Optimisation! Share your experiences, please. - page 7

New comment
 
Reshetov:
solandr:
Reshetov:
That's why I guessed that it requires a linear REGRESSION coefficient. The closer this coefficient is to 1 in absolute value, the more linear the yield curve is.
I would like to understand (for those who studied arithmetic in parochial school) what kind of linear regression coefficient of 1 are we talking about? The linear regression equation is the straight line equation y=a*x+c. On the X-axis, as I understand it, are numbers of deals (1, 2,3.....N), on the Y-axis, we put the balance in the deposit currency (1000USD, 10000USD, 100000USD....., etc.). What formula is used to say that a=1, or tends to it? What is the normalisation principle used in this?
Sorry, we are talking about the linear correlation coefficient. I apologise for the misstatement.

Oh, that's all right! It happens. It must be so easy to get correlation and regression wrong - they always go side by side, don't they? Don't they? ;o)
So, what is your idea concerning the calculation of the correlation coefficient for the variables trade number - deposit size?
What is the physical meaning of such a "linear correlation coefficient" that will tend to 1 in the best case, i.e. the balance - simply to a perfect straight sloping line? Due to purely logical considerations, it is evident that the less loss-making trades there will be, the closer this correlation coefficient will tend to 1. But this can be seen from the analysis of results of a run without any calculation of the "linear correlation", right? What is the physical meaning of this action, besides the fact that the tester has drawn a straight line? ;o)

By the way, I can add one more similar "Paradox" to your "Paradoxes" box ;o). If you take and calculate the correlation coefficient for the run that has been made at profit capitalization, i.e. with the constant increase of the lot as the deposit grows, then the correlation coefficient will decrease from 1 downwards more and more, while the yield curve will be getting steeper and steeper. This can lead to the conclusion that increasing the lot as the deposit grows is not good for you ;o))) I give all official rights on this Paradox to you.
 

how does it go... ? it's bound by some sort of distance to some sort of average... and sails with it... ... could be tied to an MA for example... say a profit should be within 50 pts of a MA and if price moves away, the profit should follow it... I'm testing with the Canadian one now... the Euro-Buck is not very good there...

 
solandr:
Reshetov:
solandr:
Reshetov:
That's why I guessed that it requires a linear REGRESSION coefficient. The closer this coefficient is to 1 in absolute value, the more linear the yield curve is.
I would like to understand (for those who studied arithmetic in parochial school) what kind of linear regression coefficient of 1 are we talking about? The linear regression equation is the straight line equation y=a*x+c. On the X-axis, as I understand it, are numbers of deals (1, 2,3.....N), on the Y-axis, we put the balance in the deposit currency (1000USD, 10000USD, 100000USD....., etc.). What formula is used to say that a=1, or tends to it? What normalization principle is used in this case?
Pardon, we are talking about the linear correlation coefficient. I apologize for mispronouncing it.

Oh, that's all right! It happens. It must be so easy to get correlation and regression wrong - they always walk side by side, don't they? Don't they? ;o)
So, what is your idea concerning the calculation of the correlation coefficient for the variables trade number - deposit size?
What is the physical meaning of such a "linear correlation coefficient" that will tend to 1 in the best case, i.e. the balance - simply to a perfect straight sloping line? Due to purely logical considerations, it is evident that the less loss-making trades there will be, the closer this correlation coefficient will tend to 1. But this can be seen in the analysis of results of a run without any calculation of the "linear correlation", right? What is the physical meaning of this action, besides the fact that the tester has drawn a straight line? ;o)

