearers of "secret knowledge" knew the school curriculum - but you can't chop up a long enough series into chunks and compare MO and variance? Too complicated?
hooper to give?
what does stationary mean? How is it defined?
The simplest definition: mo = constant, variance = constant.
Defined by the unit root test, of which there are many.
How complicated? Instead of symbols "trend" I wrote "HP".
But there are more serious considerations. The analytical straight line smoothing formula (more accurate than detrending) is very much dependent on the sample size. Let's take the EURUSD sample since 2000. Let's isolate the trend as a straight line. almost a horizontal straight line, but with deviations of about 2500 pips! This is exactly what the machine writes - the average hospital temperature. But if we take any filter we will get variance of tens of pips. Since we are not trading on time intervals of 10 years, we can do with a straight line when smoothing out 50-100 observations. But some estimations require more observations. I always apply a filter to avoid getting into details. Purely a practical consideration.
so it is understandable, to detrend the initial series, but for equity it is desirable to trend in one direction and more or less constant.
The simplest definition: mo = constant, variance = constant.
Determined by the unit root test, of which there are many.
well, here we go - have to repeat the first year's syllabus...... And I thought that all bearers of "secret knowledge" knew the school curriculum - but you can't chop a long enough series into pieces and compare Mo and variance? Too complicated?
hooper to give?
Well, with mo=constant and variance=constant it is absolutely clear what the model should be and what the deterministic component should be. I.e. a linear trend.
Well, with mo=constant and variance=constant it is absolutely clear what the model should be and what the deterministic component should be. I.e. a linear trend.
Well, in fact, stationarity - Mo and variance are not constants, but should of course float, but no further than certain limits..........
Well, in fact, stationarity - MO and dispersion are not constants, and should, of course, float, but not beyond certain limits..........
Well, that's what I replied to the automaton above.
So, you write:
"A model without normality will be correct and adequate with a certain accuracy if the series is stationary. "
If the series is stationary, then by subtracting the trend (mo), the residual will be normal. I.e. the residual analysis is the assessment of robustness or stationarity of the distribution (which is actually the same thing).
P.S. the first differences are stationary and the equity series itself has a unit root
Well, give me an example of a "good one" from your point of view, where the residuals are not normally distributed
Draw a trend line with an upward slope. Now overlay the different noise components with different distributions -- uniform, normal, binomial, Cauchy, geometric, logistic, Poisson, Weibull, ..... (continue?)
Now think -- does the type of residual distribution determine the trend component?
Draw a trend line with an upward slope. Now overlay the different noise components with different distributions -- uniform, normal, binomial, Cauchy, geometric, logistic, Poisson, Weibull, ..... (continue?)
Now think about it -- does the type of residual distribution determine the trend component?
No, it doesn't. But if deviations from this wonderful trend introduce you to kolyan... We don't want to meet him here. That's what this is about.
Draw a trend line with an upward slope. Now overlay the different noise components with different distributions -- uniform, normal, binomial, Cauchy, geometric, logistic, Poisson, Weibull, ..... (continue?)
Now think about it -- does the type of residual distribution determine the trend component?
Shit, that's what I was talking about. But I wouldn't trade this shit with Cauchy distribution, because the variance and mo are undefined ;) The trend component depends on it and it's not the point. It's about identifying the trend component and trusting it.
Draw a trend line with an upward slope. Now overlay the different noise components with different distributions -- uniform, normal, binomial, Cauchy, geometric, logistic, Poisson, Weibull, ..... (continue?)