From theory to practice - page 29

New comment
 
Евгений:

I wonder, philosophically, why the future should depend on the past. Although the future is a continuous transition from the present, as I understand it is also a non-marking process.

So what are the chances that in the future a brick will not fall on your head (even while working on a construction site), if it has not happened before and up to the current second in the present and we have taken this history into account in our formulas.

Or what is the probability that Vasya Pupkin, driving his car for a long time along the same route, will not change it in the future and every time this route will not be repeated.

It's just a thought, if anything!

Because you are twisting the facts Consciously or not it is a matter of psychology
 
Alexander_K2:


Someone - tell me how the NEPARAMETRIC coefficient of kurtosis is calculated!!!!!


Coefficient of excess

 
Alexander_K2:

A-ya-ya-ya-ya-ya! Well, how come, eh? :)))) I thought we all understood... :((((

I'm not talking about understanding, but that you suggest "aggregate current and historical variance" as something new, even though the same bollinger makes a similar comparison, just in a different implementation.
 
Alexander_K2:  Someone - tell me how the NEPARAMETRIC coefficient of kurtosis is calculated!!!!!
Probably the same as any other criterion - by replacing values with their ranks?
 
Nikolay Demko:

The coefficient of kurtosis


No, Nikolai - what is needed is a non-parametric one, like a non-parametric skew. I know it counts somehow. I've seen it in English-language literature, but I can't find it now.

 
Alexander_K2:

No, Nikolai - what is needed is a non-parametric one, like a non-parametric skew. I know it counts somehow. I've seen it in English-language literature, but I can't find it right now.


https://smart-lab.ru/uploads/images/00/00/16/2012/09/17/3119fe.gif

http://documentation.statsoft.com/STATISTICAHelp.aspx?path=Nonparametrics/NonparametricAnalysis/Dialogs/StartupPanel/NonparametricsStartupPanelOrdinalDescriptiveStatistics

 
Nikolay Demko:

https://smart-lab.ru/uploads/images/00/00/16/2012/09/17/3119fe.gif


Nikolai, you're a genius!

Get to work everyone, my friends!

Sincerely,

Alexander.

 
Alexander_K2:  Can someone describe exactly the physical meaning of choosing MA or EMA or WMA?
By the way, period averages are a bad choice. They show something adequate if the data changes little over the sample period, and does dance around something average, simply put in a flat. But there are trends in price all the time, and on a trend, the average is very much lagged, and stands very far away from the price. It's a bit like hammering nails with a screwdriver, the tool is not up to the task. MA simply "does not know" what a trend is, there are no such notions in this model. The notion of a trend is present, for example, in Holt's moving average, but it is a bad tool too, because it makes lots of errors on trend reversals. If we are going to approach the question academically, we should probably take as trend an approximation of a smooth line.
 
bas:
By the way, period averages are just a bad choice. They show something adequate if the data changes little over the sample period, and really dances around something average, simply put in a flat. But there are trends in price all the time, and on a trend, the average is very much lagged, and stands very far away from the price. It's a bit like hammering nails with a screwdriver, the tool is not up to the task. MA simply "does not know" what a trend is, there are no such notions in this model. The notion of a trend is present, for instance, in Holt's moving average, but it is a bad tool too, because it makes lots of errors on trend reversals. If we are going to approach the question academically, we should probably take as trend an approximation of a smooth line.

I work exclusively and only with WMA, where weights are calculated from the t2 probability density function of the Student's distribution for that particular currency pair. And to do this I need to know exactly the nonparametric standard deviation of a particular t2-distribution of increments for a particular pair, which I've learned to calculate only numerically from persentiles. Tiresome calculations! But they give a very accurate moving average behaviour.

So if anyone thinks it's all in the bag, they're wrong. They still have a lot of work to do.

 
bas:
By the way, period averages are a bad choice. They show something adequate if data changes insignificantly over the sample period, and really dances around something average, in other words, in a flat. But there are trends in price all the time, and on a trend, the average is very much lagged, and stands very far away from the price. It's a bit like hammering nails with a screwdriver, the tool is not up to the task. MA simply "does not know" what a trend is, there are no such notions in this model. The notion of a trend is present, for example, in Holt's moving average, but it is a bad tool too, because it makes lots of errors on trend reversals. If we are going to approach the question academically, we should probably take as trend an approximation of a smooth line.

I told Alexander at the very beginning that all these MA...WMAs have very big group and phase delays, and in most cases the average etc. is shown exactly the other way round. Zero emotion.)) And hence, all distributions are deliberately distorted.

A polynomial regression line is not bad, but then this line must be recalculated at each new point, which is not good. And you need only the last points.

1...222324252627282930313233343536...1981
New comment