Yoghurt systems and canned systems or The relationship between trading tactics and the reliability of historical test results - page 10
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
Tough question, LeoV. Probably both. But I myself would like to change the ratio of these components towards mathematics. Of course, there will never be a complete mathematical justification, and so every system will be a disaster.
Well that's so sad. The glass is half full :-). There are also a large number of mathematically complete and rigorous descriptions of how various systems and devices work. But their physical implementation constantly faces unsolvable problems, from computing device capabilities (real-time computation) to the lack of necessary a priori data. But nothing, we have solved them and we are solving them. So-called solutions with sufficient accuracy for practice. They fly and do not fall down.)
Aeronautical radio engineers save the Mathematicians :-).
I don't know... as long as it works. I have a criterion by which I clearly identify that the system has stopped working and I will stop all EAs on all accounts.
What is that criterion, if it's not a secret?
And no past experience how long such TS may work?
In my opinion, stops are by no means the most important part of the TS. It is just one of the ways to exit a losing position, some also see it as a way to limit losses, which is also correct in principle, but not suitable for every TS.
If stops and takeoffs for all positions are equal, it is a linear exit from the market, so their use in TS is not always justified.
Market noises are relative things, for one TS the 10-20 points movement is a noise, but for another one it's a way to earn money.
From the stops, or at least from their presence, depends the pattern that we will look for. A stop is a condition. If the system changes the condition, it changes itself (agreed?). So, if we run the same inputs with one stop and then another one, and then no stops at all, but with an output by the return signal, then the results will be different - different patterns work differently.
Why so sad. The glass is half full :-). There are many mathematically complete and rigorous descriptions of various systems and devices. But their physical implementation constantly faces unsolvable problems, from computational capability (real-time computation), to the lack of necessary a priori data. But nothing, we have solved them and we are solving them. So-called solutions with sufficient accuracy for practice. They fly and do not fall down.)
Aeronautical radio engineers will save the Mathematicians :-).
I think there is a catch. In the examples you cite the aim is clear. And it can be reached. In Forex the situation is somewhat different. Under constantly changing market conditions you need to achieve the target - profit. What is a profit? This is a vague goal, it is not clear, because this profit may be achieved in different ways, I mean by different trades (even on the same TF). They may be frequent, may be not so and so on......
It's always better in the morning. I will try to make my point of view more coherent.
Suppose we have a TS (trading system) that has input parameters that we select using history in the hope to make the TS profitable.
Let's consider TP(profit taking level) and SL(loss limitation) most frequently used in TS.
TP - of course you can do by selection, but think really every time (in each transaction) this level should be exactly the same, say 40 points, but maybe now it is 43 or 24. It turns out (more logical) TP must be calculated each time you enter the market, but for this you need to have an appropriate block in the TS, which gives a forecast. Suppose the price reaches 1.9789 + - 5 pips by 12:00 + - 15 min Moscow time tomorrow. But during this time different news may appear, which will radically change your forecast, i.e. you should also have a block, which predicts news and also gives a forecast about the market reaction to it. Maybe theoretically it is possible, but practically I think it is unbelievable.
IHMO the market itself tells you when to enter and when to exit. I.e. with every new tick of TC you have to make a decision - sit back in the trade (it moves in our direction) or it's time to exit.
SL - it is nothing like an emergency exit from the building in case of fire, ie you have problems with Internet, TS can not receive information for analysis and therefore work correctly = exit the market itself, if your forecast is not justified (or you have reached the zone of uncertainty, where it will rush). The reasoning is almost similar to TR.
For further clarification I will take the example of a well-known MA
iMA(NULL,0,MATrendPeriod,0,MODE_EMA,PRICE_CLOSE,0);
MATrendPeriod - let's say, an optimizer has been selected and we have a profitable TS since the time of King Gorokh. We got the value of 14. Is it really the best period at each point of time. Or maybe it should be larger on a calm market and smaller on a fast one? Here is one of the approaches to solve this problem ('Perry Kaufman AMA optimized'). The period is adapted to the RMS of the market (calculated, which is important, not set once and for all). This is adaptation in its pure form, a fast market period of averaging is small (a slow one is big), and you set the limits in which period should change.
