From theory to practice - page 1559
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
I have already written to him, he is looking for meaning in a certain hypostasis , now with a bias towards everything he remembers from university, then Gann, then the Witches, then the Almighty trying to call to account
He is looking for a grail for two years, don't be so severe, five years from now if he suffers some more he will sing or scream differently.
http://www.thealgoengineer.com/2014/online_linear_regression_kalman_filter/
there is also rolling regression and other modifications
I don't like Kalman because it always assumes some prior knowledge about the system. In the article it is the knowledge that the coupling coefficients of two quotes are described by a random walk. If our knowledge is consistent with reality then all is fine, but if not then alas. We need a Socratic approach - "I know that I know nothing")
As far as I understand it's not a particular algorithm but a general approach where coefficients are constantly recalculated regardless of whether there is a discrepancy or not. You need to look at each case on a case-by-case basis to see if there is a loss of accuracy with this simplification. The constancy of the window can potentially lead to inaccuracy.
There are usually two objectives in the search for a breakdown: 1) whether a malfunction has occurred and 2) at what point in time it has occurred. The first can be solved in a consistent (online) way, while the second seems to be solved only in an a posteriori (offline) way. Since our prices are very close to the SB, we have to solve our problem as precisely as possible, i.e. use both approaches.
By the time you find it with statistical methods, it will be too late) The best solution is a stoploss or a breakdown.
Any algorithmic approach to solving problems under uncertainty can be described as statistical (can - does not mean necessarily must). However, this is not usually referred to as a matstat, but rather as a statistical decision theory.
I dislike Kalman in that it always assumes some prior knowledge of the system. In the article it is the knowledge that the coupling coefficients of the two quotes are described by random walk. If our knowledge is consistent with reality then all is fine, but if not then alas. We need a Socratic approach - "I know that I know nothing")
As far as I understand it's not a particular algorithm but a general approach where coefficients are constantly recalculated regardless of whether there is a discrepancy or not. You need to look at each case on a case-by-case basis to see if there is a loss of accuracy with this simplification. The constancy of the window can potentially lead to inaccuracy.
There are usually two objectives in the search for a breakdown: 1) whether a malfunction has occurred and 2) at what point in time it has occurred. The first can be solved in a consistent (online) way, while the second seems to be solved only in an a posteriori (offline) way. Since we have prices very close to SB, we need to solve our problem as accurately as possible, i.e. use both approaches.
Well, a simple moving regression is run on a graph, coefficients are written and stuffed into a classifier. You get a break indicator, which you can check with new data.
This is if you don't want to invent anything, high-level so to speak)
You need concrete research, Alexei. CUSUM, Schuchart maps, etc., if you are interested and close to it.
I alone have no time to do everything. And there is less and less hope for the forum members. One - quotes Vysotsky, the other - philosophizes and follows some signals, as if this will bring them closer to the goal. Some kind of theatre of the absurd.
I am ready to participate in discussions of meaningful theoretical issues. I will not get involved in any joint projects involving a waste of time and/or money.
Well, simply run the sliding regression on a graph, record the coefficients and stuff them into a classifier. The result is a breakdown indicator which can be checked with new data.
This is if you don't want to invent anything, in a high-level way).
This approach is good for exploratory analysis of a series. The final trading system should be even simpler)
Well, simply run the sliding regression on a graph, write the coefficients and put them into a classifier. The result is a break indicator which can be checked with new data.
That is, if you don't want to invent anything, high-level, so to say).
I already suggested something about the exponent. I've been saying for a long time to make such an analysis (but I don't have enough brains to realize it, as it turned out I even cannot plot the price change according to the exponent ).
Or build an array of prices with different trends (linear, exponential, etc.) and then compare the actual price or maybe there are other ways to determine the type of trend.
Che has already suggested something about the exponent. And I told you to do such an analysis a long time ago (but I don't have enough brains to do it, as it turns out I can't even build the price change according to the exponent ).
Or build an array of prices with different trends (linear, exponent, etc.) and then compare the actual price or maybe there are other ways to determine the type of trend.
I don't know, I don't do visual dr... research. I just feed it into models and look, and optimize it. The best results are obtained just on regression fiches.
Any algorithmic approach to solving problems under uncertainty can be described as statistical (can - does not mean necessarily must). However, this is not usually referred to as a matstat, but rather as a statistical decision theory.
Che has already suggested something about the exponent. Yes and I have long said to do such an analysis (but I do not have enough brains to implement, as it turned out, even the change in price by exponent cannot build).
or build an array of prices with different trends (linear, exponent, etc.) and then compare the actual price or maybe there are other ways to determine the type of trend.
Just figure out how to calculate the regression coefficients (fixed order) using the least squares method on a fixed sample. Then count them in a sliding window of fixed size - you get a set of indicator-coefficients.