Zero sample correlation does not necessarily mean there is no linear relationship - page 51
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
Alsu, anonymous help me understand. What does it mean? It turns out that the apparent positive correlation between Bid andAsk of any symbol is a fiction. And the negative correlation between the forward and reverse quotes is also something that can be thrown out, because it has neither stationarity nor ergodicity?
1. If you didn't know how the series in the examples were obtained, how would you recognise (preferably with an example) that they are not 100% correlated quantities but collinear quantities? Does a similar constraint (investigating for collinearity) apply to autocorrelation?
2. What is the arithmetic relationship between Bid and Ask?
P.S. "The farther into the woods the thicker the partisans" it would seem, we are searching for interrelation, and there, where this interrelation is most sharply revealed, ... baffle. And you just wanted to assess the correlation of the two rows at some point in the past.
It turns out that the apparent positive correlation between Bid and Ask of any symbol is a fiction. And the negative correlation between the forward and backward quotes is also something that can be thrown out, since there is no stationarity
You can throw it away. Use stationary rows.
(3) How to quantize data
Take the first difference between candles, it is not HP BP. Why should it be normally distributed if one candle has X trades and the other has 100X trades and all with different volumes. Digging into tick history, level II history? The deeper it goes, the more differences between brokers.
You can quantize by volume if you have access to it.
You may not quantize at all. Then the formula for the correlation will be different.
In any case, you will not achieve normality by price quantization alone.
Bids and asks are simply better offers and... and so on. Can they change if there is no actual trading? Sure. Can they remain unchanged if there is a trade? Yes, absolutely (partially executed). Midprice! What about the moments when the spread increases several times, what about midprice or best-band?
If you calculate by trade prices - the results will be noisy because of bid/ask bounce.
For midprice you can use it, if the instrument is liquid enough.
The best solution is to use the arithmetic mean between the expected prices of two market orders (buy and sell) of a certain preset volume. But we need Level2 data for that.
Collinear values can't participate in statistical calculations.
Not true :P It's just that different methods are used. For example, instead of linear regression you can use principal component regression.
Correlation is a constant. If each sample of two SVs for which the correlation is calculated is statistically the same as other samples from the general population of those SVs, then we can say that the two SVs are dependent. More precisely, their behaviour is similar. This holds for normally distributed SV.
If SVs are not normal, then cointegration is applied, when the characteristic of the mutual dependence of two SVs is not a number, but a series with certain properties.
The conditions for the applicability of correlation and cointegration are not written correctly. Correlation (in particular, rank methods) is applicable irrespective of the form of distribution, stationarity and ergodicity of random variables are sufficient. The tests for cointegration also do not depend on the distribution shape, only the same order of integration of random processes is required (the order must be greater than zero).
You can throw it away. Use stationary rows.
You can quantise by volume if you have access to it.
You could not quantize at all. Then the formula for the correlation will be different.
Anyway, quantization of prices alone will not ensure normality.
If you do calculations on transaction prices - the results would be noisy due to the presence of bid/ask bounce.
You may use midprice, if the instrument is sufficiently liquid.
The best way is to use the arithmetic mean between the expected prices of two market orders (buy and sell) of some preset volume. But we need Level2 data for that.
Not true :P It's just that different methods are used. For example, principal component regression can be used instead of linear regression.
The conditions for the applicability of correlation and cointegration are not written correctly. Correlation (in particular rank methods) is applicable regardless of the form of distribution, stationarity and ergodicity of random variables suffice. Tests for cointegration also do not depend on the shape of the distribution, only the same order of integration of random processes is required (the order must be greater than zero).
Of course, your remarks are more accurate than mine.
But.
I, I think my definition is more correct as the application is more clearly seen from it, and for me this is much more important than the purity of the definitions. In general, I'm trying to forget all those definitions I was taught at the institute and take the meaning of the term in the form of program code. Taking a specific code, for example R, and executing that code to calculate cointegration is the definition of that word. This, in my mind, is the only way to disassociate myself from the flourishing pseudoscientific diversity in Russian science. This reflects my desire for profit, not dissertation.
So if you would give the specifics of any package, preferably R, to back up what you say, it would be super valuable to me.
Guys, apply at least some of the above to your trading, and then evaluate the results statistically:)
Gentlemen, can you tell me if this data series is stationary or non-stationary?
Gentlemen, can you tell me if this data series is stationary or non-stationary?
And how many observations are depicted? Just two, or are there a dozen?
How many observations are depicted? Just two, or are there a dozen?
A lot. A lot.
Well then I(1), unsteady.