A single quality indicator for the strategy - page 3

New comment
 
Yousufkhodja Sultonov:
The sum of the squares of the deviations of what?

The point is to square all deviations from the linear regression and sum them up. The larger the deviation, the worse the estimate, usually divided by the final value of the linear regression (i.e. the final value of the linear regression divided by the sum of the squared deviations). This is the classic method. They can be different (again, reference to the math).

It depends on what the estimate is for, whether it is an intermediate estimate or a final one, and what its exact value is.

There are hundreds of methods.

 
Alexandr Andreev:

The point is to square all deviations from the linear regression and sum them up. The larger the deviation, the worse the estimate, usually divided by the final value of the linear regression (i.e. the final value of the linear regression divided by the sum of the squared deviations). This is the classic method. They can be different (again, reference to the math).

It depends on what the estimate is for, whether it is an intermediate estimate or a final one, and what its exact value is.

There are hundreds of methods

I wanted to ask, what parameter are we analysing?
 
Yousufkhodja Sultonov:
I wanted to ask, which parameter are we analysing?
Ekweti, well, or balance, better so Ekweti, let's ask all the minor questions in private, otherwise the thread will suffer
 
Yousufkhodja Sultonov:

I propose that this indicator (K) is the product of the recovery factor (RV) by the expected payoff:

K=FB*MO

PV=Net Profit/Max Drawdown ;

IR = Net Profit/Number of Trades.

Example (EUR/USD, TF D1):



We need to correctly handle small or no trades, it would not fit. This is what we need to automate the process.

Here's another condition: the possibility of comparison of two or more strategies has been provided and we will use some fixed scale, for example, from 0 to 100 and we will consider 75 and higher to be a good level. Strategies will be compared when trading on the same time interval.

 
Aliaksandr Hryshyn:

You need a small or no number of trades to be handled correctly, that's not going to work. This is necessary to automate the process.

Here's another condition: it was possible to compare two or more strategies and reduce them to some fixed scale, for example, from 0 to 100 and we will consider 75 and higher to be a good level. Strategies will be compared at trading on the same time interval.

Firstly, it is nonsense to have any indicators of a strategy with zero trades, because if there are no trades, there is no indicator. And the K coefficient is already remotely estimating even the results of one trade. But it makes no sense to evaluate a strategy with less than 100 - 1000 trades.

Secondly, you do not yet have an indicator that generally evaluates the strategy, and you are already talking about its scale, which is wrong. Tell me, what feature of the strategy does not take into account the coefficient K = PF*MO? True, in the above example I have managed to get the K to be constant - decreasing year on year from 1973 kin to the present. This suggests that the strategy has become worse, that it needs to be improved. The K value varies widely, from a fraction of one (1 case) to 80. Another thing is to try to understand the physical meaning of the K coefficient. About comparing strategies - please show the best K results on other strategies.

 
Aliaksandr Hryshyn:

You need a small or no number of trades to be handled correctly, that's not going to work. This is necessary to automate the process.

Here's another condition: it was possible to compare two or more strategies and reduce them to some fixed scale, for example, from 0 to 100 and we will consider 75 and higher to be a good level. Strategies will be compared when trading on the same time interval.

Everything would be so easy, as the result very quickly take 100 000 strategies, because everyone has its own parameters, and the strategy may be considered not with one but with many parameters and have a number of high marks at the output, so 100 000 I call it little, rather evaluations get a dozen million. And what we have in the scale from 0 to 100 , with 75 estimations we will have about five thousand strategies, so we should be very careful when estimating, and if the strategy is a bit worse, we will drop it. And we need accuracy of at least 0 to 10,000. And so that the person understands why the estimate of 7,500 is worse than 7,501 and agrees with the system
 

here is my adviser for a month on a cent account



Files:
TesterGraph.gif  12 kb
 

The MT5 has different criteria for optimising the parameters in the strategy tester, which are used to evaluate the trading strategy. You can use them too. To choose one - take any more adequate Expert Advisor, optimize it by all criteria one by one, and compare the results of forward testing. Usually, sharpe ratio or recovery factor wins in such a comparison. For example, optimization by sharpe ratio always gives much better results than optimization simply by balance.

 
Igor Gurov:

here is my adviser for a month on a cent account



Please calculate the K value = PF*MO and provide it here.
 
Dr.Trader:

The MT5 has different criteria for optimising the parameters in the strategy tester, which are used to evaluate the trading strategy. You can use them too. To choose one - take any more adequate Expert Advisor, optimize it by all criteria one by one, and compare the results of forward testing. Usually, sharpe ratio or recovery factor wins in such a comparison. For example, optimization by sharpe ratio always gives much better results than optimization simply by balance.

The top starter wants a generalized criterion that takes into account both pluses (profit) and minuses (drawdown) of the strategy.
123456789
New comment
Reason: