Non-fitting system - main features - page 4
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
As you know, many TS are well-fit on a representative sample and drain on forward tests. Such systems should not be used in trading.
Today I analyzed different TS I have and this is what I found out:
The signs of a working system is that it optimises almost equally on both sampling and double sampling with a constant lot. For example, we take 10,000 bars. We divide it approximately in half, i.e. by 5000 bars. Optimize it for 5000 bars. Then we apply it to 10 000. If the profit factor remains almost unchanged or has insignificantly changed (becomes independent of the sample length), the system is likely to pass the forward test. Naturally, trades should be about 600 - 1000 on 10 000 bars (300 - 500 on 5000 bars).
It turns out that if optimization of an TS significantly worsens as the test sample increases, the probability of a successful forward test is very low.
And some TS, for example, optimize on 300 trades, and on 600 the optimization yields nothing but deep drawdown and low expected payoff. Such a system can be immediately burned.
Actually, there is already an answer concerning the necessary and sufficient sample length. The sample length must be such that optimization with a fixed lot on the whole sample, as well as on its half, yields almost the same profit factor.
Only a respectable person can resuscitate a sacred cow.
Are we optimising on stationary or non-stationary BP? What are we constantly discussing? Some kind of forward test. This test can give the following answers:
the forward section is close to the tested one - TS values are almost the same;
the forward section is not similar to the one being tested - TS is unprofitable;
the forward section is almost similar to the one being tested - we don't know whether to throw it away or feel sorry for it.
And what will be the TP on the real? Like the one being tested, similar or not?
A sacred cow can only be resuscitated by a respectable person.
But anyone can corrupt an idea.
You gave a good definition of a fitting system.
It's hard to call it a system either.
The cart from the mountain...
And we are talking about other attributes.
Adaptability and to what?
The tolerances are also worth discussing.
I've been writing here but it all seems to have faded into oblivion https://forum.mql4.com/ru/23455/page6 .
Interestingly, the market is a changing structure and its current phase in the global sense (not to be confused with the phase of the socket) may change at any time so the TS gets a fiasco. However, there is a possibility that this phase may happen again in the foreseeable future and then the EA will make a profit. In the proposed advisor implemented the principle of its admission to trading on the analysis account, it is certainly just an idea, but imho worthy of discussion, because soptimized in 1999 for 10 years on 15 minutes has not lost. You can kick me, but I am deeply convinced that we should not look for the magic optimum parameters but for the criterion of a TS admittance to real trading.
A sacred cow can only be resuscitated by a respected person.
Are we optimising on stationary or non-stationary BP? What are we constantly discussing? Some kind of forward test. This test can give the following answers:
the forward section is close to the tested one - TS values are almost the same;
the forward section is not similar to the one being tested - TS is unprofitable;
And if profitable?
the forward section is almost similar to the one being tested - we don't know whether to throw it away or feel sorry for it.
You have to look for an unlikeable one - otherwise what's the point of a forward? Yes and the test in general.
And what will be the TP on the real? Will it be the same as the one being tested, similar or not?
HZ But both similar and dissimilar will be. What is the problem?
===
Do you have anything in your résumé other than arguments about the meaninglessness of everything, because "unsteadiness" is there? From branch to branch, the same thing. Honestly, I don't remember you writing about anything else. About anything else.
Just a follow-up...
Maybe I don't get it.
How do you define similarity?
I think that's important.
And if profitable?
You have to look for an unlikeable one - otherwise what's the point of a forward? Yes and in general - the test.
HZ But both similar and dissimilar will be. What is the problem?
===
Do you have anything in your brainstuff besides reasoning about pointlessness of everything, because "unsteadiness" is there? From branch to branch, the same thing. Honestly, I don't remember you writing about anything else. On any subject.
I'm against thoughts like the Moscow-Peter rail with no welds.
All that Metacquotes offers us is for stationary markets, keying a tester. All the talk of a forward test is an attempt to hide a design flaw in the tester.
I tried to push the topic of non-stationarity several times on the forum - without result, it was tried many times before me, but each time the topic was clogged by DSP supporters with a rigid focus on the stationarity of BP.
The beginning of the branch is an attempt to get away from the primitive alorhythm of the optimizer, who can find a pimple in the pit, but above the edges of the pit. What is proposed in the thread submission, over-optimisation and all, is an attempt to escape the trap of this one pimple, called the grail.
Just a follow-up...
Maybe I don't get it.
How do you define similarity?
I think that's important.
Yes. For example, we do a CU for a Hearst coefficient > 0.7. Different way of doing it. Initially, we assume that BP = large trends, small trends, cycles (sideways) and Gaussian noise. We make a system for an hourly with trends of 200 to 300 pips.
Yes. For example, make a TC for a Hurst coefficient > 0.7. In a different way. Initially we assume that BP = big trends, small trends, cycles (sideways) and Gaussian noise. We make a system for an hourly with trends of 200 to 300 pips.
Can you give a definition of non-similarity?
Since we are talking about non- or stationarity...
Formal, I mean.
That sounds like nonsense.
I apologize for the nonsense.
Can you give a definition of non-similarity?
Since we're talking about non- or stationarity...
Formal I mean.
That sounds like nonsense to me.
Pardon my deafness.
By the way, good question. I'm really tempted to make an EA that would classify sections of the market according to how unprofitable/profitable it is on them.
Well, it's like this - (oops, I'm setting myself up!!!)) visually. You can use the prevailing frequencies of the local area. In short, you can make it up. But by eye, that's fine too.
All that Metacquotes offers us is for stationary markets, keying the tester. All the talk about the forward test is an attempt to hide a design flaw in the tester.
I tried to push the topic of non-stationarity several times on the forum - without result, it was tried many times before me, but each time the topic was blocked by supporters of DSP with a rigid focus on the stationarity of BP.
Do you have real ideas how to consider non-stationarity in the tester?
Your complaints about nasty stationarity supporters as well as your crown phrases - "dynamic system", "chaos theory", "VR is non-stationary", "sequence of digits of Pi number" I already learned by heart.