MODE_EEMA - the reasoning is the same, why EMA, but not SMA, or may be the non-linear averaging is better...
PRICE_CLOSE - this is the most interesting part from my point of view. I think you have often thought about or tried to use not only Close, but various combinations of OHLC (like (O+H+L+C)/4). So what is the conclusion which is better ? Again, selection, fitting on history. Suppose we stopped at Close, but when do we know it? Only when a new tick has come (the old bar closes and a new one appears), i.e. you analyze an obsolete price, even if it is 1, 2...5 points obsolete. And what is more important than the tick, that came at the end of the day, i.e. at 23:59:59, when there isn't almost anyone on the market? Is this tick more important than the one next to it? And it is this tick that lies in the analysis and serves as the basis for the trading system?
Every tick is important, every price movement, and every time a new tick arrives, you need to analyse it to enter or leave the market. This is the only way, otherwise you will not have time, this is the age of speed. In the past, the market was slow, and you could make decisions after reading a newspaper (stock-exchange report) in the morning, now is not the time.
TS has to be adaptive, it has to calculate everything it needs, collect the data and take it into account. Optimization on history is evil and self-deception, as 90% of traders use it.
P.S. But as Marcus Anneus Seneca said "Errare humnnum est", maybe I'm wrong.
I agree. Except that we will select stops and takes for a given period simply by counting the number of successes at each value. Which value is most repeatable, that will be used. We can plot a histogram: the number of times price moved by so-and-so points after the signal. On one axis - the amount (height of bars), on the other - the value of movement. Here is a "slice" of regularities, depending on SL and TP.
To calculate stop and take levels using good patterns, you need to teach the system to look for and use them. And to do that, you need to do it yourself at least once. And we are discussing one of the important aspects of this search - the longevity of patterns. Or how to find a good sustainable pattern. There :)
>> Yeah. How?
"The search for patterns" is an eternal and inexhaustible topic, for which no answer will ever be found (in the perpetuum mobile sense). And testing and evaluation of TC is quite a practical problem that can be solved for a given TC, even if the allegedly found patterns are not logically understood or completely unknown. It seemed to me that the topicstarter's question was asked precisely about how to adequately assess the behaviour of the TS in the future...
The search for patterns is a pointless activity if we cannot justify with any degree of reliability that the TS built on the found pattern will acceptably behave in the future.
I totally agree!
Testing on the history of quotes is not the only method for testing the system reliability. Here is an outline of an alternative testing methodology: Sandwich Toss Article. But if you're allergic to mathematics, it's better not to read it.
Thank you))) I'm a mathematician-programmer by training, so I think we'll figure it out=)
I don't know... as long as it works. I have a criterion that clearly identifies that the system has stopped working and I will stop all EAs on all accounts.
If you can tell me about the criterion, it seems to me that the most important thing is to stop in time.
What is the criterion, if it is not a secret?
In the optimisation interval, I determine the Max DDD. I check that Max DDD is not exceeded in the OOS interval. In real trading, 1.5*MaxDD is considered a stop.
No past experience of how long such a TS can work?
Agreed. Except that we will adjust the stops and takes for a given period simply by counting the number of successes at each value. Which value is the most repetitive, we will take it. We can plot a histogram: the number of times price moved by so-and-so points after the signal. On one axis - the amount (height of bars), on the other - the value of movement. Here we have a "slice" of patterns that depend on SL and TP.
To calculate stop and take levels using good patterns, we should teach the system to look for and use them. And to do that you should do it yourself at least once. And we are just discussing one of the important points of this search - durability of patterns. Or how to find a good sustainable pattern. There :)
I highlighted, do you really think that i have never used SL and TP in 8 years on forex and never looked for a pattern?
I've never used SL and TP to analyze the market, it shows when to enter and when to exit the market. (But I don't deal with them for a long time.) But I don't do them, they are long gone.