Econometrics: why co-integration is needed - page 9
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
faa1947, you are in the middle of a cointegration fit. At one time hrenfx was doing this akivno in various threads and even posted various inikators for this type
The classic use of cointegration is spread trading and arbitrage as one of its variants. As well as any other mean reversion type methods. The fitting is difficult to achieve something - you need to understand the reasons for cointegration. They may be economic, for example, or speculative. The latter are based on the fact that certain participants compare these assets, believing them to be related, and buy undervalued assets and sell overvalued ones. In other words, they are working on spread shifting. These actions are the fundamental basis of occurrence, which is modeled by error correction and the Johansen test.
The question to be answered is why one trades on the convergence of these assets. Then the spread can be returned on the right side.
Read my first post in this thread.
Dear econometrician, read the definition I gave earlier.
Read, after all, wikipedia, which says in normal language that "Before the 1980s many economists used linear regressions on (de-trended[citation needed]) non-stationary time series data, which Nobel laureate Clive Granger[1] and others showed to be a dangerous approach that could produce spurious correlation".
Co-integrated in the case of a buy-and-hold strategy. Not cointegrated in the general case, although you can build examples of profit series when it is cointegrated.
There is no connection at all with the randomness and non-randomness of the profits here.
http://www.burns.com/wcbspurcorl.htm
Well try to prove out-of-sample. I'll give you a clue as to the errors...
Where is this index traded? Or at least futures on it? Now your results are proof that for any series you can construct another one, and that pair of series will be cointegrated.
Believe me, I have read and (unlike you) have a practical idea of how to apply it. I have even written for what and how cointegration is applied.
Co-integration and correlation are different concepts altogether. One can exist without the other, and in no way does either concept entail the existence of the other.
So you've chosen the wrong model. Use a non-linear model, e.g. polynomial regression, and you will be happy.
I will disappoint you with the following fact. Fitting has nothing to do with the independence of random variables. There is a strict definition of independence in the theorist.
Reread on a fresh head, but the impression is from yesterday.
Your demonstration that you are a more literate person than I am - I admit, I won't be deterred.
But where is the constructiveness?
Links to portfolios in this forum are not interesting.
Links to Wiki are just as uninteresting as any other books, as the meaning of the word "cointegration" in this thread for me is the program that calculates that cointegration. Everything else is just opinion.
The trend is towards cointegration. You missed it in my post.
At one time hrenfx did this akivno in various threads and even posted various inikators for this type
Your reference to this man is completely unexpected to me. At one time I caught him putting his own meaning into a generally accepted term, and then began to insist that he was entitled to it. After that I am not interested in any of his statements.
You are engaging in cointegration fitting.
What is fitting or not fitting? I take it and calculate it. That there is an economic link between the Eurodollar pair and the dollar index is obvious to me. I checked this obviousness - I managed to choose such a vector that the difference between these two assets is stationary.
The Mathematician above gave us the link to the topic of spread trading. There was an indicator there. I have a stationary quote as the difference of two non-stationary quotes. It's an aid to spread trading. But I don't understand where the entry is and where the exit is.
Here is the graph.
For the upper part we create a trend-following trade? Is it so? I do not understand.
Or on the overbought/oversold principle?
What is fitting or not fitting? I take it and calculate it. That there is an economic relationship between the Eurodollar pair and the dollar index is obvious to me. I specified this obviousness - I managed to find such a vector, that the difference between these two assets was stationary.
That is the fit. The fit itself is a tool, and it cannot be good or bad. Use can be right or wrong. Correct if there is a high probability that the properties will be maintained. Even your unloved hrenfx tested models on data that did not participate in the "fit". That's one way to check robustness. Have you even done this?
Here's the chart moved.
For the top, do we build a trend trade? Is it like that? I do not understand.
Or on the overbought/oversold principle?
Of course overbought/oversold. If there is trending then use that as well. I.e. overbought oversold only in the direction of the trend.
Even your unloved hrenfx tested the models on data that did not participate in the "fit". That's one way to check robustness. Have you even done that?
No, I haven't, I haven't seen the need for it yet.
Once again, I would like to make my point about the forward test. This is not directed at you, but for the sake of the truth.
Forward test for the TS, the balance of which is not stationary, is a self-deception, it does not tell us anything. Half of the forum is moaning about drawdown, while TS systems are always tested in forward test! Did the test, did the forward test - answer why you trust those tests. You don't want to answer, can't - silently fund your account and keep doing forward tests.
Of course overbought is oversold. If there is trending, then use that as well. I.e. overbought oversold only in the direction of the trend.
That's interesting, but you will have to pick the assets you want to trade.
It turns out that when the overbought zone is reached: asset1 we sell and asset2 we buy. Then vice versa, right?
Have you even done this?
And the forward test of what? That the cointegration remains?
I'm sure the calculated cointegration won't stay. Without the forward test (another proof of the dubiousness of this tool).
This confidence is based on the fact that both initial series are I(1), which is not necessary. eurodollar I have seen both I(2) and I(3), and both assets should have the same order of integration. The problem is an old one. What's behind the right-hand edge of the quotient.
It turns out that when the overbought zone is reached: asset1 is sold and asset2 is bought. Then it's the other way round, isn't it?
Yes. In essence you are getting some kind of synthetic. If in a synthetic, a real instrument with plus enters, then it trades at overbought sells, at oversold it buys. With a minus it is the opposite, and the weights indicate the proportions in lots.
Co-integration is the basis of all spread trading. You can only trade spreads in cointegrated instruments
And the forward test of what? That the cointegration remains?
I'm sure the calculated cointegration won't stay. Without the forward test (another proof of the dubiousness of this tool).
This confidence is based on the fact that both initial series are I(1), which is not necessary. eurodollar I have seen both I(2) and I(3), and both assets must have the same order of integration. The problem is an old one. What's behind the right-hand edge of the quotient.
1. co-integration check on the out-of-sample
2. the forward built system on the out-of-sample
1. co-integrity check on the out-of-sample
2. the forward of the system built on the out-of-sample
Thank you. Communication with you is always constructive.
If you take this branch into account, it might be possible to build a TC.