expert testing of strategies - page 5

New comment
 

Notused, do you mean the inputs for the product that Artur has - or for the NS?

I will not answer the second question - I do not know myself.

And to the first - in pure form, we need a symbol, the length of the testing period, TF and the "transfer function" of the Expert Advisor being tested. More precisely, the "transfer operator", i.e. mapping of the "array of input history -> array of trades" type. Or we can call it a "transfer matrix" (probably more accurate). This information is quite sufficient for comprehensive testing of the Expert Advisor.

The form, in which this transfer operator is sent to input of the "Expert Advisor", is not so important: It can be an algorithm, i.e. ex4, mq4 (even obfuscated), code in another language, etc., or the matrix itself (of course, a proper tester must understand it).

But there is nothing close to it in the product advertised by Artur.

 
You, Matemat, are very well versed in terminology and technical erudition. But who will undertake construction of such an expert evaluator? How much time should be spent on its construction (it can take a lifetime)? To check the effectiveness (correctness of work) of expert evaluator, it should be tested at least on the knowingly profitable "transfer matrix" of the expert so that the output result of expert evaluator matches the knowingly known result (standard one). Otherwise how can we judge if the evaluator gives the correct results or not? On the other hand, if we already have a knowingly profitable Expert Advisor, why should we evaluate it?
Besides, if we are speaking about the evaluation of a strategy and not about the stability of its results, we do not need any evaluators. Any author, with the accumulated experience, is able to estimate by eye whether a strategy is viable or not.
In general, the idea of the evaluator was originally this way. For example, you have come up with a strategy, it usually has several parameters.
the parameters. If each of 4 parameters lies in the range from 0 to 100, we will get 10^8 results. How to choose the results? on Maximum Profit? on Linear Regression? on Profit percentage? as practice shows it does not work and does not give results.
 
Artur - it's like you fell out of the sky - or didn't you look at the optimiser at all? There's a genetic algorithm in there, and a report...
 
Artur писал (а): On the other hand, if we already have a knowingly profitable EA, why evaluate it?

The premise is wrong. No one in the world can say that he/she has a knowingly profitable Expert Advisor. There is no technology that proves 100% of known profitability.

P.S. In general, the optimization by 4 parameters with hundreds of values of each in one cycle of GA - is, imho, pure madness, or rather an attempt to replace our own brains with machine brains - in the hope that the machine will take care of everything.

 

The sustainability of the system has to be built in at the design stage.

There is no point in coming up with lots of strategies and then selecting the best ones.

This is unproductive.

 

Itso, not only have I not seen the optimizer, I haven't even seen what metatrader looks like. (I think in the context of this thread binding to a specific programming language and trading terminal is not necessary). But I am interested in something else. You say "optimizer", "genetic algorithm" - do you know how they work and by which algorithms? If so, I have no questions ...

I'm not saying that the Expert Advisor is obviously profitable, I'm just saying that in order to test transmission matrix tester, which you mentioned, such an Expert Advisor (a reference one) would be needed. Otherwise, how can one know that the estimator works correctly, if there is no way to compare results of the estimator with knowingly expected ones? And knowingly expected results are exactly those results that would be obtained by estimating a 100% working Expert Advisor (which, of course, does not exist)

Concerning the optimization by 4 parameters - I don't know why you think it's crazy, in my practice it is normal, here is a simple example of a strategy with a set of 4 parameters (abstract) - MACD indicator - it has two moving averages - one of them is enumerated, say, from 3 to 100, the other - from 10 to 200 and the average taken from the difference of two sliding from 5 to 50, plus RSI with enumerated from 10 to 200 - here you have the most common and quite real set of parameters, and I run a quadruple nested loop on these parameters. Yes you have to wait 20-30 minutes, but it works. So I have a question how to handle all these nearly 10^8 results. And if you put a transfer matrix analyzer into the body of such a loop (very voluminous calculations of doubtful usefulness) you might not literally survive till the end of the loop ;)

Topor - 100%

 
Artur, you please at least give a subtle hint on how you implement this check - "1 - strategy is most likely to fit the curve and has not shown itself to be robust". Thanks in advance!
 

The best analyser is a human, no one has invented a better one and hopefully never will. Since a person receives 80% of the information visually, there is no better way than a picture of quotes with the trades plotted on it. This gives more information than another black box with the name BB

If this new super method says that it is bad, screw the black box. You don't even need the numbers here.

 

Prival, what will happen to the last sell?

 
Artur:

Itso, not only have I not seen the optimiser, I haven't even seen what metatrader looks like.

?!?!??!


Artur - I haven't seen you either, but I can tell right away - you're blond...

Well, it's like - yes I have not seen your Mercedes, but on my Zaporozhets it all works! Why get attached to a particular car?

I thought you were a serious man.....

1234567
New comment