An accident or an unrecognised pattern? - page 5
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
Lots of things can be thought of. For example, you could randomly mix up the real increments and summarise them in this new order.
The question is, what is the point of it?
Well, for example, to test the stability of the algorithm. If it fails, then all questions about its positive probability would fall away at once...
A mountain off my shoulders)))
Since I happen to tightly fit the whole story, how else can I detect the fit by simulation other than by counter-tests?
What Dmitriy suggested, run it through a random series and identify the differences does not work for me, because despite the absence of indicators and filters, the algorithm itself stands at the price and repeats these points in subsequent segments.
(peculiarity of self-learning in the system)
For example, to check the robustness of the algorithm. If it fails, then all questions about its positive probability would fall away at once...
A mountain off my shoulders)))
Since I happen to tightly fit the whole story, how else can I detect the fit by simulation other than by counter-tests?
What Dmitriy suggested, run it through a random series and identify the differences does not work for me, because despite the absence of indicators and filters, the algorithm itself stands at the price and repeats these points in subsequent segments.
(Peculiarity of self-learning in the system)
This makes some sense. I have an article on the subject of stability testing. Only there it's not prices that are modelled, but sequences of trades.
But this would not be analogous to forward testing, which is done on prices from the "future". It's testing on adjusted past prices.
No.
If you see a good signal, you can never tell if the author is trading on a coin and just got lucky or if the author is trading on a good trading system.
You can never tell for sure if it's a pattern or a fluke. You can only say with a certain probability. the more beautiful the signal, the more likely it is to trade on a good system.
as for the tester results... i can make a straight line out of several currency pairs by optimization. the more currency pairs - the easier it is to make. similarly, the more optimization parameters, the easier it is to achieve a nice profit chart.
As you've been told, divide a 10-year period into two halves.
The second half may also produce a nice graph by pure chance. but this is very rare. (this is what we talked about with a certain probability).
Or try to test on other currency pairs. very often systems that work on one pair work on others as well.
they also say that:
if you've done the optimisation and picked the output parameters, then
if for a small change in them, the profit curve deteriorates sharply, then it's probably a fit.
But I haven't tested this method.
I guess randomness is the unknowable regularity. For nothing in our world happens by itself
Randomness C and regularity H are forms of necessity. Therefore, it is never possible to predict whether the trading process follows a random or regular path at a certain moment in time. Consequently, when carrying out TS testing, we determine the correlation C and Z for a certain length sample, which is valid for the specified testing period. Attempts to adapt this ratio for other lengths and periods are doomed. The use of forward testing is a gross error in evaluating the TS stability. Testing of strategies on the tester is a tool to check the efficiency of ideas put into the TS theory. Nothing else can be required, and even more so, it is stupid to expect anything from the tester. Even the result with an 80/20 profit does not guarantee the stability of the TS. At the right moment a random 20 will overtake a regular 80. It's not as simple as it seems obvious.
It won't show anything.
It's just that in its original form it was fitting all 10 years at once, and in other cases it will fit in chunks - same eggs, only in profile
+
I'm being categorical because I know what I'm talking about. If you want to waste years of your life on delusions, that's your right).
Any algorithm, with or without probabilities, predicts future price behaviour. Passing through a quote backwards makes your algorithm meaningless.
There are no other methods, except for forward, and there is no need for it. Everything was invented by clever people long ago, you just have to use it. This is not a subject for discussion at all.
If people on this forum read textbooks on machine learning before getting into discussions, a lot of questions would go away by themselves).
I'm categorical because I know what I'm talking about.
1. I believe you gladly, since you admit to being categorical, that's a good thing.
Any algorithm, with or without probabilities, predicts thefuture behavior of prices. If you pass a quote backwards, you render your algorithm meaningless.
2. This statement is questionable.
I will give you an example of an algorithm that will pass the quote straight or backwards or sideways, but the final result will be the same.
Let's perform 1000 independent tests in normal time sequence, one test every hour.
Let's take the paradox of Monty-Hall with guessing of 3 doors as our algorithm.
https://bodyonov.ru/projects/monty-hall-demo
You probably know it. And we'll guess it.
The result of these guesses is predetermined and guarantees us 66% of positive outcomes with a margin of error of ~1-2%.
Now the question is...
If we don't change the algorithm, but take all the results/outcomes in the history and shuffle the dates and put them in a different sequence, do you think
the result (66%) will change and the algorithm will work differently?
I'm being categorical because I know what I'm talking about. If you want to waste years of your life on delusions, that's your right).
Any algorithm, with or without probabilities, predicts future price behaviour. By going backwards through a quote, you render your algorithm meaningless.
