Econometrics: why co-integration is needed - 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
????
How about trading from the boundaries inside the range? Up from the level to sell, down from the level to buy.
Like oversold/oversold?
The conclusion itself has no basis in fact, because fitting and the presence of regularities are quite different things, because it is not necessary for two time series to be similar to each other, thereby "fitting" each other, in order to trade. To trade, we need regularities, while regularities should not always be present - it is enough to fit one series to another, regularities should be present at or in front of those places of a series, where we are going to open a position and obtain profit. In other places it does not matter at all what happens, up to complete non-matching of rows.
Because of this, it is not necessary at all for rows to fit each other.
The conclusion itself has no basis in fact, since fitting and the presence of a pattern are quite different things
Quite right! Fitting happens if the TC is trained on patterns that are unevenly distributed. I.e. in one area they are densely distributed and in another area they are empty. And at any stretch you can adjust it to these very regularities, the result will be too much at the other one.
In the simplest case, this is how TSs are adjusted, designed either for a rebound or for a breakdown. If trends dominate, the bounces will fall and the breakdowns will work, if sideways prevail - on the contrary.
The case is a little better with breakout systems, because breakouts depend on the width of the channels and the change of volatility will make them fail.
The mathematical notion of stationarity, i.e. stability of expectation and dispersion, is clearly not appropriate here, since it does not solve any problem - it's botany. The problem is the uneven distribution of patterns - trading signals.
Сам вывод не имеет под собой оснований, поскольку подгонка и присутствие закономерностей - совершенно разные вещи
Reshetov:
Exactly right! Fitting happens if the TC learns from patterns that are unevenly distributed. I.e. in one area they are dense, and in another area they are empty. And at any stretch you can adjust it to these very regularities, the result will be too much at the other one.
In the simplest case, this is how TSs are adjusted, intended either for a rebound or for a breakdown. If trends dominate, the bounces will fall and the breakdowns will work, if sideways prevail - on the contrary.
The case is a little better with breakout systems, because breakouts depend on the width of the channels and the change of volatility will make them fail.
The mathematical notion of stationarity, i.e. stability of expectation and dispersion, is not appropriate here, since it does not solve any problem - it's botany. The problem is in uneven distribution of patterns - trading signals.
If it is only the right edge of the quotient, then the problem is the minimum sample based on which the decision is made. In the ARIMA model the last four bars are already a lot.
Back to the topic. Stationarity is the reliability of the prediction beyond the right edge of the quotient. Got stationarity from the cointegration algorithm. How to use over the right edge? there was an idea expressed here - I'll look into it now.
How about trading from the boundaries inside the range? Up from the level to sell, down from the level to buy.
No, it does not work
to faa
(1)
You have very shitty model metrics, starting with the size of the coefficient, etc. It's understandable that to pull your eurik by the ear you have to multiply by big numbers, which means you have to be fucking accurate in your predictions, i.e. the t-statistic doesn't really tell you anything (it should be ten times smaller), it just lies and gives you the illusion again.
(2)
Further, what does "stationary" mean? In what sense? Stationary only the distribution or also the ACF? If only the former (stationarity in the narrow sense, not much use). You seem to take very seriously the figure determining the probability of stationarity. And most likely you have imaginary stationarity, the value of your sequence is 0.0132-0.0137. Frankly speaking, it is a complete phony and it is clear that it will not go far from your so-called "level" even if it really wants to, its coefficient will fail.
(3)
The presence of stationarity absolutely does not mean and does not equal predictability, everything is not so simple, a condition as if necessary, but not yet sufficient :o)
(4)
Your magic formula: cointeg = -eurusd + 119.3552 * REGRES_1 - 0.276233 - 2/112E-05*trend is bullshit. I'm not even going to explain it, I'm sick of it ...
(4)
You have two X's and one equation, i.e. you cannot pass to currencies. There is only one way out - make the model more complicated until there is no correlation or look for statistical dependencies, maybe they exist
Just a thought: presence of cointegration in the regression
eurusd = 119.3552 * REGRES_1 - 0.276233 - 2/112E-05*trend
It means that we may not worry about using this regression for the forecast - its residue is stationary.
It just occurred to me that the presence of cointegration in the regression
eurusd = 119.3552 * REGRES_1 - 0.276233 - 2/112E-05*trend
It means you may not worry about using this regression for prognosis - its residue is stationary.
Anscrambler has been tried before - nothing comes out of it :( Neither one is suitable for more than 50/50 forecasting
But I wish you success, my hands are clumsy and my knowledge is simply less.
I've tried the Ancrambler before and it doesn't work :( None of them are more than 50/50 for forecasting.
But I'd like to wish you good luck, my hands are clumsy and my knowledge is simply less.