Yoghurt systems and canned systems or The relationship between trading tactics and the reliability of historical test results - page 19
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 see, Alexander, thank you! I wanted to argue with you and drag you to the formula optimization == fitting, but I saw your rightness :-)
to LeoV
Blind parameters are those that do not depend on the indicator and are hardwired.
Examples:
-takeprofit and stoploss in the area of TS decisions.
- Trained perseptron.
Generally speaking, the period of the indicator is also a kind of "blind" parameter of TS, =="semi-blind")), since the regular TS doesn't see the market and doesn't change the settings.
Blind parameters are those that do not depend on the indicator and are rigidly set.
Examples:
-takeprofit and stoploss in the area of TS decisions.
- Trained perseptron.
Generally speaking, indicator period is also a kind of "blind" parameter of TS, =="half blind"))), because usual TS doesn't see the market and doesn't change settings.
- Well, they are also optimisable parameters.
- A trained perseptron - there is a catch here too. As with over-optimization. It may be overtrained, i.e. it may learn the rules too well in the training area and perform poorly in the future. So it's also a matter of training.
- Well, they are also optimisable parameters.
- A trained perseptron - there is a catch here too. As with over-optimization. It may be overtrained, i.e. it may learn the rules too well in the training area and perform poorly in the future. So it's also a matter of training.
In the perseptron (as I understand it) - there are no connections, which means there are no rules for it to learn.
So it is impossible to overtrain it, i.e. overtraining is just a blurring of the yes/no line.
P.S. Blurring with untrained data.
The perseptron (as I understand it) has no connections, which means there are no rules it can learn.
So it is impossible to overtrain it, i.e. overtraining is just blurring the yes/no line.
P.S. Blurring with unprepared data
Well I was referring to the network. A single perseptron, in my opinion, plays no role.
How can I tell you - a trend can essentially be taken as a pattern, and so can pullbacks on the tern - but how can you determine the duration of the trend, and whether the pullback is the start of a new opposite trend?
From the dynamics of the price series, it looks like there is no way. Maybe by levels, Fibo or something else.
I would like to propose a discussion of the relationship between trading tactics and the reliability of historical testing results.
When we create a trading system, we rely on some regularities that depend on the input variables. We often optimize the system according to these parameters at the testing stage and rejoice when we get good results. And then it turns out that the system starts to fail after a week. Some systems only start to fail after a few months. I also suggest discussing factors that affect the "shelf life" of systems. Why some systems are yoghurt, others canned tuna, and is it possible to create a perpetuum mobile or at least come close to it?
1. I do not think the very formulation of the problem is correct - pure determinism: an apple is ripe - it falls on its head, it does not ripen - it does not fall. The market, Forex in particular, has no functional clarity. Another interpretation is more suitable (I read it in somebody's book) about self-affirmation and self-annihilation of charts. For example, take Fibo. There are charts on which Fibo levels are present, and there are charts without Fibo. What is the matter? The point is that Fibo levels appear and disappear depending on the ratio of volumes of players who believe in Fibo and those who do not believe in Fibo at this point in time, because there should be those who win on Fibo at this point in time, who lose to them. This applies to any other figures that traders have discovered over several hundred years of the market. Currently these are either figures (head and shoulders, doji, etc) or indicators - saw a book that described over a thousand.
2. Each of the figures or indicators shows on the outwardly chaotic market a certain pattern, which, if found, may help to predict the further market behaviour.
3. every trader's little experience on the market convinces of the impossibility to build a trading system on one indicator - a set of indicators for a trading system is necessary. This does not change the matter, as a set of indicators identifies a new pattern, which is impossible to draw on paper, as it is multidimensional. But this multidimensional pattern has the same fate - it is subject to the law of self-affirmation and self-destruction.
4. Today the search for new patterns is automated and accelerated by mathematical methods of statistics, wave analysis and artificial intelligence. Using the tester and the optimizer we find parameters of the trading system, which would reveal the pattern, and having revealed it we predict the market and get the profit.
5. Here a question arises: does the pattern found by the TS meet often or rarely? The Papuans of New Guinea have a large number of boats and all these boats have personal names - there is no plural among the Papuans. You can't say in Papuan: these boats are made of palm trees and these boats are made on another island. So the main question on the trading system is: Are the patterns we have identified as individual as the boats of the Papuans, or do they have some generality?
6. Nobody cancelled the law of thechanalysis "History repeats itself". Any TS that has made a profit on historical data will make that profit in the future. The bugbear of over-optimization gets a different meaning. Over-optimised systems do not have a decent level of generalisation and may be too rare in the future. But in reality the situation is even worse, we will only see signs of a pattern after it has formed, when there is nothing left to predict.
7. In reality, the problem with creating a TC is not only developing rules that will identify a sufficiently generalised pattern, but also developing rules to anticipate the moments of transition from self-definition to self-affirmation of the graph and vice versa. Moreover, the longer this process of transition from self-affirmation to self-affirmation is, the better the trading system is.
8. Extremist requirements for the TS "at all periods, at all instruments, and for years" are not feasible, if we recognize the property of charts about self-destruction and self-affirmation as a law. Any trading system will be recognized by mathematical methods, losing traders will identify winning patterns and one day there will be no one to win - everyone will be smart, owning a useless trading system.
9. Considering the above, the development of TS is represented by the following steps:
a) A trading system is developed.
b) searching (optimization) of the trading system parameters
c) Time periods are searched for (timeframes are parameters), for example, January 2007, when the system is profitable.
d) timeframe, where the system is losing, is searched for.
e) the period of transition from profitability to loss and back is examined. If you were successful in finding the causes, you have a trading system, if not - you do not have a trading system.
Conclusion: we have the same canned food - both yoghurt and tuna.
Are the patterns we have identified as individual as the boats of the Papuans, or do they have some generalisation?
Run TC on OOS and you will get the answer to such an uncomplicated question.
Hello all (first post - need a drink %)
In my opinion, the system should be self-adaptive. Market behaviour is constantly changing - the system should adapt to the market itself. Not to create something finite and wait until it goes bad (and everything goes bad: even canned food, even yoghurt), but for the system to live... As in evolution: some species disappear, others appear...
Run CU on OOS and you will get the answer to such an uncomplicated question.
Faith in the results of OOS is faith in the holy grail. I believe that in principle it is impossible to design a TS that will always win - there will always be intervals when the TS will be unprofitable.
Within the concept of "self-affirmation-self-destruction" that I profess, OOS does not solve anything at all, because within this concept OOS can point to:
a) continuation of the identified TC pattern (positive OOS result)
b) the self-destruction of an identified pattern (negative result)
c) a transition from self-affirmation to self-destruction of a pattern (in the case of a deteriorating result).
That said, the OOS test says nothing about the future, which we do not yet see.
The use of OOS is more of a technique for obtaining a fairly generalised pattern. At any rate, with a negative OOS result we can try to refine the TC, which if successful will mean that we got a more generalized pattern, but no more than that. But we won't be able to continue the process until the grail is obtained - this generalized pattern will self-destruct.
In the MQL forums we can find other tips on obtaining more generalized patterns, such as testing on different currency pairs.
Metastock says that the TS should consist of trend, volatility, momentum, cycle, market strength and support-resistance indicators - that is another way to obtain a rather generalised pattern.
Again briefly my vision of the problem: a quote consists of some number (maybe infinite) of patterns which are self confirming and self destroying and the main problem of TS is the transitional process from self destroying to self confirming and vice versa.