Non-fitting system - main features - page 24
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
Avals with all due respect but come close to the formula I am using now, I would very much like to hear another solution.
PF is one .
Sharpe and Sortino coefficients are not bad either, but they also have flaws. I have no perfect formula, although more or less optimal, taking into account the TC itself, you can pick it out.
rak wrote >>
P.S. If you are interested I can send it to you in person later.
It would be interesting :)
why is it bad? we are interested in the dynamics of equity. build lR to the cumulative sum of this series and it will be up.
The main condition is again statistical validity. But the system itself can make adjustments to the minimum trade requirement. For example, the ratio of SL to TP. The more this ratio differs from 1, the more trades are necessary. The extreme case is when we trade without stops - any number of trades will not be enough. Although not exactly so, because there is a stop loss - a margin call.)
In my experiments on this subject it says differently. We have SL and TP of the same value but dynamic, calculation, say, by ATR or dispersion on one signal may open several positions. If there is a pattern, profit should be higher than loss. imho more accurate and faster estimated criterion for the strategy admittance to the real account, than, say, compared to the trailing stop, here one trade in the trend can bring the lion's share of profit. Although I agree there is a loss.
P.S. for those wishing to slur the word trend in another thread, the constructive just started .
My experiments in this regard tell a different story. Strategy tested by opening price is just easier to write the skeleton of the TS, have SL and TP of the same value but dynamic, the calculation of say by ATR or variance on one signal may open several positions. If there is a pattern, profit should be higher than loss. imho more accurate and faster estimated criterion for the strategy admittance to the real account, than, say, compared to the trailing stop, here one trade in the trend can bring the lion's share of profit. Although I agree there is a loss.
P.S., for those who want to use the word trend in another branch, just started constructive.
Well, yes, it depends on the TS and the approach to system-building in general.
Well yes, it depends on the TS and the approach to system-building in general.
Look at the personal .My experiments on this matter tell a different story. Strategy tested by opening price is just easier to write a skeleton TS, we have SL and TP of the same value but dynamic, the calculation of say by ATR or variance on one signal may open several positions. If there is a pattern, profit should be higher than loss. imho more accurate and faster estimated criterion for the strategy admittance to the real account, than, say, compared to the trailing stop, here one trade in the trend can bring the lion's share of profit. Although I agree there is a loss.
P.S. for those wishing to slant the word trend in another thread, just getting constructive .
I've seen your posts too.
Of course it is constructive, but it is a little off-topic, don't you think? Maybe a separate thread?
===
I hope the top starter will forgive me for being a little off-topic. I'm not so ashamed of the general background.)))
1. it is a good idea to examine the TS in order to determine where and most importantly why it does not work. And not even for the purpose of improving it - there is no universal, except not to trade, TS - just not to use the TS in such areas.
2. About the speed of reaction to the change of the market character was once discussed, and there is a code of specific filter in the Code Base for this. The brief essence of the problem and requirements for filtering are as follows:
Applying the usual low-frequency volatility filtering with enough smoothing to track the overall market character gives monstrous response delays (phases) that almost nullify all the benefits of the adaptation. I.e. it was necessary to create a filter which would
a) reacts with minimum delay to an increase in the degree of market passions on one side and
b) effectively suppresses noise at the current volatility level on the other hand.
Of course, one can filter not only volatility - what is more appropriate to the situation. For example, the measure of directional movement or whatever...
===
And for the future - let's try not to get sidetracked. (applies to me too!!!))
There is a variant of non amateurish...... truth the effect is the same :)
- optimization, for period N
- recording results in a file
- analysis in excel
- writing the parameters of acceptable results in another file
- testing of Expert Advisor with these parameters on a tank and forward N/2
- selection of results comparable to results obtained during the optimization period (essentially the same "nature of the curve") ....
- selection of the only one and its final optimization......
even tired of writing :) .... but there is a problem with the result, although it seems that everything is taken into account and random results are rigidly cut off.....
>> were there any other, more successful variations on the theme of Excel files with processing, saving and further use of optimization results?
It's better to start with methods. I'm studying this one now. Very interesting points, especially in the chapters about wavelets and multifractal analysis.
It's hard to program, though. But, for example, there really is something in correlation integrals...
so I guess we can wait for the mt4 library for the mathematical apparatus from this monograph by A. Pavlov's "Methods of analysis of complex signals" - similar to LibMatrix ?
were there other, more successful, variations on the theme of Excel files with processing, saving and further use of optimisation results?
If you mean automation of the process, it was automated, to the maximum, except for analysis-sampling-sampling-sampling....
And if you mean the final result, there was only a conclusion: no test/s, even the most successful one, guarantee anything in the future ..... But the systems were not overly adaptive either :))
Ok, just as an example, we have a number of trades +25, -10, +8, +5, +0, +4, +3, +1. Here PF greater than 1 seems good, but the angle of the regression results minus, imho is not good .
One criterion is not enough. For this series of deals I can say at once that when discarding one maximal profitable deal we lose half of all profits made by the system. You cannot do it that way.
I guess we can wait for the mt4 library for the mathematical apparatus from this monograph by A. Pavlov's "Methods of analysis of complex signals" - like LibMatrix ?
No such plans have been made so far.