Econometrics: why co-integration is needed - 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
faa1947: Что такое подгонка или не подгонка?
Fit (over-training, over-optimisation) is a term used exclusively for financial markets (non-stationary time series).
That's rubbish. A horse or something.
A fit is an estimate of the parameters of a parametric model. Nothing can be overfitted.
And nothing can be fitted to a non-stationary market at all.
That's rubbish. A horse or something.
A fit is an estimate of the parameters of a parametric model. Nothing can be overfitted.
And you can't fit anything to a non-stationary market at all.
I will try to explain it to you.
Any segment of time in the past is essentially stationary, because knowing the time series itself in the past, you can almost always find a function or some pattern on it that will give a profit on the past data - it's not a problem at all.
The term non-stationarity, when applied to financial markets, means that we can't know how the market will change in the future, exactly in the future.
Using the past data, knowing how the market changed in the past, we can always find a function or a pattern that will take into account these changes. But no one knows how the market will change in the future. It is not a fact that it will change according to its past changes. Moreover, it will most likely not change like that.
For this reason, by adjusting our TS (in fact, it has been reoptimized) to past data, to those market changes that were in the past, we will obtain the TS, which cannot produce a profit in the future.
In practice, after the optimization of TS, we usually want to take TS with the parameters that give the largest profit with the smallest drawdown - this is exactly the adjusted TS for the past data. On future data such TS does not work, because these optimized parameters have taken into account all the necessary changes in the past, but the market in the future becomes different, not like in the past, and our TS, with such optimized parameters, does not take it into account
Let me try to explain it to you.
Any segment of time in the past is essentially stationary, because knowing the time series itself in the past, you can almost always find a function or a pattern on it that will yield profits on past data - this is not a problem at all.
The term non-stationarity, when applied to financial markets, means that we can't know how the market will change in the future, exactly in the future.
Using the past data, knowing how the market changed in the past, we can always find a function or a pattern that will take into account these changes. But no one knows how the market will change in the future. It is not a fact that it will change according to its past changes. Moreover, it will most likely not change like that.
For this reason, by adjusting our TS (in fact, it has been reoptimized) to past data, to those market changes that were in the past, we will obtain the TS, which cannot produce a profit in the future.
In practice, after the optimization of TS, we usually want to take TS with the parameters that give the largest profit with the smallest drawdown - this is exactly the adjusted TS for the past data. On future data such TS does not work, because these optimized parameters take into account all the necessary changes in the past, but the market will be different in the future, not like in the past, and our TS with such optimized parameters does not take it into account.
I spent several years on this, until I understood: either TS identifies and models non-stationarity, and then it is TS, or it is not. All terms of optimization, overoptimization are all emotions, pure shamanism and the deeper the trance, the more faith in what I have created.
I know of several ways to account for the non-stationarity of the market. Cointegration is one method and is valuable precisely because the result is a stationary series. and here optimisation and over-optimisation are not appropriate.
Yes. You are essentially getting some kind of synthetic. If the synthetic has a real instrument with a plus entry, then it trades overbought it sells, over-sold it buys. With 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
Spread trading is the certainty that the quote will return to zero from the extremes. but after how long?
I know of several ways to account for the non-stationarity of the market. Cointegration is one way and is valuable precisely because the result is a stationary series. and here optimisation and over-optimisation are not appropriate.
These terms are not invented by me. They are commonly known terms used by traders. If you don't like them at this point in your life, it doesn't mean you won't agree with them in the future, as life flows and changes.
You are simply over-optimised by the term cointegration )))))
What you claim is essentially a prediction of the future. That is, by making a stationary series out of a non-stationary series you can predict with 100% probability what the market will be tomorrow.
Try applying for the nobel prize or something....))))
These terms were not invented by me. They are well known terms used by traders. If you don't like them at this point in your life, it doesn't mean you won't agree with them in the future, as life flows and changes.
You are simply over-optimised by the term cointegration )))))
What you claim is essentially a prediction of the future. That is, by making a non-stationary series into a stationary series you can predict what the market will do tomorrow.
Try applying for the nobel prize or something....))))
These terms were not invented by me. They are well known terms used by traders.
Certainly not by you. This is TA - a country populated by shamans and Pinocchios. Read a primer and as a revelation "forward test, forward test, over-optimisation ..."
In this thread we are discussing how the stationarity of the difference between two non-stationary series can be exploited. So far, only spread trading has emerged. But it doesn't qualify for the Nobel.
In that case, the non-stationarity of the two series must be identical, so that it (non-stationarity) is extinguished by subtraction.
If it is different, then after subtraction we get third kind of unsteady series.
How, by what parameters can we compare non-stationarity of two non-stationary series?
In that case, the non-stationarity of the two series must be identical, so that it (non-stationarity) is extinguished by subtraction.
If it is different, then after subtraction we get third kind of unsteady series.
How, by what parameters can we compare non-stationarity of two non-stationary series?
Spread trading - the certainty that from the extremes the quote will return to zero. but after how long?