Optimisation range - page 2

 
budimir >> :

picking the most vividly trending areas

.........-all this is bullshit .........:о)


bullshit... BREED ... BREED ... so it's ... Bread !--------------------->> for someone :o)

 
Vinsent_Vega писал(а) >>

to Neutron

Do these two types of errors apply to any TS or only to neural networks?

These two types of errors refer to any method of trading strategy parameter optimization. They also refer to the MT Strategy Tester but the conclusion is based on the assumption that the number of parameters to be adjusted in the tester is equal to the number of input parameters. Maybe there is a situation when the number of input parameters is less than the number of parameters set in the tester, then the formula will be changed into a more general one.

 
Neutron >> :

Listen, I don't understand how this should work in reality... Because initially we don't know the average frequency of deals of this TS with these parameters... besides, when you change these parameters the frequency of deals changes...

 
Look, let's say I have 5 parameters, one of which is the Mach Mach period... let's say I do the first optimization on a sufficiently long interval... let's say 2 years... during the optimization there are several stable zones: one zone where the Mach Mach period is equal to, let's say, 9... a second zone with a period of 80 and a third zone with a period of 240... Using the formula I have to choose the optimization period with 20 trades... I calculate the average deal frequency of each bag and get the optimization interval for the 9th bag according to the formula to be equal to, say, 2 days... since it makes 20 trades in 2 days... for the 80 it's 2 weeks and for the 240 it's 2 months...

logically i should choose the most profitable of these 3 wagons and most often it will be the 9th...

but that's just about the wagon... and I have 4 other parameters and they all affect the frequency of trades... and consequently the optimization period... say, changing stop loss from 25 to 250 will affect the frequency of trades...

What side should I come in from when optimizing?
 

No, no. Wait!

The frequency of deals by itself, and their optimal number on the test history by itself. You optimize parameters of TS by looking at trade results - find the maximum of some functionality, in this case, it may be the comulative income or profitability (number of points per transaction). Now you have a question: given the number of adjustable parameters, you need to find the most optimal number of transactions, on which the tester will optimize the strategy. Note, not the time, namely the number of entries and exits to the market.

In short, the task includes only the number of trades - they must not be more or less than the optimal number. You have found the optimum profitability - trade. After a certain period of time, you start over-optimization and do it all the time. How to implement it in the Strategy Tester? You have to think...

 
someone is looking fora"grail", someone is looking for a robust system, even with minimal, but STABLE profits...... You should dance from here.... there are also methods such as constant "disorientation of the system" under the current market parameters...... who wants to trade, so optimizes...... unfortunately, or maybe fortunately :)))) a unified approach to this case does not exist, and can not exist..... all do it for themselves.... i know one thing, until you decide on the question (which period and how many forwards, etc.) you'll never trust the trading system....... :(((( and if you don't trust, you don't trade.......
 
Neutron >> :

In short, the task is only the number of trades - they should not be more or less than the optimum. When you have found the optimum profitability, you trade. After a certain period of time, you start over-optimization and do it all the time. How to implement it in the Strategy Tester? You have to think...

ah... So... So, you mean to put a limit on the number of trades when optimising from the beginning?

 

>> Well, yeah.

rider писал(а) >>
there is no unified approach to this matter, nor can there be..... everyone does it for themselves.... i know one thing, until you decide on this issue (what period and how many forwards, etc.) you will never trust the trading system....... :(((( and if you don't trust it, you don't trade.......
The sensual approach to trading, it's not for me. In my opinion, there is mathematics that defines an optimum in a certain field of parameters. It should be adhered to, everything else is bogus.
 
Thanks, Sergei... I think I've got it... but how to implement it technically - I don't think it's a big deal... :)
 

to Vinsent_Vega & Neutron

I have some interesting ideas concerning optimization, or rather re-training, correct me if I'm wrong:

We take a trading strategy, run it on 3 intervals - 1 month, 2 weeks, 1 week (each interval is two times bigger than the previous one), save the results in the database (the same MS Access or any other) from each period to a separate table. Then we make a query that displays the deal size with the set of parameters being the same for all three tables and, theoretically, we obtain the set that will not give the maximal profit but the most stable one. The tester's periods may be arbitrary, the main thing is to keep dependence between them (reducing each of them by half). It is clear that we may not have such an option, when all three tables show successful results with the same set of parameters, then we should select all parameters that are the closest to them. If we take Fibonacci as a basis and build queries by Fibo-numbers for specified interval, with subsequent use of Fibo-numbers as weight criteria (results on the interval nearest to end have more weight than previous ones). In general I have enough thoughts so far, and I'll share results as I get them...

to Vinsent_Vega I also wondered how to know the amount of deals before running the optimization - the answer is simple - we run the Expert Advisor with the default parameters at a necessary interval and look at the amount of trades, then you can run it a second time at the same interval having changed the parameters very much, so you can estimate how many deals the EA will perform. Since the optimization is processor-intensive, I start with the minimum range and increase it further as needed

To Neutron, what if all parameters of the Expert Advisor were adjustable? Or in this case we will obtain a pure adjustment for the history?