Non-fitting system - main features - page 13
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
Actually, the answer to the topicstarter's question is NONE!
Someone is showing you a PAMM which has been giving a stable (5% drawdown) profit for several years and has earned several mio.
Does it guarantee that PAMM does not fail? - No.
What is the probability of losing this PAMM tomorrow - negligible (but it is).
In a week - more.
In a month - even more.
And so on.
An auto-optimiser system falls under the definition of such a system. That is, it optimises itself. Since this does not distinguish this system from others, apart from the complexity of implementation, this attribute is questionable.
There are different types of auto-optimization. One thing is history-based fitting and quite another thing is explicit calculation of parameters from available time-series (it may be no fitting at all).
If a system (based on TA) without external parameters, it should work on any data, even random data. And this is a contradiction. Hence, a system without external parameters cannot be profitable.
Quotes are unlikely to be random data, so the purpose of a system without external parameters is to specify an internal view of price behaviour (extract the information needed) and use it.
Auto-optimisation comes in many forms. Fitting on history is one thing, and explicitly calculating parameters from existing time series is another (it may not be a fit at all).
Suppose you have found some market pattern. I.e. the system will not work on random data.
The market pattern will appear on any synthetic instrument obtained by simple arithmetic operations. For example, on (3.5 *EURUSD +GBPJPY / USDCAD).
Are you ready to claim that the auto-optimizer system will work on any synthetic trading instrument?
Suppose you have found some kind of market pattern. That is, the system will not work on random data.
The market pattern will appear on any synthetic instrument obtained by simple arithmetic operations. For example, on (3.5 *EURUSD +GBPJPY / USDCAD).
The regularity on one instrument may appear to a greater or lesser extent on the other one. Therefore, it will have to work on a synthetic trading instrument (note that this is not random data!).
Are you ready to claim that the auto-optimizer system will work on any synthetic trading instrument?
If it is not automatically running the tester on the previous period and choosing the best variant, it probably will. Although there are variants here, too.
I think we need to talk about that, too.
why external? how are they different from internal? did the author decide at one moment that they can be removed? on what basis did he make that decision? on the fact that the system is not fitting?
And what are the signs that tell him to do it?
i tried to explain it as i understand it, but i missed it a bit :o). And imagine that you have made an aeroplane, but you have not made the control part of the System (a part of the aircraft), but you "took" the control out of it and gave it to an external optimizer. An engineer on the ground optimized everything for you and you fly, somehow you set up the controls and the whole thing flies with you. You are also told that the "controls" will not change no matter what. Would you get into such a plane? Or assume that you have made control, but your instrument readings are permanent, there are many examples of this.
Optimisation is, after all, about obtaining parameter values that maximise/minimise the target function (in our case profit), or more spatially, finding extreme values. It only makes sense within the area of constraints that have been imposed. To extend the optimal parameters to a large area is very presumptuous (and in my understanding just stupid), given the specifics of quotes (I mean non-stationarity and high order of the system). Unless there are parameters changing slowly in time, and there are none. And then we start dancing with tambourines - and if we adjust 128 parameters, it will definitely work...
With a zero spread, a system with no external parameters should work on any trading instrument, as the system uses a market pattern.
It may seem that this rule does not apply to multicurrency systems. But it is not. A multicurrency system without external parameters always lends itself to breakdown into sub-systems without external parameters. This means that all discussions about monocurrency systems without external parameters also apply to multi-currency systems.
What kind of fiction is this - external parameters? >> What is that?
The decision about the value of the parameter is made outside the system. Does it make more sense?