Mechanisation of optimal parameter selection. Finding a common denominator. - page 5

 
Avals:

Yes, the number of deals is an important parameter. The number of deals should be sufficient, and this "sufficient" varies depending on the system. It starts with dependent trades (for example one entry can be divided by 10 different volums and saddleloch will be 10 times larger) and finishes with such nuances as ratio of average profit trade to average loss trade (for systems where profit trade is larger than loss and vice versa we need much more trades than when they are approximately equal). But we should not introduce the index of the number of deals in the formula of strategy "goodness". This is an indicator of estimate quality (credibility of testing results). If there are not enough trades for the system - the results simply cannot be trusted.


I think the same way, number of deals shouldn't be calculated. Same as the ratio of average loss vs. average profit - it doesn't mean anything, unless you always place stops at the same distance from the price...

a little - a lot, big - small, a lot - not a lot, this very topic was conceived in order to get rid of subjectivity of evaluations

i.e. ratio of total orders volume in the market within testing period to the initial deposit (of course reduced to the currency of the deposit and taking the leverage into account)...

 
OnGoing:

Only the correctness of the logic can be checked in the tester, nothing more.

The criteria for the "profitability" of the TS should be contained in its very essence and should not require statistical sampling on history.


I do not understand, give me an example? If the system is formalized and profitable, then it should be profitable in the past?
 
Avals:

I don't understand, can you give me an example? If the system is formalised and profitable, does it have to be profitable in the past?
It should ALWAYS be profitable, not just in the past, that is the question.
 
Avals:

It is also important which indicators we are comparing. The indicator should assess the "goodness" of the system's results. imha, we can come up with various composite indicators or take the standard ones such as sharpe or sortino. All have certain advantages and disadvantages. That is why it is better to use a simple one like the profit factor.

Isn't it better to use Recovery Factor, because it also takes into account deposit drawdown? I also heard that if two systems have FV for example 4 and 6, then working together (simultaneously) they will give FV=4+6=10. What do you think about it.
 
OnGoing:
It has to be profitable ALWAYS, not just in the past, that's the question.
Well, this is what needs to be checked and statistics is one of the tools (not the only one)
 
Avals:
Well, this is what needs to be checked and statistics are one of the tools (not the only one)
Optimisation - is based on a single method - statistics. Here it is proposed to evaluate the TS only on the basis of it.
 
Europa:
Isn't it better to use the Recovery Factor, because it also takes into account deposit drawdown? I also heard that if two systems have FS for example 4 and 6, then working together (simultaneously) they will give FS=4+6=10. What do you think about it.
It's a matter of taste. For me, it is more suitable for the choice of MM. As I have already written - there is no perfect indicator of system quality - all have disadvantages and advantages. So you can apply one which you know the disadvantages and advantages. Well, or for some cases switch to another one. In general, it doesn't make sense to make this indicator particularly complicated - to look for an ideal. There are other more important things and methods of assessing robustness. imha
 
OnGoing:
Optimisation - based on a single method - statistics. Here it is suggested that the TC be evaluated on that basis alone.
Well, this thread is about statistics. No one said that it is the absolute truth and that you cannot supplement the selection on the basis of statistics with market logic, for example.
 

The system is a fit with historical data. If the fit is good, it will be profitable for some time. If it doesn't, it needs to be adjusted again. Trying to come up with a system that works on "all available history" is doomed.

Good fitting differs from bad fitting in that changing the adjusted parameter within a wide range leaves the system profitable.

For example: we adjust and optimize the MA. We obtained the optimum and determined a period of 100. If the system still performs in the range 50-200, then it is a good fitting system.

 
Avals:
Well this thread is about statistics. No one said that it is the absolute truth and that it is impossible to supplement the selection on the basis of statistics with market logic, for example

It is not about what it can be supplemented with, but about the fact that statistics is not an adequate method of evaluation at all. That is why it is called a fitting, a self-deception, an illusion...

If one prefers to delude oneself, one has the right to do so, of course.