The perfect mechanical trading system. - 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
Can I give you a little hint? :)
...
3. It's not the price that should be considered, but the logarithm of the price.
In logarithms everything becomes simpler and more correct.
For small changes in price, there will be no difference between working with the price and the logarithm,
With large changes in price, the difference will be significant.
And could you give another hint, i.e. describe point 3 in more detail (how to calculate this price logarithm?)
Self-adjusting indicators are a dead end. I will try to justify my opinion.
I have developed several such indicators, but they appeared to be sensitive to volatility of quotes coming from different brokerage companies. That is, these indicators work fine on data of one brokerage company and do not work at all on data of another one. Worst of all they work with TC data. For example on the standard indicators on the same quotes interval there is divergence in one brokerage company and not in another.
My research showed that volatility is the leading factor to be considered in a self-tuning indicator. However, eventually the indicator becomes dependent on the filtering method of quotes in different brokerage companies and on the changes of this method (that is not notified by brokerage companies).
Here I also faced with the fact that I cannot "harden" (as Renat always says) the self-tuning, because it is constant (linear), and not discrete.
I believe that the only way to avoid the volatility problem is to ignore the absolute values of indicators and quotes. I.e. to make a decision in MTS we should use the correlation of values in one form or another, and this is in fact pattern recognition.
And the filtering parameters will again depend on volatility in brokerage companies... You will make a filter for quotes of NS, and then you will find that your brokerage company filters stronger than your filter, etc.
Self-adjusting indicators are a dead end. I will try to justify my opinion.
I have developed several such indicators, but they appeared to be sensitive to volatility of quotes coming from different brokerage companies. That is, these indicators work fine on data of one brokerage company and do not work at all on data of another one. Worst of all they work with TC data. For example on the standard indicators on the same quotes interval there is divergence in one brokerage company and not in another.
My research showed that volatility is the leading factor to be considered in a self-tuning indicator. However, eventually the indicator becomes dependent on the way of filtering quotes in different brokerage companies and on the changes of this method (that is not notified by brokerage companies).
Here I also faced with the fact that I cannot "harden" (as Renat always says) the self-tuning, because it is constant (linear), and not discrete.
I believe that the only way to avoid the volatility problem is to ignore the absolute values of indicators and quotes. I.e. to make a decision in MTS we should use the correlation of values in one form or another, and this is in fact pattern recognition.
And filtering parameters will again depend on volatility in brokerage companies... You will make a filter for quotes of National Operator, and then it will appear that your brokerage company filters more intensively than yours.
At the end result Expert Advisor will work differently in different brokerage companies, but profitable in different ways.... So, the difference in rates of different brokerage companies is an integral part of the market and does not play any role in our task. This means that if there is an automated mechanical trading system (automatic self-adjusting system :)) with one parameter to work in one brokerage company profitable, it may be unprofitable in another. Only in the other brokerage company, it will retune and work profitable again.
ArtemRG 21.11.2006 14:01 wrote (a):
And the filtering parameters will again depend on volatility in the DC... You will make a filter
of NS quotes, and then it will turn out that your brokerage company filters stronger than your filter, etc.
Yes, probably the filter should be set up for a certain brokerage company and even more - for a certain currency pair.
But the task is not to be universal, the task is much more modest:
"The task: can I write a function that runs a month's history once a day and sets the optimal number for Stop Loss parameter?
AND THE GREATEST thing: can I use this function to check it in the tester? Will the tester work at all? It turns out that it must change the parameter of stop for a new day every day in the tester mode, is it possible? If we solve this problem, the rest is icing on the cake."
To create a universal adaptive system (if it is possible) we will need to analyze a huge amount of parameters, and not only the "noisiness" of a single day. And it will also require a fair amount of time and a considerable amount of clever heads, - as a consequence a considerable cost. But in this case everything is much more modest. For now we just need to write a tester in the EA that will analyze only one parameter.
Join us in writing the code - especially since you have a lot of experience in this!
One more argument in favor of this approach is the assumption that EAs adjusted to the history for the first time (even if not for a long time) trade quite profitable. I think the Championship can serve as an example - in the beginning there were much more profitable EAs (it seems to me that this was due to the fact that they were adjusted to the history)
Before writing an EA with a tester inside, try to check this hypothesis manually. Suppose for each of the last 10 months prooptimize some EA on the previous 6 months and report the result.
If only it were that simple...
Another argument in favour of this approach is the assumption that EAs adjusted to the history at first (albeit not for a long time) trade quite profitably. I think the Championship can serve as an example - in the beginning there were much more profitable EAs (it seems to me that this was due to the fact that they were adjusted to the history)
I totally agree.
Another argument in favour of this approach is the assumption that EAs adjusted to the story for the first time (even if not for a long time) trade quite profitably, an example I think can be the Championship - in the beginning there were many more profitable EAs (I think this was just due to the fact that they were adjusted to the story)
Before writing an EA with a tester inside, try to check this hypothesis manually. Suppose for each of the last 10 months prooptimize some EA on the previous 6 months and report the result.
If only it were that simple...
Who would do it? (so that no one wastes time)
I have to tell you right away myself, I have not been able to realize the principle described so far. I cannot conquer MQL yet :).
the idea is being developed for 4 o'clock and is as follows, first we need a well-filtered trend line, but not lagging like MA. For example we can take Kravchuk's FATL, then its derivative and the second derivative is calculated, i.e. acceleration and jerk.
Provided both acceleration and jerk are greater than zero (possibly also introducing some delta for false signals), we buy; if both acceleration and jerk are less than zero
, we sell.
But at the same time there will still be a lot of false signals. To filter them, I suggest a condition - the selected trend line (e.g. FATL) is greater than a fractal in the necessary direction, but in a smaller time frame.
Exiting a position using a trailing stop, the stop loss at entry into the position is equal to the average size of a candle with the shadows for the last 24 hours in the desired time frame.