By the way, I can add one more similar "Paradox" to your "Paradoxes" box ;o). If you take and calculate the correlation coefficient for the run that has been made at profit capitalization, i.e. with the constant increase of the lot as the deposit grows, then the correlation coefficient will decrease from 1 downwards more and more, while the yield curve will be getting steeper and steeper. This can lead to the conclusion that increasing the lot as the deposit grows is not good for you ;o))) I give all official rights on this Paradox to you.
1. If the yield curve is linear, it means that there were no sharp dips and ups, ie the strategy produced a stable signals over the entire testing period, rather than on a few random signals. For example, you look at the yield curve and there are sharp rises about once a quarter and the rest of the time it is either sinking or trying to stay afloat. Obviously, the system has adjusted to a few signals with the largest movement and simply ignores the rest. But if you analyze it, it becomes clear that this fitting will not lead to anything good on the real account because the takeoffs occurred more often on the news, i.e. on the things that are more suitable for fundamental analysis than for technical analysis. Therefore, it is almost impossible to catch similar movements in the future using the same signals. An Expert Advisor is customized for technical analysis, not for fundamentals. But if the yield curve is more linear, it means that the signals were issued on the medium movements that do not have such large ranges but they occur more frequently and are easier to analyze. Therefore linear yield curves are more stable.

2. As for curves with money management or other conditions, when the system trades with a non-permanent lot, normalization applies, i.e. the balance is calculated as follows

balance[i] = balance[i - 1] + OrderProfit() / OrderLots();
i++;

We obtain the yield curve as if the system opened trades with 1 lot. And no paradoxes.
 
solandr:

The figure shows just the initial stage of your proposal (getting the values of a and S). That is, the figure shows the result of one run in the tester. It is not difficult to obtain parameters a - linear regression coefficient and RMS for this graph. Suppose we have 1000 such charts based on the optimization results. As a result, we have an array of values 1000x2, where the first index is the run number and the second index is the values of a and S respectively. Further, what can display the obtained values of a and S on a two-dimensional graph along the axes show except for the extrema which may be several? I just want to know what you mean?


For a start we can open this file and see if our runs add up to a dense enough spot on the diagram or not. We discard the 70-80% of runs that do not fit into the needed confidence band (let's say) and carefully look at the parameters of the EA that have given us the results of the remaining runs. If these parameters will also give us a certain confidence level, instead of dancing around the entire range - then the results are quite stable and can be objective. If not - it means that some parameter(s) in the Expert Advisor are redundant and the model itself should be changed.

I think the methodology should be practiced almost manually at the initial stage in order to streamline it later.
 
Reshetov:
2. Regarding curves with money management or other conditions, when the system trades with a non-permanent lot, normalization is applied, i.e. balance is calculated:

balance[i] = balance[i - 1] + OrderProfit() / OrderLots();
i++;

We obtain the yield curve as if the system opened trades with 1 lot. And no paradoxes.
I always calculate system performance in pips. It's very easy to find difference between opening and closing prices, divide by Point and get pips.
 
In my opinion it is very difficult to create a stable EA implementing one strategy, therefore it is better to implement several trading strategies and limit the drawdown as much as possible, then the profitability of each strategy will change when the market changes, but the total result will be more or less stable. Ideally on a two-dimensional surface
with delayed on axes influencing variables is not desirable intensive spots.
 
nchnch:


To give you an example ... EA chart for 7 years... ( profit 10 p) stop 300 but profit floats with the price even if in a loss.... Profit to drawdown ratio of about 25 for seven years... it's not much in principle... but you can make 200 annual profits.
I wanted to make such a picture for one year, then understood that the profit would be too high (one friend has recently asked to add zeros in digits of the terminal), I aborted the test at 5 months.



Realistically, however, this is not the case. It's just a beautiful mirage. And there are a lot of them. This should always be borne in mind.
 

In reality, it is not. It's just a beautiful mirage. And there are a lot of them. You should always keep that in mind.
Well, that depends on who you are.... It's a mirage for some and a real one for others :)). it's a similar picture in real life...
 
nchnch:
Well, it depends on whom.... Who has a mirage and who has a real one :)). in real life, it's a similar picture...
How similar? Have you made 50-100 successful deals during the last month, or is it steadily going up in real trading for half a year or so?
 
4 months approximately running... ( also as predicted)
123456789
New comment