Two pieces of news - page 2
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
Lecture 2. Let's move on.
Optimisation
One of the means to try to solve the problem is optimization of some trading strategy using historical data. I will not specify how exactly it is achieved as there are many of them, for example: using a neural network for minimum residuals RMS, econometric models with residues refinement, genetic algorithm of extremum search, gradient method of extremum search, scientific method - exhaustive trying of variants with application of different ranking of the order of these very variants, etc.
Suppose we have a trading system in the form of an algorithm. The trading system has only one input parameter. The trading system is written in MQL4 which means that there is a small choice of tools for optimization - the strategy optimizer built into MT4 trading terminal.
We take the strategy, set the steps of the single input parameter, disable the genetic algorithm, enable the search for extrema by mathematical expectation and run it through the whole range. The result is a certain curvature: expectation along the vertical line and values of the input parameter of our TS along the horizontal line. Assume that the resulting curvature has multiple extrema, some of which are above the zero expected payoff and some are below it and it is quite possible that a certain portion of extrema falls on 0 (which is not necessarily true).
It is easy to see that zero expectation is a martingale. And the input parameter is the argument of the function (resulting curve) that manages the martingale. Many people who have used the tester have surely seen that expectation can be both zero and non-zero. I.e., using the TS input parameter, the strategy can be transferred to a martingale or it can be taken out of a martingale. Theoretically it does not seem to concern in any way the so-called tautology of Doub, because he clearly states the impossibility to control (not) martingale only by frequency of bets or their size. But, we will consider such cases later on. Our task is profit extraction, not the Doub's tautology. For now let us only consider the fact that expectation is manageable to a certain extent, but for now only on historical data.
Now a question for a puzzle: what values of an input parameter can be taken after optimization for a trading strategy?
I do not think that many will think long about the answer. Especially when it comes to the application of TS to our own real money.
I will try to answer. To make a profit even on historical data, not to mention the forwards, demo and reals, we need the extremum with the minimum.
Many people are already turning the finger upside down and have quite reasonable arguments against this as yet unsubstantiated assertion.
Therefore I do not conclude the lecture, but would like to hear arguments against or for.
One can even try to draw some kind of inferences that lead to the correct answer. Namely:
Why is it that some traders choosing an extremum with a maximum nevertheless do not obtain any profit on forward tests?
Or go further and try to substantiate why we need the minimum extremum - the very bottom - rather than an intermediate value between a high and a low?
What specific conditions must this minimum extremum satisfy in order to solve the above problem?
You see, it is important to understand the essence of the problem. I propose to scientifically summarize what has been said above:
1. Given: A row has
2. Not given, but what do we want to get?
3. did not develop ....
Colleagues, as they say, a correctly formulated problem is 50% of the solution! It seems to me that we are now again (for the umpteenth time) on the verge of ... of a new understanding .... of the Oak theorem!
Only if it is correctly formulated in its essence and not in insignificant trifles that only get in the way and distract from the solution of this very problem.
It's the same as if a teacher asked you how many times two is, and you answered (because you don't know the answer) by asking, for "scientific" purposes, what exactly the question is about: apples, whores, etc.
What are you trifling with? Put at least 1.00000001, it is more than just 1. By the way, I`d like to ask people who put "+1" what? Roughly speaking, what and in what is measured?
It's Madam Probability!
There is always a positive probability (i.e. >0) that the information submitted by the topicstarter can have a positive effect on the expectation of current trades.
Naturally, the forum form of exposition requires an additional filter to separate a seed from a husk. There is no need to cling to the rhetoric, as I'm sure the sense is quite clear to you, unless your task is to cool down, deliberately, the imulse of the author's creative search. All imho, no offence!
This is Madame Probability!
There is always a positive probability (i.e. >0) that the information given by the topicstarter can have a positive effect on the expectation of current trades.
Naturally, the forum form of presentation requires an additional filter to separate the seed from the husk. No need to cling to the rhetoric, because I'm sure that the meaning is quite clear to you, unless your task is not to cool down, deliberately, the imulse of the creative search for the author. All imho, no offence!
Not anymore, it should have been done earlier, many earlier, i.e. quite early. The example with the problem about apples and whores in the context of sutra pissed me off. I give up, let him develop creatively, I have nothing against it, just trying to understand what was the point.
PS: Be careful with probability, even the value of 1 doesn't play any role here, the observer just didn't have the patience to get the whole "presentable" sample :o)
Reshetov писал(а) >>
In order to make a profit even on historical data, let alone forwards, demo and real, we need an extremum with a minimum.
An extremum with minimum is a minimum value of IR (negative) ?
What specific conditions must this minimum extremum meet in order to solve the above task?
You can plot the dependence of MO(1) on MO(2) of forward optimization and choose the parameter, at which MO>0 && M0(1)=MO(2)
Lecture 2. Let's move on.
It is easy to see that zero expectation is a martingale.
I propose to start with a definition. In this case it is not difficult to notice the substitution of the concept.
What specific conditions must this very minimum extremum satisfy in order to solve the above problem?
it should be as large as possible, above zero preferably? :)
Why are you being so petty? Put at least 1.00000001, it's more than just 1. By the way, all wanted to ask betters, "+1" what? Roughly speaking, what and in what is measured?
+1 vote for.
That's what this entry means.
About the so-called "lectures" - it's a bit of a mishmash.
And what does martingale have to do with it?
I suggest we start with some definitions. In this case, it is not difficult to notice the substitution of the concept.
In this case, it is difficult to notice the substitution, because there is nothing to compare it with, i.e. it is not known what has been substituted for what. At least two definitions need to contradict each other to a greater or lesser extent in order to detect the substitution.
I therefore give my definition:
Martingale with respect to the particular player playing the particular game is a strategy, chosen by this player in this game and within the framework of this game rules, during which the arithmetic mean of the results of all games played by this player in this game will tend to zero with the increase of the number of games played.
If you are personally unhappy with this definition and believe it to be a substitution of terms, please provide another definition that you think is more accurate.