Discussion of article "Econometric Approach to Analysis of Charts" - page 2
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
Add. We'll republish.
Wrong, and wiki has errors. Take the uniform distribution. You can calculate the kurtosis, but there are no tails at all. So you have been correctly told that in the general case there is no relationship between gravity, length, the very presence of tails and the kurtosis coefficient. Take a triangular distribution - the same thing.
What exactly is wrong?
-Alexey-, honestly, in the field of time series research, in particular the series of returns I described in the article, I have never encountered such distributions as you are talking about (triangular and uniform) at all. It is very likely that they serve a completely different purpose. So no need to refer to some marginal examples, I can't find another word, pardon me. And if you assert something, be kind enough to give a concrete analytical example.
The distributions that are most often, if not always, used in the above analysis are the normal distribution, the Student distribution and the Cauchy distribution.
...in order to use this coefficient, it is necessary, firstly, to determine the empirical distribution function, which is not a trivial task from the point of view of mathematics, and secondly, within the framework of some probabilistic criteria, to test and accept the hypothesis that the distribution has a single mode, which is not always fulfilled for financial series and is also a non-trivial task. Without carrying out these steps, further calculation is impossible, if it requires the "Excess coefficient".
Valuable observation. If one were to do Distribution Fitting in an article, there would be enough material for a second one. Perhaps in the future either I or someone else will write an article on this topic. The purpose was simply to point out the features of a time series that require the use of non-linear models.
-Alexey-:
Nothing is written about what to do if the test says that the series is not applicable for the assumed model.
Nothing is said about why and on what basis a series of so-called "yields" and not some other one is used.
Which ones are attractive and for what purpose? Why attractive and not expedient? What is a valid mathematical criterion for choosing the type of conversion?Yes, nothing is written :-)
Because the logic is simple. If a test, here a Q-test, shows no autocorrelation, then there is no point in using a non-linear model. Use linear ones.
The series of returns is used on the basis that the stationarity of the time series is ensured. This is important, for example, for subsequent modelling.
The asset price is non-stationary. Yields are usually stationary.
In addition, it is possible to compare different series with each other, i.e., figuratively speaking, to cross hedgehogs with hedgehogs.
Usually, the frequency of the series is relatively high (daily, hourly, etc.), so the series of returns can be obtained either by continuous compounding (which is what we did) or by simple compounding. The difference would be small.
I think the question of the difference between the terms "attractive" and "expedient" is debatable.
Do you think critics should attend thesis defences, or should they just accept it without defence and start practical research? I think you will be interested to know that the correlations are not calculated correctly. Criticism is meant to make the material better, no one is arguing that the article is good. So that the profit on your systems is not small, but more.
Criticism is very necessary. For in a dispute we find the truth, as a rule.
-Alexey-, and what is the incorrectness of the calculation, can you specify?
Autocorrelation in yield series is usually weak or absent.
What exactly is wrong?
-Alexey-, honestly, in the field of time series research, in particular the series of returns I described in the article, I have never encountered such distributions as you are talking about (triangular and uniform) at all. It is very likely that they serve a completely different purpose. So no need to refer to some marginal examples, I can't find another word, pardon me. And if you assert something, be kind enough to give a concrete analytical example.
The distributions that are most often, if not always, used in the above analysis are the normal distribution, the Student distribution and the Cauchy distribution.
They may be used most often, but the real distribution is not normal, Cauchy or Student, but is usually closest to the Laplace distribution. It was correctly stated above that the task of identifying the distribution is non-trivial, but that doesn't mean it's impossible. In any case, I am personally convinced by the tests of conformity to the double exponential with the result R^2 = 0.999 quite well.
Now about this graph:
Here, by the way, only Laplace distributions are drawn, which means that they all have an excess ratio of exactly 3. Thus, it again has no connection with the "thickness of tails" - it is the same for all graphs presented here.
PS I would also talk about econometricians reprinting each other from one textbook to another and then to paedivikia, but okay, I'll leave it till next time.
question for the author of the article.
You interpret the Lewing-Box test as follows (quote):
The correct interpretation according to the criterion definition should be as follows:
They may be used most often, but the real distribution is not normal, not Cauchy or Student, but usually closest to the Laplace distribution. It was correctly stated above that the task of identifying the distribution is non-trivial, but that doesn't mean it's impossible. In any case, I am personally convinced by the tests of conformity to the double exponential with the result R^2 = 0.999 quite well.
Now about this graph...
Here, by the way, only Laplace distributions are drawn, which means that all of them have an excess ratio of exactly 3. So it again has no connection with the "thickness of the tails" - it is the same for all the graphs presented here.
alsu, I agree that the Laplace distribution always has an excess ratio of 3. I was hasty with its estimation, because I haven't seen it for a long time... But once again I repeat that econometricians in the field of research I wrote about use these distributions. If Nobel laureates are not authorities for you (e.g. Robert Engel), then I will pass.
If you do not give a concrete analytical example, I consider the argument speculative.
denkir:
What exactly is wrong?
1) Each coefficient is determined using a different amount of data, i.e. they are statistically unequal. Therefore, each coefficient should be tested for significance separately from the others. This is not the case in the Ljung-Box test.
2) The significance level for the test is chosen based on what - just like that?
-Alexey-, honestly, in the field of time series research, in particular yield series, which I described in the article, I have never met such distributions as you are talking about (triangular and uniform) at all. It is very likely that they serve a completely different purpose. So no need to refer to some marginal examples, I can't find another word, sorry.
So you haven't answered - where did this series of returns come from, what is the justification for the choice of transformation? No doubt, you can get a straight line from the price chart with the help of transformations, but what is the use? I read what you're writing:
The series of returns is used on the basis that the stationarity of the time series is ensured.
Such an approach is just a terrible cluttering with a great loss of information. What is there to forecast then? In the language of mathematics it is called - fitting the data to the forecasting method. But this is not the way to do it. The method can be used only if the initial data are acceptable for it, and not to cut them so that they become so. This is a well-known problem of modern statistics.
And if you assert something, be kind enough to give a concrete analytical example.
Here is a concrete not analytical, but a practical example on the current situation of the Euro at 4 hours. Distribution over a number of residuals that I get by other transformations, and I know why. You can see that the distribution is close to triangular.
And its shape can change in the most bizarre way, because this series is non-stationary. Where do you get the idea that the price series is distributed according to the
the normal distribution, the Student distribution and the Cauchy distribution.