There are no other methods, except for forward, and there is no need for it. Everything was invented by clever people long ago, you just have to use it. This is not a subject for discussion at all.
If people on this forum read textbooks on machine learning before getting into discussions, a lot of questions would go away by themselves).
Forward is a piece of future history that has nothing to do with real history and/or future processes.
It is foolish to look for a pattern at random. Any pattern must be based on a theory, justified by the logic of the processes that occur, or on assumptions based on analysis of observations, or on a plausible postulate. Therefore, any regularity has to be searched for consciously, with the approximate expectation of how it should appear. Consequently, the search for a pattern is painstaking and exhausting work and should begin with the formulation of the above-mentioned positions. During last almost 8 years of searching I managed to formulate only 3 assumptions in which I managed to find 3 regularities that led to positive results. But all of them confirmed my assumption that in such a perfect market as Forex it is impossible to achieve outstanding results. Profits fluctuate between 10 and 15 per cent per annum, and that is when they are compounded over 10-20 years. It is not even possible to guarantee a profit within these limits for a specific, taken at random, year in history. Conclusion - it is impossible in principle to get a stable and guaranteed profit on the market, which is much bigger than that of a bank, because Forex is first of all an interbank instrument. On the other hand, this is my personal opinion and by no means am I imposing it on other market researchers and rake-evaluators. I wish them luck in finding better results.
And the very 3 theories I have developed and researched are known to all:
1. The Universal Regression Model for predicting market price https://www.mql5.com/ru/articles/250;
2. market theory https://www.mql5.com/ru/articles/1825;
3. bull and bear strength analysis https://www.mql5.com/ru/code/19139 ,https://www.mql5.com/ru/code/19142.
Thanks, I am still familiarising myself with the materials you have suggested.
Don't consider my silence to be disrespectful)
1. You need a flat filter. And it should not depend on the period.
2. using this filter, identify the maximum movement in any direction.
3. define the ratio of the maximum movement to the average movement and set the minimum parameter of this ratio.
e.g. 10/1. This parameter is aligned and tends to 1/1 at a) crises b) flats and both are detrimental to any system.
4. introduce the absolute parameter of non-random movement (how many times higher the maximal movement should be than the average)
For example, take 600 pips as a non-random movement and search for the parameter of the maximum movement size equal to or greater than that you have specified.
In this case the distance between the orders should be maximum 200 pips, i.e. 1/3 of a non-random movement, so as to spread the volatility.
5. open an order system and set the distance between orders if you do not change the lot, or the degree of risk if you change the lot in relation to point 4.
6. gradually increase points 3,4 until you reach a stable result.
7. Adjust for each financial instrument individually.
8. the system initially with the parameters described above should work on the rebound, but if the price suddenly goes higher, work on the rebound and the continuation of the trend.
PS: looking realistically at things personally my observations:
a) nothing can grow exponentially forever.
b) after exponential growth there may be no correction, so the averaging method can turn into big losses, because you only enter the market on relatively large and pronounced movements.
c) about point 1, so you understand how difficult it is, see the number of pages in the "from theory to practice" thread
d) something can always go wrong, that is why I focused on order systems, which will smooth out mistakes in trading due to order opening peculiarities.
e) 10% a month stably is enough to become a millionaire in six months to a year. above 150-200% a year is too risky even to try to do more. if the market allows it - use it. but do not try to find a super system that makes such percentages every month, unless you want to lose everything.
But don't try to find a super system that does that every month, if you want to lose everything. f) you probably won't use my advice, because it's too complicated, and most likely you will take a simpler path - averaging, increasing lots, or sinking in the swamp in which the branch "from theory to practice" has sunk.
Go for it.
I'm trying to understand and apply the filter checking conditions you suggested, but I can't figure out how...
The problem is that I can't define the F1 condition. I don't understand how to define a regular price change even from history.
Because my algorithm works on heads and tails principle and in fact the price doesn't play a role at all, there is only the result of outcomes - guessed/unguessed.
There are also sequences of events in history - guessed/not guessed, but this history is not considered in the algorithm, otherwise we may get to a false Monte Carlo output.
That's why we have nothing to rely on but the outcome.
And we should somehow understand that the result of guessing eagle/turtle over 50% is random or logical...
But I'll think about how to apply your condition).
I'm talking about finding real patterns in price history, not about some kind of simulation or generation of something resembling such a history.
In a real situation the whole difficulty is that there are no histories of acting factors. The only thing we know clearly enough is how central bank rates have changed over time. And with knowledge of the impact factor histories, it's just a matter of doing the appropriate mathematics.
Yes, F1, 2 etc. are here as reference signals (for the logic of the regularity itself and for its identification), without their knowledge it is indeed impossible to determine what in the price history is regular (in the key of their action) and what is not.