TSR - resuscitating trading systems - page 9

 
Avals:

if the EA is worthless, then any combination of looser systems will be looser. It's like filtering one random walk with another - you'll still get SB. If at least one of the crossed systems is robust, then it might make sense. You have to compare it with a simple portfolio of the two systems - what will be more profitable and/or under what conditions

There's the whiner again. No one is forcing you to use loosely based TS. Fuffy Expert Advisor was given only as an example, i.e. I purposely changed its perceptron inputs for shitty ones, so that it could fit the history in the demo example, but it was sure to fail on forward tests without using additional filtering - the TS is absolutely not suitable for trading in its pure form according to R. Pardo's methodology. The purpose was to show the benefits of using the filter (once again for the especially gifted: the filter of trading signals, not the TS) cutting off uncoordinated trading signals. That is, to show how to get a profitable trading signal from a knowingly losing TS by partial elimination of false trade signals.


Once again for the especially gifted: The Expert Advisor attached as an example in the first message of this thread is not recommended for autotrading - it is not intended for this purpose. This is just a deliberately reduced demo version, not a workhorse.

 
Reshetov:

There's the whiner again. No one is forcing you to use Loose TS. The lousy Expert Advisor was given only as an example, i.e. I purposely replaced its perceptron inputs with shitty ones, so that it could fit the history in the demo example, but it was sure to lose on forward tests without using additional filtering - the TS is absolutely not suitable for trading in its pure form according to R. Pardo's methodology. The purpose was to show the benefits of using the filter (once again for the especially gifted: the filter of trading signals, not the TS) cutting off uncoordinated trading signals. I.e. to show how to get a profitable trading signal from a knowingly losing TS by partial elimination of false trade signals.


Once again for the especially gifted: The Expert Advisor attached as an example in the first message of this thread is not recommended for autotrading - it is not intended for this purpose. This is just a deliberately reduced demo version, not a workhorse.


weirdo, out of two sbs you generated a third one randomly profitable and pass it off as development and draw some conclusions. Kindergarten is the youngest group :)
 
joo:
Why goodbye non-stationarity? Non-stationarity will never go anywhere. That's why the grail will never appear.

Stationarity, as I understand it, means that there are probabilistic patterns. Accordingly, if BP has identified patterns, then it is no longer non-stationary. :)

 
voltair:

Stationarity, as I understand it, means that there are probabilistic patterns. Accordingly, if BP has identified patterns, then it is no longer non-stationary. :)

You put your own meaning in the concept of "non-stationarity", which is different from the generally accepted one.

If a process is non-stationary, it does not mean the absence of regularities in it. Similarly, if a process is stationary, it does not necessarily have any regularities, for example, white noise, it has no regularities, though the process is stationary (the only regularity in it is a channel, knowing it allows trading profitably).

 
Avals:

weirdo, out of two sbs you generated a third one randomly profitable and you pass it off as development and draw some conclusions. Kindergarten is the youngest group :)

Once again, for the particularly gifted from kindergarten: I have nowhere claimed that this is a super-duper development with a 100% guarantee of results in the future. Moreover, I warned that the filter above cannot filter out all of the false signals and some part of them will not be detected, which can then have a negative impact on the balance. I gave only a demo-example in which the profit was obtained not in the best trading conditions (intentionally worsened). Once again: this is not a development, but a demo, which anyone can take and double-check, in order to make an independent decision whether it is worth the trouble or not.


Those who don't like it, i.e. all the whiners who are offended and humiliated by fate and the given demonstration example, should go to GOB, instead of shitting with empty flub in the threads.

 
joo:
You put in the notion of "non-stationarity" some meaning different from the generally accepted. If a process is non-stationary, it does not mean that there are no regularities in it. Similarly, if a process is stationary, it doesn't necessarily have any regularities, for example, white noise, it has no regularities, though the process is stationary (the only regularity in it is a channel knowing it to trade profitably).

Well, actually, there's a bit of a joke there... After all, I already wrote that if we take any section of BP, we will always find "regularities" in it by (un)linear regression or by Fourier or c NC or any other way, which will fall apart already in the next section. So it is possible to find some regularities, but they have no sense...

And as for white noise, indeed, boundary values already could be used, i.e. sense appears, but for this reason it is a stable MO in a stationary process.

So when you say "I have voiced a specific method ... which allows to establish presence of found regularities unambiguously", I want to specify - are these those "senseless" regularities of non-stationary VR about which I spoke or some other, "useful"? Then how do you separate one from the other?

And please - give your (or generally accepted) definition of non-stationarity.

 
voltair:

So when you say "I have stated a specific method ... which allows to unambiguously establish the presence of found regularities", I want to specify - are those "meaningless" regularities of non-stationary VR I was talking about or some other, "useful" ones?

Useful ones that can be used outside of Sample.

voltair:

Then how do you separate one from the other?

Silently. :)

I do not separate them. I am only looking for "useful" regularities. However, the question about the lifetime and speed of changes of regularities is still open, which actually is not very important at this stage (the question is purely academic and is unlikely to be interesting/applicable in practice one day). The market will never become completely unpredictable and chaotic - it is not in the interest of the powerful, nor is it in anyone's interest. The only thing left to do is to optimise the TS more often.

voltair:

And please - give your (or a generally accepted) definition of non-stationarity.

From wiki:

"Stationarity is the property of a process not to change its characteristics with time. It makes sense in several branches of science."

Accordingly, "non-stationarity" has the opposite meaning.
 
joo:
Useful ones that can be used outside Sample.

This is, of course, understandable, but says nothing about conscious, objective criteria for "usefulness". I wish it did!

joo:
... the question about lifespan and rate of change of regularities remains, for now, open, which actually isn't that important at this stage (the question is rather purely academic and unlikely to ever be of interest/applicable in practice).

Here I disagree. The question is both interesting and applicable, imho.

joo:
... The market will never become completely unpredictable and chaotic - it is not in the interests of the powerful, nor is it in the interests of anyone. The only thing left to do is to optimise the TS more often.

In my view, it is (not entirely, but) an element of... of faith. :) And I would like... Science, objectivity, at least statistics. The same frequency of optimizations - no clear criteria, no data, right? But surely it is possible to find the optimums empirically. And perhaps, having found them, it would be possible to determine their correlation with some (global or not so global) cycles.

 
Figar0:
Me, I'm not complaining, the effect is the same, another autotrader I know, here's Reshetov more. Sometimes at the back, sometimes in the middle. Try it.

So this, Sergei:

first of all why? All this can be organised in a much simpler way, with OOS at the front, and being able to trade immediately after checking on the test, just a slightly different approach to the problem ;)

Secondly, it is one thing to optimize a profitable strategy, and quite another to play with fitting without any fundamentals at all. In the first case, the approach outlined is likely to yield benefits, but then it can all be organised quite differently and no worse.

 
TheXpert:
It could all be organised in a much simpler way, with OOS at the front, and be able to trade immediately after testing on the test, just a slightly different approach to the problem ;)
I believe. Well scribble down how (if anything - you can, in private), pls!