TSR - resuscitating trading systems - page 11
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
So the stars are not aligned simultaneously. Although what the fuck does it matter, the most important thing is that almost all results (unless you count the individual whiners who have no use in proving or demonstrating anything) have converged in profit and now the question of (un)promising TA can be put a full stop.
Where is the profit? Do you have any real-time stats for a decent period? Otherwise, as usual, you come up with what you call "nerd" shit and then the plot is "the branch and the forum can be closed" :) Once again for the gifted - do not mix shit, you will get shit as a result. And something good will not come out of it by such methods. :)
The robustness of the system needs to be assessed and there are a number of methods for this. Not to filter out the fit by fitting. And when you find robust systems, there will be no need to filter them with each other, because usually they give discrete input/output signals and practically do not overlap in time. At the same time, they can hold pose in one direction for quite a long time.
... The robustness of the system needs to be assessed and there are a number of methods for this. ...
Avals, can you cite these methods? Here I want to mathematically estimate the robustness of TS not only on history (by some specific number), but also its prediction in the future. Plus, it is desirable that this prediction coincides, at least statistically, with reality (on OOS's).
in short - each element of the system, the filter, the input-output condition can be assessed for robustness relatively separately. And each element must pass this test successfully. A good filter, for example, when tightening the filtering condition, should consistently improve the quality of the results (increase in PF, decrease in drawdown, etc.). If this does not happen - it should be discarded. This can be implemented using the usual tester and viewing the results of optimization for a particular parameter of the filter.
There are other methods, but I'm too lazy to write too much here :)
... in short - each element of the system, the filter, the input-output condition can be evaluated for robustness relatively separately. ...
If the TS has N explicit and implicit input parameters. Then when optimising all of them we will get (N + 1)-dimensional surfaces of different characteristics (profit, PF, FS, drawdown, etc.).
Each of these surfaces should be "smooth". If this is not the case, then at least some of the input parameters are a fitting filter.
OK, tell me (or give me a link) how to estimate the robustness of the filter. It is possible to consider TC as a filter of some kind.
Yes, most TCs can be thought of as a logical union of a set of filters. I.e. a set of Boolean variables and operations and, or, not with them. Also separately calculation of input/output levels. This does not apply to neurons and fuzzy logics.
I don't have the link. It was discussed on the spider. As a simple example, for example, calculate the dependence of IQ level on a person's height. Assume the rule (which needs to be checked): the higher the height, the lower the IQ. Then we have the statistics of height - level of IQ. Using it we calculate for sub-samples height>H - average IQ. If starting from some level with each step of the enumeration X the average IQ level decreases gradually, then the chance that it is justified and not a random chance increases significantly. If it goes up and down with increasing X then it probably isn't and must be found false or reformulated.
There are other statistical techniques, and there are non-statistical ones - logical ones.
... Discussed on the spider. To give a simple example, for example calculate the dependence of IQ level on the height of a person. Assume a rule (which needs to be checked) ... If starting from some level with every step of the enumeration of X average IQ level decreases gradually, then the chance that this rule is really justified and not a fluke increases a lot. ...
If a TS has N explicit and implicit input parameters. Then when optimizing all of them we will get (N + 1)-dimensional surfaces of different characteristics (profit, FF, FS, drawdown, etc.). Each of these surfaces should be "smooth". If not, then at least some of the input parameters are a fitting filter.
Thank you! In general the idea seems to be clear, but it would be good to develop it into practice for real TC. What is the dependence on what should be taken in their case in your opinion?
Have you visualised such surfaces? Or do you estimate their smoothness in general?
Visualisation for 1 and 2 input parameters is built into MT4 as standard. One of the recommendations, when entering a new input parameter (filter), optimize only it and watch the smoothness of the curve.
I estimate the smoothness of the surface (2 and 3 dimensional) visually. For more-dimensional surfaces, smoothness estimation was not done. It has always been sufficient to reduce everything to the 2 and 3 dimensional case.
How do you calculate your "smoothness factor"?
2. joo reports that his method requires constant re-optimisation. On OOS, my profits have been growing steadily for between 3 and 5 months. I.e. there is no need to systematically and continuously run the computer.
I have a couple of questions about it:
1. What gives you a stable growth of profit on OOS for these 3-5 months?
2. Are you betting money on it?